100% found this document useful (1 vote)
1K views713 pages

Smirnov - A Course of Higher Mathematics - Vol 3-2 - Complex Variables Special Functions

good text book

Uploaded by

Dung
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
100% found this document useful (1 vote)
1K views713 pages

Smirnov - A Course of Higher Mathematics - Vol 3-2 - Complex Variables Special Functions

good text book

Uploaded by

Dung
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 713

A COURSE OF

HIGHER
MATHEMATICS
V. I. Smirnov
Volume III Part 2
COMPLEX
VARIABLES
- SPECIAL
FUNCTIONS
ADIWES INTERNATIONAL SERIES
I N M A TH EM A TIC S

A. J. L o h w a t e r
Consulting Editor
A COURSE OF

Higher Mathematics
VOLUME III
PART TWO

V. I. SMIRNOV

Translated by
D. E. BROWN

Translation edited by
I. N. SNEDDON
Simson Professor in Mathematics
University o f Glasgow

PERGAMON PRESS
OXFORD • LONDON • E D I N B U R G H • NEW YORK
PARIS • F R A N K F U R T

A D D ISO N -W ESLEY P U B L IS H IN G COMPANY, INC.


READING, MASSACHUSETTS • PALO ALTO ■ LONDON
1964
Copyright (c) 1964
P E R G A M O N P R E S S LTD.

U. S. A. Edition distributed by
A D D I S O N - W E S L E Y P U B L I S H I N G COMPANY, INC.
Reading, Massachusetts • Palo A ho ■ London

P E R G A M O N P R E SS
International Series of Monographs in
P U R E A ND A P P L I E D MATHEMATICS
Volume 60

Library of Congress Catalog Card No. 63-10134

This translation has been made from


the Russian Edition of
V. I. Smirnov’s book
Kypc ebicuieu MameMamuKU (K u rs vysshei maternal iki),
published in 1957 by Fizmalgiz, Moscow
HADE IN GREAT BRITAIN
CONTENTS
I n t r o d u c t io n ix
F orew ord to the F ourth E d it io n x

CHAPTER I

THE BASIS OF THE THEORY OF FUNCTIONS


OF A COMPLEX VARIABLE 1
1. Functions of a complex variable 2. The derivative. 3. Conformal
transformation. 4. The integral. 5. Cauchy’s theorem. 6.The fundamen­
tal formula of the integral calculus. 7. Cauchy’s formula. 8. Integrals of
Cauchy’s type. 9. Corollaries of Cauchy’s formula. 10. Isolated sin­
gularities. 11. Infinite series with complex terms. 12. The Weierstrass
theorem. 13. Power series. 14. Taylor’s series. 15. Laurent’s series.
16. Examples. 17. Isolated singularities. Point at infinity. 18. Analytic
continuation. 19. Examples of many-valued functions. 20. Singularities
of analytical functions and Riemann surfaces. 21. The theorem of
residues. 22. Theorem on the number of zeros. 23. The inversion of a
power series. 24. The principle of symmetry. 25. Taylor’s series on the
circumference of the circle of convergence. 26. The principal value of
an integral. 27. The principal value of an integral (continuation).
28. Integrals of Cauchy’s type.

C H A P T E R II

CONFORMAL TRANSFORMATION
AND THE TWO-DIMENSIONAL FIELD 120

29. Conformal transformation. 30. Linear transformation. 31. Bilinear


transformation. 32. The function w—zz. 33. The function w=lc{z-\-\jz)l2
34. The biangular figure and the strip. 35. The fundamental theorem.
36. Christoffel’s formula. 37. Individual cases. 38. The exterior of the
polygon. 39. The minimum property of the transformation into a circle.
40. The method of conjugate trigonometric series. 41. The two-dimen­
sional established flow of liquids. 42. Examples. 43. The problem of
flow round a contour. 44. N. E. Zhukovskij’s formula. 45. The two-
dimensional problem of electrostatics. 46. Examples. 47. The two-
dimensional magnetic field. 48. Schwarz’s formula. 49. The core cot
(«—<)/2. 50. Limiting problems. 51. The biharmonic equation. 52.The
wave equation and analytio functions. 53. The fundamental theorem.
54. The diffraction of a two-dimensional wave. 55. The reflection of
elastic waves from rectilinear objects.
vi CONTENTS

C H A P T E R III

THE APPLICATION OF THE THEORY OF RESIDUES,


INTEGRAL AND FRACTIONAL FUNCTIONS 223

56. Fresnel’s integral. 57. Integration of expressions containing trigono­


metric functions. 58. Integration of a rational fraction. 59. Certain
new types of integrals containing trigonometric functions. 60. Jordan’s
lemma. 61. Contour integrals of certain functions. 62. Examples of
integrals of many-valued functions. 63. Integration of a system of
linear equations with constant coefficients. 64. The expansion of a
fractional function into partial fractions. 65. The function cot z. 66. The
construction of meromorphic functions. 67. Integral functions. 68. Infi­
nite products. 69. The construction of an integral function from its
given zeros. 70. Integrals which depend on parameters. 71. Euler’s
integral of the second class. 72. Euler’s integral of the first class.
73. The infinite product of the function [F(z)]“ *. 74. The representation
of r(z) by a contour integral. 75. Stirling’s formula. 76. Euler’s summa­
tion formula. 77. Bernoulli numbers. 78. Method of the steepest
descent. 79. Isolation of the principal part of an integral. 80. Examples.

C H A P T E R IV

FUNCTIONS OF SEVERAL VARIABLES AND MATRIX


FUNCTIONS 313

81. Regular functions of several variables. 82. The double integral and
Cauchy’s formula. 83. Power series. 84. Analytic continuation. 85. Mat­
rix functions. Preliminary propositions. 86. Power series of one matrix.
87. Multiplication of power series. Conversion of power series. 88. F ur­
ther investigations of convergence. 89. Interpolation polynomials.
90.Cayley’s identity and Sylvester’s formula. 91. Analytic continuation.
92. Examples of many-valued functions. 93. Systems of linear equ­
ations with constant coefficients. 94. Functions of several matrices.

CHAPTER V

LINEAR DIFFERENTIAL EQUATIONS 357

95. The expansion of a solution into a power series. 96. The analytic
continuation of the solution. 97. The neighbourhood of a singularity.
98. Regular singularity. 99. Equations of Fuchs’s class. 100. The Gauss
equation 101. The hypergeometric series. 102. The Legendre poly­
nomials. 103. Jacobian polynomials. 104. Conformal transformation
and the formula of Gauss. 105. Irregular singularities. 106. Asymptotic
expansion. 107. The Laplace transformation. 108. The choice of solu­
tions. 109. The asymptotic representation of solutions. 110. Compari­
son of results. 111. The Bessel equation. 112. The Hankel function.
113. The Bessel functions. 114. The Laplace transformation in more
general cases. 115. The generalized Laguerre polynomials. 116. Positive
CONTENTS Vll
values of the parameter. 117. The degeneration of the equation of Gauss.
118. Equations with periodic coefficients. 119. The case of analytic
coefficients. 120. Systems of linear differential equations. 121. Regular
singularities. 122. Regular systems. 123. The form of the solution in the
neighbourhood of a singularity. 124. Canonical solutions. 125. The
connection with regular solutions of Fuchs’s type. 126. The case of the
arbitrary Us. 127. Expansion in the neighbourhood of an irregular sin­
gularity. 128. Expansions into uniformly convergent series.

C I I A P T E R VJ

SPECIAL FUNCTIONS
§ 1. Spherical functions 493
129. The determination of spherical functions. 130. The definite expres­
sion for spherical functions. 131. The orthogonal properties. 132. The
Legendre polynomials. 133. The expansion in terms of spherical func­
tions. 134. Proof of convergence. 135. The connection between spherical
functions and limit problems. 136. The Dirichlet and Neumann problems.
137. The potential of voluminous masses. 138. The potential of a
spherical shell. 139. The electron in a central field. 140. Spherical
functions and the linear representation of rotating groups. 141. The
Legendre function. 142. The Legendre functions of the second kind.
§ 2. Bessel functions 537
143. The determination of Bessel functions. 144. Relationships between
Bessel functions. 145. The orthogonality of Bessel functions and their
zeros. 146. Converting function and integral representation. 147. The
Fourier—Bessel formula. 148. The Hankel and Neumann functions. 149.
The expansion of the Neumann function with an integer subscript. 150.
The case of the purely imaginary argument. 151. Integral representation
152. The asymptotic representation of Hankel functions. 153. Bessel
functions and the Laplace equation. 154. The wave equation in cylindri­
cal coordinates. 155. The wave equation in spherical coordinates.
§ 3. The Hermitian and Laguerre polynomials 584
156. The linear oscillator and the Hermitian polynomial. 157. Ortho­
gonality. 158. The conversion function. 159. Parabolic coordinates and
Hermitian functions. 160.The Laguerre polynomials. 161. The connection
between Hermitian and Laguerre polynomials. 162. The asymptotic
expression for Hermitian polynomials. 163. The asymptotic expression
for Legendre polynomials.
§ 4. Elliptic integrals and elliptic functions 604
164. The transformation of elliptic integrals into normal form. 165.
The conversionof the integrals into a trigonometric form. 166. Examples.
167. The conversion of elliptic integrals. 168. General properties of
elliptic functions. 169. Fundamental lemma. 170. The Weierstrass
CONTENTS

function. 171. The differential equation for J)(m) (608). 172. The func­
tions ak(u). 173. The expansion of a periodic integral function. 174. The
new notation. 175. The function #,(«). 176. The function &k{v). 177. The
properties of theta-functions. 178. An expression for the numbers ek in
terms of &s. 179. The elliptic Jacobian functions. 180. The fundamental
properties of Jacobian functions. 181. The differential equation for
Jacobian functions. 182. Addition formulae. 183. The connection be­
tween the functions ty(u) and sn(tt). 184. Elliptic coordinates. 185. The
introduction of elliptic functions. 186. The Lam6 equation. 187. The
simple pendulum. 188. An example of conformal transformation.

SU PP L E M E N T

THE CONVERSION OF MATRICES INTO THE


CANONICAL FORM 670
189. Auxiliary hypothesis. 190. The case of simple zeros. 191. The first
stage of the transformation in the case of repeated zeros. 192. Convers­
ion into the canonical form. 193. The determination of the structure of
the canonical form. 194. Examples.
I ndex 697
INTRODUCTION

Some observations on the aims and history of Prof. Smirnov’s five-


volume course on higher mathematics have been made in the Intro­
duction to the first volume of the present English edition.
In the present volume which forms the second part of Vol. I ll the
author comes to the discussion of one of the central subjects of modern
pure mathematics, an understanding of which is essential also to
engineers and physicists — the theory of functions of a complex
variable. Prof. Smirnov’s approach is classical and for that reason is an
excellent introduction for those students who will go on to study the
more modern developments of the theory of functions while at the
same time presenting a complete picture of those aspects of the
theory (conformal mappings, differential equations in the complex
plane, calculus of residues) which are of most direct interest to applied
mathematicians. An interesting feature of the book is that it was the first
textbook at this level to include a chapter on the theory of functions
of several complex variables, a subject which forms an essential tool
to workers in quantum field theory.
Special functions continue to play an important part in the edu­
cation of both pure and applied mathematicians. This is recognized
by Prof. Smirnov. He follows his lucid account of the general theory of
functions with a full treatment of those properties of special functions
which are necessary for the proper understanding of mathematical
physics and engineering.
Both sides of the coin are engraved with the author’s characteristic
style and the result is a work of great interest which has already been
acknowledged as a classic in the countries of Eastern Europe and is
now whole heartedly commended to students in the English-speaking
world.
I. N. S n e d d o n
FOREWORD TO THE FOURTH EDITION

I n t h i s edition Volume II I was divided into two parts. The second


part contains material from the former Volume III, beginning with
the chapter dealing with the principles of the theory of the functions
of the complex variable. I t was slightly rearranged and new material
was added. This new material is related mainly to the treatise of the
integrals of Cauchy’s type and to the approximate calculation of in­
tegrals by the method of the steepest descent.
I was greatly assisted in my treatment of the latter by Professor
G. J. Petrashen’, to whom I am deeply grateful.
References to the first part of Volume III are indicated, for
example: [IIIj, 44].
V. Sm irnov
CHAPTER I

THE BASIS OF THE THEORY


OF FUNCTIONS
OF A COMPLEX VARIABLE

1. Functions of a complex variable. In the differential and integral


calculus we assumed that the independent variable, as well as the
function, could only be real. Later, in higher algebra we analysed the
most elementary function, viz. the polynomial, in which the inde­
pendent variable took complex as well as real values. In this chapter
we shall develop the analytical basis so as to include functions of
a complex variable.
Take, for example, the polynomial

f(z) = a 0zn + ax zn~x + • . . + o„,

where ak are given complex numbers. We can assume that the inde­
pendent variable z also assumes arbitrary complex values; in this
event the function f(z) is defined for arbitrary complex values of z.
The same can be said about the rational function:
a0 zn + °i zn~ 1+ . . . + Q„
+ b1zm~ 1 + . . . + 6 m

or even about expressions containing radicals, for example:

] j z - 1.
In Chapter VI of Volume I we defined elementary transcendental
functions for the case when the independent variable assumed com­
plex values; thus we have for the exponential function:

ez = ex+ly — ex (cos y + i sin y)

and having thus defined the exponential function, we can also define
1
2 TUB BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [1
trigonometric functions the arguments of which assume complex
values:
e '2 - e ~ '2 e12 + e " ' 2
6inz = -----^ ------5 cos z= ----- ^------;
(1)
sin 2 1 e '22 - 1 COS 2 *'** +
tanz = COS z t el2z + 1 ’
cot z = sin 2 f2Z __

Let us recall the expression for the natural logarithm of a


complex number:
logz = log |z| + iargz, (2)
where | z | is the modulus of z and arg z denotes the amplitude of the
variable z. Similarly, by considering the functions inverse to (1), we
arrive at the inverse circular functions of a complex variable:
arc sin z, arc cos z, arc tan z, arc cot z.
It can easily be shown that these functions can be expressed by a
logarithm.
Assume, for example, that
e '21*’ — 1
z = tan w = i.,(e '2 + 1)
.
whence
i(ei2w + 1 )3 = ellw - 1

1 —22
Multiplying the numerator and the denominator by i and taking
logarithms, we obtain:
w = arc tan z = 21 log
° l* +, z2 .
Similarly, if we assume that:
et« _ e-<»’
z = sin w = -------- ,

we obtain a quadratic equation for e1"’:


e2iw - 2iz e‘w— 1 = 0,
whence
= iz + ]/l — z2,
and consequently:
w = - ^ log (i z + K l— 22)
where both values of the square root must be taken.
1] FUNCTIONS OF A COMPLEX VARIABLE 3

We shall see later that all the elementary functions of a complex


variable mentioned above have a derivative i.e. a definite limit
of the relationship
/(2 + Az) — f{z)
Az
exists when the complex expression A z tends to zero. This chapter
ia devoted entirely to the study of the theory of functions of a
complex variable possessing a derivative. We shall see that this
theory is distinguished by great clarity and simplicity and has wide
applications in many branches of natural science and engineering.
In this chapter we shall give a short outline of the theory itself and
in the following chapters we shall deal with its applications. In this
way we hope to achieve a clearer and more compact treatment of
the theoretical basis.
In future we shall frequently use the geometric interpretation of
complex numbers which we mentioned earlier in [I, 170].
Let us briefly recall this method of interpretation. Consider a
plane referred to perpendicular axes OX and O Y ; every point in this
plane can be defined by two real coordinates (x , y), or by one complex
coordinate x + iy, which we shall in fact use in future. In this
sense the plane is known as the plane of the complex variable,
the X axis as the real axis and the Y axis as the imaginary axis.
In the following chapters, in addition to this point representation of
a complex variable we shall also use the vector representation: when
the complex number x + iy is represented by a vector the compo­
nents components of which are equal to x and y. The relationship
between these interpretations becomes clear at once: if a vector is
drawn from the origin to the point with complex coordinate x + iy
then this vector corresponds to the same complex number x + iy.
In general, if a vector is drawn in our plane from a point A with
complex coordinate ax + ia2 to a point B with complex coordinate
bx + ib2, then this vector A B corresponds to a complex number
equal to the difference between the complex coordinates of its
origin and terminus:
(£>i Oj) i(b2 &2) •
Let us recall some earlier results from [I, 171 and 172],
The addition of complex numbers follows the laws of the geometric
addition of the vectors, corresponding to those numbers. The modulus of
the complex number is equal to the length of the corresponding vector,
4 T H E BASIS OP TH E THEORY OP FUNCTIONS OP A COMPLEX VARIABLE [1
and the amplitude to the angle which this vector makes with the
X axis.
If the complex variable varies, the corresponding point moves
in the plane.
We say that z = x + iy tends to the limit a = a + ib, where a
and b are constants, if the modulus of the difference

|a — z\ = ]/(a — x)2 + {b — y)2


tends to zero.
It follows from the above expression that, since the terms under
the radical are positive, | a — z | —>- 0 is equivalent to
x —>■a and y~+b.
Hence
x + iy~> a + ib
is equivalent to
x-> a and y -^ b .
Obviously the variable point M which corresponds to z = x + iy
will in this case tend to a limiting position represented by the point A
with complex coordinate a = a + ib. It can easily be proved, though
we shall not do so here, that the usual theorems on the limit of a
sum, product and quotient apply.
Note also that it follows from the definition of the limit that z -> 0
is equivalent to | z | —*■0. Also, if z -> a it is clear that | z | -> | o j.
Cauchy’s test for the existence of a limit also applies to a complex
variable.
Assume, for example, that we have a numbered sequence of values
of the complex variable:
zi = x 1 + y 1i;
z2 — x2 “f- y%i ) • • •;
zn = *n + yn*;---
The existence of a limit of this sequence is equivalent to the existence
of limits of the real sequences xn and yn; the necessary and sufficient
condition for such a limit to exist is that the absolute values of the
— I
differences | xn — xm \ and | yn y m should be as small as possible,
when n and m are sufficiently large [I, 31].
Taking into consideration that

K ~ zml = K(*n - *m)2+ ( y n - Vm)2


1] FUNCTIONS OF A COMPLEX VAIUABLE 5

and that the terms under the radical are positive, we can see that
the necessary and sufficient condition for the limit of the sequence
zn to exist is that | zn — zm | should tend to become as small as
desired when n and m are sufficiently large, i.e. strictly speaking,
for any given positive e an If exists such that \ zn — zm \ < e if
n and m > N. Everything said at the beginning [25] of Volume 1
about real variables also applies generally to the complex variable.
The necessary and sufficient condition for a limit of a complex
variable z to exist consists of the following [I, 31]: for any given
positive e a value of the variable z exists such that | z' — z" \ < e,
provided z' and z" are any two values which exceed the value
mentioned. We shall say in future that the complex variable z tends
to infinity when | z | -> + °°-
Consider now a function of the complex variable
w = f(z)
and let us agree on our terminology. The function f(z) can be defined
either in the whole plane or in a definite domain of the plane of the
complex variable z, for example, in a definite circle, rectangle, annulus
etc. In this domain we shall distinguish interior points and points
on the contour. Thus, for example, in the case of a circle, centre
the origin and unit radius, interior points are determined by the
condition
|z| < 1 or x 1 + y2 < 1,
whilst the contour is the circumference
\z\ = 1 or x- + y2 = 1.
An interior point has the characteristic property that not only
the point itself but also a neighbourhood of it belongs wholly to
the domain, i.e. the point M will be an interior point of a domain,
provided that a sufficiently small circle, centre at M, belongs wholly
to that domain.
Points of the contour are not interior points, though interior points
can be as close as we please to the contour. We also assume that our
domain does not break down into separate parts (connectivity of the
domain), or, to be more accurate, we assume that any two points in
the domain can be connected by a line which also lies wholly within
the domain. In future we shall understand by a domain the set of
interior points of the domain. If the contour is added to the domain
then the domain is closed.
6 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [1
We shall call a domain bounded if all its points are at a finite distance
from the origin. In future we shall enlarge further on the characteristics
of a domain.
Let us now return to the function w = f(z). Assume that it is
defined in a domain B, i.e. for every point z inside B f(z) has a definite
complex value (we are considering single-valued functions). Let z0 be
a point in B. The function f(z) is said to be continuous at the point
z0if f(z) —> f(z0) when z z0, i.e. for every given positive e a positive
value of r) exists such that | f(z) — f(z0) | < e provided | z — z0 | < rj.
A function is said to be continuous in B if it is continuous at
every point of the domain B. The function f(z) can be defined not
only in B but also on the contour I of the domain, i.e. in the closed
domain B. We say that such a function is continuous in the closed
domain B provided it is continuous at every point of the closed
domain B. When defining continuity at any given point z0 on the
contour I it must be remembered that the point z may tend to z0 in
any direction, though without leaving the domain B. The theorem
[I, 43] on real variables is also valid: if f(z) is continuous in a boun­
ded closed domain, it is uniformly continuous in this domain, i.e.
for any given positive e a positive T] exists (which is one and the
same for the whole domain such that \f(z^) — /(z2)| < e provided
| z2 — z21 < r), where z, and z2 belong to the given closed domain.
Let us write z and w = /(z), separated into real and imaginary parts:

z = x + yi)

w = f(z) = u + vi.

To specify z implies specifying x and y, and to specify /(z) implies


specifying u and v, i.e. we must consider u and v as functions of x and y:

w = /(z) = u(x, y) -f v{x, y)i. (3)

With elementary functions the separation into real and imaginary


parts can be performed by simple operations for example:

w = z2 = (* + yi)2 = (x2— y 2) + 2xyi.

Assume that z0 = x 0 + y 0 i; the condition z —>- z0 is equivalent to


x ►Xq and y y 0.
2] TH E DERIVATIVE 7

It follows from the definition of continuity at the point z0 that


when 2 -> z 0, i.e. when x —»-x„ and y -^V o we have:
/(*)-►/(*o)
or
■u(x, y) + v(x, y)i u(x0, y0) + v{xQ, y0) i,
which is equivalent to
u(x, y) u(x0, y0)
and
v{x, y ) ^ v ( x 0, y 0).
Consequently, the continuity of f(z) at the point z0 is equivalent to
the continuity of u(x, y) and v(x, y) at the point (x0, y 0).
Separating the real and imaginary parts and using the property
of continuity of elementary functions of the a variable, we can
see that the polynomials ez, sin z, cos z, are continuous functions in
the whole plane of the complex variable. A rational function is con­
tinuous everywhere except at points where its denominator vanishes.
Similarly, tan z is continuous everywhere except at the points where
cos z vanishes. As with a real variable, the sum and product of a
finite number of continuous functions is also a continuous function
and the quotient of two continuous functions is also continuous,
except for those values of z for which the denominator vanishes.
In our further treatment of the theory we shall first consider single­
valued functions, then later we shall consider specially the problem
of many-valued functions.
The function ^z — 1, the function (2) and the inverse circular
functions are all examples of many-valued functions.2

2. The derivative. Assume that f(z) is defined at a point z and a t


all points sufficiently close to z. The derivative f'(z) at the point z
is determined, as already mentioned, by the usual method, as the
limit of the equation
/(z + Az) - /(a) . .
Az ’ v ;
and this limit should be finite and determinate when the increment
Az tends to zero in any manner.
It is easy to prove that, as with a real variable, a constant factor
can be taken outside the symbol of differentiation and that the usual
laws for differentiating sums, products and quotients apply [I, 47],
8 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [2

It is also easy to show by applying Newton’s formula, that the


usual law for differentiating power functions applies when the index
is a positive integer:
(:zn)' = W2"-1 • (5)
We can thus assert that a polynomial has a derivative at any
point z and a rational fraction has a derivative everywhere, except
for those values of z for which its denominator vanishes.
Furthermore, the usual rule for differentiating a function of a func­
tion applies:
F'z {w) = F'w{w)-w'z, (6)

assuming, of course, that both derivatives on the right-hand side


of the equation exist. As with a real variable [I, 45], the existence
of a derivative at a certain point proves the continuity of f(z) at
that point.
Assume that the function f(z) is defined in a domain B, and that
it has a derivative at every point in B. If this is the case we simply
say that f(z) has a derivative in the domain B. This derivative f'(z) will
be a single-valued function in B.
We shall now introduce an important new definition. We say that
f(z) is regular or holomorphic in B if it is a single-valued function in B,
and has a continuous derivative f'{z) in B. Note that the continuity
of f(z) in B follows from the existence of a derivative. It is sometimes
said that f(z) is regular (or holomorphic) a t the point z0. This means
that f(z) is regular in a domain which contains z0.
Let us turn to formula (3) in which both z and f(z), are separated
real and imaginary parts and let us then ask the question: what con­
ditions must be satisfied by the functions u(x, y) and v(x, y) in order
that f{z) should be regular in the domain B. To start with we assume
that f(z) is regular in B and thence we draw conclusions regarding
u(x, y) and v(x, y).
As we mentioned above when defining a derivative (the existence
of which we are assuming), the increment of the independent variable
Az = Ax + Ayi may tend to zero in any manner.
Select a point M in B, the coordinate of which is z = x + yi,
and a variable point N with coordinate z + Az = (x + Ax) + (y Ay) i,
so that N tends to M.
We shall consider two cases of N tending to M, i.e. of Az tending
to zero.
2] TH E DERIVATIVE 9

In the first case we assume that N tends to M while remaining on


a straight line parallel to the X axis; we then have
Ay = 0 and Az = Az, (7)
and in the second case we assume that N tends to M while remaining
on a straight line parallel to the Y -axis; we then have
Ax = 0 and Az = %Ay. (8)
We can now construct the derivative f'(z) in both cases. In the
general case we have:

Az-+0
(9)
, [u (x + Ax, y + Ay) — u(x, y )] + i[v(x + Ax, y + Ay) — v(x, y)]
a™ Ax + iAy
A y-*0

This gives in the first case:


/ ' (z) = U m [ u(x + A x , y ) - u ( x , y)_ v(x + Ax, y) — v(x, y)
Ax

We can thus see that the real and imaginary parts on the right-hand
side of the equation must have a limit, i.e. the functions u{x, y)
and v(x, y) must have partial derivatives with respect to x, in
which case the formula applies:
!> ^ y) dv(x, y)
( 10 )
dx dx

Similarly, in the second case we have, from (8) and (9):

/' (2) = lim — [ u(xt y + Ay) — u(xt y)


+ »'
. .v(x , y + Ay) — v(x , y)
Ay-*-0 1 L Ay Ay
or
&v{x, y) y)
/'(2) 9«/ ( 11)

Comparing the expressions (10) and (11) for f'(z) we obtain the
conditions which must be satisfied by the partial derivatives of
u(x, y) and v(x, y):
9“ (X, y) _ dv(x, y) _ dv(x, y) _ du(x, y)
dx dx
( 12)
Gy dy
Notice that, by (10) and (11), the continuity of the partial
derivatives of the first order of the functions u(x, y) and v(x, y)
10 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [2

follows from the continuity of f'(z). Our earlier considerations lead us


to the following conclusion. In order that f(z) should be regular in B
the following conditions must be satisfied: u(x, y) must have con­
tinuous partial derivatives of the first order with respect to x and y
in B and these derivatives must satisfy the relationship (12).
We can now show that these conditions are not only necessary hut
are also sufficient for the regularity of f[z) in B. We shall assume that
the above conditions are satisfied and prove the existence of a con­
tinuous derivative f'(z). Taking into account the continuity of the
partial derivatives of u(x,y) and v(x,y) with respect to x and y, we
can write [I, 68]:
u{x + Ax, y + Ay) — u(x, y) =
= ^ M L Ax + ^ M L Ay + e^Alc + H A y,

v(x + Ax, y + Ay) — v(x, y) =


= i f l i , + Ay + + , <Ay.

where ek tends to zero simultaneously with Ax and Ay. Using the


latter expressions for the construction of the increment of the function
f(z + A z) —f{z) and substituting this in the equation (4), we obtain:
/(z + Az) — f(z) _
Az

(-£ -Ax + +i Ax +_l r Ay) + (Ei+ iSa) Ax +(e*+ t>4)Ay


Ax + i A y
whence, by making use of the conditions (12), we can rewrite this
relationship in the form:
/ (z + Az) —/(z) _
Az

- * L (Ax + iAy) + i ^ L ( A x + iAy) ^


Ax + iAy ^ 5 Ax + iAy ~ 6 Ax + iAy
where
Eg — —
f- iEy and £q = Ei>-f-
tend to zero simultaneously with A z.
It is easy to see that the last two terms on the right-hand side also
tend to zero.
■l\ TH E DERIVATIVE 11

In fact, for example:


p Ax —.|p |.
- — 1I Ax 1I •
a Ax + iAy 1 51 YAx2+ Ay1 ’
the first factor tends to zero and the second factor does not exceed
unity.
Thus the last formula can be rewritten in the form:
f(z + Az) — f(z) _ 9u(x, y) , dv(x, y) . , _
Az ~ dx + dx 1
where e7 tends to zero simultaneously with Az and the first two terms
on the right-hand side are independent of Az. The equation (4) thus
tends to a definite limit which is defined by formula (10). Hence
the above conditions for u(x, y) and v(x, y) are necessary and sufficient
for f(z) to be regular in B. The equations (12) are usually known as
the Cauchy-Riemann equations.
It may be recalled that we have already encountered these equa­
tions: they were satisfied by the velocity potential and the stream
function in the case of the steady-state plane flow of an ideal incom­
pressible fluid [II, 74]. We can thus see that the fundamental equa­
tions of the theory of functions of a complex variable (12) are, at the
same time, also fundamental equations in the study of the above
hydrodynamical problem. This is the basis for numerous applications
of the theory of functions of a complex variable to hydrodynamics
and we shall deal with this in the next chapter.
Let us now mention a very important circumstance which follows
from the equations (12). We shall see later that regular functions
u(x, y) and v(x, y) have derivatives of all orders. Differentiating
term by term the first of the equations (12) with respect to x and
the second with respect to y and adding, we obtain:
92w . 92u
0 X2 ' 01/2
= 0. (13i)
Similarly, it is easy to deduce from the equations (12) that
92i> . d2v -
0X2 dy*~ U‘ (132)
This shows that the real and imaginary parts of the regular function
f{z) should satisfy the Laplace equation, i.e. they should be harmonic
functions. In the following chapters we shall analyse in greater detail
the connection between the theory of functions of a complex variable
and the Laplace equation.
12 T H E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [2

Notice another important circumstance which follows from the


equations (13), viz. in the construction of a regular function we
can take its real part arbitrarily, i.e. we can take for u(x, y) any
solution of the equation (13!). We can show that in this case v(x, y)
is defined up to a constant term.
In fact, it follows from the equations (12) that
, dv , , dv , du , . du ,
d« = - a r d* + - w dy = ~ dx + "a* dy>
whence
(*. y)
v(x,y) = J- ~^~d x + - ^ - d y + C. (14)
It remains to show that the given line integral is independent of
the path and yields a certain function of its upper limit [II, 171].
Let us recall that the condition for a line integral
$ X d x + Ydy
to be independent of the path can be written as follows:
dX _ 8Y
dy dx

Applying this to the integral (14) we obtain:


d f 8 m ] __ 8 [ 8b ] 82u . 82u __ „
dy [ dy J dx ( dx J °r dx2 dy 2 ’
and this condition is satisfied since we took a harmonic function
for u(x, y). Remember that, if u(x, y) is a single-valued function
then v(x, y) may turn out to be a many-valued function, provided
the region in which we are applying formula (14) is multiply
connected [II, 72].
Let us now turn to some examples. A polynomial is obviously
a regular function in the whole 2-plane. A rational fraction is a regular
function in any domain which does not contain the zeros of its deno­
minator. If we take, for example, /(z) = z2, then u(x, y) — x2 — y2
and v(x, y) = 2xy. It can easily be shown that these functions satisfy
the equations (12).
Let us show, for example, that the exponential function
ez = e* (cos y -J- i sin y)
is regular in the whole plane. In this case:
u(x, y) = ex cos y; v(x, y) e* sin y,
3] CONFORMAL TRANSFORMATION 13

whence it follows directly that:


du Y du „ .
—- = excos y; - ^ - = _ e xsm y\
dv „ . dv r
S T = examy; - ^ - = ex cos y.
These partial derivatives are continuous and satisfy the equations
(12). We evaluate the derivative in accordance with the formula (10):
(e2)' = excos^ -f- iexsiny = ex(cosy + isiny), i.e. (ez)' = e2.
We have thus obtained the same law for differentiating an exponen­
tial function as in the case of a real variable. It is now easy to
prove that sin z and cos z also have continuous derivatives in the whole
z-plane. These derivatives are evaluated by the same laws as those
for real variables. For, on applying the rules for the differentiation
of an exponential function and a function of a function, we obtain:
, . „ ( eiz - e~ !z V e12 + e"'*
(sms) = ( ---- ----- S J = ----- 2----- = C0SZ’
. „ ( e12 + e- '2 V . e'2 — e~ iz
(cos Z) = ---- ^2-----J = 1------ 2----- = ' sin2-

3. Conformal transformation. Let us now explain the geometric


meaning of the concept of functional dependence and of a derivative.
Assume that the function/(z) is regular in a domain B in the X Y -plane.
Every value of z in the domain B corresponds to a definite value
w = f(z), and the set of all the values w = u + iv which correspond
to all the z’s in B will fill a new domain B1 which we shall draw
in a new plane of the complex variable u -f iv (Fig. 1). We can

F ig . 1
14 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [3
therefore say that our function f(z) transforms the domain B into the
domain Bv Strictly speaking we should have investigated more closely
the dependence between the points z and w as given by our function
and proved that the set of the values of w would also fill a certain
domain. Later, when we have at our disposal the necessary analytic
apparatus, we shall in fact carry out this closer investigation; for the
moment, however, we shall restrict ourselves to making general obser
vations only, which will, nevertheless, permit the reader to understand
the geometric meaning of the terms introduced. We shall see later
that, if the derivative f'(z) does not vanish at a point z, then a
sufficiently small circle, centre at z, will be transformed into a domain
in the u;-plane, which includes the corresponding point w = f(z).
Let us now explain the geometric meaning of the modulus and
amplitude of a derivative. Assume that the derivative f'(z) does not
vanish at the given point. Take two adjacent points z and z + Az.
Their corresponding points in the region BL will be w and w + Aw.
Take the lines M N and M 1N 1 which join the points z and z + Az,
and w and w + Aw respectively. These vectors correspond to the
complex numbers Az and Aw. Thus, the ratio of the lengths of these
vectors will be as follows:
_ |Aw\
|AfJV| \Az\

or, remembering that the modulus of a quotient is equal to the quotient


of the moduli:
IH/^,1 Aw
\MN\ ~~ Az ‘

In the limit, when the point N tends to M and the point M 1 tends
to N v we have:
lim
la w
|M tf| = \ f (2)|,

i.e. the modulus of the derivative f'{z) gives the change in linear dimensions
at the point z during the transformation by the function f(z). If, for
example, f(z) = z2 + z + 3, then in the course of the transformation
the linear dimensions at the point z = 1 will be magnified three times.
Let us now explain the geometric meaning of the amplitude of a
derivative. Assume that the point N tends to the point M along a
line I and let Zx be the corresponding line in the domain B1 (Fig. 2).
The amplitude of the complex number Az gives the angle between the
vector NM and the real axis and, similarly, arg w gives the angle
3] CONFORMAL TRANSFORMATION 15

between the vector MyN^ and the real axis. The difference between
the amplitudes, i.e.
arg Aw — arg/lz,
gives the angle between the vectors M 1N 1 and M N and this angle
is read from the vector M N in the counter-clockwise direction.
Remembering that the amplitude of a quotient is equal to the diffe­
rence between the amplitudes of the dividend and the divisor, we can
write:

arg Aw — arg Az = arg .

v y

In the limit the direction of the vector M N coincides with the


direction of the tangent to the curve Zat the point M, and the direction
of the vector A11N 1 coincides with the direction of the tangent to
the curve Zx at the point M v
Taking the limit of the above formula we can see that the amplitude
of the derivative arg f'(z) gives the angle of rotation at the given point z,
made as a result of the transformation by the function f(z). In other
words, if an arbitrary curve I were to be drawn through z, and
had a definite tangent at the point z then, as a result of the trans­
formation, a new curve Zj would be produced, the tangent to which
would make at the corresponding point w an angle with the above
tangent equal to the amplitude of the derivative. If we take two
curves in the domain B, which intersect at a certain angle at the
point z then, as a result of the transformation the tangents to
16 TH E BASIS OP TH E TH EO RY OP FUNCTIONS OF A COMPLEX VARIABLE [4
these curves will rotate by the same angle, equal to the amplitude
of the derivative and, consequently, the angle between the transformed
curves will remain the same in magnitude and direction as before,
i.e. transformation by a regular function conserves the angles at all
points at which the derivative of this function does not vanish. A trans­
formation under which all angles are conserved is usually called
conformal.
If we draw a net of curves in the domain B in the X F- plane then,
as a result of the transformation, we shall again obtain a net of curves,
which will, of course, be different; the angles between these curves
will, however, remain unchanged except at points where the derivative
vanishes. If we take, for example, a net of straight lines parallel to
the axes in the domain B, then we shall obtain, generally speaking,
a net of curves in the domain Bx; but the angles between these
curves will remain right angles, as before, i.e. the net will remain
orthogonal. Moreover, if we divide the domain B into small similar
squares, then each one of these squares will be transformed in the
domain B1 into a small curvilinear rectangle the sides of which will be
approximately equal to the product of the length of the side of the
square and the modulus of the derivative at an arbitrary point in the
square, i.e. this curvilinear rectangle will also be a square up to higher
order terms; however the value of | f'(z) | will be different at different
points and therefore the curvilinear squares filling B will have sides
of different lengths.
Let us consider in greater detail the problem of a function of a
function:
F(w), where w - f(z).
Let f(z) be regular in the domain B and let it transform this domain
into a domain Bv We assume further that F(w) is also regular in the
domain Bv In this case the function of a function F(w) will be regular
in the domain B and the differentiation rule, as given by formula
(6), will apply for it.

4. The integral. Let I be a curve in the X Y -plane. We shall always


assume that a curve has a parametric equation of the type:
z = <M0; y = <Pz{t),
where p^t) and ip2(t) are continuous functions with continuous deriva­
tives, or that the curve consists of a finite number of sections, each
one of which, from beginning to end, has the properties just mentioned.
4] TH E INTEGRAL 17

We already know [II, 66] that the evaluation of a line integral


J [X{x, y) dx + Y(x, y)dy]
simply involves the evaluation of the usual definite integral. It is
sufficient to substitute qj^t) and <p2(t) for x and y in the integrand,
where dx - rp[(t) d< and dy =
= <p'2(t) dt. We now have to "n-7 B
integrate with respect to the
variable t within the limits of
variation corresponding to the
curve I.
Assume that a continuous
function f(z) is given on the
curve I (Fig. 3). We shall ----
explain the concept of the con­
tour integral of the function
F ig . 3
f(z) over the curve (contour)
I. Divide the curve I into
sections between the points M v M.2........Mn_v and let zk be the
complex coordinate of the point of division M k; for the sake of
symmetry we shall denote the complex coordinate of the beginning
of the curve A by z0 and of the end of curve B by zn. Moreover,
let Ck be a point on the arc of the curve M k-.x M k. Let us write the
sum of the products:

^f(C k) {Zk-Zk-iY
k =l
The limit of this sum, when the number n of divisions increases
indefinitely and every arc M k- 2M k becomes indefinitely smaller,
is known as the contour integral of the function f(z) over the contour I:

J /(z) dz = lim J ? f(Ck) (zk — a*-!). (15)


fc=i
Denote zk = xk + yk i and t,k = + Vk i- Separating the real and
imaginary parts of f(z) we can write:

j£/(C*) (zfc—Zfc-x) =
k=l

= 2 tu^ k’ Vkj+v^k, Vk) i] [(a* ~ *fc-i) + {yk — yk-i)i]


k= 1
18 TH E BASIS OF T H E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [4
or

2 f(M (Zk — z*-i) =


k=l
n
= 2 u^ k>^ (x« — Xk~ J ~ v ^ k>^ (yk ~ y*-i) +
k=1
n
+ i 2 v^ k<^ (xk — + «(£*> *?*) (y><- y><-li-
fc-i

Bearing in mind the assumptions made with regard to the line Z


and the continuity of /(z), both sums standing on the right-hand side
of the equation tend to limits, equal to the corresponding line integrals
over Z; we thus obtain an expression for the integral (15) in the form
of the sum of the usual real line integrals:
J/(z) dz = J [u(z, y) dx — v(x, y) dt/] + i J [v{x, y) dz + u(x, y) dy]. (16)
i i i

Above, for the sake of simplicity, we assumed that the line Z has a
beginning and an end; it is evident that this definition still holds
when we integrate round closed contours.
The contour integral (15) possesses exactly the same properties as
the usual real line integral [II, 66]. Let us recall these. A constant
factor can be taken outside the sign of the integral. The integral of
a sum is equal to the sum of integrals. When the direction of the
contour of integration changes, the sign of the integral changes. If
the contour of integration were to be divided into several separate
parts, then the integral round the whole contour would be equal
to the sum of the integrals over the separate parts.
We now introduce an important inequality for the magnitude
of the integral (15). Assume that the modulus of the integrand
does not exceed a certain positive number M on the contour Z, i.e.

|/(z )|< il/ (z on Z), (17)


and let s be the length of the contour Z. In this case the following
inequality holds for the integral (15):
J
| /(z) dz | < Ms. (18)
/
For let us return to the sum (15), which gives the limit of the
integral.
5] CAUCHY’S THEOREM 19

Taking into account the fact that the modulus of a sum is less
than or equal to the sum of the moduli of the terms, we obtain:

I2 Mk) (2* - V l ) | < y |/(£*) I \zk - ZA-xl


fc=l fr=l
or, by (17):

Ifc=l
5 /(C*)(Zfc—2ft-l) I< M 2k=l h - 2*-i|.
The factor multiplying M obviously represents the perimeter of a
step line inscribed in the contour I, and by taking the limit of this
latter inequality, we obtain the inequality (18).
The integral (15) satisfies a more precise inequality, viz. denoting
by ds the differential of the arc of the curve I we obtain the
following formula:

IJ7(2) dz j < f |/(z)| ds. (19)


i i

This inequality can be obtained directly, if we replace /(z) by


| f(z) | and dz = da: + idy by | dz | = |/da?+dy2 = ds in the
integrand expression.

5. Cauchy’s theorem. We shall now put the fundamental question,


viz. under what conditions does the contour integral (16) become
independent of the path taken. It is obviously necessary and suffi­
cient that both line integrals on the right-hand side, which give
the real and imaginary parts of the contour integral, should also be
independent of the path. Applying the criterion for independence
of the path of a line integral, as given in [II, 71], we arrive at the
equations
9u{x, y) _ _ dv(x, y) _ dv(x, y) _ du(x, y)
dy dx ’ dy dx

and these are precisely the Cauchy-Riemann conditions. Hence, the


conditions for the independence of a contour integral (16) of its path
are the same as the condition for the regularity of the function /(z). This
circumstance is of fundamental importance in the integral calculus
of functions of a complex variable.
20 TH E BASIS. OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [5

Note that in deducing the conditions for the independence of a


line integral of its path we used the formula [II, 69]:

\ IP (*■y ) t e + Q (*, y) dy] = | J •


/ B ^
In deducing this formula we assumed the continuity not only
of the functions P(x, y) and Q(x, y) themselves, but also of their
partial derivatives, since they too are under the sign of the double
integral. In the case under consideration these latter are in fact
continuous, since, for the regular function/(z), both functions u(x, y)
and v(x, y) have continuous partial derivatives of the first order. In
future we shall integrate round the contour of the domain B itself.
This will be quite in order if we assume that f(z) is regular in the
domain B up to the contour i.e. it is regular in the closed domain B.
By this we mean that f(z) is regular in a somewhat larger domain,
which contains the domain B together with its contour; hence f(z)
is regular in the closed domain B if it is regular in a domain containing
B and its contour.
For a more detailed study of this problem it is necessary to take
account of the kind of domain in which f(z) is regular, i.e. in this case,
as with real contour integrals [II, 72], it is most important whether
the domain is simply or multiply connected. We shall recall the appro­
priate basic definitions and formulate the results which are analogous
with those obtained for real contour integrals.
If a bounded domain in the 2-plane has a closed curve as its contour
(in other words, has no holes) then the domain is simply connected.
If at the same time f(z) is a regular function in this domain and z0
is a point in this domain, then the integral

F(z) = f/(z')dz' (20)


i,
(where z' denotes the variable of integration), taken over an arbitrary
curve in the domain, does not depend on the path and gives a single
valued function of its upper limit of z. At the same time the value
of the integral round an arbitrary closed contour in the domain will
obviously be equal to zero. If our function f(z) is regular in a closed
domain then we can integrate round the contour of the domain B
itself, and we obtain zero as a result of the integration.
Let us now assume that our domain B is multiply connected and
bounded by one outer closed contour and by several closed interior
CAUCHY'S THEOREM 21
5]

contours. Assume, for the sake of simplicity, that there is only one
interior contour i.e. the domain is doubly connected (Fig. 4). We make
a cut X in our connecting the outer and inner contours. The cut domain
B' will now be simply connected and the expression (20) will give a
single-valued function of 2 in B'. If we assume that f(z) is regular in
the closed domain, then we can integrate round the contour of the
domain. We can then assert that the integral round the entire contour
of the simply-connected domain B ' must be zero. As indicated in
the figure, we must here integrate in the counter-clockwise direction
round the outside contour, in the clockwise direction round the inside
contour and twice in opposite directions along the cut X. The integrals
along this cut will cancel each other and we
consequently have:
f / (2 ) d z + J / ( 2 )dz = 0, (21)
O'* O'*
where I^ is the outside contour, l2is the inside
contour, and the arrows indicate the direc­
tion of integration. It follows from the
diagram that the direction of integration
for both contours can be determined from one and the same condi­
tion, viz. when describing a circuit round the contour the domain
must remain on the left-hand side. This direction will be termed positive
withrespect to the domain. Using the equation (21) we can say, thateven
in the case of a multiply-connected domain, the integral round the
contour will be equal to zero, provided we integrate everywhere in
the positive direction with respect to the domain.
If the direction of integration round the inner contour is reversed
then instead of the formula (21) we can write:

C/( 2)d 2 = f / ( z ) d z , (22)


& O'*
i.e. the integral round the outer contour is equal to the sum of the
integrals round the inner contours (in this case only one), provided
we integrate round all contours in the counter-clockwise direction.
The results obtained give the fundamental theorem in the study
of the theory of functions which is usually known as Cauchy’s theorem.
We can formulate it in several different ways.
C auchy’s Theorem I. I f a function is regular in a closed simply-connected
domain then its integral round the contour of this domain is equal to zero.
22 TH E BASIS OP TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE K
C a u c h y ’s T h e o re m II. I f a function is regular in a closed multiply,
connected domain then its integral round the entire contour of thii
domain in the positive direction is equal to zero.
C a u c h y ’s T h e o re m III. I f a function is regular in a closed multiply,
connected domain, then its integral round the outer contour is equal to
the sum of the integrals round all the inner contours, provided we integ-
rate round all contours in the counter-clockwise direction.
Let us now explain a lemma of Cauchy’s theorem which is of practi-
cal importance. Assume that different contours V and I" have the same
beginnings and ends A and B. Assume that I' can be transformed
into I’ by means of a continuous deformation without leaving
the domain in which f(z) is regular and without changing the beginning
A and end B in any way. It follows from Cauchy’s theorem that the
value of the integral of f(z) will not be affected i.e. if a contour with
fixed ends is continuously deformed without leaving the domain in which
f(z) is regular, the value of the integral of the function f(z) round the
contour will remain unchanged as a result of this deformation. The same
thing holds in the case of deformation of a closed contour, as long as
it remains closed at all times.
To conclude this we shall make a statement of fundamental impor­
tance. When deducing Cauchy’s theorem we assumed the continuity
as well as existence of the derivative f'(z). This continuity of f'(z) enters
into the definition of a regular function. By employing a different
method of proof it is possible to prove Cauchy’s theorem by using the
existence alone of f'(z) without assuming its continuity. However, we
shall see later that it follows from Cauchy’s theorem that f(z) has
derivatives of all orders, which means that f'(z) must be continuous.
Thus the second method of proof of Cauchy’s theorem, which we do
not give here, is theoretically important in that it does not use the con­
tinuity of f'(z); on the other hand, one of the consequences of this
proof is that the existence of the derivative f'(z), implies that the deriv­
ative must be continuous. In future,unless stated otherwise, we shall al­
ways integrate round a closed contour in the counter-clockwise direction.

6. The fundamental formula of the integral calculus. Assume that /(z)


is regular in a domain, and consider the function given by the formula
(20). If our domain is multiply connected, we can still assume that F(z)
is a single-valued function, by making appropriate cuts. In the same way
as in the integral calculus for functions of real variables [I, 96] it can
be shown that F(z) is a primitive for f(z), i.e. F'(z) = f(z).
6] TH E FUNDAMENTAL FORMULA OF TH E INTEGRAL CALCULUS 23

To do so we note, first of all, that it follows from the definition


of an integral as the limit of a sum that:
J dz = /S — a ,
i
where /9 and a are respectively the complex coordinates of the end
and the beginning of the contour I. We evidently have:
z+Az
F{z + Az) — F (2 ) = J / ( ') d ' ,
2 2
Z

where we can integrate, for example, along the straight line con­
necting the points 2 and 2 + Az. We can write:
z+Az
F( z + A z ) - F ( z ) = J [f(z')-f(z)+f{z)]dz’ =
Z

z+Jz z+^lz
= f(z) J dZ' + J
[/ (2') - / ( 2)]d 2 \
z z

where f(z) is taken outside the integral, Bince it does not contain the
variable of integration 2'. The latter formula can be rewritten as
follows:
z+Az
F_( z + A £ - F ( z ) = / ( 3 ) + i J [f (s0 _ f (z)] dz>. (23)
Z

It remains to be shown that the last term on the right-hand side


tends to zero when Az -*■ 0. Using the inequality for an integral
given in [4] and taking into consideration that, in this case, the length
of the path of integration is equal to | Az |, we can write:
Z +Az
i J [/(*') -/( Z)] d2' < - r ^ r - m a x | / ( 2') - / ( 2 ) | • \Az\ =
= max | / (2') — / (z) | .
We must take the maximum of the modulus of the difference
| f(z') — f(z) | when z' varies along the straight line connecting 2 and
(2 + Az). The continuous positive function | f(z') — f(z) \ of z’ attains
its maximum value on the given line at a point z' = z'0\ i.e.
max |f(z')—f(z) | = | f(zo)—f(z) |. But when Az-+0, the point 2q, which
lies on the given line, tends to 2 and, by virtue of the continuity of
f(z), the difference f(z'0) — f(z) —+ 0; it follows that the last term on
the right-hand side of the expression (23) tends to zero i.e. F'(z) = f(z).
24 TH E B ARTH 0 1 T H E TH EORY OP FUNCTIONS OF A COMPLEX VARIABLE [6

Let us now show th at when the function f(z) has two primitives ^ ( 2)
and F 2(z) they will differ by a constant term. We have by hypothesis:

F[ (2) = / (2) and F2 (2) = / (2) ,


i.e.
[^ 1 (2) — (z)]’ = 0 .
It thus remains to be shown that if the derivative of a function is
identically zero in the domain B then this function is constant in B.
Hence let f x(z) = m1(x , y) + ivL(x, y) and

/;(* )= 0 .
We can write the following two expressions for the derivative:

11 ' > 0X + 3x dy dy

and we have consequently:


9^. = n . 9«L = n . dv,
0; 0,
dx ~ ’ dy ~ ’ dx dy

from which it follows that ux and v1are independent of x and y, i.e. they
are constants; hence the function f^z) will be constant.
Assume that we have a primitive F^fz) for the function f(z). It will
differ from the primitive (2 0 ) only by a constant term, i.e.

J f(z')dz' = F 1[z) + C.
z.
To determine this constant term we assume that the end 2 coincides
with the beginning of the path z0, which gives

0 = Ft (z0) + C or C = — F x (20) ,
and the previous formula can be rewritten in the form:

J f(z') d 2' = F 1 (2) — F 1 (20) , (24)


z,

i.e. the contour integral is equal to the increment of the primitive over the
path of integration. We are assuming, of course, that the primitive F(z)
is a single-valued function and that it is regular in the domain which
contains the path of integration.
t h e f u n d a m e n t a l f o r m u l a o f t h e in t e g r a l c a l c u l u s 25
6]

Example: Consider the integral


j ( z — a)ndz, (25)
i
where n is an integer and I is a closed contour. If n differs from (—1)
the primitive will be:
- ( z - a ) n+1. (26)

This will be a regular single-valued function everywhere if n > 0, and


everywhere except at z = a when n < —1 . We assume that the contour I
does not pass through z = a. On describ­
ing a circuit round a closed contour the
single-valued function (26) will obviously
receive a zero increment and, conse­
quently, the integral (25) round an ar­
bitrary closed contour will be equal to
zero when n / —I. When n > 0 this
follows directly from Cauchy’s theorem.
When n < — 1 the result also follows from
Cauchy’s theorem as long as the point
z = a does not lie within the contour I. But the above arguments show
that if n is negative and not equal to (—1) the integral (25) will be
equal to zero, even if a lies within the contour I. In this case the
integrand is no longer regular at the point z — a since at this point
it becomes infinite.
Let us now consider the case when n = —1 , i.e. consider an integral
of the type
J t ^ T - <2 7 >

If a lies outside the closed contour I, then, in accordance with


Cauchy’s theorem, the integral will be equal to zero. Assume that
the point a lies inside the contour I (Fig. 5). Let us draw a circle C,
centre at a and small radius q. The integrand (z —a) -1 will be regular
in the annulus between the contour I and the circle C and, con­
sequently, in accordance with Cauchy’s theorem, we can integrate
round the circle C when evaluating the integral (27). On this circle
z — a = ge1*,
where y> varies in the interval (0 , 2 tt). Hence
dz = ipe1’’dip.
26 TH E BASES OP TH E T H E O B T OP FUNCTIONS OP A COMPLEX VARIABLE [7
Substituting in the integral (27) we obtain;

= 2 ni,
J 2 -0 J oe'’’
0
i.e. finally
J ¥ iA - = 2» . (28)
I
7. Cauchy’s formula. Let f(z) be a regular function in a closed
domain B, which for the moment, for the sake of simplicity, we con­
sider to be simply-connected. Let I be the contour and a an interior
point of this domain.
Let us construct the new function:
/(z) (29)
2 —0

This new function is also regular everywhere in B, except, perhaps,


at the point z = a, since at this point the denominator of the fraction
(29) vanishes. Isolate this point by a circle, centre at a and small
radius e, and let C, be its circumference. In the annulus between the
contours I and C, our function (29) will be regular without exception
and, consequently, in accordance with Cauchy’s theorem, we can write:

r_
) 2
^ _ d2
— a
= Ir 2^ —- ad 2.
I C,
In the integral on the right-hand side we put f(z) = f(a) + f(z) —
— /(a), so that
f-O E l-dj _/(< ,) f _ ^ 2_ + f ' M - ' W ^ ,
J z - f l ] 2 — a I z —a
I c, CB
or, by (28):

| —— d 2 = / 2m + [ t(z)z Z fa {n) dz-(30)


J c.
Let us now pay attention to the following circumstance: the integral
on the left-hand side of formula (30) and the first term on the right-
hand side are independent of the choice of the radius e ; we can there­
fore assert th at the second term on the right-hand side is also
independent of e. We shall now prove, however, that it tends to
zero when c —>- 0 . It then follows that it is simply equal to zero.
7]
CAUCHY'S FORMULA 27

Applying the inequality from [4] and keeping in mind the fact that
as z moves round the circumference C„ centre at a, | z — a | = e,
we get:
max |/ (2 ) - / (a) |
/ (g) - / (°) d2 < --------------- 2 tie - max | / (z) —/ (a) I • 2n.
II. 2 —0 e on ce

When e becomes infinitesimally small, the point z on the circum­


ference tends to a, and the maximum of the modulus of the difference
| f(z) — f(a) | tends to zero, i.e. the second term on the right-hand side
of the formula (30) does in fact tend to zero together with e and,
in accordance with the above argument, it will be simply equal to zero.
Thus formula (30) can be rewritten in the form:

i
Let us now somewhat modify our notation, viz. we shall now denote
the variable of integration by z' and an arbitrary point in our domain
by z. In this case the above formula takes the form:

/ ( * > = 2 5 r J ^ S r '* - <31>

This is Cauchy’s formula and it expresses the value of a regular


function at any interior point z of the domain in terms of its value on the
contour of the domain. The integral which forms part of Cauchy’s
formula contains z as a parameter in the integrand in an exceptionally
simple form.
The point z lies inside the domain and the variable point z ' on the
contour of the domain. Thus z' — z # 0 and the integral in Cauchy’s
formula is an integral of a continuous function; hence it can be differ­
entiated with respect to z under the sign of the integral as many
times as we please. Differentiating successively we obtain:

and, in general, when n is an arbitrary positive integer:

(32)
28 T H E BASIS OP TH E THEORY OP FUNCTIONS OP A COMPLEX VARIABLE [7
We can thus see that a regular function has derivatives of all orders
and these derivatives are expressed by the contour values of the func­
tion in accordance with the formula (32).

Let us prove formally the possibility of differentiating under the sign of


the integral in order to determine /'(z). We have:

/; (z+ Az) - f K(z)=


'
~2m rJ ,z ^(g/) dz' - —2m rJj jz'e —
— z — zlz
i . dz'=
z
I
= / (zQ dz'
2jii J (z' — z) (z' — z — Az)

/ (z + Az) — f (z) 1 f / (*0 dz'.


Az 2ni J (z' — z) (z' — z —■Az)

If we take the limit as Az 0 under the integral sign on the right-hand


side we obtain the expression:
/(*') (32.)
r (z) = 2m I (z' —z)2 dz'.
I
I t remains to prove the possibility of passing to the lim it under the integral
sign, i.e. it must be shown th at the difference

l
J
a = J _ r _ / £ L d*>__L ___ / (g")_____ dz'
27ii J (z' — z)2 2jii (z' —z) (z' — z — 4z)
l
tends to zero when Az 0.
After elementary rearrangements we obtain:
— Az /(*')
<5= dz'
2m z)2 [z' - (Z + dz)]

The function /(z') is always continuous on I, and its modulus is bounded,


i.e. |/(z') | < M . We denote by 2d a positive number equal to the shortest
distance between the point z and the contour I, i.e. | s ' — z \ > 2d. The point
(z + Az), when Az is close to zero, is close to z and we have | z' — (z + Az) \ > d.
Applying the usual inequality to the integral, we have:
\Az\ M -a
l<51< 2n 4d3 ’
where a is the length of the contour; it follows th at S -<■ 0 when Az -*■ 0. I t
can be shown similarly by starting with the formula (32t), th at f'(z) also has
a derivative
r(=)
'
21 r /(g°
2m J (z' —z)a
d~'
which it was required to prove.
7] CAUCHY’S FORMULA 29
The formulae (31) and (32), like Cauchy’s theorem, are also directly
applicable to a multiply-connected domain; all we have to do is
integrate round the entire contour of the domain in the positive
direction, i.e. so that the domain remains on the left.
Let us now extend Cauchy’s theorem to the case of an infinite
domain. Let f(z) be regular in the domain B formed by the part of
the plane outside the closed contour I, and let it be subjected to an
additional condition, viz. when the point z
moves to infinity the function f(z) tends to zero:
f (z) -►0 when z - > - o o . (33)
We shall show that Cauchy’s formula still
holds:

<M >

a
where we integrate so that the domain B (in
this case the part of the plane outside I)
remains on the left. To prove this we draw a
circle, centre the origin and large radius R. Our function f(z) is regu­
lar in the annulus between the contour I and the circumference CR (Fig.
6 of the circle, and for an arbitrary point z in this annulus we have

d✓ +1 *2m Jr m (35)
z' —z d
G c*
As in the proof of Cauchy’s formula we shall see that the second
term on the right-hand side is essentially independent of R and, con­
sequently, if we can prove that it tends to zero when R increases in­
definitely, it follows that it must be identically zero and formula (35)
becomes formula (34). Let us find an upper bound for the second
term on the right-hand side of the formula (35). To do so we replace
the modulus | z' — z \ in the denominator by a smaller quantity,
viz. by a difference of moduli \ z' \ — \ z | = R — \ z \ . We then
obtain the upper bound in the form:
2nR
J 2 -2 * < max |/(s')
on C b 3- 1 * 1
Cb
or
2n
I Jf 24 ^—- 2d z ' < max| /(z' )
on C b
Cb
30 TH E BASIS OF THB THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [V
If R increases indefinitely the above fraction tends to 2 n and the first
cfator max | f(z') | tends to zero in accordance with the condition
onCj
(33). We have thus proved Cauchy’s formula for an infinite domain.
Note that it follows from the proof that the condition (33) must be
satisfied uniformly with respect to z. In other words, this condition
can be fully formulated as follows: for any given s an Re exists such
that | f(z) | < e when \ z \ > R,.
Sometimes we have to deal with functions which are regular inside a
domain and have definite limits on the contour of the domain so that
they are continuous functions throughout the closed domain, although
one cannot say that they are regular in the closed domain, i.e. that they
remain regular when the domain is enlarged. Note that Cauchy's theorem
and Cauchy’s formula apply to such functions i.e. those which are regular
in a domain and continuous in a closed domain. In fact, if we compress
the contour slightly, the function now remains regular on the contour
and therefore Cauchy’s theorem evidently applies, i.e. the integral round
the contour is equal to zero. If the contour is now expanded continu­
ously so that it eventually coincides with its initial position, in the
Urnit the value of the integral round the initial contour of the domain
will also be zero. Here, of course, we can pass to the limit, because
the function is uniformly continuous in the closed domain.
It can be said that almost all the further results of this chapter are
a direct result of Cauchy’s formula. We shall return to it on many
occasions. Below we give two examples of applications of this formula.

Let us prove Cauchy’s theorem in greater detail, when /(z) is regular inside the
circle | z | <1?, centre the origin and radius B, and continuous in the closed
circle | z | < B. The function/(z) is regular in the closed circle | z | < B lt where
B t is any positive number less than B. Cauchy’s formula is applicable, and
we have:
J / (z) dz = 0.
1*1
On the circumference of this circle, z = B t e^ and dz = B x iei<r dtp, so that
2*
iBt J / {Blellp) e(?l dip = 0.
0
Since /(z) is uniformly continuous in the closed circle [1] we can prove the
possibility of passing to the limit under the integral sign as B l -►B [II, 84];
we then obtain in the l i m i t :
7] CAUCHY’S FORMULA 31

or returning to the variable z we can write:


I / (2 ) dz = 0,
1*1 = R
which is what we wished to prove. When the contours axe of a more compli­
cated type the proof becomes more involved. Cauchy’s formula for functions
which are regular inside a region and continuous in a closed region follows
from Cauchy’s theorem, as above.
Example 1. Take the exponential function f(z) = ez. This function is regular
in the whole plane and we can apply formula (32); we take as I and
arbitrary closed contour containing z as an interior point:
e2= ni_r—
2m j (r ' - z)n+1
Let us take as I a circle, centre at z and with a fixed radius gI. We now
have:
2 ' —2 = ge^; e2’ = e2 e&cos v + s’n ^ ; dz' = ige1
*^ dg>
and substituting in the above formula we obtain:
2n
1 _ w’ ( ee cos <P+ le Si n <P- in <p d
2nonJ
0
whence
n f
2n — = e ocos <p + 1 (e sin p - nip) d
nl J

On separating the real part, we obtain a definite integral of a fairly compli­


cated type:
2n
f e*cos <pcos (g sin <p— nip) dcp = 2n . (36)
.1 nl
2. Consider the rational fraction

W ) ~ n ‘h (37)
where the degree of the polynomial y>(z) in the denominator is higher than
the degree of the polynomial q>(z). This function evidently satisfies the condition
(33). Assume also th at I is a closed contour containing as interior points all
the zeros of the polynomial y>{z). We can then say th at the function (37) is
regular in the part of the plane outside I and th at Cauchy’s formula for an
infinite region holds for it. The integration over I in this formula must be
carried out in such a way that the domain outside I remains on the left, i.e.
in the clockwise direction. If we integrate in the counter-clockwise direction
the result will have the opposite sign and we thus obtain:
_ p(z) 1 f <p(z') , (38)
Y»(z) 2?nJ y(z') (2 ' —2 )
32 TH E BASIS OF TH E TH EORY OF FUNCTIONS OF A COMPLEX VARIABLE [8

Consider the integrand in this latter formula. Regarded as a function of z',


it ceases to be regular or, as is usually said, it has singularities inside I at points
where tp(z') vanishes. The point z is not a singularity since it lies outside the
contour I (in the infinite region B). The existence of these singularities, which
are zeros of the polynomial y(z'), implies th at the value of the integral (38) round
the closed contour I is not zero.

8. Integrals of Cauchy’s type. In Cauchy’s formula (31) the numerator


of the integrand repesented the value of the regular function f{z') on the
contour I. At the same time, according to Cauchy’s formula, the value
of the integral reproduced precisely the function f(z) at a point in the
domain. We shall regard the integral in Cauchy’s formula as a com­
putational device, and consider what it will yield if we substitute in
the numerator of its integrand a function which is continuous and
specified in a purely arbitrary way on the contour, about which not­
hing is known other than that it is specified and continuous on the
contour. Denote this function by co(z'). Our integral will evidently
be a function of z:

<3 9 >
I
Bearing in mind the general assumptions made with regard to the
function co(z'), the integral on the right-hand side is known as an
integral of Cauchy’s type. As in the previous section, we can differ­
entiate with respect to z under the integral sign as many times as we
please, and obtain formulae analogous to (32):

f W <*> = -35r[ p . - ’r . *'■ m

i.e. F(z) is always a regular function inside the domain B, bounded


by the closed contour I. We could, of course, have assumed that z lies
outside the contour I. In this case we should again have obtained the
formula (40) together with the formula (39), i.e. formula (39) also
defines a regular function for points z which lie outside the contour I. If
we assume that z lies on the contour I, the integral (39) becomes
meaningless, since the integrand becomes infinite on the contour of
integration. This brings us to the following result: the integral of
Cauchy’s type (39) determines two regular functions: one inside the
contour I and the other outside the contour.
Notice th at these two regular functions will generally be different.
To explain this circumstance consider the simplest case, viz. when the
8] IN'TEQRAiS OF CAUCHY’S TYPE 33

“density” to(z') in the integral of Cauchy’s type is the same as the


value on the contour of a function f(z), which is regular in the closed
domain bounded by the contour I. Thus let co(z') = f(z') be a regular
function in the closed domain bounded by the contour I. If z lies inside I
then Cauchy’s formula (31) applies and the integral of Cauchy’s type
1
2jii (41)

gives the function f(z) inside the contour. Let us now assume that z
lies outside the contour I and examine the integrand in the integral
(4 1 ) as a function of z'. Its numerator f(z') is regular in and on I and
its denominator z — z' does not vanish in or on I since we have
assumed that z lies outside I. We can therefore apply Cauchy’s theorem
and assert that the integral (41) is equal to zero provided z lies
outside I, i.e. in this case the integral of Cauchy’s type (41) gives /(z)
inside I and zero when z lies outside I.
Let us now return to Cauchy’s formula (31). In this formula the
“ density” f(z') in the integral of Cauchy’s type was the same as the
values of the function f(z) itself on the contour I. In the general case of
Cauchy’s integral (39) when co(z') is assumed to be an arbitrary continu­
ous function on the contour I, this situation obviously no longer holds.
In the case of formula (39) we have to distinguish two functions: the
function fx(z) defined by the formula (39) inside I and the function / 2(z)
defined outside I. If z tends to the point z' on the contour I from inside,
the question arises as to whether fx(z) will tend to a limit at all and if
it does tend to a limit then what will be the connection between it and
the values of co(z'). The same question can be asked with regard to the
function / 2(z) when z tends to z' from outside the contour. In this chapter
we shall not concern ourselves with this problem. By making certain
additional assumptions with regard to the functions fx(z') and / 2(z')
we find that they must have limit values, though the connection of these
with co(z') is relatively complicated. The difference between these limit
values, or to be more precise, the difference between the limit values
of /j(z) and / 2(z) when z tends to z' along the normal to the curve I,
will be exactly equal to co(z'). This is confirmed by the example of an
integral of Cauchy’s type given by (41). Here the interior limit is
f(z') and the exterior limit zero.
Integrals of Cauchy's type are frequently used in the analytical
representation of functions. Note that this representation is many-
valued, i.e. to be more precise, one function can be represented by
34 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [8

different integrals of Cauchy’s type. Let us show this by the following


example. Let I be a closed contour which encircles the origin 2 = 0
and let us define a regular function in I which is identically zero. This
function can obviously be represented by an integral of Cauchy’s type
(39) with a “densisy” co(z') = 0 . We shall show that this function, i.e.
zero, can be represented by an integral of Cauchy’s type with a
“density” co(z') = 1fz'. In fact consider the integral

,42)

and let us show that it is equal to zero whatever the position of z


inside I (we recall that the origin also lies in I). On decomposing the
rational fraction into partial fractions we can write:
1 1
z' (2 ' — z) z (z' - z)
and, consequently:
1
F (z) = — 2niz 2niz
l
From the example in [6 ] we obtain:

^ • - - T + T - 0'
Hence the integral of Cauchy’s type (42) also gives zero in I. On
adding this integral to another integral of Cauchy’s type (39) which
yields a regular function F(z), we obtain another integral of Cauchy’s
type, yielding the same function F(z). Thus we cannot conclude from
the equality of two integrals of Cauchy’s type:
1
2ni
(z inside I) (43)
I I
for any z inside I, are, that the “densities” of these integrals are the
same. This will only be the case if we impose certain additional condi­
tions on the densities. Thus, for example, the following theorem by
Harnak applies: if o 1(2 /) and co2(z') are continuous real functions and
I is the circumference of a circle, the equation (43) is equivalent
to the identity a>x(z') = co2{z').
At the end of this chapter we shall consider the problem of the limit
ing values of integrals of Cauchy’s type when the contour of the
domain is approached.
9]
COROLLARIES OP CAUCHY’S FORMULA 36

9.Corollaries of Cauchy’s formula. Let f(z) be a continuous


function which is regular in the closed domain B with the contour I, or
let it be regular inside the domain and continuous in the closed domain.
Consider the regular function [/(z)]n, where n is a positive integer, and
apply Cauchy’s formula to this function:

I
Let M be the maximum of the modulus | f(z') | on the contour I and
denote the minimum of the modulus | z' — z | by 6, i.e. the shortest
distance from the point z to the contour I.
Applying the usual inequality, we have:

where S is the length of the contour I. The above inequality can be


rewritten as follows:

l/W K^tro)'
When the positive integer n tends to infinity we obtain the following
inequality at the limit
|/(z) | < M , (44)
i.e. if f(z) is a function which is regular in a domain and continuous in
a closed domain, its maximum modulus is attained on the contour, i.e.
the modulus at any interior point of the domain is not greater than
its maximum modulus on the contour.
It can be shown that the sign of equality in the formula (44) can
only be obtained when f(z) is constant. The above property is usually
known as the principle of the maximum [or maximum modulus theorem].
Let us now consider a second corollary of Cauchy’s formula. The
function ez and a polynomial in z are examples of functions which
are regular in the whole plane. We shall show that the moduli of these
functions cannot be bounded except in the trivial case of f(z) being
constant. In other words the following theorem, generally known as
Liouville’s theorem, applies: if f(z) is regular in the whole plane and
is hounded, i.e. a positive number N exists so that for every z

\ f W\ <N. (45)
then f(z) is a constant.
36 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COJ1PLEX VARIABLE [10

We apply Cauchy’s formula to f'(z):

I
Owing to the fact th at f(z) is regular in the whole plane we can take
an arbitrary contour for the contour I which encircles the point z.
We take a circle, centre at z and radius R for the contour I and enlarge
it indefinitely. We then have:
| z' — z\ = R
and therefore:
max | / (2 ') |
| / ' ( z ) | < ~ - ^R2------2nR.

Taking into account the inequality (45), we obtain the following


result:
!/'(*)! < 4 -
The left-hand side of this inequality is independent of R and the
right-hand side tends to zero when R increases indefinitely. It there­
fore follows that f'(z) = 0 and that f(z) is a constant [6 ].
Take, for example, the function cos z. It follows from formula (1)
that its modulus increases indefinitely when z tends to infinity along
the imaginary axis. For, putting z = iy, we have
e-y + e*
cos iy --------^-----

10. Isolated singularities. Let us finally turn to a third corollary


of Cauchy’s formula, viz. to a study of the singularities of a regular
function. Assume that f(z) is a single-valued function which is regular
in the neighbourhood of the point z — a but not at the point z = a
itself. This singularity of the function is generally known as an isolated
singularity. For example, for the function

the point z — 0 will be an isolated singularity. We shall now consider


the possible types of isolated singularities.
There can be three cases: ( 1) the modulus of the function f(z) remains
bounded when the values of z are close to a; (2 ) the modulus of the
function f{z) tends to infinity when z tends to a; (3) the modulus
10] ISOLATED SINGULARITIES 37

| /(z) | does not remain bounded when z approaches a; the function,


however, does not tend to infinity but oscillates.
Let us consider the first case and show that in this event the point
z = a is not a singularity of the function f(z). In other words if f(z)
is single-valued and regular in the neighbourhood of z = a and if its
modulus is bounded in this neighbourhood then it is also regular at
the point z = a. In fact, encircle the point z =■ a by two circles with
radii R and g respectively, centre at z = a, where g < R. If z lies in
the annulus between these circles then according to Cauchy’s for­
mula we have:

qc * O ce
We shall show that the second term on the right-hand side tends
to zero when g tends to zero. As in the proof of Cauchy’s formula it
will follow from this that the second term is simply equal to zero.
The condition the modulus of the function /(z) is bounded in the
neighbourhood of z = a gives | f(z) | < N, where N is a positive
number.
We have: (z' — z) = (z' — a) — (z — a); we replace the modulus
of this difference by the smaller quantity:
| ( z ' —a) — (z — o) | > |z — a\ — \z' — o| = |z —a\ — g,
where J z' — a \ = g on Cg. We thus get the following inequality for
the given term:
i r /(»') 2 N Nq
2m J z" — dz' < 2n z — a — o ■2 ng = Iz —° I —Q
which shows that this term tends to zero when g 0. Hence the above
formula gives:

Cs
i.e. for all values of z close to a the function f(z) is expressed by an
integral of Cauchy’s type and therefore /(z) represents a function
which is regular everywhere, including the point z = a. Strictly
speaking, if f(z) is single-valued and regular near z = a and if also
its modulus is bounded then f(z) tends to a definite finite limit, when
z -> a; if we assume that f(a) is this limit then /(z) will be regular
everywhere, including the point z = a.
Let us now consider the second and third cases. The function
l(z —a) is an example of the second case and singularities of this type
38 TH E BASIS OP TH E T H E O ET OF FUNCTIONS OF A COMPLEX VARIABLE [10

are known as poles, i.e. if f(z) is single-valued and regular near the point
z = a and if it tends to infinity when z tends to a then the point a is a pole
of the function f(z).
We shall now give an example of a singularity of the third kind.
We shall show, in particular, that the point 2 = 0 will be a singularity
for the function
/(*)=e*. (46)
In fact, if 2 approaches zero from the positive direction then the
function (46) tends to + ° ° , and when z approaches zero from the
negative direction it tends to zero. Singularities of this type are known
as essential singularities, i.e. the point z = a is an essential singularity
of the function f(z) if the function is single-valued and regular in the
neighbourhood of the point z — a, yet is not hounded in this neighbour­
hood and does not tend to infinity when z — a.
We shall now prove a theorem on the values of a function in the
neighbourhood of an essential singularity. This theorem was first
proved by Sokhotskii.
Theorem. I f z = a is an essential singularity of f(z) then when z
varies in an arbitrarily small circle, centre z = a, values of f(z) are
obtained which can be as close as we please to any previously assigned
complex number.
The assertion of this theorem amounts to the following. Let y be an
arbitrarily chosen complex number and let e be an arbitrarily chosen
positive number. In this case there will be points z in an arbitrarily
small circle, centre z = a, where | /(z) — y \ < e. Let us use reductio
ad absurdum. Assume that there is a positive number ft such that at all
points of a circle C, centre at a, the following inequality holds:
| f(z) —/? | > m, where m is a positive number. Let us construct the
new function:

e h r
This function is regular in the circle 0 and its modulus is bounded:

r/V)‘- 7 T < W -
It therefore follows from the above proof that it is regular at the
point z — a; moreover, when z —>■a the function ep(z) tends to a finite
limit. Thus
1
f (z) —P + <P(z)
IN FIN ITE SERIES W IT H COMPLEX TERMS 39
11]

must also tend to a finite limit when z -»• a provided the limit of
<p(z) is not zero; or it will tend to infinity when the limit of g.(z) vanishes;
both these possibilities, however, contradict the definition of an essen­
tial singularity.
We can prove a more accurate theorem, viz:
P icard’s Theorem: I f z = a is an essential singularity of f{z) then
in any small circle, centre at a, f(z) assumes an infinite number of times
every complex value with the possible exception of one.
The proof of this theorem is vastly more complicated than the
proof of the previous theorem and we shall not attempt to give it
here. We shall only test this theorem for the function (46) which has
an essential singularity at the point 2 = 0 .
Take any complex number a, other than zero, and write the equation

e* = a. (46j)
Remembering the rules for taking the logarithms of complex
numbers we obtain the roots of the equation (46J:

_ ______ 1______
2 log | a | i (q>-(- 2kn) ’

where < p is the amplitude of a within the interval (0,2 n) and k is


an arbitrary integer. By taking its absolute value as large as we
please, we obtain zeros of the equation (46j) as close to zero as
we please. Thus the function (46) assumes an infinite number of times
any previously assigned number, except zero, in any circle, which
can be as small as we please, centre the origin. It can be shown that
the function sin 1/2 assumes an infinite number of times every value
without exception in a circle with the centre at the origin.
Poles and essential singularities are isolated singularities i.e. the
function is regular in the neighbourhood of these points. In future
when examining many-valued functions we shall deal with yet another
type of isolated singularitiy, viz. with branch points.

11. Infinite series with complex terms. Having explained some


fundamental points connected with the concept of the integral we
shall now examine infinite series with complex terms. Consider the
infinite series with complex terms:

(®i + ib 1) -(- (a2 ib2) + • • • + (o n + *"&n) + ■■■ (47)


40 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [11

This series is said to be convergent when the sum of its first n terms
S„ = (a l + a2 + • • •+ a n) + + ^ + + &n) (^ 8 )

tends to a finite limit as n increases indefinitely. It follows from


this definition that the series (47) will only converge when the series
of real terms converges
®i + + • • • and b1 + b2 + • • • . (49)
These series consist of the real and imaginary parts of the terms of
the series (47). If we denote by A and 5 the sums of the series (49) then
the finite sum of (48) will evidently tend to the limit A + iB which
represents the sum of the series (47).
Let us now explain the concept of absolute convergence of the
series (47). On replacing every term in the series (47) by its modulus
we obtain a series with positive terms:

K + + &! + ■•• (®o)


We can show that if this series converges then the initial series
(47) will also converge. In fact, from the evident inequality:
]/a2n + &n > | an I and | br \
we can see [I, 120 and 124] that the convergence of the series (50)
implies the convergence (even absolute) of the series (49), and hence
the convergence of the series (47).
I f the series (50) converges then the convergent series (47) is said to be
absolutely convergent. Such absolutely convergent series have proper­
ties analogous to those of absolutely convergent series with real terms.
If the series (47) is absolutely convergent then, as we have just
seen, the series (49) will also be absolutely convergent and their
sums are independent of the order of terms [I, 137]. Therefore we can
say the same about the sum of the series (47).
Similarly, using arguments analogous to those in [I, 138], we
can prove a theorem on the multiplication of absolutely convergent
series. In fact, if we have two absolutely convergent series of complex
terms
S = cq -f- a 2 -f- . . . and T = f}x -\- 4" • • • <
then the series
a lP l + i a l@2 + a 2^l) + ( a l/? 3 + a 2^2 + a3@l) + • • ■

will also be absolutely convergent and its sum will be equal to ST m


11] IN FIN ITE SERIES W ITH COMPLEX TERMS 41
We shall not prove this theorem in detail. Cauchy’s test for the exis­
tence of a limit applies for a complex variable and, as in the case
of a real variable, gives the necessary and sufficient condition: for the
convergence of a series of complex terms: the necessary and sufficient
condition for the series (47) to be convergent is that for any given posi­
tive e a positive N exists such that
n+p

2 (ak + ibk) < s ,


k=n+1
provided n > N, where p is an arbitrary positive integer.
Let us now examine a series with variable terms, i.e. a series the
terms of which contain the variable z:
(z) + u2 (z) + .. . (51)
If this series converges for all values of z in a domain B (on a curve I)
then it is said that the series (51) converges in the domain B (on the
curve I).
Let us now introduce the concept of uniform convergence as
we did for the real variable [I, 143]. The series (51) is said to be
uniformly convergent in the domain B (on the curve I) if for any assigned
positive e a positive N exists, which is one and the same for all values
of z in the domain B (on I), such that
n+ p

2 UkW < e , (52)


k= n+ 1

provided n > N and p is an arbitrary positive integer. Uniformly


convergent series of a complex variable have the same properties
as uniformly convergent series of a real variable [I, 146]. We shall
give two fundamental properties which can be proved in exactly the
same way as in the case of a real variable.
If the terms of the series (51) are continuous functions of z in the
domain B (on the curve I) and the series is uniformly convergent in
this domain (on the curve I) the sum of the series will also be a con­
tinuous function.
If the series (51), which consists of continuous functions, converges
uniformly on a curve I, it can be integrated term by term along this
curve.
Let us finally indicate a sufficient condition for the absolute and
uniform convergence of the series (51), which is exactly analogous
to that for the real variable [I, 147]. If for all values of z in the
42 TH E BASIS OP TH E THEORY" OP FUNCTIONS OP A COMPLEX VARIABLE [12

domain B (on the curve I) the terms of the series (51) are bounded:
\uk (z)\ < m k (k = 1, 2, ...),
where mk are positive numbers which form a convergent series, then
the series (51) is absolutely and uniformly convergent in the domain B
(on the curve I).
Let us draw attention to one further circumstance which follows
directly from the above viz. if the series (51) is uniformly con­
vergent on the curve I and we multiply all its terms by a certain
function v(z), the modulus of which is bounded on the curve, e.g. is
continuous, the new series will also be uniformly convergent.
In fact, as a result of this multiplication we obtain instead of the series
(51) the following series:
ux (z) v (z) + u2 (z) v (z) + . . . .
where | v{z) | < iV. It follows from the inequality (52) that we have
for the new series:
n+p n+p

2 Uk (Z) V (z)
k= n+ 1
IV (z) I
k = n +1
< Ne,

from which follows the uniform convergence of this series, since N


is a definite positive number and e is as small as we please when n
is large.
Having explained the elementary concepts concerning series
with complex terms we shall now prove a fundamental theorem for
series, the terms of which are regular functions of z.

12. The Weierstrass theorem. I f the terms of the series (51) are
regular functions in a closed domain B with the contour I and if this
series converges uniformly on the contour I then it converges uniformly
in the whole closed domain B, its sum is a regular function in the
domain B and each term of this series can be differentiated as many
times as we please.
Denote by z' a variable point on the curve I. It is given that
the series
(z') + u2 (z') + .. . (53)
is uniformly convergent and we therefore have the following inequality:
n+p
2 U« V) < s (for n > N and any p > 0 ).
k= n
12] TH E WKIERSTRASS THEOREM 43

The above finite sum of regular functions is also a regular function


in the closed domain B and therefore, in accordance with the prin­
ciple of the maximum, the above inequality implies the same inequ­
ality for the whole domain [9]:
n+p
£ (Z ) <e (for n > N and any p > 0 ),
h=n
from which it follows that the series (51) is uniformly convergent
in the whole closed domain.
Denoting the sum of the series (53) by 95(2 ') (a continuous function
on I), we multiply all the terms of the series by
1 1
2ni {z' — z) ’
where 2 is an interior point of the domain B:
1 y (z'l _ 1 «i (z') ,__ 1 M; (g') ,
2ni (z' —z) 2ni (z' —z ) 2m (z'—g) ‘

This series will also converge uniformly on the contour I and by


integrating it term by term round this contour we obtain:
1
. r ^ i d 2' = -2m^ Jf z'^ —- z
2ni J z ’ — 2
d2 ' + . . .
I I
But tfe have Cauchy’s formula for regular functions of the type
uk(z) and therefore the latter formula can be rewritten in the form:
1 f y ( z ')
d 2 ' = «! (2) + u2 (2) +
2m J z' — z

This shows that the sum of the series (51) can be represented by
an integral of Cauchy’s type in the domain B and that it is a regular
function. Denote this sum by <p(z):
m
2 M 2) = W 2 ) = ~ J ^ r d 2\ (54)

Note that owing to the uniform convergence of the series (51)


throughout the closed domain B, which we proved above, <p(z) is con­
tinuous in the closed domain B and formula (54) is simply Cauchy’s
formula for the function <p(z).
It only remains to be shown that the series (51) can be differentiated
term by term as many times as we please. To do this we multiply (51) by
44 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [1 2

the factor
ml
2ni (z' - z)m+l
where m is a positive integer, and integrate round Z:
ml r <p ( 2 ')
y(g/) d _, _
_ _ mi
m l_ r _
r mi (g,i _ d2' -u dz' 4 -
2m J (z '—z)m+1 (2 ' — z)'/"+1
2m J (z' + 2m J ( z ' - z ) m+1 + "■

It follows from Cauchy’s formula and from formula (54) that the
latter expression can be rewritten in the form:
9>
<m>(z) = u<m>(z) + (z) + . .. ,(55)
which shows that the series can be differentiated m times, term by term,
inside the domain. In the next section we shall apply this theorem
to a particular type of series, with which we shall deal almost exclu­
sively in future viz. to power series.
Note 1. By using the usual inequ­
ality for integrals it is easy to show
that the series (55) which is composed from
derivatives, is uniformly convergent in
any domain Bl which, together with its
contour, lies in B. We can construct
the usual expression for the series (55)
n+P
^ 4 m) (z)-
k= n

By using the Cauchy form of the derivative we obtain


n+p —t f l n+p
2 -S -J 2 * * iW -
k= n
k= n I
Let 6 be the shortest distance from the contour Zj of the domain B1
to the contour Z (Fig. 7). Applying the usual inequality to the above
integral we obtain:
n+p m\S n+p
^ 4 m)(z) <max ’
k= n 2 ndm+1 on ( k= n

where S is the length of the contour Z. Owing to the uniform con­


vergence of the series (53), the last factor on the right-hand side will
be as small as we please when n is large and this gives the condition
for the uniform convergence of the series (55). It is also easy to show
that when B is a simply-connected domain, the series obtained by term-
13] POW ER SERIE6 45

by-term integration:
z z

J ux («') dz' + J uz (z') dz' + ... ,


a a

where a is a point in B, will converge uniformly in B [c.f. I, 146].


The terms of this series are regular single-valued functions in B [6 ].
Note 2 . The Weierstrass theorem could have been formulated by using
sequences of functions instead of series [1,144]: if we have a sequence
of functions sk(z) (k = 1 , 2 , . ..), which are regular in the closed
domain B with the contour I, and if the sequence tends uniformly to a
limit on the contour I, it will tend uniformly to a limit in the whole
of the closed domain B; the limiting function s(z) will be regular in B
and for every positive integer m we have in B:
lim (z) = s<m>(z).
k~* °°

13. Power series. A power series is a series of the type


o0 + ax{z — b) + a2(z — b)2 + . . . , (56)
where ak and b are given numbers. Let us consider, first of all, the
region of convergence of the series (56). We
shall prove the following theorem:
A b e l’s T h e o re m . I f the series (56) con­
verges at a point z — z0, then it will converge
absolutely at every point z which lies nearer
to b than z0, i.e.
\z — b\ < \ z0 — b \ ,
and it will converge uniformly in any circle, cen­
tre at b and radius g smaller than \z0— 6 |, i.e.
smaller than the distance from z0 to b (Fig. 8 ).
It follows from the condition of this theorem that the series

«o + ®i (*o ~ b ) + a2 (z0 — b)2 + .. .

converges and, consequently its general term tends to zero when the
number of the term increases indefinitely. We can therefore assert
that a positive number N exists such that for every k:

Iak (zo - &)fc| < N. (57)


46 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [1 3

Consider now a circle Ce, centre at b and radius g smaller than


| z0 — 6 |, so that this radius can be denoted by g < G \ z0 — b |,
where 0 < 6 < 1 . We have for every z in this circle Cg:

| 2 — 6 | < 0 120 — 6 | . (58)

We consider the upper bounds of the terms of the series (56) in


the circle CQ. From (57) and (58) we can write:

z —b k < N6k,
I (z — b)k | = | ak (z0 — b)k
—b

from which it follows that in the circle CQ, the terms of the series
(56) have moduli less than a decreasing geometric progression com­
posed of positive numbers, i.e. the series (56) converges absolutely and
uniformly in the circle Cg. It is obvious that every point z which
lies nearer to b than z 0 can be considered as belonging to a circle
Ce and, consequently, it follows from the above, that at every such
point the series (56) converges absolutely. Abel’s theorem is thus
fully proved. We will now discuss some corollaries of this theorem.
Cobollary 1 . If the series (56) diverges at a point z = zxthen it will
obviously also diverge at every other point which is further removed
from b than z1. For, if the series converges at this latter point then,
as a result of Abel’s theorem, it must also converge at the point Zj.
We can therefore say that the following applies to the series (56):
the convergence of the series at a certain point implies its absolute
convergence in the circle which passes through this point and has its
centre at b; and the divergence of the series at a certain point implies
its divergence outside the circle which passes through this point and has
its centre at b. It follows that for every series of the type (56) a positive
number R exists such that the series (56) converges absolutely when
| z — b | < R and diverges when \ z — b\ > R , while in any circle
with a radius smaller than R, i.e. when | z — 6 | < OR (0 < 6 < 1 ),
the series (56) is convergent uniformly. This number R is known as
the radius of convergence of the series (56) and the circle | z — b | < R
as the circle of convergence of the series (compare with analogous
results obtained for a real variable) [I, 148],
Note that the above arguments do not provide conditions for the
uniform convergence of the series (56) in the whole circle of con­
vergence but only in a concentric circle drawn with a smaller radius.
We express this fact by simply saying that the series (56) converges
13] PO W ER SERIES 47

uniformly inside its circle of convergence. In general we shall say that a


series is uniformly convergent in a domain if it converges uniformly in any
closed domain which, together with its contour, lies inside the given
domain.
Note another important point in connection with the above
results. The radius of convergence R can, in certain cases, be infinite.
In that case the series (56) will converge absolutely at every point
in the plane, and it will converge uniformly in a circle drawn with
any finite radius. Note also the second extreme case, viz. when R = 0 .
In this case the series (56) diverges at every point except at the point
z = b. At this point the series will reduce to its first term. In future
we shall not deal with power series for which R — 0 .
C o r o l l a r y 2. The series (56) converges uniformly in its circle of
convergence and, consequently, the Weierstrass theorem applies to it,
i.e. the sum of the series (56) in the circle of convergence is a regular
function of z and of the series can be differentiated term by term as
many times as we please. By virtue of being a uniformly convergent
series it can also be integrated term by term. Furthermore, as a result
of their absolute convergence, power series can be multiplied term by
term like polynomials.
It follows from what has been said above that term-by-term differ­
entiation and integration of the series (56) will not affect the con­
vergence In the circle of convergence, i.e. the series:

a1 + 2a2 (z — b) + 3a3(z — b)2+ .. .; (59)

a0(z — b) (z — 6)2 + . .. (59j)

have radii of convergence which are not less than that of the series
(56). It is easy to see that the radius of convergence of series (59)
cannot be greater than the radius of convergence of the series (56).
For, suppose that the radius of convergence g of the series (59J
is greater than R, i.e. g > R. In view of what has just been said,
we do not decrease the radius of convergence on differentiating this
series and we return to the series (56); hence p < R, which contra­
dicts g > R. We can thus maintain that term-by-term differentiation
and integration of the series (56) do not alter its radius of convergence.
Note in conclusion that nothing has been said above about the
convergence of the series (56) on the circumference \ z — b\ = R of
its circle of convergence. We shall consider this problem later.
48 TH E BASIE OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [14
14. Taylor’s series. We saw above that the sum of the series (56)
is a regular function in the circle of convergence of this series. We shall
now prove the converse proposition: any function f(z) which is regular
in a circle \z — b \ < R , centre at b, can be represented in this circle by a
power series of the form (56) and this representation is unique.
Take a fixed point z in the circle
| z — b | < R. Draw a circle CRl, centre at b
and radius R 1 which is smaller than R, that
z lies in CRl (Fig. 9). We can express f(z) by
Cauchy’s formula, by integrating round CRl:

<6 0 >

In CRl we have | z' — b \ = R1 and, on the


other hand, | z — b | < Rlt since z lies in
CRl. Using the formula which gives the
sum of the terms of an infinitely decreasing geometric progression,
we can write:

1 1 _ 1 _ ^ (z - b)k
z' — z z' — b , s —b (z' — b)k+1 7
(61)
1 ——--- =
— k—0
z —b
where we have the following expression for the moduli of the terms
of this series:
(s - b)k z —b
1 ak (a
(z' - b)k+1 ~ rJ \q - z' — b

and it follows from the above that 0 < q < 1. Hence the infinite
series (61) converges uniformly with respect to z' in CRl. Multiplying
both sides by

and integrating term by term round to CRl we obtain from (60):

2 " ( * - » )'• 5 3 -J T ^ - i k d2'


k=0 c*.
or
f(z) = ak{z - b)k, (62)
A=0
14]
tatloh ’s series 49

where, by Cauchy’s formula [7]:


1 f Z(z') j . t k)(b)
2ni | ( r ' -& )* + * dz lcl ’ (62J
Cjt,
i e. the value of f(z) at an arbitrary point in the circle \ z — b \ < R ,
where f(z) is regular, is represented by the Taylor series:

f { z ) = m + I ^ L { z ~ b ) + J l ^ L { z - b Y + ... (63)

We can show that the representation of f(z) by a power series is


unique. Assume that f(z) can be represented in a circle, centre at b,
by a series of the type (62). We will show that the coefficients ak
are uniquely determined viz. they must be Taylor’s coefficients. In
fact, assuming in (62) that z = b, we obtain a0 = f(b). Differentiating
the power series (62):
/' («) = ^ hak (z - b)k~l .
k=l
On again putting z =
we obtain in general:
ak = / wk\(b)
and formula (62) must be the same as Taylor’s series (63). Hence
if we could obtain in two different ways expansions of one and the
same function into power series in positive integral powers of (z — 6 )
the coefficients of the same powers of (z — b) must be the same in
both expansions.
The above arguments show that Taylor’s series (63) for the function
f(z) converges in a circle, centre at b, in which f(z) is regular; in this
circle the sum of the Taylor’s series is equal to f(z).
Upper bounds for the coefficients in Taylor’s series follow directly
from the above expression. Let R be the radius of convergence of
the series (62). In formula (62!) we take a circle, centre at b and radius
(R— e), for CRl, where e is a small fixed positive number. On this
circle our function /(z) is regular, its modulus does not exceed a
positive number M and, evidently, | z ' — b \ = (R — e). The usual
upper bound for the integral gives:
71f
HI < T3T±lj.- ■ <M>
e can be taken as close to zero as we please though the value of
M evidently depends on the choice of e.
50 TH E BASIS OP TH E THEORY OF FUNCTIONS OP A COMPLEX VARIABLE [15

Let us apply the Weierstrass theorem which we proved in [ 12 ] to


the case of a power series. Assume that we are given regular functions
in a circle CR, centre at b:
uk (z) = (z — b) + 4 k>(2 — by + ----
and assume that the series

2 Uk W
k= 1
converges uniformly in this circle. In this case, in accordance with
the Weierstrass theorem, its sum is also a regular function in this
circle and it can therefore be represented by a power series:

2 t°ofc>+ °ik) (z - b) + 4 fc) ( z - b ) * + . . . ] =


k-1
= a0 + ax{z — b) + a2(z — b)2 + . . .
In accordance with the Weierstrass theorem we can differentiate
this series term by term as many times as we please. Differentiating
and putting z = b, we obtain the following expression for the sum of
the coefficients of this series:

a0 = 2 aok) • ai = 2 aik) ; = 2 aik) - • • •,


A=1 fc=l fc=l

i.e. given the above assumptions, we can add these infinite series like
ordinary polynomials.

15. Laurent’s series. I t is not difficult to obtain results similar to


the above for power series of a more general type:
. . . + o_ 2 (2 — b)~2 + a_L (2 — 6 ) - 1 + a0 +
+ o 1 (2 - 6) + o2 (2 - 6)2 + ..., (65)
which contain negative as well as positive integral powers of (z — 6 ).
The series (65) is usually known as Laurent’s series. We shall first of
all determine its domain of convergence. The series (65) consists of
two series:
a 0 + a 1 (2 — 6 ) + a 2 (2 — 6 )2+ . . . (6 6 x)
and
15] LAURENT'S SERIES 61
and we have to find the domain in which these two series converge.
This will be the domain of convergence of the series (65). The series
(66!) is the usual power series of the type we considered above and
its domain of convergence is a circle, centre at b. Let this circle be
| z — b | < i?!- To analyse the series (66 2) we replace z by a new
variable z' in accordance with the formula z' = (z —ft)-1. This trans­
forms the series (66 2) into the usual power series of the type:
0-1 Z' + 0 _ 2 Z’2 + ...
Its domain of convergence in the z'-plane is a circle, centre the
origin (zero takes the place of b). Denote the radius of this circle
by 1 IB2, s o that the domain of convergence of this latter series will
be | z’ | < l/ii!2 or 1/| z' | > i?2. Returning to the former variable z,
we obtain the domain of convergence in the form | z — b | > R2.
Thus the domain of convergence of the complete series (65) is given
by two inequalities:
|z - 6 | < R2, |z — b \ > R 2. (67)
The first inequality defines the interior of the circle, centre at
b and radius Rv and this is the domain of convergence of the series
(662). The second inequality (67) defines the part of the plane outside
the circle with centre at b and radius R2 and this is the domain of
convergence of the series (66 2). When R2 < R2 the inequalities (67)
do not define a domain When R± > R2 the inequalities (67) define
a circular annulus
R 2< \ z — b \ < R v (6 8 )
bounded by two concentric circles, centre at b and radii R2 and Rl
respectively. Hence the domain of convergence of a series of the type
(65) is the circular annulus (6 8 ).
Above we split the series (65) into two power series; it follows
from the theory of power series that the series (65) converges absolutely
and uniformly in its annulus of convergence, the sum of the series
is a regular function and the series can be differentiated term by
term. Note that in the inequality (6 8 ) which defines the dimensions
of the annulus, the inner radius R2 can vanish and in that event the
series (65) will converge for all z sufficiently close to b. Similarly
the outside radius S x can become infinite, in which case the series
(65) will converge for all values of z which satisfy the condition
I z — b | > R2. If the annulus is defined by the inequality
52 TH E BASIS OP TH E THEORY OP FUNCTIONS OF A COMPLEX VARIABLE [1 5

0 < | z — 6 | < oo then the series (65) will converge in the whole
2 -plane except at the point 2 = 6 .
Note also that that part of Laurent’s series (65) which contains
positive powers of (2 — b) converges not only in the annulus (6 8 ) but
also everywhere in the outer circle, i.e. where | 2 — b | < iJjj the
part of the series which contains negative powers of (2 — b) converges
everywhere outside the inner circle, i.e. where | 2 — b | > R2. If, for ex­
ample, the series contains a finite number of
terms with negative powers, the condition
R2 = 0 applies, and if it contains a finite
number of terms with positive powers of
(2 — 6 ), then the condition R1= °° applies.
We must emphasize once again the fact
that we are only considering Laurent’s se­
ries for which R2 < R1 since otherwise the
series has no domain of convergence.
The converse may be proved in the same
way as for power series, viz. if f(z) is regular
in the annulus (6 8 ), it can be represented by a Laurent's series in this
annulus and this representation is unique.
If we slightly compress the outer circumference of the annulus and
slightly enlarge the inner circumference, /(z) will also be regular on
both contours of the annulus. Denote these contours by (7/^ and CRl.
Applying Cauchy’s formula to an arbitrary point 2 in this annulus
(Fig. 10), we obtain:

/(z) = ~2m f J -z'^—L2d z ' - f —


2m J
f 4 ^ - d 2' .
z —z
(69)
C*i Cr%

When integrating round the circle CRl we have

and in the same way as in the proof of Taylor’s theorem we can


represent the fraction in the integrand by a series which converges
uniformly on the circle CR :

■v ( z - b ) k
2' —2 kt 0 (2' - b)k+1
Multiplying by
(70)
15] LAURENT’S SERIES 53

and integrating round CRl we obtain for the first term on the right-
hand side of the formula (69) a representation as a power series in
positive powers of (z — b):

"2^i" J I ' - z d z ' — a o + a i (z — b) + °2 (z ~ &)2 + • ■ • >

where
i r ___ m n
ak 27ii J iz' _ ,k+i ■dz'.
( * ' - by

We have on the contrary, on integrating round CRi:


z '- b
z —b < 1,

and instead of the above fraction we must write another expansion,


uniformly convergent on the circle CRi:
1 _ J__ 1__ _ “ (z ' - b )k
z' —z z —b ' , z '- b ~ ( Z - b)k+1 ’
1~ ,_ h k- 0 V

whence again multiplying by the factor (70), we get an expression for


the second term on the right-hand side of the formula (69) in the form
of a power series in negative integral powers of (z — b):

IST
o c*.
JiA— d2' = 0 - 1( z - 6 ) - 1 + a_2(3- 6 ) - 2 + . . . ,

where

Combining both terms we obtain an expression for the function


/(z) in the annulus in the form of Laurent’s series:

/ ( *) = ^ “x i z - W - (71)
fc=—OO
It remains to be shown that this representation is unique. For this
purpose we shall show, in the same way as for Taylor’s series,
that the formula (71) gives well-defined expressions for the coeffi­
cients of the expansion ak. Let 1 be a closed contour in the
annulus (6 8 ) which encircles b. On this contour the series (71) con­
verges uniformly. We select an integer m, multiply both sides of the
54 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [ 16

equation (71) by (z — and integrate round I in the counter­


clockwise direction:

J (z - b)-m~1f(z) dz = ak J (z - h)*-”- 1dz.


I k ---- -- /

We know from [6 ] that all the integrals on the right-hand side will
be equal to zero except one which contains (z — 6 )-1, the inte­
grand of. This integral will be obtained in the term corresponding to
k = m, and will be equal, as we know, to 2 ni. Hence the above
formula gives:
f (z — b)~m~1f(z) dz = 2niam,
i
whence well defined expressions for the coefficients are obtained:

<*rn= ~ J ( z - 6 r" -V (z )d z (m = 0, ± 1 , ± 2 , . . . ) . (72)


i

16. Examples. Applying the expansion into Taylor’s series to the


elementary transcendental functions, we obtain expansions into power
series which are familiar from the differential calculus; these series will
also hold when the independent variable assumes complex values.
Example 1. For the function f(z) = e2 we evidently have f n\z) = ez,
and consequently, / (n)(0) = 1. Formula (63), when 6 = 0, gives
(McLaurin’s series):

e 2 = l + T T + ~li- + ■•• (73)


Our function e2 is regular in the whole plane and, consequently,
the expansion (73) holds in the whole plane.
Similarly we can obtain expansions for the trigonometric functions
which also hold in the whole plane:

sm z — —2 S3 , 2s
gj- + -gj ... (74)

c o s z = l - — + -^- —. . . (75)

2. The formula for a geometric progression

- ^ = 1 + * + =? + . . .
serves as an example of a series with a circle of convergence | z \ < 1.
16] exam ples 65

We replace z by (—2) in this series and integrate between the


limits 0 and 2 :
Z
/ \ f dz z z1 . z*
VW ~ J 1+z ~ 1 2~ + 1 ••• (7 6 )
0
We thus obtain a new power series with the same circle of con­
vergen ce | z | < 1. When 2 assumes real values its sum, as we know
from [I, 132], is equal to log (1 + 2). We will show that the same
thing applies to all complex values of 2 in the circle | z | < 1 , i.e.
strictly speaking, we will show that the sum of our series

? < z ) = / l ( ">
0
satisfies the equation
e*<2>= 1 + 3 . (78)
We take the function e9(z) = f(z) which is regular in the circle
| z | < 1 and write its expansion into a McLaurin’s series. To do
so we have to find the derivatives of this function. Taking into
account that <p'{z) = 1/(1 + 2), we evidently have:
/ ' ( 2) = e ^ ) . T — (79)
and furthermore
. f” (2 ) = e'p<2) • --------- e^z)--------- 0
1 ' > (1+z)2 (1 + z)2 ~ ’
i.e. / (n)(z) = 0 when n > 2 . It also follows from the formulae (77)
and (79) that /(0) = e° = 1 and /'(0) = 1. Thus the expansion of f{z)
into McLaurin’s series does, in fact, give us:
/( 2) = e*<2>= 1 + 2.
We can thus see that the sum of the series (76) is one of the possible
values of log(l + 2). This latter function is many-valued, but the
power series (76) singles out a single-valued branch which is regular
in the circle | 2 | < 1 :

log(l + *) = - f - 4 + - T - — (8°)
Values of the logarithm as given by this formula are sometimes known
as the principal values of the logarithm. The circumference of our circle
of convergence passes through the singular point 2 = —1 which belongs
to the function log (1 + 2). The character of this singularity will be
explained later.
66 TH E BASIS OP TH E THEORY OP FUNCTIONS OF A COMPLEX VARIABLE [16

3. Consider the function (1 + z)m- When m is a positive integer


its expansion in integral positive powers of z is given by the usual
Newton’s binomial formula. When m is a negative integer the func­
tion will have a pole at the point z = —1 ; on evaluating successively
its derivatives and constructing McLaurin’s series, we obtain an exp.
ansion in the circle | z | < 1 [I, 131]:

(i + z r = i + ^ - z + z * + m(m- 31,)(m- 2) z3 + . . .
(81)
When m is not an integer, our function will be many-valued. For
example, when m = l / 2 we have |/l+ z . In general, for an arbitrary
value of the constant m we can write our function in the form [I, 176]:
(1 + z)m = eml0g<1+2\ (82)
and this function will be many-valued since the function log (1 + z)
is many-valued. Replace log(l -f z) by the value given by the
equation (80). In this case the function (82) will be a single-valued
function which is regular in the circle | z | < 1. Evaluating successively
the derivatives of the function (82), we have from (77):

[(1 + z ) my - em l°e<1+*). - me*'"-1' l0g<1+2>= m ( l + z ) m~ \

[(1 + z)mY = m(m - 1 ) e*™-2) logd+z) = m(m - 1) (1 + z )m~ 2


and in general:
[(1 + z)m] « = m(m - 1) .... (m - k + 1 ) e<m- k>log<1+i>=
- m(m — 1 ) . . . (m — k + 1 ) (1 + z)m~k ,
where log (1 + z) is defined by the series (80). Note that the series
(80) gives th at value of log (1 + z) which vanishes when z = 0.
Thus formula (82) and the subsequent formulae give:

and
(i + z)mU = i; [(i+ z)m]'Uo = ™
[(1 + zrF>|z=o = m(m - 1) . . . (m - k + 1 ).

This shows that for our function (82) McLaurin’s series coincides
with the series (81), i.e. formula (81) gives a regular single-valued
value for the function (82) in the circle | z | < 1 for an arbitrary
index m. In future we shall call formula (81) Newton’s binomial
formula.
16] EXAMPLES 67

4. Replacing z b y (—z2)in the formula for a progression, we obtain


the expansion, valid in the circle | z | < 1 :

T f T " = 1 — z2 + z4 — . . .

On integrating between the limits 0 and z we obtain a new expan­


sion, which also holds in the same circle:

I T T ? = “T------T + JS— • " <“ >


0

We shall see later that the sum of the above series gives one of
the possible values of arc tan z, and formula (83) thus defines one
branch of a many-valued function in the circle | z | < 1 , namely a
branch which is regular and single-valued in the given circle.
The expansion for one of the branches of the many-valued function
arc sin z in the same circle can be obtained similarly:
£
dz 1 Z& 1- 3
+ ~2 ~ T + 2 - (84)
f V r=i 4

5. Consider the function

^ = z(z - 1) (z - 2) •
The poles z = 0, z = 1, z = 2, are singularities of this function but otherwise
it is single-valued and regular in the whole plane. Consider three circular
annuli, centre the origin:
(Kt) 0 < |z| < 1; (KJ l<|z|<2; (K3) 2 < |z| < + oo.
In each one of these our function can be expanded into a Laurent’s series
in integral powers of z. For example, on decomposing /(z) into partial frac­
tions in the annulus K it we have:

2 1
z z -2 ’
whence, since 1 < | z | < 2, we have in the annulus:

z —1 2 Zk+1 ’
2 1 - 1 k=o z
z
1 1 1 “ z*
z- 2 2 . z “ 2 2 o* ’
1 --- =- k=o z
58 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [16

and finally, in the annulus K t we have:

Similarly, in the annulus K 3, where [ z [ > 2, we have an expansion in nega^


tive powers of z only:
1 _ - i 1 _ 1 1 _ “ 2*
z —1 — 2 _fc+i ’ z —2 z , 2 — 2 ,k+i
k=0 z 1 -----------k=0 z
z
or

/(*) = 2 (2*- 1- 1)
fc=2
Our function will also be regular, for example, in an annulus, centre at
z = l , inner radius ii2 = 0 and outer radius fJ, = 1. In this annulus it can
easily be represented by a Laurent’s series in integral powers of (z — 1).
6. Consider the quotient of two power series:
6„ + b,z + b2z » + . . . g
o0 + % z + -2+ • • ■
Let the radii of convergence of both series be not less than a positive
number g. Assume also that the constant term a 0 of the series which appears
in the denominator does not vanish. In that case the function in the denominator
will not vanish at the origin or in some circle with centre at the origin. Assume
th at it is regular and not equal to zero in the circle | z | < gL. We can
assert th at the fraction (85) will be regular in a circle, centre the origin, the
radius of which is equal to the lesser of the two numbers g and (or, per­
haps, even in a larger circle). In this circle we have an expansion of the
function into a power series:
b0 + b, z + 5; z2 + ■■.
= c„ + cx z + c2z2 + . . .
®o+ °i z + “***+•••
To evaluate the coefficients ck we multiply the quotient by the divisor and
obtain the product in the form of a power series-
Op c0 + (a, c„ + a„ c,) z + (o2 c0 + a, c, + o0 c2) z! + . . . = 60 + b, z + b2 z2 + . . .
The product obtained must coincide with the dividend and by virtue of
the uniqueness of the expansion into a power series, we can simply equate
the coefficients of like powers of z. This gives a series of equations for the coeffi­
cients ck of the quotient:
°o co = bp
°i co + ap ci = bi ( 86)
at C P "t” ° i ci "f- a o cz = b2

The coefficients ck can be evaluated successively from the above formulae.


The first (n + 1) equations (86) can be treated as a system of (n + 1) equations
ISOLATED SINGULARITIES. POINT AT INFIN ITY 59
17]

in the unknowns c01 c,, ..., cn. On solving them with the aid of Cramer’s for­
mula we can write an expression for the coefficient c„ in the form of a quotient
0f two determinants:
o, 0, .... . 0, K
°1. o 0> 0, .... . 0, bt
o 2, o „ °o- • ■. . . 0, b,

a n - l< a 77-2> a n -3’ • '■ ■» O 0 ,


a„ . a n - 1> a n - 2 ’ • ■ *» O j , bn
CTo» 0, 0, 0, 0

<xlt a„. 0, ., 0, 0

o 2, O il a 0, .. 0, 0

a n - 1’ ° n - 2 > a n - 3> • • • i O 0, 0

aw °n -i> a n - 2» • • . i O f,

On applying these arguments to the expansion

2 Z3
, sm 2 H 3T + ‘ "
tan 2 = -------
cos 2
2! ■+•••

we obtain a power series for tan z in the circle | z | <ji/2, since, os we shall
see later, the function cos z has only the real roots which are familiar from
trigonometry.

17. Isolated singularities. Point at infinity. Assume that the function


f(z) is single-valued and regular in the neighbourhood of the point
z = b, but not at the point z = b itself. It will thus be regular in an
annulus, centre at b and inner radius zero, and it can expanded
in this annulus, i.e. in the neighbourhood of the point b, into a Laurent’s
series in integral powers of (z —b). In this case there are three
possibilities: (1) the series does not contain terms with negative powers
of (z — 6 ), (2 ) the series contains a finite number of terms with negative
powers of (z — b) and (3) the series contains an infinite number of
terms with negative powers of (z — b).
In the first case the series which gives the function /(z) and does
not contain negative powers of (z — b) is simply Taylor’s series and
our function will in fact be regular at the point z = 6 . Let us consider
the second case, when the series has the form:

/(* )= 2 ak ( z - b ) k, ( 88 )
fc=—m
60 TH E BASIS OP TH E THEORY OP FUNCTIONS OP A COMPLEX VARIABLE [17

where the coefficient a_m does not vanish. Formula (8 8 ) can be


rewritten in the form:

/(*) J7 l> -m + a-m +1(«~ b) + a-m+2 (* ~ W + •■•


]•
(2 —b)
When z tends to 6 the factor outside the square bracket tends to
infinity and the square bracket tends to the limit a_m, which is
finite and non-zero (the sum of a power series is a continuous func­
tion) ; therefore the product tends to infinity. Thus in the second case,
in accordance with our earlier terminology [ 1 0 ], the point 6 is a pole
of the function f(z). Let us introduce a new terminology which is in
general use, viz. in the expansion (8 8 ) the point b is known as a
pole of order m; the sum of terms in negative powers
a-m____ I____°-m+t_____i_ i a-\
(2 - b)m "r (2 - b)m-1 ^ ^ Z-b
(a—m ^ 0 )

is known as the infinite part corresponding to this pole. The coefficient


of (z — 6 )-1, viz. a_x has a special name, viz. it is known as the residue
of the function f(z) at the pole b.
We will now show that an expansion of the form (8 8 ) always holds
when 6 is a pole of the function. Thus, let f(z) be single-valued and
regular in the neighbourhood of b and let it tend to infinity when
z —►b. We will show that an expansion of the form (8 8 ) holds. Consider
the function
l
<p{z) -
/(2)
It is regular in the neighbourhood of the point b and tends to zero
when z-*-b. Consequently <p(z) is also regular at the point b [10],
where it vanishes. Let us write its expansion into a Taylor’s series.
In this expansion the constant term will certainly be absent. Assume
that the first term of this expansion, which is not zero, contains
(z - b)m, i.e.
<p(z) = bm(z - b)m + bm+1(z - 6)m+ 1 + ••• (bm * 0 ).
From above we have the following formula for f(z):

ftz\ = 1 = 1 . !________
n <P(Z) (2 — b)m bm + bm+l(z — b ) + . . .

The denominator of the second fraction written does not vanish when
2 = 6 and therefore this fraction can be expanded into a Taylor’s
series in positive powers of (z — 6 ). Dividing this Taylor’s series by
ISOLATED SINGULARITIES. POINT AT INFINITY 61
17]
(z _ b)m we obtain an expansion for /(z) in the form (8 8 ). Comparing
the last result with the one above, we can assert that the term
“pole”, introduced by us in [ 1 0 ] is equivalent to the concept of
a singularity in the neighbourhood of which the function can be
expanded into a Laurent’s series with a finite number of terms in
negative powers of (z — b). Consequently an essential singularity will
be a point in the neighbourhood of which the function f(z) can be
expanded into a Laurent’s series with an infinite number of terms in
negative powers of (z — b). Here, as with a pole, the coefficient of
(z _b)-1 is known as the residue of f(z) at the essential singularity b.
Note that in the expansion of 97(2) there must be a non-vanishing
coefficient bm, since otherwise 97(2) would be identically zero in a circle,
centre at b, and this contradicts the equation 97(2) = 1/f(z), since f(z),
according to the given conditions, must be regular in the neighbour­
hood of 2 = b.
We shall now introduce the concept of the point at infinity. We
consider that a plane has one point at infinity. The neighbourhood
of this point at infinity is defined as the part of the plane outside
a circle, centre the origin. This neighbourhood is defined by an
inequality of the form | 2 | > R. We could, of course, take the centre
of the circle at a point other than the origin, i.e. instead of the above
inequality the neighbourhood of the point at infinity could be defined
by an inequality of the form | 2 — a | > R, and this would not cause
any fundamental changes. However, we shall use the first condition
| z | > R.
Let /(z) be a single-valued function which is regular in the neigh­
bourhood of the point at infinity. We can regard this neighbourhood
as a circular annulus, centre the origin, inner radius R and outer
radius infinity. In this annulus it must be possible to expand /(z)
into a Laurent’s series in integral powers of z and, as before, three
different cases arise.
In the first instance we consider the case when the Laurent’s
series contains no terms in positive powers of 2 , i.e. when the expan­
sion has the form:
/(* )= B o+ - ^ + - 5 - + . . . (89)

When 2 tends to infinity, /(z) tends to a finite limit a0 and it is


said that f(z) is regular at the point at infinity, where f(°°) = a0.
In the second instance we consider the case when the expansion
of f(z) into a Laurent’s series contains a finite number of terms in
62 TH E BASIS OP TH E THEORY OF FUNCTIONS OP A COMPLEX VARIABLE [17

positive powers of z:
f{z) = a_mzm+ a _ m+1 zm~1+ ... +o_! z + a 0+ — +

+ -%- + ■■• (a_m* 0 ) . (90)


Taking zm outside the bracket, we can see, as above, that /(z) tends
to infinity when z —►°o, while the quotient f(z)jzm tends to a finite
non-zero limit a_m,. In this case the point at infinity is known as a pole
of f(z) of order m and the set of points (a_m zm + . . . + a-i z) is known
as the infinite part at this pole.
Finally, if the expansion contains an infinite number of terms in
positive powers of z:
/(z) = . . . + a_ 2 z2 + a_xz + a 0 -1— ^— b + • ■•» (91)
then the point at infinity is known as an essential singularity of the
function f(z). If we replace z by a new independent variable t according
to the formula

the neighbourhood of the point at infinity of the z-plane will


be transformed into the neighbourhood of the origin' in the plane
and the expansion (91) will contain an infinite number of terms in
negative powers of t. It follows that if, z = °° is an essential singularity
of f(z), on variation of z outside any circle, centre the origin, radius
as large as we please, a value of f(z) may be obtained as close as we
please to any arbitrary, previously assigned complex number, and
indeed, /(z) takes any complex value, with the possible exception of
one [10] an infinite number of times. In all three cases the coefficient
a± of z -1 with the reverse sign is known as the residue at the point
at infinity, i.e. (—a^. The purpose of this definition of the residue will
be explained later.
Note that, when the point z = a is a pole of the function f(z), wo
write f(a) = °° and say that w = /(z) transforms z = a into the point
at infinity. When z = °o is a pole of /(z), this is usually written as
/(oo) — oo and it is said that w = f(z) transforms the point at in­
finity into itself, i.e. the point remains in its former position.
Returning to [7] we can see that the condition for the applicability
of Cauchy’s formula to a domain containing the point at infinity can
be formulated as follows:
f(z) -*■ 0 uniformly when z-> oo,
18] ANALYTIC CONTINUATION 63

and means that f(z) is regular at the point at infinity, whilst in the
expansion (89) a0 = 0, i.e. /( » ) = 0.
Example 1. We said earlier with reference to the function e* that
it is regular in the whole plane, but in saying this we excluded the
point at infinity. The expansion of the function ez holds everywhere
and, in particular, in the neighbourhood of the point at infinity.
It contains an infinite number of terms in positive powers of z
and, consequently, the point at infinity is an essential singularity
of ez. The same can be said, for example, about sin z and cos z.
2. Every polynomial is a regular function in the whole plane and
evidently has a pole at infinity, the order of which is equal to the
order of the polynomial.
Consider the following rational function which is a quotient of two
polynomials:
y(z)
V(z)
= /(*),
where the fraction cannot be simplified,i.e. the zeros of the numerator
and the denominator are different. Our function will have singularities
at a finite distance these being the zeros of the polynomial y>(z), and
these points will be the poles of the function. The behaviour of the
function at the point at infinity will depend on the degrees of the
polynomials in the numerator and the denominator. If the degree
of <p(z) is higher than the degree of yi(z) by m, then f(z) will tend
to infinity when z->- but the ratio f(z)fzm will tend to a finite
non-zero limit, i.e. our function will have a pole of order m at infinity.
However, if the degree power of <p(z) is higher than the degree of ip(z)
the function will be regular at infinity.

18. Analytic continuation. If a function/(z) is regular in a domain B,


the question arises as to whether is it permissible to extend the domain
in which the function is defined, i.e. is it permissible to create a larger
domain C which contains B within itself and to define the regular
function F(z) in this larger domain so that it coincides with f(z)
in the original domain B. This extension of the domain in which the
regular function is defined, or, as it is sometimes said, the extra­
polation of the regular function, is known as analytic continuation
of the function. It happens that, if analytic continuation is pos­
sible, it will be fully defined and unique. In this respect regular
functions of a complex variable differ fundamentally from say continu­
ous functions of a real variable. For suppose we are given a continuous
64 THE BASIS OF THE THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [18

function co{x) of the real variable x in the interval a < x < b. We can
evidently continue the graph of this function outside the given interval
in an infinite number of ways without affecting its continuity. How­
ever, in the case of a regular function f(z) of a complex variable the
values in the original domain B will fully determine the values of the
function outside this domain provided such extension of the domain,
i.e. analytic continuation, is at all possible. It only needs to be remar­
ked that in the course of analytic continuation one can arrive at many­
valued functions. In this section all the circumstances which can
arise during analytic continuation are explained and the proof of the
uniqueness of this continuation is given.
To begin with let us explain some properties of regular functions.
Let the point z = b be a zero of the regular function f{z). In this
case the constant term will be absent in the Taylor’s series with the
centre at b and, perhaps, some of the succeeding terms will also be
missing. Assume that the first non-zero term is of the order (z — b)m, i.e.

/(*) = + 0«+ i (* - &)m+1 + • • • («m * 0) (92)


or
f{z) = { z - b)m [am -f am+1 (z - b) + . . . ]. (93)

In this case z = b is known as a zero of multiplicity m. Let us


turn to the formula (93) and assume that z is equal to a number
which is close to, but different from, b. In this case the factor (z — b)m
will be non-zero, and the value of the sum inside the square brackets
will be close to am, which is also non-zero, i.e. the sum is non-zero.
In other words, at all points sufficiently close to the zero of a regular
function, the function must be non-zero. Hence the zeros of a regular
function are isolated points. We naturally assumed in the above argu­
ments that the expansion into Taylor’s series (92) contains at least
one term which is non: zero. Otherwise, we must evidently regard the
function as identically zero, at least in the circle in which Taylor’s
expansion holds. Keeping this in mind we can now proceed to prove
the theorem which is of fundamental importance in the problem of
the uniqueness of analytic continuation.
T h e o re m . I f f(z) is regular in a domain B and vanishes in a domain
/?, which is part of B, then f(z) is identically zero in the whole domain B.
We shall use reductio ad absurdum. Assume that f(z) does not
vanish at the point c in the domain B. Let us take a point b in /? and
connect this point with the point c by the curve I, which belongs
18]
ANALYTIC CONTINUATION 65

to the domain B. On a section of this curve, adjacent to the point b,


0ur function will vanish and on another section, adjacent to the
point c, it will be non-zero. Therefore there must be a point d on the
curve I, such that our function is zero along the whole section bd;
whilst there are points on dc as close as desired to d at which the
function is non-zero. A regular function is at the same time contin­
uous and, consequently, there must be a zero at the point d itself.
However this zero is not isolated since the whole arc bd of the
curve I consists of zeros of our function. It follows from earlier consi­
derations that the expansion of our function into a Taylor’s series
with centre at d must be identically zero, and consequently, our
function must vanish in a circle with centre at d, i.e. it must also
vanish on a section of the curve adjacent to the point d which is
part of the section dc. This circumstance contradicts the property of
the point d, that on the section dc there are points which can be as
near as we please to d, where f(z) does not vanish. The theorem is
thus proved.
Note. In the proof of the above theorem the conditions could
have been restricted to the fact that f(z) vanishes on a curve in B.
In this case the function would have vanished in a circle with the
centre at one of the points of the curve.
It is even sufficient to assume that the zeros of f{z) have a point
of accumulation in B, i.e. there is a point b in B such that in a circle,
centre at b and radius, as small as we please, f(z) has an infinite
number of zeros. By virtue of our earlier considerations, Taylor’s
series for f(z) with centre at b would, in this case, be identically
zero, i.e. f(z) would vanish in a circle, centre at b, i.e. would vanish
everywhere in B.
C o r o l l a r y . Let the two functions f^z) and / 2(z) be regular in B
and coincide in a part /? of this domain or on a curve. Their differ­
ence must be equal to zero in /3 and hence, by the above theorem, it
must be equal to zero in the whole domain, i.e. if two functions,
which are regular in a domain, coincide in part of that domain (or on
a curve) then they coincide in the whole domain.
Suppose that all the values of our two functions and the values of all
their derivatives coincide at a point b in the domain B. In this case
their expansions into Taylor’s series with the centre at b will also
coincide, i.e. our functions will coincide in a circle, centre at 6 , and
therefore they will coincide in the whole domain; hence if the values
of two functions and of all their derivatives are the same at a given point
66 TH E BASIS OP TH E THEORY OF FUNCTIONS OF A CO UPLES VARIABLE [18

of a domain, the values of the functions will be the same throughout the
domain in which the functions are regular.
Let us now turn to the problem of analytic continuation. Let
/x(s) be regular in the domain By and suppose that we have succeeded in
constructing a new domain B2 which has the part By 2 in common
with the domain By (Fig. 11 ), and that we have defined the regular
function f2(z) in the domain B, which coincides with f y ( z ) in B1 2. We
may term f2(z) the direct analytic continuation of fy(z) from By into B2
via By 2. The function defined as fy(z) in B v and as f2(z) in B2, gives a
unique regular function throughout the extended domain. We will show
that there cannot be two dif­
ferent analytic continuations. In
fact, assume that we have two
different analytic continuations
of f2(z) from Bl into B, via By 2.
These two functions f£\z) and
/ 22>( z ) which are regular in B2must
coincide with fy(z), and consequ­
ently, they must coincide with
each other in B1 2. But now, by what has been proved above, they
must coincide in the whole domain B2, i.e. they give the same ana­
lytic continuation.
Suppose now that we have a chain of domains Bv B2, B3, . . . where
By and B2 have the part By 2, in common, B2 and B3 have the part
B2 3 in common etc. In the domain B2 we have a regular function
f2(z) which coincides with fy(z) in By 2. In the domain B3 we have a
regular function f3(z) which coincides with f2{z) in B2 3, etc. Here we
have the analytic continuation of f x(z) via a chain of domain and this
analytic continuation is unique. Note that, generally speaking, the
overlapping of the domains Bs need not be confined to the parts B k k+l
which we have mentioned above. Let us consider, for example, a chain of
domains consisting of three domains By, B2and B3, and assume that f ?3
overlaps with By (Fig. 12 ). In this common part, which is shaded in the
diagram, the values of fy(z) as defined in By and the values of f3(z) as
defined in B3, can be different. Here we obtain a many-valued
function as a result of the analytic continuation. However we can
avoid this geometrically, viz. if the values of fy(z) and f3(z) are different
in the shaded part then we assume that this shaded part consists
of two sheets, as it were, one of which belongs to By and the other
to B3.
18] ANALYTIC CONTINUATION 67

The problem of many-valuedness can arise even during the


first step of the analytic continuation. Suppose we have the
analytic continuation of fx(z) from B1 into B2 via Bx 2 (Fig. 13),
but that B1and B2have also a part /9in common. In /3the values of f2(z)
may or may not coincide with the values of fx(z). The set of all values,
obtained as a result of all the possible analytic continuations of

the initial function fx(z), yields a unique function which we shall call
an analytic function and which we denote by f(z). As we have said
already, this function f{z) may prove to be many-valued.
Instead of analytic continuation via a chain of domains one
frequently speaks about analytic continuation along a curve. Consider
a line I divided into successive sections: P i Q y P 2Q 2, • • • > P n Qn>
so that the sections P kQ k and P k + i Q k + i have the part P k+1 Q k in
common (Fig. 14). Assume that this curve I is covered by a chain of
domains B v B , .............. B k , . . . , so that the section P k Q k lies in B k .
Denote by B k , fc+1 the domain in which B k and B k+1 overlap and
which contains the section P k + i Q k of the line I . (There can be sever­
al or even an infinite number of domains in which B k and P k + i
overlap, but we only take the one which contains P k+1 Q k .)
Suppose there exists a regular function fx(z) in B x and that this
function can be continued with the aid of the chain of domains
B v B 2, . . . , B k , . . . , B n via B x< 2, B 2 3 .............. B n ~ l m . Instead of this, we
say that f(z) can be continued along the line I. The values of the func­
tion on the section P x Q x (and in the neighbourhood of this section) are
given and by applying the fundamental theorem of this paragraph we
can see, as before, that there can be only one analytic continuation
68 THE BASIS OF THE THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [18

along I. It does not depend on the way in which we divide I into sections
and cover it by domains possessing the above properties.
Let us return to the analytic continuation along I via a definite chain
of domains B k. In the neighbourhood of every point on the line I the
analytic function f(z) will have definite representation as a Taylor’s se­
ries. We shall call this series the function-element at the corresponding
point on the line I. If this line I is slightly deformed while its ends Px and

Qn remain fixed, it will remain in the domains B k and the function-ele­


ment of f(z) at the point Qn will be the same, as before. This shows
clearly that, in general, if we deform this curve continuously whilst
keeping its ends Pl and Qn fixed, and if the analytic continuation
of the initial element at the point Px along the line is possible in any
of its positions, the element obtained at the point Qn as a result of the
analytic continuation will always be the same.
Suppose that, on analytic continuation along a curve I from the point
Pu we can only proceed as far as a point C, beyond which analytic
continuation along this curve is no longer possible. In this case the
point C is known as a singularity of the function. But note an
important circumstance, viz. had we performed the analytic conti­
nuation from the point P± to the point C along another curve
instead of the curve I, the point G might not have been a singu­
larity, i.e. generally speaking, a singularity is defined not only by
its position in the plane but also by the path by which it is reached
18] ANALYTIC) CONTINUATION 69

in the course of the analytic continuation (c.f. the example in [19]).


In future we shall almost always be concerned with the simpler case,
when the position of the singularities can be fixed in advance and
they do not depend on the path of the analytic continuation.
A theorem of great importance in the theory of analytic conti­
nuation follows at once from the above. It is known as the uniqueness
theorem. I f the analytic continuation of an initial function-element is
possible along any path in a simply connected domain B, these
analytic continuations along paths in B yield a unique function
in B.
In fact, let the initial function-element be defined in the neigh­
bourhood of a point Px and let us take two different directions l2
and l2 for the analytic continuation from Pt to Qn. Owing to the fact
that the domain is connected we can, by means of a continuous
transformation, transform the contour into the contour l2 without
leaving the domain B and, by hypothesis such analytic continuation
along the contour will always be possible. But, as we have said
already, the final result of the analytic continuation at the point
Qn will be the same, i.e. in fact the different paths selected for the
analytic continuation give the same final result and we obtain a
unique function f(z).
In the above arguments we have sometimes restricted ourselves
to the simple outlines without going into the details of the proof
since this would require much more space. We hope, however, that
the reader formed a picture the basic ideas about analytic continua­
tion. We must emphasize that all that has been said above is only
theoretical in character and does not give any indication as to the
practical methods of analytic continuation.
Let us now mention another principle in the theory of functions
which is closely connected with analytic continuation and which is
known as the principle of permanency. Assume that the initial function-
element f y(z) satisfies a certain equation, e.g. the differential equation
of the second order:

Po(«) + Pi(z) + Pi (2) /(2) = o. (94)

the coefficients pk(z) of which are given polynomials in z. On


analytical continuation of f^(z), the derivatives f{(z) and fl(z),
and also the whole of the left-hand side of our equation are analy-
70 T H E BASIS OP TH E THEORY OP FUNCTIONS OP A COMPLEX VARIABLE [18

tically continued. Consequently, if the left-hand side vanished in


the initial domain, it would also vanish on analytic continuation,
in other words, if the initial analytic function-element satisfies the
equation (94) the equation will be satisfied by the analytic function,
obtained from the initial element as a result of the analytic con­
tinuation.
Let us now turn to a definite method of analytic continuation.
Here we shall only use circular domains and Taylor’s expansion in such

domains (Fig. 15). Let the initial function-element be given in the


form of a Taylor series, centre at 61:

/i(* )= j ? (95)
fc=o

Let us draw a contour I from the point bx and perform the analytic
continuation of our function along this contour. We proceed as fol­
lows: we take a point b2 on the curve I such that the arc b2 lies in
the circle K l which is the circle of convergence of the series (95). By
using this series we can evaluate the derivatives f[n\b 2) and write the
expansion of our function, centre at b2:
w « /(ft)
h (*) = 2 , w (2 - h*)k = 2 - n t H - (z - b*)k ■ (96)
fc= 0 k=0

This new function will be defined in a circle K 2, centre at b2. If this


circle lies outside the circle K lt the function (96) will provide an
analytic continuation of fx(z). At the point b2 the values of the
functions f^z) and f2(z) and the values of all their derivatives coincide,
and the functions will be the same in the overlapping portion of the
18] ANALYTIC CONTINUATION 71

two circles. Notice that the series (96) can be obtained from the
series (95) as follows: we rewrite the series (95) in the form:

2 ’o i» [ ( z - 6 a) + (6, - 6 1)]fc. (97)

On expanding | (z — b2) + (b2 — ij) |k by the binomial formula and


collecting terms in like powers in the sum (97) of (z — b2), we
obtain the series (96).
Having performed the first analytic continuation we proceed to
the next. We choose on the curve I a new point b3 such that the
arcb2b3 belongs to the circle K 2. The series (96) can be rearranged,
as shown above, in powers of (z — b3), when a new function-element
is obtained:

fa (2 ) = 2 aW - b^ ’
k= 0

which is defined in a circle K3, centre at b3, etc. As a simple example,


consider the series
7— 7 = 1 + z + z 2 + ... (98)

This series is convergent and defines a function which is regular


only in the circle | z |< 1 . But its sum 1/(1 —2) is a regular function
in the whole plane except at the point 2 = 1 and, consequently,
we can continue the series (98) in the whole plane. If we take a point b2
in the circle | 2 | < 1 and reconstruct the series (98) in powers of
(2 — b2), we obtain a new series of the form:

1
(z — b2)k .
2
k=0
(1 - bt)k+1

This series converges in a circle, centre at b2 and radius equal to


the distance from this point to the point 2 = 1 . If the point b2 does
not lie on the segment (0 , 1 ) of the real axis, this new circle will
lie outside the old circle and we obtain an analytic continuation which
can be continued further. In practice, in a case of this kind, it is
evidently unnecessary to use analytic continuation of the series (98),
— it is natural to use the finite form 1/(1—z). However, if the
function is only given in the form of a power series and no other ex­
pression is known for it only the process of analytic continuation
72 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [19
remains open. Many attempts have been made in this field to find
an easier practical way of performing analytic continuation. Later
we shall give one practical method for particular case. For the present
we take the analytic continuation of elementary many-valued func­
tions as an example.

19. Examples of many-valued functions. Consider the function


z = w2 (99)
and suppose that the variable w varies in the upper half-plane i.e.
in that part of the plane in which the coefficient of the imaginary
part is positive (above the real axis), so that arg w varies from 0
to 71. On squaring, the modulus \w\ is squared, and the amplitude
is multiplied by two i.e. the values of z will fill the whole plane, ahd
both the positive and negative parts of the real axis in the w-plane
will be transformed into the positive part of the real axis in the
2-plane. We can thus see that, as a result of the transformation (99),
the upper half of the w-plane is transformed into the whole z-plane with
a cut along the positive part of the real axis from 0 to + ° ° . Denote
the plane with this cut by T v Conversely, we can regard w as a
single-valued function of z in the domain T2.
w = Yz, ( 100 )
where those values of the radical must be taken which give the positive
coefficient of the imaginary part of ]/z. Real positive values of z lie
both on the upper and the lower edges of our cut. On the upper edge
positive values of the function J/2 must be taken, and on the lower
edge negative values must be taken. The limit of the ratio A w/A z
will evidently be equal to the reciprocal of the ratio A zjA w, i.e.
the usual law for differentiating inverse functions will apply and the
function ( 10 0 ) will be regular in our domain:
d2 0 dw 1
( 101 )
d^T - ZW] I T _ Ip T •
Returning to the beginning of our discussion, we now assume that
w varies in the lower half-plane. On squaring, we evidently obtain for
2 a second copy of the same former region T v Let us denote this
by T 2. In this new region T 2 our function (10 0 ) will again be regular
and single-valued, provided that value of the radical is taken which
gives the negative coefficient of the imaginary part of |/i.
EXAMPLES OP MAXY-VALUED FUNCTIONS 73
19]

It also follows from the above that the values of the function (100)
along the upper edge of the cut in the domain T 1 coincide with the
values of that function along the lower edge of the cut in the domain
f and vice versa.
We can thus see that, by cutting from 0 to + ° ° , we obtain a domain
in which our function ( 100 ) is single-valued, but to obtain all the values
of the function we must regard it as two different functions, which
are defined in the domains T t and T2 respectively as above. Such
a division of the function ( 100 ) into two separate single-valued
functions appears artificial and we shall now combine these two
functions into a single analytic function which is single-valued and
regular in a two-sheeted plane. In order to produce this two-sheeted
plane T, imagine the copy Tx on top of the copy T2 with their edges
joined together cross-wise along the cut, viz. the upper edge of the
cut in is joined to the lower edge of the cut in T2 and vice versa.
We assume that the point z = 0 coincides on both copies. The con­
structed two-sheeted domain T is evidently obtained from the u>-plane
as a result of the transformation (99) and the function (100) will be
regular and single-valued in the entire domain T, except at the point
2 = 0. Note the special importance of this point. If, starting from
a point z0 we draw a closed contour about z = 0 , on returning to
the point z0 we find ourselves on another sheet as compared with
that from which we started to draw our contour. Here, the values
of the function J/z, as defined above on our contour, will evidently
give the analytic continuation of the function along that contour,
and the final value of the function at the point z0 will be of oppo­
site sign as compared with the initial element at that point. The point
2 = 0 has the property that the function |fz is continuous and
has a derivative in the neighbourhood of the point, but on the ana­
lytic continuation round a closed contour about this point it changes
its values. Such a point is known as a branch-point of the function.
In the case under consideration we return to the original values of
the function by describing another circuit about the point z = 0
and such a branch-point is known as a branch-point of the first
order. The domain T evidently represents the total domain of
existence of the function (100). In this case we have been able
to obtain this domain relatively simply, since the function (100 )
is the inverse of the very simple function (99). Figure 16 shows the
appearance of a two-sheeted plane near a branch-point of the
first order.
74 TH E BASIS OP TH E THF.OEY OP FUNCTIONS OF A COMPLEX VARIABLE [ 19

Generally speaking, if the function


z = <p(w) ( 102 )

is single-valued and regular in the whole ic-plane, as a result of the


transformation ( 102 ) this plane may change into the many-sheeted
z-plane and the inverse function, to
( 102 ) :

w = f(z) (103)
will be regular in this many-sheeted plane
and will have the derivative

r w = -? w -
This regularity will only be lost at
points which correspond to values of w
where cp'(w) = 0 . These points correspond
to the branch-points of the inverse func­
tion (103). We will explain this in greater detail in one of the following
paragraphs. The above many-sheeted planes are usually known as
Riemann surfaces (Riemann was a mid-nineteenth century German
mathematician). Consider the following example:

m “ 7 T+T ’ <I04»
The many-valuedness of this function is solely due to the presence
of jR, and, consequently, in the two-sheeted plane T, which we
constructed above for the function (100), the function (104) will also
be single-valued. This function will have a singularity at the point
z = 0 (branch-point) and also at one of the points z = 4. There will
be two of these points (on both sheets and T2). On one of the
sheets ^ 4 = + 2 and on the other J / 4 = —2 . On the latter sheet
the point z = 4 will be a pole of the function (104). Had we not
the two-sheeted plane T, as a result of the analytic continuation of
(104) we would have obtained different values for this function
and the point z = 4 would have been a singularity for those paths
of the analytic continuation for which Jfz is equal to —2 when z = 4.
The function (104) can be regarded as the inverse of the function

z = (2ww
- 1)*
(104J
19] EXAMPLES OF MANY-VALUED FUNCTIONS 75

which is regular in the whole plane except at w = 0 , where it has a


pole of the second order; it can be shown that this function transforms
the w-plane into a two-sheeted T-plane of the type described above.
In the case under consideration the point w = 1/2 will be transformed
into the branch-point 2 = 0 ; the point w = 0 will be transformed
into the point z = °o. The point w — °° gives the point 2 = 4 on
one of the sheets. A point with the same coordinate 2 = 4 on the
other sheet will be obtained when w = 1/4. Note that in the given
two-sheeted plane we must regard not only the point 2 = 0 but also
2 = °o as coinciding on both sheets, i.e. the points 2 = °° and 2 = 0
are both branch-points of the first order. The first of these points can
be obtained from the formula (99) only when w = 0, and from the
formula (104j) only when w = 1/ 2 . The point z = °°, can be obtained
from the formula (99) only when w = and from the formula (104^
only when w = 0 .
Consider a function of the following kind:
w = f(z) = Y(z — a) (z — b). (105)
For this function the points a and b are branch-points. By describ­
ing a circuit round a closed contour encircling one of these points
we change the sign of the expression (105) but by describing simul­
taneously circuits round both points we leave the function unaltered.
In fact, assume:
z — a = Qr e1*1; z — b = g2elVi,
whence
I
r— 2
/(*) = V e i^ e
If we describe a circuit round a closed contour I encircling both points
in the counter-clockwise direction, 2 n will be added to the amplitudes
<py and <p2, the Bum (q>x + <p2) will receive an increment of 4ji and the
amplitude of the expression (105) an increment of 2n, i.e. the value
of the function will be unchanged. In order to make the function (105)
single-valued it is sufficient to make a cut from the point a to the
point b. This cut prevents us, as it were, from making separate
circuits round the points a and b. The function (105) is two-valued
at all points except z = a and b; in order to obtain all the values
of this function we must take two copies of the plane cut in the
way described above. On each of these (105) will be a single-valued
function and the values of this function on different copies will differ
76 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [19

from each other in sign only. If we superimpose one copy on top of


the other and imagine the edges of the cuts to be joined cross-wise,
we obtain a two-sheeted Riemann surface, with branch-points
of the first order at a and b\ on this surface the function (105) will
be single-valued and regular (except at the branch points). The point
at infinity will not be a branch-point and each sheet will have its
own point at infinity. In the neighbourhood of this point at infinity
we can rewrite the function (105) in the form:

/ (z) = ± z ( l - ~ ) 2 (l - } ) 2-

Expanding the differences by the binomial formula, which is


possible since in the neighbourhood of the point at infinity | a/z | and
| bjz | are smaller than unity, we obtain our function in the following
form in the neighbourhood of the point at infinity:
1 a 1 b2 1 • 3 o*
/ (z) = ± z 2 - 4 - 6 z3
X
2 z 2 - 4 z* - )
1 b 1 a2 1 - 3 b*
X
2 z 2 - 4 22 2 - 4 - 6 Z3

i.e. on multiplying the series we can see that the point at infinity is
a pole of the first order on both sheets.
Notice that, on solving the equation (105) with respect to z, we
obtain a many-valued function w, i.e. the function (105) is not the
inverse of a function which is single-valued in the whole plane. The
Riemann surface, on which it is single-valued, will have two branch­
points z — a and z = b of the first order. This Riemann surface can
be obtained by the transformation
bw2 — a
z= —
w2.—
— 1r

of the w-plane and the inverse function of the above:

will have the same Riemann surface as the function (105).


Consider the function
f(z) = y ^ ~ a (10fi)
where n is a positive integer. Every circuit round the point z = a
changes the value of the function, and on performing n such circuits
19] e x a m p l e s o p m a n y -v a l u e d f u n c t i o n s 77
in the same direction, we return to the original value of the function,
i.e. for the function (106) the point a will be a branch-point of the
order n — 1. For, denoting the modulus and amplitude of z — a by g
and <prespectively, we obtain:
* ni
— a = Yqq n.
On describing a circuit round z = a ,n times in the positive direction,
n____
we add 2nn to <pand, consequently, the amplitude of Jfz — a receives
an increment of 2 jt, which does not alter the value of the function.

We shall now consider another many-valued function which is of


special importance in the theory of functions, viz. the logarithm.
This function is obtained as a result of the inversion of an exponen­
tial function
z = e*. (107)
We shall first explain some properties of the exponential function.
It is easy to see that it has a pure imaginary period of 2ni. In fact:
e w+27,i _ eH
<e2nt _ ew(coa 27i -j- i sin 27i) = e“\
We divide the plane w — u + iv into strips, 27i in width, by straight
lines parallel to the real axis. As the fundamental, strip take say the strip
U bounded by the lines v = 0 and v = 2n. We can transform this funda-
78 TH E BASIS OP TH E THEORY OP FUNCTIONS OF A COMPLEX VARIABLE [19
mental strip into another strip by adding 2nni to w, where n is an integer.
The values of the function (107) thereby remain unaltered because
of the periodicity, i.e. the values of the function in every strip will
be the same as in the initial strip. We shall now consider the way
in which the fundamental strip is transformed by the function (107).
Draw in this strip a line AUl, parallel to the imaginary axis with the
abscissa u = u0. Along this line we have:
u = u0 (0 < v < 2n)
and, consequently:
eu' = eu° e,B (0 < v < 2n).
i.e. our line will be transformed into a full circle, centre the origin and
radius e“*, whilst the same point on the circle corresponds to the
ends of the line XUn. If we take the part of the strip U bounded by two
lines, parallel to the axis u = 0 with the abscissae u = ux and u2,
as a result of the transformation (107) we obtain a circular annulus
in the 2-plane, centre the origin and radii e“‘ and e“2 (Fig. 17). Finally,
the entire strip U will be transformed into the 2-plane with the excep­
tion of the origin. The upper and lower edges of the strip will be
transformed into the positive part of the real axis. We make a cut
along this part of the real axis. We can say then that the upper edge
of this cut corresponds to the lower edge of the strip and the lower
to the upper edge of the strip. Denote by Tx the plane with the cut
without the origin. In this domain Tx the function inverse to (107):
w = log 2 (108)
will be single-valued and regular and its derivative can be found by
the usual rule for differentiating inverse functions:
dw 1 1 1
(109)
dz ~ (i*y
We have, as we know:
log 2 = log | 2 1 -f *arg 2 .
On analytical continuation of this function along a contour we must
preserve continuous variation of the amplitude arg z.
In the domain Tx variations of the amplitude is restricted to
0 < arg 2 < 2 jt, and we obtain a single-valued definition for the
function (108). The point 2 = 0 is evidently a branch-point for our
19] EXAMPLES OF MANY-VALUED FUNCTIONS 79

function (108), viz. on analytical continuation of this function


round a closed contour about the origin and on encircling the origin
n times in the counter-clockwise direction, we add 2 nni to the function
(108); every subsequent circuit will give new values of the function,
i.e. in this case the point z = 0 is a branch-point of infinite order.
Let us return to the transformation of the in-plane by the function
(107). Every strip into which we have divided the m-plane gives a
new domain T1 in the z-plane and there will consequently be an in­
finite number of such domains. We superimpose them on top of each
other and number them so that the domain corresponding to the
initial strip is number one, the next one placed on top of it is number
two, etc. while domains corresponding to the lower strips are numbered
0 , (—1). (—2 ), . . . Let us now join in imagination the edges of the
cuts as follows: join the upper cut of T x with the lower cut of T0,
and the lower cut of Tx with the upper cut of T2; join the upper cut
of T 0 with the lower cut of T_i and the lower cut of T 2 with the upper
cut of T3 etc. In this way we obtain a Riemann surface with an in­
finite number of sheets and with branch-points of an infinite order
at z = 0 and z = » . On this Riemann surface T our function (108)
is regular and single-valued. The surface T is obtained from the
w-plane as a result of the transformation (107).
The function w = log (z — a) evidently has branch-points of an
infinite order at z = a and z = °°. Let us also consider the function

w = loS ^ r | = log (* - a) ~ log (z - b). ( 110 )

This function has branch-points at z = a and z = b. If we describe


a circuit in the positive direction round a closed contour which en­
circles both points, both terms in the above expression receive an
increment of 2 ni and the difference remains unchanged, i.e. infinity
is not a branch-point of the function ( 110 ).
We can rewrite our function in the following form:

w = log (l - —) - log (l -
and we can expand both terms according to formula (80) for all values
of z, the moduli of which are greater than | a | and | b |. As a result
we obtain the following form for our function in the neighbourhood
of the point at infinity:
( 111 )
80 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [20
where
ak — k ■
Formula (111) gives one of the branches of our many-valued function
in the neighbourhood of infinity. To obtain the remaining branches
it is sufficient to add 2 nni to the above expression. For each fixed
integer n we obtain another branch of our function.
Consider the function

w = arctan z = 4 ^-log - 2

which has branch-points of infinite order at z = i and z = —i. The


derivative of this function, as in the case of the real variable, is
dw
dz 1 +z2
or
dw
dz (i + z) (i - z)

20. Singularities of analytic functions and Riemann surfaces.


In the preceding paragraphs we dealt with examples of many-valued
functions and constructed Riemann surfaces for these functions, on
which they were single-valued. We shall now consider the general
problem. Owing to lack of
space we shall not go into de­
\
\ tails. To begin with we shall
\
\ explain the concept of an iso­
KI
/ lated singular point in analytic
/
/ / continuation.
Assume that we are given
the initial analytic function-
element f(z) at a point z — a
and that we propose to
continue it along a line I. Suppose that analytic continuation is
possible up to point 2 = 6 but no further, so that the point 2 = 6
is a singularity in the analytic continuation along I [18]. Assume
the existence of a circle K, centre at 2 = 6 , such that the func­
tion-elements f(z), corresponding to points on the section cb of the
line I in K, (Fig. 18) can be analytically continued along any
line in K, which does not pass through the point 2 = 6 . In this
case the point 2 = 6 is known as an isolated singularity of f(z)
20] SINGULARITIES OP ANALYTIC FUNCTIONS AND RIEMANN SURFACES 81

(corresponding to the path I). The analytic continuation along any


line in K can lead to single-valued or many-valued functions in
K. In the first case the single-valued function obtained in K is regu­
lar everywhere in K except at z = b and it can be expanded into a
Laurent’s series in integral powers of (z — b); the point z = 6 is ei­
ther a pole or an essential singularity of our analytic function
f[z) (on analytic continuation along I). In the second case, when
the function in K is many-valued, the point z = b is known as a
branch-point. Suppose that, given all the possible analytic continua­
tions in K, we obtain a finite number of different elements at a
point z = a inside K. Denote this number by m. It is easy to see that
at any other point z = /9 in K, we again obtain m different elements.
This is due to the fact that, on analytic continuation of different initial
elements along the same path from a to or from /? to a, different ele­
ments are obtained at the end-point. In the case under consideration
the point z — b is known as a branch-point of the (m — 1 )th order. If
the number of different elements obtained on analytical continuation
in E is not finite at every point in K then z = b is a branch-point of
infinite order.
Let us consider in greater detail the case of a branch-point of the
finite (m — l)th order. By hypothesis, analytic continuation can be
performed in E , except at the point z = b. A circle E with an isolated
point z = 6 is a doubly-connected domain. Take m copies of the
circle E and cut each copy along the same radius. Such a circle K v
cut along the radius, is a simply connected domain. Let us take the
same point z = a in each copy K v At that point we have m elements of
our analytic function. In each copy we take a definite element at the
point z = a and continue it analytically in K v In accordance with the
uniqueness theorem [18] we obtain a definite single-valued function
in every copy. We shall term one edge of the cut in every copy the left-
hand edge and the other the right-hand edge. For example, we shall
describe as the right-hand edge that edge from the points of which we
can reach the left-hand edge by moving inside K ± and encircling the
point 2 = 6 in the counter-clockwise direction. Take a copy of the circle
Ex with a single-valued function defined in it. Call this copy the first
copy and denote the definite single-valued function b y / 1(z). The values
of the function fx(z) on the left-hand edge coincide with the values
of our function on the right-hand edge of the cut in some other copy
of Kv Call this latter copy of E xthe second copy and denote the function
defined in it by / 2(z). Join in imagination the left-hand edge of the
82 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [20
first copy with the right-hand edge of the second copy. The values
of the function / 2(z) on the left-hand edge of the second copy coincide
with values of our function on the right-hand edge of another copy of
K v Let us call this copy the third copy and denote the function defined
in this copy by f3(z). J oin in imagination the left-hand edge of the second
copy with the right-hand edge of the third copy. Continuing in this way
we eventually reach the last copy m. It can easily be seen that the values
of the function f m(z) on the left-hand edge of the mth copy coincide with
the values of/(z) on the right-hand edge of the first copy. Join in imagi­
nation these two edges. We thus obtain an m-sheeted circle L with the
branchpoint 2 = 6 of the (m— l)th order. This point must coincide on
all copies. In the m-sheeted circle L our function will be single-valued
everywhere except at the point z = b. Replace 2 by a new independent
variable
2' = = Ye e (112 )
where q = \ z — b | and <p = arg (2 — b), where cp is fixed in a definite
way at a point of L. The point 2 = b will be transformed into the point
2 ' = 0. In general, on describing a circuit round the point 2 = 6,
the amplitude changes by 2nm and on describing a circuit round
the point z' = 0 it changes by 2n. In the z'-plane the m-sheeted circle
L will be transformed into a one-sheeted circle C, centre at z' = 0
m_
and radius JIR, where B is the radius of L. In this one-sheeted circle C
our function will be single-valued and regular with the possible ex­
ception of the point z' = 0 . Consequently it is possible to expand it
in O into a Laurent’s series:
f{z)=
2 anz’n
+
n= —®
or, returning to the former variable:

f(z)= 2 an( y ^ r s y = 2 an ( z - b ) ^ , (113)


n = —» n=» —oe
i.e. in the neighbourhood of the branch-point of the (m — l)th order
our function can be expanded in integral powers of the argument (112 ).
The value of the argument (11 2 ) at a point z, which lies in the neigh­
bourhood of the point z = b, can be fixed in an arbitrary yet definite
manner. The expansion (113) can have different forms. It can happen
that the expansion contains no terms with negative values of n:

/ (z) = a0 + Oi yg — b + a2(jh — b f + . . .
20] SINGULARITIES OF ANALYTIC FUNCTIONS AND RIEMANN SURFACES 83
jt is evident here that f(z)^-a0when z->b, while 2 may tend to b in any
manner as long as it remains in L. In the case under consideration
we assume that f(b) = a0 and call the point z = b a branch-point of the
regular type. If the expansion (113) contains only a finite number of
terms with negative values of n, then f(z) -> °° when 2 b. In this
case we put f(b) = 00 and call the point 2 = 6 0 branch-point of the
polar type. If the expansion (113) contains an infinite number of
terms with negative values of n, the
point z = b is a branch-point of the
essential singularity type.
All these definitions can be extended
to the point at infinity. Suppose that
f(z) is continued analytically along the
contour I and that a neighbourhood
K( 12 1> -B) of the point at infinity (Fig.
19) exists, such that the function-ele­
ments f(z), corresponding to points on
the arc I in K, can be analytically con­
tinued along any path inside K. If
this analytic continuation gives a
single-valued function then the point
2 = co will be either a regular point of f(z), or a pole or an essential
singularity [ 1 0 ]. If the analytic continuation produces a many-valued
function then 2 = 00 is a branch-point. If this branch-point is of the
finite (m — l)th order then in its neighbourhood the expansion given
below applies:
+“ / l \n + «= _5
f(z)= 2 “» h r = 2 m-
n= — 00 ( |/g ) n= —a

where everything that was said above regarding such an expansion


again applies. The character of the point 2 = °° may, of course,
depend on the path I of the analytic continuation, by which we
reached the neighbourhood of the point at infinity.
We shall now explain in general terms the concept of a Riemann sur­
face for a given many-valued analytic function f(z). Suppose that, on
analytic continuation of the initial element, we have reached a point
2 = a. At this point we are given a certain element, i.e. a series
expanded in positive integral powers of (2 — a). This series can be
rearranged in positive integral powers of (2 — /?), where is
any point in the neighbourhood of 2 = a, i.e. the element at the
84 TH E BASIS OP TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [21

point z = a also gives the elements at all points which are sufficient­
ly close to a. To every such element we ascribe a point z which acts
as the centre of the corresponding circle of convergence of the power
series (element). To the given element with the centre at z = a we
ascribe the point a. To elements obtained from it at the neighbouring
points z = /?, we ascribe points z = /?, which belong to the neighbour­
hood of z = a, i.e. which lie on the same sheet as the point z = a.
In the course of analytic continuation we keep on obtaining further ele­
ments and, consequently, further points z on the Riemann surface. If,
on returning to the point z = a, we obtain the function-element
which we had earlier, we identify it with the former point z = a.
If, however, this element proves to be different from the former, then
we regard the new point z = a as being different from the former point
z = a (we consider it to be situated on a different sheet) i.e. on
analytic continuation we regard two points z, which have the same
complex coordinates, as different if we have different elements of
our analytic function at those points. In this way we construct a
Riemann surface during the analytic continuation, which corresponds
to the given analytic function f{z). On this surface f(z) is single-valued
and regular. A Riemann surface usually includes the poles of f(z) as
well as the finite order, branch-points of the regular and polar types.
Note that, in general, the Riemann surface cannot be obtained from
the w-plane as a result of the transformation z = <p(w), where q>(w)
is single-valued and regular in the tu-plane (with possible poles), as
could be done in the simpler cases in [19].
Above we have only considered isolated singularities. It can happen
that these points completely fill certain lines in the course of analytic
continuation. For example, it can happen that the initial element, given
by a power series, cannot be continued in any direction, i.e. that any
point of the circumference of the circle of convergence of this power
series is a singularity. The series given below can serve as an example
of a series which cannot be continued:

zn' = z + z1'2 + z1'2'3 + z1'2'3'4 + . . . = Z + Z2 + Za + Z24 + ...


n =1

21. The theorem of residues. Let us now return to the expansion


of a function into a Laurent’s series in the neighbourhood of a singula­
rity (a pole or an essential singularity). In this expansion we separated
the coefficient of (z — 6 ) -1 and called it the residue of the function
at the given singularity. We shall now explain the significance of this
TH E THEOREM OF RESIDUES 85
21]

coefficient. Thus, let us assume that in the neighbourhood of the point


b the following expansion holds:

/(*)= 2
f t — — eo

Integrate this formula round a small closed contour l0, which surrounds
the point b, and on which the given expansion converges uniformly:

J / (z) dz = a* J (2 - b)k dz.


I* k = — oo U

As we saw earlier [6], all the integrals on the right-hand side are equal
to zero, except one, which corresponds to k = — 1; this integral is equal
to 2ni, i.e. we have:

[ / (z) dz = a_x• 2ni.


h
Let us now consider the more general case. Assume that /(z) is
regular in a closed domain B with the contour I, save for a finite number
of points bv .. .,bm in that domain, which are poles or essential singu­
larities of the function. Denote by (s = 1, . . . , m) the residues
at these singularities. Isolate each singularity by a small closed con­
tour ls. In accordance with Cauchy’s theorem we can write:

J/(z ) dz = ^
l s= 1 /,
But, as we saw above, the value of every integral round each contour
ls is equal to • 2ni and, consequently, the above equation ex­
presses the value of the integral round the contour of the domain in
terms of the residues of the function at the singularities in the domain:

[*/ (z) dz = 2ni ^ a^ . (114)


■/ s=l
T h eo rem o f R e s id u e s . I f a function is regular in a closed domain
save for a finite number of points (poles or essential singularities) then
the value of the integral of this function round the contour of the domain
is equal to the product of 2ni and the sum of residues at the above singu­
larities.
In future we shall find countless applications for this theorem.
At the moment, however, we shall only establish some theoretical
consequences which are necessary for our further treatment. To begin
86 THE BASIS OP THE THEORY OP FUNCTIONS OF A COMPLEX VARIABLE [21

with we shall give the practical rules for calculating residues without
using the expansion of the function into a Laurent’s series.
As the first example consider the following function:

= (U 5)

where ip(z) and xp(z) are regular at the point b and \p{b) = 0 , so that
the function (115), generally speaking, has a pole at the point b.
Suppose, furthermore, that the point z = b is a simple zero of f(z),
i.e. the expansion of the function ip(z) into a Taylor’s series begins
with a first degree term:
V (z) = cx (z — b) + c2 (z - 6)2 + ... (cx # 0).
In this case the function (115) has a simple pole (of multiplicity'one)
in the neighbourhood of z = b:

<P(b) + ( z - b ) + ...
1 (Z) = (z - b) [c, + cs (z - b) + ... ] •
It follows from the last formula that we can write for the residue o_x:
<p{b)
a- i = / (z) (z — b) |2_ 6
or, taking into account that cxis equal to y>'(b):
t p( b)
a. = (116)
V' (b)
As a second example, consider the case when the function f(z)
has a pole of an arbitrary order m at the point b:

f(z) = 2 M z - f c ) fc.
k = —m

The product /(z) (z — b)m is a regular function at the point b, and the
coefficient a_x is the coefficient of (z —b)m~x in this product, whence
recalling the expression for the coefficients in Taylor’s series, we have
the following formula for the residue of our function:

a_, = ^ [ /( z ) (z-br] (117)


(m - 1) 1 d21 z^b
Let us consider one more example. Suppose that /(z) has a zero of
the mth order at the point b, i.e. Taylor’s series with centre at b begins
22]
THEOREM ON TH E NUMBER OP ZEROS 87
with a term containing (z — b)m. In this case our function has the
following form in the neighbourhood of the point 6 :
f(z) = ( z - b r <p ( z ) (<p(b)*0), (118)
where <p(z) is regular at b and is not equal to zero. Let us construct
the logarithmic derivative of our function:
/'(z) _ m , <p' (z)
f{s) z —b ?(z) (119)

It can easily be seen that the point 6 is a simple pole of the logarithmic
derivative, with a residue equal to the order of the zero of the function
f{z). If, instead of having a zero, our function has a pole of multiplicity
m at the point b, formula (118) still holds, except that to must be
replaced by (—to); also, all the subsequent working is the same, i.e. if
at a given point the function has a pole of multiplicity n that its loga­
rithmic derivative has a simple pole with a residue (—n) at this point.

22. Theorem on the number of zeros. Assume that f(z) is regular


in the closed domain B with the contour I and that it does not vanish
on the contour. Assume that it has zeros bt........bm in the domain
of orders kv . . . , km. Its logarithmic derivative has simple poles at
these points bs with residues ks. From the theorem of residues we
have:

2^ JT ^ “ d2 _ + + ‘" +km’ (12°)


/
If every multiple zero is counted as many times as there are units in
its multiplicity, the number on the right-hand side gives the number
of zeros of our function in the domain, i.e. given our assumptions
with regard to the function f(z), the integral on the left-hand side gives
the number of zeros of the function included within the contour I.
The integrand here evidently has the primitive log f(z) and we
obtain the value of the integral by determining the increment that
this primitive receives on describing a circuit round the contour I.
Here we have to consider the single-valued branch of the function,
which means, as we know already, that when describing a circuit
round the contour I we must preserve a continuous change of arg
/(z); here,
log / (Z) = log If (2) I + i arg f ( z ) .
88 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [22

After the circuit is completed log | /(z) | returns to its former value
and receives no increment; consequently, the total increment received
by our primitive is equal to the product of i and the increment of arg
f{z). In accordance with formula (120), we must also divide the incre­
ment of the primitive function by 2ni, when we finally obtain the
following result:
C a u c h y ’s T h e o re m . I f the function f(z) is regular in the closed domain
B and does not vanish on the contour of this domain, the number of
zeros of the function in the domain is equal to the change in the amplitude
of the function on describing the contour, divided by 2n, or, in other words,
it is equal to the change in the amplitude expressed as parts of 2n.
The above theorem is evident in the case of polynomials. Take, for
example, a polynomial of the third degree and represent it as a
product of first degree factors:
c 0 + axz + a2z2 + O3Z3 = a 3 (z — 6j) (z — b2) (z - b3).
Suppose that the zeros bl and b2 lie within the contour I and that the
zero 63 lies outside the contour I. Every difference (z — bk) corresponds
to a vector drawn from bk to z. When the point z describes a circuit
round the contour I the amplitudes of the vectors (z — 61) and (z — &2)
evidently receive an increment of 2n, but the amplitudes of (z — b3)
remains unaltered. Hence the total increment of the function is equal
to 4?r (the amplitude of a product is equal to the sum of the amplitudes
of the factors) or, expressed as parts of 2 n, this increment is equal to
2 , i.e. it is equal to the number of zeros inside I.
Let us establish a further theorem on the number of zeros of a
regular function; this is a direct result of Cauchy’s theorem. Assume,
as before, that f(z) is regular in a closed domain and does not vanish
on the contour. Assume that we have another function <p{z), which is
regular in a closed domain, and the modulus of which on the contour
I is less than that of /(z), i.e.
\<P(z) I < |/(2) I on I. (121)
Notice that, when this condition applies, /(z) evidently cannot
vanish on I. Consider the two functions:
/ (z) and / (z) -f- tp (z). (122)
Both functions satisfy the conditions of Cauchy’s theorem. We have
shown this to be so for the first function and the second function
cannot vanish on the contour because of the condition (121 ). We shall
22 ] THEOREM ON TH E NUMBER OP ZEROS 89
show that the second function has the same number of zeros within
the contour as the first function. For this purpose consider the ampli­
tude of this function on the contour, remembering that on the contour
f(z) * 0 : ^
arg [/(z) + <p (2 )] = a rg /( 2 ) + arg[l + j ^ ] .

To prove our assertion it is sufficient to show that, on completing a


circuit round the contour I, the change in the amplitude

arg[ 1 + /l$ ] (123)


is equal to zero. In accordance with the condition (121 ), the modulus
of the fraction <p(z)/f(z) is less than unity and, consequently, on
describing a circuit round the contour I the variable point

will always lie inside a circle C, centre z’ = 1 and unit radius.


This variable point describes a closed curve which lies within the
circle C and which evidently does not encircle the origin. We can
thus see that the variation in the amplitude of the expression (123)
is in fact equal to zero.
R o u c h e ’s T h e o re m . I f f(z) is regular in a closed domain with a contour
I and <p(z) is also regular in a closed domain and satisfies on the contour
I the condition (121 ) then the functions f(z) and f(z) + <p(z) have the same
number of zeros within the domain.
Note that from Rouche’s theorem follows the basic theorem of algebra,
viz. any polynomial of the nth degree:
a0 + a1z + . . . + a „ 2n K # 0) (124)
has exactly n zeros in the plane. For suppose we put here f(z) = an zn
and <p(z) = a0 -f- ax z + . . . -+- a n-1 zn~1. On any circle, centre the
origin and a sufficiently large radius, we evidently have | <p(z) | <
If(z) |, since the degree of the polynomial <p(z) is lower than the
degree of the polynomial f(z). In accordance with Rouche’s theorem, the
polynomial (124) has the same number of zeros inside the circle as
the polynomial f(z) = a„ zn, and the latter polynomial has a zero of
multiplicity n at the origin.
We shall also point out a consequence of Cauchy’s theorem
which is of great importance in the theory of conformal transformation.
Suppose that the function
w = f{z) (125)
90 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [22

is regular in a closed domain and, when the point z describes a circuit


round the contour I, the point w describes a simple closed circuit
round the contour lv which does not cut itself (Fig. 20). We shall
show that the function (125) transforms in this case the initial region
B into the domain Bv bounded by the contour lv Take a point w1
inside the contour and a point
w2 outside the contour lv We have
to prove that the function
F 1( z ) = f { z ) - w l
has one zero in the domain B
and that the function
F2( z ) = f ( z ) - w 2 •
has no zeros. When the point z
describes a circuit round the con­
tour I the difference f(z) —u\ = w —w1will correspond to a vector drawn
from the point wx to the variable point w on the contour l2. Two
cases are conceivable: the point w can describe the circuit round the con­
tour lx in the counter-clockwise direction or the circuit can be described
in the clockwise direction, assuming that the point z describes the cir­
cuit in the positive direction, i.e. in the counter-clockwise direction. In
the first case the variation in the amplitude of the function F2{z)
is evidently equal to 2 .i, and consequently, this function has one
zero within I. In the second case we obtain a negative number ( —27i)
as the variation in the amplitude of the function F\(z) and it ap­
pears that the function F2(z) has minus one zero in the domain which
does not make sense, since the number of zeros must either be equal
to zero or be a positive integer. Hence the second case is impossible,
and when the point z describes a circuit round the contour I in the
positive sense then the corresponding point w must describe a circuit
round l2 also in the positive sense. Let us now turn to the function
F2(z). The amplitude of the corresponding vector from w2 to the
variable point w on the contour receives no increment on describing
a circuit round the contour Zx and, consequently, the function F2(z)
has no zeros within I. We thus arrive at the following theorem: if
f(z) is regular in a closed domain B with a contour I and transforms
I into a simple closed contour lx which does not cut itself, then on de­
scribing a circuit in the positive sense round the contour I, a circuit
in the positive sense is also described round the contour llt and the
22] THEOREM ON TH E NUMBER OF ZEROS 91
function f(z) transforms the domain B into the part of the plane
bounded by the contour I.
We obtained Cauchy’s theorem by considering the integral on the
left-hand side of formula (120) and assuming that f(z) is regular
in a closed domain and does not vanish on the contour. Let us now
assume that f(z) has a finite number of poles in the domain but that
it is otherwise regular and that it is also regular on the contour, on
which it does not vanish. In this case, as we have seen, the integrand
has simple poles in the domain at the zeros of the function f(z), with
a residue equal to the order of the zero, and at the poles of the
function f(z) with a residue equal to minus the order of the pole.
Applying the fundamental theorem of residues to this integral, we
obtain in the case under consideration the following formula instead
of the formula (120):
i r /' (2) j
- 2 r i j - f k dZ==m~ n ’ (126>

where m is the total number of zeros and n the total number of


poles of our function in the domain. Assume that these zeros are situated
at the points bv . . bm, and the poles at the points cv c2........ c„,
where multiple poles and zeros are counted several times. It is easy
to prove the following formula by using the fundamental theorem
of residues:

Jz / fj- d2 = (&i + &2 + • • ■ + bm) — (cx + c2 + . . . + cn), (127)


i
i.e. the integral on the left-hand side expresses the difference between
the sum of the coordinates of the zeros and the sum of the coordinates
of the poles. In fact, in the case of a zero b of multiplicity k we have
the following expansion in the neighbourhood of that point:

2J W = + (z ~ h ~ r + «o + «i (2 - &) + • • •].
whence it follows that the residue at this point is equal to kb.
The same argument also applies fora pole.
In conclusion we shall make an addition to the above theorem
about the conformal transformation of one domain into another domain.
It is given that f(z) has one simple pole in the domain B, i.e. in formula
(126) n = 1 , and that f(z) transforms the contour I into a simple closed
contour which does not cut itself, but when describing a circuit
92 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [2 3

round I in the positive sense a circuit in the negative sense will be


described round the contour Let us consider once again the functions
F2(z) and F2(z). They both have the same simple pole in the domain
as the function f(z). For the first of these functions the variation in the
amplitude, expressed as parts of 2n, is equal to minus unity but,
on the other hand, in accordance with formula (126), this change in
the amplitude must give the difference between the number of zeros
and the number of poles; thus by hypothesis, the function has one
pole. It therefore follows that the function F2(z) has no zeros. Con­
versely, the variation in the amplitude of the function F2{z) when
the point z describes a circuit round I is equal to zero i.e. the difference
between the number of zeros and poles is, for this function, equal
to zero. But this function has one pole and, consequently, it^must
also have one zero. Thus in the case under consideration the func­
tion f(z) transforms the part of the plane inside the contour I into
the part of the plane outside the contour lv while the pole is trans­
formed into the point at infinity.

23. The inversion of a power series. We shall now apply Rouche’s


theorem to the investigation of a function which is the inverse of a power
series:
w = o0 + a2 (z — b) + a2 (z — b)2 + . . . = F{z). (128)
To begin with we assume that the coefficient ax is not equal to zero,
i.e. F'(b) 0 . For values of z which are close to b we obtain values
of w which are close to a0. We shall show that in the case under con­
sideration a neighbourhood of the point b will be transformed into a
one-sheeted neighbourhood of the point a0 which contains this point.
It will then follow that the inverse function to (128) will be single­
valued and regular in the neighbourhood of the point a0 and there­
fore can be expanded into a Taylor’s series in powers of (w — a 0).
The function
/ (z) = ax (z — b) + a2{z — b)2 + . . .
has a simple zero at the point b and it will certainly not vanish in the
neighbourhood of this point [18]. Let if be a circle, centre at b, in
which the function f(z) is regular and where it has a single zero z = b.
On the circumference C of this circle | /(z) | does not vanish; a positive
uumber m exists such that on the circumference | f(z) | > m. Assume
further that K x is a circle in the m-plane, centre at a0 and radius o,
which is smaller than m. Take a fixed point w0 in this circle.
TH E INVERSION OP A POW ER SERIES 93
23]

We consequently have | a 0 — w0 | < q < to, i.e. on the circum­


ference C of the circle E we have I a0 — w0 | < | f(z) | since | f(z) | > m
on C. In accordance with Rouche’s theorem the function
a 0 - wo + / ( z) = a0 + f ( z ) — wo = F (z) - w0
has the same number of zeros in the circle K as the function f(z), i.e.
one zero. In other words, the values of w = F(z) provide a one-
sheeted covering of the circle K x when z varies in the neighbourhood
of the point z = b, i.e. the one-sheeted circle i ^ i n the ui-plane corres­
ponds to non-circular neighbourhood of the point z = b in the z-plane
(which contains the point z = b). Our assertion has thus been proved,
i.e. if in the series (128) the coefficient ax # 0 , then the neighbourhood
of the point z — b will be transformed into a one-sheeted neighbourhood of
the point w = a0 and the inversion of the series (128), when the values
of w are close to w = a0, will have the form:

z = b + 2 cn(w — a0)n. (129)


n=l

We shall now consider the case when the first few coefficients vanish
in the series (128):
w - a0 = am (z — b)m + am+1 (z — b)m+1 +
+ am+2( z - b r ^ + . . . (am =f=0), (130)
i.e.
w - a 0 = am( z - b ) m\\ + ^ ± L {z - b ) + ^ ( z - b ^ + . . . ] .
L u rn J

This formula can be rewritten in the form:

= fan (2 - 6 ) jl + (z-b) + ^ - (z-&)2+ • • •])", (131)


m_
where for |Aam we take a definite value of the radical, whilst the
expression {l -t- [. . has m different values for z close to b,
these values being all obtained from any one of them by multiplying
by the values of the with root of unity [I, 175]. The latter equation
is equivalent to (130). On the right-hand side the sum in the square
brackets is close to zero for values of z close to 6 , and we can apply
Newton’s binomial formula to this square bracket [16]:

— f— - l )
94 T H E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [2 3

A circle which can be as small as we please, centre at z = 6 , can be


found in which the square bracket is a regular function, the modulus
of which does not exceed a number q, which is smaller than unity.
In this circle the above series converges absolutely and uniformly.
The terms of the series in this circle form power series and by using
the Weierstrass theorem as applied to power series, we obtain an ex­
pansion of the figured bracket on the right-hand side of formula (131)
in the form of a power series in the above circle:

{1 + [••■]}* = l + c 1( z - b ) + c i ( z - b ) ' - + . . . .
and this formula (131) can be rewritten in the form:

\ w — a0 = dx (2 — 6 ) + d2 (z — 6)2 + . .. , '(131!)
m ___
where dx = |fam ^ 0. When applying Newton’s binomial formula, we
took a definite value of the radical on the right-hand side of formula
(130), and formula (lSlj) gives us the same value of the radical on the
m___
left-hand side of (lSlj). Denote this value of —a0 by w'\

w' = |tw — a0 = dx(z — 6 ) + d2(z — 6)2 + . . . (132)


From what was proved above (dj# 0 ), the one-sheeted neighbourhood
of the point 2 = 6 will be transformed into the one-sheeted neighbour­
hood of w ' = 0 and owing to the fact that w — a0= w 'm, the one-sheeted
neighbourhood of w' = 0 will be transformed into the m-sheeted
neighbourhood of the point w = a0 [19], if we take all the values of
Yw—a0i.e. in the case (130) the one-sheeted neighbourhood of the point z = b
will be transformed into the m-sheeted neighbourhood of the point w = a 0.
Furthermore the derivative of the function (132) does not vanish
when 2 = 6 , and consequently, under transformation by this function,
angles at the point 2 = 6 remain unaltered [3]. The further trans­
formation w — a0 = w m magnifies angles at the point w' = 0 m
times since, raising to the mth power, the amplitude of the complex
number w’ is multiplied by m, i.e. under transformation by the function
(130), angles at the point z = 6 are magnified m times.
Finally, in accordance with what has been proved, the inversion
of the power series (132) has the form:

3= b+ en w’n,
n=1
24] t h e p r in c ip le o f s y m m e try 95

or, returning to the variable w, we obtain the inversion of the power


series in the form:
CD «*_____________

z = b + ^ en (Yw — b)n. (133)


n=l
m ___
As mentioned above, the formula w' = )Jw —b transforms the
m-sheeted neighbourhood w = b into a one-sheeted neighbourhood
y/ = 0 if we take all the possible values of the radical; in the
expansion (133) we must also take all the values of the radical
appearing on the right-hand side. Only then do we obtain aone-sheeted
neighbourhood of the point 2 = 6 .
Above we considered the case when the point 6 and the correspond­
ing point a0 lie at a finite distance apart. Exactly the same results are
obtained when one or both these points lie at infinity. Suppose, for
example, that 6 = °° and that a0 is finite. In this case we have the
following expansion instead of the expansion (130):
w - a 0^ a m— + am+1-— + ( r o > 0 ; a m# 0 ) . (134)

When m = 1 , the one-sheeted neighbourhood of 2 = °° is transformed


into a one-sheeted neighbourhood of w = a0. When a0 = °° and 6 is
finite our function has a pole at the point z = 6 . If this pole is
simple, i.e. when the expansion has the form:
«> = -— y + a 0 + ° i ( 2 - & ) - f ••• , (135)
then the one-sheeted neighbourhood of the point2 = 6 is transformed
into the one-sheeted neighbourhood of the point w — o°. Finally, when
6 = a0 = °° our function is defined at infinity and has a pole at that
point. If this pole is simple then the expansion has the following form:
w = az + a0 + ^ - + - % - + . . . , (136)
and the one-sheeted neighbourhood of the point 2 = °° is transformed
into a one-sheeted neighbourhood of the point w = °°. The function
inversa to (136) has an expansion of the same form:2
2 = —
o tv -)- 60u -) w—|----
ur1—b • • ■ (137)
24. The principle of symmetry. In [18] we described the analytic
continuation of the domain B1into a new domain B 2in the event when
this new region overlaps the initial domain; but at that time we
did not give a practical method for performing the process of
analytic continuation in the general case. We shall now describe one
96 THE BASIS OF THE THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [24

method for performing analytic continuation in a special case, when


the new domain does not overlap the initial domain but merely
touches it along a contour. However, to begin with, we must prove an
auxiliary theorem.
R i e m a n n ’s T h e o r e m . I f f^z) is regular on one side of an arc of the
curve L and also on the curve while f2(z) has the same property on the other
side't of the curve, and the values of these functions coincide on the arc L,
then these two functions jointly define a
single regular function in a domain
which contains the given arc or, in
other words, f2(z) is the analytic con­
tinuation of f2(z).
Draw the contours l2 and l2 which
have common ends on L\ one con­
tour lies in the domain where the
function /x(z) is regular and the
other in the domain where the function f2(z) is regular, so that our
functions fx(z) and f2(z) are regular in the domains B1and B2, bounded
by the closed contours l2 and L and l2 and L respectively (Fig. 21 ).
Take a point z in Bv This point lies outside B2 and we can therefore
write [7]:

/i+L
0 = j2m
_ r /, (o dz'.
J z' —z
l.+L

If we add these two equations, on the right-hand side we have to


integrate twice along the arc L in opposite directions, while the integ­
rands are the same in both cases, since, by hypothesis, the values of
/ 1(z/) and f2(z') coincide on L. The two integrals thus cancel each other
and only integrals along the arcs l2 and l2 remain. For the sake of
simplicity, denote by f(z') the function which is equal to / 1(z') on the
arc and equal to / 2(z') on the arc l2. Adding the above equations, we
obtain:

h +l.
Similarly, by taking the point z in the domain B2, we obtain:

k (z) = 2ni J+i. ^ ’ dz',


/,
24] TH E PRIN C IPLE OF SYMMETR7 97

i.e. our functions f x(z ) and / 2(z) are expressed by the same integral of
Cauchy’s type round the closed contour (^ + Z2)• Consequently, the first
of these functions can be analytically continued from the domain B l in
the domain B 2, while the second function can be continued from the do­
main B 2 into the domain B x and, as a result of the analytic continuation,
a single analytic function is obtained, which proves Riemann’s theorem.
Note that we used Cauchy’s
formula in the above proof; this
formula also applies when the
function is not regular on the
contour but when it is conti­
nuous in a closed domain and
regular in that domain. We are
thus under no obligation to
assume that the two given func­
tions f i ( z ) and f 2(z) are regular
on the arc itself, as is given
in the conditions of Riemann’s
theorem. It is sufficient that
/j(z) is regular on one side
of the arc L and that it is continuous as far as the arc; the same applies
to f 2(z ) on the other side of the arc; also, the values of these functions
must coincide on the arc L . Riemann’s theorem thus proves that each
function can be analytically continued across the arc and that one of
these functions is the analytic continuation of the other.
Let us now consider the principle of symmetry.
T h e P r i n c i p l e o f S y m m e try . I f / 1(z) i s re g u la r o n o n e s id e o f a se g m e n t
(a , b) o f th e r e a l a x i s a n d i f i t i s a l s o c o n t in u o u s u p t o t h i s l i n e , w h i l e o n
th e l in e i t s e l f i t s v a l u e s a r e r e a l, th e n t h i s f u n c t i o n c a n b e a n a l y t i c a l l y
c o n tin u e d a c r o s s t h i s l i n e , a n d a t p o i n t s s y m m e t r i c a l w i t h r e s p e c t t o th e
r e a l a x i s , t h i s f u n c t i o n h a s c o m p l e x c o n j u g a t e v a lu e s .
Assume, for convenience, that our function f x{z ) is regular in a domain
B lt which touches the line (a , b ) and lies above it (Fig. 2 2 ). Construct the
domain B 2 so as to be symmetrical with B x with respect to the real axis
and proceed to define the function f 2(z ) in that domain according to the
following rule: assume that at every point A 2 of the domain B 2 the
function f 2(z ) has a complex value, which is conjugate with the value
of the function f x(z) at the point A v symmetrical with respect to the real
axis. These symmetrical points evidently have complex conjugate
coordinates and denoting, as usual, the complex number which is
98 TH E BASIS OP TH E TH EO BY OP FUNCTIONS OF A COMPLEX VARIABLE [24

conjugate with a by a, we can define our function f 2(z ) in the domain


b 2 as follows:
/*(*> = M * ) -
This newly constructed function will be regular in the domain B v
for the increments A z of the independent variable and A w of the
new function are complex conjugates to the analogous magnitudes
for the function fy (z ) at the symmetrical point. The same can be said
with regard to the ratio of their increments. Consequently, this ratio
tends for the function f 2(z ) to a definite limit, equal to the complex
conjugate to the analogous limit for f x( z ), i.e. it tends to a limit equal
to /'(z); therefore the function f 2( z ) will be regular in the domain B 2.
The values of f 2( z ) coincide with the values of fy (z ) on the line (a , b),
since on this line the values of fy (z ) are real. In accordance with
Riemann’s theorem, f 2(z ) is the analytic continuation of fy (z ) across
the line, which proves the principle of symmetry.
The principle of symmetry can be formulated geometrically as fol­
lows: if f y(z ) is regular on one side of the segment ( a , b ) of the real axis
and if it transforms this segment into another segment on the real axis,
then this function can be analytically continued across this segment;
now, points symmetrical with respect
to the real axis will be transformed
into other points, symmetrical with
respect to the real axis. The principle
of symmetry can be formulated in a
more general way, by introducing
the concept of p o i n t s s y m m e t r i c a l
w i t h r e s p e c t t o a c ir c le , viz. two points
are symmetrical with respect to a
circle if these points lie on the same
radius of the circle (one point must
lie on the radius itself and the other
on its continuation) and the product of their distances to the
centre of the circle is equal to the square of the radius of the circle
(Fig. 23).
Let and A 2 be two points symmetrical with respect to the circle
C . Draw a circle C ' through these points and let M be one of the
points of intersection of this circle and the circle C .
Taking into account the fact that the product of O A 2 and its outer
part O A y must be equal to the square of the tangent and, on the other
25] TAYLOR’S SERIES 021 TH E CIRCUMFERENCE OF TH E CIRCLE OF CONVERGENCE 99

hand, that by definition this product must also be equal to the


square of the radius O M 2, we can assert that O M is a tangent of the
circle C , i.e. the circle C ' is orthogonal to the circle C . It is thus
easy to see that two points A x and A 2, symmetrical with respect to
the circle C, have this characteristic property: any circle drawn

F i g . 24

through these points is orthogonal to C, or in other words, a fa m ily of


c ir c le s th r o u g h p o i n t s , s y m m e t r i c a l w i t h r e s p e c t t o th e c ir c le C , c o n s i s t s o f
Two points, symmetrical with respect to a
c ir c le s o r th o g o n a l t o C .
straight line, have the same characteristic property, viz. a f a m i l y o f
c ir c le s d r a w n th r o u g h th e s e t w o p o i n t s c o n s i s t s o f c ir c l e s o r th o g o n a l to
(Fig. 24).
the s t r a i g h t l i n e
In this general form the principle of symmetry reads as follows:
if f x(z) is regular on one side of an arc ( a , b) of the circle C v if it is
continuous as far as the arc and transforms this arc into an arc of
another circle C 2, then f x(z ) can be analytically continued beyond this
arc (a, b) and points, symmetrical with respect to the circle C 2, are
transformed into points, symmetrical with respect to the circle C 2.
In this definition of the principle of symmetry we can understand by
the term “circle” a circle in its ordinary sense or a straight line.
The proof of this general definition of the principle of symmetry is
given at the beginning of the following chapter.

25, Taylor’s series on the circumference of the circle of convergence.


Consider Taylor’s series

(138)
2 a k (z -
k=0
b )k
100 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [25

with radius of convergence R. Putting z — b = q e"p, we can write


the series in the form:

2 a kek^ (139)
k=0
or
00

^ Oft (cos Tc<p+ i sin k<p) Qk.


k=0
It is given that this series converges when g < R. When q = R, i.e.
when we are on the circumference of the circle of convergence, we can
say nothing definite about convergence. If we take, for example,
the series
1 + z + z2 + . . . *(140)

with the radius of convergence R = 1, then on the circumference of


the circle of convergence, i.e. when | z | = 1, the moduli of all terms
of the series are equal to unity, and the series, evidently, diverges
on the entire circumference of the circle of convergence. Consider the
following series as the reverse example:

1 + -p- + - |r + • • • (141)

In this series the ratio of the modulus of one term to the modulus
of the preceding term is
zn+1
(n + 1)* '

and this ratio tends to | z [; therefore, in accordance with d’Alam-


bert’s test, the radius of convergence of this series is also equal to
unity. By substituting z = e,ip we obtain a series, the moduli of the
terms of which are equal to positive numbers 1/n2, forming a con­
vergent series, i.e. the series (141) converges absolutely and uniformly
not only inside the circle of convergence but also throughout the closed
domain including the circumference. We can thus see that the con­
ditions relevant to the convergence of a power series on the circumfer­
ence of the circle of convergence can be very varied.
We saw earlier that differentiation and integration of a power
series does not alter the circle of convergence. However, these processes
can have a definite effect upon the convergence of a series on the
25] TAYLOR'S SERIES ON THE CIRCUMFERENCE OF TH E CIRCLE OF CONVERGENCE 101

circumference of the circle of convergence. Thus, for example, by


integrating the series (140) twice we obtain the series
g3 |__ 2=4 L
_Z.___ I__
1-2 ' 2 -3 ' 3*4 ~ ’
which, like the series (141), converges absolutely and uniformly
throughout the closed circle.
We shall now describe a theorem which deals with the sum of a
power series, when the latter converges on the circumference of the
the circle of convergence. We proved an analogous theorem earlier
for the real variable [I, 149] and we shall therefore not prove this
theorem again: below we just formulate the result.
A b e l’s S e c o n d T h e o re m . I f the power series (138) converges at a
point z = b — Re'^o on the circumference of the circle of convergence,
then it will converge uniformly along the whole length of the radius
arg (z — b) = <pQ. It follows that the sum of the series is a continuous
function along the whole length of the radius, i.e. the value of the sum
of the series at a point Relip< ‘ on the circumference is equal to the limit
to which the interior value of the sum of the series tends when ap­
proaching the point Rel<Poalong the radius from the interior. A simple
evaluation of the sums of certain trigonometric series is based on this
theorem.

Consider one example. In the expansion


2 22 23 Z4
,0 8 (1 + z) = T - T +T - T + -
replace z by (—2 ) and subtract the series thus obtained from the one above.
We thus obtain the expansion

(142)
(T + T + f + - )
in the circle of convergence | 2 | < 1. In this expansion, put z = e,?> and
separate the real and imaginary parts:
( cos <p cos Zip cos 5<p 'V .. ( sin <p sin 3<p sin 5<p 'I
^1 I ' ^ 3 5 h • • -J + ^ j 3 5 h • • -J •

It can be shown, though we are not going to do so here, that both these trig­
onometric series converge when <p differs from k n (k = 0 , ± 1 , ± 2 , ...) .
Let us determine the sums of these series. Separating the real and imaginary
parts of the function, we have:

log 1l _ 2z = .log | l + z | + i arg


1 1 +*
|1- 2| 1 —2
102 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [25

From Fig. 25, when z = e,p we obtain directly:

11 + 2 I = 2
2
(0 < q>< 2n) ,
[1 - 2 |= 2 sin ^ ■

The amplitude of the fraction (1 + z)/( 1 —z) is equal to the angle between
the vectors A M '(—z — 1 ) and AM (z — 1); when z = 0 the sum of the series
(142) is equal to zero, and in this case the angle must also be equal to zero.

When the point z coincides with the point elv this angle rests on the diameter
and is evidently equal to (± jt/2.) We have thus found the sum of the above
trigonometric series:
i . 9 c, ( cos <p cos 39? 'I
log c o tT = 2 ^— - + — 3 — + - - J
(0 < <
p < n)
71 n ( sin q> , sin 3<p ,
T = 21 I 1 3 *■'
Note one other circumstance, connected with the representation of
the trigonometric series in the form (139). Separate the real and imagi­
nary parts in the coefficient ak, viz. ak = ak — i(ik. Substituting in
formula (139) and separating the real and imaginary parts in the sum,
we obtain the following formula:
00

/ (z) = 2 (“*cos k<P + Pk sin ek +


k= 0
00

+ i^ (—j3k cos kq> ak sin k<p) Qk. (143)


k= 0
26] TH E PRINCIPAL VALUE OF AN INTEGRAL 103

The second trigonometric series differs from the first series only by
the fact that the coefficients of cos kip and sin k<p are interchanged
and the coefficient of sin kip has its sign reversed. The second trigono­
metric series is usually said to be conjugate with the first series. Note
that we introduced the minus sign in the coefficients ak for greater
simplicity in subsequent formulae. This is of no fundamental import­
ance, since the real number can be both positive and negative.

26. The principal value of an integral. We shall now consider the limiting
values of Cauchy’s integrals. To begin with let us introduce a new concept in
connection with integrals of discontinuous functions. Let x = c be a point
in a finite interval (a, b) and /(x) a function which is defined in this interval.
Assume further th at the integrals
C— 6

J f (x) dx an d J" f (x) dx (144)


«+•
exist for any e > 0. Assume, for example, th at f(x) is continuous in the whole
interval (a, b) except a t the point x = c and that it becomes infinite when
x tends to c. The improper integral of f(x) over the interval (a, b) can be defined
as follows: if when e +0, the integrals (144) tend to finite limits then the sum
of these limits is, in fact, equal to the integral of f(x) in the interval (a, b)
[1, 97]. If the integrals taken separately have no limits but the sum of the in­
tegrals, when e + 0 , tends to a finite limit, then this limit

lim J" / (x) dx j / (x) dx


B—
*■+0 £+•
is know n as th e principal value o f th e integral over th e in te rv al (a, 6):
b c —s b
v. p. J / (x) dx = lim f / (x) d x + j" / (x) dx (145)
a «—+0 X -i .
where v. and p. are the first letters of the French words valeur principale which,
in English means “principal value” .
In future, for the sake of compactness, we shall not write the letters v. p.
in front of the integral. The main feature of definition (146) is that the limits
of integration on the right-hand side of the formula contain the same number,
which tends to (+0).
The principal value of an integral can be defined similarly when f(x) has
several discontinuities in the interval. If the usual improper integral of the func­
tion f(x) exists over the whole interval (a, b) [I, 97] then the principal value
(146) of the integral evidently coincides with this improper integral.lt follows
from the definition (145) that a common factor can be taken outside the integral
and th at the integral of a finite number of terms is equal to the sum of the
integrals of the individual terms, on the assumption that the integrals of the
terms exist only in the principal value sense.
104 TH E BASIS OP TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [26

We shall now give some simple examples of the principal value of an integral.
Consider the integral:
b
c dl
(146)

where a < x < b and where p is a positive number. When p > 1 we have:
x —6 b
rAt r H/
dt li |r li l
J {t — x)p J (t — x)p — p — 1 [ (b — x)p_1 (a —x)p -T +
x+ e

+ [ ( - l ) p- 1 - l ] ^ I - j -
When p is even the last term on the right-hand side is ( —2): ep~l, and the right-
hand side increases indefinitely when e -* (+0), while the integral (146) does
not exist. However, when p is odd the right-hand side of the above formula
does not contain e and we have:
b

J (t — x)p 1 —p [ ( 6 — x)p_1 (o — x ) p _ 1 ] (P is o d d ) .

When p = 1, we obtain:
t = x —e
f dt , r dt , , b —x
+ log (t - x) = log
x+e t= a <=X+e

f d( b —x
J t —x x —a

We say th at the function to(x) satisfies a Lipschitz condition of order a in the


interval (a, b), where 0 < a < 1, provided th at for arbitrary values of xl and x,
in the interval, the conditions shown below are satisfied:
| £0 (Xj) — CO(x,) I < l | x ! - x 11“, (147)
where & is a constant. We introduced the same condition earlier, when a = 1,
and we saw th at it is satisfied when co(x) has a bounded derivative in the
interval [II, 51], Consider the integral:

(148)
/ ( x ) = j T = l T d(’

which can be rewritten in the form:

CO(i) — CO(x) f dt
- df + “ (*) J 7— ^
26] TH E PRINCIPAL VALUE OP AN INTEGRAL 105

Using the condition (147) we obtain the following inequality for the integ
r a n d of the first integral in the neighbourhood of the point t = x:

I co(t) — co (x ) k
(149)
I t —x ^ 11 — x | 1 - a

and, consequently, this integral is absolutely convergent in the usual sense


[II, 82]. The second integral is equal to:
b —x
co (x) log
x — a

Hence the integral (148) has a meaning for any x in (a, 6) provided that tu(<)
satisfies a Lipschitz condition (147). The function f ( x ) , which is defined by the
equation (148), is defined for all values of a; in (a, 6). Construct the expression:
x —« 6

a x+e

When e is positive the integrand is a continuous function of t and x , provided


that x belongs to an arbitrary closed interval within the interval (a, b), and
that t belongs to the interval (a, x — e) or ( x + e, 6); consequently, the expres­
sion (150) is a continuous function of x [II, 80]. Using the identity:
” >(*) = 01 ( t ) - <u (x) m , . 1
t — X t — X ~ K>t —X

and the condition (147), it is easy to show th at when e -*■ (+0), the expression
(150) tends uniformly to the limit f ( x ) with respect to x and, consequently, the
function f ( x ) , defined by the formula (148), is a continuous function in any
closed interval contained in (a, 6); in other words, the function f ( x ) is a contin­
uous function in the interval (a, b). Later we shall prove the more precise
result viz. when o>(t) satisfies a Lipschitz condition of order a < 1 then the
function f ( x ) will also satisfy a Lipschitz condition of the same order a in any
interval within (a, b). When a = 1 in the condition (147), f ( z ) satisfies a Lip­
schitz condition of an order less than unity.
The continuity of the function co(x) evidently follows from the condition
(147). On the contrary, it does not follow from the continuity of the function that
it satisfies a Lipschitz condition, i.e. a Lipschitz condition is a stronger condi­
tion than mere continuity. Note, that if the integral (148) is to exist a t a point
x it is sufficient that co(t) satisfies a Lipschitz condition in a neighbourhood
of the point x and th at in the remaining part of the interval (a, b) it is continuous
or merely integrable for the integral (148) exists when the inequality (149)
applies to all values of t sufficiently close to x . If every point x in the inter­
val (o, 6) belongs to the interval in which a Lipschitz condition (147) is
satisfied with a given a and k, then the integral (148) exists for all values of
x in (a, 6). Here, the constants a and k can be different in different intervals
contained in (a, b).
106 TH E BASIS OP TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [26

We shall now investigate the possibility of a change of variables in the


integral (148). To begin with, let us prove the lemma: if 77,( 6) and rj2(e) are such
that the ratios 77,( 6); e and 772(e): e tend to zero when t ( + 0 ). then
X-e+fJiC*)
[ J i £ L d « = ue_m+0 J C tJ^W_dl+
J t —X - X J
r pt —> - At
a L a x+«+i7,(0
To prove the lemma it is sufficient to prove that
x-«+i,(«) X+£+17,(1)
co(t)
lim rr j “o e
W d( = 0 and lim f - At = 0.
e-»+0 J t~ i £ —►t 0 J t
x—« X+£
We shall prove, for example, the first of these equations. Assume that 77,( 6) >0,
then | ( — x \ > e — 77,( 6) when x — e < t < x — e + 77,( 6) and, conse­
quently:
X-£+*7i(£)
m • 77, (e) _ m 77, (e) ; Q
e - *h (*) 1 _ 2i(f) e
X—£ 6

where m is the maximum value of | ui(t) |. When 77,( 6 ) < 0 , we can write:
X—£+77,(0

J
x-a
“>(0
t —x
„ m • 177, (e) |
Al s ------------
e
0 ,

and the lemma is thus proved.


By using this lemma it is easy to prove the formula for the change of vari­
ables in the integral (148).
Theorem. Let t = p(z) be a monotonically increasing junction of r, varying
in the interval (a, b) when a < t < (3, while ft( t) has continuous derivatives up
to the second order in the interval (a, fi) and p'( t) p 0 in (o, /?). In this case the
formula for the change of variables applies:
b p
C a (t) , f <0 [p (t)] p' (t
(151)
J t —x J p (T )-p {£ )

where x = fi(ti) and the integral on the right-hand side is to be understood in the
principal value sense.
In accordance with the definition of the principal value of an integral, we
form the sum:
f- b

J <*>[/* (*)] ^ (*) d , r


/*(*)-/< (£) ^ J
£+8
L« (*)] ^ (t) dr.
ju(T)-A«(f)
(152)

Denote: — e) = x — e' and /i(f + e) = x + e ' + 77. According to Taylor’s


formula:
/£({ + h) = p (f) + v ( f ) + E - p r (f + eh) (0 < e < i).
27J TH E PRINCIPAL VALUE OF AN INTEGRAL (CONTINUATION) 107

Putting h = —£ and later, h = + e, we obtain:

x - e ' = x - e/i' (|) + (f - 0,£) ;

e2
® + e’ + 7) = a: + e/i’ ({) -f — /t" (£ + 02e) (0 < 0, and 02 < 1) ,

whence:

= 8 [/*' (f) - - fi" (S - ».«)] ; »? = - [ / « '( ! + M + # • '( * - M ] ,

and, consequently, the ratio r : e' tends to zero when e' -► 0. Transforming
the integrals in the sum (152) to the variable t, we can write this sum as follows:

a x +e’+rj

and by using the lemma proved above we can assert that the sum (152)
gives in the limit the integral on the left-hand side of (1^1), which proves the
above formula. In the conditions of the theorem the “monotonically increasing
function can evidently be changed to “monotonically decreasing” .

27. The principal value of an integral (continuation). The concept of the


principal value of an integral can also be applied to a line integral. We shall
only consider integrals of Cauchy’s type:

f (f) = j 7 — | - d r - (153)
L

where L is a closed or open contour in the plane of the complex variable r


and | is a point on that contour, which does not ooincide with its end when the
contour L is open. Let s be the length of the arc L, measured from a certain
point. In future we shall assume that the functions x(s) and y(s) in the para­
metric equation of the contour t(«) = x(a) + y(a)i have continuous derivatives
up to the second order. Assume th at the point r = f corresponds to the point
a = a0. The principal value of the integral (153) can be defined as the principal
value of the integral with respect to the real variable a:

f w [ t (a)]
t' (a) da, (154)
J r (a) —r (s0)

where I is the length of the contour L, and we can assume th at a0 lies in the
interval of integration. Precisely as in [26], it can be shown that the integral
(153) exists if the function co(r) satisfies a Lipschitz condition on L:

I 0} (Tj) —to (Tt) I < k I Tj — T, I1 (0 < a < 1). (155


108 TH E BASIS OP TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [27

Using the theorem for the change of variables proved in [26], it is easy to
show th at if, in the parametric equation of the contour t (1) = x(t) + y(t)i,
the functions have continuous derivatives up to the second order and T'(t) 5A 0 ,
then the principal value of the integral (153) reduces to the principal value of
the integral:
f w [r (t)]
t' (t) d l ,
J v (t) —r (t0)
a
where (a, fS) is the interval of variation of the parameter t and t = t0 corresponds
to the point r = f. If tu(r) is identically unity then we have the primitive
log (t — t 0) for the integral (153) and we obtain the following result when
the contour is closed:
P <1t
| • (156)
L
We must always remember th at we integrate round a closed contour in the
counter-clockwise direction. Similarly, as in the case of a straight line, we can
assert th at, when the condition (155) applies, then formula (153) defines a
function / ( f) which is continuous at all interior points of L, when the curve
is open, or at all points of L, when the curve is closed. In this case, as in the
case of the straight line, the more exact theorem, proved by I. I. Privalov
(Dokl. Akad. Nauk, S.S.S.R., X X III, No. 9, 1939) applies:
I f the condition (155) applies then the function /(f) satisfies on a closed contour
L a Lipschitz condition for the same a when a < 1, or of any order, less than
unity, when a = 1. I f L is an open curve then the same condition applies for f(£)
on any closed arc of the curve within L.
We shall prove this theorem for a section of a straight line. The proof is
analogous for contour integrals. As a preliminary let us make some remarks
with regard to a Lipschitz condition. I t is easy to see th at a Lipschitz condition;

[/(f+ - / ( f ) | < k \A S \a (157)


can be tested for sufficiently small values of [ zlf |. In fact, let (157) be proved
for | A{ | < m, where m is a positive constant. When | d f | > m then the
relationship:
| / ( g - Mf ) - / ( g ) l
m *r
remains bounded, i.e.
|/(f-M f)-/(f)|< * iM f|° (M f|> m ),
where kl is a constant. Choosing the greater of the two constants k and kt
we obtain a Lipschitz condition for all permissible values of A\. Suppose also
that f) < a < 1. For values of A£, with modulus less than unity we have
| | jS > | A f | a, and therefore, if/(f) satisfies a Lipschitz condition of order
a, then it will also satisfy this condition of order /S. Suppose th at the two
functions / / f) and / 2(f) satisfy a Lipschitz condition of the same order a.
I t is easy to see th at their sum and their product also satisfy this condition of
27] THE PRINCIPAL VALUE OF AN INTEGRAL (CONTINUATION) 109

the same order. In the case of the sum this is directly due to the fact th at the
modulus of a sum is less than or equal to the sum of the moduli, and in the
case of the product we can write:
h («+ w /. <f + ^f) - /. (« h (f) = /, (f + m [a a + as ) - h (f)] +
+ A(« [ M l - M f ) -/*(£)].
from which our remark about the product follows directly.
We shall now prove the above theorem. We have:

N fl-f-R * '

l ( l ) ~ +
a
where ai(t) satisfies a Lipschitz condition of order a. Assume th at f belongs
to an interval I, in (a, b). In the second term on the right-hand side the factor
ai(£) satisfies a Lipschitz condition of order a and the second factor has a
bounded derivative which satisfies the same Lipschitz condition of the first
order. Thus the whole product satisfies a Lipschitz condition of order a and
it is sufficient to prove the theorem for the function:
b

expressed by the usual bounded integral. We have to find the upper bound of
the modulus of the difference:

v (g+ jf) - v(g)= j [ * (t)( ~ - w(i)t z r (f)] &» <i58>


a
where | | is sufficiently small. On separating the interval (£ — £,<? + £)
from the interval of integration, where e = 2 | |, we take the upper bound
of the integral (158) in this part of the interval. By using the condition (155)
we obtain:
t+e

{-*
The integral of the second term can be represented in the form:
f *+«
J ( f - t ) “" 1d / + J (t — i)a~1dl = (2° | d f + 2“ | d | |a).
f-« (
no TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [27

We can proceed similarly with the integral of the first term and the modulus
of the integral (158) in the interval (f — e, { + e) has an upper bound:
fc, | A£ | a, where is a constant. Upper bounds of the integrals in the added
intervals (a, £ — e) and (£ + e, 6), remain to be found. For this purpose we shall
represent the integrand in the form:

[« w - “ <f + ^)] (f-e-At) - H* + T^T• <159>


Using (155) we obtain the following inequality for the modulus of the integral
of the second term:

dt dt b —£
k +
II
t - S h -f lo® £ — a
{+«
where k2 is a constant. We recall th at the modulus of the above logarithm
remains finite during variations of £ in the interval I. The upper bound of the
integral of the first term in the expression (169) remains to be found in the
added intervals (a, £ — e ) and (£ + e, b). Let us find the upper bound of the
integral in the first interval. The inequality in the second interval will be exactly
the same. From (155) we have the following inequality for the first term of the
expression (159):

[co(t) — oi(£ + A£)] <
( ! - { ) ( ( - { - A£)
\A£ | k M* 1
1—a
1
Mr

Mr

a
1

w
9
1

1
1

1
1

t- £

When t varies in the interval (a, £ — e) then (£ — t) > e, i.e. {£ — t) > 2 | J f |,


and, consequently | A£ |: | t — £ | < 1/2; hence:

Thus during the variation of t in the first interval (o, £ — e) the modulus of
the first term of the expression (159) does not exceed
2 1~gfc | .dg |
( { - ( > 0),
(£ - 0 *"“

and the modulus of the integral of the above first term can be written as fol­
lows:

2 ak \A£ i r — - — (159,)

When a < 1, then the inequality is as follows:


21~° k l 1 k
+ < A£\a.
1 —a (f - o)1-a 2l~a \A £\1~a 1 — a
27] TH E PRINCIPAL VALUE OF AN INTEGRAL (CONTINUATION) 111

We thus obtain the required upper bound for the difference (158) when a < 1.
When a = 1 the upper bound (159j) takes the following form:
f c | d f | [ l o g ( f - o ) - l o g ( 2 | d f |) ] ,
the difference (168) when a = 1 can be written as follows:

*1 | d * | + * 1 M f | i o g - J f p

where k3 and kt are constants. Bearing in mind that as /If ->- 0, log (l|zlf |),
tends to infinity more slowly than any negative power of | d f | we can write:

I+ MS I log < *» M f I'5.

where f) is any number which satisfies the condition 0 < /? < 1; the theorem
has thus been proved for the case when a = 1.
We shall now investigate the behaviour of the function / ( f ) when the point
f aproaches the ends of the line, for example, when it approaches the end
t = a. We are assuming, as we did above, th at ai(t) satisfies a Lipschitz condi­
tion of order a on the whole closed section (a, b). To start with we assume that
co(o) = 0. In doing so we can extend this to saying that the function is zero
when t < a, i.e. we can assume that co(t) = 0 when t < a. In this case cu(t) is
defined on a section (o„ 6) where ot < a and the Lipschitz condition is not
affected by the above extension. The integral

a, a

gives the former function/(f), and bearing in mind th at the point t = a lies
within the line (alt b) we can maintain on the strength of what was proved
above, th a t/(f) satisfies a Lipschitz condition of order a (we assume that a < 1)
and on any line (a, &,) where 6, < 6. Suppose now that co(o) # 0.
We can write:
b b
, ... f co It) — (o (a) , f d(
'<« = J - y - f— d( + J T ^ r •
a a

The numerator of the first integrand vanishes when t = a and this integral
gives a function which satisfies a Lipschitz condition of order a up to the point
f = a. As we saw in [26] the second term on the right-hand side is equal to:
co (a) log (6 — f) — co (a) log (f —a ).
The minuend in this difference satisfies a Lipschitz condition of order one
up to f = a.
Thus in the neighbourhood of the point f = a the function /(f) represents
a sum:
- o> (a) log (f - a) + /, (f) ,
112 t h e b a s is o f t h e t h e o r y o f f u n c t io n s o f a c o m p l e x v a r ia b l e [28

where / t(£) satisfies a Lipschitz condition of order a up to the point £ = a.


When the end £ = 6 is considered, we obtain the analogous result:

a. (b) log (b - £) + /, (£),

where/»(£) satisfies a Lipschitz condition up to the point £ = 6.


The behaviour of the function/(£) near the ends of the line was also considered
when more general assumptions were made with regard to co(t). We only quote
the result here, but the proof can be found in Muskhelishvili’s book Singular
Integral Equations, which contains the results of the first investigations into
integrals of Cauchy’s type.
T h e o r e m . Let w(t) satisfy a Lipschitz condition (147) of order a on any
closed line (o', b') within (a, b); the constant k depends on the choice of the line
(o', 6') (k grows indefinitely when a' ->■ a or when 6' ->- 6). Assume further
th at near the ends of the line, the function a>(t) can be represented ip the form:

(160>
where c denotes either a or b, y — yt -f- y2 i (y#0) when 0 < yt < 1, and
satisfies a Lipschitz condition up to t = c. At the same time /(£) satisfies a
Lipschitz condition of order a if a < 1, or of any order less than unity, when
a = 1, on any closed line within (o, 6); in the neighbourhood of £ = c it has
the form:
/ (£) = ± n cot yn ™ ^ ■+ /, (£),
(£ - c)v
and when y, = 0 then /j(£) satisfies a Lipschitz condition up to t c; if
y, ^ 0 then
r (*)
/. (f) = | £ - c | v .

where / (*,(£) satisfies a Lipschitz condition up to £ = c, and y0< y; the sign


(+ ) refers to the case when c — a and the sign ( —) to the case when c = 6.
All this applies when the line is replaced by a sufficiently smooth arc with ends
t = a and t = 6, and when we integrate with respect to the complex variable t.
Note th at when y = 0 the result shown above applies:

/ (£) = ± co (c) log ^ e + A (£),

where /(£) satisfies a Lipschitz condition up to £ = c.

28. Integrals of Cauchy’s type. Consider an integral of Cauchy’s type (8):

w (t)
F dr, (161)
T— Z
L

where z does not lie on L. If L is a closed contour, then this integral defines two
different regular functions: one within L and the other outside L. If the
28] INTEGRALS OP CAUCHY'S TYPE 113

contour is not closed then F(z) is regular outside L. In either case F(oo) = 0.
When z = £ lies on the contour we have the principal value of the integral
and we can rewrite it in the form:

1 f m (T) a- _ m ($) f d r , 1 f “>M - to (£)


2m Ji:-£ 2 m j T - £ " r 2 mJ t-£
I. L L

i.e. from (156)

L L

To begin with we assume the contour to be closed. We shall prove the theorem:
if z tends towards a point £ on L then the integral (161) has the limit:

± y “ <{>+ -s s - J r = I j d' ’ <l63)


L

where the (+ ) sign is taken when z —■ £ from inside L and the ( —) sign when
z -* £ from outside L. Consider the first case. The integral (161) can be rewritten
in the form:
1 f co(r) dT = to (£) f dr 1 f to (t) - co (£)dT
2ni J r —z 2ni J r — z ' 2ni J t —z

or

Consider the difference:

J 1^
_ rf co (r) — co (£) ^ 1 to (tt) —co
f to i (£)- dr =
2m J 2mJ t—£
L
_ 1 f to (r) — <u (£) z — S dT
(165)
2m ,J t —£ t —z

On either side of the point £ mark small arcs g. Denote the part of the contour
thus formed by L x and the remaining part by L 2. Denoting the difference (165)
by a single letter A we can write:

co (t) — co (£) _ z — £ dr + J _ f co (t) — co (£) z —£ dr. (166)


r —£ r —z 2m J t —£ r —z

Suppose that z tends to £ along the normal to the contour L. In this case the
distance from z to £ is less than the distance from z to other points on the
contour, i.e. | z — £ | < | r — z |. Also d r = | x'(s) + y'{s) i | ds, where
114 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [28

| x'{a) + y ’(a) i | = 1. By finding in the usual way an upper bound of the


First integral in the formula (166) we obtain:
s*+s
1 r co (r) -c o (() z - £ dr < f ____ d«____
2m J r— f r— z 2jiJ | t (a) — t (a0) |l““
L)
where a = a0 corresponds to the point r = |. Bearing in mind that the ratio
of the length of the chord | r(s) — t(«0) | to the length of the arc | s — a0 \
tends to unity we can assert th at the latter integral converges and con­
sequently, for any given positive e we can choose tj so small that the modulus
of the integral along L x is smaller than e/2. Having thus fixed r) we obtain
the usual integral along L 2 in which j t — f ( and | r — z | remain greater
than a given positive number and therefore the modulus of the integral along
Z/2 will be smaller than e/2 for all z’s which are sufficiently close to Owing to
the lack of restriction on e we can maintain that the difference (166) tends to
zero when z -*■ £ along the normal, i.e.

1 f CO ( t ) — CO (f) dT = 1 f CO (t) - CO (I) dT


lim 2ni J r —z 2m J r —£
z-f L L
or, from (156)
1 f
lim 2m J
co (t) -
r —z
CO {()
dr - 2mi Jr r<o—
(?)
£
z-f L
and formula (164) gives the required result:

lim f W ^ dr = — co (|) + 2m f dr. (167)


2_ i 2m J r — z 2 J r —f
L L

In the case when the point tends towards the contour from the outside the
proof is exactly the same but we must remember that:
1 f dr j 1, when z lies inside L
2m J r —z ~~ ( 0, when z lies outside L ■ '

Until now we assumed th at z -► { along the normal. I t can be shown that


formula (167) still applies when z tends to £ in an arbitrary way. In order to
do so it is sufficient to show that the limiting tendency of the point towards
the contour is along the normal; the integral (161) tends uniformly to the limit
(163) for all values of $ on the contour L. We shall only consider the circle
| z | = 1. To begin with we suppose that z -► f along the normal. In this case
r = e^; | = el‘Pa and ds = dip. I t is easy to show th at when 0 < x < n/2 then
sin x > 2x/jt. Using this we can write:

| T - 11 = 2 sin I y ~ y° I > ■
— I cp - <Pv | ( I <P ~ <P» I < at) ,
28] INTEGRALS OF CAUCHY’S TYPE 115

and the modulus of the integral along L will be smaller:


Vi+’l 9,+n
Jl f n ° d<p _ k r dy _ k ?„
271 J 21_a \<p — <p0 |1-a ~~ 21 a 7ia J (¥>-7>0)1_a ~ 21~a7itt a n ‘

On the contour L2, provided z lies sufficiently close to £ we have::

| r - f | > — sin ; 2I> y 8ln v ; | co (r) — to (f) | < 2M ,

where M is the maximum value of | co(r) | on L. Assuming th at 6 = | z — £ |


we obtain
1 f (o(r) — co {£) z —£ 1 SMS SMd
2m J r —£ T ----Z
dr <
2tt sin2Tj
(2n - 2 r,) < sin2rj
L,

and finally:
Ml < 21-°k na a ria + sin2
8M
T]
To begin with r] is so chosen that the first term is less than c/2; as a result
of this fixed rj the second term will be less than e/2 provided <5 > (esin2 rj):
(16M). These inequalities do not contain f and therefore the difference (166)
tends to zero uniformly with respect to | as z tends towards the circum­
ference along the radius. Consequently this tending to a limit in formula
(167) also takes place uniformly with respect to f. One result of this is th at the
right-hand side of formula (167) and the integral (161) represent a continuous
function of £ [1, 145]. In [26] we mentioned the fact th at this function satisfies
a Lipschitz condition.
Denote the right-hand side of formula (167) by a),(£) and assume th at
z -*■£ in any manner. Let £' be a variable point on the circumference which lies
on the same radius as z. Evidently £'-*■£ and | z — {' | -► 0. Using the fact
proved above th at the tending to a limit in (167) proceeds uniformly when
z tends towards f along the radius we can maintain that for any given positive
e we have for all z’s sufficiently close to f :
£
27ii T—Z dr — n>i (£') ~2 '
L

On the other hand, as a result of the continuity of co1(£), the modulus


| tot(£) — ail(£') j < e/2 holds for all z’s sufficiently close to £ and therefore:
1 f co(r) dr — cot (£)
<e
2m J t —z
L

for all z’s sufficiently close to £. Owing to the lack of restriction on £ this shows
that the tend to a limit in formula (167) takes place when z tends towards £
from the inside uniformly and in any manner with respect to £. In other words
we can maintain th at the function F(z) which is defined in the circle by the
116 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [28

integral (161) is continuous up to the circumference and its limiting values on


the circumference are given by the formula (167). The same condition applies
when the limit is approached from outside.
This property of integrals of Cauchy’s type can be proved for any closed
contour L when the assumptions given in [27] with regard to x(a) and y(s)
are made. We can also assume th at L has a finite number of angular
points. Let M be an angular point on L and suppose th at when describing a
circuit round L in the counter-clockwise direotion, the direction of the tangent
a t M revolves through an angle ji0, where —1 < 0 < + 1 . From this it is
easy to see th at on the right-hand side of formula (156) we have (1 — 0) ni
instead of ni, and th at the expression (163) at the point M should be repla­
ced by the following expression (I. I. Privaloff, Dokl. Akad. Nauk, SSSR.,
X X III, No. 9, 1939)

± "Mr^ ®(f) + 2^r I t = t dT ’


L

where the positive or negative signs should be taken simultaneously.


If we denote by F ^ f) and Fe{() the limiting values of the function (161)
on the contour, defined within and outside L, then the theorem which we
proved above can be rewritten in the form:

r.w --y « < e+ -sr j T ^ * - <>«•>


L

This theorem can be proved similarly for an open contour. Let us consider
a finite interval (a, b) of the real axis:
b
(170)
a

If m(t) is identically unity then, instead of formula (168), we can write:

b
1 f di 1 , b —z /17n
~2ni J T ^ T = ~ 2 n ilt>g ~ a ^ z ’
a
(1?1)
where we must take those values of the logarithm which vanish when z = oo.
If $ lies in the interval (o, 6), then instead of formula (156) we have:

b
1 C dl 1 6 —1
2ni J t — £ 2 ni f —a ’
a
28] INTEGRALS OP CAUCHY'S TYPE 117

where real values of the logarithm are taken. Repeating the previous arguments
word for word we obtain:

a a

The function (171) has different limits when z tends to £ from above or below
(a, b), viz..
b —z b -£
log log ± ni ,
z — a z-f £ —a
where the positive sign refers to the case when z tends to £ from above, i.e. to
values with a positive imaginary part, and the negative sign to the case when
z tends to £ from below. When integrating from a to b the upper half-plane
lies to the left and therefore, the tendency of z to £ from above is analogous
with the tendency of z to £ from inside a closed curve. Similarly, the tendency
from below is analogous with the tendency from outside a closed curve.
Denoting by .?,•(£) and Fe(£) the limiting values of the function (170) when
the tendency of z to £ is from above or below, we obtain formulae which are
analogous with the formulae (169):

(172)

If co{t) satisfies the conditions given at the end of [27] in the interval and if near
the ends of the interval it has the form (160), then for points z near the ends
of the interval, the following statements apply:
1. If y = 0, then

J,w - ± T i f 1<,8 7 i r + f'"<2>-


where the (+ ) sign refers to the case when c = a and the ( —) sign to the
case when c = 6, whilst F(z) is a bounded function which has a definite limit
when z —>- c. For log (z — c) we can take any branch, which is single-valued in
the neighbourhood of z = c in a plane with the cut (a, b).
2. If y = yl + yt i ^ 0 then
e +y*f co* (c)
F (z) = ± Fo (z) ,
2£ sinyzz (z - c)Y
where the signs are chosen in the way described above and (z — c)v denotes a
single-valued branch in the neighbourhood of the point z = c in the plane with
the cut (a, b); the value of (z — c)Y on the upper (left) edge of the cut is equal
to that value of (< — c)y, which enters the formula (160). Furthermore F a(z)
118 THB BASIS OP TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [28

has the following properties: if y, = 0 then F 0(z) is bounded and has a definite
limit when z -* c; if y, > 0, then

F0(2)| < -c\Y»


where c and y0 are constants and y# < 1. Using the concept of Leber’s integral,
the values of integrals of Cauchy’s type can be investigated for any integrable
function co(t) and for a great variety of contours (see Privaloff, Integrals of
Cauchy's type, 1918).
Let us notice one particular case. If a>(r) are the limiting values on I of a
function, which is regular within the closed contour L and which is continuous
up to L, while cu(t ) satisfies a Lipschitz condition, then F[(£) = a>(£), and
the first of the formulae (169) shows that co(t) is the solution of a homo­
geneous integral equation of the second kind:

(O(T)
>(£) = dr. (173)
m Jf -t

L et L be a simple closed contour, as described above. The principal value of


th e integral
co(t)
j_r. dr (174)
2m J

transforms any function a>(r) which is given on L and which satisfies a Lipschitz
condition into another function <ot(£), which is also defined on L and which also
satisfies a Lipschitz condition. In other words, the integral (174) is an operator
for the function <d( t ). To the function thus obtained we can again apply an
operator with Cauchy’s nucleus. In this case the following formula applies:

In other words, as a result of the two transformations by Cauchy’s nucleus,


we obtain the initial function multiplied by 1/4. To prove (175) we rewrite the
first of the formula (169) in the following form:

— (176)
2ni Jf -r

The right-hand side gives the result of the linear transformation of the function
co(r) by Cauchy’s nucleus. This operation can again be applied to the right-
hand side:

_1 (f)-
d |, (177)
2m f- n
28] INTEGRALS OF CAUCHY’S TYPE 119

where rj lies on L and the integral is in the principal value sense as before.
Owing to the fact th at F /(£) gives the limiting values on I of a function which
is regular within L, from (173) we have the following:
1 f F,(£)
2m ) f —T] - y */(«»>•
L
On the other hand, from (176):

1
2ni ..
L
and, finally, the integral (177) appears to be equal to 0 ( 77) : *-e- we obtain
the formula (175).
CHAPTER II

CONFORMAL TRANSFORMATION
AND THE
TWO-DIMENSIONAL FIELD

29. Conformal transformation. In this chapter we shall consider


some applications of the theory of functions of a complex variable
to problems of two-dimensional hydrodynamics, electrostatics and the
theory of elasticity. Conformal transformation plays an important part
in these applications and we shall therefore begin this chapter with a
detailed account of conformal transformation. We explained the basic
characteristics of the transformation of a regular function in [3] and
later in [22]. We considered in greater detail transformations at points
where the derivative did not vanish and at other points, where it was
zero. At points of the first kind all angles remain unchanged, while
at points of the second kind all angles are magnified, as we described
in [23]. Let
w = f(z) (1 )
be a regular function which conformally transforms the domain B into
the domain Bv I f f'(z) does not vanish anywhere in the domain B, then
the domain BLhas no branch-points, but it can still have several sheets,
i.e. it can overlap itself. Consider in the domain B a curve I, a function
tp(s) given on this curve and a line integral
f <P($) ds,
1
where ds is an element of the arc of the curve I. As a result of the
transformation (1 ) the curve I is transformed into another curve
in the domain Bv and an element of arc of the new curve is ex­
pressed by the product dsx = [f'(z) | ds since \f'(z) | gives the change
in linear dimensions [3].
Introducing the function
2 = F (w), ( 2)
120
29] C0NF0RMAJ. TRANSFORMATION 121

which is the inverse of (1 ), we evidently have F'(w) = 1[f'(z) and,


consequently, we can write ds = | F'(w) | d^. The integral obtained
after the transformation can be written in the form:
f V (*) ds = J <p {Sj) | F' (w) | dsr (3v
i i.
Similarly, bearing in mind that | f'(z) |2 is multiplied by the change
in surface area at the given point we obtain the following formula
for the conformal transformation of the double integral:
jT 9>(*) d<r = J J <px (to) | F' (to) |2 dff1( (4 )
B B,

and the following formula applies to an element of area:


d<*i = \ f (2) |2 dcr. (5)
When the real and imaginary parts are separated
w = f(z) = u (x ,y ) + iv(x,y), (6 )
it can easily be seen that | f'(z) |2 is equal to the functional determinant
of the functions u(x, y) and v(x, y) of the variables x and y. In fact,
this functional determinant is expressed by the following formula:
D (m, v ) du dv du dv
D (x, y) 3x 3y dy 9x ’
or, from the Cauchy-Riemann equations, by the formula:
D ( u , v ) _( du V f dv y
D (x, y) ~ I, 3x J 1 3x J ’
and this is, in fact, the square of the modulus of the derivative
du . . dv
|/'( z ) | 2 = dx ' * 3x
Consider in the z = x -{- iy plane two families of lines:
u {x ,y) = Cj; v ( x ,y ) = C 2, (7)
where Ct and C2 are arbitrary constants. In the w = u + iv plane they
correspond to the lines u = CX and v = C2, which are parallel to the
axes of coordinates; therefore the lines (7) are obtained from the net
of straight lines parallel to the axes, as a result of the transformation
(2 ). One of the consequences of this is that the lines (7), which belong
to different families, are orthogonal except at the points where f'(z)
vanishes. Conversely, if we take the equations
u = u(x, y)\ v = v(x,y)
122 CONFOHHAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [29
and assume that x = Cx or y = C2 on the right-hand side of these
equations, where C1 and C2 are arbitrary constants, we obtain a net
of lines in the w = u + iv plane, consistsing of two orthogonal
families of lines. This net is obtained as a result of the transformation
effected by the function (1 ) from a net of straight lines, parallel to the
axes of coordinates in the z-plane. These two nets of lines which are of
great importance in what follows, are usually known as isothermic

nets. We shall explain the meaning of this - term. The real part
u(x, y) (or the imaginary part) of a regular function should satisfy the
Laplace equation [2 ]:
d2u (x , y) 82u (x , V) _ n
8x2 ^ dy2

This equation is satisfied by the temperature of an established heat


current [II, 117] and we suppose that this case is two-dimensional, i.e.
the temperature u is independent of one of the coordinates. When
interpreting the function u(x, y ) as the temperature of an established
heat current, the lines of the first family (7) will be the isothermic lines
and this is how the name “isothermic net” is derived. In the case under
consideration, lines of the second family (7), orthogonal with the first
family of lines, serve as vectorial lines for the vectors which we con­
sidered in [II, 117] and which we called “the vectors of the heat current” .
Two lines u{x, y) = u 0 and u(x, y) = ux are transformed by (1)
into the lines u = u0 and u = ux which are parallel to the u = 0 axis
29] CONFORMAL TRANSFORMATION 123

and part of the domain B bounded by the above lines, is transformed


into part of a strip, bounded by the lines parallel to the u = 0 axis.
The curved rectangle, bounded by four lines of the isothermic net,
is transformed by (1 ) into a rectangle, bounded by straight lines,
parallel to the axes (Fig. 26)
u = u0\ u = u1\ v = v0; v = vv

Let us make one further addition to the fundamental principles of


conformal transformation before considering any examples. We saw
that the transformation of a regular function f(z) conserved the angles
both in magnitude and in sign at points where the derivative did not
vanish. Sometimes the transformation of a plane is considered where
the magnitude of the angles is conserved but their sign is reversed.
This transformation is sometimes called a conformal transformation of
the second class. For example, the symmetrical transformation of the
real axis is, evidently, a conformal transformation of the second
class (Fig. 27). This transformation is given by the formula w = z.
Generally speaking, if f(z) is a regular function in the domain B then
the formula
w = f(z) (8 )
gives a conformal transformation of the second class, defined in the
domain B', which with, the domain B is symmetrical about the
real axis. In fact, the transition from z to z transforms B ' into B while
124 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [30

conserving the magnitude of angles but not their sign. The subsequent
transition from z to /(z), in accordance with formula (8 ), will neither
alter the magnitude of angles nor their sign; thus in the final trans­
formation from z to w we have a conformal transformation of the
second class.

30. Linear transformation. As the first example of conformal


transformation we shall consider the simplest linear function
w = az + 6 (a # 0), (9)
from where
1 b
z=—
a
w -----
a
.
This function transforms the whole plane, including the point at
infinity, into itself, the point at infinity remaining in its former
position. In particular, when a = 1 , the function w = z + 6 gives
the parallel transition of the plane along a vector corresponding to
the complex number 6 . In the case 6 = 0 and a = e‘v (where ip is a
real number), the number ip must be added to the amplitude of z,
and so the transformation w = eiv z, will involve the rotation of the
plane about the origin by an angle ip. In general, the movement of a
plane as a whole is obtained by a combined rotation and parallel
transition:
w = e ‘v z+ 6. ( 10 )
If a = e ‘v # 1, i.e. the transition is not purely parallel, the
coordinates of the stationary point of the transformation i.e. of the
point which remains in its former position during the transformation,
can easily be determined by the formula (10). The coordinates of this
point are determined from the equation
z0 = e ‘v z0 + 6 , whence z„ = .

It can easily be shown that the transformation (10) can be written


in the form
w — z0 = e 'v (z — Zq),
i.e. in general the transformation ( 10 ) can be regarded as the rotation
of the plane about the point z0 through an angle ip. Note that there
will be a second stationary point at infinity for the transformation (10 ).
We shall now consider the case when the modulus of the coefficient
a in the linear transformation (9) is not unity. Introducing the modulus
31] BILINEAR TRANSFORMATION 125
and amplitude of a, we shall consider the transformation when 6 = 0:

'w = pe‘v z
In this case the length of the vector from the origin to the point z
must be multiplied by q and the plane rotated about the origin by
an angle y>. This transformation is known as an identity transfor­
mation with the origin as the centre of similarity and with a coeffi­
cient of similarity q.
We shall now consider the general case of the linear transformation
(9 ) when a ^ 1 . Introducing the stationary point of the transformation:
, t b
z0 - az0 + h, i.e. Zq = y-—^ ,

we can rewrite formula (9), in the form:


w — z0 = a (z — z0,)
and we evidently have here an identity transformation with the centre
not at the origin but at a point z0. We leave the reader to show that
the isothermic net will, in this case, consist of two families of parallel
straight lines; this is obvious and purely geometrical.

31. Bilinear transformation. A bilinear transformation is a trans­


formation which can be expressed by a quotient of two linear functions
+b
( 11 )
+d
Here ad — be ^ 0 since otherwise the fraction in the equation ( 11 )
could be simplified and would simply be a constant. Solving the
equation ( 11 ) with respect to z we obtain a formula for the transfor­
mation of the inverse of (11 ), which will also be a bilinear trans­
formation
— d w -\- b
z = ------- —.
cm) — a

Every point in the z-plane has a corresponding point in the m-plane


and vice versa, i.e. the transformation (11 ) transforms the whole
plane, including the point at infinity, into itself.
If in formula (11 ) c = 0 , then the transformation is simply a linear
transformation. Otherwise the point z = °° is transformed into the
point a/c and the point z — —dje gives the point w = i.e. in
general the point at infinity does not remain stationary during the
bilinear transformation.
126 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [31

We shall now give one characteristic property of the bilinear


transformation, viz. that it transforms a circle into a circle, where
by a “ circle” we shall now and in future understand not only a circle
in the usual sense but also a straight line. This property is quite
obvious for a linear transformation where the plane moves as a whole,
or in an identity transformation; a linear transformation transforms
a straight line into a straight line and a circle into a circle, in the
usual sense of the word. Before proving this property of the bilinear
transformation we shall describe it in a slightly different way. Let us
suppose that c ^ 0 and divide the numerator by the denominator;
formula (11 ) can then be written in the form:
i f i a , . be — ad
w = e -\----- — r
, d
where e = —
c
and '/ = ----=
—.
c1
g+ T
Our transformation thus involves a parallel transition w L = z -\-d /c ,
a transformation of the form w 2 = f j w 1 and another parallel transition
w = w 2 + e . It is therefore sufficient to consider the simple trans­
formation:
w= ± (13)

and prove that it transforms a circle into a circle. An equation of a


circle can be written as follows:
A (*= + y 2) + 2B x + 2C y + D = 0,
where A = 0 for a straight line. The equation can then be written
as follows:
A z z + d z + 8 z + D = 0 , where d = B — i C , (14)
where the line above the letters shows that complex conjugate numbers
were taken. We now suppose that we have a circle I in the 2-plane.
To obtain the equation of the transformed circle in the ui-plane it is
sufficient to determine z from the equation (13) and substitute the
expression obtained in the equation (14). We then obtain a curve l 2
in the ui-plane which is given by the equation:
A yy -f- 8 y w + dyw + Dww — 0.
This equation is of the same type as the equation (14) i.e. it corres­
ponds to a circle (or to a straight line). Thus e v e r y t r a n s f o r m a t i o n
o f th e t y p e (11 ) t r a n s f o r m s a c ir c le i n t o a c ir c le ( a s t r a i g h t l i n e i s a
c ir c l e w h i c h p a s s e s th r o u g h th e p o i n t a t i n f i n i t y ) .
31] BILINEAR TRANSFORMATION 127

Suppose that the transformation (11) transforms a circle I into a


circle lx and that both circles are circles in the usual sense of the
word. Bearing in mind what was said in [2 2 ] we can maintain that if
the completion of a circuit in the positive direction round I corresponds
to the completion of a circuit in the positive direction round lv then
the interior of I is transformed into the interior of lv However, if
the circuits round I and lx are described in opposite directions, then
the interior of I is transformed into the exterior of lx and vice versa.
If one of the above circles is a straight line or, if both are straight
lines, then in order to determine the domains of the plane which are
transformed into each other, we have to take the corresponding
direction of circuits along both lines, when parts of the plane which
lie to one side of the moving observer, for example, to the left, are
transformed into each other.
Let us consider two points Ax and A2 which are symmetrical with
respect to the circle I. Suppose that after the transformation they
become the points B1 and B2. We shall show that these points are
also symmetrical with respect to the transformed circle lv In fact,
a family of circles through the points A x and A 2 will, as we said
in [24], consist of circles orthogonal to I. After the transformation
we evidently obtain a family of circles through the points Bx and B2
and, as a result of conformity of circles belonging to a family, these
circles will be orthogonal with the circle lx; this, as we know, is one
of the characteristics of symmetry. Thus if the circle I is transformed
by (11 ) into the circle lx then points, symmetrical with respect to the
circle I, are transformed into points, symmetrical with respect to
the circle lv Notice that the point at infinity is symmetrical with the
centre of the circle. In the case under consideration a family of circles
which passes through these two points is transformed into a family of
lines which passes through the centre of this circle and these lines
are, obviously, orthogonal to the circle itself.
If a and c are both non-zero then the transformation (11) can be
written in the following form:

<1 6 >

The numbers a and /3 have a simple geometric meaning, viz. the


point z = a is transformed into the origin w = 0 and the point z =
into the point at infinity.
Let us consider a family of concentric circles, centre the origin,
in the w-plane. The equation of these circles is \ w \ = 0 and the
128 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [31
points mentioned above, viz. w = 0 and w = are symmetrical
with respect to these circles. It follows that these circles correspond
to circles in the z-plane for which the points z — a and z = /? are
symmetrical; the equation of this family of circles has the form:
z —a
z- P
= c. (16)

where C is an arbitrary constant. Thus th e e q u a t io n (16) co rresp o n d s


/? a r e s y m ­
t o a f a m i l y o f c ir c l e s w i t h r e s p e c t t o w h i c h th e p o i n t s a a n d
m e tr ic a l (Fig. 28). The straight line which is perpendicular to the

line from a to /? a t its midpoint also belongs to this family. Let us now
consider a family of straight lines in the w-plane which passes through
the origin or, in other words, a family of circles which passes through
the points w — 0 and w = °°. This family of circles is evidently
given by the equation arg w = C . It corresponds to a family of
circles in the z-plane which passes through the points a and and
the equation of this family is (since the amplitude of A; is a constant):

= (H )

Hence th e e q u a t io n (17) d e s c r i b e s a f a m i l y o f c ir c l e s i n th e z - p l a n e
31] BILINEAR TRANSFORMATION 129
/?. The circles of the family (17)
w h ic h p a s s e s th r o u g h th e p o i n t s a a n d
evidently cut the circles of the family (16) at right angles (Fig. 28).
Let us now define the isothermic net in the z-plane. It corresponds
to two families of straight lines in the w-plane which are parallel to
the axes. Each family can be regarded as a family of circles which
touch at infinity. In the z-plane each family corresponds to a fa m ily
of circles which touch at the point z = — d j c . Hence th e r e q u i r e d
i s o th e r m ic n e t c o n sists o f tw o fa m ilie s of c ir c le s , th e c ir c l e s o f e a c h
f a m i l y to u c h i n g a t th e p o i n t z= — d j c ; c ir c l e s b e lo n g in g t o th e tw o
d iffe r e n t f a m ilie s in te r se c t a t th is p o in t a t r ig h t a n g le s (Fig. 29).

The accurate definition of one family, and consequently, also of


the other family, depends on the values of the complex coefficient
of transformation (11 ).
The transformation (11) contains three arbitrary complex parameters,
viz. the relationship of three of the coefficients a , b , c, d , to the fourth.
Hence the transformation (11 ) can be defined if a corresponding
number ofauxiliary conditions is given. We can,for example,
assume that three given points z v z2, z3 in the z-plane should be
transformed into three given points w v w 2, w 3 in the w-plane. It is
easy to write down the bilinear transformation which takes these
conditions into account. It has the form:
W — U>, w3 — w2 _ Z — Z, _ Z j - I j
( 18 )
w — w3 w3 — wt z —z2 z3 - Zl
130 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [31
For, by solving this equation with respect to w we obtain a bilinear
transformation of the type (11). Also, when substituting z = z1 and
w = wv we obtain zero on both sides of formula (18). When substi­
tuting the second pair of points z = z3 and w = w3 we obtain unity
on both sides of the equation, and when substituting the third pair
of points, we obtain infinity on both sides of the equation. It can
be seen from this that the bilinear transformation, as given by
formula (18), does in fact, satisfy the required conditions. It is
also easy to show that these conditions define the bilinear trans­
formation uniquely. Obviously the constructed transformation trans­
forms a circle, defined by the three points zv z2, z3 into a circle,
defined by the three points wv w2, w3. If both sets of points are
taken on the same circle, then the bilinear transformation transforms
that circle into the same circle. If, in addition, the sequence of the
points zk on this circle gives the same direction in which the circuit
is to be described as the sequence of the points wk, then the interior
of the constructed circle is transformed into itself.
Consider, for example, the upper half-plane bounded by the real
axis where interior points are defined by the condition that the
coefficients of the imaginary parts of the coordinates are positive.
In the case under consideration any transformation which transforms
the upper half-plane into itself should also transform the real axis
into itself, i.e. real z’s should have corresponding real w’a; consequ­
ently we can take it that all four coefficients are real in formula (11 ).
However, this is not sufficient; in addition it is also necessary that
w should increase as z increases along the positive part of the real
axis. Otherwise the upper half plane z will be transformed into the
lower half-plane w.
Substituting z — x iy in formula ( 11 ) we obtain:
w _ (as + b) + iay
(cx + d) + icy ’

or, separating the real and imaginary parts:

I _ (ax + b) (cx + d) + ocj/* , . (ad —be)


u-riv— (ca: + d)2 + ct y* ^ ( c x + d y + c*y*

It follows that when y > 0 the coefficient of the imaginary part


of w will also be positive, provided the following condition is satisfied:

ad — bc> 0 . (19)
31] BILINEAR TRANSFORMATION 131
Hence the general form of the bilinear transformation which trans­
forms the upper half-plane into itself is (11 ), where the real coefficients
can be arbitrary provided they satisfy the condition (19).
The transformation of a unit circle into itself takes place similarly,
i.e. a unit circle is a circle with unit radius and centre the origin,
the equation of which can be written as follows: | z \ < 1 . To begin
with we shall explain certain properties of points symmetrical with
respect to the circumference G of this circle.
Let Axand A2be two such points
and M a point on the circumference ~"'\W_________
C. We have OA1 • OA2 - OM2, f / f \
which can be written as a propor- / /
tion (Fig. 30): / ]
0 T X _ OM \ J
OM ~ OA~2 ‘ \ /
It follows that the triangles ^ ------- ^
OAtM and OA2M, which have a Fig. 30
common angle A fiM and propor­
tional sides adjacent to this angle,
are similar; from this similarity we obtain the following proportion:
M A X = OAx
( 20 )
M A t ~ OM

Denote by a the complex coordinate of Ax and let a = q elv.


For the symmetrical point A2 we must have /? = el,p[q or, in
other words, the complex coordinate of this symmetrical point can
be expressed by the fraction fi = 1fa. We shall construct a bilinear
transformation so as to transform the unit circle into itself and the
point a into the origin. This transformation should transform the
symmetrical point fi into the point at infinity, i.e. it should have
the form:
w = kT= J (21)
or substituting § — 1/a,
w = k , ( 22 )

where A; is a factor, the form of which is still to be determined


from the condition that the right-hand side of the formula (21 ) for
any z on the circumference C should have a modulus equal to
132 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [3 1

unity, i.e.
1*1yZ = 1 when M= 1-
But from (2 0 ):
|2 - g | _ [a|
\z-f)\ 1 ’
hence | ka \ = 1 . We can see that the product ha must have a unit
modulus, i.e. it must have the form ha = e'v’, where ip can take
every real (value. Thus returning to the formula (22 ) we obtain the
following formula for the required transformation:
z — a
w = e'v a z - 1’
(23)

in which the point a and the real parameter ip can have any value we
please. In particular when a = 0 , i.e. when the origin is transformed into
itself, we have the simple transformation w = e,(v+J,) z, i.e. the
unit circle revolves about the origin by an angle (ip + 7t). The general
transformation (23) can be separated into two parts, viz. into the
transformation
z —a
W : az — 1 1 (24)

which transforms a unit circle into itself and which transforms the
point a into the origin, and, subsequently, into the rotation about
the origin by an angle ip.
We can also construct a countless number of transformations which
transform a circle K 2 into another circle K2. For this purpose it is
sufficient to construct one of these transformations to transform K 1
into K2 and, subsequently, to apply any bilinear transformation to
the result so obtained, in order to transform the circle K 2 into itself.
It is important to note that two bilinear transformations applied in
succession also produce a bilinear transformation. In fact, suppose
that we have a bilinear transformation of the variable z to the
variable w of the form ( 11 ) and, subsequently, the following bilinear
transformation
a, w + bt
ct w-{-dl
(25)

of the variable w to the variable wv Substituting the expression


(11 ) in the above formula we obtain, after elementary transpositions,
the final bilinear transformation of the variable z to the variable wx:
(a , a + 6, c) z + (Oj b + bx d)
(c t a + dl c)z + (Cj 6 + dl d)
31] BILINEAR TRANSFORMATION 133

This is usually known as the product of the bilinear transformations


(11) and (25) and, generally speaking, this product depends on the
order of factors, i.e. on the order in which the bilinear transformations
(11) and (25) are carried out.
Let us now return to the unit circle and the upper half-plane; we
shall construct a bilinear transformation to transform a half-plane
into a unit circle. We take for this purpose the following transforma­
tion:
®= - S r - (2fi)
It is easy to see that the point z = i in the upper half-plane will
be transformed into the origin and that real values of z correspond
to values of w with modulus equal to unity. In fact

where the numerator and the denominator of the fraction are respect­
ively equal to the distances from the point z to the points i and (—i);
when the point z lies on the real axis then these distances are equal
and, consequently [ w | = 1 . If an arbitrary bilinear transformation
which transforms a unit circle into itself, is applied to the variable w,
we obtain the general form of the transformation which transforms
the upper half-plane into a unit circle.
In conclusion we shall prove the principle of symmetry in the general
form as formulated in [24]. Let the function f[z) be regular on one
side of an arc A B of the circle C, let it be continuous as far aB the
arc A B and let it transform it into another arc A ^ ^ of the circle Cv
Subjecting 2 to a bilinear transformation which transforms C into
the real axis we obtain:
02 4- b
Zl~ cz + d '
and performing a similar bilinear transformation on the function
itself we transform the circle C1 into the real axis. We thus obtain
a new function fi(Zj) of a new independent variable zx:
r \ _ <*7(2) +&
i W - c'f{z)+dr
This new function f^fzf) is regular on one side of the real axis and
continuous as far as the line which it transforms into part of the
real axis. In accordance with our earlier definition of the principle
of symmetry given in [24] this function can be analytically continued
beyond the above line; at points symmetrical with respect to the
134 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD L31

real axis its values are also symmetrical with respect to the real
axis. Bearing in mind that the two bilinear transformations mentioned
above transform symmetrical points into other symmetrical points,
we can maintain that the initial function /(z) can be analytically
continued beyond the arc A B of the circle C, and points symmetrical
with respect to this circle are transformed into points, symmetrical
with respect to the circle Cv
Bilinear transformations, as we shall see later, are of great impor­
tance in the theory of a complex variable. They are often used
in the same way as the transformation of coordinates is used in
analytical geometry viz. before starting to solve a problem, the
plane of the complex variable in the problem, is subjeoted to the
bilinear transformation so as to obtain the simplest possible conditions.
Thus, for example, by using the bilinear transformation we reduced
the general case of the principle of symmetry to the particular case
considered above.
Let us call the reflection in a circle or in a straight line the trans­
formation of a plane where every point A is transformed into a point
Av symmetrical with it with respect to G. Let z be the complex
coordinate of A and w the complex coordinate of Av Let us assume
that C is a circle, centre B(z = a) and radius R. The vectors BA
and B A X should have the same amplitudes and the product of their
lengths should be equal to R 2. It is easy to see that this leads to the
following formula which expresses w in terms of z:

”> - » = T ^ 5 ’ <27>
i.e. the reflection of the circle is expressed by a bilinear function of z:
_ oz + (fi2 — aa)
z —a
and, consequently, it is a conformal transformation of the second
class. Let us now consider the reflection in a straight line. Assume
that the straight line passes through the origin and makes an angle
ip with the positive direction of the real axis (Fig. 31). In this case
the point z is transformed into the point w which has the same
modulus | w | = | z | and amplitude arg w = 2y> — arg z, i.e. the
transformation can, in this case, be written as:
w = ei2vz, (28)
where it is expressed as a simple linear function of z. It is clear that the
same result is obtained when reflection takes place in any straight line.
32] TH E FUNCTION to = ** 135

If we produce two successive reflections in two different circles


or straight lines, we obtain a bilinear transformation. Let us consider
in greater detail the case when two successive reflections are produced
in two intersecting straight lines. It can always be assumed that
the point of intersection lies at the origin. Let y>x and ip2 be the
angles made by these straight lines with the positive direction of
the real axis. When producing these successive reflections we move
from the point z to the point wx and from the point wx to the point w,
according to the formulae:
wx - - eilVl z; w = el2v*wx.
Substituting the expression for wxon the right-hand side of the second
formula we obtain the final transformation for z to w in the form
w=
this involves the rotation about the origin by an angle 2(y>2 — y>j), i.e.
two successive reflections in intersecting straight lines involve the
rotation of the plane about the point of intersection by an angle
equal to twice the angle between these straight lines. I t is also easy
to show that two successive reflections in parallel straight lines give
parallel transformation of the plane.

32. The function w = z2. Using a different notation we have already


investigated the function
w = z2 (29)
136 CONFORMAL TRANSFORMATION AND TWO-DIMENSIONAL FIELD [32

and we saw that it transforms the 2-plane into a two-sheeted Riemann


surface in the id-plane with branch-points of the first order at id = 0
and w = oo. We shall now establish the form of the isothermic nets
in the z- and iD-planes. On separating the real and imaginary parts we
obtain:
w = u(x, y) + iv(x, y) = (x + iy)2 = (x2 — y2) + i2xy.
The isothermic net in the 2-plane consists of two families of rect­
angular hyperbolae (Fig. 32):
x 2 — y2 = Cy, 2xy = Cv

Let us now consider the isothermic net in the w-plane. In the


formulae:
u = x2 — y2; v = 2xy, x = C y
let us assume that x = C y and eliminate y, then if we assume that
y = C 2 and eliminate x, we obtain two families of parabolae (Fig. 33):
v2 = 4C\ (Cl — u); v2 = 4C2(Cl + u),
which have been obtained as a result of the transformation of the
straight lines x = C y and y = C 2 from the 2-plane.
We can evidently regard the isothermic net formed by these
parabolae as an isothermic net in the tc-plane of the function id = |/i
which is the inverse of the function (29).
Let use consider in Fig. 32 any rectangular hyperbola represented
by the dotted line for which OX is the major axis. Its equation is x2 —
— y2 = Cy, where C y is a positive constant. Let us consider its right
branch. If in the equation x2 — y2 = C we alter C from C y to ( + °°), we
obtain one of the hyperbolae which are shown by the dotted lines; its
33] TH E FUNCTION u = k (z + l/»)/2 137

right branch will lie further to the right than the right branch of the
hyperbola x 2 — y2 = C^; it follows that the function (29) conformally
transforms that part of the 3-plane within the right branch of the
hyperbola into the half-plane u > Cx of the w-plane. Similarly the
function (29) conformally transforms part of the z-plane within the left
branch of the hyperbola x2 — y2 = Cx into the half-plane u > Cv
Let us now consider in Fig. 33 any parabola shown by a dotted
line. Its equation has the form v2 = 4C\ ( C \ -f- u) and there will be
a corresponding straight line y — C 2 in the 2-plane, where the con­
stant C. j can be regarded as positive since only C \ enters the equation
of the parabola. If in the equation v2 = 4G2 ( C 2 + u), C is altered
from C \ to (+ °°) then parabola represented by dotted lines will
be obtained which lies further to the left than the parabolae
v2 = C \ ( C \ + u)\ it follows directly from what was said above that
the function z = ]fw conformally transforms that part of the wi-plane
outside the parabola v2 = C \ ( C 2 + u) into the half-plane y > C z of
the 2-plane.

33. The function w = k (z + l/z)/2 . Let us consider the trans­


formation of a plane by the function

w = 4 (z + t - ) ' (30)
where k is a given positive number. Let us consider the form into
which the net of polar coordinates in the 2-plane will be transformed,
i.e. consider the form into which the circles | z \ = q, centre the origin,
and the family of straight lines arg z = tp, which passes through the
origin will be transformed. Substituting in formula (30) z — q e1’’
and separating the real and imaginary parts we obtain the equations:

U= (e + ——
) cos 7>; 0 = ——-jsinp. (31)
Consider the circle g = g0. On eliminating <p from the equation
(31) the following equation is obtained:

i.e. in the tu-plane the circle is transformed into an ellipse, the semi­
axes of which are
138 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [3 3

where we take the positive value for b, since the difference can be
either positive or negative. When g = g0, equation (31) gives
the parametric equation for this ellipse. In the case of a unit
circle, when q — 1, the equation (31) gives u = k cos <p and v = 0,
i.e. the ellipse degenerates into a line (—k, -\-k) on the real axis described
twice, or, as we shall say in future, it degenerates into a double line.
When g decreases from unity to zero the ellipses grow indefinitely
until they cover the whole plane, i.e. the whole ui-plane with a cut
(—k, -f k) corresponds to the interior of the unit circle. Similarly,
when g increases from unity to infinity we also obtain indefinitely
growing ellipses, i.e. the whole w-plane with the cut (—k, -ffc)
corresponds to that part of the 2-plane outside the unit circle. The whole
2 -plane is transformed into a two-sheeted Riemann surface in the
ic-plane with branch-points at w = —k and w = -j-fc. It follows that
the function, which is the inverse of (30):
w ± f w2 —k2 (30,)
k
is two-valued and has the same branch-points. Let us investigate
the ellipses (31) more closely. The foci of these ellipses lie on the
real axis and their abscissae c are determined, as usual, by means of
the semi-axes a and b, according to the formula: c = i j / a 2 — b2.
In this case we have:

i.e. for every value p0 the foci lie on the ends of the line (—k, -\-k)
or, in other words, the ellipses (32) have coinciding foci.
Let us now consider the straight lines g = 9>0underthe transformation.
Eliminating the variable g from the equations (31) we have:
___ If!________ * ____ 1 (3 Q\
fc* cos2q>0 k2 sin2tp0 ’ ' '

i.e. we obtain a family of hyperbolae with semi-axes a = k \ cos <p0 |


and 6 = k | sin <p0 \. We shall show that the foci of these hyper­
bolae coincide with the foci of the above ellipses. We know
that the foci of the hyperbolae (33) lie on the real axis and that the
abscissae of the foci are found from the formula c = ± |fo 2 + b2.
In this case c = -±_k, i.e. the ellipses and the hyperbolae do, in
fact, have coinciding foci. The hyperbolae, which correspond to
33] t h e FUNCTION » = *(* + l/z)/2 139

the axes of coordinates in the z-plane

[9 = n and— ) ,

degenerate into the u = 0 axis and into the lines (—°°, —k) and
(k, + °°) on the real axis. Hence we can finally say that the net of
polar coordinates of the z-plane is transformed by (30) into a net of
ellipses and hyperbolae with foci at the points (Fig. 34).

F ig . 34

It is easy to construct a function for which the net of ellipses and


hyperbolae with coinciding foci serves as an isothermic net. In
order to do this we recall what we know already about an exponential
function [19]:
w = ez,
with a period of 271i. From the formula

w = exey'

it follows that the lines x = x 0 are transformed into circles, centre


the origin and radii e*°, whilst the lines y = yQare transformed into
the straight lines <p = y 0, which pass through the origin, i.e. the
function ez transforms the net of Cartesian coordinates in the z-plane
into a net of polar coordinates in the tc-plane.
Consider the function:
tUj = e‘z = eix e~y. (34)
140 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [34

with period 2tz. It follows from the above formula that the net of
Cartesian coordinates of this function will be transformed into a net
of polar coordinates for the lines y = y 0 will be transformed into
circles, whilst the lines x = x 0 will be transformed into straight lines.
Let us now consider the function

w = Jt {w i + ^ ;) = -) = * coBg- (35)

As a result of the transformation (34) the net of Cartesian coordinates


will be transformed into a net of polar coordinates and subsequently,
as a result of the transformation (35), the net of polar coordinates
will be transformed into a net of ellipses and hyperbolae with coincid­
ing foci as mentioned above. The application of these two trans­
formations viz. from z to w1 and from w1 to w gives, as the final
result, the transformation w = k cos z; thus the function k cos z
transforms the net of Cartesian coordinates into a net of ellipses
and hyperbolae with coinciding foci, i.e. this latter net is the
isothermic net for the function w = k cos z in the w-plane. If we
were considering the inverse function w = arc cos (z/fc), then the
net of ellipses and hyperbolae with coinciding foci would be the
isothermic net of this function in the z-plane.
Proceeding as in the previous paragraph, we find that one value
of the function (301) transforms the w-plane with the cut ( —k , +fc)
into the interior of a unit circle in the z-plane. The same function, for
any fixed q0, transforms the part of the plane outside the ellipse
(32) into the interior part of a circle, centre the origin and radius
g0, where g0 < 1 . If we take the other value of the function (30j)
we obtain part of the plane outside the above circle, provided p0> 1.
Similarly, one value of the function (30!) conformally transforms part
of the ui-plane between the two branches of the hyperbola (33), into
a corner of the z-plane, defined by the inequalities: g0 < arg z < jt —ip0,
where 0 < <js0 <
A detailed investigation of conformal transformations of curves of
the second order can be found in Privalloff’s book T h e I n tr o d u c tio n
O f T h e C o m p le x V a r ia b le I n to T h e T h e o r y O f F u n c tio n s .

34. The bi-angular figure and the strip. Let us consider the bi-angular
figure formed by two arcs of the circles C x and C 2 (Fig. 35). Let y>
be the angle of this bi-angular figure and ax and cq the coordinates
34] TH E BI-ANGULAR FIGURE AND TH E STRIP 141

of its vertices. On effecting the bilinear transformation

we transform the points ax and a 2 into w2 = 0 and = °°, so that


the arcs of this bi-angular figure are transformed into straight lines
which pass from the origin to infinity, the angle between these
straight lines being xp. If we subsequently effect the transformation

F ig . 35 F ig . 36

wi = Wilv, then this angle will be equal to n and the bi-angular figure
will be transformed into a half-plane. By multiplying w2 by the
factor e'*3, we can make this half-plane into the upper half-plane,
bounded by the real axis. Grouping together all these transfor­
mations we finally obtain a formula for the transformation of our
bi-angular figure into the upper half-plane:

<3 6 >

Here <p0 is a real number which depends on the position of our


bi-angular figure. Performing the bilinear transformation of w as
shown in [31] we can transform our bi-angular figure into a unit
circle.
Notice that we are considering a bi-angular figure confined inside
the contour formed by the arcs of two circles. Figure 35 shows part
of the plane outside the closed contour which can be regarded as a
bi-angular figure defined by the arcs of two circles. However, the
angle of this bi-angular figure will no longer be xp but (2n — xp).
We assumed above that the angle of this bi-angular figure is not
zero. Let us now consider the case when this angle is zero. Assume
that the two circles Cx and C2 touch from inside (Fig. 36). In this case
142 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [34
that part of the plane confined within the closed contour, is a bi-angular
figure with zero angle. Similarly, when two circles touch from outside
(Fig. 37), that part of the plane outside those circles also is a bi-angular
figure with a zero angle. If a is the coordinate of the point of contact,
then by performing the bilinear transformation

we transform the circles into parallel straight lines, while the bi-angular
figure itself is transformed into a strip, bounded by the two parallel
lines. If subsequently we perform an identity transformation and
also a parallel transition and rotation, i.e.
a linear transformation, we can always cause
the given strip to be bounded between two
given parallel lines, for example, between
the lines
y= 0 and y - 2n.

Let us now try to find a regular function


which would transform this strip into the
upper half-plane. We know that the function
w = ez transforms our strip into the 10-
plane with the cut (0 , + ° ° ) along the po­
sitive part of the real axis. Performing sub­
sequently the transformation |fw, we obtain the upper half-plane, i.e.
the final function which transforms our strip into the upper half­
plane will be
w - e*l2 .
It follows from above that the function ez transforms a strip,
bounded by the straight lines y = 0 and y = 71 into the upper half­
plane. Performing the bilinear transformation of the variable ez,
which transforms the upper half-plane into a unit circle [31], we
obtain the function
ez —i (37)
w - e* + i ’

which transforms the strip, bounded by the straight lines y — 0 and


y = n into a unit circle.
We shall consider in greater detail one case of the bi-angular figure,
viz. the upper half-circle, constructed on the line (—1 , + 1 ) of the
jg ] TH E FUNDAMENTAL THEOREM 143

real axis as its diameter. The function

transforms the vertices of this bi-angular figure, viz. z = —1 and


z== -{-1 , into the points w — 0 and w = + 00, while the diameter
and the semi-circle are transformed into two straight lines; the angle
between these lines is equal to twice the corresponding angle of the
semi-circle, i.e. it is equal to n.
In other words our two halves of straight lines form a single straight
line, viz. the real axis, as can easily be seen; when describing a circuit
round the contour in the counter-clockwise direction we move along
the real axis from —°° to + ° ° , i.e. the function (38) transforms
the semi-circle into the upper half-plane. On performing, in addition,
the bilinear transformation (26) we obtain the function
(z + l)* -i(z -l)2
(* + l)* + i(z-l)« ’
which transforms our semi-circle into a unit circle.

35. The fundamental theorem. We dealt above with the cases of


connected domains being conformally transformed into half-planes or
into unit circles, and we considered both bounded connected domains
(semicircle) and connected domains containing the point at infinity
(the outside of an ellipse, the outside of the bi-angular figure). Let
us now try to solve the problem of the transformation of any given
connected domain in the z-plane into a unit circle or into a half­
plane in the m-plane. In doing so we exclude two cases, viz. when the
given domain is the whole z-plane, including the point at infinity,
and when the given domain is the whole plane except for one point,
for example, the point at infinity. In all other cases a regular func­
tion w = /(z) exists in the given connected domain B, which trans­
forms this domain into the unit circle | w | < 1 . However, we can
subsequently transform this unit circle into itself by a bilinear trans­
formation and we then obtain a new conformal transformation of the
region B into a unit circle. Let us mark within the region a definite
point A and assume that during the transformation by our function
m = /(z), (39)
this point is transformed into the point a, which lies in the unit circle.
Similarly, performing a suitably chosen bilinear transformation of
144 CONFORMAlj TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [36

this circle, we can always translate the point a to the origin without
changing the unit circle [31]. This new transformation transforms
the point A into the origin. Also, by rotating the unit circle about
the origin we can cause the linear elements to remain stationary
during the transition of the point A to the origin i.e. f'(z) should
be positive at the point A. We can thus see that from one conformal
transformation of the region B into a unit circle we can construct
a countless number of similar transformations among which there
is one transformation which transforms any given point A in B,
into the centre of the unit circle, without altering the direction at that
point. It can be shown that with this additional condition the function
effecting the conformal transformation is defined uniquely, viz. the
following fundamental theorem in the theory of conformal trans­
formation applies:
R iemann’s T heorem. I f B is a given connected domain in the z-plane
and z0 is a point in B, then a function f(z) exists, (except for the two
cases mentioned above) which is regular in B ; this function transforms
the domain B into a unit circle so that z0 is transformed into the origin
and the value of the derivative f'(z) is positive.
We shall accept this theorem without proof. Note, that the function
mentioned in this theorem can only in exceptional cases and for the
simplest domains be expressed as an elementary function. The usual
proof of Riemann’s theorem establishes the existence of this function
but it is of little use for even the approximate construction of the
function. We shall deal later with the practical problem of the approxi­
mate construction of the function which performs the conformal trans­
formation.
Let us now make one important addition to Riemann’s theorem.
If the contour of the domain is a simple closed curve and has the
properties mentioned in [4], then the function f(z) is continuous up
to the contour of the domain B and transforms this contour into
the circumference of a unit circle.
The inverse function is, in this case, not only regular in the unit
circle but it is also continuous in the closed circle.
As we said above, the function which performs the conformal
transformation of the given domain B into a unit circle can only
be fully defined when the additional condition mentioned in the
above formulation of Riemann’s theorem is given. We can replace
this additional condition by another condition, but we must still
assume that the function effecting the conformal transformation is
35] TH E FUNDAMENTAL THEOREM 145

continuous up to the contour. In doing so we can use the bilinear


transformation of a unit circle into itself, so that three given points on
the contour of the domain B are transformed into three given points on
the circumference of the unit circle. In this case the function effecting
the conformal transformation is fully defined. The former condition
can also be formulated differently, viz. in the first place it is necessary
that the given point z0 in B should be transformed into the origin.
We can then rotate the unit circle about the origin so that a given
point on the contour of the domain B is transformed into a given
point on the circumference of the unit circle, whence it can be shown
that the function is fully defined.
Thus when fulfilling the conditions which guarantee the continuity
of the function performing the conformal transformation as far as
the contour of the domain B we can fully define this function by freely
choosing three points on the contour of the domain B which must correspond
to three points on the circumference of the unit circle, or by freely choosing
one interior point and one point on the contour of the domain B, which
must correspond to two similar points of the unit circle. If we have
two connected domain B1 and B, in the z-plane, then according
to Riemann's theorem, we have two regular functions:

w\ = fi(zi) and wi = fi izi)’ (40)


which transform these regions into a unit circle \w x \ < 1. Elimina­
ting, in theory, the variable wx in the above equations, we obtain
the regular function z2 = 9>(Zi)> which is transforms B L into B2.
In this case every point zx corresponds to a point z2, so that one
and the same wx can be defined by both zx and z2, according to (40).
We can thus achieve the conformal transformation of any two connec­
ted domains (except for the two cases mentioned above) into each other.
Simultaneously we can, of course, make the same additional condi­
tions which are mentioned above in connection with the transforma­
tion of a region into a circle.
Notice one important property of the function f(z) which transforms
a connected domain into a circle or into another connected domain.
We take it that our domains are one-sheeted domains or, more
strictly, that they can overlap but do not contain any branch-points.
Also the derivative f'(z) cannot vanish in the domain, since the vanish­
ing of a derivative indicates the formation of a branch-point in the
transformed domain [23]. We also notice that the functions log f'(z)
and ]ff'(z), can have no singularities in the course of the analytic
146 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [36
continuation within our connected domain B, and they will, there­
fore, be single-valued [18] and regular functions in this domain.
If we have not a connected domain but a domain with two boun­
daries in the 2-plane, e.g. an annulus confined between two closed
curves then it is obviously impossible to transform it conformally
into a connected domain, so that every point of the annulus should
correspond to a definite point of the connected domain and vice versa.
In the case of a multiply-connected domain there is one circum­
stance which makes this case different from the case of the simply-
connected domain, viz. not every two domain with the same number
of boundaries can be conformally transformed into each other. Thus,
for example, two annulae confined between concentric circles can be con­
formally transformed into each other only when the proportion of the
radii of the circles which confine these annulae, is the same in both cases.
In the case of a multiply-connected domain, one domain can still be
transformed into another domain of a definite type, viz. any n-bounded
domain can be transformed into a plane with n cuts which have the
appearance of parallel sections of straight lines, and some of these
cuts can degenerate into points.
Before considering the approximate methods for constructing the
functions effecting a conformal transformation we shall develop
an analytic expression for a function effecting the conformal trans­
formation of a unit circle or of the upper half-plane into a domain
confined by a broken line, i.e. a polygon. This formula frequently
occurs in various applications.

36. Christoffel’s formula. Let us suppose that we have a polygon


A v A2........An (Fig. 38) in the 2-plane and let the angles of this
polygon be a1n, a2ii, . . ann. Consider the function
z = f(t), (41)
which conformally transforms the upper half-plane t into a polygon.
We have to construct an analytic expression for this function.
Assume that the following points A * which lie on the real axis,
correspond to the vertices of this polygon
t = ak (Jc — 1 , 2 ,/. .,n),
and take it that all these points lie at a finite distance; this can
always be achieved by a bilinear transformation of the £-plane.
Let cq be the extreme point on the left and an the extreme
point on the right. We shall now consider the problem of
36] CHBISTOFFEL’S FORMULA 147

analytic continuation of the function f { t ) across the real axis. Take


a certain part a k a k+1 of the real axis which corresponds to the
side A k A k+1 of the polygon. In accordance with the principle of
symmetry we can analytically continue the function /(f) across the
line a k a k+1 and we obtain as a result a new polygon in the lower
half-plane which is produced by the reflection of the initial side in
the side A k A k+1. We can then analytically continue the new function
so obtained from the lower half-plane into the
upper half-plane, across the line a t oI+1 on the
real axis. As a result of the principle of symmetry
the new values of /(f) again give a polygon in
the upper half-plane; this polygon is obtained
from the second polygon by its reflection in that
side of the second polygon which corresponds to
the line a/ a t+1 of the real axis, etc. We can thus
see that it is possible to continue our function
/(f) without difficulty across the real axis and
in doing so the function transforms any plane
into a polygon, which is obtained from the ini­
tial polygon, after several reflections in sides
which correspond to those parts of the real axis across which the
analytic continuation was performed. Note that the side A n A 1
corresponds to a line on the real axis from a n to °° and from ° ° to a v
so that the point at infinity of the
As f-plane corresponds to a point on
the side A „ A x of the polygon.
The points a k will, in general,
be singularities of the function f ( t ) .
Let us investigate the character of
these singularities. Take a point a2
and describe a circuit around this
point by starting from and return­
ing to the upper half-plane. In
doing this we have to cross from
the upper half-plane to the lower
half-plane by cutting across the
line a 2a 2 and, subsequently, we
return from the lower half-plane to the upper half-plane across
the line a 2 a 3. It follows from what was said above that the values
of /(f) in the lower half-plane form a polygon A 1A 2A 3 . . . A „ , which
148 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [36
is obtained from the initial polygon as a result of the reflection in
the side A lA2; the return to the upper half-plane involves a reflec­
tion in the side A2A3 of this new polygon (Fig. 39).
Thus the above circuit round the point a2, corresponds to the
reflection in the straight lines A2AX and A2A3, i.e. it follows from
[31] that a linear transformation of the form z' — b2 = elv (z — b2)
is involved, where b2 is the coordinate of the point A2.
It follows from this that

/(<) = e'V(f) + y>


where y is a constant (y = b2 — et<pb2) and /*(<) is a new branch of
f{t) in the upper half-plane.
It also follows that
HO _ HO
HO n t) ’
i.e. the function
HO (42)
HO
is regular and single-valued in the neighbourhood of the point a2
this point can either be a pole or an essential singularity of the func­
tion (42). We shall show that this point is a simple pole with a residue
(a2 — 1). In fact we can replace z by a new complex variable z':

z' = (z — b2)a*,
where b2is the coordinate of the vertex ^ .T h e value z' = 0 corresponds
to this vertex and the sides A2AX and A2A3, which cut at an angle
a2 n are transformed into two straight lines, which cut at an angle
7i i.e. the above two straight lines are transformed into two parts of one
and the same line I in the z'-plane, which go from the origin in opposite
directions. If we now return to the £-plane, we can see that the neigh­
bourhood of the point o2, situated above the real axis is transformed
into the neighbourhood of the point z' = 0 in the z'-plane which
lies to one side of the straight line I. It follows from the principle
of symmetry that the same condition applies in the neighbourhood
of the points t = a2 and z' = 0 which lie on the other side of the
above lines. Hence the neighbourhood of the point t = a2 is trans­
formed into a one-sheeted neighbourhood of the point z' = 0 and
we have an expansion of the form:

2 — (z ^2)“* — c2(t — a2) -)- c2(t —r <z2)2 -j- . . . (Cj ^ 0 ).


36] CHRISTO FFKL’S FORMULA 149
It follows that:
z = b2 + c°s (t - a2)“* jl + -?*-(<- a2) + (t - a2y + . .

or, by applying the binomial expansion [cf. 23]:


f(t) = b2 +
where fx(t) is regular and does not vanish at the point t = o2. Hence:
/ ' (t) = a2( t - a ,)0*- 1 h(t) + ( t - a 2)“2/] (<),
f" (t) = a2 (a2 - l)(i - a,)02- 2/! (<) + 2 a 2 (t - a2)a^ f[ (t) +
+ ( t - a 2)°>n(t)
and consequently:
H O _ 1 K ~ 1) h (t) + 2 a 2 (f - a2) f j (t) + (t - o 2) 2 / i (Q
/ ' (f) t - oa ‘ a 2 /[ (0 + (t - <ij) («)

The second factor on the right hand side is a regular function at the
point t = a2 where it is equal to (a2 — 1 ), i.e. in the neighbourhood
of the point t = a2 the following expansion applies:
H O
+ P(t — a2)>
m
where P(t — a2) is a regular function at the point t = a2.
It can similarly be shown that the function (42) has at every point
ak on the real axis a pole of the first order with a residue (ak — 1 ).
Also, as we know, our function has no other singularities at a finite
distance and therefore the difference
HO ^ 1
(43)
no ^ t ~ as
is a regular single-valued function in the whole plane. We shall now
explain the behaviour of the function (43) at infinity. As we saw
above, the function f(t) tends to a definite value at infinity, viz. to
the coordinate of that point in the side A„AV which corresponds
to t = °o; consequently, in the neighbourhood of the point at infinity
we have an expansion of the type

f(t) — bm H----- + - + ■• •

It follows that the function f “(t)jf'(t) can be expanded as follows at


infinity:
H O ^ i
/'(/) t ’ t2 f ‘’
150 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [36

i.e. the function tends to zero when t-*-°°. We thus see that the
function (43), which is regular in the whole plane, tends to zero
when t —> °° and, consequently, it is bounded in the whole plane.
Liouville’s theorem [9], therefore states that the expression (43) must
be constant, but from what we have just seen, it tends to zero as
t —> 0 0 ; it therefore follows that the constant must be_ equal to zero.
We thus have the equation:
HO <*i - l , eg —i , , an - i (44)
/ ' (t) ' t —at l —a2 ' ' ‘ ‘ ' t —a„
Integrating once we obtain:
log / ' (t) = (cq - 1 ) log (t — ax) + (a2 - 1 ) log (t - a2)-f •+
+ (an ~ 1 ) log — an) + @

/' (t) = A ( i - a,)**-1 (t - a2)°'-K . . ( t - anr ~ \


and, finally, by integrating once again, we get the result:
t
z = /(f) = A J (t - aj )" !-1 (f - o,)"’- 1 . . . ( < - dt + B, (46)
o
where A and B are constants. Our problem has thus been solved
and the conformal transformation of the upper half-plane t into a polygon
with angles ak n is given by the formula (45), where ak are points on
the real axis, and A and B are complex constants.
We will, first of all, explain the importance of these constants.
Above we only used the magnitude of the angles of our polygon.
Thus when the polygon is subjected to movement or even to an
identity transformation we conserve the angles and therefore formula
(45) should also apply to the new polygon. The importance of the
constants A and B is due to the fact that when altering their magnitude
wo pass from one polygon to another similar polygon. The part
played by the numbers ak in formula (45) is of much greater importance.
The position of these points on the real axis together with the value
of the constant A give the lengths of the sides of the polygon. We shall
return to this problem later.
In deducing formula (45) we assumed that all the vertices of the poly­
gon have corresponding points on the real axis which lie within a finite
distance. We now suppose that one of the vertices, say An, corresponds
to the point at infinity. This transformation can easily be obtained from
36] CHBISTOFFEL’S FORMULA. 161

the one above by replacing ( by a new variable r according to the


formula
. 1 i
t = — — + an>
Bince when t = an we have r =
Completing the change of variables we obtain:

/(*) = ATJfl (an- «i - -tP 1•••


f 1 \ ( 1 \ a*~l dr D

Notice also that as a result of the well-known property of the


sum of angles of a polygon we have:

ai + a2 + • • • + an = 71 — 2. (46)
Using this relationship and changing the symbols of the constants,
the above formula can be rewritten in the following form:

J (r -
T

f(x) = A' a tf* -1 . . . (r —a ^ 1)a"-'~1dr + B '. (47)


o
This formula applies when one vertex of the polygon corresponds to
the point at infinity r = °°.
From formula (45) it is not difficult to obtain a formula which
gives the conformal transformation of a unit circle | w \ < 1 into our
polygon. It is sufficient to apply the bilinear transformation which
transforms the upper half-plane t into the unit circle \w \ < 1 . This
transformation takes the form:
w +l
w = tt +
—i 1
i ’ or i w —1
Substituting the expression for t in formula (45) and using (46) we
arrive at the following formula:
w
z = A" J [w — a £)“1-1 (w — alY*-1. .. (w — On)a"- 1 dm -f B", (48)
o
where the points oj lie on the circumference of a unit circle and are
defined with respect to the points ak by the formula
ak —*
a"k ak + *
152 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [36

In the formulae (47) and (48) we altered the lower limit of integra­
tion; this is of no great significance as it only affects the values
of the constants B ' and B".
Let us recall the assumptions made in deducing the formula
(45). We assumed th at there is a function /(<) which transforms
the upper half-plane into our polygon and we then obtained the
expression (45) for this function. Let us now investigate the formula
(45); assume that ak are given points on the real axis and ak are
positive numbers which satisfy the condition (46). We shall show
that under these circumstances formula (45) transforms the upper
half-plane into a domain without branch-points (one-sheeted or many-
sheeted), the contour of which is a broken line with angles ak n
(k = 1 , 2 , . . . , n). To begin with note that every factor (t — ak)°l~ 1 of
the integrand is a regular and single-valued function in the upper
half-plane, and the derivative
m = A(t - aJ - i- i (t - a , ) - - 1 . . . ( * - an)°*~i
does not vanish anywhere in the upper half-plane. Formula (45) thus
gives the conformal transformation of the upper half-plane into a
domain B in the z-plane which contains no branch-points. Let us now
consider the form which the contour of the half-plane, viz. the real
axis, acquires as a result of the transformation. Let us suppose that
Ovaries in the interval < t < a2 on the real axis. The corresponding
part of the contour of the domain B can be represented by the equation
t
z= A }(< - a ( t - . . . ( * - a „)a"—1 dt + C, (49)

where C is a constant, which is given in terms of the former constants


by the formula:
ai
C = B + A $ ( t - a ^ - 1 (t - a tf* -1 . . . ( < - a „)n— 1 d*.
0

When t varies in the above interval each of the differences t — ak has


a constant amplitude which we denote by <pk. We can, obviously
take it that g>t = 0 and <pk = n when k > 1 (ax < o2 < . . . < an).
The amplitude of the integrand in the expression (49) always remains
constant, viz. it is equal to

(«1 — 1) <Pl + (« 2 — !) <P2 + ■• • + K — 1)<pn = <p.


36] OHBISTOFFEL’S FORMULA 153
Formula (49) can therefore be written in the form:

2 = Aei<p J 11 - Uil"1- 111 - a 2 10’" 1 . .. - a ,,]0"-1 d< + C, (50)


ai
where we integrate along the interval of the real axis and in fact the
above integral is real. It can be seen directly from formula (50) that the
line a2 < t < o2 on the real axis corresponds to the line A 1A 2 in the
half-plane z , which starts at the origin and makes an angle arg ( A e 1*)
with the real axis. In the transition from < t < a2 to a2 < t < a3
we have to translate the point a2 from the upper half-plane. The
amplitude of the difference (t — a2) then receives an increment (—n)
and the amplitude of the factor (t — a2)a,_I receives an increment
—n (a^ — 1). Thus in the next interval a 2 < £ < a 3 we have a formula
analogous with (50) and only the amplitude q>has a new value which
differs from the one above by the term —n (a2 — 1 ), i.e. this interval
a2 < t < a3 corresponds to the line A 2A 3 in the 2-plane so that the
angle between the lines AXA2 and ^42^3 is equal to (71 — a2n).
We finally consider the point at infinity of the <-plane. For this
purpose we can rewrite the integrand in the formula (45) in the form
ffti+at...+an— _ °l j QJ—1 ^

Applying Newton’s binomial theorem and using the relationship


(46) we obtain the following expansion for the integrand in the
neighbourhood of the point at infinity:

1 I 03 1 c*
t* T t* ' (*

and after integration the right-hand side of the formula (45) is as


follows:
do + T " + ' i t + ‘ ' ' ’

i.e. the point t = 00 given by formula (45) is a regular point for the
function f(t). Hence when the real axis of the i-plane passes through
00 we obtain, as for ether segments of the real axis, part of a straight
line in the 2-plane. Note that as a result of the condition ak > 0 the
integral (45) has a fully defined finite value at the point t = ak.
Hence the above hypothesis with regard to the transformation by
the function (45) has been proved. As we have already mentioned
154 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [37

above the polygon obtained can cross itself (Fig. 40). The same applies
also to the formulae (47) and (48).
Thus, for example, formula (48) gives the conformal transformation
of the unit circle into the domain B bounded by a broken line which
contains no branch-points, provided the points a"k on the unit circum­
ference and the constants ak are chosen ar­
bitrarily to satisfy the condition (46).*

37. Individual cases. Let us begin with the


simplest case, viz. the triangle. Applying the
bilinear transformation to the i-plane we can
always simplify our problem by making the
vertices of the triangle correspond to the
points t = 0 , 1 and To do this we use the
formula (47) and assume that a[ = 0 and = 1, whence we
obtain the following formula

2 = A ' j'r°i- 1 ( T - l y ^ d T + .B'. (51)


6

In this case our formula only includes the arbitrary constants A ’


and B'; these constants are of no significance since they are connected
with an identity transformation of
a triangle. The comparative simpli­
city of formula (51) is due to the fact
that any two triangles with equal angles
are necessarily similar. In the case of
a quadrangle this circumstance no
longer applies and the integrand in the
general formula for a quadrangle with given angles, contains an
undefined parameter, which depends on the lengths of the sides of
the polygon.
Formula (51) also applies to an infinite triangle with angles n/2,
?r/2 and 0. This triangle represent a strip, bounded by two parallel
halves of straight lines and a perpendicular line (Fig. 41). Assuming
that cq = cq = 1/2 we have in formula (51):

dr
z= A + B'.
7 Yt(t —1)
0
37] INDIVIDUAL CASES 166

Let us consider this rectangle in greater detail. Assume that the


vertices of the rectangle B have the following coordinates (Fig. 42):
eo, to, a>, , . co. , .
T< T ~2--- 1“ ta>2 < ----- 2--^

where co1 and co2 are given real positive numbers. Take the right
half of this rectangle with the vertices
„ (U, CO, . .
°’ 2 ’ “ 2“ + *" 2’ l(02’

F ig . 42

and assume that it is conformally reflected in the right half of the


upper half-plane t, i.e. in that half of the upper half-plane t where
the real parts of all points are positive. We can take it that the vertices
0 , 0 J 2 and ico2 correspond to the points 0 , 1 and 00 on the contour
of the above right part of the upper half-plane. The vertex coj/2 + ica2
corresponds to a point on the real axis, which lies between the points
1 and 00 . Denote this point by 1/fc, where 0 < fc < 1. As a result
of the principles of symmetry the left side of our rectangle corresponds
to the left half of the upper half-plane t and the vertices —0)J 2 ,
—u>'/2 -f i co2correspond to the points t = —1 and t = —1/fc. It follows
from above that we can always conformally transform the upper
half-plane into a rectangle B, so that the points t = —1 , 0 , 1, 00
should correspond to the points 2 = coj2, 0, coj2, i co2, while the
points t = 1/fc and t = —1/fc should correspond to 2 = coj2 + i co2
and 2 = —ojJ2 -f- i co2. We can now apply formula (45), assuming
that a 1 = —1/fc; a2 = —1 ; a3 = 1 ; ai = 1/fc and a2 = a 2 = a3 =
= = 1/ 2 -
156 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [3 7

We thus obtain the following formula, bearing in mind that t = 0


when 3 = 0 :
t
t d
o f <>-<•> ( A - - ••)
i.e. the following formula:
_ _ _ _ dt___ (52)
* - Af

The values of t in the interval —1 < t < 1 on the real axis, give
the line (—a>j2, -\-aij2) on the real axis in the 2-plane. It follows
that we can assume in formula (52) that A is a positive constant
and that the radical is unity when t = 0 . Other values of the radical
in the upper half-plane are unique, since this radical is a regular
function without branch-points in this half-plane. Bearing in mind
that the vertices coj/2 and coj2 + i oi2 correspond to the values 2 = 1
and t — 1jk, we obtain the following formulae:
l
CO, dt
2 A f /(I - t2) (1 - Jfc2 t2)
(53)
dt______ _
"2
~1) (1- ifc2t2)
The lengths of the sides of our rectangle are equal to tOj and co2;
we can therefore construct an equation to determine the parameter
k, which enters the integrand, from our knowledge of the relationship
of the lengths of the sides of the rectangle:

oi, : oi. dt
= 2. f w - (1 - k1P)
i*) 'J K - )d«- fc22)
(<2 1 (1 1
(54)

Having thus determined k (theoretically speaking) we can proceed


to determine A from one of the equations (53).
The integral in formula (52) cannot be expressed by elementary func­
tions and is known as an elliptic integral of the first class in the
Legendre form. We shall deal with these integrals later and therefore
we shall not investigate the problem of finding k in the equations
(54) in greater detail here. The above argument was introduced so
37] INDIVIDUAL CASES 157

as to explain more clearly the problem of the determination of Tc in


Christoffel’s formula.
Let us consider one more case. Suppose that we have a regular
71-polygon AXA2 . . . An in the 2 -plane and let 2 = 0 be its centre
(Fig. 43 for n — 6 ). Take the conformal transformation of the triangle
0A1A2 into the sector 0 'A {A 2 of a unit circle with the angle at the
centre equal to 2jiln, so that the vertices of the triangle 0, Al and A 2
should correspond to the centre of the circle O' and to the ends A[ and
Ai of the arc. According to the principle of symmetry the reflection of

the triangle in its sides gives the reflection of the sector in the corres­
ponding radii. Thus in the course of analytic continuation the function
reflects the whole regular polygon into a unit circle. It follows directly
from these considerations that by reflecting a regular polygon into
a unit circle, the vertices of the polygon correspond to points, which
divide the circumference of a unit circle into equal parts. In this case
we must also assume in formula (48) that

Rotating the unit circle about the origin, we can, of course, take
it that the vertix Ax corresponds to any arbitrary point on the circum­
ference, for example to the point w = 1. At the same time the points
el2nkln (lc= 1, 2 ........n — 1 ), on the circumference also correspond to
vertices of the polygon so that the integrand in formula (48) has
the following form:

\, '- U 1— 1 (
L(m> — lH w — e n J \w — e n) . . . \w — e
i(n-D— 11—-
n JJ n.
If we suppose that the origin is the centre of the polygon we obtain
the following formula for the transformation of a unit circle into a
158 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [38

regular n-polygon
W
dw
z= A (55)
'/ y(wn _ l)i

The modulus of the constant A" is determined from the dimensions


of the polygon and the amplitude of this constant simply gives the
rotation of the polygon about the origin.

38. The exterior of the polygon. Let us now consider that part of the
plane outside the broken line (Fig. 44). In this case our domain, which
we can also call a polygon, contains the point at infinity. We const­
ruct the function z = f(w) which conformally transforms a unit circle
into our infinite polygon. In this case the sum of the angles of the
polygon is equal to n (n + 2 ); denoting the angles by ak x, as before,
we obtain instead of (46) the following re-
— lationship:
ai + a2 + • • • "t" a n — n + 2 . (56)
We suppose that the origin w = 0 is
transformed into the point at infinity. The
function f(w) will then have a simple pole at
the origin and the function f'{w) can be ex­
panded as follows in the neighbourhood of
the origin:
/ , (M’) = -^ r + co + c i W’ + ••• (57)

^ We now suppose that ak are the points on


the circumference of the unit circle which cor­
respond to the vertices of our polygon. We
construct the function as before. By proceeding in the same
way as in [36] we can see that this function has a simple pole at every
point a"k with a re s id u e s — 1 . Also, as a result of(57),we see that it has
a simple pole at the origin with a residue ( —2 ) and at all other points
it is regular, as before. Let us investigate its behaviour at infinity. The
function f{w) is equal to infinity at the origin, and in the course of ana­
lytic continuation across any arbitrary arc a"k ak+1 of the unit circle it
acquires the value infinity, symmetrical with the value of this function
at the origin with respect to that side of the polygon which corresponds
to the arc ak ak+1 of the unit circle, i.e. the value of the function f(w)
at infinity is equal to infinity and our function transforms the neigh-
38] THE EXTERIO R OF TH E POLYGON 159

bourhood of infinity into a one-sheet neighbourhood of infinity. (This


is that part of the polygon obtained from the given polygon as the
image in the side Ak Ak+l). Hence the values of f(w) obtained after
the above analytic continuation in the neighbourhood of infinity can
be expanded as follows:
f(w) = d ^ w + d0 + ~ + ••• (d -^ O ). (58)

The function f"(w)jf'{w) is single-valued and regular in the whole


plane except at the above poles. Differentiating formula (58) we obtain
the following expansion for the function f(w)jf'(w) in the neighbour­
hood of infinity:
r(w) _ h ht
f ' (w ) tv3 “T" W* ^ ' • • ’ (59)
where the direction of analytic continuation is irrelevant owing to
the single-valuedness of the function. Thus in the case under considera­
tion, the functionf"(w)lf'(w)h&s the above poles and vanishes at infinity,
but otherwise it is regular. Carrying our argument further, as in [36]
instead of formula (44) we obtain the following:
f" (™) _ 2 I °i —1 I q2 - 1 , | an- l (60)
f (w) w ' w — a\ w — a'i ' " " * ' w — ’
this gives the following formula instead of (45):
W dto
z = A § (w — a ")01-1 (w — a £)a,_1 (w - a"ny + B. (61)
i
If we apply the transformation tv = 1/r to the variable xd then
the interior of the unit circle is transformed into the exterior of the
unit circle and, after performing corresponding transpositions in the
integral (61), we obtain a formula which transforms the exterior of
a unit circle into that part of the plane outside the broken line, where
the points at infinity correspond with each other:

3 = A ' / (T - ai )“1-1 (T ~ aa)a2-1 ■ • • ( * - + B. (62)


l T
The form of this formula is the same as that of formula (61).
Let us consider, for example, the transformation of that part of the
plane outside a square. Owing to symmetry we obtain points ak, which
divide the circumference of the unit circle into equal parts and, by
rotating this circumference, we can take it that these points will be
CL j — 1 J d 2 — %, Q jj — 1 , ■ %•
160 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [38

In this case we have the following relationship for the angles:


3
— O3 — €£4 — 2 »
and therefore the final formula has the form:

t = + . (63)
1

The values of the constants A ' and B depend on the dimensions


of the square and its position in the plane.
Note that at the beginning of [36] we considered the transformation
of the half-plane into a polygon and in this paragraph we applied
the result obtained to the transformation of a circle into a polygon.
It can easily be seen that all arguments used in [36] also apply here.
Let us make one remark in connection with formula (62). We ex­
pand the integrand in the neighbourhood of the point at infinity.
From (56) we can rewrite this function in the form:
1

(■
or applying Newton’s binomial theorem, we obtain the following
expansion:
(°i — 1) o t + (a, — 1) a , + . . . + (a„ — 1) an Cj_
1 — T ' T» ' t3 ~
When integrating the term containing 1/r we obtain the function
log x and, consequently, in order that the neighbourhood of the point
t = 00 should correspond to a one-sheeted domain it is necessary (and
sufficient) that the constants a* should satisfy the relationship:

(a l — 1) ° l + (a 2 — 1) a 2 + • ■• + (a rt — 1) an — 0 ■ (64)

If this relationship is not satisfied then formula (62) transforms the


outside of a unit circle | t | > 1 into the domain B bounded by a
broken line with a branch-point of the logarithmic type at infinity.
Formula (62), as we noticed above, can also be used for the trans­
formation of the interior of the unit circle |t| < 1 into an infinite poly­
gon, when the point t = 0 is transformed into the point at infinity.
If we expand the integrand in the neighbourhood of the point r = 0
and cancel the terms containing 1 / t , we obtain the condition:
39] TH E MINIMUM PROPERTY OP TH E TRANSFORMATION INTO A CIRCLE 161

It is identical with (64) since it is given that \ak \ = 1; hence it is


given that ak 1 = d* and that the numbers ak, are real (and positive).

39. The minimum property of the transformation into a circle. Consider the
function
e = f (r) = r + c2r2 ---- (65)

which is regular in the circle | r | < R. This function transforms the circle into
a domain B which can have several sheets and which can contain branch-points.
The circle | r | < R lt where < R, is transformed by (65) into a part of the
domain B, which we denote by B v Let us determine the surface area of this
domain. As we know it can be expressed by the integral [29]:

S i - J J l/'M I 2d*.
M<*.
where we integrate round the circle | r | < R 1. We can rewrite this integral
as follows:
J?, 2*

Si = J J (1 + 2%™'* + 3csr2el2v + . . . ) (1 + 25jre~/’’+ 3c3r2e~i2^ + . . . ) r dr d<p.


«o
Owing to the absolute and uniform convergence of the series in the circle
| t | < R r we can multiply our two series term by term and we can also integrate
them term by term. Notice that by integrating the function elklp through the
interval (0, 2ti), where k is an integer other than zero, we obtain zero. Thus when
multiplying the above series it is sufficient to retain only those terms which
do not contain factors of the type elk,p, and integration with respect to <p will
then simply involve multiplication by 2n. We thus obtain:

S, = 2jiJ(1 + 2 2|c2 |2r 2+ . . . + n2 | c„ |2r2n—2 + . . . ) r d r


o
or
S ^ n R f + n 2n\c„\*R\n. (66)
71 = 2

As tends to R, the latter increases and tends either to a finite limit or to


infinity. In any case this limit, which gives the surface area of the domain
B, will be greater than nR2, i.e. greater than the surface area of the initial circle
| r | < R, provided th at in the expansion (65) at least one of the coefficients
ck is other than zero. We thus obtain the following result. In the transformation
of the circle \ r | < R by the function (65), which is regular in this circle, the
surface area of the domain increases, provided at least one of the coefficients ck
is other than zero.
Having established this preliminary theorem we can now explain one very
important property of a function performing conformal transformation. Let
£ be a connected bounded domain in the z-plane and assume that the origin
z = 0 lies in this domain. Suppose also th at F fz) is a function which
1G2 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [39
conformally transforms B into a unit circle and that the origin z = 0 is trans­
formed into the centre of this circle. In the neighbourhood of the point z = o
this function can be expanded as follows:
F 1 (z) = dxs + dt z2 + . . . .
where dx > 0. Let us now consider a new function

This function transforms B into the unit circle | r | < R, where R = l/du
and its expansion in the neighbourhood of z = 0 is:
r = F (2 ) = 2 + az z* + a3z3 + . . . (67)
Its inverse function is regular in the circle | r | < B where it can be expanded as
follows:
z = / (r) C= T + c2T* + c3 t3 + . . . (68)
Tho double integral
J j | ^ ( 2 ) | 2ds, (69)
B

which gives the surface area of the circle must be equal to 71R2. If instead of
the function F(z) we take any other function <p(z), which is regular in B and
has an expansion of the form (67) in the neighbourhood of the point 2 = 0,
then by substituting for 2 in the expansion (68) we obtain a function r which is
regular in the circle | r | < R where it can be expanded as follows:
<P(*) = ?>[/ (*)] = t + esT' + e3z3 + . . . = / , (r). (70)
Let us evaluate the double integral (69) for this new function <p(z). Changing
to the r-plane and recalling the expression for an element of surface area in
the r-plane in terms of an element in the 2 -plane [29]:
dsz= |/'(t) I*dsT,
we find:
j‘J|^( )|Jdsz= J J | <p' (z)•/'(t) I2dsT= J J |/i(T)|2dsT,
2
B M < rt M < R

and in accordance with the above hypothesis this integral will be greater than
nR2, provided th at at least one of the coefficients in the expansion (70) is
other than zero. If all coefficients are equal to zero, i.e. <p(z) = r then evidently
<p(z) = F(z). We thus arrive at the following theorem.
Theorem. Among all the junctions which are regular in B and have in the
neighbourhood of z — 0 an expansion of the form (67) there is one function which
conformally transforms B into a circle, centre the origin, and which gives the integral
(69) its minimum value.
This theorem can be used for the construction of an approximate expression
for the function F(z) which transforms B into a circle, in the form of a poly­
nomial. Hence F(z) can be approximately represented by a polynomial of
the nth degree:
F (z) = 2 + Oj z2 + . . . + an zn. (71)
39] TH E MINIMUM PROPERTY OF TH E TRANSFORMATION INTO A CIRCLE 163

and we shall determine the coefficients ak of this polynomial from the condition
that the polynomial (71) alone among all other polynomials of a similar kind,
gives the integral (69) its minimum value. Let us construct an arbitrary poly­
nomial
(o (z) = b2 z2 + b3 z3 + . . . + bnzn
and subsequently construct a new polynomial which has the same form (71)
as the polynomial F(z):
<P(z) — F (z) + ecu (z),
where e is a real parameter. Construct the integral (69) for this new polynomial

J J [-F' (z) + «*>' (2)] [&' (2 ) + «"»' (2 )] d*.


B

This function of e should reach its minimum value when e = 0. Equating


to zero its derivative with respect to e when e = 0 we obtain the following
condition:
J J” [F' (z) (o' <z)] + [F’ (z) (o' (z)] da = 0 , (72,)
B

which should apply for every arbitrary choice of the polynomial cu(z).
Similarly, substituting ie fore, where e is real, we obtain instead of (72,)
the condition
J J IF' - F 7^ ) (o' (z)] da = 0. (72,)
B

On adding, we obtain the condition


j’ J F ' (z) (o>(z) da = 0 .
B

Assuming th at <o{z) is, in turn, equal to


co (z) = z2, z3, . . . , zn
and introducing the symbol:
Ptk = .f J z ' V d s , (73)
B

we obtain the following system of equations of the first order for the required
coefficients ak of the polynomial (71):

Pio + 2P u + 3P iz « * + • • • + MPi, n-i a n =


P20 + 2p 2,o 2 + 3p2j a s + . . . + npj n., an = 0, ^74 )

Pn-1,0 + 2 P n -i,l ®2 + 3 P n -i, 2 ° 3 + + n Pn-i, n -i°n = 0 -

Hence this problem must involve the evaluation of integrals of the type (73).
If a simple closed curve, which does not cross itself, serves as the contour
of the domain, then it can be shown that the above polynomials tend uniformly
as n ->■ 00, to a function in B, which reflects B into a circle.
164 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [40
In conclusion we shall make a remark about the first theorem proved in
this section. The function (65) transforms the circle | t | < R into a domain
B which can have extremely complicated geometric properties, viz. it can
have several sheets and the form of the contour can be very complex. I t can
be shown th at such a domain may not even have a surface area in the usual sense
of the word and what we have called the surface area of the domain must be
understood as the limit of the surface areas of the domains B u which lie in B
and which expand in such a way th at every point of B which falls within these
domains makes them tend to B as their limit. If B has a surface area in the
usual sense then the latter evidently coincides with the above limit.

40. The method of conjugate trigonometric series. We shall now give


another method for the approximate construction of a function which con­
formally transforms the connected domain B into a circle. In this case we obtain
this approximate representation in the form of a polynomial which does not
lie in the z-plane of the domain B, as was the case above, but in the r-plane of
the unit circle. For the sake of simplicity we assume that the centre of the circle
is transformed i nto the origin which lies in B. Let
Z = O, T -)- Oj T2 + . . . (75)
be a function which transforms the unit circle C{\ r | < 1) into B. If the contour
of £ is a simple closed curve, then it can be shown th at the series (75) will
converge uniformly within and on the circumference of the closed circle G. On
the circumference we assume that t = el'p, where 0 < <p < 2n; we then obtain
the equation for the contour r of our domain B:
z = x + iy = Oj e,p + o2 el3p + a3e,3p + . . . (76)

or, separating the real and imaginary parts in the coefficients = afc — i(ikt
we can write the equation of the contour in the form:

x = 2} (°A cos k<p + f}k sin k<p); y = £ (—Pk 003 k<p + aftsin k<p). (77)
k= 1 k= 1

In a particular case a, can be real, i.e. ^ = 0. The equations (77) give the
parametric representation of the contour r of the domain £ in a special form,
viz. they give the parametric iepresentation in the form of conjugate trig­
onometric series [25]. This is known as the normal parametric representation
of a curve. In its complex form this can be written in the form (76). Conversely,
if we have the normal parametric representation of the contour J1of the domain
in the form (76) or (77) we can construct the function itself by substituting
e'kp by r* in the series (76). In this case the series (76) must be uniformly con­
vergent. The problem thus involves the finding of the normal parametric repre­
sentation for the contour r of the given domain B.
We assume th at we have an equation for the contour f i n an indefinite
form and th at this equation is as follows:
x2 + y* - 1 + AP (x \ v2) = 0, (78)
40] TH E METHOD OP CONJUGATE TRIGONOMETRIC SERIES 165

where Ais a constant and P(x2, y 1) a polynomial containing only even powers
of x and y. The equation (78) can be rewritten in complex form. Note that
p(x2, y2) can be considered to be a polynomial of two arguments:
x1 -f- y 2 = zz and 2 (x2 —y 2) = z2 -f- z1,
bo that the equation (78) can be rewritten in the form:

*2 —1 + A 2 2 Au (zzf (z2 + z2)1= 0, (79)


I =0 k=0
where A kl are given real coefficients. I t is given th at our curve r is symmetrical
with respect to the axes of coordinates and by repeating the arguments analo­
gous with those in [37], when we considered a regular polygon it can be shown
that in the formulae (77) fik = 0 and a2k = 0 so th at the complex form of the
equation of the contour is as follows:
2 = a, ei<p+ a, e'3’’ + . . . (at > 0), (80)
where a^ +1 are real coefficients; consequently:
£ = a 1 e - ,* + a s e - ,3»’ + ... (81)

By direct multiplication we obtain the expressions:

+ c*
2 2J aij+i atj'+i I .*iP9
P” —“ L,j - j = p J

(2p+2)w _|_ (82)


Z2 + S*- 2 2 a2j+l +1j
p-0 J+j’=P
+ CO
e -((2p+2)?>_
+ 2 2 a2]+l a2j'-¥l
P =0 j+ r-p J
In each of the above expressions summation with respect to j and j ' is done
from 0 to + oo but only those values are taken which satisfy the equations writ­
ten below the symbol of summation. Substituting the expression (82) in the
left-hand side of (79) we should, by multiplying the series once again and col­
lecting terms with identical powers of eiip, equate to zero all terms with dif­
ferent powers of e“f . Note th at in the formulae (82) the coefficients of positive
and negative powers will be the same but only positive powers of e1’’ enter
the formulae. The same evidently applies to the expansion of the left-hand side
of the equation (79), so that we have only the constant term and the coeffici­
ents of ei2Plp where j> > 0 to equate to zero.
Without actually performing all of the evaluations in the general case we find
that from the first equation in (82) we obtain a system of equations in the form:

ai + a3 + af + •• • + AT0 («2j+i) — A
ai a3+ as a5+ ••■+ ^ Pi (a2j+i) = 0 (83)
«6 + aa a, + ■■• + (a ,;+ l) = 0
166 OONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [40
where T p(a2j+1) are definite expressions containing the given coefficients Aki
and the unknown coefficients o2y+I. We shall not write them out for the general
case. Let us rewrite the above system of equations leaving on the left-hand
side of each equation only the first term, extracting the square root in the first
equation and dividing all remaining equations by at:

ai —Vl —[a| + a! + • • • + ^ 2\>(a2j+i)]»


a3 a 3 as _ a s a 7
^ ( a 2j +1) ,
«i ui
as —— AT 2 (a2j+i),
“i

Expanding the radical by using Newton’s binomial theorem we obtain:

ai = 1 — [al + <*%+...+ ).T„ (a2(+t)] +

+ [a| + a| + . . . + AT 0 (a2y+1)]2 + ■• •
(84)
• ~ ~ ^ ^ 1 (a 2]+l)
ui
• ^(a2y+i)
ai
-

We shall solve this system by the method of successive approximations and


by taking the following for the initial values:

a<<» = 1; a(3°) = = . . . = 0. (85)

Substituting the expressions (85) in the right-hand sides of the equations


(84) and rejecting all terms which contain powers of A higher than the first
degree we obtain the first approximation for the coefficients:

+i + My+i> (86>
whence, by using the expressions T k(azj+i), it can be shown th at all expressions
(86), are equal to zero for sufficiently large values of j.
By substituting again the expressions (86) in the right-hand sides of the
equations (84) and rejecting all terms containing powers of Ahigher than the
second degree we obtain the second approximation for the coefficients in the
form:
“iyVi + MV+1 + ^ a2j+i >
where again all expressions will be zero for large j ’s etc. I t can be shown that
the infinite series thus obtained for a2y+, converge for all values of A which
are sufficiently close to zero and that they give the solution of the problem.
40] TH E METHOD OF CONJUGATE TRIGONOMETRIC SERIES 167

Example 1. To clarify the above method we shall consider an example,


viz. we shall find a function which reflects a unit circle into the interior of the
following ellipse (87)
a^ + y2 - ! - A(a;2 - i/2) = 0. (87)

In its complex form this equation can be written as follows:

z2 + z2
zz — A 1,
2

and by using the formulae (82) we obtain an infinite system of equations


in the form:

<*! + <*! + °! + “ ? + a? + a J i + . . .

a, a , + ag a 6 + a 5 a, + a 7 a 9 -f- a 9 a n +

“ i a5 + a a a7 + as + a, a u + ... = * (“l Og)


a? + “s a»+ at au + • •• = A ( y “S + a i a s)
( 88)

<*i a9 + a3 a„ + ■• • = A (a, a, + a3 a5)

“i an+ agaig + ■•• = A fa t a , + a3 a, + y a| j

Let us introduce the new unknown gk, assuming that

eo = ; pi = ; p2 = • (89 )
al a\

In this case the system (88) can be rewritten as follows:


JL
<?o = ( i + e ! + e l + . . . ) *
i ,
S i ----2" A Pi e3 62 Pa 6s 61 646s • ••

62 — ^61 6163 6264 Pa 65 •••


, (1 , , 1 (90)
e a = ^ [ y e! + e 2J - ei Pi - e2 e5 - •••

p4 = a (g! e 2 + p3) - Pi p6 - •••

Pt = A ( p i + Pi Pa + y Pl) - Pi Pg - • • •

Without paying any attention to the first equation for the present, we can
solve the remaining equations by using the above method of successive approx­
imations.
168 OONFOILMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [40
Thus, on going as far as terms containing A5 we obtain:
1 , 1 ,. 3 ,
e' ~ Y x ~ T x + ~32x;
o -- ^ 16
e» A«■
6*~ 8 X‘!
where all remaining gk are equal to zero. We took as the initial values
p,01= Pj0> = . . . = 0 . Substituting the expressions obtained for gk in the right-
hand side of the first equation of the system (90) and using Newton’s binomial
formula, we obtain an expression for g0 which is accurate up to terms in A5.

By knowing the values of gk, and from (89) we can construct ak:

— 001 °a — 00 01 I a5 — 00 02 J ••■

Thus the unknown function which reflects the interior of the unit circle
into the exterior of the ellipse (87) can be approximately represented by a
polynomial of the eleventh degree:

2. Consider the conformal transformation of a unit circle into the interior


of a square bounded by straight lines x = ± 1 and y = ± 1 parallel to the
axes. The equation of this square can be written in the form:

(1 — x2) (y2 — 1) = 0 or ar2 -fj/2 — 1 — i 2j/2 = 0.

Introducing the parameter A we obtain a family of lines:

x* + V2 — 1 — Ax2y* = 0.
In the complex form this can be written as follows:

In this case the square is symmetrical with respect to the axes of coordinates
and the bisectors of the angles between these axes. Bearing this in mind and
using the same arguments as in [37] we find that the normal representation of
the contour of our square must have the form:

z = a, el<p + a5 e,5p + a9 + . .. (<h > 0),


40] th e method of conjugate trigonometric series 169

where aik+l are the required real coefficients. The above method gives the
following infinite system of equations for these coefficients:
a| + “5 + “S+ ■• ■=
= 1 +y [(y~] + (“i“s)2+ (ai “»+ y as) + (“i“is+ asas)2+ ■
••J
al a5 + a5a9 + “l, + • • • =

“ l[“ T (a' “■>(t ai) + (“'" • + T ]


(92)

al a13 + • • • =
= y [ - ("T") (a. a®+ y al) ~ y (“i “,)2+ («i “i3+ “6“,) (y “2) + -••]

In this case we act somewhat differently, viz. we suppose immediately that


A= 1 and we then solve the system (92) so obtained by the method of successive
approximations; the initial values are as follows:
ai = l; “5= “, = ... = 0 .
Substituting this in the system (91) we have:

af = 1 + i ( y ) 2: a« = T - [ - T ( y ) 2] ; o,-o.,= ...= o
or
a, = 1.0607; a5 = — 0.0626; a9 = aI3 = . . . = 0.
Substituting these approximations in the system (92) we have:

a \ + ( - 0.0625)2 = 1 + y [ (L0^°7)- + 1.0607)2 • ( - 0.0625)2 + -i- ( - 0.0625)‘j ;

1.0607a, = — [^ -— ^1'0^07)< + y (1.0607)2 • (-0 .0 6 2 5 ) + - ( - 0 . 0 6 2 5 ) • I.O6O7J ;

1.0607ag = i - j^_ -L (1.0607)2 (-0 .0 6 2 5 ) + -i- (-0 .0 6 2 5 )2 • (1.0607)2j ;

1.0607ala= y [ “ T t1-0607)2 ( ~ 0.0625)2j ;

1.0607a,,= 0.

This gives the following approximations:


a, = 1.0672; a5 = - 0.0922; a3 = 0.0181; als = - 0.0016 ; a 17 = 0.

These successive approximations can obviously be continued further but to


evaluate successive approximations the system (92) must be written in a more
specific form, by adding new equations and writing out a greater number of terms
170 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [40
in each equation. To evaluate successive approximations the preceding approx­
imation is substituted in all terms of the system (92) except m the first term
on the left, although beginning with the second equation, the values of a,
are substituted in the first term.
In the case under consideration the values accurate to the first four decimal
places, will be as follows:
a, = 1.0807; a5 = - 0.1081; a, = 0.0450; au = - 0 . 0 2 4 2 ; a „ = 0.0174;
a2l - -0.0125.
Notice th at when applying the method of successive approximations with
initial values a, = 1, ao = a9 = . . . = 0, all the coefficients aik+l in every
approximation from a certain coefficient onwards will be equal to zero.
Instead of the above method an expansion in powers of A can be used, not
for the coefficients a2y+I but for the right-hand sides of the equations (77) or
(78). This will lead to somewhat different results when the method of succes­
sive approximations is used. We are trying to find the normal parametric repre­
sentation for the curve (78) in the form of series in whole powers of the para­
meter A.
x = xB(?) + xx (?) A+ x2 (?) A* + . . . ; y = yB(?) + y, (?) A+ j/2 (?) A2+ . . . , (93)
where x 0(?) and j/0(?) are functions giving the normal parametric representation
of the curve (78) when A = 0, i.e. they give the circles x z -f- y* — 1 = 0. In
other words, in formula (93) we have:
x 0 (?) = cos ? ; yB(?) = sin ?.
Further coefficients xk(<p) and 2/*(?) must be conjugate functions, i.e.
they should be represented by conjugate trigonometric series. Substituting the
expression (93) in the left-hand side of the equation (78) and equating to zero
terms with like powers of A we obtain equations for the determination of
the coefficients of the expansion.
3. Let us apply the above method to the ellipse:

«5 - y (z2 + z2) = 1 , (94)

which we have also considered in connection with the first method. We shall
try to find the normal parametric equation for this curve in the form:
z = * + iy = e'51+ zt (?) A+ s2 (?) A2 -f . . . , (95)
where every value z^.(?) is an expression of the form:
zk (tp) = e"p + e'3v - ( - . . . (96)
Substituting (95) in the left-hand side of (94) we have:

(e'> + 2lA+ 22A2+ . . . ) ( e - ,’, + z1A+ z2A2+ . . . ) _ A [(e'*, + zl A + . . . ) 2 +

+ (e -> + 2lA + . . . ) 2] = 1. (97)


40] TH E METHOD OF CONJUGATE TRIGONOMETRIC SERIES 171
Equating the coefficients of A to zero we obtain:

e' ^ 1 + e- ' ^ 1 = i - ( e ^ + e - ^ )
or
CB[e-,?’z1] = y cos 2<p,

where tji is the symbol of the real part. In agreement with (96) this gives

HI [af1) + a^1) ei2q>+ . . . ] = y cos 2tp,


whence
af!) = y ; aiji) = a(D = aW = . . . = 0 .
Finally this gives:
21 = y e ' 3p. (98)

Let us return to formula (97) and equate the coefficients of A1 to zero:


e~,p z2 + e,?l z, + 2 , ^ = e/v zt + e~lv
or
' Kfe-^Zj] = '£[e'*’z1] - y z , z 1.
From (98) we have:
rn [e‘ i? SjJ = - y + y cos 4<p,
i.e. remembering (96):

Vi [a(2) + a(2) e'2?1 -f . . . ] = — y + y cos 4y,

whence a.™ = — 1/8, aj.2’ = 1/2 and the remaining a^’+1 are equal to zero, i.e.

22 = -y e '* + y e ^ .

Continuing in the same way we obtain

^ = elSlp + | ei1lp; z4 = ^ e'*> - A e'1- + I e'»*


16 8 8

and, finally, substituting in (95) and replacing e'k<p by t * we obtain the approxi­
mate expression for the unknown conformal transformation:

The method described above was developed by Prof. L. V. Kantorovich.


The detailed description of this method with the proof of convergence can be
172 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [41
found in the work of the author (Mathematical Manual, vol. 40 : 3). Notice that
if in formula (91) we consider no terms containing powers of A higher than
the fourth power then formula (99) will be obtained.

41. The two-dimensional established flow of liquids. Having explained


the theoretical basis of the theory of conformal transformation we
shall now consider the application of the theory of a complex
variable to hydrodynamics. Let there be a two-dimensional estab­
lished flow of liquid, where the potential of velocity is equal to
(p(x, y) and the function of the current is equal to y>(x, y) [II, 74]. Let
us recall that the components of current at every point are ex­
pressed by the formulae:
_ d<P(x,y) . _. 9(p(x,y)
X~ dX > Vy ~ Qy

and the difference


V {x v V i ) — V> (®0>Vo)
= V { M i ) — V { M 0) (101)
gives the quantity of liquid which flows in unit time across an arbi­
trary contour connecting the points M Qand Mv The flow is indepen­
dent of time and is the same in all planes parallel to the X Y -plane;
the liquid is of unit density. More strictly, the expression ( 101 )
gives the quantity of liquid which flows in unit time across a cylindri­
cal surface, parallel to the Z-axis, of unit height, the surface having
the shape of a contour I in the XT-plane which connects the points
M 0(x0, y0) and M ^x^ yx). Hence from above the functions <p(x, y) and
y>(x, y) are connected by the relationships:
0ip dtp d<p dtp
dx dy ’ dy dx ’
which are exactly the same as the Cauchy-Riemann equations. We
can therefore say that the function of the complex variable
f (z) = <p(x, y) + iy [x, y) (102 )
has a derivative in the domain occupied by the flowing liquid. This
function ( 102 ) is usually known as the complex potential of flow.
As we said above, the functions <p(x, y) and y>(x, y) can be many­
valued viz. they can acquire constant terms by encircling a certain
point, or more generally, by encircling certain holes in the region
under consideration. For the function tp(x, y) this many-valuedness
indicates the presence of a source at the corresponding point and for
the function <p{x, y), the presence of an elementary turbulence at
41] TH E TWO-DIMENSIONAL ESTABLISHED FLOW OF LIQ U ID S 173
that point. In such cases the function f(z) will also be many-valued,
i.e. it will acquire constant terms by encircling certain points (or holes).
From (100) we can see that the velocity vector corresponds to the
complex number
9© . . 9© 9© ,9w
"9x 1~dy ~ Hx ^ ~dx
The latter expression is the function conjugate with the derivative
fjz) [2 ], Hence a function g(z), conjugate with the derivative f'{z),
gives the velocity vector of the flow.
Consider the isothermic net which corresponds to the function (102):
<p(x,y) = C1; ip (x, y) = C2. (103)
The first family of lines represents a family of lines of equal velocity
potential or, in other words, a family of equipotential lines. The
second family (lines of current) represents, as can easily be seen, a
family of trajectories of liquid particles. In fact, as we know, the two
families will be orthogonal but the velocity vector, which is equal to
grad <p(x,y), is directed, as it happens, along the normal to <p(x,y) = cv
which corresponds to a line of the second family (103). Thus in the given
established flow the velocity vector at every point is directed along
the tangent to the line from the second family (103), which passes
through that point, i.e. in fact this family is a family of lines of current,
and the latter, in an established flow, gives the trajectories of liquid
particles.
Until now we have considered the kinematic picture and thought that
any kinematically possible picture of movement can be given by a com­
plex potential which is a regular function; conversely, any complex
potential gives a kinematically possible picture of movement. We
will now show that we can in this way also satisfy other hydrodynamic
equations which give us the value of the pressure. Let us write down the
hydrodynamic equations for the two-dimensional established flow,
assuming that exterior capacity forces have a potential U{x, y).
Bearing (100 ) in mind we obtain two hydrodynamic equations and a
continuity equation [II, 115]:
dip 92© 9© 92© __ dU 1 dp
dx dx2 9y dxdy dx q dx ’
9© 92© 9© 92© __ 9(7 1dp
dx dx dy 9y dy* dy q dy ’
d2y , &<P _ n
9s2 f dy2 *
174 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [42

where q is the density of the liquid and p ( x , y ) is the pressure. The


continuity equation is evidently satisfied, for the real part of the
regular function is a harmonic function. The first two equations can
be rewritten in the form:

wW+GtfM+T'K
It follows that the expression inside the shaped brackets, must be
a constant and we thus obtain the following integral:

T [ ( * r + » ) ,] - " + 7 ’ - * <«*>
which determines the value of the pressure p ( x , y ) . If the capacity
forces are absent and if we assume that g = 1, we obtain the formula:

(105)
where | V | denotes the velocity.
Note that if instead of f ( z ) = <p + iip we take the complex potential
i f ( z ) = — ip -f- irp, then the equipotential lines will be transformed into
lines of current and vice versa. Hence e v e r y i s o t h e r m ic n e t o f a r e g u l a r
f u n c t i o n e s s e n t i a l l y g i v e s t w o d i f f e r e n t p i c t u r e s o f th e f l o w o f a l i q u i d .

42. Examples. 1. All examples of isothermic nets which we considered earlier


can now be interpreted from the point of view of hydrodynamics when, as we
have shown above, every example gives two hydrodynamic pictures.
Let us now consider some other examples. To begin with we consider the ele­
mentary function
/ (z) = A log (z — a) = A log | z — a \ -+- iA arg (z —a),
where a is a point of the plane and A is a real constant. In this case the equi­
potential lines are circles, centre a t a, and the lines of current are straight lines
which originate at that point. The function /(z) acquires the constant i2 nA by
describing this point and therefore the imaginary part of the complex poten­
tial <p(x, y) (the function of current) acquires the term 2jiA. The velocity vector
is determined by the complex number

If we denote by q and <p the modulus and amplitude of the complex number
z — a, then the velocity vector corresponds to the complex number Ae'^/g.
One result of this is that on approaching the source, the velocity tends to
infinity and, when A is positive, this velocity is directed from the source to
infinity, i.e. we have a source but no flow.
42] exam ples 175

Let us now consider the more general function:


z — a +. l.^Aa r g _z _—_a
f(z) = A log * = A log ( 106)
z —b
where a and b are distinct points in the plane and A is a real constant. In this
case the isothermic net is defined by the equations:
= Ci ; arg - ■= C,
As we know the first of these equations corresponds to a family of circles
with respect to which a and b are symmetrical and the second equation to
y
I

a family of circles which pass through the points a and b [31]. In the case
under consideration we have a source of intensity 2nA at the point a and a
flow of the same intensity at the point b.
2. Let us suppose that the points a and b lie at the points —h and 0 on
the real axis and take A = 1/h. In this case the function (106) has the form:
/ (2) = log (z + h) — log z _

Taking the limit as h 0 we obtain the complex potential which charac­


terizes a so-called dipole a t the origin:

I t can easily be shown that in this case the isothermic net consists of circles
passing through the origin and touching the Y axis (equipotential lines) and
of circles passing through the origin and touching the X axis (lines of current)
(Fig. 45) [31].
176 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [42
3. Consider the function:
/ (s) = iA log (z — a) = — A arg (z — a) + iA log ] z —a \,

where A , as before, is a real constant. In this case circles, centre at a, serve as


lines of current and the straight lines which radiate from the point a are equi-
potential lines. By encircling the point a in the positive sense the real part of
f(z) (velocity potential) receives an increment —27iA and we have at the point
A an elementary turbulence of intensity —2nA.
4. Take the function
/ ( 2) = - ( 2 + - j , (107)

which we investigated in [33], Separating the real and imaginary parts we obtain
an equation for the lines of current in the form:

~z[y ~ * 4 , 0 = c
or
ky [x2 + y t - l ) - 2 C (a;2 + y * ) = 0.
In the general case these lines are certain curves of the third order. In the
particular case when C = 0 we have a circle x 2 + y2 = 1 and the axis y = 0.
We are only considering th at part of the plane outside the above circle. We can
say that the lines of current consist of the lines (— oo, —1) and (1, oo) on the
y = 0 axis and of the above circle. In this case we have considered the flow
of liquid outside the circle with the liquid circulating round the circle. Evalua­
ting the derivative

we see th at the velocity of flow at infinity is equal to k/2 (where k is real) and
this velocity is equal to zero at the points z = ± 1, i.e. at points where the lines
of current enter the circle.
We now add a logarithmic term to our function and thus construct a new
function
L{z) = ^ { z + ~ ) - i A \ ° Z Z . (108)

The real part of the second term also remains constant on the above circle,
i.e. this circle, even for the complex potential (108) is one of the lines of current,
but in the case under consideration the velocity potential receives an inorement
2 tiA on encircling this circle, i.e. the potential (108) gives the flow round
our circle with an elementary turbulence. Figures 46,, 462 and 46a give the
appearance of the lines of flow for various values of the constant A/k. The
flow represented in Fig. 462 shows th at the points of entrance and exit of the
lines of flow coincide on the circle round which the liquid circulates.
5. As we saw earlier in [33] the isothermic net for the function
/(z) = arc cos zjk consists of confocal ellipses and hyperbolae with foci at
42] exam ples 177

on the real axis. This net is shown in Fig. 47. If we take the hyperbolae as the
the lines of current we obtain the picture of flow through the aperture
(—le, + k) on the real axis. If we take the ellipses as the lines of current we
obtain the picture of flow round the ellipse or round the line (—k, -(-fc).
6. Frequently when studying the hydrodynamic picture it is more con­
venient to give not the complex potential w = /(z) but its inverse function

F ig . 46a

z = <p(w). Consider an example of this kind. Suppose that the complex poten
tial is given by its inverse function
z = w -(- e“ .
Separating the real and imaginary parts
z = x + iy; w = <p+ iip,
wo have:
x = (p + eip cos y>; y = xp + e<psin ip.
178 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [4 3

Assuming th at y> = C we obtain an equation for the lines of current in the


parametric form
* = <p+ e'p cos C ; y = C -f- e9 sin C,
where g> is the variable parameter. Consider two lines of current, viz. the lines
corresponding to C = n and C — —n. In the first case we have
x = q>—elp; y = Jt.
I t can easily be seen th at in this case the lines of current consist of a double
line — 00 < x < —1 on the line y = n. In the second case when G = —n the
lines of current consist of a double line — o o < * < — 1 on the line y = ~ n.

Also, when C? = 0, the axis y = 0 itself serves as a line of current. Figure 48


represents the appearance of the lines of current in this case.

43. The problem of flow round a contour. Suppose that we are given
a simple closed contour 2 in a plane and that we are investigating
the flow of liquid outside this contour, which must satisfy the follow­
ing two conditions: (l)the contour I must be one of the lines of current
and (2 ) the velocity at infinity must be finite and have a definite
direction. It is also necessary that the complex potential f(z) be a
single-valued function. We assume that the velocity at infinity is given
by a positive real number c (i.e. we choose the positive direction of
the real axis as the direction of the velocity at infinity).
Suppose that we know the function which conformally transforms
that part of the z-plane outside I, into the outside of the unit circle
| r | > 1 . We know that there is an infinite number of such functions
and wo choose the function which translates the point at infinity into
itself and has no direction at that point. o/(°°) is a real positive number
44] N. B. ZHUKOVS K IJ’B FORMULA 179

for this function and we have the following expansion for it in the
neighbourhood of the point z = <=°:
r = co (z) = bz + bQH— + ... (6 > 0 ). (109)
As we already know, the complex potential in the problem of flow
round a circle, can be expressed as follows:

/ i W = t (t + t )- (110)
where k is a real constant which we shall determine later. If we sub­
stitute r by its equivalent expression from (109) in the equation (110)
we obtain a single-valued function which is regular outside the contour l\
its imaginary part remains constant on the contour I, in the same way
as the imaginary part of (110 ) remained constant on the circle | r | = 1 :

/ (* )= h [® ( * ) ] = 4 [«> (2) + - $ ] ■ (in)


The constant k only remains to be chosen so that the velocity at
infinity is equal to c, i.e. so that f'(°°) = c. From the formulae (109)
and (110 ) we must have at infinity

]»'<*) “ d f H = r 1'
from which it follows directly that we should take k = 2cjb. We thus
see that the problem of flow round a contour involves the conformal
transformation of that part of the plane outside that contour into the
exterior of a unit circle.
It can be shown that if the function f(z) is single-valued the solution
of the problem is unique provided that f(z) has no singularities outside I
other than the simple pole z =

44. N. E. Zhukovskij’s formula. Let f(z) be the complex potential


which gives the flow round the contour I and let the velocity at in­
finity be equal to the positive number c. We assume that f(z) is not a
single-valued function but that in describing a circuit round the
contour I its real part <p{z, y) gains a constant term y. The component
of pressure on the body about which the liquid circulates can be
expressed by line integrals:
Fx = $p{x, y) cos (», *) ds; Fy = J p (x, y) cos (n, y) ds , ( 11 2 )
i t
where p(x, y) is the pressure and n is the direction of the normal
inside the contour.
180 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [44
The element of the contour ds, like the vector, corresponds to a
a complex number dz = e'9 ds, where 9 is the angle between the
tangent to the contour and the OX axis. The multiplication of a
complex number by i is equivalent to the addition of ti/2 to the ampli­
tude and therefore the complex number ieie ds corresponds to a
vector ds, directed along the inside normal to I, and we evidently
have:
F x + iF y = J p id z . (113)
i
According to formula (105)

p= o - 4 i / - (S) p - c - i . | - g
and therefore
Fx + iF y = i J C dz — y1 *Jri dz
d/ dz.
I i
It is obvious that
J dz = 0 ;

it is convenient to change to complex conjugate values in the above


equation, after which we obtain:

= | ‘ d2 = Y i J ^ d/ - ( 114>
i / i
The contour I is a line of current, and therefore y>(x, y) is a constant
on this line; y>(x, y) = Cx and therefore, on I:
f{z) = <P(x , y) + iClt f[z) = <p(x, y) — iCv
from which it follows that df = df. Multiplying both sides of (114)
by i we obtain a complex value which fully characterizes the vector
of the total pressure upon the body:

B = F , + iFx = - ^ % i f .
I
or finally:
A = F > + i ^ = - i | ( i ) !di . (115)
I
The function f'{z) is regular and single-valued outside I. In the neigh­
bourhood of infinity it can be expanded as follows:

/ , ( 2 ) = c + 4 l + ^ - + •••> (116)
46] TH E TWO-DIMENSIONAL PROBLEM OP ELECTROSTATICS 181

ffhere c is the given value of the velocity at infinity. For the function
f(z) we have the following expression in the neighbourhood of infinity:

/ (z) = C + cz + b±logs — A - + . . . ,

and by describing a circuit in the positive sense around I, the function


j{z) acquires the term %2nbv which we earlier denoted by y. Thus
= (l/ 2m) y and instead of (116) we can write:

n*> = « + T C + - ? - + - - -

whence, by squaring, we obtain an expansion in the form:

[/'(*>]* = + 4- + -... (H7)

When evaluating the integral (114) we can, as a result of Cauchy’s


theorem, integrate not round the contour I but round a closed curve
which surrounds I and which lies in the neighbourhood of the point
at infinity. We can then use the expansion (117) when we obtain the
following expression for R:
R = Fy + iF x = - -^r- 2ni = - cy,
i.e.
Fy = — cy; Fx = 0 . (118)

45. The two-dimensional problem of electrostatics. We shall now


apply the theory of functions of a complex variable to the problems of
electrostatics. We frequently meet here problems analogous with
those considered above. First of all we shall explain what is meant by
the two-dimensional problem of electrostatics. As we know the point
charge e creates lines of force in space which act in accordance with
Coulomb’s law and the intensity of the field is expressed by the well
known formula

where q is the distance from the charge e to the point M, at which the
vector of force is determined. This vector of force takes the direction
of the line connecting the charge and the point M. Imagine now that
we have a charged straight line parallel to the Z-axis which crosses
the X Y -plane at the point 0, and that the density of the charge is the
same at every point. Denote this charge, which is proportional to
182 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [45
unit length of the line, by e. The picture of the electrostatic field is
obviously the same in all planes parallel to the XT-plane; therefore it
is sufficient to consider the XT-plane alone; here again, as a result of
the principle of symmetry, the vector of force must lie in this plane
and take the direction of the line joining the point 0 with the point
M in the plane at which the force is calculated. The elementary charge
on the section dz of the straight line is expressed by the product e • dz,
and the value of the force at the point M with coordinates (x, y, 0 ) by
the sum of the projections of component forces multiplied by the
direction O M of the above line.
We have the following expression for the force:
edz
+ j/2 -f- 22 ’
when 0 is the origin. The above expression must be multiplied by
the cosine of the angle f , made by the direction N M , from the variable
point N on the X-axis, and the direction O M \ from the right-angled
triangle O N M we have:

cos <p= __ and z — r tan <p,


Yx* + y2+ z2
where r = j/x2 -f- y2. Replacing the variable <p in the integral
T OO
e cos<pdz
J ** + V* + 2*

by z we obtain the following expression for the force:


n
2

/ = -y J cos <pdip
n
2
or
f = ^~ (r = R T ? ) . (H9)
The corresponding potential of force is:
F(x,y) = 2e l o g ^ , (120 )

where r 0 is an arbitrary constant which we assume to be positive.


Hence the logarithmic potential (120) is the elementary potential
and it originates, as it were, from this point charge, if we disregard
the whole space and consider the XT-plane alone. Note that this
45] TH E TWO-DIMENSIONAL PROBLEM OP ELECTROSTATICS 183

elementary potential (120 ) does not vanish at infinity like the usual
Newton’s three-dimensional potential 1jr, but becomes infinity; this
ia the essential difference of two-dimensional electrostatic problems.
If, instead of a charged line, we have a charged cylinder, the base B
of which lies in the X Y -plane, then instead of the elementary potential
(120 ) we obtain a potential expressed by a double integral

V (x, y) = 2 j j e ( £ , i?)log — df dy, (121)


B

where q(£, y) is the density and r the distance from a variable point
(f, y) in the domain B to the point M{x, y):
r= — * ) 2 + {y — y)2.
Similarly, if the surface of a cylinder is charged, then the potential
is expressed by a line. We also know that the functions log r and
(120) satisfy the Laplace equation [II, 119):
82 V 82 V _
8aj2 ‘ 8i/2

The potential (121) outside the charge, i.e. outside the domain B,
also satisfies this equation.
We can assume that any harmonic function is the real or imaginary
part of a regular function of a complex variable. In this case we shall
consider the potential V(x, y) to be the imaginary part of a regular
function
f(z) = U(x, y) + i V( x, y) . (122 )
Hence every electrostatic picture outside the charge gives a regular
function f{z) (complex potential) and, conversely, any such regular
function gives the electrostatic picture of the two-dimensional field.
In this case both families of the isothermic net of functions
U{x, y) = C1-, V(x, y) = C2 (123)
have a simple physical meaning. The second family in (123) gives
a family of equipotential lines and the first, which as we know is
orthogonal to the second, gives a family of lines of force, i.e. it
gives lines the tangents to which define at every point the direction
of the acting force. The components of the vectors of force can be
expressed as follows
184 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [46

or, from the Cauchy-Riemann equations:


8F 8U
dx
F
J y= dx
Thus the vector of force corresponds to the complex number
av 8U
Fx + iF y =
dx dx (124)
If we have a closed finite conductor then inside this conductor, as
we know, the potential remains constant and the density of charge
on its surface, as proved in electrostatics, is calculated according to
the formula:
e = ^ -] /F i + F l

or, with the aid of the complex potential, by the formula

<?= — |/'(z )|- (125)


The analogy between the above concepts and the corresponding
concepts in problems of two-dimensional hydrodynamics can readily
be seen.

46. Examples 1. All examples of isothermic nets which we considered above


can now be interpreted from the point of view of electrostatics. Consider, for
example, the function
/(z) = i 2 e l o g ^ | - . (126)

Its imaginary part remains constant on the circles with respect to which
the points a and b are symmetrical [31], Take two such circles C2and C2and sup­
pose th at the imaginary part of the function (126) has constant values F t and F2
on these circles. If we imagine two cylinders formed by lines parallel to the
z axis for which the above circles serve as bases, then the complex potential
(126) gives the picture of the electrostatic field between two such cylinders,
where Ft and F2 are the values of the potentials on the respective cylinders.
Notice th at in the general determination of the electrostatic field in an annulus
between two conducting curves ll and lz we have a complex potential, the
imaginary part of which remains constant on the curves Z, and Z2. Thus the
complex potential f ( z ) transforms the above annulus into a strip, bounded
by two straight lines, parallel to the real axis. Such a transformation cannot
be single-valued, for an annulus is a multiply connected region and a strip is a
connected region. The function (126) in the above example is evidently many­
valued in the annulus confined between the circles Gl and C2.
Note one other property of the field defined by the function (126). This
function can be written as follows:
/ (z) = i 2e log (z — o) — i 2e log (z — b) .
46] EXAMPLES 185

By using this expression it can be shown th at both conductors have equal charges
0f opposite sign. In agreement with this the function (126):

/ (z) = i 2e lo g ------- £ -
^ z
will be regular at z = oo.
2. If we want to determine the electrostatic field between two conductors,
each of which goes off to infinity (Fig. 49) then the domain between these conduc­
tors will be connected and the problem
essentially involves the transformation
of the region into a strip bounded by
two straight lines, parallel to the real
■axis. Thus, for instance, when these
lines are the lines (—oo < x < —1) on
the straight lines y = n and y = —n,
then the formula z = w + e"' gives
the inverse function for the un-

Fio. 50

known function and Fig. 48 gives the picture of equipotential lines in this case.
Notice th at at the end of our lines we have w = and ew = —1. But
formula (124) gives the magnitude of the force as:
Aw dz
Y * l + n = I/' (*>I dz du>
in this case
in + = 11 + ewr 1,
i.e. at the ends of the above lines the force becomes infinity.
This is a special case of a more general example which we shall now consider.
Assume th at our two conductors have the appearance shown in Fig. 50: A B
and AC are two parallel halves of straight lines, so th at the points B and C lie
on their common perpendicular. The directions BD and CD make the same
angle a with A B and AC, where a = fin. Draw a straight line PQ, parallel
to the above straight lines at an equal distance from either line. P art of the plane
bounded by PQ, A B and BD can be regarded as a triangle and the angles at
the vertices B and P are respectively equal to ((t + 1) ji and zero. Let us trans­
form this triangle into the upper half-plane and let the vertices B, P and Q
correspond to the points t = —1, 0 and oo. Using formula (47) we have

z = a J (r + 1)'' t xdr, (127)


o
186 CONFOBMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [46

where the constant a can be regarded as positive; this can be achieved by rota­
ting the 2 -plane. The z-plane is th at shown in Fig. 50 and r corresponds to the
plane in which our triangle is represented by the upper half-plane. If we reflect
the above triangle in the line PQ, then the half-plane will be reflected in the line
0 < r < +oo on the real axis, and part of the z-plane between our two con­
ductors will be reflected in the r-plane as a plane with the cut ( —oo, 0). If
we now assume th at
T = e“’,

and recall the image produced by an exponential function [19], we obtain


a strip in the w-plane, bounded by straight lines parallel to the real axis and
at a distance jr from it, i.e. we obtain the strip
— 71 < I [id ] < 71,

where I is the symbol of the imaginary part. Thus w, as a function of z, has


a constant imaginary part f j i o n our conductors and represents the complex
electrostatic potential of the two-dimensional field between our two conductors.
Owing to the fact th at t = e* we can rewrite (127) in the form:
H>
z = o J (ow+ l)"du>. (128)
The value of the constant a evidently depends on the distance between the
straight lines A B and AC. Suppose that this distance is equal to 26. For points
in the z-plane near the point at — oo, the isothermic net giving the
families of equipotential lines and the lines of force will evidently be close
to the net of Cartesian coordinates and this net corresponds to the net of Carte­
sian coordinates in a strip of width 2ji in the w-plane. The point z = —oo
corresponds to i = 0 and therefore to w = —oo. Bearing in mind th at the width
of the strip in the z-plane is 26 and th at when w tends to —oo the function
e* tends to zero and, consequently, from (128), dz/dw tends to a, we can see
th at the value of the constant a is equal to 6/jr. Consider the particular case
when ft = 1/2, i.e. when the straight lines BD and CD are perpendicular to
A B and AC.
In this case we have:
w
z = — J*Yew + 1 dty, (129)

whence the integral can easily be evaluated by substituting e"1+ 1 = tt .


The absence of the lower limit of integration means th at any constant
number can be added to z, i.e. it involves a parallel transition of the z-plane,
which is of no significance.
3. Let I be a simple closed contour which is the trace of the cylindrical
conductor in the XF-plane. Let e be a given charge on that conductor, per
unit length along the z-axis. We have to evaluate the two-dimensional electro­
static field outside I. Reflect th at part of the z-plane outside I onto the outside
of a unit circle in the r-plane, i.e. into the domain | r | > 1; we take it that the
point at infinity is translated into itself, so that in the neighbourhood of z = «>
the expansion of the function which performs the conformal transformation
46] examples 187
ia as follows:
T = CO(2 ) = CZ + C0 -[---H-------f + ■• • (130)
z z‘
We shall show that it is possible to construct the complex potential for our
electrostatic field from a knowledge of the above conformal transformation.
In fact, let us construct the function:

/ (z) - i 2e log [t = <0 (z)],

where t 0 is a constant which plays no essential part. The imaginary part of the
above function will obviously be:
I [/ («)] = 2e [log | t 0 | - log | r |].
' Since | t | = 1 lies on the curve I we can say th at the imaginary part remains
constant on I. Let us now determine the value of our potential in the neighbour­
hood of the point at infinity. From the expansion (130) we have the following
expression for /(z) near z = 0 0 :

/ (z) = — i 2e log z + d0 + — + • • •
z
The first term of the above expansion gives the potential —2e log | z | which,
according to (120), corresponds to the given quantity of electricity on the con­
ductor. Formula (126) gives the following expression for the density of distribu­
tion of the charge along I:
1 , e 1 dr
(*)i 271 T dz
or, since | t | = 1:

/ e = 2ti
e dr
dz
e
2n
d z —l
dr (131)

If / is a square then the relationship between r and z can be written in


the form [38]:
J Yt*+ 1 dr, (132)

when we take on the circle | t | = 1 the points


[ (?t + 2kn)i \
exp -----— ------------ (fc= 0,1, 2, 3),

which correspond to the vertices of the square.


Formula (132) gives:
dz , Vtj + 1
dT = h— T ~ ’
and, in this case, we can rewrite (131) as follows:

£ — 2 71 a / t*+ 1
188 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [47

■where a is a constant which depends on the length o f side o f the square.


On the contour o f the square we have t = e^, where <p is a polar angle of the
unit circle. A side o f the square corresponds to the change o f this angle
through the interval (ji/4, 3ji/4) and we therefore have the following expression
for the length o f side o f the square:
3*
4 _______
r fan? .p i .
s= a j — ie d?’
n
4
from which, after performing a simple change of variables we arrive without
difficulty a t th e following formula which connects the lengths of the sides of
the square a w ith the constant a:
It
2
s = a ] /2 ^ )^cos#d#. (133 )
o
The integral on the right-hand side cannot be expressed in a finite form and
belongs to a class o f so called elliptic integrals.

47. The two-dimensional magnetic field. Above we explained the connection


between the analytic functions o f a complex variable and a two-dimensional
electrostatic field. In exactly the same way we can consider a two-dimensional
magnetic field originating from infinite straight currents, which are perpendicular
to the X T -plane. We shall give the fundamental results which apply here
without proof. For the vector of magnetic force we have the formulae:
dtp
H Hy = (134)
dx'
where the function tp satisfies the Laplace equation outside the field’s source
and is therefore the real part o f a certain analytic function:
f (z) = tp + iip. (135)
Using the Cauchy-Riem ann equations we can rewrite (134) in the form:
dip
Hy
dip
&T : w
or
H = — grad ip,
so that ip is the potential of the field.
The lines tp(x, y) = Cl which are orthogonal to ip(x, y) = C2 are thus lines
o f force o f the field. When we have one straight current o f force q directed
along the Z axis, we have for the function (135):
/ ( 2 ) = — 2g log z ,
hence tp = —2q log r and ip — —2q arg z. The lines of force are circles, centre
the origin, and by encircling the origin the potential receives an increment
( —47iq). In this case one o f the lines <p(x, y) = C2 m ust lie on the surface o f the
magnetic conductor (magnetic permeability equal to infinity).
48] SCHWARZ’S FORMULA 189

48. Schwarz’s formula. The above applications of analytic functions


0f a complex variable to problems of hydrodynamics and electrostat­
ics were essentially based on the close connection existing between
harmonic and analytic functions of a complex variable. We men­
tioned this earlier in [2],
Let us formulate once again the main points: the real and imaginary
parts of an analytic function are harmonic functions and, conversely,
every harmonic function can he regarded as the real part of an analytic
function; its imaginary part can be determined accurately as far as the
constant term, i.e. the function can he determined as far as the purely
imaginary constant term. From what was said earlier in [II, 194] a
harmonic function is defined uniquely in a connected domain by its
limiting values on the contour of that domain (Dirichlet’s problem).
We can therefore say that the function f(z), which is regular in a domain
B with contour I, is defined accurately as far as its purely imaginary
constant term hy the given values of its real part on the contour I. There is
no simple formula which would give the solution of this problem for
every domain, i.e. which would define the regular function from the
given contour values of its real part. It is not difficult however to
construct such a function for a circle and we shall do this now.
Consider a circle, centre the origin and radius R. Let u(x, y) be the
part of the unknown analytic function. This harmonic function is
defined from its contour values u(cp) by Poisson’s integral, which as
we know, has the following form [II, 196]:

= = {r< R ). (136)
—n
It can easily be seen that the core of Poisson’s integral i.e. the frac­
tion integrand, is the real part of an analytical function viz.:

= reai \ ^ + 21 (z = rei&—-x -\-iy ).


B? — 2r R cos (cp — &) + r 2 L .Re*?’ — z J

If, instead of Poisson’s core wre substitute an analytic function of


the complex variable z in the integral, we obtain a function of the
complex variable z, the real part of which coincides with u(x, y). This
function is given below:
n
f{z) = u {x, y) + iv {x, y) = (v) J« d(p- (137)
190 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [48
By assuming that z = 0 in this formula we obtain a purely real
value for f(z), i.e. formula (137) gives one solution of our problem
where the real value is at the origin. If we denote by Ci the imaginary
part of the unknown function at the origin, then the general solution
of the problem takes the form:
n
(?) b I iI - Y d(P + d . (138)
— 71

This formula is usually known as Schwarz’s formula.


If we separate the imaginary part of the fraction in the integrand
. , I" Rely -f- z"I__ 2tR sin (# — <p)
im. par ^ i<p_ z j Jt2— 2rR cos (<p— &) r2 ’

we obtain an expression for the imaginary part of the regular function


in the circle in terms values of its real part on the contour:

V(*’ y)= i r —f71UM + d(P+ C- (139)

All we said above is closely connected with the concept of conjugate


trigonometric series.
Let
eo
-~- + ^ ( an cos n<p + bn sin n<p)
z n=l

be a Fourier series of the function u(<p) representing the limiting values


of the real part of f(z). As we know from [II, 195] the same real part
can be represented in the circle not by Poisson’s integral but by a
series of the type:
to

u (x, y) = u (r, &) = — + [an cos n & + bn sin n &) rxt. (140)
1 n =1
We have for the imaginary part of a conjugate trigonometric series
[25]:
v(x,y) = v (r, ■&) = C + ^ (— bn cos n ■&-f an sin n d) r" . (141)
n=l

If the function u{<p) has satisfactory properties, for example its first
derivative satisfies Dirichlet’s conditions, then the series (141), like
the series (140), is uniformly convergent in the whole closed circle and
the function v(r, 9) is harmonic in the circle and continuous in the
49] TH E CORE COT (s - 1)12 191
closed circle. v(r, 6) is usually known as a function conjugate with u(r, 0)
[2], and the same name also applies to all its limiting values v(\, q>)
with respect to u(<p).
Assume that two Schwarz’s integrals give one and the same function
which is regular in the circle
n n
I f , > Retv + z , I f . . i?e‘> + z ,
*T J M J “2M d?- (142)

where mx(<p) and u2(cp) are continuous real functions. It can be seen
that these functions coincide for they are the limiting values of the
same harmonic function, viz. of the real part of our regular function.
Therefore the identity (142) of z is equivalent to the identity u^rp) = u2(<p)
of <j>. This is essentially Harnak’s theorem which we mentioned in [8].

49. The core cot (a—<)/2. We shall now apply the fundamental theorem of
limiting values of Cauchy’s integrals [28] to the circle | z | = 1, centre the origin
and unit radius. Assume th at we are given a real function w(r) on this circle
where r = els, which satisfies a Lipschitz condition. By using Schwarz’s for­
mula [48] we can construct a function which is regular in this circle and the
real part of which has the limiting value v ( t ) on the circle:
n
u (re,9) -f- v (re^) i = J u (t) * da (z = retv'), (143)

or, since d r = irda,

u (re'*) + a (re'>) i = J u (T) dr.


M -i
Putting r- [-z = (r — z) + 2z and breaking the right-hand side up into two
integrals we obtain:
71

u(rel*) + v(rel*)i = ± - j u (t) ds + - ^ J ^ . _ i_ dr.


—» lrl = i
Let us suppose th at the point z = ro"p tends to a point £ = ei( on the circle
| z | = 1. Using the theorem of limiting values of Cauchy’s integrals [28] we
obtain the limiting value for our function:

u (e1') + v (e") * = ^ - [ u (r) da + £ — ^ + - |i [ - ^ ■ - ^ j d r .


M=i
n
u (e") + v (e“) i = J u (r) da + u (£) + J u (r) ~ —j d®- (I*4)
192 CONFORMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [49
but
2g _ 2e“ — 1 + * cot
I —a
t —f e 's — eif 2
and by separating the imaginary part in (144) we obtain a formula where the
limiting values of the imaginary parts are expressed in terms of the real part:
n
V (e") = — J U (e,s) cot da,

and where the integral must be taken in its principal value sense. We shall
write u(a) and v(t) instead of u(e‘s) and lie'1):
n
1 — 8
cot d$. (145)

We recall that (143) gives a regular function in the circle | 2 I < !. the imaginary
part of which is zero at the centre of the circle. Bearing in mind that the value
of a harmonic function at the centre of a circle is equal to the average arithmetic
mean on the circle [II, 194], we can write:
n
J v (<) dt = 0. (146)
—n
The function u(a) is a periodic function of period 2ji and the function v(t)
is also obtained in periodic form; in formula (145) we can therefore take any
period, 2n in length, for the interval of integration. The function cot z, has
a simple pole with unit residue when z = 0, [21] and we can express the core
of the linear transformation by Cauchy’s core:

i cot ± ^ = _ _ ^ _ + P (t_ s)i (147)

where P(z) is an analytic function, regular at all points on the line


2jt < z < 27i. I t can be shown in the same way as in [27] that if the periodic
function u(a) satisfies a Lipschitz condition of order a, then v(t) also satisfies
a Lipschitz condition of the same order, when a < 1, or of any order smaller
than unity, when a = 1. I t appears from (147) th at this statement can also
be derived from an analogous case for Cauchy’s core.
On applying the linear transformation (145) to the function v(t) we obtain
a new function which satisfies a Lipschitz condition:
n
w^ = J v ^ cot ' V di-
- ji
The function ir(<,) gives the limiting values of the imaginary part if we take
v(t) for the limiting values of the real part when:
n
f w (£,) dq = 0. (148)
—n
49] TH E CORE COT (j - 0/2 193

On the other hand, if the regular function (143) is multiplied by (—i) we obtain
the regular function v(rellp) — u(rellp) i. Having thus taken the real part, the
imaginary part can be determined accurately as far as the constant term and
we can therefore ye&te:
w ( t,) = — u (<,) + G.
To determine the constant C we integrate both sides of this equation over the
interval ( —n, + 31) and, remembering (148) we have:
n
0 = — j" u (tY) + 2n C,
and finally:
n 71

dt = —u («,) + J u (a ) da, (149)


— 71

i.e. the two applications of the transformation (145) give us minus the original
function accurately as far as the constant term. The result can be written in
the form:
71 r n 71

1 u (a) cot
t —a
d a cot d1= J u (a) d s. (160)
47l2 2 U{t') + ~2^
—n
This formula is known as Hilbert’s formula and the core of the transformation
(145) is usually known as Hilbert’s core. Notice that on the left-hand side of
formula (149), as in Fourier’s integral, we cannot change the order of integra­
tion. Denoting the transformation (145) by a single letter h we can write
formula (145) in the form:
v(s) = h [u (a)],
where a denotes the amplitude of both functions. In this case Hilbert’s formula
(150) can be written:
n
h? [u (a)] = u ( a ) ------ J' u da.
(a )

Formula (145) can be regarded as an integral equation of w(a), where v(t) is


the given function. I t follows from above th at this equation is soluble only when
the condition (146) is satisfied. One solution of this equation, according to
(149), will be given by the function:
n
u,(a) = — ~ J v (t) cot - 2- * df. (151)

This is also the solution of the equation (149) which satisfies the condition
71

J" u (a ) d a = 0 .
-JT
194 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [49
In other words, this is the imaginary part of the real function v(relip) — iu(re,lf)t
which vanishes at the origin. I f the value of the function u(reilp) at the origin
is equal to C then
n
u (3 ) = C ---- J v (t) cot a 2~ dt, (152)

where u(s) = const, is the solution of the homogeneous equation


n

for when u(s) — const., the imaginary part v, which vanishes at the origin,
must be equal to zero. Formula (152) gives all the solutions of the equation
(145), for the imaginary part is determined accurately as far as the constant
term in terms of its real part. We have assumed throughout that both the
given and the unknown functions satisfy a Lipschitz condition.
The transformation (145) can be written in the form of an indefinite integral
similar to the one used in the core of Cauchy’s integral. In fact, taking into
consideration th at
71

_ I fj cot t —
g « da
, = 0,
-----71

since the homogeneous equation (145) has a solution equal to a constant


u(s) = c we can rewrite the formula (145) in the form:
71

v (0 = — J [u (s) — u (I)] cot 1 2 8 d«. (153)

Let us suppose th at the function u{s) has a continuous derivative. Taking


into consideration that

and applying the formula for integration by parts in the intervals ( —n, t — e),
and (t + e, n) to the integral (145) and also taking into account the formula:

lim [u{t + e) — u (t — £)] log (sin2 4 -) = u' (f) 2s log sin2


• -. + 0 \ *) "
(t — e < e < t + e),
we obtain the following expression for v(t):
+n
v (/.) = — J (O' (S) log (sin2 1 2 ds,

where the integral on the right-hand side is undefined.


If w(t) satisfies a Lipschitz condition, then the function of the complex vari­
able z = relv, given by formula (143), is continuous as far as the circumference
50] LIMITING PROBLEMS 195

| 2 | = 1, as wo saw above. Let

-5- + J ? (ak cos k8 + bk sin ks) (154t)


1 k=1
be a Fourier series of the function w(s). For the function v(s) we then have the
following Fourier series [48]:

y (— bk cos ks + ak sin ks). (1542)


*=l
As a result of the equation [II, 147]:
n n
— f u2 (s) ds = (a* + bk) and — ( v2(s) ds = (bk -f <4),
71 J
— 71
2 fctl 51 J 71
k=1
and consequently:
ft ft
j" t’2 (s) ds < J w2 (s) ds,

where the sign of equality applies only when a 0 = 0. Thus, a6 a result of the
transformation (145), the integral of the square of the function in the interval
( —n, + 7i) can only decrease. Notice th at we have supposed that the
function u(s) is real. We can thus see that the transformation (145) is equivalent
to the transition from the Fourier series (154!) to the series (154.).

50. Limiting problems. Diricl-ilet’s problem is the simplest case of limit


problems in which harmonic functions are involved. Let us formulate the general
limiting problem for harmonic functions where Dirichlet’s problem is a particular
case: it is necessary to find a harmonic function in a connected domain B with
contour I, which satisfies on this contour a limiting condition of the form:

OM+6^ - + C l | ~ = d’ (155)
where a, b, c and d are real functions given on the contour I, which we assume
to be functions of the length of the arc s of th at contour. We also assume that
u is the real part of a regular function
/ (z) = u (x, y) + iv (x, y).
As we know:
0U . 8u
0£/ ’
and, consequently, we have

h -S ’ + ^ = ® l(b + ic) /' (2)]>


where H is the symbol of the real part.
196 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [50
Condition (155) can be rewritten in the form:
« [<z/ (z) + (b + ic) / ' (z)] = d, (1 56)

and therefore the problem involves the finding of a function, regular in B


which would satisfy the condition (144) on the contour.
Let z = tu(r) be the known function which conformally transforms our domain
B into the unit circle | t ] < 1 . We can assume that the unknown function
is a function F( t) which is regular in the unit circle
1
F (r) = / [co (r)] ; f ' (s) = F ' (t ) ■
oS (t)
We have instead of (144)

^ K (T)+- y ^ F,(T)] =d d Ti = i )-
where, as a result of the transformation z = a>(r), we can take it that a, b, c
and d are defined on the circle | r | = 1. Hence our problem simply involves
a circle.
Consider in greater detail the case when the limiting condition (155), which
applies to the circle \ z \ = 1, does not contain the unknown function u. In
this case the problem can be formulated as follows: it is necessary to find a
harmonic function u(x, y) in a unit circle, which satisfies on this circle a limit­
ing condition of the form:
, du 3u ,
6 a^ + c W = d-
We suppose that u is the real part of a regular functionf(z). In this case dujdx
and —duj'&y are the real and imaginary parts respectively of a regular function
f'(z) and the above problem is thus equivalent to the following problem, usually
known as Hilbert’s problem: fin d a function f(z) which is regular in a unit circle,
the real and imaginary parts of which satisfy on the circle a limiting condition of
the form:
I (<p) u {ip) + m (<p) v (<p) = d {<p) (0 < 95 < 2n) , (157)
where l(<p), m(<p) and d(<p) are the given functions of the polar angle <p on the
unit circle. We assume th at the coefficients are continuous functions and that
l(<p) and m(ip) do not vanish simultaneously. When both sides of the equation
(157) are divided the coefficients satisfy the condition:
I2 (<p) + m* (<p) = 1. (158)
We can assume:
I (<p) = cos ft) (<p); m (<p) = — sin to (<p) , (159)
where a>(<p) is a function of <p, viz.

to (tp) = — arc tan


m (<p) ( 160)
I (<p)
Let us consider in detail the case when formula (159) gives 10(95) as a single-valued
function of <p. This will be so, for example, when l(<p) and m(<p) do not vanish in the
50] LIMITING PROBLEMS 197

interval ( —re, +?t). Using the function <0 (95) we can write the limiting condition
(157) in the form:
f (z)\ = d (<p) (z = e'*). (161)
Let us construct a function re(z) from its real part co(<p) by using Schwarz’s
formula:
n
{z) = i J 10 (y) d?I’ (162)

Denote by wx{(p) the limiting values of its imaginary part. The function
/(*)
has a real part on the unit circumference z = efq>which is equal to

e - “i<*’> a [ e imMf{z)]z = eiv

and, consequently, the limiting condition (161) is equivalent to the following


limiting condition:
0>[einMf(z)] = d(<p)
By knowing the real part on the contour we can define the function inside
the contour by again using Schwarz’s formula
n
e ^ )f(z ) ~ - L J d ( , ) e—*<*> d<P+ iC,

where (0 ,(95) are the limiting values of the imaginary part of the function (162):

co, (<p) - lim 3


r->l Jr 40 <*>
, , -&elv
e'vZ+Z re^ dy (163)

where 3 is the symbol of the imaginary part.


Finally we obtain the following expression for /(z):
n
/ (s) = e i - f d (<p) e~a' M eiV + Z dp + iC (164)
2re J e‘V — z Y
—n
Consider now the case when the function co(p) acquires the increment
(—2 rere) on describing a circuit round the unit circle, where re is a positive integer
co (re) — co ( — re) = — 2rere. (165)

Let us construct a single-valued function on the emit circle:


X (<P) = <*>(V) + n<p,
and construct a similar corresponding function for the complex variable 5(z),
for which X(rp) is the limiting value of the real part. The limiting values of
198 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [50
the real part of the function
Cj (z) = a (z) + in log z
are equal to to(<p) and the limiting values of its imaginary part are evidently the
same as those of the function <5(z). We denote them again by £ul(p). I t can be
shown similarly th at the limiting values of the real part of the function
e '^ l / f z ) = z - 'V ',(2)/(z).
must be equal to
r2[z~neia(2)f(z )] = d ((p )e -a,lM (166)
Owing to the presence of the factor z~n this function can have a pole of an
order not higher than n a t the origin. To begin with, let us construct with the
aid of Schwarz’s formula a function, regular in a unit circle, the limiting values
of the real part of which are
71

— \ d ( V) e— *M ^ 1 ± L d<p + iC. (167)

We must add to the above function a term, the real part of which is equal
to zero on the unit circle but which can have a pole of order n at the origin.
I t can readily be seen th at this term will have the form

2j + iB* ( l F + **)]>
where A k and are arbitrary real constants.
Adding the latter expression to the expression (167) we obtain the general
solution of the problem

/(z) = z ne - ,0«

+ T5T del. (168)


-» J
When n is a negative integer in formula (165) the solution of the problem
will be different. The function under the symbol of the real part in the expres­
sion (166) will then not only be regular in the unit circle but it will also have
zero of order n at the origin. When constructing a regular function on the right-
hand side of formula (166) with the aid of Schwarz’s integral, we must also
write down the condition th at the function obtained must have a zero of order
n at the origin. We therefore have several conditions which must be satisfied
by the function d(<p) if the problem is to be solved.
Consider another particular case, viz.: suppose that the limiting values of a
harmonic function on the unit circle have the form :
du du
-Q^ + l ^ + mn = d W ' (169)
60] LIMITING PROBLEMS 199

•where I and m are constants, d(<p) is the given function, n is the direction of
the outside normal to the circle and s is the direction of the tangent to the
circle. Instead of taking derivatives along the axes of coordinates we take
them, in this case, in the directions connected with the boundary curve which
are indicated above. As we know from [II, 108] these derivatives are expressed
in terms of each other. The limiting conditions, as expressed by formula (169),
are frequently used in mathematical physics. Differentiation along the normal
n coincides with differentiation along the radius-vector r, and differentiation
along 8 coincides with differentiation with respect to the polar angle <p, when
r = 1. In general by assuming th at z - reiip, and u = |/ ( 2 ) |, we have
the following, when 3 |/(0) | = 0:

-|jf- = «[*' /' (*')]; = «[*' if’ (2')] (*' = e'*),


and the limiting condition (169) can be rewritten in the form:
* [(1 + H) *' / ' (*') + mj (2 ')] = d (<p) (z* = e'*).

We multiply both sides of the equation by

1 2' + z
dtp
2n z' — z

and integrate with respect to <p. We then obtain a new equation which is equi­
valent to the one above [48]. Using Schwarz’s formula it can readily be seen
that this new equation will have the form:

( 1 + « ) * / '(a)+ « /( « ) - * ■ ( * ) , (170)
where

■F(z)
v'
= — f d (<p) — V—
2n J ^ eif —2 v
d<p- ——
—— f d(g>) -z' z-(2 '+—
2m J
Z - dz*.
2)
(171)
' '
12*1-1
Equation (170) is a linear differential equation of the first order. Solving
with the usual formula [II, 4] we obtain the following expression for the
unknown function:

/(*) = *"* C + ± j iS^ F{e)dz (172)


Zo
where
m
k = 1+ il •
In equation (172) the arbitrary constants can be determined from the condition
that the point z = 0 is a regular point for/(z). If

n
d (95) = A 0 ^ (A s cos s<p + B s sin s<p),
s= 1
200 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [51

then we have for F(z)


f ( 2 ) = A 0 + ^ ( A s - i B s)zs.
s-l
Substituting in (172) and integrating we finally obtain the following expan­
sion for f(z ):
Ao_ n As iBs zs
/(*) = m + 2 m + s (1 + il)

51. The biharmonic equation. We shall now consider the connection between
the theory of analytic functions of the complex variable and the theory of
the so called biharmonic functions, i.e. functions which satisfy the condition
44 u (x, y) = 0, (173)
where A is the Laplace operator, which expresses the sum of the second deri
vatives of the variables x and y (we are considering the two-dimensiona
case). Equation (173) can be written as follows:

I. dx2 ^ 9y2 ) I. 9x2 dy2 )


or
91u 9*u 91u _ -
(174)
dx* dx2dy2 dy*

Let u be a function, which together with its derivatives, is continuous in


a connected domain B where it satisfies the equation (174). According to (173)
the function
4 u = p (x ,y ) (175)
is a harmonic function. Suppose that q(x, y) is a conjugate function so that

P (*. y) + *?(*» y) = / (2) (176)


is an analytic function of the complex variable z = x + iy.
Construct the analytic function

<p(z) = —J /( z ) dz = r (x, y) + is(x, y). (177)

We obviously have:

4r = 4, = ° ; | L = | L = _ L 2 [,(2) ] = -L p . (178)

We now evaluate the Laplace operator for the expression u — (rx -f- ay). We
have from (178):
4 [u - (rx + sy)] = P ~ 2 ~ 2 = °-
51] THE BIHAR.MONIC EQUATION 201

i.e. the above expression is a harmonic function which we shall denote by p v


Introducing the conjugate function ql and the corresponding function of the
complex variable y>(z) = p v + iql wo can write:
u — (rx + sy) = p i; u = (rx + sy) + p L= 01 [(a: — iy) (r + is)] + p,
or
u = Ol\zq> (z) + y>(z)]. (179)
Hence according to formula (179) every biharmonic function can be expressed
by two functions of a complex variable. The converse can also be readily proved
by choosing arbitrarily the analytic functions <p(z) and y>(z), where formula
(179) gives the biharmonic function, i.e. formula (179), which contains two arbit­
rary analytic functions, gives the general expression for a biharmonic function.
This formula is usually known as Hurse’s formula.
For any given biharmonic function u the functions tp(z) and yi(z) in formula
(179) are not fully defined, for they do not contain arbitrary constants. The
real function q(x, y) is defined accurately except for the constant term, i.e.
the function f(z) is defined except for the purely imaginary constant term.
Also, when defining the function <p(z) in formula (177) the arbitrary constant
complex term must be taken into consideration. In its final form the function
<p(z) will contain arbitrary elements of the following form:
C + iaz,
where C is an arbitrary complex constant and a an arbitrary real constant.
These arbitrary constants can be determined when certain additional conditions
are made, for example conditions of the form
y (0 )= 0 ; g [q>' (0)] = 0 (g being the imaginary part) (180)
Similarly, when determining y>{z), we obtain the arbitrary constant as a
purely imaginary constant, which can be determined if the function yi(z) is,
for example, subjected to the condition
5 [» (0 )]= 0 . (181)
The conditions (180) and (181) fully define the functions q>(z) and y>(z)\ we
do, of course, assume that the point z = 0 belongs to our region.
Let us now consider the fundamental limiting problem for biharmonic
functions. I t can be formulated as follows: find a biharmonic function inside
a closed contour I from the given values of the function and its normal derivative
on this contour:
u = o), (s); - |^ = a>2 (») (on I) . (182)

We will show th at the limiting conditions (182) also give directly the limiting
values of the coordinates of the derivatives of the function u. In fact, we have
202 c o n f o b m a I j t r a n s f o r m a t i o n a n d t h e t w o -d i m e n s i o n a l f i e l d [5 1

■where 8 is the direction tangential to the contour I. Hence from the limiting
conditions (182) follow the following limiting conditions:

-g^- - a>[ cos (a , x) + co2 cos (n , x) = co3 (a),


(183)
= o>l cos (s, y) + Co, cos (n , y) = co4 (s).

The functions to3(s) and co4(s) cannot be taken arbitrarily in the above
expressions, viz. the line integral

\ i dx + w dy’ (184)
which gives the increments of the function round the closed contour, must
be equal to zero, since the function u must be single-valued. We thus arrive
at the following condition for the functions co3(s) and co4(s) in the limiting
conditions (183):

J [ f03 (s) cos (s>*) + “ 4 (®) cos (,s<V)] ds = 0. (185)


I
Apart from this the choice of these functions can be arbitrary.
We shall try to find the biharmonic function by using Hurse’s formula
(179). Differentiating with respect to x and y and using z and z we have:

-fa = <H[<p(z) + z <p' (z) + y>' (z)],


(186)
— = m [ - i<p (z) + iz q>' (z) + iy>' (z)] = 3 [<p(z) - z y ’ (z) - y>' (z)].

We thus obtain two equations which must be satisfied by the unknown


functions <p(z) and y>(z) on the contour I:

- fa ~ * = 9 (« ) + 2 <P' (z ) + V' (* ) = (® ) ~ iaU ( » ) .


(187)
8u -------- ------
-fa + * -0^- = 9 (z) + 2 9' (2) + V>' (z) = «°3 («) + »"4 («)■

The second of these equations is obviously obtained from the first as a


result of the transition to conjugate values. We thus obtain a limiting problem
for two analytic functions.
Here, as in the case of harmonic functions, we are only considering the interior
problem, i.e. the bounded part of the plane.
In the two-dimensional problem of the theory of elasticity the tensions
X x, Yy and X y are expressed in terms of a biharmonic function according to
the formulae:
62] THB WAVE-EQUATION AND ANALYTIC FUNCTIONS 203

and by using Hurse’s formula, the tension can be expressed by two analytic
functions. Without going into details of the proof we shall just give the final
result. Using the symbols from formula (179) we have:
X x + Yy = 4 K (2)].
(189)
2Xy + * (X x - Y y) = - 2* [y," (2 ) + hp’’ (2 )].
With the aid of these formulae the two-dimensional statics problems in
the theory of elasticity when tensions are given on the contour, can be solved
as a limiting problem in the theory of functions of a complex variable.
An explanation of the connection between the theory of functions of a
complex variable and the two-dimensional statics problems in the theory of
elasticity was given by Prof. G. V. Kolosoff in his work: “One application of
the theory of functions of a complex variable to the two-dimensional mathema­
tical problem in the theory of elasticity". A systematic account of the applica­
tions of the theory of functions of a complex variable to problems of the theory
of elasticity can be found in the book by Prof. N. I. Muskhelishvili Some Funda­
mental Problems in The Mathematical Theory of Elasticity.

52. The wave-equation and analytic functions. We saw in Volume II


that for spreading waves, for example, acoustic or electromagnetic
waves, the following equation is of the greatest importance

it is usually known as the wave-equation. We shall now only con­


sider the two-dimensional case, i.e. when the unknown function u
does not depend on one of the coordinates, e.g. the z coordinate. In
this case the wave-equation has the form:
02 U 9* U . 02U i 2 1
~ W ~ Ih F + Ify2" [c ~ ~ a^)' (191)

where u is a function of the variables t, x and y. By using the analytic


functions of the complex variable we can separate a certain class of
solutions of the equation (191), which have important applications in
physics; the use of analytic functions considerably simplifies all
operations with this class of solutions.
Let us construct an auxiliary equation which plays an important
part in all that follows:
I (t) t + m (t) x + n (t) y + p (r) = 0, (192)
where l(r), n(t) and p(r) are analytic functions of the complex
variable r. The equation (192) defines x as a function of the variables
t, x and y. Let us suppose that we have an analytic function /(r) which, in
204 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [52

the final analysis, is a function of the variables t, x and y. We shall


derive the formulae for the derivatives of this function. Denoting by
8' the derivative of the left-hand side of the equation (192) with
respect to the variable t and applying the usual rules for the differen­
tiation of complicated and undefined functions we obtain without
difficulty the following expressions for the derivatives of the function r:
fo _ l(*) . 9r _ m(t) . dr n (r)
dt 6' ’ dx 6' ’ dy 6' ' {
When evaluating derivatives of the second order it must be remem­
bered that
8' = I' (t) t + m' (r) x + n' (r) y + p' (r) (194)
depends, for example, on t both directly and through r:
92r 3 [ i ( r ) ] I(t) l{ r)l'(r) 2l(r)l'(r) I2(t) 8
~W dr L 6' J d' + 8'2 ~ 8'2 S'* " (195)
which can be written as follows:
9*t 1 d p (* )1
9(2 — 8’ 9t [ <5' ' J (196)
We obtain similarly:
92r I. 9 r m2(r) "I 92r 1 9 [ n2(t) 1
9a:2 <5' 9r L 8' J ’ dy2 ~ 8' dr I 8' J’

( 1 9 7 >

The given analytic function /( t) depends on t, x and y through r,


and its derivatives are obtained by using the rules for differentiating
complicated formulae. Bearing in mind earlier formulae we obtain:

aw )
which can be written as follows:
92/( t) (199)
dt* 8'
and in exactly the same way
9*/W 92Hr) 1__ 9_
dx* <5' dy2 8' dr
& l(r) _ I 9 |V ™ (t)w(t)"|
( 200 )
9x9y ~ 8' dr \J w 8' J’
63] THE FUNDAMENTAL THEOREM 205

If we suppose that u = f( x) and substitute in (191) we obtain


an equation of the form:
1 3 /_\ (T) + w2 M ~ «212(T) 1 r>
5' dr Y' \ > <5' J — U’
from which it follows that /(x) is the solution of the equation (191)
if the coefficients of the equation (192) satisfy the relationship
m2 (t) + n2 (t) = a212 (x). (201)
If we want to obtain a real solution, we must only take the real
part of /( t), which should separately satisfy the equation (191) in the
same way as the imaginary part satisfies it.
Consider now the three-dimensional space (S) with coordinates
(t, x, y). If in a domain B of this space the equation (192) gives real
values for x, then we wrongly supposed that the function f(x) was an
analytic function, since its argument takes real values only. It is
sufficient to assume that f(x) is an arbitrary function of a real
variable with continuous derivatives of the second order.
This brings us to the following theorem which states the class of
the solutions of the equation (191) which we mentioned above.

53. The fundamental theorem. I f in a domain B in the space {S)


the equation (192) and the condition (201) define r as a complex function
of the variables t, x and y, then the real and imaginary parts of any analy­
tic function f(x) give a solution of the equation (191). If, however, r is
a real function of (t, x, y) in a given domain, then any arbitrary real
function of t with continuous derivatives up to the second order gives
a solution of the equation (191).
If l(r) # 0, then dividing both sides of (192) by l(t), we can assume
that l(t ) = 1. We can also put m(r) as a new complex variable (—0).
In this case the condition (201) gives n2{x) = a2 — 02, since the equation
(192) can be rewritten, for example, in the form:

t — 6x + Ya2 - 92y + p [G) = 0 (202)


where p(Q) is any analytic function of 0. Instead of /(r) we should,
of course, write /(0).
Let us consider in greater detail the particular case when p(0) = 0.
In this case the equation (202) has the form:

t - G x + Y a 2 - G 2y = 0 or 1 _ Q± + f a 2 - 02 f = 0, (203)
206 CONFORMAL TRANSFORMATION AND TIIE TWO-DIMENSIONAL FIELD [63

which defines 0 as a function of two arguments

f= V= (204)
In this case the constructed solutions f(9) of the equation (191) are
also functions of the arguments (204), i.e. they are homogeneous func­
tions of zero order of t, x and y. Such functions [I, 154] are defined
from the relationship
u (kt, kx, ky) = u (t, x, y ) ,
which should be an identity. The converse can also be shown, viz.
that any such homogeneous solution of the equation (191) can be
obtained in the way described above. In future we shall simply call
such solutions homogeneous solutions.
Let us consider the equation (203) in greater detail. The radical
]/d2 —02, will be a single-valued function in the 0-plane with a cut
(—a, -fa) along the real axis [19]. We can fix the value of the above
radical by the condition that it should be positive above the imaginary
axis, i.e. when 0 = ib, where 6 > 0. This condition is equivalent to
the fact that the above radical must be negative imaginary when
0 > a or positive imaginary when 0 < —a on the real axis. This
can easily be proved by considering the continuous change of the argu­
ment of the above radical. The equation (203) can be rewritten in the
form:
1 —0£ + f a 2 - 0 2J? = O. (205)
Eliminating the radical and solving the quadratic equation so ob­
tained we derive the following expression for 0:
a _ £ —i y Y 1 —a 2(£2+ Tj*) _ xi — iy ft2—a2 (s2+ y 2)
£2+ >?2 “ z2+ y2 ‘ 1 '
We suppose that one of the following inequalities applies;

f* + Vz < (207)
or
x2 + y2 < -A-1
U> 2. (208)

In formula (206) the sign of the radical must be taken as positive.


This can easily be proved by considering the equation (205) in which
the radical must have a definite value. In fact, if we suppose in the
equation (205) that £ = 0, we obtain purely imaginary values for 0
53] THE FUNDAMENTAL THEOREM 207

and, from (205), the sign of the radical |Aa2 — 02 must be opposite
to the sign of r), i.e. if, for example, rj < 0, then according to the above
condition, 0 should lie above the imaginary axis, which coincides with
the choice of the sign in formula (206), where we assumed that the
radical is positive.
When the values of f and r\ are fixed we have, from (204), a straight line
in the space (S), which passes through the origin. We shall only con­
sider that part of the straight line where t > 0 and we shall call this
line a ray. It appears from the conditions (207) and (208) that these
rays form a conical beam with its apex at the origin and an angle
equal to arc tan lja at the apex, and the £-axis as the axis of the beam.
.The equations (205) or (206) give complex values corresponding to the
rays of this beam in the 0-plane with the cut (—a, +a). By using
formula (206) this relationship can be followed more accurately. Let us
emphasize some essential facts which follow directly from formula
(206). Notice, first of all, that the rays which form the surface of the
conical beam, i.e. the rays which satisfy the equations

£- + V2 = ^ r or *2 + t/2 = -|-< 2,

correspond to points of the cut in the 0-plane. The axes of our conical
beam, which are defined by the values x = y = 0 o r ^ = r] = 0, cor­
respond to the point at infinity of the 0-plane. Notice finally, that rays
situated in the y = 0 plane for which rj = 0, correspond to real values
of 0, the modulus of which is greater than a, i.e. they correspond to
points on the real axis of the 0-plane which lie outside the cut
{—a, +a). If we divide our beam of rays into two parts by the plane,
y = 0, then one part corresponds to the upper half-plane 0, and the
other to the lower half-plane, viz. the half where y > 0 corresponds
to the lower half-plane and the half where y < 0 to the upper half­
plane.
If we take the solution of the equation (191) constructed by the
above method, i.e. the solution which is the real part of a certain
analytic function /(0), then this solution will have a constant value
on each one of the above rays.
Let us now investigate the values of 0 for points of the space (S)
which lie outside the above conical beam, i.e. for all points which
satisfy the inequalities

£'2 + V2 > -^r or xZ + V~ > t2■


208 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [53

The equation (205) gives us two real zeros, which lie on the line
(—a, -fa):
Q__ £ ± V Ya* (£2 + V1) — 1 _ Xt ± yt fa2(s2+ y 2) —
f2f 7j2 x1+ y2 ''

This line (—a, -fa) is the cut of the plane, and on opposite edges
of this cut the radical ]/a2 —62has opposite signs, so that in the equation
(205) we should take into consideration the double sign of the radical;
we must also take both signs of the radical in formula (209). Let
M0(t0, x0, y0) be a point outside our conical beam and 9± and 92 the
corresponding values of 9, obtained from formula (209). If we substitute
these values 9 = 9Xand d2 in the left-hand side of the equation (205)
we obtain two real equations of the first order with respect to t, x and y
and, consequently, we have two planes through the point M 0. This can
be expressed in a different way, viz.any value 0 = 0 „on the cut (—a, -fa)
corresponds to a plane P in the space (S ). Let Abe the generating line
which corresponds to the point 9 = 90 on the cut. This generating line
A must lie in the plane P. It is not difficult to show that the plane
P will be tangential to the surface of our conical beam along the gener­
ating line A. In fact, if the plane P is not tangential to the surface
of the cone along A then it would cut this surface, and part of the
plane would then lie within the conical beam. In that case points
within the conical beam would correspond to real values of 9 = 90
in the interval (—a, + a ) which, as we saw above, is not possible.
Hence [from (205)] any real 9 on the cut (—a, -fa) corresponds to a
plane tangential to the surface of the conical beam along the generating
line which corresponds to the given value of 9.
Instead of talking about a conical beam and tangential planes to its
surface we can use a two-dimensional diagram, i.e. we can cut our
conical beam with a plane perpendicular to the t axis. In this case
the conical beam is represented by a circle to which the tangential
planes are tangents. In particular, we can use the variables f and rj
in the transition to the two-dimensional diagram. Instead of the coni­
cal beam we have in the (|, rj) plane the circle K:

P + v2 < (210)

where every point of this circle corresponds to a definite ray of our


beam and vice versa. The tangent to the above circle corresponds to
the tangential plane to the surface of the beam. The half-plane 77 > 0
53] THE FUNDAMENTAL THEOREM 209

corresponds to that part of the space where y > 0. The axis r) = 0


corresponds to the plane y = 0.
Let f(9) be an analytic single-valued function in the 0-plane with the
cut (—a, -fa). Take the corresponding solution of the equation (194):
u= < 7 2 \f(& )\ (fHbeing the real p art). (211)
This solution will be defined within our conical beam or, in the case
of a plane (|, r\), within the circle (210). We shall give one method for
continuing this solution outside the conical beam which has very
important applications. In order to do this we draw a family of tangen­
tial half-planes to the surface
of our conical beam, all in the
same direction, i.e. their cor­
responding tangents to the circle

£2 + V2 = p - (212)

should have the appearance


shown in Fig. 51. These tangen­
tial half-planes do not cross
one another and fill part of the
space (S ) outside the conical beam. /(0) remains constant on every
one of these half-planes and we can thus define the solution of u out­
side the conical beam uniquely by using the same formula (211) which
gave the solution within the conical beam. In this case the values of the
solution remain constant not on the rays but on half-planes outside the
conical beam. Notice that the directions of the tangents to the circle
(212) can evidently be selected and we therefore obtain two different
methods for continuing the solution, when the above method is used.
The corresponding surfaces of the conical beam 0 belong to the cut
—a < 0 < -fa. We can, in this case, separate the values of u as given
by formula (211) into two real terms u = %(0) + u2(9) and continue one
of the terms along the half-tangents I (Fig. 51) and the other along
half-tangents II. This also gives a solution of the equation outside
the circle. We therefore have an infinite number of different ways
of continuation and in all cases the continuity of the solution u is
preserved in transit through the circle. In actual problems the method
of continuation is determined from the movement of the frontal wave.
All that was said above referred to the solution in a space (S ).
Let us now suppose that we are only interested in the half of the space
210 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [63

where y > 0, or in the plane (|, r]), where r) > 0. Assume that formula
(211) gives the solution in a semicircle and that it is equal to zero on
an arc A B of this semicircle, as shown in Fig. 52. This case has many
applications in problems of propagation of vibrations and we arrive
at a single-valued continuation of the solution (211) by using half­
tangents to the circle shown in Fig. 52, i.e. by using the corresponding
half-planes which are tangential to the surface of the conical beam.
In this case the solution will be equal to zero outside the contour
Aj^ABByA^
Analogous considerations can also be applied to the general case
of the equation (202) but, instead of the conical beam, we shall, of

course, have a much more complicated geometrical picture (a family


of straight lines with two parameters), which depends on the choice
of p(0).
In the equations (202) and (203) we can substitute another variable z
instead of 0, which is connected with 0 by a functional dependence.
We shall give here one particularly convenient choice of this complex
variable. Let the connection between z and 0 be in accordance with
the formula
0 = t (3 + t )- <213>
In this case, as we know from [33], instead of the 0-plane with the
cut {—a, + a) we have a unit circle | z | < 1 for the variable z. By
using formula (213) it is not difficult to see that for our choice of the
value of the radical we have the formula

(214)
63] THE FUNDAMENTAL THEOREM 211

Let us consider in greater detail equation (203) in this case. It will


have the following form:

<- y ( z + T ) a: + i f [z - \ ) y = ° (215)
or
1 - y (2 + y ) £ + i T ( z - l h = 0 ’
which can also be rewritten as follows:

= (216)

We introduce polar coordinates for the circle (210) according to the


formulae
f = g cos <p\ r] = g sin <p (o < g < -i-j.
In this case the equation (216) can be rewritten as follows:
age~l*z2 — 2z + age'9 = 0,
and we have a solution for z in the form z = re*9, where r is deter­
mined from the quadratic equation
agr2 — 2r + ag = 0 (0 < r < 1),
i.e. in this case every point on the circle (210) (i.e. every ray) corresponds
to the value of the complex variable z = re1* with the same amplitude,
and points on the circle (212) correspond to points on the unit circle
with the same amplitude. In other words, every radius of the circle
(210) corresponds to the radius of the unit circle | z \ < 1 with the
same polar angle.
The fundamental ideas about applications of the theory of functions
of a complex variable to the solution of the wave-equation (191)
with which we have dealt in this section, have wide applications in
problems of propagation of vibrations (acoustic, electromagnetic),
as well as in more complicated problems of propagation of elastic
vibrations. The above method only gives one class of solutions of
the equation (191) but it so happens that this class includes the
solutions which are of great importance in physics; by using the
latter it is possible to bring problems connected with the reflection
and diffraction of waves, to a final form suitable for evaluation.
The equation (191) is an equation of a two-dimensional wave
(cylindrical wave), but by using the principle of superposition it is
possible to construct new solutions from solutions of the above type,
212 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [54

and thus to investigate the wave equation in three dimensions. The


theoretical basis of the above method can be found in works by
S. L. Sobolev and in papers by this author printed in T h e W o r k s of
T h e S e i s m o l o g i c a l I n s t i t u t e a t th e A c a d e m y o f S c ie n c e . Its applications
to definite problems can be found in works by E. A. Naryshkina
and S. L. Sobolev. Without going into details which would take us
outside the scope of this book, we shall describe briefly the application
of this method to two problems: the diffraction of a two-dimensional
wave and the reflection of elastic vibrations from a flat object.

54. The diffraction of a two-dimensional ware. Consider the (x, j/)-plane,


out in the direction of the straight line y = x, where x > 0. Assume further
th at in the remaining part of the plane where l < 0 we have a two-dimensional
wave, which is propagated parallel to the X axis with a velocity l/o, so that
at the instant when t= 0, it reaches the end of the cut (the origin). Assume that
this two-dimensional wave has the following olementary form:

u = 1 when x < — t\ u = 0 when x > — t , (217)


a a '
Behind the frontal of propagation u is constant and equal to 1 and in front of
the frontal, which has not yet been disturbed, u = 0 .
In the case under consideration the function u satisfies the equation (191)
when t < 0 and it is the homogeneous solution of this equation; it depends only
on f and 77 and is defined by the conditions

u= 1 when £ < —, u = 0 when £ > — . (218)


Q CL

The frontal of this wave moves with the velocity l/o, which agrees with the
wave equation (191).
We shall now investigate the problem of diffraction of the wave (217) at the
above cut and we shall suppose that after diffraction, i.e. when t > 0 , the
wave will still be represented by a homogeneous solution of the equation (191),
i.e. by the real part of an analytic function /(z) of the complex variable z,
as defined by the equation (216). This assumption is quite natural, for the line
which causes diffraction is a cut which ends at the origin. We take it that on
both sides of the cut the following condition is satisfied:
u = 0 (on the cut). (219)
At the instant when t = 0 our two-dimensional wave reaches the out
after which diffraction takes place. Take any positive time t > 0. Bearing in
mind that according to the wave-equation (191) the velocity of propagation
of the disturbance is equal to l/o, we have at the given instant the following
picture of the disturbances. To begin with, the straight frontal A BCD is tom
in two by the obstacle through which the frontal has passed. The line of this
frontal is perpendicular to the X-axis and OB = (l/o) t. We next have a straight
frontal formed by the wave reflected from OO, according to the usual law,
54 J THE DIFFRACTION OF A TWO-DIMENSIONAL WAVE 213

(Fig. 53). This -will be the straight line EG, parallel to the X axis. Also tho
presence of O creates an additional disturbance in the circle, centre the origin
and radius (1ja)t. I t is the main object of this problem to determine the function
Min this circle. Let us list those values of u, which apply outside the circle.
In front of the line A B F below the cut OG we evidently have u = 0. Also u = 0
above this cut and in front of the line CD. Now in the part of the plane bounded
by the contour ECFE, the falling wave is
joined by the reflected wave and from the
limiting condition (219) we have again
M= 0. In the part of the plane outside
the above circle and behind the frontal
of the wave, u = 1 everywhere except
in the above domain ECFE. The circle,
centre the origin and radius (1/a) t happens
to be the circle (210). In this case, how­
ever, it is cut along the radius
arc tan ( 7 7 /<?) = ji/4.
According to the equation (216) we
have on transit to the z-plane, a unit
circle z < 1 cut along the radius
arg z = jr/4. We know from above that the
radii of the circle (210) correspond to the
radii of the unit circle | z j < 1 with the
same central angle.
Bearing in mind the above values of
u and the limiting conditions and making
the transition to the z-plane we obtain the following problem: find a function
f(z), regular in the cut circle | z | < 1 and — 7 ji /4 < arg z < ji/4 , so th at its
real part should vanish on both edges of the cut, i.e. on the radii:
n . In
arg z = — and arg z = ---- —,
and also on the arcs
7 in , „ n
— — < arg z < ---- y > and 0 < arg z< 4 ’

and be equal to unity on the remaining part of the circle | z | = 1. I t is not dif­
ficult to write the solution of this problem in a definite form.
Rotating the z-plane about the origin by an angle 7ji/4:

wx = e'T z,
we obtain the circle | wl | < 1 and 0 < arg wt < 2ji, which is cut along the
radius arg wl = 0. By extracting the square root this cut is transformed into
the interval (—1, +1) of the real axis and the circle is transformed into the
upper part of the unit circle. Therefore the transformation:
__ i In 1
;= = e 8
214 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [54
transforms our cut circle in the 2 -plane into an upper semicircle in the 10-plane.
The boundary conditions for the unknown function/(to) will then be that the
real part of/(to), should be zero in the interval ( —1 , + 1 ) of the real axis and

[/ (e,p)] = 0 when 0 <?> < and - ^ - : n < y < j r ,

4L [/ (e,<p)] = 1 when < <p< n.

f(w) thus transforms the interval (—1 , + 1 ) of the real axis into an interval of
the imaginary axis and, according to Schwarz’s principle of symmetry, /(to)
can be analytically continued into the lower part of the unit circle when at
points to, symmetrical with respect to the real axis, it acquires values sym­
metrical with respect to the imaginary axis [24].
We thus obtain the following equation:

« [ / (e-'*)] = - « [ / («'*)]•
Bearing this in mind we arrive at the following boundary conditions for
f(z) on the unit circle:

<2 [/ (e'9)] = 0 — y < q>< y and-^-Ji < q>< -jj-Ji

( 220)
*[/(•>'*)] = 1

« [ / ( o'')] = - l - — <<p<

To construct the solution of this limiting problem consider the function

1. a —w 1 + arg
— log -5 ------- 221)
p —w= —
% log P — w
(
% j3 — w

where a and P are points on the unit circle, situated a t opposite ends of the same
diameter A B (Fig. 54). Let M be the variable point w. The real part of (221)

arg ^ ^ = arg (a — w) — arg (p — w)

represents the angle between the vectors M A and M B, measured from MB.
The function (221) is single-valued and regular in the circle | w \ < 1 . When
w = 0 it is equal to ji, and has a period 2jt. We suppose th at in the circle | w | < 1,
it is equal to n and we thus fix a definite branch of the function (221). For this
choice of branch we have:
a —w 1 1 — a 1 to __
log
P —w n i 1 —P~rtv
= 71 + -4- log (1 — a 1w ) --- log (1 — P 1to),
% %
64] THE DIFFRACTION' OF A TWO-DIMENSIONAL WAVE 215

where we take the principal value of both logarithms, as defined by the usual
power series. If w lies on the arc A P B , then the above angle BM A is equal to
ji/2 and when it lies on the arc AQB it is equal to 3ji/2, i.e. for the given choice
of the single-valued branch of the function (221) in the circle | w | < 1 its real
part is equal to nj2 on the arc A P B and to 3ji/2
on the arc AQB.
Let us apply this result to the function:
.Tn
. 1i e 8 —w 1 . e- ‘ T
8 —w
v (w) = T l og— ------+ — log
*o
e 8 —w e 8 —w

Denoting by M lt M,, M 3 and M i the points


.7n .In
e"
e ‘ «, e“
e 8,

we can say th at the real part of y>(w) is equal


to 2ji on the arcs M lM 2 and M 3M t, to n on
the arc M XM 3 and to 3ji on the arc M 2M i . Bearing this in mind we directly
obtain the solution of the limiting problem (220) in the following form:

/(<")= — VM 2.
71

Returning to the former variable z we obtain the solution of the diffraction


problem in the circle
x* + y 2<
in the form
In In l\
8_ e 8 z2J
&
11

-717% log n In 1\
( i n t ln 1'I ( i
(e 8 —e 8 z2J(e 8 _ e M

The above considerations have no strict theoretical basis and the concept
of an elementary two-dimensional wave u which is equal to unity behind the
frontal and to zero in front of the frontal seems at first to be rather artificial.
It can, however, be shown that any two-dimensional wave can be represented
by an integral which contains the elementary two-dimensional wave. The result
so obtained can therefore be made to include the diffraction of a two-dimensional
wave of the most general kind by reducing the problem to the case we considered
above.
Let us consider the general appearance of a two-dimensional wave which
moves parallel to the X axis. This wave is given by the function / (tja — x)
and we assume th a t/(r) = 0 when t < 0. The function/ (</a — x) certainly satis­
fies the equation (191). Above we have considered the elementary case, viz.
216 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [65
f(r) = 1 when t > 0 and / ( t) = 0 when r < 0. Denote /(t) by u(r) in that
particular case as we did in formula (217):
0 when r < 0
u (t ) = (218,)
1 when r > 0

Let us now suppose th at f(a) is a continuous function with a continuous deriv­


ative and th at this function is equal to zero when r < 0. We can write:

/ ( t ) = j « ( t - a ) r w <u-
0
In fact, taking into consideration the definition of u(r) and the condition
/ ( 0 ) = 0 we obtain:

J u (r —A) y (A) dA = J / ' (A) dA = f (t) - / (0 ) = / (r ) .


0 0
We can therefore write:

f ( 1 1 - *) = \ u ( t ‘ - x - A) f' W dA = \ u ~ * ) r (A) dA•


o o
I t can be seen from this formula th at the general type of falling two-
dimensional wave is a sum (an integral, strictly speaking) of elementary fall­
ing waves
- *) /' W dA.
If we denote by U(x, y, t) the above result obtained for the diffraction of an
elementary wave, then in the case of the falling wave / (tja — x) we have a
solution of the form:
V = $ U { x ,y ,t- a A ) / ' (A) dA.
0
We are only considering the result of diffraction with respect to the origin
and we denote this by U(x, y, t). When t > 0, this takes place in a circle, centre
the origin and radius (l/o) t, i.e. we suppose th at U(x, y, t) = 0 when t < 0 for
any (x, y) and also th at U(x, y, t) = 0 when x* + y2> (l/o2) t2 and t > 0 . Hence
in the expression for V the integral containing A will, in fact, be propagated
in the finite interval in which A varies.
The above method can be used for solving the problem of diffraction of a
two-dimensional wave falling in any arbitrary direction at any given angle.

55. The reflection of elastic waves from rectilinear objects. In two-dimensional


problems in the theory of elasticity the component displacements u and v
can be expressed by the formulae
9© , dtp dg> 9y ( 222 )
u = “a — h - 5 - 1
ox oy « = -5 —
oy dx
55] THE REFLECTION OP ELASTIC WAVES PROM RECTILINEAR OBJECTS 217

where the function <p is usually known as the potential of longitudinal waves
and the function y>as the potential of transverse waves. These potentials should
satisfy wave-equations of the form
92 <p 92<p
(223)
012 dx2 ^ 9y- ’
9> 92v . 92V (224)
91* ~ dx1 1 dy2 ’
where
(225)

where g is the density of the medium and Aand n are Lamp’s elasticity constants.
The numbers l/o and 1/6, as we know from the theory of elasticity, give the
velocity of propagation of the longitudinal and transverse waves and formula
(2 2 2 ) gives the subdivision of the general agitation into longitudinal and trans­
verse agitations.
f; We shall also state formulae which express the tension in the elastic body in
terms of the potential. We shall only consider the vector of tension which acts
on a surface perpendicular to the Y axis. The components of this vector can
be expressed by the following formulae:

After these preliminaries let us formulate the problem. Let us suppose th at at


the instant when ( = 0 an agitation, purely longitudinal in character, is
propagated from the point x = 0 , y = y 0; this propagation has a potential
<p, which satisfies the equation (223) and which gives the homogeneous solu­
tion of the equation with respect to the arguments t, x and (y — y 0), i.e. it
is defined as the real part of the analytic function
<p = «[<P(0)], (227)
in which the complex variable 0 is determined from the equation:
t — 8x + fa? — 02 (y — y0) = 0 . (228)
This latter equation differs from the equation (213) only in so far as y
is replaced by (y — y„). This shows that the potential (227) corresponds to
a force which at the time t = 0 is concentrated at the point x = 0 , y = y a.
We shall not explain this here from the point of view of the mechanical charac­
teristic of the source.
We assume that the given function <2>(0) in formula (227) is regular in the
0 -plane with the cut (—o, + a), except at the point at infinity, and that its
real part vanishes on the cut. This later circumstance shows th at the given
potential <p vanishes on the surface of the conical beam with the vertex of
angle arc tan l/o at i = 0; x = 0, y — y 0. This surface corresponds to the
frontal of the spreading agitation. We assume, of course, that the potential
218 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [55
is equal to zero everywhere outside the conical beam. Let us suppose that
we are given not the whole plane of the spreading agitation, but only the
half-plane y > 0, with the centre of agitation x = 0; y = y„ > 0 . The potential
<p fully defines the movement only at the instance when t > ay0. When
t = ay 0 the frontal of agitation reaches the line y = 0 which is the edge of our
medium, and reflected waves appear; the reflection laws should be obtained
from the limiting conditions at this edge. We assume th at the medium is free
of tension and in future, when writing the corresponding limiting conditions,
we shall equate to zero the expression (226) when y = 0 .
As a result of reflection two other potentials must be added to the given poten­
tial <p: one is the reflected longitudinal potential <p1and the other is the reflected
transverse potential yl. We assume th at both potentials are expressed as real
parts of analytic functions of a complex variable
Vi = * [ 0 , ( 9 , ) ] ; V i-a p P ito )]. (229)
We have to find equations for both complex variables 0 2and 02, as well as the
form of the analytic functions 0 [(®i) and !f,I(02); the latter are obtained from
the given falling potential rp and from the limiting conditions. From [53]
and also owing to the fact th at the wave-equation for the transverse potential
y> contains the constant b instead of a, the above complex variables will be
found from equations of the type:
t - x ± / a 2- 0? y + p t (0J = 0 ,

t — 02x ±]/b* — 6 \ y + p 2 (0 2) = 0,

and we m ust, first o f all, select the form o f the functions p l ( 0 1) and p 2(0 2) and
the signs of the radicals, bearing in mind the fact that th e values of radicals
in cut planes are always determined in the way explained in [53],
Consider the conical beam o f rays which corresponds to the equation (228),
with vertex at the point t = x = 0, y = y 0. In this case the difference
(y — Vo) replaces the letter y, if we make comparisons w ith [63]. The plane
y = y a divides our beam into tw o parts and the part o f the beam where y > y 0
will never m eet the edge y = 0 in the space (S) with the coordinates (t, x, y).
The second part o f the beam where y < y0will m eet this plane, and the points
of intersection of the straight lines o f the beam and the plane will fill a whole
domain o f this plane defined b y the inequality (Fig. 55)

& + (231)

This follows directly from the fact that the equation o f the beam will, in this
case, have the form

x 2 + (y - Vo)2

The domain (231) therefore represents the interior o f a hyperbola in the y = 0


plane in the space (S ). It follows from [63], that that part of the conical beam
which intersects the plane y = 0,
where y — y a < 0,
corresponds to the upper
half-plane of the complex variable 0. A t the same time y decreases while t
65] THE REFLECTION OF ELASTIC WAVES FROM RECTILINEAR OBJECTS 219
simultaneously increases along every ray. We select in the equations (230) the signs
of the radicals so that they should be opposite to those in the equation (228);
the functions p x{8x) and p 2(02) are determined in such a way th at the equations
(232) and (228) should coincide when y = 0. We thus obtain for the new complex
variables the following equations:

t —0! x - Ka2—0J (y + y0) = 0, (232)

t - e t x -Y b * - d l y - Y a * - 6 \ y a = 0. (233)

Consider a point M ^ ^ x ^ in the domain (231) in the plane y = 0, in the space


(S ). This point can be reached by a ray belonging to one half of the beam, which
corresponds to a definite value 9 = 9'.
If we substitute the coordinates of the
point t = x = xx and y = 0 in the
equations (232) and (233), we obtain the
identical value for the complex variables
0 , and 02. If we now substitute these
values = 0 ' and 02 = 8' into the
complete equations (232) and (233), then
the equations so obtained will determine
two rays which we shall in future call the
reflected longitudinal and the reflected
transverse ray [all this takes place
in the space (5)]. Notice one important cir­
cumstance, viz. as a result of the definite
choice of the signs of radicals in the equa­
tions (232) and (233) it is evident th at
t and y increase simultaneously along
the reflected rays, i.e. the reflected rays
travel into the depth of our half-plane,
as time goes on or, in other words, the
reflected waves alter nothing in the picture of disturbance which existed before
reflection occurred. Let us test this circumstance for the equation (232). By
comparing it with the equation (228) it can easily be shown th at it corresponds
to a conical beam with vertex at the point t = x = 0 , y = —y 0, symmetrical with
the centre of agitation -with respect to the y = 0 plane. Bearing in mind the
fact th at the sign of the radical in the equation (232) is different from the sign
of the radical in the equation (228) we can say th at the values of 0 in the upper
half-plane, which we obtained as a result of reflection, correspond to rays,
where t > 0 and y y 0 > 0 ; also when t increases y increases along the ray.
An analogous circumstance also applies to rays defined by the equation (233),
but in this case the beam of rays will no longer be conical. Thus from every
point M x of the region (231) two different rays will radiate. We are trying to find
the potentials of the reflected waves in accordance with the formulae (229),
so th at they remain constant along the reflected rays. The form of the
functions in the formulae (229) remains to be found. As we have already said
220 OONFOBMAL TRANSFORMATION AND TH E TWO-DIMENSIONAL FIELD [55
above, we are considering in this case limiting conditions of the form:
2 32 (P + <Pi) _|_ 32Vi 02Vl
0.
dx dy &y2 6r2 y-o =

(M[ 32 (<P+ <Pi) , 0! (<P + Pi) ‘


8x2 3y2
+ 2 92(<8y2
P+ Pi)

To evaluate the derivatives of the functions <p, <plt and y>lt as determined by the
92yt
0x 0y 3>-o = 0.

formulae (227) and (229) we can use the formulae (200) by substituting i(r),
m(z) and n(r) by the corresponding coefficients from the equations (228),
(232) and (233). Notice also th at in the reflected transverse potential rp1we must
replace a by b. When y = 0 our complex variables 0, 0„ 02 coincide and we can
denote them by the same letter 0. We thus arrive at conditions of the following
type:
1 8 — 29 if a2 — 02 [&' (0) — <Pj (0)1 + (5* — 202) (0) ] _
01 S' 80 S' J ’
(234)
1 8 (b* - 20*) \0' (0) + 0X(0)] - 20 Yb2 - 02 ^ (0) ] _
01 d' 80 S' J

where
0' = - X + Vo-

The conditions (234) should be satisfied in the whole domain (231), i.e. in the
whole upper half-plane 0 .
We obviously obtain the solution of the equations (234) if we determine the
unknown functions <Pt(0) and 0'1(0) from the equations:

- 20 y a 2 - d2 [ 0 ' (0 ) - 0 i (0 )] + (b 2 - 2 0 2) V ' i (0 ) = 0 ,
(b 2 - 2 0*) [ 0 ' (0 ) + 0't (0 ) ] - 20 yb2 - 0* I P i (0 ) = 0 .

I t can be shown th at these equations are not only sufficient but are also
necessary if the conditions (234) are to be satisfied. By solving them we obtain
expressions for the derivatives of the unknown functions

0 ' (6)= _ -(2 glr- bi)2 +492 ^ rJL 0 (0), '


' F(6) v' (235)
* (» )— 4e(29,- ^ ^ L ^(0).
where
F (0) = (20* - 6 *)* + 40* ya2 - 0* Yb2' (236)
To obtain the solution of the problem we are only interested in the derivatives
of the potentials. From formula (222) we obtain the following formulae for
the displacements:

« - » W <»>H +»;<«.) + r , («,) -S fl • (237)


55] THE REFLECTION OF ELASTIC WAVES FROM RECTILINEAR OBJECT'S 221

If neither the falling, nor the reflected ray passes through the point M(t, x, y)
then we must cross out the corresponding term in the expressions (237). Notice
one important circumstance, viz. from the given condition the real part of
0'1(&) is equal to zero when —a < 0 < + o. From the formulae (235) and
(225) it is clear th at the relationship 6 > a follows directly; it also applies to
# '( 0 ) and ¥^( 0 ), so th at the reflected potentials <pr and xpl are constant on the
surfaces of the reflected beams of rays and we can assume th at they are equal
to zero on these surfaces and outside the beams.
If we were to consider the source of transverse vibration instead of the source
of longitudinal vibration then the picture would be somewhat different. In
this case we would be given the potential of transverse vibrations in the form
of the real part of an analytic function
* = «[1P(0)]P (238)
regular in the 0 -plane with the cut (—6 , + 6 ) and the complex variable 0 is
determined from the equation
t - ex + Kb2 - 0! (V “ Vo) = 0, (239)
where the real part y>(8) is equal to zero when — 6 < 0 < + 6 . We are looking
for reflected longitudinal and transverse potentials of the form:
^ = ^ [ ^ ( 9 ,) ] ; Vl = n [V, (0)], (240)
where 0 ! and 02 are determined from the equations
t - e l X - Ya* - 91 y - Y v - 0 } y0 = 0 , (241)
t - 8 t x - V b ^ 9 l ( y + y0) = 0 . (242)
Similarly for functions in the expressions (240) we obtain the following
expression instead of formula (235):

(243)
(202 - 52)2 + 402 Ya2 - 02 / 6 2 — 02
y ; («) - F(9)
V ' ( 0) .

In this case the cut in the 0 -plane, points of which correspond to rays on
the surface of the conical beam, will be —6 < 0 < + 6 . The coefficients of
y>'(8) in both expressions (243) contain the radical Ya*—&2 and therefore
these coefficients, which remain real when —a < 0 < + o , cease to be constant
when —b < 9 < —a and a < 0 < 6 . At the same time the product of the
imaginary part of the coefficient and the imaginary part of !P'(0 ) gives the real
part of <Pj(0) and W[{9) which is other than zero when
— b < 8 < — a and a < 9 < b. (244)
If we substitute these values of 0 in the left-hand side of the equation (241)
then, after the separation of the real and imaginary parts, we have:
t — 8x - Yb2 - 02 y0 = 0 ; y = 0.
222 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [65

i.e. for the reflected longitudinal potential these critical rays, on which the
potential is other than zero, do not penetrate into the medium but travel in
the y = 0 plane (Fig. 56). For the reflected transverse potential the reflected
beam of rays, given by equation (242), will simply be a conical beam with the
vertex at t = x = 0 , y = —y 0; along the generating lines of the surface of
this beam, which correspond to values of 0 satisfying the conditions (244),
the values of the reflected potential will be other than zero. In this case we
shall have to continue the reflected transverse potential outside the above

conical beam by the method described in [53]. This circumstance has a simple
mechanical meaning, viz. the transverse waves radiating from the source of
vibration originate longitudinal refleoted waves when falling on the edge
y = 0 ; these are propagated along the edge faster than the transverse waves
and they, in their turn, produce a transverse wave, which travels in front of
the reflected wave and follows the usual laws for transverse waves.
We have only given here brief indications and not a detailed mechanical
investigation of the formulae (235) and (243). Note th at the denominator of
F(8), as given by the equation (236), has real zeros 0 = ±c, which satisfy
the inequality c > 6 ; the existence of these zeros produce the phenomenon
known as the phenomenon of surface waves.
CHAPTER III

THE APPLICATION
OF THE THEORY OF RESIDUES,
INTEGRAL
AND FRACTIONAL FUNCTIONS

56. Fresnel’s integral. In [21] we proved the fundamental theorem


of residues which is the basis of the application of the theory of
analytic functions to various calculations and analytic representations
of functions. We will now deal with the problems of evaluating definite
integrals, integrating linear differential equations, expanding functions
into infinite series and representing them by contour integrals.
Let us start with the evaluation of the definite integral [H, 83]
a>
j" sin (z2) dx (1 )
0
usually known as Fresnel’s integral and met with in problems of light dif­
fraction. Consider the integral
Je"*dz, (2 )
l
where I is a closed contour, which consists of the section OA of the real axis, the
arc A B of a circle centre at O, and radius R = OA and the section BO, and we
take the angle AOB as equal to w/4. Inside this contour the integrand e '!’ has no
singularities and therefore integral (2) is equal to zero. Divide this integral
into three parts which correspond to the above three parts of the contour.
The variable z will be real along OA and we suppose here th at z = x, where
0 < x < R. Along BO we have z = xe,n^ \ z2 = ix2and dz = e " 1/ 4 dx. Finally,
along A B
z = Re1* (o<p<^),

hence z2 = R2 e'2* and dz = iRe'* dp. We thus obtain the following equation
n
(3)
223
224 the a p p l ic a t io n op the theory o f r e s id u e s , f r a c t io n a l p u n c h o n s [6 6

We will show th at the third integral in the above equation tends to zero as
R increases indefinitely. Bearing in mind th at the modulus of eT, when r is
purely imaginary, is unity and substituting the integrand by its modulus, we
obtain the inequality:
n 71

T 4
J iRe~R‘(cos 2* + ''6lnW dtp < R J 0 —R*cos 2<p
0 o
We will prove th at the expression on the right-hand side tends to zero as R -►oo.
Substituting a new variable y>= 2tp for tp and rejecting the constant factor,
which is of no importance, we obtain:
31
Y
R j e - R ' C0*'f d v > .
o
We now divide the interval of integration into two parts (0, a) and (a, ji/2),
where a is a certain number between 0 and n/ 2 :
n n
Y a Y
R £ 0 -R*cosP dy» = f R e -R‘cosvdtp+ J iJe- fi,C0S*’dip. (4)
0 6 a

In the first of the above integrals we substitute the negative index by its
greatest value, i.e. by the smallest absolute value, viz. by (—I ?1 cos a). We
multiply the integrand of the second integral by the fraction sin i/i/sin a, which
is always greater than unity in the interval a < yi< ji/2 . Having thus increased
the sum (4) we obtain:
31

J i?e-R*cosadV + | siny e-R- co. y dy)>


0 a sin a
and it is sufficient to show th at this latter sum tends to zero. Both integrals
can be fully evaluated and their sum will be
31
v=T
1 - e —R* COS a
a fle -R,cosa + -5 4---- re- * ,cH = a R e-R' cosa +
R sin a L Jv=a R sin a
from which it follows directly that this sum tends to zero as R increases
indefinitely. We have thus shown that the third term on the left-hand side
of (3) tends to zero as R -*■ o o . The first term on the left-hand side has the
l i m i t.

J e J' dx,
o
which, as we know from [11, 78], is equal to (1/2) . We can therefore say that
the second term has a definite limit; this gives us
o
57] INTEGRATION OF EXPRESSIONS CONTAINING TRIGONOMETRIC FUNCTIONS 225

or, separating the real and imaginary parts under the integral:

( -y- + i J [cos (x2) - i sin (x2)] d i = | ^ .


'0
Equating the real and imaginary parts we obtain Fresnel’s integral:
es co

J cos (x2) dx = J sin (x2) dx = y j / y . (5)


0 '

57. Integration of expressions containing trigonometric functions.


Consider now the integral:
2n
j R (cos x, sin x) dx (6)
o
where R (cos x, sin x) is a rational function of cos x and sin x. Let
us substitute the complex variable z — elx for the real variable x. As x
varies in the interval (0,2;r) the complex variable z varies on the
circumference of the unit circle |z| = 1. Also, from Euler’s formula,
we know that
z + z- 1 . z — z~ x
cos x = —- — ; sin x = — ^ — •

and therefore dx = (1jiz) dx. Substituting all this in (6) we obtain the
integral of a rational fraction on the unit circle | z | = 1, which we
denote by 0.
This integral is equal to the product of 2jm and the sum of residues
of the integrand at poles inside the unit circle.

Example 1. Consider the integral


2n
dx
(0 < e < 1 ).
I t + e cos x

Performing the transformations shown above we can obtain it in the form:


dz
J
C iz [(l1 +^ )

2_ C dz
i J «z2 + 2 z -(- e '
c
226 TH E APPLICATION OP T H E THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [5 7

The poles of the integrand will be the same as the zeros of the quadratic
equation
ez* + 2 z + e = 0 , (7 )
in which one of the zeros has a modulus smaller than unity. This zero is deter­
mined by the formula
- 1+

where the positive sign of the radical must be taken. The residue of the inte-
grand can be determined by the rule stated in [2 1 ], viz. this residue is equal to the
quotient obtained by dividing the numerater of the integrand by the derivative
of the denominator when z = z0, i.e. in this case the residue is
II 1
T ~ 2ez0 + 2 — 2 / l — ez ’
and we finally obtain the following result:

dx
( 8)
I t+ e cos x

2. Let us also consider the integral


2*
I dx
(1 -|- e cos x )2
(0 < e<l).

On performing the same transformations as above we obtain the integral:

A f g dz.
i J (ez2 + 2 z + e)*
c
In this case z = z 0 will be the only pole inside the unit circle and it will be
a pole of order two. From [21], to determine the residue r 2 a t this pole we must
multiply the integrand by (z — z#)*, then take the first derivative of this
product and put z = z0. Let z = zY be the other zero of equation (7):

9. —
- 1-
------------------ ' ___________ •

its modulus being greater than unity. On performing the above operations we
obtain in this case:

r = g + zt
[ (z - zi)2l-o e2 (z — Zj)s z-o
and, subsequently, putting z = z 0 and bearing in mind the expressions for
z 0 and z1( we obtain the residue:
1
r =
4(1 - £2)a« ■
58] INTEGRATION OP A RATIONAL FRACTION 227
The theorem of residues gives us finally:
2ji
J (1
dx
+ e cos x)*
271
(1 - e2)3/2 (9)

58. Integration of a rational fraction. Consider the integral of a


rational fraction
+“
f <p (x)
J v(*) dx. ( 10)

If this integral is to have a meaning it is necessary and sufficient


[II, 82] that the polynomial y>(x) in the denominator should have no
real zeros and that its order should be at least two units higher than
that of the polynomial <p(x). If we also consider the function of the
complex variable
= y(z)
V(z) ’
then it will have the same property as the product zf(z) in that
it will tend uniformly to zero as z-> °°, i.e. it will not depend on
the manner in which z tends to infinity. In other words: for any
small positive e there exists a positive R „ such that | zf(z) \ < e when
| z | > R , . We will show that providing the function /(z) satisfies
this condition, its integral along any arc of the circle | z \ = R tends to
zero as R increases indefinitely.
Lemma. I f f(z) is continuous in the neighbourhood of the point at
infinity u and zf(z) -> 0 uniformly as z ->■ °o, then the integral of
f(z) along any arc of the circle \ z | = R tends to zero as R increases
indefinitely.
Writing down the upper bound of the integral in the usual way
as we did in [4], we have:

J7(z)dz = Jz/(z)idz < max | zf(z) j - -p-s,


on I a

w h e re s is le n g th o f th e a b o v e a rc I w h ic h e v id e n tly , d o e s n o t e x c e e d
2 tiR, s o t h a t f i n a l l y

Bearing in mind that zf(z) tends to zero on the arc as R increases


indefinitely, we have the direct proof of our lemma.
228 TH E APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [5 8

Let us now return to our example and integrate the rational fraction
9o(z): y>(z) over the contour consisting of the section (—R, -\-R) of
the real axis and the semicircle in the upper half-plane, for which the
above section is the diameter. We can take R so large that all poles
of the function, f(z), in the upper half-plane will be included in its
constructed semicircle. Denoting it by CR we have:
R
(11)

where S r denotes the sum of residues of the function f(z) at its


respective poles in the upper half-plane. As R increases indefinitely,
the right-hand side of the equation will not alter and the second term
on the left side will, according to the lemma, tend to zero, so that
we obtain the limit

i.e. integral ( 1 0 ), which is an integral of a rational fraction, is equal


to the product of 2 n i and the sum of the residues of the integrand at
its respective poles in the upper half-plane.

Example. Let us consider the integral

In this case the integrand has a single pole z = i of order to in the upper
half-plane. To determine the residue at this pole we must, according to [21],
multiply the integrand (z2 + l)~n by (z — l)n; the product thus obtained
must be differentiated (to — 1 ) I times with respect to z and it we can then
put z = i, i.e. the required residue is determined by the formula:
1 1 d n_1 (z — i)n ___ 1___ d"~‘ (z -f i)-”
i(» ~ 1)!. dz«-i (z2+ 1)" 2-i (to — 1 )! dzn-1
or
( - TO) ( - TO- 1 ) . . . ( - TO- n + 2) (2i)_2n+1 to (to + 1) . . . ( 2to — 2) .
r =
(TO- 1)! (TO - 1)! 2 2" - 1 1‘
and we finally obtain:

( 12)
59] CERTAIN NEW TYPES OP INTEGRALS CONTAINING TRIGONOMETRIC FUNCTIONS 229

59. Certain new types of integrals containing trigonometric func­


tions. Note that in the deduction of the above rule for evaluating
infinite integrals we did not use anywhere the fact that the integrand
f(z) was a rational fraction. It is sufficient for our purpose if the
function f(z) satisfies the following two conditions: it is regular in
the upper half-plane and on the real axis, except at a finite number
of poles, and secondly, as z —*■00 in the above domain, zf(z) -> 0
uniformly. In this case, as before, we obtain equation (11), in which
the second term on the left-hand side tends to zero, so that taking
the limit we have:
R
lim J / (a:) da; = 2ni r,(13)
R—*a> —R
where Z r is the sum of the residues of f(z) at its poles in the upper
half-plane. Dividing the interval of integration (—R, +i?) into two
parts {—R, 0) and (0, +f?) and substituting (—x) for x in the first
integral we can, instead of equation (13), write:
R
lim J f/(a;) + / ( — a;)] da; = 2n i ^ r
R-* = 0
or
j [/(*) + / (~ z)] da; = 2ni ^ r . (14)

Let us apply this result to the particular case when the inte­
grand is
f(z) = F (z) eimz, (m > 0) (15)

where the function F(z) satisfies the two conditions above. At the
same time, as can readily be seen, the function f(z) will also satisfy
these two conditions. To prove this it is sufficient to show that the
factor e,mz is regular in the whole plane and that it remains bounded
in the upper half-plane and on the real axis. We must have

e imz = e im ( * + . » and | e im z | = e ~my ( m > 0 ; y > 0 ) ,

from which it follows directly that | eimz | < 1 when z > 0. If F(z)
satisfies the two conditions above we can write:
es
J [.F (x) e,mx + F (— x) e ~ im*l d z = 2ni ^ r, ( 16 )
0
230 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNOTIONB [69

where E r is the sum of the residues of the function (15) in the upper
half-plane. Let us consider two particular cases. To start with we
suppose that F(z) is an even function, i.e. F(—z) = F(z), whence
CD

J F (x) cos mx dx = ni ^ r . (17)

If, however, F(z) is an odd function, i.e. F(—z) = —F(z) then:


co
f F (a;) sin mxdx = n (18)
6
Example 1. Consider the integral

cos mx
J X2-j- o2 dx (a > 0 ; m > 0 ).

In this case the function


F (z) =
1
o2 + s 2
obviously satisfies the above two conditions and is an even function, so that
we can apply formula (17). The only pole of the function
efmz
/(*) = a 2 + z2 (19)

in the upper half-plane is a simple pole z — ia. We can determine the residue at
this pole by the rule which we have used already and which can briefly be formu­
lated as follows: the numerator, divided by the derivative of the denominator.
In this ease, the above rule gives us the following expression for the residue of
the function (19):
0-ma
T = — — -------,
i2a
and we finally obtain:

cos mx
x2+ o2 dx 2a
( 20)
0
2. Let us consider the integral

I x sin mx
(x2 + a 2) 2
dx.

In this case formula (18) will be valid and the function


Imz
^ ^ = (z2 + a 2) 2
59] CERTAIN NEW TYPES OF INTEGRALS CONTAINING TRIGONOMETRIC FUNCTIONS 231

will have a single pole 2 = ia of order two in the upper half-plane. The residue
at this pole is determined by the formula
d f 2e,mz n
“ d2 W + a*)* (Z m ) \ \ 2. ia

d \ I! _ _Z!Le-m
d2 L (2 + ia)z J Iz-ta 4o

which gives us directly the final result:

x sin mx _ nm _ ma
J (x2 + a 2) 2 d x ~ ~ l5" e
(21 >

Note. Generally speaking we have no right to write formula (13) in


the form
+ CD

J f (x) da; = 2ni ^ r . (22)

In fact, the infinite integral

J / (*) d *

is determined as the sum of the limits of the integrals


r 0
J f (x) dx and J f ( x) dx
0 —R

as R tends to (+°°). If these limits do not exist separately, but


the sum of the above integrals tends to a finite limit, i.e. a finite
limit exists:
R
lim f / (x) dx,
R-f+~ —R
then this limit is known as the principal value of the integral in the
infinite interval and is denoted as follows:
+ 00 +ii
v. p. J f[x) dx = lim J /(x) dx. (23)
—00 Jt—
►+» —R
The integral in formula (13) should be taken in the sense of its principal
value. But if for any particular reason we know that this integral
exists as an ordinary indefinite integral, then this should not be
assumed, for then the principal value of the integral will be the same as
232 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [60

the usual indefinite integral. In [26] we defined the principal value


of an integral in the case when the continuity of f(x) is disrupted at
any particular point at a finite distance.

60. Jordan’s lemma. The conditions placed on the function F(z) in


the above paragraph can be made less restrictive so that formulae
(17) and (18) hold, by using a lemma which will be of great impor­
tance in what follows:
J o r d a n ’s Lemma. I f in the upper half-plane and on the real axis
F(z) satisfies the condition: F(z) -> 0 uniformly as 2 -*-°°, and m
is a certain positive number, then as R —> + ° °
J F (z)eimzdz^>-0, (24)
Cb
where Cr is a semicircle in the upper half-plane, centre the origin and
radius R.
Introducing the polar coordinates z = Re1*, we can rewrite integral
(24) as follows:
J F (Re1*) elmR^cos,f'+isin,p) iRei,pd<p,
o
and therefore, taking into consideration the fact that | ieimRcos *+“p| _ i
we have
71

| f F( z ) e imsdz\ < J |i?’ (^e'> )|e-mWsin',,I? d ^


CB 0
or
J F{z)elm2dz max F(z) J e~mRain,p Rdp. (25)
Cr o n Cb o
It is given that | F(Rel,p) | tends to zero as R —»- °° uniformly with
respect to <p where 0 < <p < n and, consequently, it is sufficient to
prove that, as R-> °°, the integral

J e~mRsin'>’Rd<p (26)
o
will be bounded. Dividing the interval of integration into two parts:
(0, jt/2) and (nj2, n) and substituting in the second integral the variable
rp by (ji — <p), we obtain integral (26) in the form:

2
2 J e - mRsin,p Rd<p.
o
60] JORDAN'S LEMMA 233

We will now proceed as we did in [56], Dividing the interval of


integration into two parts and increasing the positive integrand,,
we obtain the inequality:
31 31

~2 a ~2
2 j e - mRsin,pB dcp < 2 J e - m*sin,,. f f d < p + 2J e- mRsina R d<p.
0 0 a

The last two integrals are in their final form and we obtain the
following inequality:
31

2"
2 f e- mRsi(1fiJdffi < ---------[— e- m/?sin’T =a + 2e—mRsinai? f-J —a] .
J r mcosaL |,,=0 12 )
0

The second of the above terms tends to zero as B -*■<=•°, and


the first term tends to the finite limit 2/m cos a so that the whole
sum remains bounded as R —►00. The same may be said for integral
(26) from which the result of the lemma is derived.
By using the lemma we can, for example, prove formula (18),
when less restrictive requirements are made with respect to the
function F(z). In fact, we required^ earlier that in the upper half­
plane and on the real axis zF(z) —>■0 as | 2 | -* •00 and this condition
was necessary for the integral
J F[z) elm2dz
Cr

to tend to zero in the upper semicircle as R —>- 00. The lemma shows
that it is sufficient for F(z) -*■0 as z -*■0 0 and therefore it is sufficient
to use formula (18) in the assumption.

Example. Consider the integral

Jf xa*sin+ mx
a» ^, (o,> .
0 ; m > 0).

In this case the function


F(z) =
z2 + o2
satisfies all conditions of formula (18) and therefore we need, as before,
determine only the residue of the function
Z6lmZ
F (2 ) eimz =
22+O 2
234 THE APPLICATION' OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [61

a t the pole z = ia in the upper half-plane. This will be a pole of order one
and the corresponding residue is determined by the usual rule: the numerator
divided by the derivative of the denominator, i.e.
1 _
2z z-ia = T e
and finally

f x s*n Tnx ,i _ _ .i L P~ma (27)


J x2 + a2 2

61. Contour integrals of certain functions. From a knowledge of the theory of


residues it is easy to construct contour integrals for non-continuous functions.
Consider, for example, the function <p(t), which is zero when t < 0, and
unity when t > 0 , i.e.
tp (t) =

Fio. 58

We will show th at such a function can be represented by the contour integral

i29'

in which t appears as a parameter in the integral. The contour of integration


includes the whole real axis but the origin z = 0 which is a pole of the integrand
is avoided by the circumference of a small circle, centre the origin in the
lower half-plane (Fig. 57). Consider the auxiliary contour lR, which consists
not of the whole real axis but only of its section (—R, + i 2), surrounds the origin
and includes the semicircle Cr in the upper half-plane, centre the origin and
radius M. If t > 0, then Jordan’s lemma can be applied to integral (29), so
that the integral over the semicircle will tend to zero as R increases. The
61] CONTOUR INTEGRALS OP CERTAIN FUNCTIONS 235

integrand has a single pole inside the contour, viz. at the origin z = 0, where
the residue is unity. Therefore

1.
h
Passing to the limit we obtain
1
2ni (*> 0).

Let us now suppose that t< 0. Consider the closed contour which consists
of the above section (—R, + iJ) of the real axis, surrounds the origin and includes
the semicircle C'R of radius R, not in the upper but in the lower half-plane
(Fig. 58). Inside this contour our function has no singularities and therefore
the integral along this contour is zero.
We will now show th at as R increases indefinitely, the integral over the
lower semicircle will tend to zero. In fact, if we change the variable of integra­
tion z so th at z' = —z, then the lower semicircle CR will be transformed into
the upper semicircle <7# and we have:

Ck CB

I t is given th at t < 0, i.e. —t > 0, and Jordan’s lemma shows that the
latter integral does, in fact, tend to zero. Passing to the limit, as before, we get:
1 f eitz
2 ^ J — d2 = ° <*«»•

Consider also the integral when t = 0.

* f-dz (30)
2m J z

If we consider, as before, the section (—R, + R) of the real axis wo have to


evaluate the increment of log z as it moves along this section and round the
origin. At the end of the path the real part of log z will be log R and, con­
sequently, it has received no increment. The imaginary part, which is equal to
i arg z, must receive an increment iti when surrounding the origin along the
semicircle, but on other sections of the path it remains unaltered. Thus integral
(30) is equal to 1/2 along the section (—R, +JJ). Consequently we should obtain
the same result in the limit as R -*• oo, i.e.

-2m
i _ Jr izd z = i2. (31)

In this case it was essential th at both the upper and lower limits should tend
to oo and the same absolute values, i.e. integral (31) should be taken in its
principal value sense in the interval ( —oo, -foo) when surrounding z = 0 .
236 THE APPLICATION OP THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [61

Integral (29), when t ^ 0, will be convergent in the usual sense of the word
with respect to infinite limits. In fact, separating the real and imaginary parts
we obtain
f cos tz , , f sin tz ,
— 2— “2 nn“ — -— dz (a > 0 ).
a a
We proved the convergence of the second integral in [11, 83], The convergence
of the first integral can be proved in exactly the same way.

a a b

F i g . 59

Thus, when t^O integral (29) gives the function (28). When t = 0 the integ­
ral only has a meaning in the principal value sense and is equal to 1/ 2 .
Consider now another example where the function is zero everywhere except
on a certain finite section where it is unity, i.e.

y>(t) = 0 when t < a and t> b ; y>(t) = 1 when a < < < 6 . (32)

I t is not difficult to represent the above function as the difference of two func­
tions of this kind; hence
e(Q-t)z
1
v>(0 = 2ni J z
d z — 1 r e,(a- ° z ,
2ni z J (33)

Both terms vanish when t > b. In the interval a < t < 6 the first term on
the right-hand side is unity and the second term is zero so th at the difference
is unity. Lastly, when t < a and both these terms are unity, the difference is
zero and we have, in fact, obtained function (32). The graph of this function
is shown in Fig. 59.
Consider now the function which is zero when t < 0 and which, by starting
with t = 0 , begins to decrease from unity exponentially:

9>l ( < ) = 0 (t< 0 ); 9>,(t) = e'—at (t > 0 , a > 0 ). (34)


61] CONTOUR INTEGRALS OF CERTAIN FUNCTIONS 237

The graph of this function is illustrated in Fig. 60. I t can readily be shown
that this function can be represented by the contour integral:
+,so
(35)
f

where the real axis is the contour of integration. The proof is exactly the same
as for formula (29). In this case the residue of the function

at the pole z = ia is equal to e~al.


Finally let us consider the function which is zero when t < 0 and which gives
the sine curve when t > 0 (Fig. 61).
Vi (t) = 0 when t < 0 ;
(36)
Vi (t) — sin at when t > 0 (a being real) . )
In the same way as before it can readily be shown that this latter function
can be represented by the contour integral:

* ■ < '> - * [ - & <37>

where the contour of integration is the real axis, which surrounds the pole
z = a of the integrand. In this case
the residue is

e,,a = cos at -f- i sin at,

so that, separating the real and


imaginary parts, we obtaiD formula
(37), where <H is the symbol of the
real part.
In some cases the above for­
mulae are written in a different
way, viz. we integrate not along
the real but along the imaginary'
axis and the pole is described
from the right side, i.e. from th at side of the imaginary axis near which the
real part of the complex number is positive. To obtain this new contour of
integration it is sufficient to rotate the plane about the origin through an angle
jz/ 2 in the counter-clockwise direction, i.e. to replace z by a new variable z',
where z' = iz or z = z’ji. By introducing this new variable we have in place
of formula (29)
1 f e*
238 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [62

In formula (35) we now obtain the pole ia not on the imaginary axis but on
the negative part of the real axis and we thus obtain the following expression
for the function rp^t):

J ■ V T 7d*'- <35'>
- - /
Similarly, we have the following expression for the function

>
The contents of this paragraph are directly connected with Laplace’s trans­
formation with which we shall deal in Vol TV.

62. Examples of integrals of many-valued functions. We will now consider


a few examples where the integrand functions are many-valued functions of
the complex variable. For the first example consider the integral

J ( - * ) a - 1 G(*)<k, (38)
I
where a is a certain real number and Q(z) is a rational function, which is such
th at za Q(z) 0 if either z 0 or z oo. The integrand is many-valued so
th at by describing a circuit round z = 0 in the counter-clockwise direction,
(—z) describes the same circuit and consequently the amplitude of this
expression acquires the term 2 n\ the expression itself acquires the factor e2ltf
and(—z)a_1 becomes (—z)a_1 e2*”“ 1)nlt i.e. in this case the function acquires
the factor e^ a~1)71', which is other than unity, providing a is not an integer.
The origin is therefore the branch-point of the integrand. To make the func­
tion single-valued cut along the real axis starting from z = 0. In the cut T-
plane our function will be single-valued and to determine it fully we have to
fix the amplitude of (—2) for a point in the T-plane. Let us agree th at on the
upper edge of the cut, where z is positive, the amplitude of the negative number
(—z) is equal to ( —ji). Describing a circuit about the origin round a closed
contour we come from the upper edge of the cut to the lower edge and in
the course of this the amplitude of (—z) acquires an additional term 2 ji, so
th at the amplitude of (—z) on the lower edge of the cut will be equal to n.
Denoting the modulus of z by x we have:
(— z) = xe~,n on the upper edge,
(—z) = xeln on the lower edge
and consequently
(—z)“ - 1 = x a~x on the upper edge,
(—z)a~‘ = xa~ x e1 n on the lower edge.
Let us now select the contour of integration for integral (38). We take for the
contour of integration the following curve which consists of four parts: the
section (e, R) of the upper edge of the cut, the circumference CR, centre the
62] EXAMPLES OP INTEGRALS OF MANY-VALUED FUNCTIONS 239

origin and radius R, in the counter-clockwise direction, the section (R, e) of


the lower edge of the cut and, finally, the circumference Ce, centre the origin
and radius e, in the clockwise direction (Fig. 62). To integrate along the positive
part of the real axis we assume th at the rational fraction Q(z) has no poles there.
In accordance with the fundamental theorem of residues the integral (38)
will be equal to the product of 2ni and the sum of the residues of the integrand
a t all poles of the rational fraction Q(z); the latter are also poles of the integrand.
We are assuming all the time that
e is taken so small and R so large

th at all the above poles will be


included in the region bounded b y '
our contour of integration. We will
now show th at the integrals along
the circles CR and Ca tend to zero
as R -*■ oo and e 0. In fact,
applying the usual inequality we
have:

I J (-* )a-1«(*)da|<
Ca
< 2nR • Ba _ 1 max | Q (z) | =
on Ca

= 2 JtRa max | Q (z) |.


on Cg
F ig . 62
I t is given th at za Q(z) -►0
as | z | oo; therefore the above
expression also tends to zero as oo. Similarly, on the circumference
Qe we have the inequality:

I] " ( - « ) “ 1 Q (2) d z I < 2 ; r s a max | Q (z) |,


c£ on c„

since za Q(z)->- 0 as z->-0 ; therefore the above expression also tends to zero
as e -<■ 0. Hence in the limit only integrals along the upper and lower edges
of the cut remain and the value of the integrand on these cuts is determined
by formula (39); this gives us:

R
lim J [x ® - 1 e“ '“ (a-1) Q (x) - xa - 1 e,'n(a" 1) Q (x) 1 dx = 2 ni 2 r,
«-0 «

where Hr denotes the sum of residues of the function (—z)a~1Q(z) at all poles
within a finite distance.
Bearing in mind that e ' h = eiK = —1, we can rewrite the above formula
as follows:

(e,,ta — e~ina) J xa _ 1 Q (x) dx = 2ni 2 t


0
240 THE APPLICATION OP THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [62

or, (from Euler’s formula):


00

J x a~ 1Q (x) dx = r> (4 0 )

Formula (40) makes it possible to evaluate many definite integrals in which


the primitive is not expressed in final form. Let us remind you once again
of the conditions to which the function Q(z) is subject, for the above formula
to be valid. The function Q(z) must be a rational fraction with no poles on the
positive part of the real axis and it must also satisfy the conditions:
za Q (z) -»• 0 as z -> 0 and z -*■ oo.
As a particular example let us consider the integral

J x dx (0 < a < 1 ). (41)


0
I t can readily be seen th at in this case the function

Q (z) = 1 + 2
satisfies all the above conditions and has a single pole z = —1. At this pole
the function
(-z )8- 1
1+ Z
has a residue which is evaluated by the rule: numerator divided by the derivative
of the denominator, i.e. this residue is

Note th at when evaluating the function (—z)0 -1 at the point z = —1 we


must bear in mind the definition of many-valued functions which was given
above, viz. on the upper edge of the cut the amplitude of (—z) is equal to ( —it)
and, consequently, when describing half the circuit round the origin on the
negative part of the real axis the amplitude of (—z) will be zero. In other words
r= 1 .
We finally obtain the following expression for integral (41) from formula
(40):
“ „a-i _
J t+X
■dx = (42)

As the second example of an integral of a many-valued function consider the


integral:

[/ A + 2 — + dz. (43)
62] EXAMPLES OF INTEGRALS OF MANY-VALUED FUNCTIONS 241

and suppose th at the trinomial A + 2 Bjz -)- Cjzi has real coefficients and
distinct real zeros z = z2 and z = z2, where 0 < z, < z2.
We also suppose that A < 0 from which follows directly th at the above
trinomial will be positive when z, < z < z2. We integrate (43) along the section
z, < z < z2 of the real axis on which the radical is taken to be positive.
The integrand

will have branch-points of order one at the points zt and z2. If we make a cut
along the section (Zj, z2) of the real axis then the function (44) will be regular
and single-valued in the cut T-plane [19].
Let us suppose th at the radical is positive on the lower edge of the cut. To
reach the upper edge where the radical is negative we have to pass one of the
branch-points [19]. Let us take our integral along the whole contour of the cut
in the positive direction, i.e. we take the integral of the function (44) along the
lower edge from zx to z2 and along the upper edge from z2 to zl. The first part
of this integral will give integral (43). When integrating along the upper edge
the integrand will change its sign but the direction of integration will also be
reversed so that the value of the integral along the upper and lower edges
will be the same, i.e. the value of the integral along the whole contour of the
cut will be twice the value of the integral (43).
According to Cauchy’s theorem we can, without changing the value of the
integral, continuously deform our closed contour, providing it does not leave
the domain in which the function (44) is regular. If I is any closed contour, which
includes the above cut and the point z = 0 , which is a pole of the function
(44), remains outside I, then it follows from above:

J =Y j\lA + 2^ + i r dz- (45>


We will now expand function (44) in the neighbourhood of z = co and of
z = 0. In the first case we can write:

^A+2^- + — = U [ l + ^ ^ + ^ ) \ ,
and using Newton’s binomial formula we obtain:

^ + + + ,46,

Let us determine the value of the radical )/A in this formula. We use for
this purpose the left-hand side of formula (44). We know that the radical is posi­
tive along the lower edge of the section (zx, z2). To reach the section (z2, + co) of
the real axis the point z = z2 must be circumscribed in a counter-clockwise direc­
tion. As a result the amplitude of (z — z2) is increased by n and the amplitude
of the expression (44) by ji/2, i.e. instead of zero this amplitude becomes ji/2. In
other words the function (44) must be considered to be positively imaginary on
242 THE APPLICATION OP TH E TH EORY OP RESIDUES, FRACTIONAL FUNCTIONS [6 3

the section (z2, +co) of the real axis. (By a positively imaginary number we mean
ai when a > 0.) I t follows from (46) th at the radical yA is positively imaginary.
Similarly, to reach the section (0, z,) from the lower edge of the section (zv z2)
the point z = zt must be described in the clockwise direction, as a result of
which the amplitude of the expression (44) will be ( —n/2), i.e. this expression
will be negatively imaginary on the section (0 , zt).
Expanding the function (44) near z = 0:

or, using the binomial expansion:

IIa + 2T + ^ = ^T~{1 + 1Ts + ---)’ (47)


and, in agreement with what was said above, yo must be negatively imaginary.
I t is given th at A < 0 and it follows from z2 > zt > 0 that C < 0.
According to Cauchy’s theorem the integral of the function (44) along a
large closed contour L near z = oo is equal to the sum of integrals along the
above contour I and along the contour A which surrounds z = 0; every integral
being taken in the counter-clockwise direction. The integrals along L and A, are
equal to the product of 2ni and the coefficient of z” 1 in the expansions (46)
and (47) and consequently:

l | ^ + 2 T + 5 - dZ = 2^

hence the formula (45) gives us the final value of our integral (43):
____________
J = \ (48)

63. Integration of a system of linear equations with constant


coefficients. We will now apply the theory of residues to the integration
of a system of linear homogeneous equations with constant coefficients.
Consider such a system:
xx = an xx -f- a12 x2 + . •+ Oln '
x2 = a21 xx + a22 x2 + • ■+ °2n Xn

x„ = anl xx -|- an2 x2 -)-. • ^nn ^n> .


where the aik are constant coefficients and the x's are derivatives of the
required function xs with respect to the independent variable t. We
require the solution of this system in the form:
X,s = J v > s (2)e'z (s = 1, 2, .. ,,n), (50)
R
63] INTEGRATION OP A SYSTEM OF EQUATIONS W ITH CONSTANT COEFFICIENTS 243

where f s(z) are the required rational functions of z; the symbol

x ^ /(z )
R

will denote, here and in future, the sum of residues of the function
f(z) with respect to all singularities within a finite distance. In formulae
(50) the functions after the symbol of the sum of residues depend
not only on the complex variable z, with respect to which we are
calculating the residues, but also on the real parameter t, so that
the sum of the residues will, generally speaking, also be a function
of this parameter t. Owing to the fact that z and t are not connected
in any way with each other, we can, when differentiating the function
(50) with respect to t, differentiate under the symbol of the sum of
residues, i.e. we obtain one and the same result if we differentiate the
function
<Mz)e'z (51)
with respect to t first and then take the sum of its residues, or if
we take the sum of residues of the function (51) first and then dif­
ferentiate with respect to t. Therefore, apart from the formulae (50)
we also have the following formulae:

x's = ^ z<Ps (z) efz (« = 1,2, ...,» ) . (52)


R

We substitute all this into our system (49) and collect all terms
together:
^ [(®u ~ z) <h (z) + aa <P2 (z) + • • ■+ aln <pn (z)] etz = 0,
i?
2 K i <Pi (z) + (o22 - z) <p2 (z) + • • • + a2n <pn (z)] e,z = 0,
R

2 k i <Pi (z) + an2 <p2 (z) + ... + (ann - z) (pn (z)] e'z = 0.
R
These equations will obviously be satisfied if we equate the expres­
sions in the square brackets to any arbitrary constant, so that we
get under the symbol of the sum of residues a function in the form
Cetz, which has no singularities at all within a finite distance. Denoting
these arbitrary constants by —Cv —C2, . . . , —Cn we obtain a system
of ordinary algebraic equations of the first order for the determination
244 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [63

of <ps{z):
(On — z) <Pi (2) + o12<p2(z) + . . . + aln <pn (z) = - Cv
an <Pi (z) + (o22 — z)<p2{z) + . . . + a2n <pn (z) = - C2,

«m <Pi (2) + on2 <p2 (z) + .. . + (ann - z) <pn (z) = - C„.


We solve this system by Kramer’s formula:
Ag(z) (5 = 1,2, (53)
A{z)
where
®12........ a l n
<3
1
ii—

11

Zl(z) = °21> ^22 — • • •>°2n (54)


G'nl* a n2 ..............

and As(z) is obtained by changing the elements of the sth column by


terms (—Ck) in the determinant A(z). Note that the determinant A(z)
represents the left-hand side of a familiar equation [IIIx, 17]. The
expressions obtained in (53) must now be substituted in formula (50).
Making these substitutions we obtain the solution of our system
in the form:
= ( « = 1.2. ••■.»), (55)

where the meanings of A(z) and As(z) are explained above.


We will now show that the result obtained satisfies the original
conditions
* 1 |f = 0 = C i> x 2 |f = 0 = ^2> • • • I X n |t = 0 = • (56)

We will only test this for xv We have:

M»=0 =js4trr-
xJ ^ A(z) (57)

The denominator of the above fraction is given by formula (54)


and it is evidently a polynomial of the nth order, the last term of
which is (—l)n zn. The numerator of the fraction in formula (57) is:

- C v a i2, ■> a ln
C 2 1 ^22 2, • • ■< a in
AAs ) =

- C n, °n2> ■» ®nn ^
63] INTEGRATION OF ASYSTEM OF EQUATIONS WITH CONSTANT COEFFICIENTS 245

Multiplying the elements of the first column we obtain a polynomial


of the ( n — l)th order, the last term of which is (—1)" C x s'1-1; we
can therefore rewrite formula (57) as follows:
( - 1 )»0|g"-» + ...
*i ( - 1)nz”+... (58)

where the dots indicate terms of the polynomial with lower powers,
which are of no importance in subsequent calculations.
We will now establish a certain general rule for the sum of residues
of a rational fraction.
Lemma. T h e s u m o f the r e s i d u e s o f a r a t i o n a l f r a c t i o n w i t h re feren ce to
i t s p‘ o l e s a t a f i n i t e d i s t a n c e , i s e q u a l t o t h e c o e f f i c i e n t o f z-1 i n the
e x p a n s i o n o f the r a t i o n a l f r a c t i o n i n th e n e ig h b o u r h o o d o f th e p o i n t a t
in fin ity .
In fact, suppose that our rational fraction has the following
expansion in the neighbourhood of the point at infinity:
/(z) = J 2 M * . (59)
k
Consider the integral
asr I * ® * 2’
where C R is the circumference of a circle, centre the origin and radius
R . If R is sufficiently large all the poles of f ( z ) will lie inside C R and the
integral will give the sum of the residues at these poles. Also, if R is suffi­
ciently large C R will be near the point at infinity and we can therefore
apply expansion (59) to solve the integral; it follows directly from
this that the integral will be equal to 6^, which proves the lemma.
N o t e . Earlier in [17] we called the coefficient in the expansion
(59) with its sign reversed the residue of the function f ( z ) at the
point at infinity, i.e. this residue is equal to (—6_j). For this reason
our lemma can be formulated as follows: t h e s u m o f t h e r e s i d u e s o f a
r a tio n a l f r a c tio n a t a l l i t s p o les, i n c lu d in g th e p o i n t a t i n f i n i t y , i s e q u a l
to zero.
Let us now apply the above lemma to the expression (58). Note
that near the point at infinity the fraction can be expanded as
follows:
( - 1 )nCl zn- ' + ... 0, ^
( - l)nz" + ... z Z* '1“ ■•
and the above lemma gives us directly | <=0 = C^, it can similarly
be shown that x s | ,_0 = C s . Hence the solution given by formula
246 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [63

(55) satisfies the original conditions (56), i.e. the arbitrary constants
Gs in the polynomial As(z) take the place of the original conditions.
Therefore our formula (55) gives the general solution of the system.

Example. Consider the system


XX — X2 + X3
x2 = x i + X3
x3 = Xi ~j~ x3.
In this case
- 2 , 1, 1
d(2) = 1, —2, 1 ,
1, 1,-2,
or
A (2) = - 2 (2* - 1) + 2 (z + 1) = (Z + 1) ( - 2* + 2 + 2),
and for the first of the required functions we obtain the formula:
- C i, 1, 1
~ C 3,~ z , 1
____ ^3» b 2> _____ tz
xi = 2 (2 + 1) (-2 = + 2 -4- 2) ’
R
or, by solving the determinant and cancelling by (1 + z):

x (1 2) ~ —0 3 iz
R — z2+ z + 2

The denominator has zeros z = — 1 and z = 2. We will calculate the residues


at these points by the usual rule: the numerator divided by the derivative of
the denominator, and we obtain:

- (4 °-' T - T 0*) + (t °- + 1 °- + T c -) •”
Notice that in this case the polynomial A{z) has a double zero at r = —1,
but in spite of this the coefficient of in the expression for x l is not a poly­
nomial in the £th power but simply a constant.

In those cases when the equations are not homogeneous (compulsory


vibrations):
x's = aslx1 + . . . + asn + fs (t) {s = 1,2, . . .,n), (60)
where fs(t) is a given function of t, the required solution should have
the following form:
x _ _ (0 Ats (z) + ■. ■+ C„ (t) A m (2) (61)
83] INTEGRATION OF ASYSTEM OF EQUATIONS WITH CONSTANT COEFFICIENTS 247

where A ik(z) is the algebraic complement of the elements of the deter­


minant A ( z ) , and C k(t) is a required function of t (the variation
method of arbitrary constants) [II, 25]. Substituting (61) in (60)
and bearing in mind that when the C k are arbitrary constants the
formulae (61) give the solution of the homogeneous system, we obtain
the following equations for the derivatives C k (t):
O 'l M A l s (z ) + • • •+ 0'n ( * ) -Arts ( z ) „ tz ___ 4 t*\
~ 2 - Z(i) e ~/sW
R
1,2, . . . , » ) . (s= (62)
We will show that this system can be satisfied, providing:
C [ (t) = e-'*/, (t); • ■• ; Cn (0 = e ~ tzf n ( t ) . (63)
In fact, making these substitutions, we get on the left side of (62):
/ . (t ) A l s ( z ) + . . . + f n ( t ) A ns(z)
~ 2 - A(z) (64)
R
If i t6 k , then in the algebraic complement A tk( z) two elements
situated on the main diagonal of the determinant A ( z ) will be cancelled,
viz. ( a n — z ) and ( a kk — z ) ; therefore A ik( z) will be a polynomial of
the order (n — 2) in z . As a result of the above lemma:
A f k (z) _~
(i¥= k),
4 *(*) ~
since the expansion of A lk( z) : A ( z ) near the point at infinity begins
with the term a/z2 and there is therefore no term in z-1.
The algebraic complement A u ( z ) will be a polynomial of order ( n — 1)
with the last term (—l)n- 1zn_1 (cf. above) and consequently:
A n (z )
A (z)
1.

It follows directly from this that the preceding expressions (64)


are equal to f s(t). The formulae for C k (t) give:
t
C k (t) = J e~TZ/fc(r) dr (k = 1,2,..., n),

and we chose the constant of integration so that C k( 0) = 0 (purely


compulsory vibrations).
Substituting in (61) we finally have:

xs = —^ J fl (t) Als (s) +a ~(g)+ fn (t) Ans (z) e(<~T)z d r . (65)


248 TH E APPLICATION OP TH E TH EORY OP RESIDUES, FRACTIONAL FUNCTIONS [6 4

64. The expansion of a fractional function into partial fractions.


We shall now apply the basic theorem of residues to the expansion
of a function into an infinite series. Let the function f(z) be uniform
and homogeneous in the whole plane except at a few isolated points
which are its poles. Such a function is usually known as a fractional
or meromorphic function. A rational fraction is one example of a
fractional function. As the second example take cot z = cosz/sinz
which has poles at the points where sin z vanishes.
This latter meromorphic function has an infinite number of poles.
Notice that when a meromorphic function has an infinite number of
poles then in any confined part B of the plane the number of poles
should, at any rate, be finite. Otherwise we would have in B at
least one limit-point for these poles, i.e. such a point z = c, that any
small circle with centre at z = c would contain an infinite number
of poles of the function f(z). This point z = c would be a singularity
of f(z) other than a pole, for it follows from the definition of a pole
[17], that it should be an isolated singularity. But it is given that
f(z) has no singularities other than poles. Once we have in a confined
part of the plane a finite number of poles we can number them in
the order of their non-decreasing moduli, so that denoting the poles
by ak we have:
| I ^ Ia2 I ^ I I ^ • • •>

where | an \ —>- + ° ° as n increases indefinitely. At every pole z = ak


our function will have a definite infinite part, which will be a poly­
nomial with respect to the argument 1/(z—ak), without the constant
term [17]. Denote this polynomial by

<66>
We shall now show that by making certain additional assumptions
the fractional function /(z) can be represented by a simple infinite
series, the terms of which are expressed by infinite parts (66). Let us
formulate the condition to which we must subject the function f(z).
Suppose that a sequence of closed contours Cn which surround the origin
exists and which are such that every contour Cn lies inside the contour
^n+l- Let ln be length of the contour Cn and 8n be its shortest distance
from the origin. We assume that 8n —*■°°, i.e. that the contours Cn
widen indefinitely in all directions as n increases. We also suppose
64] THE EXPANSION OP A VRAOTIONAL FUNCTION INTO PARTIAL FRACTIONS 249
I
that the relationship ln : dn remains bounded as n increases indefinitely,
i.e. a positive number m exists such that

is - C m . (67)

If, for example, Cn are circles, centres the origin and radii rn, then
ln = 27i Tn and <$„ = rn, so that ln : 6n = 2n. We now suppose that
the modulus of our fractional function f(z) remains bounded on all
contours Cn, in other words, a positive number M exists, such that
on any contour Cn the inequality given below is satisfied:
1/(2) I < AT (on <7n) . ( 68 )
Consider the integral:
i r /(«')
2m J z' —z dz', (69)
Cft
where we integrate in the positive direction and the point z lies
inside Cn and is other than ak. Consider also the sum of infinite parts
which refer to poles ak, inside Cn:

i« =2 (Cn)
4 (d z ).
v k)
P°>
where (Cn) below the symbol of the sum shows that only poles situated
inside Cn must be added.
The integrand of (69), which is a function of z', has in Cn a
simple pole z' = z, which is due to the vanishing of the denominator,
and poles z' = ak, which are due to infinite parts of f(z'). The residue
at the pole z' = z is determined by the rule: the numerator divided
by the derivative of the denominator:
/(«') /(«')
2'=2 1 ^=2 = /(z).
The residues at the poles z' — ak will be the same as for the function
(gy) (71)
z' —z
In this latter function con(z') is a rational fraction in which the
order of the numerator is lower than the order of the denominator
and all poles are situated inside Cn. We will show that in this case
the sum of residues of the function (71) at the poles ak will be
250 THE APPLICATION OP THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [64

In fact, the function (71) is a rational fraction of 2 ' in which the


order of the denominator is at least two units higher than that of
the numerator, for <on(z') is already a rational fraction in which the
order of the denominator is higher than the order of the numerator.
In the neighbourhood of z’ — °° we therefore have the following
expansion:
<0n (*') _ ° 2 | a3 I
2' - z g'2 ~r Z'3 1" • • • >

and the integral of the function 171) around a circle of a sufficiently


large radius will be equal to zero, i.e. the sum of residues of the
function (71) at all its poles within a finite distance, is zero. Its
residue at the point z' = z is, obviously, equal to con (z) and, con­
sequently, the sum of residues at the remaining poles ak is equal
to the expression (72). Applying the basic theorem of residues to
integral [69] we obtain:
i r /(»') 1
2ni J z' —r dz' = j { z ) ~ 2 G k ( z - a k )■

Suppose in the above formula that z = 0, where the point 2 = 0


is not a pole of f(z):
1
2ni

Subtracting this equation from the one above we have:

25- J T $ h j “ = <<*> - / (0) - 2 [ff. (—


(C„) <cn)
) - ( - i ) • <™>

We will now show that the integral on the left side of the
above equation tends to zero as n increases indefinitely. In fact,
bearing in mind that
| 2' | > <5„ and | z' — 2 1> | z' | — | 2 1> <5„ — | 2 ],
we have from (68)
Min
Jr 2' (2' - 2) 02
d2'
1*I) ’
c„
or from (67)
r /(*'> d^i < Mm
J 2 / (C/ 2) a Z|
65] TH E FUNCTION cot 2 251
from which it follows directly that the integral tends to zero, as
dn —► Therefore in the limit formula (73) gives

/ (*) - / (0) - t a 2 h ( t^ ) - ( - ■£)] =


or

''"/I''
As n increases indefinitely, the contour On will widen indefinitely,
and more and more new poles a k will find their way within Gn, so that
in the limit we have on the right-hand side of (74) an infinite series;
hence formula (74) gives f ( z ) in the form of an infinite series:

+ (7t)

Strictly speaking we should, according to (74), group together all


terms in the infinite series (75) which refer to poles situated between
C n and C n+1. However, if we are convinced that the series (75) is
convergent without performing this grouping of its terms, then we
can deal with the infinite series (75) in the usual way.
If instead of condition (68), which tells us that the modulus of the
function f ( z ) is bounded on the contours C n, we use a wider condition,
viz. that f ( z ) does not grow on the contours G n faster than a certain
positive power z n, i.e. on all contours C n the following inequality
holds:
LQ- < M (on C n),

then in place of formula (75) we obtain the following expansion formula:

f W = f m + + ... + +

+ <"»

where the symbol x [ PH2 ) denotes the first ( p + 1) terms in the expan­
sion of the function G k [1j ( z — a*)] into a McLaurin’s series.

65. The function cot g. Consider the fractional function


252 TH E APPLICATION OP TH E TH EO RY OF RESIDUES FRACTIONAL FUNCTIONS [6 5

From Euler’s formula we have:

----- 2i
and from this it follows directly th at the equation sin z = 0 is equivalent to
et2Z= 1; this has zeros z = kn (k = 0, ± 1, ± 2 , .. .)> i-e- sin z has real zeros only,
the values of which, which are well known from trigonometry. The function (77)
will therefore have poles at the points:
z= 0, ± ji , ± 2.t , (78)
We can show th at the modulus of
the function (77) is bounded in the
whole plane if we isolate the points
(78) by small circles Ag of same radius
g, where g is any given positive num­
ber. Owing to the fact th at the func­
tion (77) has a period j i , it is sufficient
to investigate it in the strip K , boun­
ded by the straight lines x = 0 and
x = 7i (Fig. 63), in which the poles
z = 0 and z = ji are isolated by the
above circles of radius g with centres
at the respective points. In any con­
fined part of the strip K our function
(77) is continuous and therefore also
bounded. I t remains to be shown th at by moving up or down the strip ad infini­
tum, the modulus of the function (77) remains bounded. Suppose, for example,
th at we move towards infinity up the strip K , i.e. if we assume that z = x + iy,
then y -- co and x varies in the interval 0 < x < n. We have:
. e'2 + e -iz Bixe~y + e~ix ey
cot z = »
e'2 - e~iz pfx e - y ■e~lx oy ’
whence, substituting the modulus of the numerator by the sum of the
moduli, and the modulus of the denominator by the difference of the moduli,
we have:
e* + e-* 1 + e~2y
I cot z I <
e y - e -y 1 - e"2?
As y increases indefinitely the right side tends to the limit I and, conse­
quently, for all sufficiently large y ’s we have, for example, the inequality:
| cot z | < 1.5 .
In exactly the same way we can act in the lower part of the strip K , and our
proposition is thus proved.
Notice th at the same proof also applies to the fractional function

(79)
sin z
65] THB FUNCTTION cot t 263
with poles at the same points and a period of 2n. The modulus of the function
(79) will be bounded if its poles are isolated by small circles of the same radius,
which can be as small as we please.
Let us return to the function (77) and adopt as the contour the circles Gn,
centres the origin and radii (n + 1/2) n. These circles satisfy the condition (67).
Also by taking q sufficiently small (for example, smaller than jr/2), we can
say th at the circles Cn will not pass through the circles Ag, which are isolated
in the plane; hence as a result of the above proof, the modulus of the function
(77) will be bounded. The same can evidently bo said about the function

/ (z) = cot z ----—, (80)


z
for z_1 tends to zero as z oo. I t can readily be seen th at the function (80)
no longer has a pole at the origin z = 0 and we can therefore apply the expansion
(75) to this function. We then calculate the infinite parts of the function (77) a t
its poles z = kn. Each of its poles will be a simple zero of sin z and the resi­
due at this pole will be calculated by the usual formula
cos z
= 1.
— (sin z)' z=kn
Hence the infinite part of the function (77) at the pole z — kn will be
1 (k = 0, ± 1, ± 2, ...).
z — kn
In particular, at the pole z = 0 the infinite part will be z~ 1 and therefore
the function (80) will no longer have a pole at z = 0. With regard to the other
poles z = kn, the infinite part of the function (80) will be the same as for the
function (77). If we are to apply formula (76) /(0) remains to be calculated.
Function (80), which is an odd function, can be expanded as follows near
z = 0:
/ (z) = y,z + y a z3 + ...,
from which it follows directly th at /(0) = 0. Formula (76) finally gives

cot z (81)
)•
where the accent above the symbol of the sum shows th at the term correspond­
ing to k = 0 is excluded.
I t can readily be proved th at the series on the right-hand side will converge
absolutely and uniformly in any bounded part of the plane providing the first
few terms with poles in this part of the plane are rejected. In fact, the general
term of the series will be
z
(z — kn) kn '
In any bounded part of the plane we have | z | < M; if we assume that the
absolute value of A: is sufficiently large we can write:
z _1________ M
(z — kn) kn k2 n (n — M k ~ l)
254 TH E APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [6 6

The coefficient of 1jk1 tends to a finite limit M jn2 as k increases indefinitely,


and the series

as we know, is convergent. Consequently [I, 147] the series (81) converges


absolutely and regularly in any bounded part of the plane.
If in formula (81) we substitute nz for z, then

7i cot nz = ---- f- ( 81 . )
z

Grouping together in pairs terms which refer to values of k, which have


opposite signs but the moduli of which are equal, we can rewrite this for­
mula as follows:
CO O -

Jtcot^z y __
z & *- w
In exactly the same way the following formula can be proved

—”n 1 + +«
= _L y ' (-
sin nz z

Differentiating the uniformly convergent series (81) we also have the formula
1 +»
sin1nz = ^r+
Z
S '
A=-»
_ L _= y . 1
(z-W k±La ( z - k y

We recall that the above formulae were deduced by a different method in


the theory of trigonometric series [II, 145].

66. The construction of meromorphic functions. We shall now deal with the
construction of a meromorphic function when its poles a* and infinite parts
a t these poles are given by

(82)
If only a finite number of poles ak (it = 1, 2, . . . , n) are given, then the function

will give the obvious solution of the problem, where the above function is a
rational fraction. Let us now suppose that we are given an infinite number of
poles ak and a corresponding number of infinite parts. We saw in [64] th at in
every bounded part of the plane there must only be a limited number of poles
which can be numbered in the order of non-decreasing moduli, i.e.
Iail < I“2I < (I «/I I-►+ “ )•
66] th I b c o n s t r u c t i o n o p m e r o m o r p h i o f u n c t i o n s 255

There are no further restrictions as to the position of the poles or the given
infinite parts. We only suppose th at there is no pole at z = 0 among them.
Every infinite part (82) represents a function which is regular in the circle
I z I < I ak l> i11 which it can be expanded into a McLaurin’s series:

9k = a{k) + a[k) z + a[k) z* + • • • (I 2 I < I «* I) • (83)

Take any sequence of positive numbers ek which form a convergent series:

2 «*■ (84)
*=l
As a result of the uniform convergence of the series (83) in the circle [13]

121< y I I
we can take a segment of this series
qk (z) = a(k) + a[k) z + a[k) z2 + . . . + a%l zm k,
such that
< Bk in the circle | z (85)
<yi
Construct the series

9 > w = i [ ^ ( F — ) - ? * (z)] (86>


and consider any circle CR, centre the origin and radius R. Since | ak | -► + oo
there exists an N such that R < (1/2) | ak | when k > N , and for these values of k
the inequality (85) holds in the circle CR; consequently, as a result of the con­
vergence of the series (84), series (86) is absolutely and uniformly convergent
in CR, providing its first N terms are rejected. These first N terms give poles
ak in the circle CR with infinite parts (82). An absolutely and uniformly convergent
series will give a regular function in CR. Owing to the fact th at the radius R
is arbitrary we see that the sum (86) solves the problem of constructing a
meromorphic function from its given poles and infinite parts. Notice th at the
polynomials qk(z) do not add new characteristics.
If the pole z = 0 is also given and its infinite part is

9° ( t ) ’
then it is sufficient to add this infinite part to the series (86). This solution
of the problem is due to the Swedish mathematician Mittag-LSffler.
In [64] we gave a formula for the expansion of a meromorphic function into
partial fractions, when certain additional assumptions were made. We shall now
give an analogous formula for the general case.
Let /(z) be a certain meromorphic function. By using the method given
above we shall construct the meromorphic function <p(z) which has the same poles
and infinite parts as /(z). This meromorphic function <p(z) will be given by a
formula similar to (86). The difference/(z) — <p(z) must be a regular function
256 TH E APPLICATION OP T H E TH EORY OP RESIDUES, FRACTIONAL FUNCTIONS [67

in the whole plane (except a t z = oo). Such a function is called an integral


function. I t is expressed in the whole plane by a McLaurin series. Putting
f(z) — <p(2 ) = F (2 ),
we obtain the following formula for the meromorphic function:

<’>+«• (t ) + j , W r = s r ) - »* <’>] • <87>


where F(z) is a certain integral function. This last formula is of great interest in
theory though it is more convenient to use the formulae (75) and (76) in particular
cases. If F(z) is any integral function then (87) gives the general formula for
all meromorphic functions for which the poles and infinite parts are given.

67. Integral functions. As we said above an integral function is a


function which is regular in the whole plane. It is expressed in the plane
by a McLaurin’s series. If this series is disrupted then the function is
simply a polynomial. Otherwise the point at infinity will be an essential
singularity of our function and in this case the function is sometimes
called an integral transcendent function. The functions e2 and sin z
are examples of integral transcendent functions. In future we shall
simply use the term “integral function” .
We know that every polynomial has zeros. An integral function
must not necessarily have this property. For example e2 has no zeros
at all. We will now construct a general expression for integral
functions without zeros. Let g(z) be a certain integral function. In this
case the function
/ (z) = e*<2> (88)
is an integral function without zeros. Let us show, conversely, that
every integral function f{z) without zeros is of the form (88), where
g(z) is a certain integral function, i.e. formula (88), where g{z) is any
arbitrary integral function, gives the general form of integral functions
f(z) without zeros.
When the integral function f(z) has no zeros then the function
V («)
/(*)
will also be an integral function. When integrating this integral
function we also obtain an integral function

0(3) = \ j $ - dz = 1° g f( z)>
from which (88) follows directly.
67] INTEGRA!, FUNCTIONS 257

Suppose now that the integral function f(z) has a finite number
of zeros, other than 2 = 0:
2 = Oj, , dm,

where multiple zeros are counted as many times as there are units
in their number. The relationship

k= 1 v
m
where the symbol _// denotes a product, which embraces all integral
k=l
values of k from 1 to m, is an integral function without zeros, i.e. it
has the form (88). We therefore have the following expression for
our function f(z)
/(*) = e > » 7 7 ( 1 ~ £ ) , (89)

where g(z) is a certain integral function.


Let us suppose that the point z = 0 is not a zero of /(z). If this point
is a zero of order p, then in place of formula (89) we must have

f(z) = e*V>zP (90)

In the most interesting case when f(z) has an infinite number of


zeros, we can no longer apply formula (90) directly, for on the right-hand
side we have an infinite product, which may not have any meaning.
To make this product convergent we have to add to the factors
(1 — zjdit) additional exponential factors which do not introduce new
zeros but make the infinite product convergent.

Let us investigate this for sin z. Rewrite the formula (81):

cot 2 ------
1
2 Icn
In this case both sides will be regular a t the point 2 = 0 and we can integralc
the infinite series term by term, from z = 0 to the variable point 2 . As a result
of integration we have
s i n 2 |2=* 2= 2
lo g i
log <2 - kn) + 2= 0

or, taking the principal value of the logarithm in the neighbourhood of the origin
, s in 2
lo g —
JIN
258 TH E APPLICATION OP TH E TH EO RY OP RESIDUES, FRACTIONAL FUNCTIONS [68

Hence, discarding the logarithms, we obtain an expression for sin z in the


form of an infinite product:

sin z = z (91)

where the accent above the symbol of the product shows th at the factor cor­
responding to k = 0, has to be excluded. In this case factors of the exponential
type ezlkn guarantee the convergence of the infinite product.
Grouping together in pairs factors corresponding to values of k the moduli
of which are equal, we have:

a in z = z U (92)
Substituting jiz for z we can rewrite the formula in the following form
sin nz
(93)
n

To give a more detailed explanation of the expansion of an integral function


into an infinite product we have to explain certain basio facts about infinite
products.

68. Infinite products. Consider the infinite product


00

ck = c1 c2 • • ■' (94)
fc=l
where ck are certain complex numbers, other than zero. The con­
vergence of the product (94) is analogous with the convergence of a
series. Let us construct the finite products:
n
Pn = ^ ck — ci c2 • • ■cn• (95)
k= 1

If, as n increases indefinitely, the product P„ tends to a finite


limit P, which is not zero, then the infinite product (94) is said to
be convergent and its limit is P.
If there are zeros among the numbers ck, then the infinite product
(94) is said to be convergent if, after the exclusion of the zeros, the pro­
duct remains infinite and convergent in the above sense. In this case
the value of the infinite product with zeros is taken to be equal
to zero. The above remark, that the limit P of the product P„ should
not be zero was made so that infinite convergent products should
have the usual properties of finite products, viz. they should be
equal to zero only when at least one of the factors is zero.
68] INFINITE PRODUCTS 259

Let us suppose that all of the terms of the product (94) are other
than zero and construct the infinite series:

J ^ lo g ck, (96)
fc=i
where the value of the logarithm in every term is determined in
a certain way. The sum of the first n terms of the series (96) will be

S „ = ^ I o g ck. (97)
fc=i
Suppose that for certain values of the logarithms the series (96)
is convergent, i.e. a limit 8n —>S exists. Formula (95) gives P„ = es",
and consequently, there is a limit Pn -> es, other than zero, i.e. it
follows from the convergence of the series (96) that the product (94)
will also be convergent. Conversely, let us now suppose that the
infinite product (94) is convergent, i.e. there is a limit Pn -*■ P, other
than zero. We determine the values of the logarithms in the series (96)
in such a way that the right side of formula (97) always contains
the principal value of the logarithm of the product c1c2 . . . cn:
S n = log | Clc2 • • • c„ | + i arg (ct c2 . . . c„ ),
where
— n < arg (cxc2 .. . c„) < n.
In thus case Sn will also tend to a definite limit, viz.:
lim Sn = log | P | + i arg P = log P ,
and, consequently, the series (96) will be convergent.
We are assuming all the time that P is not a real negative number,
for arg P lies in the interval (—n, + n). When P is a real negative
number we should select the amplitudes in such a way that arg (cjC2. . .
c„) would be confined to the interval (0,2ji). The proof would be
the same as above.
We thus arrive at the following general proposition: if all the
numbers c* are other than zero, then for the infinite product (94) to he
convergent it is necessary and sufficient that the series (96) should be
convergent when the values of the logarithms are determined in a certain
way. This infinite product will be equal to
P = es . (98)
The general term of the series (96) will be:
log ck = log | ck | + i arg ck
258 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [68

Hence, discarding the logarithms, we obtain an expression for sin z in the


form of an infinite product:

sin z = z l l ^ 1 - ekn , (91)

where the accent above the symbol of the product shows that the factor cor­
responding to k = 0, has to be excluded. In this case factors of the exponential
type ezlkn guarantee the convergence of the infinite product.
Grouping together in pairs factors corresponding to values of k the moduli
of which are equal, we have:

sin z = z (92)
V ti*)
Substituting jiz for z we can rewrite the formula in the following form
sin nz
(93)
71

To give a more detailed explanation of the expansion of an integral function


into an infinite product we have to explain certain basio facts about infinite
products.

68. Infinite products. Consider the infinite product

J ~ £ ck = cl c2 • • • > (9 4 )
fc=l
where ck are certain complex numbers, other than zero. The con­
vergence of the product (94) is analogous with the convergence of a
series. Let us construct the finite products:
n
P n = £ £ c k — C1 C2 • • • c n- (9 5 )
*=1
If, as n increases indefinitely, the product Pn tends to a finite
limit P, which is not zero, then the infinite product (94) is said to
be convergent and its limit is P.
If there are zeros among the numbers ck, then the infinite product
(94) is said to be convergent if, after the exclusion of the zeros, the pro­
duct remains infinite and convergent in the above sense. In this case
the value of the infinite product with zeros is taken to be equal
to zero. The above remark, that the limit P of the product Pn should
not be zero was made so that infinite convergent products should
have the usual properties of finite products, viz. they should be
equal to zero only when at least one of the factors is zero.
68] INFINITE PRODUCTS 259

Let us suppose that all of the terms of the product (94) are other
than zero and construct the infinite series:

^ l o g c A, (96)
fc=i
where the value of the logarithm in every term is determined in
a certain way. The sum of the first n terms of the series (96) will be

S„ = J l ° g C*. (97)
fc=i
Suppose that for certain values of the logarithms the series (96)
is convergent, i.e. a limit 8n ^>- S exists. Formula (95) gives Pn = es",
and consequently, there is a limit Pn —*■es, other than zero, i.e. it
follows from the convergence of the series (96) that the product (94)
will also be convergent. Conversely, let us now suppose that the
infinite product (94) is convergent, i.e. there is a limit Pn — P, other
than zero. We determine the values of the logarithms in the series (96)
in such a way that the right side of formula (97) always contains
the principal value of the logarithm of the product CjC2 . . . cn:
Sn = log | cxc2 • • • cn | + iarg (ct c2 . . . c„),
where
— n < arg ( c ^ . . . cn) < n.
In thus case Sn will also tend to a definite limit, viz.:
lim Sn = log | P | + i arg P = log P,
and, consequently, the series (96) will be convergent.
We are assuming all the time that P is not a real negative number,
for arg P lies in the interval (—jt, +7i). When P is a real negative
number we should select the amplitudes in such a way that arg ( c ^ .. .
cn) would be confined to the interval (0,2tt). The proof would be
the same as above.
We thus arrive at the following general proposition: if all the
numbers ck are other than zero, then for the infinite product (94) to he
convergent it is necessary and sufficient that the series (96) should he
convergent when the values of the logarithms are determined in a certain
way. This infinite product will he equal to
P = es . (98)
The general term of the series (96) will be:
260 THB APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [6 8

and, bearing in mind that the general term of a convergent series


tends to zero, we should, in any case, have arg ck -> 0, i.e. the series
(96) can only be convergent providing that, by starting from a certain
place, we take the principal values of logarithms. The determination
of the values of logarithms for a finite number of the first few terms
can, naturally, have no effect on the convergence of the series and
will only add 2 m m to the sum of the series, where m is a certain
integer. This additional term will not alter the value S and, accord­
ing to the formula (98), it will not affect P . Therefore the determina­
tion of the values of logarithms in the series (96) is only significant
in so far that b y s t a r t i n g f r o m a c e r t a i n p l a c e w h i c h c a n be c h o s e n
a r b i t r a r i l y , o n l y th e p r i n c i p a l v a lu e s o f l o g a r i th m s s h o u ld be ta k en .
Let us now consider the infinite product, the terms of which are
integral functions of z:

F ( Z) = 77 “ A W
fc=l
= W ■u 2 ( 2 ) • • • (9 9 )

We take a circle C R in the z-plane, centre the origin and radius R .


Suppose that for any R the terms u k( z ) , starting with a certain value
of k , no longer have zeros inside the circle C R . Let, for example,
this start with k = k 0 for the given radius R (this number will,
generally speaking, depend on R ) . Consider the infinite series

s (2) = 2fc=l log (z) . (100)


which can be rewritten as follows:

^ log u k (z) + J ? log u k (z). (101)


fc = 1 k=k.

The terms of this latter sum are regular and single-valued functions
in the circle C R , for u k( z) does not vanish in this circle. Assume that
for a certain set of values of these regular functions log u k( z) the
latter series will uniformly converge in the circle C R . Denoting its
sum by f kt( z ) , where /^(z) is a certain regular function [12] we have:

/ 7 uk (z) = eW2>, f j u k (z),


fc=i fc=i

i.e. in this case (99) will be a regular function in the circle C R and
its zeros inside the circle will be determined by the zeros of the terms
69] THE CONSTRUCTION OP AN INTEGRAL FUNCTION FROM ITS GIVEN ZEROS 261

Uk(z), when k < k 0. Since R is chosen arbitrarily, we can, in general


say that if the series (100) converges uniformly in any confined part
of the plane (when the first few terms are excluded which is unimportant)
the infinite product (99) will be convergent in the whole plane; it is an
integral function and its zeros are fully determined by the factors uk(z).
Differentiating the uniformly convergent series (100) we obtain:

S ’ (z) = V% n'k (2)


*=i uk(z) ’
but
F (z) = es « and F ' (z) = S ’ (z) F (a),
i.e.
F '(z) = F(z) ( 102 )
fcTi M* &
This formula shows that when the series (100) is uniformly con­
vergent, the infinite product (99) can be differentiated as shown
by (102), which is analogous to the differentiation of a finite product.

69. The construction of an integral function from its given zeros.


Using the above considerations we can construct an integral function
from its given zeros. Notice, first of all, that the zeros of an integral
function cannot have limit-points at a finite distance. If such a point
z — c were to exist, i.e. if any small circle with the centre at z = c
contained an infinite number of zeros of an integral function, then
it would be identically zero [18]. Repeating the same arguments
as in [64] we can see that, in any case, the zeros ak of the integral
function can be grouped in the order of their non-decreasing moduli:
I^1 I ^ I^2 I ^ • • • >
where | an | —;>- -f-°° as n —►°°. Notice that if a certain number a
appears q times among the numbers ak, this shows that the corres­
ponding zero a is a zero of order q. We are also supposing, for the
moment, that z = 0 does not appear among the given numbers ak.
We will only consider one particular case which is of great practical
importance, viz. when ak moves towards infinity so quickly that a
positive number m exists, such that the series

2
(103)
k=1
is convergent. We assume that m > 2.
262 th e application of t h e theory of residues , fractional functions [69

Let us construct the infinite product:

F{z) = / j ( l + -+ (104)

and show that it will satisfy all the conditions mentioned in the previous
section. Consider a certain circle C R such that starting with a
certain symbol k = k 0, the numbers a k will lie outside the circle C R ,
Then when k > k 0, the terms of the product (104) will have no zeros
in the circle C R and for any z in C R we will have:
z
<&<l ( k > k 0) , (105)
ak
where ■& is a definite positive number smaller than unity. Consider
the series (100) in this case:
£
“ak)I e°k +5(3 (106)

From (105) we can use the expansion of a logarithm into a power


series and for this determination of the logarithm we obtain the fol­
lowing formula for the series (106):

= L _ f ^ r +,_ 1
k~k, L m ' m+1 {ak ) J

Consider the general term of this series


, , \ (Z 1 ( z Vn+I
vk ( z) m \ak) m 1 (, ak J
We obviously have:
m+1
\ Vk(z)\ < — + m -f- 1 + ...

or, from (105) taking ( i j m ) \ z j a k \m outside the brackets and bearing


in mind that in the circle C R \ z | < R :
Rm
IVk {z) I (1 +& + &2+ . . . ) ,
i.e.
Rn
Iv k ( z ) I <
m ( ] - 0 ) | o&I171
As a result of the convergence of the series (103), the positive num­
bers on the right-hand side of the above inequality form a convergent
69] THE CONSTRUCTION OP AN INTEGRAL FUNCTION FROM ITS GIVEN ZEROS 263

series and, consequently, series (106) will be absolutely and uniformly


convergent in the circle CR. We can therefore say that the infinite
product (104) is an integral function and that its zeros are determined
from the zeros of the factor (1 —z/afc), i.e. the numbers ak are the
zeros of this integral function.
If we have any arbitrary function /(z) the zeros of which are ak,
then the quotient f(z) : F(z) will be an integral function without
zeros, i.e. this quotient will have the form e®^, and we will have the
following formula for the integral function f(z):

f(z) = e ' V j j e»*+ 2*0*1 +'" + m -iU ) (107)


fc=i -i)
where g(z) is a certain integral function. We have so far that the point
z = 0 is not a zero of the function. If this point is in fact a zero of
order p, then we should add the factor z** to the right-hand sides
of the formulae (104) and (107).

Let us consider, for example, the function sin z. I t has a simple zero
2 = 0 and zeros simple z = kn (h = ± 1 , ± 2 , . . . ).
In this case we have m = 2, so th at the series
1
2' | A:*I*
k = -

as we have already said above, is convergent. Using formula (107) and adding
the factor z we have
sin z = e1 (2) z / / ' (l ~ e*" ‘

The integral function g(z) cannot be determined from the above considera­
tions. The results [67] show th at in this case the function is identically zero.
Note th at when m = 1, i.e. the series below is convergent
CD
i
2
*=i Kl
we can by using the same arguments as before, write instead of formula (107):

We shall give more examples of the application of formula (104) later; it


is usually known as the infinite product of Weierstrass.
I t may happen that the numbers ak are such th at the series (103) will be
divergent for any positive m. This will happen if we suppose, for example,
264 TH E APPLICATION OP TH E TH E O B T OP RESIDUES, FRACTIONAL FUNCTIONS [7 0

th at a* = log (k + 1) (k = 1, 2, . . . ) . In fact a series with the general term


[log (k + l)]~m will be divergent for any positive m, so th at the sum of its
first few terms is greater than
k
[log(fc + I)]m ’

and this latter expression can easily be shown to grow indefinitely with k by
applying, for instance, l’Hfipital’s rule [I, 66]. I f the series (103) is divergent
for any positive m we can construct the infinite product:
z
) , (108)
~“k
where
zk
Qk (z) = ----
ak h 2o> + • • • + mkak k
and mk depends on k. Repeating the above inequalities we can see th at the
infinite produot (108) will be convergent providing th at the series given below
is convergent for any R > 0
» ( R 'vmt+l

For this purpose it is sufficient to take mk = k — 1. In fact, applying Cauchy’s


analysis [I, 121] to the series

we obtain

0,

i.e. the series does, in fact, converge. I t can be shown th at the series will be
convergent providing mk is so chosen that the inequality mk + 1 > log k
holds.

70. Integrals which depend on parameters. In future we shall have


to determine functions in the form of integrals which depend on
parameters. We have already met these functions in [61]. We have
also considered this problem for real variables and determined condi­
tions for such a function to have a derivative and when differentiation
under the integral is possible [II, 84],
We shall now consider an analogous process for complex variables.
T h eo rem . Let f(t, z) be a continuous function of two variables t and z,
where z belongs to the closed domain B with contour I and t to a finite
interval a < t < b of the real axis. Also f(t, z) is a regular function of
70] INTEGRALS WHICH DEPEND ON PARAMETERS 265

z in the closed domain B for any t in the above interval. In this case the
function oo(z) given by the inequality:
6
co (z) = J f (t, z) d f, (109)
a

is a regular function in B ; when evaluating its derivative we can


differentiate under the integral, i.e.

» ' (*) = (109j)

According to Cauchy’s formula we can write:


1 \ f(t,z')
/(*. 2) 2ni J z' —z dz',

where z lies inside B and t is any point in the interval a < t < 6.
Consequently:

a I
When integrating a continuous function we can change the order
of integration [II, 78 and 97]:
b
J / (<, z') dt
= z'-z
i
This formula gives co(z) in the form of Cauchy’s integral and, con­
sequently, co(z) is a regular function in B; its derivative is determined
by the formula [8]:
b
J/(t,Z')d«
m W = J (s' - z)1 dz •
I
Changing the order of integration we can write:

a I
According to Cauchy’s formula, the expression in the square
brackets gives the derivative 8f(t, z)/dz, the above formula is the same
as formula (109!) and the theorem is thus proved. Notice that we
could have assumed that t varies not within a finite interval (o, b)
266 TH E APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [7 0

of the real axis but along any finite curve. This would not have altered
the proof of the theorem. Notice also that in the above proof the
integral
b
J f (t , z’)dt,
a

which appears in the numerator of Cauchy’s integral for co(z), is a


regular function of z on I. This is directly due to the fact that f(t, z)
is given as a continuous function of its two arguments [II, 80].
Let us now consider indefinite integrals. To prove the theorem for
this case it is sufficient to add the condition that the integral (109)
must be convergent. To be more specific, we will consider the integral
in the infinite interval (a, + °°), but the proof is equally valid for
other kinds of indefinite integrals.
T h e o re m . Let f(t, z) be a continuous function of two variables, where
z belongs to the closed domain B and <> a. Also f(t, z) is a regular function
in the closed domain B for any t > a and the integral

J f{t,z)dt
a

is uniformly convergent with domain to z in the closed domain B. Then


eo
co(z) = § f(t, z)dt (110)
a

is a regular function of z in B and

w '{z)=\*t^-dt.
a

Construct the sequence of functions

" n(z) = ja7 (*.z)


where an is any sequence of numbers greater than o, which tends to
(-(-<»), It follows from the theorem we have just proved that con(z)
is a regular function in B and

a>'n { z ) = ] ^ ^ - d t .
a

It follows from the condition of uniform convergence of the integral


(110) that oin(z) tends uniformly to the function co(z) as given by
71] EULER'S INTEGRAL OF THE SECOND CLASS 267

the formula (110) and, according to Weierstrass’s theorem, this


function co(z) is a regular function in B; also u>'n(z) -*■ (o'(z), i.e.

lim f ^ f.- z- ■d t = co' (z )


n—“ aJ
when an tends to ( + °°) in any manner. It follows from this that

id‘'
a
where the integral on the right-hand side must have a meaning. The
theorem is thus fully proved.
In this proof of the theorem we could have assumed that integration
with respect to t takes place along a certain infinite contour G. Such
an indefinite integral must be taken as the limit of integrals along
finite contours which form part of the contour C. This theorem
applies, word for word, to indefinite integrals in which the integrand
f(t, z) becomes infinite, for example, when t approaches a.
Notice, finally, that the following rule which gives the sufficient
condition for the integral to be absolutely and uniformly convergent
applies [II, 84]: if we integrate with respect to t along the real axis
and if, when £> a and z belongs to the closed domain B, the inequ­
ality | /(£, z) | < (f(t) is valid, when the integral

J 9 {t)
a

is convergent, then the integral (110) will converge absolutely and


uniformly. Absolute convergence is determined in exactly the same
way as for the real f(t, z).

71. Euler’s integral of the second class. Consider the function given
by Euler’s integral of the second class:

r{z) = $ e~‘t*-1d t , (111)


o
where tz~l = e(z-1)Iog<, and the real value of the logarithm of the
positive number t is taken. We write the above integral as the sum
of two integrals:
1 co
r (z) = f e-( tz_1 dt + J e~‘ t2-1 d£. ( 112 )
o i
268 TH E APPLICATION OP TH E TH EO RY OP RESIDUES, FRACTIONAL FUNCTIONS [71

Consider, first of all, the second term on the right-hand side:


00
co (z) = j e- '^ -1 d<. (113)
1
When t > 1 the integrand:
e - ' f - 1 = e-,+(z— l0B' (114)
is a continuous function of t and z for any z and t > 1, and it is an
integral function of z for any t > 1. Suppose that z belongs to a
certain confined domain B of the z plane and that z = x + iy. In the
closed domain B the abscissa has its greatest value which we denote
by x 0. Bearing in mind that log t > 0 when t > 1 and that the modulus
of the exponential function 2<pt with a purely imaginary index is
unity, we obtain for z in B:
| e~lfz_11= | e-,+(x_Iog t+‘y log' | < e~'+<x»_1),0B 1— e—1' .
The integral
CD

J e—ttxo-1 di
i
is convergent [II, 82], and consequently the integral (113) is
uniformly convergent with respect to z in B. Bearing in mind the
second theorem of the previous paragraph and our complete freedom
in choosing B we can say that oo(z), as given by formula (113), is an
integral function which can be differentiated under the integral.
Consider now the first term of the formula (112):
i
<p(2) = J e_ '<z_1 d<. (116)
0

In this case the continuity of the integrand (114) may be disrupted


when t = 0, for log t, when t = 0, becomes (—°°). As before, the
modulus of the function (114) will be:
e~ nx- 1.
If x > 1 when t = 0, then the continuity of the integrand will not
be disrupted and, applying the first theorem of the preceding section,
we can see that the function (115) will be regular when x > 1, i.e.
when it lies to the right of the line x = 1. We shall now prove that
it will also be regular when it lies to the right of the imaginary axis.
In fact, take any finite domain B to the right of the imaginary axis.
Let be the smallest abscissa of points of the closed domain B .
71] EULER'S INTEGRAL OF THE SECOND CLASS 269
Since the closed domain lies to the right of the imaginary axis
> 0. bearing in mind that log t < 0, when t < 1, we obtain:
| e~‘ <z_11< e~' tXl~ 1,
if 2 lies in B. But when xx > 0 the integral
i
J e -'P * -1d t
o
is convergent and, consequently, as before, the function (115) is
regular to the right of the imaginary axis and can be differentiated
under the integral. It follows from all that was said above that
formula (111) determines a regular function r(z) to the right of the
imaginary axis.
We shall now analytically continue the function to the left of the
imaginary axis and show that r(z) is a meromorphic function with
simple poles at the points
2 = 0 , - 1 , - 2 , ... (116)
Owing to the fact that the second term on the right-hand side of the
formula (112) is an integral function, we have to deal with the
function (115).
Along the finite interval 0 < t < 1 the function e~f can be expanded
into a uniformly convergent series:
as fn
a—f
n—0
where, as always, we take 0 ! = 1. Multiplying by f ' and integrating
by parts along the interval (0, 1) we obtain:
'[ tn+z "|f-l
J V '! ld< = ^ - n\ Ln + zJt=o’
n=0

We suppose that z lies to the right of the imaginary axis and,


consequently, the real axis (n + z) is positive and tn+z = 0 when
t = 0, i.e.
l
^e- ' tz~ 1d< =
o
We thus obtain the following expression for r(z) to the right of the
imaginary axis:
Dn 1 (117)
n=0 nl z + n + J e~* <Z_I d t.
270 THE APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [7 1

Owing to the presence of n ! in the denominator, the infinite sum


on the right-hand side will be absolutely and uniformly convergent in
any bounded part of the plane, providing the first few terms, with poles
at the points (116) are rejected. This sum will, consequently, give a
meromorphie function with simple poles (116) where the residue at
the pole z = —n is (—l)"/wl. The second term on the right-hand side is,
as we have said already, an integral function. The right-hand side of
formula (117) thus gives the analytic continuation of the function r(z)
in the whole plane of the complex variable z, which formula (111) only
defines to the right of the imaginary axis; F{z) appears to be a mero-
morphic function with simple poles (116) and with a residue ( —l)"/rel
at the pole z = —n. The values of r(z), when the argument is a
positive integer, can readily be obtained. Assume that z = n 1,
where n is a positive integer. We obtain in this case [II, 81]:
QO
r ( n + 1) = J e~‘ tn d£ = n\
o
and
r { 1) = j V 'd * = 1.
o
Hence values of r(z), when z is a positive integer, give the factorials
of integers:
J ’(l) = 1; r(n+ 1) =7il (n=l,2,3,...). (118)

We shall now derive the fundamental properties of the function


r(z). Assuming that z > 0 and integrating by parts we have:
a? «
r ( z + 1) = J e~f tz d< - [— e~' tz~\ “ + z j" e- ' tz~x d<,
o r=0 o
i.e.
T ( z + 1) = zr(z). (119)

We have proved this equality only for the right side of the real axis.
However, if two analytic functions coincide on a certain line, then
they coincide everywhere [18] and, consequently, we can take it
that the formula (119) is established for all values of z. Let n be a
certain positive integer. Applying formula (119) several times we
obtain a more general formula which is valid for all complex z’s:

r ( z + n) = [z + n —- 1) (z + n — 2). . .(z + 1) zP(z). (120)


71] EULER'S INTEGRAL OF THE SECOND CLASS 271

Let us now suppose that z lies within the section (0, 1) of the real
axis, and return to the fundamental formula (111); we replace the
variable of the integration t by u, where t = u2. We thus obtain
the following results:

r{z) = 2 f e - ^ 't t ^ d i t .
6
Substituting 1 — z for z we can write:
CD

r ( l - z ) = 2 J e - ’V - 22di>.
0
Hence, by multiplying these equations together, we obtain:
» CO

/ » r ( l - z ) = 4 j J e-(“!+' l) ( v ) 2 2 1 dMdw. (121 )


00
The integral on the right-hand side can be treated as a double
integral in the plane (u, v), where the domain of integration is the
first quadrant, i.e. that part of the plane where u > 0 and v > 0.
We now introduce polar coordinates in place of u and v:
u = gcosq>; v = gsinq>.
Formula (121 ) can be rewritten in the form:
n
00 2
r (z) r (1 — z) = 4 f f e~e*cot 22-1 (pQdp d<p,
6 6
where we integrate with respect to g from 0 to + °° and with respect
to 9>from 0 to nj2 , i.e.
n
2 BO
r (z) r (1 — z) = 4 f cot22-1 <pdip f e- ®2g dg.
6 6
It can easily be seen that
co
[ e -®2g dg = y ,
6
and consequently

r (z) r (1 — z) = 2 j cot22-1 <p dip.


272 TH E APPLICATION OP TH E TH EORY OP RESIDUES, FRACTIONAL FUNCTIONS [72
We now replace <p by a new variable, according to the formula

<p = arc cot dq> = dx


2f x( l + x)
The above result can, in this case, be rewritten as follows:

T(z)r(l - 2) =
+■
o
But, as we know from [62], the integral on the right-hand side is
equal to njainnz, and consequently we obtain the following formula:

r {xz ) r t i ~ z )' = S^ill —


712
. ( 122 )

We have proved this formula only for the section (0, 1) of the rea
axis. But, as before, using the process of analytic continuation,
we can show that it is valid for all z’a.
Formula (120) enables us to transform the evaluation of r(z) for
all real values of z to values of r(z) on the section (0, 1). Formula
(122) makes it possible to transfer the section (0, 1) to the section
(0, 1/2). Assuming z = l/2 in the formula (122) we obtain:

/ ’ (i.) = (’e - <r i c k = Ytz. (123)


6

72. Euler’s integral of the first class. Euler’s integral of the first
class is an integral of the following form:
i
B (p, q) = J xp_1 (1 — x)q~x dx. (124)
o
As in the case of the integral (111) we suppose that the real parts
of p and q are greater than zero and also that:
x?-1( l _ x y - 1 = e(P~ 1) l o s x + f a - 1) i ° g ( i - x ) ;

where real values of the logarithms are taken.


Replacing a; by a new variable t, according to the formula t = 1 — x
we obtain in place of (124):
l
B{p,q) = ^ ~ 1( l - t ) P - 1dt,
o
i.e.
B (p, q) = B (q, p) . (125)
72] EULER’S INTEGRAL OF THE FIRST CLASS 273

We will introduce yet another formula which explains the funda­


mental property of the function B(p, q). Integrating by parts we can
write:

J X?-1(1 - x)*dx = [ lP(1~ x)*] ^ + -J J X>(1 - X)*-' dx.


o 0
As a result of the assumptions we made with regard to p and q we
can say that the term outside the integral is zero and that the above
formula gives us the following property of the function B(p, q):

B(p,q+l) = fB ( p + l ,q ) . (126)
We will now establish the connection between the function B(p, q)
and the function (111). Applying the same transformation as in the
previous section we can represent the product r(p) T(q) in the
form:
= 4 ]■ J e-<“a+,’s>w2' - 1f t * - 1 du d v
r ( p) r (q)
oo
and introducing polar coordinates we obtain
n
OB 2
r (p) r (q) = 4 J e- ®*£2(p+9>-1 dg J cos2p—1p sin2?- 1p d p .
o o
Replacing g by a new variable t according to the formula g = yft
we can write

J e- ®2g2(P+«)-1dg = — e~' tp+<i- 1dt = ^ T ( p + q ),


o 6
and consequently:
2
r ( p) r (q) = 2r (p + 3) f cos2p_1 p sin29-1 p dp.
6
If p is now replaced by a new variable of integration x where
x = cos2 p, then the above relationship gives
1
r (P)r (q) = r iP + 3) I xp~1{1 — x ^ - ' d x ,
0

and this gives a formula where B(p, q) is expressed as a function


of r(z):
P(p)r(q) _
B (p, q) (127)
r ( p + q)
274 TH E APPLICATION OP TH E THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [7 3

73. The infinite product of the function [/'(z)]- *. Let us return


to the fundamental definition of the function r(z) as given by the
formula (111) and assume, for the sake of simplicity, that z > 0.
The factor e~l is, as we already know, the limit as n~> oo of the
expression [I, 38]

rt-+oo N '
Replacing the interval (0, + °°) by the finite section (0, n) we obtain
the integral:
n

P n (z) = f ( i - ± y r - i d t . (128)
0
It is to be expected that as n increases indefinitely the integral
will tend to an integral which appears on the right-hand side of
formula (111). At the end of this section we shall prove this theorem
in detail but for the moment we shall use its results.
Replacing i by a new variable r where t = nr. we can rewrite
(128) in the form:
i

P n (z) = nz J (1 - x)n t z—1d r . (129)


o
We now suppose that n tends to + ° ° by taking positive integral
values. Integrating by parts we obtain:
l i

J (1 — r)n rz—1 dr = [-1 t z (1 - T)n^ “‘ + — [(1 — r y - ^ d r ,


o o
or, bearing in mind that the term outside the integral vanishes
(z 0 > ):
l i l

J (1 — r)n t z 1 dr = —J (1 — t ) " - 1 t z dr = - J (1 — r ) n ~ 1 drz+1.


o o o
Similarly, continuing to integrate by parts we obtain
i i

J (1 - r)nr 2- 1dr =
J(1 - t ) " - 2 t z + 1 d r ,
0 ’0
and, in general, we obtain the following expression for the integral
(129)
l

f ( l — r)n t z_1 dr = — nz .
tiz
o
Jv ’ z(z+■L'1)2... (z

+ n)
73] THE INFINITE PRODUCT OF THE FUNCTION [!»]-> 276
As n increases indefinitely the limit of this expression will be
r(z), i.e.
+ (13°)
or
— F = lim Z{Z\ \ ; ; _ (Z
n + n) n-* (ra-* = e-*'°en). (131)
** n~*oe
To alter the above expression slightly we multiply and divide it
bye2(1+1/a+-- +1/n). Having done this we can rewrite formula (131)
as follows:
,(1+ £+ i +-"+ H-logn) z „ z + l ^
r(z) = lim
2+ 2 ' + 1+1+■+ h)1
X

( ‘ + ! + !+•••+ - - loen)z " I M -jH


5V 3 ' zll\ + k Je M. (132)
T5T = k=i J
As the integer n increases indefinitely the above finite product
becomes an infinite product

n { i + t ) = - ‘- w
fc=i v J
The above infinite product is constructed in exactly the same way
as Weierstrass’s infinite product [69] with in this case ak = —k and
the series

converging when m = 2. Hence on the right side of (132) the last


factor tends to a definite limit (133). We will now show that the
variable
« „ = l + y + T + ... + - - l o g * (134)
also tends to a definite limit. To prove this it is sufficient to show
that the variable
0 . - 1 + 4 + -I-+ --.+ — iy-log» = » „ - i (135)
has a definite limit. The variable un will obviously have the samt
limit. Consider the branch of the hyperbola y = 1jx in the firse
quadrant. The number 1jk will be the ordinate of this branch when
276 TH E APPLICATION OF TH E TH EO RY OF RESIDUES, FRACTIONAL FUNCTIONS [7 3

x = k . Evidently log n is equal to the surface area bounded by our


hyperbola, the O X axis and the ordinates x = 1 and x = n and
the sum
l
1 + Y + T + ••• + n —1
represents the sum of the surface areas of outgoing rectangles with
unit bases on the X axis (Fig. 64). It follows directly that the difference
(135) increases together with n . On the other hand this difference
must be less than the difference of the surface areas of the
outgoing and incoming rectangles
y and this latter difference is equal
to (1 — 1/ji). Hence v n will be
an increasing bounded variable
and, consequently, it has a limit.
This limit C is usually known
as Euler’s constant. It is equal,
with an accuracy to the seventh
decimal place, to
G = 0.5772157... (136)
Formula (132) finally gives us
the limit

On the right-hand side of the above formula is an integral function


of z with simple zeros 2 = 0, —1, —2, . . . We established formula
(137) only for the positive part of the real axis. As a result of the
fundamental principle of analytic continuation we can say that it
is valid for all values of z and we thus see that t h e f u n c t i o n T ( z )
i s a n i n t e g r a l f u n c t i o n a n d f o r m u l a (137) r e p r e s e n t s i t i n t h e f o r m of
a n in fin ite produ ct.
We have thus proved that l / r ( z ) is an integral function and it
follows directly that the function T ( z ) does not vanish anywhere,
i.e. it has no zeros at all.
Using the infinite product (137) we can easily prove the formula
(122) in [71]. In fact, formula (137) gives us directly1

1
r(z)r(-z)
73] THE INFINITE PRODUCT OF THE FUNCTION f/■(?)]-1 277

or, from (93) in [67]:


1 z sin m
r{z)T(-z) - “ •
Also formula (119) gives us the following if in it we replace z by
(-z ):
r { - z ) = - r- £ ^ .
Substituting this expression for r ( —z) in the above formula we
obtain formula (122):
r ( 'z )' r {' i - z )' = sm
- r ^nz- .

We now have to show th at the integral (128) tends to the integral (111) as
n increases indefinitely; it is sufficient if we prove this when z > 0. Let us
first of all, find upper and lower bounds for the difference

I t can readily be shown th at the function

- ( - i f
is a primitive for the function

V n) n
and consequently:

0
If 0 < t < n, then the integrand is positive and therefore the same may be
said about the left-hand side. Replacing under the integral e® by e* and
(1 — v/n)n~ l by unity we obtain:

or
o (138)
Construct the difference:
n
r w - i>„ (z) . J [.-■ - (i - £)* ] . - 1d, + j > « - d,. (1Jt)
6 n
As n increases indefinitely the second integral on the right-hand side tends
to zero because the integral
J e , tz~1dl
o
278 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [73

converges. I t rem ain s to be show n th a t th e firs t integ ral will also te n d to zero
as »->• oo. F ix n = n 0 so th a t

J e ' tz 1 d« < y ,
no

where e is an y a rb itra rily chosen sm all whole num ber.


W e can w rite:

J0 [ ° " =
no n
-/[•" - ( ‘ - f ) “ ] ‘" l d ‘ + f [•" ' - ( ‘ d‘
0 no

an d from (138):
n no n
0<J ['" ~ ~ +JV v “
0 0 no

where in th e second in teg ral on th e rig h t-h an d side we h av e replaced th e differ­


ence b y a single m inuend. T he in te g ran d in th is integral is positive so th a t,
extending th e in te rv al o f in te g ratio n we ob tain :
n no »

o < / [ • ■ ' - ( ' - y ’] '■■■'“ < i . i > i d ‘


0 6 no

W hen n is large th e firs t te rm is less th a n e/2, so th a t for all sufficiently large


n ’s:

°< J t* - 1d * < « ,
6
i.e. owing to th e a rb itra rily sm all e in form ula (139) th e firs t te rm will also tend
to zero, i.e. in fact:
n
r (z) = ^ -i-j"
lim (l - tz~ l dt. (140)

Notice some other consequences of the formulae we have proved.


Taking the logarithmic derivative of both sides of formula (137) we
obtain:

G z + z 2 k{z + k )- (141)
k= 1 v 1 '
74] THE REPRESENTATION OP AO DT A CONTOUR INTEGRAL 279

Differentiating both sides:

dz! k g -H 2) ^ Q(Z + k)2' (142)

Using formula (130) we shall prove the so called doubling formula:


2 * * -ir(8 )r(z + i ) = KSi, (2z). (143)
Expressing the functions r(z) and F{z + 1/2) by formula (130) and
the function r(2z) by the formula, derived from formula (130) by
replacing n by 2n, we obtain:
2” - 1r ( 2) r ( z + - )
T(2z)
2 z +±
(n!)2 2z (2z + 1 ) ... (2z + 2n)
= lim
n~ ” 2nl z \ z + —j (z + 1) (z + y j . . .(2 + n) (z + n + y j (2 n )“
or
222- 1 r (z) T (z + y )
= lim lim (144)
r (2z) 2nlyn ; 2z + 2n + 1
now n 1
lim 2z+ 2W+ 1
;r: 2 ■
and we can see that the left-hand side of formula (144) is independent
of 2. Assuming z = 1/2 we obtain:
2“ - 1r ( z ) r ( z + i ]
r(2 z )

and from this follows formula (143). In exactly the same way as
above the following, more general formula, can be proved:

= (27r)2(m-1) mz r(mz). (145)

74. The representation of / ’(z) by a contour integral. We will now


give an expression for / ’(z) in the form of a contour integral which
holds for all values of z. If z lies to the right of the imaginary axis
then
r (2) = J e~‘ tz~x d£. (146)
280 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [74

Consider the integrand


e~,tz~1 = e~* e(z~1)log' (147)
which is a function of the complex variable 2. This function has a
branch-point at t = 0. Make a cut in the 2-plane along the positive
part of the real axis 2. In the cut plane function (147) will be single­
valued; we have assumed that log 2is real on the upper edge of the cut,
i.e. we have assumed that arg 2= 0
y on this edge. Instead of integrating
along the upper edge of the real
axis, consider a new contour of
integration I, illustrated in Fig.
65. This contour starts at + °°,
surrounds the origin and returns
to + ° ° . We know from Cauchy’s
theorem that we can, without
changing the value of the integral
J e - '2 z- 1d2 (148)
1 (2Z_1 = e (z-1 )1 °Ef),

deform the contour I in any arbit­


rary manner providing the singu­
larity 2=0 is not altered and both ends of the contour remain at
We shall now explain the connection between the integral (148)
and the function r(z)\ we will suppose that z lies to the right of the
imaginary axis. By deforming the contour I we can obtain a path of
integration which consists of the following three parts: (1) the section
(+ °°> e) of the upper edge of the cut; (2) the circle Xe, centre the
origin and radius e and (3) the section (e, + °°) of the lower edge
of the cut. On the upper edge, log 2 of the integrand (147) is real.
In transit to the lower edge log 2 acquires the term 2m, for on the
jower edge the integrand will be:
e (z - l) 2 n l e - ( + ( r - l ) l og I

where log 2 is real, as before. We thus have:

J e~' 2Z“' d2 = J e -' t 2- 1 d2 + e^ -1)2*' J e"' t z~ x d2 +


‘ ~ ' + j e -' 2Z“ 2d2, (149)
A.
where e is a certain given positive number. We will show that as
e — 0 the integral over the circle / e tends to zero. In fact on this
74] TH E REPRESENTATION OP A O B T A CONTOUR INTEGRA!, 281
circle the modulus of the factor e-t is bounded independently of z
and the bound of the factor tz~l can be calculated
| p - l | = e ( x—1) l og | l | - y a r g / _ g X - 1 g —y a r g ^

i.e. it will be infinitely small if x > 1, or it will tend to infinity of


the order 1/e1-*. Bearing in mind that we are given that x > 0
and that the length of the path of integration is 27ie we can readily
see that the above integral does, in fact, tend to zero. Hence formula
(149) gives us the limit
30

(ienni — 1) f e - ' ^ d f = f e -'F ^ d *


6 i
or, remembering the definition of r(z):
J e -' tz~' d* = (ezl7ri - 1) T (2). (150)
i
The above formula can also be written as follows:

r ^ ) = -j^hrr- (i5i)
l
The contour I does not pass through the origin and therefore we do
not have to consider only those values of z which lie to the right of
the imaginary axis. In the same way as for integral (113) in [71]
we can show that integral (148) represents an integral function of z.
We proved formula (150) only for values of z which lie to the right
of the imaginary axis, but as a result of analytic continuation it will
also hold in the whole z plane. Formula (151) represents a mero-
morphic function in the form of a quotient of two integral functions.
The denominator (ez2nt — 1) vanishes for all positive and negative
integral values of z. Whole negative values of z and z = 0 give the
polarity of J ’(z). If z is a positive integral number then the integrand
(147) will be a single-valued and regular function of t in the whole
plane (i.e. it is an integral function of t), and, according to Cauchy’s
theorem, its integral along the closed contour I will be equal to zero,
i.e. when z is a whole positive number the numerator and denominator
on the right-hand side of formula (151) vanish and therefore these
values will not be poles of the function r(z).
Let us replace z by (1 — z) in formula (150):

i
J e -' t~z dt = (e-*’2"1'- 1) r (1 — z). (152)
282 TH E APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [75

Replace / by a new variable of integration r, assuming that


I = e nl r — —t :
j e - ‘t- zd t= - J eT( e ni r ) ~ z dr = - e ~ Z!,i f eTT-2d r, (153)
i v I
where V is the contour illustrated in Fig. 66. The T-plane is obtained
from the 2-plane by rotation about the origin through an angle (—n).
The cut along the positive part of the
real axis in the 2-plane is transformed
into a cut along the negative part of the
F i g . 66
real axis in the r-plane, where the lower
edge of the new cut corresponds to the
upper edge of the former cut. On this
lower edge of the new cut we take arg (eni t) = 0, i.e. arg t = —n.
Substituting the expression (153) in formula (152) and multiplying
both sides of the equation by (—enzi), we have:

J eTt~ zdr = (e”zi — e~nzi) F(1 — z)


r
or
| eTt -2 dr = 2i sin nz r (1 — z ) ,
r
whence, using formula (122) we obtain the expression / ’(z)-1 in the
form of the contour integral:

r W = i - J eTT“7dT- (154)
/'
75. Stirling’s formula. In this section we shall give an approximate expres­
sion for log J ’(z) when z assumes large positive values. As a preliminary we
shall prove a formula which establishes the connection between the sum of
equidistant values of a certain function and the integral o f this function.
Let f ( x ) be a function which is determined when x > 0 and which has a
continuous derivative. Denoting positive integral numbers by n and k, where
k < n we can write:
n
f(n)-f(k)=Sf'(x)dx
k

and summing with respect to k from k = 0 to k = n we h a v e:

n n n
( n + \ ) 1 { n ) - 2 f ( k ) = J } J /' (x) dx. (155)
k-o k-ok
75] STIRLING’S FORMULA 283

In the expanded form the right-hand side can be written as follows:

2 J /' (*) dx = J /' (x) dx + $/' (x) dx + $ /' (x) dx + • • • +


k-o k o l 2

+ j’/' (x) dx + I /' (*) d^ ,


n —i n

where the last term on the right-hand side is obviously equal to zero. If to is a
certain positive integral number less than n, then when integrating the
above sum in the interval (m, m -)- 1) it will be repeated (m + 1) times and
we can write formula (155) in the form:

( « | l ) / ( » ) - i / ( 4 ) = J {[*] + 1}/' <*) d* . (156)


A -o 0

where by [ 2 ] we denote the integral part of the positive number x, so th at


[x] = m in the interval (m, m + 1) and [w»] = m. Let us now introduce the
function
P (x) = [x] — x ,
which represents minus the fractional part of the number x. If we add unity
to x then [x] and x increase by unity and P(x) remains unaltered, i.e. P(x)
has a period of unity. The function P(x) is determined when x > 0,
but from the law of periodicity when the period is unity, we can natur­
ally extend its definition to include also negative values of x. We know
from [II, 142] th at the value of the integral of P(x) along any interval of unit
length does not depend on the position of this interval. This value gives the so
called mean-value of our periodic function. In the interval (0, 1) we have
P(x) = —x and the mean-value of P(x) is:
1 1
J P (x) d i = —J" x dx = ------ ■
0 0

Let us construct a new function with a unit period

P,(x) = [ x ] - X + y , (157)

F i g . 67

the mean-value of which is zero. The graph of P,(x) is illustrated in Fig. 67.
Let us replace [x] in the integrand of formula (156) by its expression from
formula (157):
n
(« + !)/ (n) - 2 g / W = J {x + y + pi (x)j /' (x) d1- (158)
0
284 THE APPLICATION OP THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [75
We evidently have:

f y / ' (*) d * = y [/(«)-/(<>)]


0
and, integrating by parts, we have:
n n
J x f ' (x) dx = n f (n) — j* / (x) dx.
o o
Substituting this in formula (158) we obtain:

2 /< *) = ! / (*) dX + - [/ («) + / ( 0 ) ] - J P , (X) / ' (x) d x , (159)


k-0 0 2 o
which establishes the connection between the sum of the equidistant values
f(k) of the function/(x) and the integral of this function.
Let us choose the function /(x) as follows:
/ (x) = log (z + x ) ,
where z is a certain positive number and the values of the logarithm are real.
Substituting into formula (159) we obtain:

2 ! lo6 (2 + *) = (« + n + y ) log (z + n) - (z - y j log z - n - J dx.


o
P ut z = 1 in this formula and subtract term by term the new formula thus
obtained from the one above. Also subtract (z — 1) log n from both sides of
this equation. We then have:

• I l0<? - (2 - 1) log * = (Z - 1) log (1 + -1) + | log (l + j ~ ) +

+ (I + « ) l o g ( l + ^ ) - ( z - l ) l o g z - J | ^ d x + J A ^ < t o .
0 0
As n tends to infinity the first two terms on the right-hand side tend to zero
and the third term has the limit [I, 38]:

lim (1 + n) log = lim log 1 = log ez 1


n—*® (1 + [ + D n—* \
We can therefore write:
f z (z + 1) ..
lim log • <2 + ” ) n -Z. n 1 _
n— *® L 1 -2 . .. n n + 1J

f - r £ * + . Jf f1 ■+<*X> Cx
= (2 • z) log z —
-■ > + ( 4 -
I) o
75] STIRLING'S FORMULA 285

or [73]:

log r(z) 2+ 1 - f P . (g ) P\(x)


J 1+ * dx 2 + X
dx. (160)

Now consider the function:

Q(x) = J P, (x)dx. (161)


0
Owing to the foot that the mean-value of P t(x) is zero, the funotion Q{x) is
a continuous periodic function with a unit period and <J(0) = 0. The modulus
of this function is thus bounded. If 0 < x < 1 then [x] = 0 and formula ( 1 6 7 )
gives:
x

Q (x) = | ( 4 - *) dx = Y - (0 < x < 1) .


0
I t follows directly from this that

o < « ( * ) < -g- • (1 6 2 )

Integrating by parts we obtain:

f A M dx = f « ^ > d x =
.! z + x J z+ x
0 o
f Q(x) , r <3W 7- f Q (x ) . _
= \ ^ + ± ^ d x + [ T + ^ L , - ) - ^ + - xy dx’ (163>
where the term outside the integral vanishes when x = oo.
These considerations show us, by the way, th at the above integrals have a
meaning [cf. n , 83]. Replacing x by a new variable of integration t where
x = zl we obtain:
Fp . (*) d- = 1 f Q(zt) d«. (164)
J 2+ X 2 J (1+t)2
Also from (162):

I P .(« )
2 + X
dx 1 f Qm j.
2 J (1 + i ) 2 d t < 82
1 f
J0 ( 1
dt
+ t )2 “
1
82

This shows th at the integral (164) tends to zero as the positive number z
increases indefinitely; moreover the product of this integral and z is bounded.
This is usually written as follows:

f P . (* )
dx = 0
J 2 -j- X
286 TH E APPLICATION OF TH E THBORY OF RESIDUES, FRACTIONAL FUNCTIONS [75

Formula (160) can also be written in the form:

logT(z) = (z - y ) logz - z + C + O ( y ) , (165)

or
log F (z) = (z — y ) log z - z + C + to (z), (166)
where
I “>(*) I < - 5 -. (167)
and C denotes the constant:

C= 1 f Pi (a) r] r .
J 1+x
Let us now determine the value of this constant. For this purpose we shall
use the so called Wallis formula which expresses ji/2 as the limit of a certain
fraction:
n 22•42. . .(2n —2)2 • 2n
(168)
2 l2• 32. . .(2n —l)2
We will prove this theorem at the end of this section so as not to interrupt the
continuity of presentation.
Formula (168) can be rewritten in the form:
_ 2n—1 -i
1/ n 2 2 (n!)2n 2
I T “ n™ (2n)I ’
from where, by taking logarithms and remembering that m ! = r(?n -f- 1) for
whole positive values of m we obtain:

lim [2 log r (n -f 1) —log r (2n + 1) + {^n - y j log 2 - y logwj = log j/y.


Using formula (165) we can rewrite this equation in the form:

lim [(2n + 1) log (n + 1) — ^2n -f y j log (2n + 1) — 1 + C +

+ [2n - y j log 2 - y log n j = log f y ,


or
lim l2n [log (n + 1) + log 2 — log (2n + 1)] +
n-»® I

+ [log (w + 1) - y l o g ( 2 n + 1) - y logn] + C - 1 - y log 2j = log y ,

or

{l0g ( 1 + 2 ^ T r ) !n + T log ^ ^ T T ) + C - 1 - y log 2 } = log .


75] STIRLING’S FORMULA. 287

The first term in the figured brackets tends to loge( = 1) and the second
to (— 1/2 log 2) so that we obtain the equation

l - i l o g 2 + C - l --lo g 2 = lo g / -
' 2
whence C = log ^2jr. Substituting this into formula (165) we obtain Stirling’s
formula:
log r (z) = log + (z — log z —z + w (z), (169)
or, rejecting the logarithms:
r (z) = ]f2nzz ~ 2 e~2e (z), (170)
where the factor e(z) = e“<z) tends to zero as z increases indefinitely. If z is
equal to a whole positive number m, then multiplying both sides of the equation
by m we obtain:
(171)
where em -* 1 as m increases.
We already know that the function r ( z ) has no zeros and th at log / ’(z) is
a regular single-valued function in the z plane cut along the negative part of the
real axis. If this cut were to be isolated by a sector, apex the origin, which can
be as small as we please, then formula (169) can be applied to the remainder of
the plane. The proof is exactly the same as the proof of formula (169) when
z > 0. In this case we should take those values of log z and log r ( z ) in the above
cut plane which are real when z > 0.
W a l l i s ’s F o r m u l a . We shall now prove Wallis’s formula which we used
above. We obtained the following formula in [I, 100]:

(2fc - l)(2Jfe - 3 ) . . . 3 - 1 7i
J sin2*x dx =
2k (2k — 2).. .4 • 2 ' 2 ’

2k (2k — 2).. .4 • 2
J sin2*+1 xdx =
(2k + 1) (2k — 1)... 5 • 3 ’

Bearing in mind th at as n increases the value of sin" x decreases we can


write:
n n n
j" sin2*+1 x dx < J sin2* x dx < J sin2*-1 x dx,
0 0 0

i.e.
2k (2 k- 2 ) . . A - 2 (2k - !)• (2k - 3). . . 3- 1 n ^
(2k + 1) • (2k — 1).. .5 • 3 2k • (2k — 2).. .4 • 2
(2 k - 2 ) (2k —4 ) . . . 4 ■2
(2k — 1) (2A: — 3)... 5 • 3 ’
288 TH E APPLICATION OP TH E TH E O R T OP RESIDUES, FRACTIONAL FUNCTIONS [7 6

w h e n c e , re p la c in g k by n:
71 2 2 4 4 2n 2m
T ' 3 3 T ‘' ' 2 n — 1 2m + 1 ’
n 2 2 4 4 2n — 2 2m — 2 2m
2 1 3 3 5 2m — 3 2m - 1 2m — 1

Putting
D 2 2 4 4 2m — 2 2m - 2 2m
3 3 5 '" 2m - 3 2m - 1 2m - 1

we can write
p . ' < _ <- p
rn 2m + 1 < 2 < "
As n -► + oo the fraction on the left-hand side tends to unity and con­
sequently
lim Pn = Y <

which gives Wallis’s formula.

76. Euler’s summation formula. Lot us return to formula (159). Integrating


the last integral on the right-hand side by parts several times over, we can write
it in an expanded form. By using formula (161) we determined the function
Q(x) with a unit period which was such that Q'(x) = P 1(x). If we add a constant
term to Q(x) we can make the mean-value of this function, and of the function
Pi(x), zero. Changing the sign of the function obtained we get the function
P t(x) with a unit period and mean-value zero, so that P'(x) = —P,(x). We
know th at Px{x) = —x + 1/2 where 0 < x < 1, so that

*.(*) = X “ f + ° (0 < x < 1),


and determining C from the condition
1
f P t (x ) tlx = 0,
o
we finally obtain
P . M - X - T + IT (0<x<i).

In this case P 2(0) = P 2(l) = 1/12, and by periodic repetition Pt(x) gives a
continuous periodic function; hence the above formula holds in the whole
closed interval 0 < x < 1. Furthermore we can similarly determine the function
Pj(*) with a unit period and mean-value zero, so that P'^(x) = P2(x). We obtain
the following expression for this function in the main interval (0, 1):
7G] EULER’S SUMMATION FORMULA 289

Continuing this further we can construct the functions Pn(x) with unit periods
and mean-values equal to zero so that

-Pjm Or) = Pun - 1 (x)> P 2 in+i (x) = P im (x ) • (1^2)

All these periodic functions can be expanded into Fourier series, in which the
constant terms will be zero, since the mean-values of the functions are zero.
Figure 67 shows th at Pl(z) is an odd function. Determining its Fourier coef­
ficients by the usual Fourier rule we obtain:

sin 2nnx
Pi (x) = 2 nn
n= 1

Similarly we have for the next function P 2(#):

cos 2nnx
P 2 (x) =
n = l 2 » tf"

Notice th at the above series can be obtained directly from the series P^x)
by integration and change of sign; this corresponds to the relationship
P'z(x) = —P t(*). The series for P2(x) is uniformly convergent for all real values
of x. Bearing in mind the relationship (172) we can also obtain the Fourier
series for the succeeding functions Pn{x) by successive integration; the
constant terms of the Fourier series must be assumed to be zero.
We thus have:

p v CQ3 ^nnx “ sin 2rmx


2 m\ ) £ o 2/ n— » 2m’1 ' ' 9 2 / n - , 2/ n+i 2m+i • * 1 '
n=1 Z n 71 n=l Z 7i 7i

These formulae, by the way, give us:

P i m (o ) = 1 i p ™,x(o)=o.

For future convenience we write:

Bm
Am(0) = V (174)
(2 m) 1

where Bm are the so called Bernoulli numbers.


Let us return to formula (159). Integrating by parts and remembering that

Bm .
p un (0) —p im(n) —(2m)! ’ P2m-ri (0) —Pim+i (n) —
290 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [76
we obtain:

- J Pi (x) f' (x) dx = J P’2(x) / ' (x) dx =


0 0
n

= 4 r [ / ' (w) - *' (0>] - J p *(*) r (*)dx=


0
n
= -ff- [/' («) - / ' (0)] - f P3 (X) / ' (X) dx =
6
n
= - f p [/' (») - /'(0)] + J p s (*) / '" (*) dx =
0
n
= -fr[// <n) - f' (°)] - o\ p *(*) (*) d* =
n
= - f p [/' (») - / ' (0)] - [ /'" (») - / " ' (0)] + J P 4 (X ) /('“>(x) dx,
0
continuing this further we obtain Euler’s summation formula:
n
1 m = J / (x) dx + i . [/ (o)+ / ( » ) ] + A [/' (») - r (o)] - [/" ' («) -
1 o
- / " ' (0)] + . . . + ( - 1)" — [/<2m+l) (») - /(2m+1) (0)] +

+ ( - l)mJ P 2mf3 ( x ) / 2m+2) (x) dx. (175)


o
In these evaluations we have naturally assumed th at when x > 0, f(x) has
continuous derivatives up to the (2m + 3)th order inclusively.
The last term on the right-hand side gives the final term of Euler’s formula.
Formula (174) shows clearly th at the numbers Bn grow quickly as n increases
and the infinite series, which corresponds to Euler’s formula, is usually divergent.
Nevertheless, it is still sometimes convenient to use formula (175) for the approxi­
mate evaluation of the sum on the left-hand side of the formula.
Rewrite formula (160) replacing C by the expression we obtained above:

logP (z) = log ][2n + (z — log z —z + J dx.


o
Integrating the above integral by parts, as before, and remembering formula
(174) and also that P n(x) remains bounded for all real values of x we have,
77] BERNOULLI NUMBERS 291

when z > 0:

log r (z) = log l2n + (a - —) log z - z + • -1 — +

_1_ l)m- ] 1
•6 z* (2m - 1) 2m 22m-i

+ (-1)" -1(2m) 1J Bjm+i (x) dx.


(z + x)""+1

In exactly the same way as in the previous paragraph we oan show th at the
last integral multiplied by zim+1 remains bounded as z +oo, i.e.

f F‘m+l (X) - dx = O (— -— 1,
J (z + x)am+l Uim+1'
and the above formula can be rewritten in the form

log r (z) = log l2 h + (z - 1 ) log z - z + . 1 _ -j5s_. + ... +

+ ( - l ) m“ 1------^ ------------ — + o (— -— ]• (176)


(2m— \)2m z2m~ x [ z 2m+1j ’
If we reject the last term and write the corresponding infinite series then this
series will be divergent for all values of z. If, however, we fix m then the last
term, as z + ° ° » wdl be an infinitesimally small quantity of a higher
power, viz. of the order than the power of the remaining terms, which
are of the order 1/z2m_1.
Formula (176), like formula (169), holds in the z plane from which an arbi­
trarily small but fixed sector is cut out, the bisector of which is directed along
the negative part of the real axis. If z is positive then the last term can be
approximated to with greater accuracy and the formula given below holds:
Bx 1
log r (z) = log \[%7i + {z —-i-j log z - z + —
2 z 3-4 z3 + • • • +

+ (-!)" (2m — 1) 2m z2m~1 '


^ + 0-n ( - 1) (2m + 1) (2m + 2) z2m+7> (176.)

where 0 < 6m < 1. We shall not prove this formula here.

77. Bernoulli numbers. Let us define the Bernoulli numbers by the equations:
(2m) 1 “ 1 (177)
22m~ 1n'm ,|£, n 2m
We shall show, for the moment, that these numbers can be defined in an
elementary way and th at they are all rational numbers.
292 TH E APPLICATION OP TH E TH E O B T OF RESIDUES, FRACTIONAL FUNCTIONS [7 7

Write out the expansion of cot z into partial fractions [65]:


1 +» ( 1 1 "v
cofcz = T + k2 m + te ) ’
or
1 - 2z
cot2 = T +

or, changing to exponential functions according to Euler’s formula:


. e2' + e“ 21 _ 1 “ 2.-r
e21 — e“ 2' z*_i z2 — k1)r2
Let as suppose th at z = u/2i:

- 2| ! V
u u M)/T‘ 4fc2JI2 -f- u1

^ ± 1 = 1 + 4 V ____ “____
e“ —1 m jfTi 4&2.’i2 + u 2

2 2 " m

+ l= - + 4 £
eu — 1 t( /£i 4 i27i2 u2

The formula above may be rewritten in the form:

U tl “ 1 (178)
- 1 + - = 2u22 '
eu — 1 2 h-i 4kz n*-{-u2
We can write:

Substituting into formula (178) we obtain:

— - 1+f = - 2i ; [ i ( - ^ ) '] -

Applying the lemma from the theorem of Weierstrass about the addition of
power series [14] we can represent the right-hand side in the form of a series
in whole positive powers of u where \ u \ < 2n

u _ i + Jt = 2 \ *2u2 _ a*u* *8U<S


e" - 1 ^2 [ (2ji)2 (2«)< ^ (2»)«
78] METHOD OF THE STEEPEST DESCENT 293

where we have assumed for briefness that:

S» =1 + ^F + -W +
Remembering formula (177) we can write:
u
= 1- + 2 ( - i)m-1 (179)
e“ — 1 (2m)l
The function on the left-hand side has singularities in the neighbourhood of the
origin: u = ± 2w 'p so th at the above power series has a circle of convergence
( u | < 27t. Dividing u by the series
u3
e“ - 1 = TT + ^ r + T T + ---'
we can obtain the successive Bernoulli numbers Bt.r Below we give the first
few of these numbers:
1 691
42 ’ 2730 '

78. Method of the steepest descent. In the following few sections we shall
explain a method of approximate evaluation of definite contour integrals.
As a preliminary let us explain certain problems connected with the variation of
the real and imaginary parts of a regular function. Suppose that we are given
the following function in a domain B:
/ (z) = w (x, y) + v (x, y) i.
At every point of B at which the derivative/'(z) is not zero there will be a direction
I along which u(x, y) changes most rapidly. This direction I is given by the direc­
tion of the vector grad u(x, y) and the derivative, taken in this direction (and in
the opposite direction), has the gieatest absolute value. This derivative of
u(x, y) in the direction n, perpendicular to I, must be zero [II, 108]. The
plane of the directions n is defined by lines at the level u(x, y) = const, and
the orthogonal plane I by the family of trajectories which are perpendicular to
the level lines, i.e. by a family v(x, y) — const [29]. We can therefore say that
at every point where f'(z) is not zero u(x, y) varies most steeply along the line
v(x, y) = const. Notice that in this case du/dl along the above line is not zero.
If it should happen that at a certain point not only 9ujdn but also du/dl is zero
then at this point the derivative of u in any direction would be zero and it
would follow th at the derivative /'(z) at this point would also vanish.
Let us now investigate the position of our line in the neighbourhood of the
point z0 at w hich/'(z0) = 0. In this neighbourhood we have:
/ (z) - / (Zo) = (2 - z0)p [&o + (z - z0) + . .. ] (p>2;b„#0). (180)
Putting
bv = rv ei?r ; z - z0 = oe‘" (r0 # 0) (181)
294 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [78

and equating the real and imaginary parts of the difference f(z) —/(z0) to
zero, we obtain the following equations for the lines u(x, y) = const, and v(x, y) =
= const, in the neighbourhood of z0:
<Pi (e, *>) = r0cos (0O+ pco) + r,g cos [0, + (p + 1) co] +
+ r2o2cos [02 + (p + 2) co] + . . . = 0 , (182)

<t>j (g, to) = r0sin (0O+ pco) + r,g sin [ft + (p + 1) co] +
+ r2o2sin [ f t + (P + 2 ) c o ] + . . . = 0 . (183)

Consider the equation (182). When g = 0 we obtain:


cos (ft + pco) = 0 ,
i.e.
ft> + pa> = (2m + 1) — ,

where m is any integer. Assuming m = 0, 1, . . . , 2p — 1, we obtain all the dif­


ferent solutions of the equation (182) with respect to co when g = 0:
2m + 1 7i (m = 0, 1, 2, 2p-l). (184)
2P

I t is not difficult to see that


9<Z>, *0,
3co e=o, w=iom
and, consequently, it follows from the theorem of undefined functions [I, 159]
th at equation (182) has 2p solutions for co which are continuous with respect
to e and which tend to com when g = 0, i.e. equation (182) corresponds to
2p lines, which radiate from the point z0_ and which have a t th at point tangents
with amplitudes com. But com+p = com + jt and we can have p lines which pass
through the point z0 and have definite tangents at th at point. These lines will
divide the neighbourhood of the point z0 into 2p curvilinear sectors with equal
angles n/p at the apex. Inside these sectors, in the neighbourhood of z0, we have
alternatively cP^g, co) < 0 and g, to) > 0, viz.

when — + mji < p0 + pco < — + (m + 1) 7i:

< 0, if m is even.
(g, co)
1 > 0, if m is odd.
This is directly due to the fact th at the sign on the left-hand side of equation
(182) for the given co, which is other than (184), and for g sufficiently close
to zero, is determined by the sign of the first term.
Similarly equation (183) determines p lines which pass through the point
z„, where the tangents to these lines serve as bisectors of the angles defined by
the tangents to the lines (182).
78] METHOD OF THE STEEPEST DESCENT 295

The point z0 we shall call a saddle-point, the sectors to) < 0 negative
and the sectors 0 x(e, to) > 0 positive sectors.
Let us now consider an integral of the following type:

/„ = ! ( * - 2o)a-1 F (*) to (*)]" dz = J (Z - «o)a_1 (*) en/(z) dz, (185)


I I
where F(z), <p(z) and/(z) = log j>(z) are regular at the point z0; F(z0) and <p(z0) are
not zero, and n is the greatest positive number. Suppose that the contour
I originates at the saddle-point z0 and ends at a certain point z, and that it
lines in the negative sector. When | en^(2) | = enu(x'y> it has its maximum at
the point z0 and for larger n ’s this maximum is abrupt. I t is therefore to be
expected that the principal value of integral (186) is due to integration along
the contour I near the point z0 and it is convenient to select this section to be
along the line v(x, y) = const, along which u(a;, y) has its steepest descent. Instead
of taking the line itself we can take a small section of the tangent to this line.
We must remember, th at the contour can be deformed in accordance with
Cauchy’s theorem. Thus integral (185) can be broken up into two terms: an
integral along the small section V near the point z0 and an integral along the
remaining part of the section I" up to the point zx. The upper bound of the
integral along V is found from its modulus, and the integral along I', from
which the principal part of I n is obtained, is evaluated approximately, and the
inequality of error should be given. The integral along I' is usually evaluated
approximately by expanding the integrated into a Taylor’s series and finding
the upper bound of the last term of this series. In the following sectino we
shall apply the above system to the evaluation of an integral and, restricting
ourselves to relatively approximate upper bounds, we shall isolate the principal
part of I n whilst for the remainder we shall obtain an inequality of a certain
order with respect to a small quantity 1/n.
Let us now make certain general remarks with reference to the evalution
of integrals of the type (186). The contour I can pass through the saddle-point
and from one negative sector into another. In this case, as before, the principal
part of J„ will be determined by the integral over the small section near the
point z0, and the contour must either pass along the line v(x, y) = const, or along
the tangent to this line. If the contour I lies in the positive sector, then the
main part of I n will be obtained by integration over the small section near
the point zlt and this contour should pass along the line of the steepest descent
of u(x, y) [the line v(x, y) = const]. If the integrand is many-valued then
by integrating along the lines of the steepest descent we must bear in
mind any cuts made because of the many-valuedness of the function and a
section of the path of integration should be directed along these cuts; in making
cuts we must take into account the above general concept of stationary points
and the steepest descent of u(x, y). If, there are several saddle-points in the
domain through which the contour I passes, then a comparison of the modulus
of the integrand domain at these saddle-points should be made and the path of
integration should be selected with these general considerations in mind.
We shall explain this by a number of examples. Let us now try to evaluate
the integral along the contour I in the negative sector and, for the moment, we are
only interested in low orders like that of the small quantity 1/n. Without loss
296 TH E APPLICATION OP TH E THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [7 9

of generality we can assume th at z0 = 0. Also, denoting /(z) by the form


/(z) = /(z0) + [/(z) — /(z„)] and taking e^z°’ outside the integral we can
assume /(z0) = 0 , i.e. <p(z0) = 1.
79. Isolation of the principal part of an integral. Let us now consider the
integral
Z, z,
I n = J za —1 (z) [ ? (z) ]" dz = J za _ 1 F (z) en/(2) dz, (186)
a 0

where the path of integration goes from the saddle-point z = 0 to the point
z, in the negative sector with to even (to = 21). We suppose th at the func­
tions F(z) and <p(z) are regular in a certain domain in which the path of
integration lies. Suppose that near the point z = 0
F [ z ) = a o + a,z + ajZ2+ . . . ; / (z) = log <p(z) = z? (b0 + btz + b^z* + . . . ) , (187)
where a 0 and 6 0 are not zero. If the integral is to exist at the lower limit z = 0
we must assume th at the real part of the number a, which we denote by R(a),
is positive. According to Cauchy’s theorem [5] we can deform the path of
integration in the neighbourhood of the origin and direct it from z = 0 along
the tangent to the line of steepest descent, i.e. along the bisector of the above
sector which corresponds to to = 21, to the point z = p0 eia>o and from there
to the point z = zl. Along the second part of the path, max | <p{z) \ < 1 — 77,
where 77 is a certain positive number which depends on the choice of g0. In
future we shall select q0 independently of n, for along the second section of
the path the modulus of the integral (186) will not exceed the product M( 1 — rj)n,
where M is a certain constant, which is independent of n, so that

In = F (z) <P(z)" dz + 0 [(1 — rj)n]. (188)


0
where by 0 [ ( 1 — rj)n] we denote a number which tends to zero as n increases
or, more accurately, it is such that the quotient obtained by dividing it by
(1 — T])n remains bounded as n + 0 0 . In future we shall denote by the
symbol 0 (an) a number which is such that the quotient obtained by dividing
this number by an remains bounded as n ->■ + 0 0 . The amplitude ca' of the
bisector of the sector with the number to = 21 is expressed by the formula:
/?„ . 21+1
71. (189)
P + P
Let a be a number which is less than the radius of convergence of the series
(187). We select, in any case, g0 < a. The series ^ b^z" and the differentiated
- v-0
series ’»byzv~ 1 have the same radii of convergence as the second of the
i»=i
series (187).
Applying the inequality to every modulus of the coefficients of the power
series [14] we can write
M . M_(v = 0 , \ , 2 , . . . \
IM < (190)
' fio " U = 1,2,3,...J
79] ISOLATION OP THE PRINCIPAL PART OP AN INTEGRAL 297

where M is a constant. For further evaluations we shall introduce a new con­


cept.
We shall say that the power series 2 gv zv (or a corresponding function)
v=o

is dominating a power series (or function) 2 hv zv, providing the gv are


l>=0
positive numbers and | hv | < gv. In this case we obviously have the inequality
00
2 1v < i t s , i*r.
P-0 **“0
and we are naturally assuming th at the series converge.
Let us write the product F(z) <p(z)n in the form:
F (z) <p(z)n = (a, + ol2 + . . . ) en^ P+6‘2P+'+•••>=
= a0e < 2P+ en4»2P[(a0 + « ,2 + ...) _ aJ . (191)
The difference in the square brackets can be expanded into a power series
without a constant term:
V (2) = K + 0 ,2 + . . . ) e" ^ + ^ ' + ■■■)-a0 = clz + c22 2 + . . . (192)
Bearing in mind that ez can be expanded into a power series with positive
coefficients we obtain a dominating series for the series (192), if we substitute
the series in the index of e and in the factor in front of e by the dominating series
( z Z* \ „ W - - *-+ ¥■■ Z
- +. . A
A f + I - + M ^ + ... e )-M ,
or
M (l - i - ) - ' ^ “* 0 “ I ) _ M = M [ ( l - L y X~MnlP _ i ] , (193)

and in this last series there will also be no constant term. Denoting, for the sake
of briefness nzp — z \ we can write the dominating series (or, still better) the
dominating function in the form:
2 ri + Mz' , (1 + Mz’) (2 + Mz') ( z \ ,
o L II + 21 +
+ (1 + Mz') (2 + Mz') (3 + Mz') +

On taking (1 + Mz') outside the bracket we can write the dominating function
for (192) in the form:

<* + + (• + * ] ( - ) + (* + ¥ ) ( > + ¥ ) & ) ' + " ] •


and, finally, decreasing the denominators, we obtain the dominating function:

(194)
298 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [79

We thus have the inequality:


\z\( I 2 |\ —1—Mn| z| P
|V>(z)| < ( l + M n | 2 | P ) M L J [ l - L - J J . (195)

According to formula (191) integral (188) can be broken up into two integrals:

J n = a0 S za~ ' e nlvTd z + J za- l e " ^ V (z)dz + O U l - r i ) n].


0 0
We now replace z by a new variable of integration t, according to the formula:

= e'^o ][— .
I nra
where, from (189):
nzP = -i-.
On
We obtain:
I„ = A n + B„ + 0[ ( l -*?)"], (196)
where
nrctS a
f e -'tP~ d t,
o
(197)
nr*l a .I
| e-'tP y>(z) d<.
0
Taking into account the fact th at | z \ = ( tjn r ^ lP , we obtain from (195):
p ___ _Ml

We took Q0 < o and therefore along the path of integration we have


| z | = (</nr„)>/P < e„ < a so that, replacing | z | by the greater quantity ga
we have:
p ___ _ Mt

Bearing in mind th at for q > 0 and a complex y we have | q |y = gR<v), where


R(y) is the real part of y, we obtain the following inequality for Bn:

*(<■>+1
. ln,“
I6 oa 11-
( - 1 ) r i
o1
o |

a ) ' *

nr"l Mt J?(a) +i
ro P
/ • - ( >-*) ■
79] ISOLATION OP THE PRINCIPAL PART OF AN INTEGRAL 299

If we also subject o0 to the following condition apart from the condition p0 < a

we can take a fixed g0, which satisfies both these conditions. In the last integral
the integrand will contain a factor a~l and we can integrate to (+°o)
The integral will be convergent and the value of the integral of a positive func­
tion will increase owing to the extension of the interval of integration. The
integral will no longer depend on n and we obtain the inequality:
R(a)+1

where M x is a certain constant which does not depend on n, i.e.

The quantity 0[(1 — jj)rt] having been divided by (1 — r])n remains bounded
as n -*■ +oo. I t remains the more bounded for being divided by (l/n)'/'>(a)+1)',p,
for the ratio (1 — T])n: (l/n )^<a,+1^p —>- 0 as n -* oo, i.e.

Bn + o [(i - n)n] = o
and formula (196) can be rewritten in the form:
K(a)+i
-In — +O (198)

Let us now consider A n. We can write:

A„ = Jm'oa f— V tp dt - e,a,o' dt ,
P
nr
or have [71]:

(199)

where
a ao
.^L eia,'oa 1
P 1ll—
nr0Y
J .)
nr0Q°a
and consequently:
f?(°)
< K _L|eK a| ( n p r e " 'i KM P
,
dt.
P \ n r 0) J
nr09o
300 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [79

For large positive values of t the function


RM l
e -'t P
decreases and therefore, isolating the factor t~2 under the integral and replacing
the remainder of the integrand by its value a t the lower limit of
integration we can say th at for large n the value of the integral is less than:

W + 1 f" di RM
e~nro<io(nrcoP) P J — = e nrefl (nraQP) P .
wcf>5
This gives us the inequality for Cn:
\Cn \ < M 2e - nr<>e5,
where the constant M 2 does not depend on n, i.e.
Cn = 0 (e- nr°<?5).
Bearing in mind that, as n increases, the exponential function e~nr° pjj decreases
faster than any negative power of n we can write:
«(< +n
0 r »(“)+n
cn+ o
@' Ma)' 1
The formulae (198) and (199) give us finally:
ft(o)+1
P
In ( 200 )

In tliis formula the principal term is of order (1jn)R(a^lp and the last term is
of a smaller order still.
If more accurate inequalities are required several more terms can be isolated
from /„ in the order of increasing powers of (1/n). This brings us to the fol­
lowing general formula which we do not intend to prove here:
a+ v r /?(q)+m
P
In = - V dv (— ] P + O ( - ) ( 201)

In this formula
d _ eK(a+>') y - Sv,li r fq v +p
*’ ~ e 4 ( - V“0 I p
gv, 0 = av and gy are coefficients of zv in the expansion:

-j-j- K + °i2 + ■• ■) (V + biz*+ •••)"•


We shall consider the simplest path of integration from the point z = 0 to the
point z = z,; at the beginning of the path, i.e. at the point z = 0 the modulus of
79] isolation op th b principal pa rt op an integral 301

g>(z) has its greatest value. Consider now the integral:

I n’ = ] z a- 1F(z)q,{z)n 6z, ( 202 )

where z, and z2 lie in sectors where [ <p{z) | < 1. The path takes us from z, to
3 = 0, surrounds this point along the arc of a small circle, centre at z = 0 and
proceeds to z = z2. If R(a) > 0 then, as the radius of the above circle tends
to zero, the integral along the aro of the circle will also tend to zero and we can
simply integrate from z = z, to z = 0 along the sector with the number m = 2 / i ;
an d from z = 0 to z = z* along the sector with the number to = 2P we thus obtain

where I n (a and I n/l are integrals of the former kind over contours which lie
in the above sectors. Hence:

(203)
where

I t can be shown that formula (203) is valid also when R(a) < 0.
Example. Consider
CD

r (n + 1) = j e x i"dx.
0
On substituting x = ny we obtain:
90

0
The function ye~y has a maximum when y = 1. Substituting y = 1 + z:
SO

-I
Divide the interval of integration into two parts: ( —1, +1) and (+1, + °°).
For the second interval we have:

I [(1 + 2) e " T dz = j- [(1 + z) e -2]"-1(1 + z) e"2dz <


l 1

for, when z > 1


(1 + 2 )e -z< c
302 THE APPLICATION OF THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [79

Therefore

J [ ( l + 2) e - z]"d2 = 0 [ ( - - | ]. (204)
1
The integral
l
J[(l+*)e-*]"d! (205)
-l

has the form we considered above, but in this case

o0 = 1; a = 1; .F (2 ) = 1 and log <p(2 ) = log (1 + z) — 2 ,

from which it follows that p = 2 and b0 = 1/2, i.e. r 0 = 1/2. I t can also be
seen th at co' = and co' = 0. We finally have from (203):

^ - (iM )+0£)+ [£)’]’


7 7 0

or, taking into account th at (2/e)n decreases more rapidly than 1/n and that
r(l/2) = , we obtain:
e"r(n + l) V2h ,
m„ +1 = - y + 0 |- j ,
vr
or
,1
7\'w + 1) = nF(n) = /2hn 2 e~" > + J o (1 )]-

= l^ m " +2e-nr i + o f - U ’
I VKw

r(ri) —Y2nn %e "


L 'U)\
The last term 0(l/^w) should, in fact, be 0(l/n).
To obtain this result the integral (205) must be divided in two, viz. from
2 = 0 to z = 1 and from 2 = —1 to z = 0, and formula (201) should be applied
to each of these integrals when m = 2. At the same time terms corresponding
to v = 1 in these integrals will cancel each other and we shall obtain the same
principal term and last term 0(l/n).
All the details of the above method, the proof of the general formula (201)
and examples can be found in Perron’s paper “Ueber die naherungsweiae
Berechnung von Funktionen grosser Zahlen’’ (News of the Bavarian Academy,
1917).
One of the first works in this field was a work by P. A. Nekrasov.
80] EXAMPLES 303
80. Examples. 1. Consider the integral
i
a
1=
I z + ia
1 en <z' - 2’>dz , (206)

where a is a small positive number and n is a large positive number. The


function /(z) = z3 — z* has its maximum when z = 0 and the real axis is the
line of the steepest descent of v(x, y) = 0.
Bearing this in mind let us represent I in the form:
i
a

_1
a
where
_i
2
« -
J
f —~
2 + 10
e" (2*-2*) dz.

Noting th at z3 — z2 < —z2 when z < 0 we obtain:


_X _n _i
M <
y«’ + T — f a’ + T - -

dz.
r ^ T rf
Replacing z —1/2 by ( —1) and introducing a new variable of integration
t = —(z + 1/2) we obtain:

* 2e *
r e-"'d ( = ■ °
to <
rl — < (208)

To evaluate the first term on the right-hand side of formula (207) we suppose
th at e'u3 = 1 + A, where
. nz3 (nz3)1
J = T1 - + - 2 T - + - ’
hence

M l < n | z | 3 (l + + --W g2, — . . . + ) < W| z | 3e " l*i \ (209)


304 THE APPLICATION OP THE THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [80
We obtain:

( 210 )
z + ia z + io

Also

r *~m' j r e~nz* j , ( 211 )


J —
2 +r~^~
B ds = J —
2 +r-1-
tO d* + wi »

where

I“ i I < • -nz«dz
fj-
a*+ ^r i
and

e —n r! =e-i re~ " (Z- I)(2+ I)d2 ,


dz = e

or, replacing z -f 1/2 by unity:

r _ ? r” e _a
e_n22 dz < e t i e ' 2'dz = — *,

whence, finally,
2e 4
co, < — < — e *. ( 212 )
n
" f a*+ T
Consider the first term on the right-hand side of formula (211). Separating
the real and imaginary parts of the integrand and noting th at the integral
of the real part is zero since it is odd, we obtain:
+® +
r **—nza r P— _
J T + ^ r dz = - i S - ^ + r dt ^ = a ^ )-

On assuming th at /S* = y and differentiating the integral

C M - J ’ - S ? ! d(

with respect to the parameter y we obtain:

d/(r)
dy = I (y) -
80] EXAMPLES 305

On integrating this equation, bearing in mind that I(y) = 0 when y = -j- o o ,


we obtain:
00
I ({}*) = 2][n eP* J e~xS dx ,
P
and, finally,
4* 00 oo
f —:— dz = — £2 f n ea2n f e~x*dx. (213)
J z + la J
~“ a\n
The integral on the right-hand side (the incomplete Laplace integral) is
tabulated.
We come lastly to the upper bound of the second term on the right-hand side
of formula (120). We shall carry out this investigation in two different ways.
On recalling (209) and the fact that | z | 3< | z |2/2 when | z | <1/2 we
obtain
| d | e_nz2 < n | z |3e 2 ,
whence
1
OD
2n 2 22
J d e -" 22
z + ia
dz <
0
| z3e
— n —
3 dz < z3 e 2 dz = — '
an
(214)

On the other hand, taking into account the inequality:


2
--------- - -— M ' 1 - < —- 1 ,
when ,z . < —
1
z + ia / | z | 2 + a2 Y I -f 4a2 2
we obtain

d e - 2122 , |Z 3 - , 2n
---d2 - e 3 dz < x2e 3 dx,
z -+- ia n \ Iz + ia | ^1 + 4a2

whence

J de—2122
z -|- ia
)2ji
dz < ~
/n
(214,)

The principal part of (213) is of the order 1/a^n; if a is not a small number then,
using formulae (211), (212), (213) and (214), we can write:
306 TH E APPLICATION OP TH E THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [80

where
(216)

When a takes small positive values we can write:

(216,)

2. When considering the asymptotic representation of Hankel’s functions


we find if necessary to evaluate approximately the integral
—a+ t

—a — e

in which
/ (z) = sinh z — ?z, (218)
where the parameter £ > 1 and n is a large positive number.
a in (217) is the positive zero of the equation f'(z) = 0, i.e. cosh a =
The number e > 0 we shall consider right from the beginning to be less
than unity and later on we shall subject it to even more restrictive conditions.
I t can readily be shown th at the path of integration in (217) coincides with
the major part of the line of steepest descent of the integrand. For this reason
the evaluation of the integral involves the rational transformation of the inte­
grand and the evaluation of elementary integrals.
We shall consider two methods by means of which integral (217) can be eval­
uated. In the first method our main object will be to find the principal term
and to obtain a relatively simple assessment of error. In doing this we shall
disregard certain essential properties of the integrand.
In the second method we shall take these properties into consideration and
consequently the result obtained will be more accurate.
First method. Expand the function (128) into a power series in where
x = z + a. We obtain

(219)

or
/(z) = / ( - a ) - 5 ^ x2[i + U]> ( 220 )

where R satisfies the following inequality:

( 221 )

if | x | < 1.
We represent the integrand (217) in the following form:

( 222 )
en/(z) = ef>/(-o) (l + «5«),
80] EXAMPLES 307

where
sinha
-n—— fix' (223)
<5„ = e 2 - 1,
If we use the inequality
I e* - 1 | < I y I e1|yl
and take (221) into account then we obtain for <50 the following simple inequ­
ality:
sinha.
, - . n cosh a , n 5 x
I <50 | < ---- | re |3 e 2 . (224)

Substituting (222) in (217) we obtain:


sinha sinha
I = enf(~a) j e " 2 **dx + J <50e ™ 2 * d (225)

Subject e and n to the following two conditions:


sinh a
■e* = N §> 1 (226)

and
cosh a
» £ < 1. (226,)
sinh a
In a more accurate method of finding the upper bound the condition (226)
is preserved but condition (226,) is replaced by a stricter condition.
If condition (226,) is satisfied then when | x | < e, R satisfies the inequality
| R | < 1/2 and from (224):
sinha sinha
n cosh a . .
<5„e < ----:--- X (I X I < £). (224,)

If condition (226) is satisfied then the integrand in (225) becomes exceedingly


small near the ends of the interval of integration. For this reason the transi­
tion to an infinite interval of integration has no essential influence on the
result. Carrying out the necessary calculations we have
sinha + OO sinha

F
t
—r— X* 2n
2 dx 2 dx + d. (227)
J* J • n sinh a +
-e —OS
where
co sinha
l, = - 2 j e " 2 ** dx,
e

and
sinha
|d,| = 2 e-N J e " 2 dx,
308 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [80

whence, taking into account that x2 — e2 > (x — e)2 when x > e


slnha
------ y» . 271 , -N
M i I < 2e~N [ e 2 dt/ = (228)
I n sinh a
6
Further, on taking into account (224t) we find:
slnha slnha
. ¥* n cosh a
2 dx < x3e
4

whence
slnha .

—0
I <5„e
—— x»
2 dx
2 cosh a 2
7i sinh a n sinh a "
(229)

Using formulae (225), (227), (228) and (229) we obtain the following expres­
sion for the required integral:

+ <230'
where
2 _ c ^ |/ ^ *
7i sinh a \ n sinh a

If conditions (226) and (226,) are satisfied then | co \ is found to be appreciably


less than unity.
The above method of evaluation is very rough but it is also very simple as
far as calculations are concerned. The disadvantage of this method lies in the
fact th at it does not take into account the change of sign in the expansion
(219) and the presence of odd powers in this expansion.
We shall now remove these disadvantages.
Second method. Write the expansion (219) in the form:

/(r) = / ( - a ) - ^ J i ! . x 2 + I?1- J ? !, (232)

where
cosh a x2 x1
x3
*i 3! 4^5 4 • 5 ■6 • 7 + •
sinh n x*
x1 (232,)
R, 4! 5-6-7.8 + •
To start with suppose th at in integral (217) e is only subject to the condition
0 < e < 1. But even so we can say th at R r and R2 differ very little from the
first terms of their respective expansions. Substitute the value of the integrand

e " /M = e n / ( - a ) e~nR‘ (l + + . . .)
80J EXAMPLES 309

in integral (217). Owing to the fact that R 1 is odd we obtain:

/ = J e - . (l + ^ ^ + . . .) dx. (233)
—8
Consider the expression:

+ 5 jS + . . . ) -

(234)
- ( ‘ - T ^ + T ? — •) (> + = £ + =?+■■•}•
Select the positive e in such a way th at the inequalities given below should
be satisfied simultaneously:
e < 1; n ‘Rl < 5; nR2 < 1. (235)
Using formula (232J we can see th at these inequalities will be satisfied providing
l
s< 1 and £ < [----- — — V . (235.)
^ n oosh a ) 1
If this is so then the terms of the series with alternate signs

o i nRt n tR\
61 ~ 1 if + l i
will rapidly decrease. Hence
Sx = 1 —nRz + alt (236)
where
n n*R‘
° < “1< V -
We find from (2321):
, _ , sinh a . sinh a „ (. , x x4
1 —nR2 = 1 — n — — x4 — n — ^ — x® 11 +
7-8 1 7 • 8 • 9 • 10
Comparing this with (236) we obtain:
o . sinh a (237)
S ,= l - » —jj— *4+ <5„
where
n sinh a
I<5.1 61 *, (1+ ^ + ---)
which is, in any case, smaller than the greater of the two quantities:
56 n sinh
nh a
------x® and
('30') w2sinh2a .
I— I -------------x®.
55 5! (.29J 2 • (41)2
310 TH E APPLICATION OF TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [8 0

I t can be shown th at the latter quantity gives the greater of the errors shown
below. We can therefore take it th at
30V n 2sinh2a
x8. (238)
W < © 2 • (4!)*

For the second series, from (234):

9 - 1 4 - n*R*_LniRi 4_
= 1 + ~2~ + “J r + '
we have:
o _i i n2Rj . (236,)
--- 2--- a
where
0 < a, < 6 n4R\
We find from (232,):
n 2R\ , , w2
n2cosh2 a f x2 V
1 + ^ “ “ 1 + 2.(3i)2~ X 11 + 4T5+ - - - J =

= 1 + T W - * ‘ ' 1 + 2' + r’).


where
— ** f, | X2 X* \
4 . 5 i i‘ 6 - 7 i ' 6 - 7 . 8 . 9 i' ‘ ‘ 'j'

It is evident th at r < (42/41) (x2/4.5) and


_ . , 42 x2 [42V x4 _ 11 x2
<
r + T < 4 l ‘ TO + I4 1 J ’ (20)* 10 10 ■

We therefore obtain:
„ n 2Rj , . n 2cosh2 a
(237,)
S2 = 1 + V + °2= 1 + 2 • (3!)2~ * + **’
where 8, > 0 and
s. , /n , ,, n 2cosh2a
s2 = a2 + (2r + r2) — —x <

„ /20y«4 cosh4 a , 11 n 2cosh2a /oo0 ,


< N ~5 • 63 • 4! * + 15 • ' 20.'6» * ’ (238l)
Bearing in mind the inequalities (235) and multiplying expressions (237)
and (237,) together we obtain for (232) the following formula:

-nR.a i sinh a n2cosh2 a - (239)


R St = 1 “ Ij— *' + ■2 . ( 3!)-a- + 15’
where
| 5 | < 5 | <5, | + <52 , (239,)
80] EXAMPLES 311

and | <5, | and | <521 satisfy the inequalities (238) and (238,). Let us now substi­
tute (239) in (233). We have:

elnh a
I= sinh a cosh1a
n xl + n2 e 2 dr -f-
4! 2 • 62
(240)
where
CD
Blnha _
. „ f (. sinh a , , , cosh2a
d0= 2J ( l- n - j i- ^ + f * * ~27erx ) 6 2 x da?,
e
9
(241)
f fllnha
Ax= 2 J 6e~n~2 x da:.
o
The integrals in (240) have to be evaluated and the error (241) must be assessed.
Subject the numbers n and e which satisfy (235,) to yet another most important
condition:
n sinh a
e2 = N » 1. (242)
2

When this is done the assessment of error is obtained in an elementary form and
this gives us the formulae:

2 N 2cosh2a ~J
(243)
18 n sinh3a J ’
5
i a i ./—( 2 W (1 cosh2 a cosh1a
1 < 71\ n sinh a 8 J 1. 25 sinh2a 8 sinh2a
(243,)

whore we assume in (243) th at N > 8.


Evaluating the fundamental integral in (240) we arrive at the following
final formula:
5 cosh2a \ 1 , m; (244)
3 sinh2 a J n sinh a
where
s
e~N fa N2'cosh2 a'i f 2 \ 2 /U cosh2 a cosh1a
l< Yn I 6n sinh3a) V.Msinh a) (,8 25 sinh2 a 8 sinh1 a

Wo obtained (244) and (235) on the assumption th at the conditions (242) and
(243,) are satisfied. Suppose in what follows that
1
12
E (246)
In cosh af -
312 TH E APPLICATION OP TH E THEORY OF RESIDUES, FRACTIONAL FUNCTIONS [8 0

We find
N 3 cosh2a
n sinh a = (247)
18 sinh2a
and
N 2cosh2a 3
(247,)
6n sinh3 a y r
I t can be seen from (245) th at if we take N = 8 then the terms in the second
line of (254) will take values which exceed the value of the term containing the
factor e 'n.
Therefore, if we agree to take N > 8 the formula giving the error can be
rewritten in a simpler form:
i i „( 2 V (1 , cosh2a , cosh4a '\ /n,0.
' " I < 2 U sinh a ) (.IT + 25 sinh2a + 8 sinh3 a J ’ *^

This inequality will be valid providing th at N > 8 and this can be written
as follows:
l 2
3 ^ 3
n sinh a > j— cosh a
yr&
or
1 2
n 3 sinh a > 3 cosh* a. (249)
In the chapter devoted to Bessel functions we shall make use of the results
obtained above but substitute

cosh a = — and n = z,
z
where p is a symbol and z is the argument of Bessel’s function.
In this case z sinh a = )/p1 — z2 and the condition (249) acquires the form;
2
Kp2 - z2 > 3p*. (249,)

If we take into account the fact th at th at p > |!ps — z2 then it becomes evident
th at condition (249,) can be satisfied only when p > p 0, where p 0 = 3p*1*, i.e.
p 0 = 27. When making more accurate calculations this borderline can be
somewhat lowered.
C H A P T E R IV

FUNCTIONS OF SEVERAL VARIABLES


AND MATRIX FUNCTIONS

81. Regular functions of several variables. The theory of analytic


functions of several variables is similar in its fundamental concepts
to the theory of functions of a single variable. However, when developed
further it has several characteristic differences. We shall only deal
with fundamental concepts of this theory and consider power series
of several variables in greater detail. To simplify arguments we shall
take two independent variables. The definitions and proofs will also
be valid for a greater number of variables.
Let zx and z2 be two complex variables and

/( z i. z 2) (1)

be a function of these variables. Let us suppose that the variable zx


varies in a certain domain Bx and the variable z2 in a certain domain
B2. If, at the same time, function (1) is a single-valued and continuous
function of its two arguments and, if for any arbitrary values of the
variables in the above domains, the relationships
/ ( zt + d z 1,z2) - / ( Z i , Z ; ) and f ( z l, z 2 + Az2) - f ( z l, z t)
Azx Az j

tend to a definite limit as the complex increments A zx and A z2


tend to zero, then function (1) is said to be regular or holomorphic
as zx and z2 vary in the domains B x and B2. The limits of the above
functions give the individual derivatives of function (1) with respect to
the variables zx and z2:
9/ (zi. gt) and 9/(Zi.z2)
9*i 3z2

82. The double integral and Cauchy’s formula. Let lx and l2 be two
contours which lie in the domains Bxand B2respectively. We construct
a double integral which is obtained by the successive evaluation of
313
316 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [83

The variables z1 and z2 appear under the integral as parameters.


Differentiating with respect to these parameters we can see that our
function f{zv z2) has derivatives of all orders and we obtain expressions
for these derivatives in the form of double integrals:

dP+i f (g', z2) pj_g! r . f ________ 1 z 'i)_______ dz'


02p dz% (3)
4** /,J *JI, (£1'- Z 1)P+1^ - 2 2 ) ?+1

As for functions of one complex variable the principle of the


modulus follows from Cauchy’s formula: if the function f(zv z.2) is
regular in the closed domains B l and B, and | f(z[, z2) | < M, when
z[ belongs to lx and z2 belongs to l2, then | f(zv z2)\ < M when zx
belongs to the closed domain Bx and z2 to the closed domain B2.
The theorem of Weierstrass is proved in exactly the same way as
for functions of one complex variable: if the terms of the series
CD

2=1 vk fa*22)
fc
are regular functions in the closed domains B x and B2 and the series
converges uniformly in these domains, then the sum of the series is
a regular function in the above domains and the series can be differenti­
ated term by term with respect to zx and z2 as often as we please,
providing zr lies inside Bx and z2 inside B2.
The series remains uniformly convergent in any closed domains B[
and B2, which lie inside Bx and B2.
All the above statements can be extended to include functions of
several independent variables. Below we shall only consider power
series.

83. Power series. A power series of two independent variables zx and


z2 with centres at bx and b2 can be written as follows:

2 2 aPi(zi - W z 2- & 2)9. (4)


p --- 0 0

where the variables of summation p and q are independent of each


other; they take all positive integral values, starting with zero. Series
(4) is a double series.
We dealt with such series earlier in [I, 142] when the terms of the
series were real.
83] POWER SERIES 317

Let us suppose that the series, constructed from the moduli of the
terms of the series

2 2 1ap»\ 13i - 6i lp 1 - b*\v-


p=00=0 (6)
converges.
In such a case, as we know from [11], the series, constructed from
real and imaginary parts of series (4) will be absolutely convergent
and the sums of these double series with real terms will not depend
on the order of summation. Consequently in this case, i.e. when
series (5) converges, series (4) will also converge and its sum will be
fully determined for any order of summation. In future we shall only
consider the case when series (5) converges, i.e. when series (4) con­
verges absolutely.
Abel’s theorem can readily be established in the same way as in [13].
Suppose that the series (4) is absolutely convergent when z1 = ax
and z2 = cq. It follows directly that the moduli of the terms of this
series remain bounded for the above values of the independent vari­
ables, i.e. a number M exists such that for any arbitrary numbers p
and q the inequality holds:
| apq 11cti — bl |p | a2 — b2 < M
i.e.
\ aPl\ < \al ~ b x |P|a2- 6 , |« ’ (6>
Consider now two circles K x and K 2 of the variables z1 and z2:
Izi — 6i I < I«i — &i I: | 3 a - & 2 l < l ® i - & i l - (7)
The first of these circles includes points zx which are nearer to bx
than to cq and the second includes points z2 which are nearer to b2
than to cq.
Take a certain point zx in the circle K x and a certain point z2 in
the circle K 2, i.e.

\zi ~ biI = Iai - I and I*a — &21= ft I“a — I•


h bi
where 0 < qx and q2 < 1. At the same time, using (6), we obtain the
following inequality for the moduli of the terms of series (4):
\apq\\Zl - bl\P\Z2 - h \ q (8)
It can readily be seen that the double series with positive terms
QD CD

2 2
p= 00= 0
318 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [83
will be convergent. In fact this series is obtained by multiplying two
series with positive terms [I, 138]:
M (1 + q2 + q\ + . . . ) and (1 + ?2 + <A+ • • •) >
and its sum is
M
(1 - ffl) (1 - ffl) •
Hence in the case under consideration, series (5) will be convergent
and series (4) absolutely convergent. It also follows from the ine­
quality (8) that series (4) will be uniformly convergent in the circles
K[ and K 2 with centres at bx and b2 and radii @i and q2, which are
less than the radii of K y and K2. In this proof we did not use the
absolute convergence of series (4) when zx = cq and z2 = Oj, but we
used the inequality
Ia pq ( a l - bl Y (a2 ~ h Y I< M -
i.e. we used the fact that the terms of this series are bounded when
z2 — and z2 — flj.
We thus arrive at the following conclusion : if the moduli of all terms
of series (4) are bounded when zx = cq and z2= a2 by one and the same
number, then series ( 4 ) will be absolutely convergent inside the circles
(7) and uniformly convergent inside the circles:
I *i — h I < (x — £) I a i ~ bi I; I — &-21 < ( 1 — b) I <*2 — h I .
where e is any small fixed positive number.
Notice that if, when z2 = cq and z2 = a2, series (4) converges (not
necessarily absolutely) for any arbitrary order of summation, then
its terms tend to zero as they move away from the origin; hence
their moduli are bounded by one and the same number and
inside the circles (7) the series will be absolutely convergent.
What was said above brings us to the concept of the radius of
convergence of series (4) in exactly the same way as in [13].
In this case there exist two positive numbers Jt1 and R2 which
show that series (4) converges absolutely when |zi — | < Rx
and | z2 — b2 | < R2, and that it diverges when \zx — \ \ > R2 and
[ z2 — b2 | > R2. Notice that in this case, the region of absolute con­
vergence of series (4) is simultaneously determined by two radii of
convergence S 2 and R2 and these radii, generally speaking, cannot
be determined separately, for the value of one depends on the other.
If I?! is made smaller then it is very probable that R2 can be made
83] POWER SERIES 319
bigger. In other words in this case we can speak only of combined
radii of convergence R1 and R2 or, which comes to the same thing, of
combined circles of convergence. Consider, for example, the power series

P=0g=0 (9)
Series (5) will become

(10)
p = 0 q -0

Group together all terms of this series in which the sum p + q is


equal to a given number s. According to the binomial theorem the
sum of these terms will be

(lZll + I*2 I)*.


and series (10) can be rewritten in the form

i , ( i zii + iz2i)s.
5= 0

from which follows directly that it will converge when, and only
when 1 | + \z2 \ < 1. Hence for series (9) the combined radii
of convergence are determined by the equation Rx + Ri = 1- If we
take, for example, Rl = 6, where 0 < 9 < 1, we shall have
R2 = 1 — 0.
Consider as a second example the series

2 2 ZW -
p = o g=o

It can readily be seen that here the necessary and sufficient con­
dition for convergence is expressed by the inequalities | zx | < 1 and
| z2 | < 1, i.e. in this case R^ = 1 and R2 = 1 and the radii of con­
vergence are determined separately.
Consider series (4). As a result of uniform convergence and the
theorem of Weierstrass the sum of series (4) inside the combined
circles of convergence is a regular function f(zv z2) of two variables.
As in [13], we can differentiate series (4) with respect to both variables
as often as we please inside the circles of convergence. This dif­
ferentiation does not alter the circles of convergence.
320 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [83

Differentiating several times and assuming subsequently that


Zj = 6Xand z2 = b2 we obtain, as in [14], the following expressions
for the coefficients of the terms of the series:

aPQ 1 9p+?(z„g2)
p ig I dzP dzv (ii)
h.—bi>
i.e. series (4) is the Taylor’s series for the function f(zv z2).
If and R2 are the combined radii of convergence of series (4),
then this series will be absolutely and uniformly convergent when
I | < i?i — e and | z2 — b2 | < R2 — e, where e is any small
positive fixed number. At the same time, from (3) and (11), we have
the following inequality for the coefficients of the series:

I aPi I < (R, - e) p (R, - e)i ’


where M is a positive constant, the value of which must depend
on the choice of e.
Substituting the coefficients apq in series (4) by positive numbers
which exceed the moduli | apq |, we obtain a power series
°° » M
^ ^ R 'P R ’Q(Z1 — bi)p (z2 — bi)q (-^1 = — e; R'z = R2 — e)> (13)
p=-0 q = 0 1 2

usually known as dominating or major for series (4). It can readily


be seen that the sum of series (13) is equal to
M

and this latter function is known as the dominating function of series (4).
Coefficients in the expansion of powers of (zx— bx) and (z2 — b2) are
positive and greater than the moduli of the coefficients of apq.
The result [14] can readily be generalized to include two variables.
Let f(zv z2) be a function, regular in the circles \z1 — bx \ < Rr and
I z2 — I < Rz with centres at bl and b2 and let Zx and l2 be the
circumferences of these circles. On fixing any two points zx and z2
inside the above circles we can obtain Cauchy’s formula

/ [zv zi) = ~ dz'. (15)


- Zl) K - z t)

Consider the rational fraction


l
- 2i) (K - 2s) ‘
83] POWER SERIES 321

As in [14] we can expand it into a series of powers of the dif­


ferences (z1 — 6X) and (z2 — b2)
1 “ (z, - &i)p (z2- bj _ _
k - *.> k - **> £ i k - 6i)P+1 ^ - 6*)?+l ’
this converges uniformly with respect to z[ and z'2, providing these
points lie on the circumferences li and Z2. Substituting this expression
in formula (15) and integrating term by term, we can obtain our
function f(zv z2) in the above circles in the form of a power series

/(z 1>s2) = (ie )


p ~ 0 q=0

The coefficients of this series are determined by the formulae

api ~ ______/(zj.Za)______ d z2
(«[ - Mp+1 K - bt)q+1
1 dp+qf (zt, z2) I
(17)
p! g! 9zP929 k=ft,;2,=*i‘
Thus any function f(zlt z2) which is regular inside two circles can be
expanded inside these circles into a power series. As in [14] it can
readily be seen that this expansion is unique, for the coefficients
must be determined by formula (11).
In series (4) we can group together terms which are equal with
respect to the differences (z1 — 6X) and (z2 — b2), i.e. we can write
series (4) in the form

2 2 ^ ( 2i - 6i)P(z2 - 6 2)9. (18)


s = 0 p+q= s

where the inside finite sum includes those values of p and q the sum
of which is s. Formula (18) gives the function f(zv z2) inside the
circles of convergence in the form homogeneous polynomials with
respect to (z1 — b2) and (z2 — b2). Suppose now, conversely, that
series (18) is given in the form of homogeneous polynomials and
that it converges uniformly in certain circles | zx — bL | < Rx and
[ 22 — Z>2 | < R2. According to the theorem of Weierstrass the sum
of this series will be a regular function of f(zv z2) in these circles.
We can also differentiate our series (18) as often as we please
with respect to both variables. Differentiating and assuming sub­
sequently that zx = &! and 22 = b2 we obtain formula (11) for the
coefficients apq, i.e. these coefficients will be the coefficients of
322 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [83

the Taylor’s series for the function f(zv z2) and we can rewrite series (18)
in the form of the double series (4); this series will converge absolutely
and uniformly inside the above circles. We can therefore say that
if a series given by homogeneous polynomials converges uniformly
inside certain circles then this series can be simply rewritten in the
form of a double series, which is the usual power series and which
converges absolutely inside the above circles.
If we separate the real and imaginary parts of zt = x2 + tyi and
z2= x2 + iy* then in a four dimensional space with coordinates {x1, yv
x2, y2) the region of uniform convergence of series (18) given by
homogeneous polynomials, can be wider than that for series (4).
In example (9) series (18) has the form:

(Z1 + 32)*-
5= 0

and the domain of its uniform convergence is given by the inequality

| “f~%%| ^ 11
i.e.
[xi + x2)2 + {Vi + Vi)2 < 1 ■ (19)
For series (9) itself we must have + .ff2 = 1; its domain of
convergence is determined by the inequality:

1*1 I + I*2 I < 1-


i-e - 1 ^ ! + y\ + Y*i + y\ < i .
or
*i + y\ + xi + yi + 2 ]jx\ + y\ yxl + y\ < 1. (2 0 )
The inequality (19) determines a wider domain than (20), i.e. if the
numbers xk and yk satisfy (20) they will also satisfy (19) and vice
versa. In fact, from the inequality

(z ix2 + yiViY < (xl + y\) ixi + yl)


it follows directly that the left-hand side of inequality (19) is less
than or equal to the left-hand side of inequality (20).
All the above arguments can be extended to include n variables,
when the domain of absolute and uniform convergence of the series
will be a set of n circles.
84] ANALYTIC CONTINUATION 323
84. Analytic continuation. A function of two variables f(zv z2),
determined by a power series (4) inside the circles of convergence,
can be regular in a wider domain; as for a function of a single
variable the question of analytic continuation of this function arises.
Here, as in the case of a single variable [18], the fundamental
proposition that if two functions are regular in certain domains,
they coincide in these domains, if at any two points z1 = bt and
z2 = b2 in these domains, the values of the above functions and of
all their derivatives are the same.
Let us now consider once again the function f(zv z2) which is given
by a power series. Let z1= cx and z2= c2 be two points belonging to
the circles of convergence. Using series (4) we can determine the
values of the derivatives

0Z{ 02? 2j —CjJ Zg—C%

and construct the Taylor’s series for the function f(zv z2) in integral
powers of the differences (z2 — cx) and (z2 — c2)

2 2 a'p^zi - ci)P(22- c 2)L (21)


p =0 9=. 0

It can readily be shown that such a reconstruction of the power


series is equivalent to making the following assumptions in series (4):

(Zi - b tf = [(z2 - cx) + (c, - &!)]",


(z2 - b2)i = [(z2 - c2) + (c2 - b2)]i.

We expand the brackets using Newton’s binomial theorem and group


together terms with the same powers of (z1 — Cj) and (z2 — c2). Series
(21) will certainly converge and its sum inside the circles with
centres at c1 and c2, which belong to the circles of convergence
of series (4), will be f(zv z2). It may happen, however, that these
new circles fall outside the old circles of convergence. In that case
we obtain our function f(zv z2) in a wider domain, i.e. we extend the
domain in which our regular function can exist. Applying the above
process of analytic continuation by means of circles several times we
can, in certain cases, extend the domain in which our regular
function can exist; the whole set of values which we obtain in this
process will give us the analytic function obtained from its element
as given by series (4). We shall not consider here in greater detail
324 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [85

analytic continuation with regard to singularities. All that has been


said holds for any number of independent variables. Notice that the
paths L x and L 2, along which z 1 and z 2 vary in the course of the analytic
continuation of f ( z v z 2), do not determine the result of this continua­
tion. It is important to know exactly the way in which z1 and z2
move with respect to each other along L x and L 2. This is all we are
going to say about the general theory of functions of several complex
variables. At present this branch of the theory of functions has
developed widely. A more detailed description of fundamental facts
in this field can be found in the C o u r s e o f M a t h e m a t i c a l A n a l y s i s by
Hurse. There is also a special book by B. A. Fuchs T h e T h e o r y o f
A n a l y t i c a l F u n c t i o n s o f S e v e r a l C o m p l e x V a r i a b l e s (1948), where
extensive bibliographical references are given.

85. Matrix functions. Preliminary propositions. We shall now


consider the case when the argument of the function consists of
one or several matrices; we shall start with a single matrix. Earlier
in [IIId 44] we considered the simplest cases, viz. the polynomial
and the rational function of one matrix. Before going on to consider
more complicated functions we shall establish some fundamental
propositions. In future we shall denote by n the order of the matrix.
Suppose that we are given an infinite sequence of matrices

X v X 2 ...

We say that the limit of this sequence is the matrix X if for any
arbitrary values of the symbols i and k :

lim { X m} ik = {A}lfc, (22)


m~*00

i.e. the elements of the matrix X m have as their limits the correspond­
ing elements of the matrix X . At the same time we shall always
assume that the matrices under consideration are of the same order.
We shall now introduce certain new symbols which will be useful
in future. The symbol || a || will denote a matrix, all elements of
which are equal to the number a . The symbol | X | will denote a
matrix the elements of which are equal to the moduli of the elements
of the matrix X , i.e.

{|A |},.fc= |{ A } ,J . (23)


86] POWER SERIES OF ONE MATRIX 326

If a certain matrix Y has positive elements which are greater than


the elements of the matrix | X |, we shall write this down in the form
of an inequality
m < r .
In other words, this inequality is equivalent to the following system
of n2 inequalities:
I {-X}/fc| < (®> ^ = 1> 2, .. . , n).
Consider the infinite series the terms of which are the matrices:
•Zj + 2 2 + . . .
This series is said to be convergent if the sum of its first n terms
(matrices) tends to a definite limiting matrix Z. Such a matrix is the
sum of the series
Y = Zl + Z2 + . . . (24)
This equation (24) is evidently equivalent to the following n2
equations:
{Z}ik = {Zi}ik + {^ 2)1* + ■■■ {i, k = 1, 2...........n). (25)
Let us suppose that in the neighbourhood of the matrix A lie
all the matrices X which satisfy the condition
|X — .4 I < ||e ||, (26)
where g is a given positive number. The inequality (26) is equivalent
to the following n2 inequalities:
| { * - ,! } « | < e.

Of fundamental importance in the determination of functions of


matrices in future will be power series of these matrices and we shall
now consider these series.

86. Power series of one matrix. A power series of a single matrix


has the form:
a o + &i (X — a) + ®2 (-X^ — a)2 + • • • . (27)

where ak and a are given numbers. To simplify our notation we shall


assume in future that a = 0. Instead of series (27) we shall then have
a series
a0 + a1X + a2X- + • - • (28)
326 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [86

In accordance with the multiplication law of matrices we have:

{*2},* = 2 { x U i x U
s= 1
and, in general

{ * ")« = , , 2 , M I U - P U - P U
JuJnt • •

where the summation includes all symbols jk independently, from 1 to


n. Hence the elements of the matrix given by the sum of the series
(28) can be expressed by the series

°(A* + (29)
Jit ]%t •*•Jm—i
where 6,* denotes a number determined by the formula
„ 0 when i # k
0:u = (30)
1 when i = k .
The last circumstance is directly due to the fact that the constant
term of series (28) is the number a0, i.e. it is a diagonal matrix, all diag­
onal elements of which are equal to a0. Formula (29) shows that series
(28) is equivalent to n2 power series of a special kind with n2 variables
{X}**. Notice that when m = 1 the terms of the sum in formula (29)
have the form ax {-5T}ik and the inside sum disappears.
Consider now the problem of the convergence of series (28). We shall
consider, first of all, its absolute convergence, i.e. together with series
(28) we shall also consider the series
I°o I + I°i II X | + | a21| X |2 + . . . (31)
or its corresponding n2 series

\a0 \dik+ ^ \ a m\ ^ (32)


m_1 Aiin — t j m —l
If these series converge then series (29) will converge all the more,
i.e. the convergence of series (31) guarantees the convergence of
series (28) and in this case series (28) will be absolutely convergent.
In agreement with the definition of the matrix | X | we have:
{ |X |} ift= |{ X } ,.ft|,
i.e. expression (32) is obtained from (29) by replacing all numbers
by their moduli.
86] POWER SERIES OP ONE MATRIX 327
We shall now explain the sufficient condition for absolute con­
vergence of series (28). Construct a power series for the complex
variable
o0 + a^z + a2z2 + . . . , (33)
and let the radius of convergence of this series be equal to n q , where
n is the order of our matrix and g is a positive number. We know
from [14] that we have in this case the following inequality for the
coefficients of series (33):
i ^ M
am I („g _ £)m ’
where e is any small fixed positive number and M is a positive
number which depends on the choice of e . Let us now take the matrix
|| b ]|, where 6 is a certain number, and determine its positive integral
powers
{ll&ll2 },* = &&+ && + + bb = t& , II 6 II1 = II n6* ||,
and generally
|| 6 ||»" = || n " -16" ||. (35)
Let us now suppose that b = qt > 0 and take a certain matrix
X which satisfies the condition | X [ < 11 ||. In this case we obviously
have:
| X |m < || Pi II* i.e. I X \m < II 7im- 1er,||.
As a result of the inequality (34)

If Qi < q, then by taking e sufficiently small we have


0 < tigne—i e < 1,

and in this case series (31) will be convergent and series (28) will be
absolutely convergent. If the radius of convergence of series (33)
is infinite then it is said that the sum of this series is an integral
function of z . It follows from what was said above that in this case
series (28) will also be absolutely convergent for any matrix X .
We thus obtain the following theorem.
T h e o re m . I f th e r a d i u s o f c o n v e r g e n c e o f s e r i e s (33) i s e q u a l to n q ,
th e n s e r i e s (28) w i ll be a b s o lu te ly c o n v e rg e n t f o r a ll m a tr ic e s s itu a te d
i n th e n e ig h b o u r h o o d o f th e o r i g i n .

(36)
328 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [86

I f series (33) determines an integral function, then series (28) will


be absolutely convergent for all matrices.
When considering the absolute convergency of series (28) in the
domain (36) we can say that the sum of this series f(X) will be a
regular function in the above domain.
Consider, for example, the exponential function of the matrix:

eX = 1 + 4 + T T + ••• (3?i)
The corresponding power series (33) has an infinite radius of con­
vergence and, consequently, series (37J will be absolutely convergent
for any matrix X, or, as is generally said, it will be an integral function
of this matrix.
Consider also the exponential function for any base
a X = e X log a—! + jf h p + + ^ (372)

where log a is a certain fixed value of the logarithm of the complex


number a. Function (372) is also an integral function of the matrix X .
We shall now explain the uniqueness of an expansion into a power
series. Suppose that we are given two power series

2 aDd 2 bmXm>
m =0 m=0

each one of which is absolutely convergent in the neighbourhood of


(36); we also suppose that in this neighbourhood the sums of the
series coincide, i.e.
2 amXm= 2 amXm■
m =0 m=0
We shall now prove that in this case the coefficients a'm must
coincide with the coefficients am. Notice that condition (36) is also
satisfied by the diagonal matrices
X = z = [z,z, . . . ,z],
in which | z | < g. Hence the above assumption gives:2

2 amZm= 2 a'mZm (|z|<e)-


m-o m=0
We know that the expansion of a function of a complex variable
into a power series is unique in any particular circle and, consequently,
a'm — am. We thus obtain the following theorem.
87] MULTIPLICATION OP POWER SERIES. CONVERSION OP POWER SERIES 329

U n iq u e n e s s T I f two power series, which are absolutely


heorem .
convergent in a certain neighbourhood (36), have in this neighbourhood
the same sum, then all coefficients of these two series will be the same.
If we use the self-explanatory formula

(iS X S -1)* = S X ^ - 1,

then for the function f(X), given by the power series (28) or (27)
we have, as in [Hip 44], the equation:
f { S X8 ~ 1) = S f ( X ) £ -!.

87. Multiplication of power series. Conversion of power series.


Suppose that we are given two power series

/ 1( i ) = /27 i = 0 a" xm and A (A ) = 2m = 0 b" xm’

which are absolutely convergent in the neighbourhood (36). We can


construct a new matrix, which is obtained by multiplying their sums

Y = f 2(X)f1(X).

The elements of this matrix will be determined by the formulae:

{F }« = j £ { M * ) M / i ( * ) U (38)
S= 1

where
{ /i( X ) } sfc = a0<5sfc+ ^ a m 2
/7I —1 ji ..

{f2(X)}is = b0sia+ 2 K 2 {A M A W - - -
TTl—l j i t . . . , ju t—i

Owing to the absolute convergence of the above two series we


can multiply them term by term, so that for elements of the matrix Y
we have, from (38):

{Y = aoMifc + 2 (a^m + + •••


m=1
. . . + a mb0) 2
330 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [87
and the matrix itself may be written in the form:

Y — a0b0dik + ^ (aobm + + • • • + anPo) .


m=»1
It follows from this that absolutely convergent power series of matrices
can be multiplied like the power series of numerical variables and the
product does not depend on the relative position of factors.
We shall now construct a function which is the inverse of the given
function f(X), determined by a certain power series
Y = / (X ) = o0 + axX + a2X 2 + . . . , (39)
and we assume that in this series the coefficient ax is not zero.
Consider the power series of the usual complex variable
w = aa + a2z + a2z2 -f . . . (40)
We know that there is only one power series which satisfies the
condition ax # 0:
z = Cl{w — a0) + c2 (w — a„)2 + . . . (41)
which determines the inverse function of function (40) in a certain
neighbourhood | w — a0 | < ng. If we substitute series (41) in the
right-hand side of formula (40) we obtain:
72
w = ao + ax ^ Ck (w _ a<))ft _|_ a2 ^ ck iw ~ ao)k +
k= 1 Lft=l
if we now raise the series to their corresponding powers according to
the multiplication law for series, and isolate similar terms by grouping
together terms with equal powers of (w —a0), we arrive at the identity
w = w. If in all the above calculations we replace z by the matrix
X and w by the matrix Y , then all calculations involving power series
of matrices in powers of the differences (F —a0) will be the same as
those in operations with power series of numerical variables (w — a0)
and, consequently, the results will be the same, i.e. when al ^ 0 the
power series (39), which is determined in the neighbourhood X = 0,
permits a unique conversion of the kind

x = 2 c*(Y ~ a°)k’ (42)


k=l
and this latter series will be absolutely convergent in the neighbourhood
87] MULTIPLICATION OF POWER SERIES. CONVERSION OF POWER SERIES 331
This neighbourhood must be determined by the radius of conver­
gence of series (41).
The coincidence of the above formal operations between power series
containing ( Y — a0)k and power series containing ( w — a0)k, is due
to the fact that the above matrix series contains numerical elements
and one power matrix Y — o0. In fact numbers can be commuted in
any matrix and powers of one and the same matrix can also be com­
muted. This gives us the above coincidence of formal operations. For
example, for any whole positive k we can apply to the term

(Y - Go)*
Newton’s binomial expansion. But, in general, this formula can no
longer be applied to the term

(tfi + U 2)*,
where U 1 and U 2 are different matrices.
We shall apply the above remarks to the series
2 1 . 2 1 Z* I
w ~ e = 1 + t t + ”2r + • • •

The inversion of this series gives log to and gives us, as we know,
the power series

log = log [1 + (w — 1)] = — {w ~ 1)8 + • .. ,

which converges in the circle \ w — 1 | < 1.


Thus the inversion of the exponential function

F = e x = l + -IT + — + . . .
enables us to determine the logarithm of the matrix in the form of
the power series

= (44)

which converges absolutely in the region

|F-1|< n
The equation of the matrix
ex = Y (46)
332 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [88

with respect to X has an infinite number of solutions for a given Y.


Series (44) gives one of the solutions of this equation, viz. it gives that
solution which is a regular function of Y in the neighbourhood of a
unitary matrix and becomes a zero matrix when Y = 1. The problem
of the remaining solutions of the equation, both in the neighbourhood
of a unitary matrix and outside that neighbourhood is connected
with the analytic continuation of series (44) or, which comes to the
same thing, with the analytic continuation of n2 power series, to which
series (44) is equivalent. We shall deal with this problem later.
Let us now determine the power function of a matrix. It can be
determined by means of the logarithm of a matrix as follows:
X a = ealogX (47)
If we have a numerical variable z
z° = ea Iog z
then substituting ologz in the expansion of the exponential function

logz = log[l + (2 )]
- 1 = ~ - + ^- ^
3 • •>
we obtain a power series in the form:

2“ = [ l + ( z - 1 ) ] “ = 1 + - - ( * - 1) + ^ V ± ) (Z- 1 ) 2 +
which converges when | z — 1 | < 1. Bearing in mind the coincidence
of formal operations in power series with one matrix described above
we obtain:
+ (48)

and this expansion will be absolutely convergent in the region

(49)

88. Further investigations of convergence. We have already


mentioned above that the power series (28) is equivalent to the n2 series
(29) of the variables consider the inner sum of series (29):
(50)
i n • • •! j m - l
88] FU RTH ER INVESTIGATIONS OF CONVERGENCE 333
Grouping together similar terms in the above sum we can obtain
series (29) in the form of the usual power series of n2 variables {.£},•*.
If we replace all terms in the sum (50) by their moduli and the numbers
am by | am \, then this is evidently equivalent to substituting all terms
by their moduli in the above power series. It follows that if the series
(29) are converted into the usual power series of the variables { X then
the absolute convergence of these series is equivalent to the absolute con­
vergence of series (32), i.e. it is equivalent to the absolute convergence
of series (28).
In general, the convergence of series (28) in the widest sense of this
word implies the existence of a limit in the form of a succession of
matrices
a-o + 2 a"'Xm (51)
m=1
where I increases indefinitely. The addition of a term m = I + 1 to the
sum (51) is equivalent to adding to the sums

fc+
a<A am ^ {X },7l { X }hh . . . {X |jm —ik (52)
tn = l j i, . . . i jo t—i
(i, k = l , 2, . . . , n)
a homogeneous polynomial with respect to
«/+l 2 {X}iVl{Z}yu, . . . { Z } M (53)
i ...... h
of the power (I + 1).
Hence the convergence of series (28) in the above sense is equi­
valent to the convergence of the n2 series (29) in which all terms of the
type (53) are collected together. We shall consider, first of all, the con­
vergence of series (28) in a special domain, determined by the inequality

\X\ <A, (54)

where A is a given matrix with positive elements.The inequality (54)


is equivalent to n2 inequalities

|{ Z } f* |< { 4 } ttt (55)

which determine n2 concentric circles with centre the origin, for the
complex variables {Z}(fc. We can therefore assume that series (28)
converges in the domain (54). Let 6 be any positive number less
334 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [88

than unity. It is given that series (28) will converge when X = QA,
i.e. n 2 equations of the type will converge

aAk + 2 Qma™ 2 {A U {A U • • • M )j~ * ■


ill

The latter series can be regarded as power series of Q and we can


therefore say that their convergence will be absolute, i.e. the following
series will also converge:

\< h » o '\+ 2
*
0mK I .
jli ••
2

all terms of which are positive. We thus see that the series (29) will be
absolutely convergent for the matrix Q A . They will be all the more
convergent for all matrices which satisfy the condition | X | < QA.
Remembering that we can choose 0 as near to unity as we like, we can
say that the series (29) will be absolutely convergent for all matrices
in the domain (54). At the same time the series (28) will also be abso­
lutely convergent. We thus have the following theorem.
T h e o re m . I f s e r i e s (28) c o n v e r g e s i n a d o m a i n o f th e t y p e (54) th e n
i t w i l l c o n v e r g e a b s o l u t e l y i n t h i s d o m a i n o r , i n o th e r w o r d s , th e n 2 p o w e r
(29) w i l l c o n v e r g e a b s o l u t e l y i n th e c o n c e n tr ic c ir c l e s (55).
se r ie s
Until now we investigated the convergence of a power series in
special domains which were determined by the inequality (54) or by the
inequality (36), which is a particular case of inequality (54). We shall
now consider the general case of convergence of a power series and
assume that the matrix X can be converted into the purely diagonal
form as we did for unitary matrices and also for matrices, all
the characteristic zeros of which were different. Our condition may also
be formulated as follows: we shall only consider matrices with simple
elements. Such matrices can be written in the form [III^ 27]:
X = S [Xv X2, . . . , Xn] S - \ (56)
where S is a certain matrix the determinant of which is not zero and
?.(are the characteristic zeros of the matrix X . To simplify notation we
shall introduce symbols for segments of the series:

ft (X) = a0 + 2 amXm; ft (z) = a0 + 2 a mzm •


m =1 m =1

and the sums of the series we shall denote, as before, by


f(X) and f(z).
88 ] FURTHER INVESTIGATIONS OF CONVERGENCE 335

Substituting expressions (56) in f x( X ) we have [IIIj^, 44]:

/, (X) = a 0 + S ( £ OmI K , # ])
m~l
or
fi (X) = 8 [/, (AJ; /, (A2), (A„)] 5 ~ i. (57)
If all characteristic zeros A( lie inside the circle of convergence of
series (33) then expression (57) will have a definite limit, viz.
/(Z ) = 5[/(A1),/(A2), . . . , / ( A n)]5 -i, (58)
and, consequently, the series (28) will converge. Let us now suppose
that at least one of the characteristic zeros Ax lies outside the circle
of convergence of series (33) and we will show that (57) cannot tend
to a definite limit. In fact, we can rewrite equation (57) in the follow­
ing form:
[/«(*l). / f W ........ M*n)] = 5 -1 /,(X )S .
If fx(X) tends to a limit, then the left-hand side of the above equation
would also have a limit, i.e. all elements of the diagonal matrix on the
left would have a definite limit. However, this cannot be so for the
element /i(Ax), since Aj lies outside the circle of convergence of the
series (33). We thus obtain the following theorem.
T h e o rem . T h e p o w e r s e r i e s ( 2 8 ) c o n v e r g e s i f a l l th e c h a r a c t e r is ti c
z e r o s o f th e m a t r i x X l i e i n s i d e th e c ir c le o f c o n v e r g e n c e o f th e s e r i e s
(33) a n d i t d i v e r g e s i f a t l e a s t o n e o f th e s e z e r o s l i e s o u t s i d e th e a b o v e
c ir c le .
We proved this theorem when the matrix X had simple elements,
i.e. when it was given in the form (56). The proof can be extended to
include the general case but we do not intend to do this here.
Let us now consider the general case of absolute convergence, i.e.
the convergence of the series (31).
Bearing in mind that a power series of the usual complex variable
is absolutely convergent inside its circle of convergence we can say
that the radius of convergence of the series

m=0
will coincide with the radius of convergence of the series (33). Applying
the theorem which we have just proved to the series (31) we obtain
the following theorem for absolute convergence.
336 FUNCTIONS OF SEVERAL 7 A~RTABT.ES AND MATRIX FUNCTIONS [88

T h e o re m . The se r ie s (28) w i l l c o n v e r g e a b s o l u t e l y i f a l l c h a r a c t e r is ti c
z e r o s o f th e m a t r i x \ X \ l ie i n s i d e th e c ir c le o f c o n v e r g e n c e o f th e s e r i e s (33)
a n d i t w i l l n o t c o n v e r g e a b s o l u t e l y w h e n a t l e a s t o n e o f th e s e c h a r a c t e r is ti c
z e r o s l i e s o u t s i d e th e a b o v e c ir c le .
It follows from what was said at the beginning of this paragraph
that the absolute convergence of the series (28) implies that this series
will also be convergent in the general sense. Using this circumstance
it can readily be shown that the maximum modulus of the character­
istic zeros of the matrix | X | is not less than the maximum modulus
of the characteristic zeros of the matrix X . In fact, let q1 be the
maximum modulus of the characteristic zeros of the matrix | X | and
p2 be have a similar meaning X . If we suppose that q2 > Qv we will
show that this brings us to a contradiction. Select in series (33) the
coefficients a m in such a way that this series has a radius of con­
vergence equal to q, where q satisfies the condition g2 > g >
Such will be, for example, the power series obtained by expanding
the fraction
1

e
As a result of the above theorems series (31) will, in this case, con­
verge and series (28) diverge; this contradicts the fact that absolute
convergence also implies ordinary convergence.
Let us turn to formula (58). It shows that when the matrix X has
characteristic zeros Xt and all its elements are simple then the matrix
f ( X ) , determined by a convergent power series, will have character­
istic zeros /(Aj) and its elements will also be simple. This property,
with a certain clause, can be extended to include elements which are
not simple, viz. the following proposition holds: i f th e e le m e n ts o f th e
m a tr ix X are
(A -A ,)* , (A -A J ft, .... (A — As ) p s,

t h e n th e e le m e n ts o f th e m a t r i x f ( X ) , g i v e n b y a p o w e r s e r i e s , w i l l be

[* - / ( * i ) ] ft. [ W W ] Pa........ [ * - / & ) ] * ,


p r o v i d i n g th e d e r i v a t i v e s f ' ( X k) d o n o t v a n i s h . Formula (58) may also be
used for the analytic continuation of the function f ( X ) , when given
by a power series. Suppose that this series converges absolutely in a
domain of the type (54); take a certain matrix -3T0in this domain, and
continuously vary the elements of this matrix by a definite law. The
characteristic zeros A/ will vary continuously at the same time.
89] INTERPOLATION POLYNOMIALS 337

Suppose that the elements of the matrix S which appears in formula


(56), are doing likewise. The analytic continuation of the matrix f(X)
thus involves, according to formula (58), the analytic continuation
of the function f{X) of one complex variable.
The analytic continuation described above is very inconvenient
owing to the fact that formula (58) contains the matrix S, which has
no definite connection with the given matrix X. In fact we already
saw on the example of Hermitian matrices that the matrix S can be
selected in different ways. In certain cases the continuation described
above will not coincide with the analytic continuation of n2 series (29).
Below we shall explain in greater detail the problem of analytic con­
tinuation for which we shall use a new important formula. To derive
some auxiliary propositions for the proof of this formula we have to
deal first with some simple formulae connected with interpolation.

89. Interpolation polynomials. The fundamental and simplest


problem of interpolation consists of the following: we are required to
construct a polynomial expression with a power not higher than
(n — 1), which would take given values at n points in the plane of the
complex variable. Suppose that at the points zk (1c = 1, 2, . . . , n) it
takes the value wk. Notice, first of all, that there can only be one such
polynomial. In fact, we know [I, 185] that two polynomials with powers
not higher than (n — 1) are identical if their values at n different
points coincide. The solution of the problem of interpolation can be
given by the following simple formula:
P (~\ _ (2 - Zi) (Z — Z j) ■• ■ (g — Zft-1) (z - 2ft 11) • • • (g - Zfl)
"-1 ' ' (2ft - * l) (2ft - **) • • ■ (2ft - Z f t - l ) (2ft - Zfc+l) • • ■(2 - Zn) k

It can be seen directly that the expression on the right-hand side


is a polynomial of z with a power not higher than (n — 1). If we
put, for example, z = zv then on the right-hand side all terms, except
the first, vanish and the fraction in the first term will obviously be
unity, i.e. Pn_1(z1) = w1 and, similarly, Pn_1(zft) = wk.
If f(z) is a regular function in a certain domain and the points zk
belong to this domain then the formula
" ( Z - Z , ) . . . ( Z - Z A_1) ( Z - Z f t . 1) (2 ~ 2|l) (60)
^ (2ft - 2,)... (2* - Zft_,) (2 - z*+1) (2ft 2/j) /(2ft)
gives that unique polynomial with a power not higher than (n — 1),
the values of which at points zk coincide with values of the function
338 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [89
f(z). This polynomial is usually known as Lagrange’s interpolation
polynomial for points zk and formula (60) is known as Lagrange’s
interpolation formula.
The general polynomial of order (n — 1)
O0 + Qi2 + . . . + an-l Z"- 1
has n parameters as. In Lagrange’s formula these parameters are
determined from n conditions, viz. from the conditions that at the
points zk the values of the polynomial should be equal to f(zk). Let us
now formulate the problem in a more general way. Suppose again that
f(z) is regular in a certain domain and that we are given j points zv z2,
. . . , Zj inside this domain; we are required to construct a polynomial
of order not higher than (n — 1), the values of which at the points
zk must coincide with the values of all its derivatives up to the degree
(pk — 1), with analogous values for the function f(z), i.e. in this case
we have for the polynomial P(z) the following condition
p (Zk) = / {Zk)\ • . •; PlP*-» (zfc) = /(PJT-1) (zk) ( k = 1, 2, , j),

where we assume that Pi + p2 + • • • + Pj = n, so that the total


number of conditions will again be equal to n. It can readily be shown,
as above, that such a polynomial must be unique. In fact, if there
were two such polynomials, then their difference would be a poly­
nomial of order not higher than (ra — 1) and zk zeros of multipli­
city pk, i.e. we would find that this polynomial with a power not higher
than (n — 1) would have n zeros. Hence even with this new wider
formulation the interpolation problem can only have one solution.
We shall give the method by which the required interpolation poly­
nomial can be constructed. Construct for this purpose the polynomial
of degree n:
v (z) = (2 - Z ,)* (z - 22)P> • • • (2 - Zj)PJ
and the function
?(’> = £ § • <6 1 >

This latter function has poles at points zk of order not greater


than pk. The sum of the infinite parts of this function with respect
to the above poles can be represented by a certain fraction in which
the degree of the numerator is lower than the degree of the denomi­
nator and where the denominator has the form
(z - g j * (z - . . . (z - Zj)V,
90] CAYLEY’S IDENTITY AND SYLVESTER'S FORMULA 339

where the integers q k are not greater than the integers p k. Multiplying
the numerator and denominator by one and the same factor we
can reduce the above sum of the infinite parts of the function 95(2)
to the form:

P/1-1 (g)
P(Z) ’

where Pn_j(2) is a certain polynomial of degree not higher than (n — 1).


After this formula (61) can be rewritten in the form:
m fii-i (2) + ft) (2),
p( z) p(z)

where co(z ) is a certain function which is regular in the whole domain,


including the points z k. Rewrite the above formula as follows:

f { z ) =P„_ 1{z)+p(z)co{z). (62)

The second term on the right-hand side in the neighbourhood of the


point zk can be written as a product of (2 — pkzk) and a certain function
which is regular at the point zk, i.e. the second term on the right-hand
side vanishes together with the derivatives at the point zk up to the
order (p k — 1). Thus at these points the value of the polynomial
P„_ 1(2) and the derivatives up to the order (pk — 1) coincide with the
corresponding values of the function f(z), i.e. the polynomial Pn_x(2)
is the required interpolation polynomial. In future we shall sometimes
denote it by h ( z , zv . . . , zn). If all values zk are different then this is
simply Lagrange’s polynomial. If, however, there are equal values
among the values zk, e.g. if a certain number zk appears pk times then
at the point zk the values of the polynomial and its derivatives up
to the order (pk — 1) coincide with the corresponding values of the
function f(z). When n = 2 and z1 # z2 we have

H z ', z v *a)=iTZ1T —z8/(*i) + z2


——tZ/1 ( 4
and when zx = z2

h{ z - , z v 21) = H z 1) + L ^ 1 f ' { z 1) -

90. Cayley’s identity and Sylvester’s formula. Let X be a certain


matrix and
D { X - XI) = 0 (63)
340 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [90

be its characteristic equation where D ( Y ) denotes the determinant of


the matrix Y . Denote the zeros of this equation by Xy, X2, . . . , Xn . The
left-hand side can be written in the form
(_ l)n (An + n.A"-1 + . . . + a n_ ,X + o„) = ( - 1)>(A ), (64)
where the a k are simply expressed by elements of the matrix X or
by zeros of the equation (63). Thus, for example
®1 = — (^ 1 + ^2 + • • • + K) > a2 = ^ 1^2 + ^ 1^3 + • • • + ^ n —l ^ n •

The expressions a k are examples of a numerical function or of a


matrix, i.e. they represent a function which takes a numerical value
for the given matrix X . We have already considered such functions
[IIIj, 27]. Let us recall that (—1)na n is the determinant of the
matrix and a x is the trace of the matrix equal to the sum of its diagonal
elements.
Cayley’s identity consists of the following: if in the polynomial
y>(X) = Xn -f- a x + . . . + a n we replace the letter Aby the matrix
X , then we obtain a zero matrix as the result, i.e. the following iden­
tity holds:
y > ( X ) = X n + a 1X n- i + . . . + a n = 0. (65)
Let us suppose that the characteristic roots Xk are different or, more
generally, that the matrix X can be represented in the form:
X = S [ X V X2, . . . , Xn] S - K

At the same time, as we saw in [III1( 44] we have:


V ( Z ) = 5 [ v ( A 1), y> (A2), . . . , y (;.„)] 5-1.
But the zeros Xk are zeros of the polynomial ip(z) and therefore:
y>{X) = 5[0, 0...........0 ]5 -!.
In the centre stands a diagonal matrix in which all elements on the
main diagonal are zeros. Such a matrix consists of zeros throughout
and, in general, the right-hand side of the above equation is also a
matrix which consists of zeros, i.e. formula (65) does, in fact, apply.
It is not difficult to extend the proof of this identity to include the
general case when, by using the method of limit transition, we con­
sider, to start with, matrices with different characteristic zeros.
Consider now a certain function f ( X ) which is determined in the
domain
\ X \ < A (66)
90] CAYLEY'S IDENTITY AND SYLVESTER’S FORMULA 341

by an absolutely convergent power series


/ (X) = <z0 + a xX + a 2X 2 + ... (67)
Take a certain matrix X belonging to (66) and assume that the
characteristic zeros Xk are different. We now suppose that in the iden­
tity (62)
P (z) = (2 - *i) (z - h ) ■• • (z - **) = V (z) •
We then have the identity
f { z ) = P n_ 1 {z) +y(z)co(z)> (68)
where P„_1(z) is Lagrange’s interpolation polynomial for the points Xk.
Formula (68) obviously remains an identity if we replace the variable
z by the matrix X , for in the product on the right-hand side there is
only one matrix X , the powers of which commute and therefore a
product can be constructed in the same way as before when z is
replaced by the matrix X . In this case the polynomial y ( X ) coincides
with the polynomial (65) and from Cayley’s identity we have
f(X) = P n_ 1 ( X ) .

In the expanded form we have in the domain (66) for any matrix X
with different characteristic zeros:
,,T. " , (X — A.) . . . (X — Aa„ ) (X - Xk+l) . . . (X - ,,, ,
(69)
n (**-*.)•••(*»- V i) (** - **+.) (h - K) 1 ( k>
Formula (69), which is known S y l v e s t e r ’s f o r m u l a g i v e s th e i n f i n i t e
(67) i n th e f o r m o f a p o l y n o m i a l m a t r i x and the infinite power
se r ie s
series enter the formula through the expressions f ( A k), which are ordi­
nary power series of the complex variable.
If among the characteristic zeros Xk of the matrix X there are similar
values then we shall have on the right-hand side of formula (69)
not Lagrange’s polynomial but the more general interpolation poly­
nomial which we mentioned in the previous section; we have a
similar representation for series (67) in the form of a polynomial of
the matrix
f ( X ) = h ( X ; Xly A2, . . . , A„). (70)
For a matrix of the second order this gives us, when Ax # A2:

= + (71)
and when Ax =
f ( X ) = f { X 1) + ( X - X 1) f (A,). (72)
342 p ra o n o N S o f s e v e r a l v a r ia b l e s a n d m a t r ix p u n c t io n s [9 1

Thus, for example, for an exponential matrix of the second order,


when Xx / X2, we have:
e ca e\(73)
+
- A 2
Note that the general formula (70) which should be applied when
there are similar zeros among the zeros Xk, can be obtained from
formula (69) by the method of limit transitions, when certain groups
of zeros Xk tend to a common value.

91. Analytic continuation. Formula (67) which determines a regular


function f ( X ) in the domain (66) is equivalent to the n 2 power series
(29) which are absolutely convergent in the concentric circles:
| { X} rt| < { 4 } (*.
By performing analytic continuation of these n 2 power series we
determine the matrix f ( X ) in a wider d o m a in a l th e w h o le s e t o f m a t r i c e s
o b t a in e d a s a r e s u l t o f t h i s a n a l y t i c c o n t i n u a t i o n w i l l d e te r m i n e th e a n a l y t i c
(67) i n d o m a i n
f u n c t i o n f ( X ) a s g i v e n b y i t s o r i g i n a l e le m e n t, v i z . b y s e r i e s
(66). Consider Sylvester's formula. When the elements of the matrix X
vary according to a definite law, its characteristic zeros Xk will also vary
continuously in a definite manner and, according to (69), the analytic
continuation of the function f ( X ) will, in fact, involve the analytic con­
tinuation of the function /(z) of one complex variable. If in the course
of this analytic continuation some of the zeros Xk coincide then instead
of formula (69) we must turn to formula (70). If the function of the
complex variable /(z) remains single-valued after analytic continuation
then the only difficulty caused by using Sylvester’s formula for analytic
continuation is due to those matrices X , the characteristic zeros of
which contain values of X which are singularities of the function /(z).
Thus, for example, on analytic continuation of the function which is
determined in the neighbourhood of zero by the series
f(X) = ..., I + X + X 2+

those matrices will be singularities, the characteristic zeros of which


include at least one which is equal to unity. It can be proved that in
the case under consideration the analytic continuation by Sylvester’s
formula is fully equivalent to the analytic continuation of the above
n 2 power series and that it gives all values of the analytic function.
Let us now consider the case when /(z) is an analytic function which
becomes many-valued in the course of its analytic continuation. In
this case the function /(z), as we saw above, can be single-valued not in
91J ANALYTIC CONTINUATION 343
the ordinary plane of the complex variable but on a certain Riemann
surface R with several sheets; this shows that our functions are many­
valued. When the elements of the matrix X vary continuously its
characteristic zeros will do likewise on the above Riemann surface R .
If we are determining the values of our analytic functions f ( X ) for
a certain particular value of the matrix X 0 then we must know not
only the matrix X0 but also the process of analytic continuation
of the function f ( X ) which produced this matrix X 0, by starting with
a certain matrix in the domain (66), in which the function was given
by series (67). In other words, we must know not only the matrix X 0
but also the path of the analytic continuation which produces this
matrix. Let us suppose that this path was such that Sylvester’s formula
was used throughout the analytic continuation, but that when X tends
to -3l 0 two characteristic zeros X2 and X2 tend to coincide, i.e. they tend
to one and the same complex value A0, although they lie on different
sheets of the Riemann surface R . In the limit, the characteristic
zeros X2 and X2 of the matrix X 0 coincide, but the function f ( z )
in the neighbourhood of this common value A0 will be determined by
two different Taylor’s series, since the corresponding points lie on diffe­
rent sheets of the Riemann surface R . In general in the limit although
Xx = X2 we have f ( X 2) ^ f ( X2). In this case the Lagrange-Sylvester
formula is devoid of any meaning and we assume that the matrix X 0
along the above path of analytic continuation is a singularity of the
function f ( X ) . It may, of course, sometimes happen that f ( X1) = f ( X ^
in the limit and the difference becomes apparent only in certain deriv­
atives, i.e. for a certain s we have / (s)(Ax) ^ f s){X2), although Xt = X2.
In this case by moving the characteristic zero away from the common
value X 0 by as little as we please, we can obtain such a value X 0 on
different sheets, for which f ( Xf ) # f ( X 2) although Xx = X2 \ we assume,
as before, that the matrix X 0 is a singularity of the analytic func­
tion f ( X ) . Hence i n th e c a s e o f a m a n y - v a l u e d f u n c t i o n c e r ta i n m a t r i ­
c e s m u s t b e a d d e d to s i n g u l a r m a t r i c e s o f th e a n a l y t i c f u n c t i o n f { X )
w h ic h a r e d e te r m i n e d b y th o s e p a t h s o f a n a l y t i c c o n t i n u a t i o n , i n w h i c h
s i m i l a r c h a r a c t e r is ti c z e r o s o f th e m a t r i x X c o r r e s p o n d to d iffe r e n t a n a ­
l y t i c e le m e n ts o f th e f u n c t i o n f ( z ) .
We will not explain in greater detail the above characteristics of a
many-valued analytic function f ( X ) . Let us investigate only one simple
special case. Consider the matrix

X = S [Xv X2 ........ Xn] S -i,


344 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [91

where S is a definite matrix, the determinant of which is not zero and


whose zeros A* are different. We suppose that this matrix lies in the
domain (6 6 ) where our function f(X) is determined by series (67). We
shall now continuously vary the matrix X by a definite law, viz. when
fixing S the zeros Xk will vary in such a way that they will always be
different and will never coincide with the singularities of the function
f(z)\ also, in the limit, all zeros Xk will tend to the common value A0,
but will lie on different sheets of the Riemann surface R of the function
f(z). Suppose, for simplicity, that on these sheets the values of the
function f(z) at the point A0 differ in pairs. At the limit we have the
matrix
X 0 = S [A0, A0, . . . , Au] iS—1 = A0,

i.e. simply the zero A0. The values of the function are determined in
the original domain (6 6 ) by the formula

£[/<;.,), f(k 2)------ ;(;.„)] s - i . (74)


The problem involves the analytic continuation of f(Ak) for a fixed S.
In this case we obtain a definite limit for the function, which must
be equal to:
S f/V .........................................................(75)
where fik denotes the value of the analytic function /(A0) on that sheet
of the Riemann surface on which the zero Allies. Notice that the final
result (75) depends on the choice of the matrix S. By changing the
elements of the matrix S as little as we please we also change the
final result (75), which can be readily shown by bearing in mind that
fit / p.k when i =£ k. The fixing of the matrix S involves the choice
of a definite law of variation of the matrix X in the course of analytic
continuation. In doing this we obtain a definite limit for the function
f(X) at the singularity X = X 0. By slightly altering the elements of S
we arrive at a different limit. It also follows from this that series (29)
cannot be analytically continued over the point X = X 0. This singu­
larity is, of course, connected with the path of analytic continuation
which brings us to that point. It can be shown generally that the
singularities of f(X) as determined above for the case of the many­
valued function f(z), will also be singularities in the analytic continu­
ation of n2 power series (29) and vice versa. In other words, the use
of Sylvester’s formula in analytic continuation is equivalent to the
analytic continuation of tl2 series (29). Those matrices will be singular
92] EXAMPLES OP MANY-VALUED FUNCTIONS 345

in that continuation among the characteristic zeros of which there are


singular points of f(z) and also matrices in which similar characteristic
zeros lie on different sheets of the Riemann surface of the function f(z).

92. Examples of many-valued functions. The logarithm of the matrix


Y = log X (76)
is, by definition, the solution of the equation
er = X. (77)
Let us suppose th at the matrix X has simple elements:
X = S [Au Aj........ A„] S-i, (78)
and th at none of the zeros Afc is zero. I t can readily be seen th at we obtain
the solution of equation (77) by assuming
Y = S [log A,, log A,........log A„] S ' 1, (79)
for, as we saw above
eY = S [el«« h , e'«« * « , . . . , e'»» *»] S ' 1 = S [A„ A..........A„] S ' 1,
i.e. the matrix (79) satisfies equation (77). We can in formula (79) take any value
of log X k so th at we have
Y = S [log A, + 2jtrji.........log A„ + 2nrni~] S ' 1, (80)
where log Xk always denotes the principal value of the logarithm
— n < arg X k < n,
and rk are any integers.
The many-valuedness of formula (80) is due to two causes. In the first place
it is due to the freedom of choice of the numbers rk and, secondly, to a certain
arbitrariness of the matrix S, which enters formula (78), when the matrix X
is fixed. If, when A( = Xk, we have r, = rk then the values of log X are said
to be regular. We will show that a regular value of a logarithm is fully deter­
mined by the choice of the integers rk so that it is wholly independent of the
choice of the matrix S. In fact, let /i1( (i2l ftj be different characteristic
zeros of the matrix X and rv r2, . . . , rj be the corresponding integers in formula
(80). Construct Lagrange’s interpolation polynomial with a degree not higher
than (j — 1), with the condition
P(fik) = logfik + 2nrki (k = 1, 2, , j).
According to formula (78) we have:
P ( X ) = ,S[P(AI), . . . , P(A„)]S ' 1,
or
P (X) = S [log A, + 2jir,i . .. , log A„ + 2nr„i] S ' 1,
i.e. P(X) = Y, from which it follows directly that the assumed value of the
logarithm does not depend on the choice of the matrix S, for this latter matrix
does not appear in the construction of the polynomial P(X).
346 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [92

Applying Lagrange’s formula we have:

. ^ ( X - A 1) . . . ( X - A ft_1) ( X - A ifc+1) (X - A„) (81)


k ? i (; fc - * 1 ) • •• (A* ~ V i ) (** - **+») (A* — An)

if all the characteristic zeros are different. Using this formula it can be shown
th at any matrix, in which at least one characteristic zero is zero, is singular for
the function log X.
Let us suppose th at it is impossible to represent the matrix X in the form
(78), i.e. we suppose that the matrix has multiple elements. Using the argu­
ments of [88] it can be shown that the elements of X are
(A - A,)Pl.........(A — Am) Pn,, (82)

and th at the elements of the matrix log X , which give the solution of equation
(77), will be:
(A - log A1)Pl, . . . . (A - log AJP-. (83)

If, when the values of A* are equal, we take equal values of log A* then the
corresponding values of log X are said to be regular. I t can also be shown that
formula (81) gives all the regular values of the logarithm in the course of analytic
continuation and that it gives regular values only.
Consider, for example, the simplest irregular values of a logarithm. Take as
the matrix X a certain zero A, i.e. a diagonal matrix with elements A. We can
write this matrix as follows:
X = S [A, A.........A] S ' 1 = SXS-» = A/,
where S is any matrix, the determinant of which is not zero. Fixing the numbers
q in a definite way we obtain the value of the logarithm:

log X = S [log A+ 2w,t, log A+ 2nr2i .........log A-f 2nrni] S~l


or
log X = S [log A, log A, . . . log A] S~l + [2jW|Z, 2nrti, . . . , 2jir„i] S ' 1,

and finally:
log X = log AJ + 2niS [rj, r2, ... , r„] S ' 1.

If the zeros are not equal then the second term will essentially depend on
the matrix S, which can be chosen quite arbitrarily.
We saw above th at formula (79) gives the solution of equation (77) if the
matrix X has the form (78). I t can be shown that in this case formula (79)
gives all the solutions of equation (77) [S in (79) is chosen arbitrarily].
Consider now the square root of the matrix
l
Y = X2
as the solution of the equation (84)
Y2= X .
92] EXAMPLES OP MANY-VALUED FUNCTIONS 347

For the matrix X, situated near a unit matrix, we can represent one branch
of this many-valued function in the form of the following power series:

Y = [f + ( X - I ) ] 2 = J + i ( X - I ) + I (; - 1) ( X - I ) * + . . . (85)
2!

This series can be transformed by Sylvester’s formula providing the charac­


teristic zeros of the matrix X are different:

V _ x \ _ V (X - A,) . . . (X - Aa_,)(X - A,fl) . ■■(X - A„) ,r


<**-*>>■■ ■ ~ ~ Xk+i) •• • (A* - *„) ' k' {Vb)
To simplify the notation we suppose th at the matrix is of second order.
Let the matrix X be of the form:
| X = S [A„ A2] S"1 (At and A2 ^ 0).
I t can easily be proved th at equation (84) has the solution

S[±^T, ± (87)

where arbitrary values of the radicals may be taken.


I t can be shown that this formula gives all the solutions of equation (84)
and th at S, as in (79), is chosen arbitrarily.
If we take in formula (87) only regular values, i.e. when Xl and A2 coincide
we take equal values of the radicals only, then it can be shown in the same way
as for a logarithm th at formula (87) will give a definite solution which depends
on the choice of the radical but which is independent of the arbitrary choice
of S.
If A, and A2 are different, then formula (87) gives, in general, four different
solutions of equation (84). On supposing th at At and A2 are equal, we have for
the matrix X:
X = S [A„ A,] S-> = XJ,
where S is any matrix the determinant of which is not zero.
Formula (87) gives
_ _
X * = S [ ± / A l, ± / A , ] S - 1.
If we take equal values of the radicals then this is equivalent to the fol­
lowing:

X2 - ±Y^1- (88)
Consider now the case when different values of the radicals are taken:

X* = / A i S [ l , - l ] S - » , (89)

X2 = - / A ^ S [ 1 , - 1] S~l (90)
348 FUNCTIONS OP SEVERAL VARIABLES AND MATRIX FUNCTIONS [93

where S ie any matrix, the determinant of which is not zero. Write out this
matrix and its inverse in the expanded form:

M
1
*11. *12 *22D~l, - *„D-> *22»
: 5-» = = £)->
*21. *22 — 8tlD~l, *n-D-1 — 8tl. *11
where
D = D (S) =
*i*
— * 11*22 — * 12*21 •
*2U *22

Formula (89) can be rewritten in the following form


1^
X* *11. *12 *22. —*12
[1. - 1 ]
*21. *22 *21. *11

_ 2*11*12
■X* —Y^l (*11*22 *12*2l) 1 *11*22 4" *12*21. (91)
2 * 21* 22 . — (*11*22 + *12*2l) I

We can thus see th at in this case the square root X 1/2 has an infinite number
of values and th at these values contain arbitrary elements 8lk of the matrix S.
If the elements of X are (82) then the elements X 1/2 will be:

(A - Y ^)”'........ (A -
and if, when the values Xk are equal, equal values of yxk are taken then the
values of X 1/2 are said to be regular.
Formula (86) gives all the regular values of X 1/2 in the course of analytio
continuation. We assume th at not one of the zeros A^ is zero, for the point z = 0
is a singularity of the function yzl

93. Systems of linear equations with constant coefficients. Suppose


t h a t we are given a system o f lin ear differential eq u atio n s w ith con­
s ta n t coefficients:
x[ = a llx 1 + a L2x 2 + . • + a lnx n .

*2 = a 21X l + a 22X 2 + ■ ■ + °2 nXn>

xn = + a n2X2 + • * • 1
" ” >

w here x k are fu n ctions o f th e in d ep en d en t variable t a n d x'k th e d e ri­


v ativ es o f th ese functions.
L e t us suppose t h a t th e req u ired functions {xl ......... x„) are com po­
n en ts o f a certain vecto r
*(*i. x2...........*n) •
The com ponents x k are, in th is case, functions o f t a n d we d eterm ine
th e d ifferen tiatio n o f th e v ecto r x w ith respect to t, as a new vector
93] SYSTEMS OF LIN EA R EQUATIONS WITH CONSTANT COEFFICIENTS 349

with components (x[, ...,x 'n):


dx
dt o-
We lastly introduce a matrix A with elements aik. With this notation
we can rewrite the system (92) as follows:
d*_
dt Ax. (93)

Suppose that there is a solution of this equation which satisfies the


original conditions:
xk\t=0 = 4 0) (* = 1. 2, . . . , n ). (94)
These original conditions form a certain vector which we denote by
x(0)(40)- - 4 0)) •
It can readily be shown that the solution of the system (93) for the
given original conditions (94) has the form:

or, introducing the matrix


,, T . At A*t2
e = 1 + TT + “2T + • • ■’
we can rewrite the solution (95) as follows:
x = eA,x(0). (96)
In fact, formula (95) gives

x = x0 + Ax0 + A2x0 + .. •

Differentiating with respect to t we have:

+ -£j- A?xo + — ^ 3Xq + . • .


or

and from (95):



dt -~AA*'
x

At the same time the original conditions must also be satisfied,


for when t = 0 formula (96) gives x | ,_ 0 = x(o>.
350 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [93

We can rewrite the system (92) by using the matrix notation in


another form. As a preliminary we shall explain the f u n d a m e n t a l r u le s
o f d i f f e r e n t i a t i n g a m a t r i x . Suppose that the elements of a certain
matrix X are functions of the variable t . We determine the derivative
d X / d t as a matrix, the elements of which are obtained by differentia­
ting the elements of the matrix X with respect to t , i.e. [IIIj, 83]
( d X ) _ d { X ) lk
\ dt jik di
The usual law for differentiating a sum follows directly from this
definition, viz. if X and Y are two matrices the elements of which are
functions of t then
d ( X + Y) dX dY
(97)
At di ' dt

In exactly the same way the formula for differentiating a product


can easily be proved

y + x% . (98)

but it must be remembered that, in general, the positions of the factors


in the above formula (98) must not be interchanged. From the defi­
nition of the product we have:

= 2 { X U T U
$=1
whence
d (x r> ,A = ^ d ( X } ls
dt ^ dt {YU+
1 s=l

which gives us formula (98) immediately. This formula can readily be


generalized to include any number of factors. Thus, for example,
for three factors we have:

A (z r z ) = ^ F Z + A ^ z + z r - ^ . . (9 9 )

We shall also introduce a formula for the differentiation of the


reciprocal matrix. Suppose that the determinant of the matrix X is
not zero so that we have the reciprocal matrix X -1 where:

X X ! = /.
93] SYSTEMS OF LINEAR EQUATIONS W ITH CONSTANT COEFFICIENTS 351

Differentiating this identity with respect to t we obtain:


dX -*r_
‘ + * T = °.
from which follows the law for differentiating the inverse matrix:
dX-i
( 100 )
d/
Let us now return to the system (92). Consider n solutions of this
system. They form a square table consisting of n2 functions:
x n (0 . X 12 ( 0 ............... x in ( 0

X 21 ( 0 . X 22 ( 0 > • • • i X 2n ( 0

x m (0 . X n2 ( 0 ............... x nn ( 0

Where that the first subscript of the function denotes the number
of the function and the second subscript the number of the solution
into which the function enters, i.e. for example x23(t) denotes an ex­
pression for the second function x2 which enters into a solution with
number 3. We therefore have:

—ap- = <*/!*!* ( < ) + • • • + ainxnk (<) (*, k = 1 , 2 .........n)t

and formula (92) can be rewritten in the form of the following matrix:

IT " AX' I102*


where X is the matrix ( 101 ). Let us recall, once again, that for
this system of notation the matrix X gives n solutions of the system
(93) since every column of this matrix gives a certain solution of the
system (93). In this case the original condition will be the statement
that when t = 0
* U , 0 = * (0\ (103)
where X(0) is an arbitrarily chosen matrix with constant elements.
In the same way as above it can be shown that the solution of the
system (102), the original condition of which is (103) has the form
X = eMX™. (104)
Let us suppose that the determinant of the matrix X^°\ which gives
the original conditions, is not zero. We will show that in this case the
determinant of the matrix X will not be zero for any t. Bearing in
352 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [93

mind formula (104) we can see that in order to do so it is sufficient


to show that the determinant of the matrix eAI is not zero, for the
determinant of the product of two matrices is equal to the product
of the determinants of these matrices.
It can readily be shown that, in general, the determinant of the
exponential matrix eKis never zero.
In fact, together with the matrix

eK= ' + £ + £ + <105>


we construct the matrix
Yn
e -Y= I - + ( " 1) 4 + (106)

Multiplying the two series on the right-hand sides of the above


formulae we only deal with numbers and powers of one matrix Y,
60 that the places of all factors can be interchanged.
Thus we formally obtain the same result as would have been
obtained had we substituted the variable matrix Y by the variable
number z. But, at the same time, as a result of the identity e 'e '! = 1,
the multiplication of the right-hand sides of formulae (105) and (106)
will give unity and, consequently, in the case under consideration, we
shall also have the following equation:

6 ^ - ^ = I,

which holds for any matrix Y. It follows directly from this equation
that the matrix e~Y is the inverse of the matrix ev and that the
determinant of the matrix ev is not zero. Notice that if Y and Z are
two different matrices which do not commute, then the product evez
will not, in general, be equal to ey+z.
Thus it follows directly from formula (104) and from the proved
property of an exponential matrix that if the determinant of the
matrix X^s\ which gives the original conditions, is not zero then the
determinant of the matrix X will not be zero for any t. In this case
the matrix X will give n linearly-independent solutions of the system
(102 ). We shall now show that if Y is a matrix which gives any n
solutions of the system (102 ) then it can be expressed in terms of
the above matrix X by means of the formula

Y = XB, (107)
94] FUNCTIONS OF SEVERAL MATRICES 353
where B is a certain matrix with constant elements. Formula (107)
obviously shows that any solution of the system can be expressed line­
arly by n linearly-independent solutions of the system. To prove formula
(107) we notice, first of all, that from the given conditions, Y should
satisfy equation (102 ), i.e.

— = AY. (108)

It is also given that the determinant of the matrix X which satisfies


equation (102 ) is not zero and consequently, the inverse matrix X -1
exists. From the law for differentiating inverse matrices we have:
dX-i
di
x -1

or, bearing in mind formula (102 ), we obtain:

— = - I - M Z I - 1 = - X - 'A . (109)

Let us now construct the derivative of the product

dt dt
where, from (108) and (109)

— (X -JF) = - X ~ 'A Y + X - 'A Y ,

w < x - ' r >= °-


We thus see that the product X _1Y is a certain matrix B, the
elements of which do not depend on t; from this formula (107)
follows directly.

94. Functions of several matrices. We shall now explain the


fundamental ideas and facts connected with functions of several
matrices. Owing to the impossibility of commuting, the theory of
functions of several variable matrices is considerably more compli­
cated than the theory of functions of one variable matrix; we shall
therefore only consider the most fundamental facts of this theory.
Let us begin with the polynomial. The general form of a homogene­
ous polynomial of the second order in two matrices will be
aXi -j-bX^X2 -f- cX2X 1 -|- dX
354 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [94

A homogeneous polynomial of the second ordei in I variable matrices


will have the form
" 2 a i k X j X k ,
i,k=1
where the summation is carried out with respect to the variables i and
k, which independently of each other run through all the integers,
from unity to I. A homogeneous polynomial of order m in 1 variable
matrices will have the form:

2 an - - - j mXj l . . . X Jm. (110)


h.
Here, as before, ajt ■■■jm denote numerical coefficients. In formula
( 110 ) each of the variables of summation jk takes all integral values
from 1 to I, so that the above sum will contain lm terms in all. Let us
now consider that particular case when in formula ( 110 ) all the coeffi­
cients djl........ are unity:

y x h . . . x jM. (in )
jt •••»jrn~^
It can readily be shown that the sum (111) represents a power of
the sum of the matrix X jk, i.e.

(■ * !+ ••■ + *«)m = 2 X h ••• x j ~- (112)


Jit • • • r j m = 1
Thus, for example
(X i + x 2y = (X j + X 2) ( X 1 + x 2) = x f + x , x 2 -j- x 2x x + x 2-
Consider now a power series of I matrices. Such a series can be
written as follows:

a o + 2 a h' ‘ jm X ji ■ •■ X Jm. (H3)


m=l Ju ...,j™=i
A full investigation of the convergence of this series is considerably
more difficult than for a power series of one matrix and we shall only
prove the sufficient condition for absolute convergence of series (113).
Notice that series (113), like a series of one matrix, is absolutely con­
vergent, providing the series below is convergent

K l + 2 2 i . \ \ x h \ - - - \ X j m \. (ii4)
fn—1 jit
94] FUNCTIONS OF SEVERAL MATRICES 355

where the convergence of this latter series guarantees both the con­
vergence of series (113) and the independence of the sum of this series
from the position of its terms. Let us fix the integer m and denote
by the maximum modulus | a J l t . . . |, i.e.
K , ........(115)
We now construct the series of the usual complex variable

£ a (m)z m, (116)
m=l

and let n g be its radius of convergence, where n is the order of the


matrix. Replacing the coefficients | a,j , .. , j m | by greater coefficients,
viz. by a(m) in series (114), we obtain the series

a»+ 2 a(m) i
m=1 j , ..........

which can obviously be rewritten in the form:

«o+ ••• + l Z ' l ) m- ( U7)


m =1

This series can be regarded as a power series of one matrix


Z = \X,\ + ... + \ X l \, (118)
and bearing in mind that the radius of convergence of series (116) is
equal to n g , we can say [8 6 ], that series (117) will converge provided
that
l-^iI + • • • + l-^ii < IIell-
In this case series (114) will be all the more convergent. We thus
obtain the following theorem.
T h e o r e m . I f th e p o s i t i v e n u m b e r s a (m) a r e d e t e r m i n e d f r o m th e c o n ­
d itio n (115), a n d series (116) h a s a r a d i u s o f c o n v e r g e n c e e q u a l to n g , th e n
(113) w i l l b e a b s o l u t e l y
th e p o w e r s e r i e s c o n v e r g e n t p r o v i d e d c o n d i ti o n
(118) i s s a t i s f i e d .
In the particular case when the radius of convergence of the series
(116) is equal to infinity, series (113) will be absolutely convergent for
any matrix X k.
Notice also that the function f ( X v . . . , X [ ) , which is defined as
the sum of the series (113), satisfies the obvious relationship
f i S X . S - 1, .... S X ' S - 1) = 8f(Xv .... X t) S - \
356 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [94

where S is any matrix, the determinant of which is not zero. We had


an analogous property before for the analytic function of one variable
matrix.
We notice in conclusion, without attempting to give the proof, one
characteristic of a power series of several matrices in relation to the
uniqueness theorem. Here the uniqueness theorem is formulated as
follows: if the equation

«0 + 2 2 a h ............... j - X h • • • X Jm =
m“ 1jit •*'tjt»—1
= \ bh, , jmX Jl . . . X jm
m = l ht ■• - *jm~ 1

holds for all matrices


X lt • • • i X t
of any order, which are sufficiently near to a zero matrix, then b0 = a0
and bjlt • ••,;•„ = • • •,jm-
Had we omitted the conditions of “any order” in our formulation
then the theorem would have been false. In a particular case a homo­
geneous polynomial can be constructed, the coefficients of which are
not zero
2 °h’ ' ‘ ‘ ’ A ' ‘ ‘ X jm'
Jit .. •1 I
which makes the identity equal to zero for all matrices of a definite
order.
A description of the general theory of analytic functions of matrices
and their applications to the theory of systems of linear differential
equations was given in the works of the late I. A. Lappo-Danilevskij,
printed in the Journal of the Leningrad Physico-mathematical Society.
At present all papers remaining after the death of I. A. Lappo-
Danilevskij are published in the works of the Institute of Mathematics
at the Academy of Science, U. S. S. R.
CHAPTER V

LINEAR DIFFERENTIAL EQUATIONS

95. The expansion of a solution into a power series. In Volume II we


dealt with linear differential equations of the second order with variable
coefficients and, in particular, with the solution of these equations
by means of power series. At the time we only considered equations
which could be formally satisfied by a certain power series, without
proving the convergence of this series. In this chapter we shall give
a full and systematic account of linear equations of the second order,
the coefficients of which are analytic functions of a complex
variable. We therefore assume that the independent variable in a
differential equation is a complex variable and that the unknown
function and the coefficients are analytic functions.
Let us write a linear differential equation of the second order in the
form:
w " + p ( z) w ' + q (z) w = 0, (1)
where w ' and w " are derivatives of the function w with respect to the
variable z.
We are also given the following initial conditions:
W| 2~z„=c0; w \ 2_Z0 = <V (2)
Let us suppose that the coefficients p(z) and q ( z ) are regular functions
in the circle | z — z0 | < R . We will show that there is a solution of the
equation (1) in this circle (regular function), which satisfies the
equations (2). By introducing a new unknown function u = w ' , we
can rewrite equation (1) as a system of two equations of the first
order:
du . . , .6w
-& = - p ( z ) u - q { z ) w , ~ ^ = u .

For the sake of symmetry we shall consider the general case of a


system of two linear equations with two dependent variables:
— = a (z) u + b (z) v ; = c (z) u + d (z) v (3)
857
358 LINEAR DIFFERENTIAL EQUATIONS [95

and show that this system has a regular solution in the circle
| z — z„ | < R, which satisfies all initial conditions:

^ I 2 — 20 — a 't v \ 2— 20 = P > (^)

provided the coefficients of the system (3) are regular functions in


the above circle.
We shall again use the method of successive approximations which
we used earlier in Volume II. The proof will be exactly the same as
before. Instead of the system (3) with the initial conditions (4) let lis
write the equations in the form of integrals:
Z Z

M= a + J [a (z) u + b (z) u] d2 ; v = fi + j- [c (z) u d (z) v\ d z. (5)


2, 2,

Consider the circle K : | z —z „| < Rv where R1 is a positive number,


smaller than R. At every point strictly inside this circle, the coefficients
are regular functions and, consequently, the following inequalities
hold:
\a (z )\< M ; \b (z)\< M\ \c {z)\< M \ \d (z )\< M , (6 )
where M is a fixed positive number. Applying the method of successive
approximations we assume:
u0 (z) = a; v0 (2) = /?, (V
and generally
Z
«n+l (3) = a + J [a (2) Un + b (2) vn\ d2 .
( 8)
l ’n + 1 ( 2 ) = P+ ] [c (Z ) « „ + d (2 ) Vn\ d2.
z.
At every stage the functions of z to be integrated are regular and
in no case does the value of the integral depend on the path of inte­
gration in the circle K. Suppose also that to is a positive number,
satisfying the inequalities
| a | < to; | /? | < to. (9)
To simplify the working we assume that z0= 0 and integrate from
0 to 2 along a straight line. In this case
z = ge1'!’; d 2 = eiv dp (0 < g < R x) . (10)
95] THE EXPANSION OP A SOLUTION INTO A POWER SERIES 359

The first of the formulae (8 ), when n = 0 , gives:


c
(z) a + b (z) /J] e,ipdp.
o
Substituting under the integral all terms by their moduli and using
(6 ) and (7) we obtain the inequality
| ux (z) — u0 (2 ) | < 2Mmg (llj)
and, similarly
K (2) — v0{z) | < 2Mmg. (112)
The first of the equations (8 ), when n = 1, gives
Z
u2 (2) = a + f [a (2) ux + b (2) Vy\ d2,
0
and, subtracting from this the first of the equations (8 ) when n = 0 ,
we obtain:
Z
( )-
“ ■2 2 «1 (2) = J [a (2) K ~ U 0) + b (z) (»! - »„)] d2 .
6
Replacing each term by its modulus and using the inequalities ( 11A)
and ( 112) we have:
Q
I w2 (2) — Uy (2) I < (2M)2TOJ p dp
or
|«2 (*)“ «! (Z)| < TO (2y
and similarly
i®2( 2 ) - » i (2)1 <TO-(2y - .
Continuing in this way we obtain the following results:

K + i (2 ) - un (2) I < m {2{™ l\+


y_ ,

I v n+ i (2) - Vn (2 ) 1< •
It follows that the terms of the series
Wo + [« 1 (z) — M0] + [« 2 (2) — Wl (2 )] + ••• (1 2 )

in the circle | z — z0 | < By have smaller moduli than the positive


numbers
(2Me)n+:
(n + 1)1
360 LINEAR DIFFERENTIAL EQUATIONS [96

which form a convergent series, i.e. the series (12) converges absolutely
and uniformly in the circle | z — z0 \ < Rv The sum of the first (n + 1)
terms of this series is equal to un{z) and, consequently, un(z) tends
uniformly to a function u(z) in the circle | z — z0 | < Rv Similarly
vn(z) tends uniformly to a function v(z). According to the theorem of
Weierstrass on uniformly convergent series, these limit functions will
also be regular in the circle K. Consider now the equations (8). In the
first of these the integrand tends uniformly to the limit function
a (z) u + b (z) v.
But we know from [I, 146] that as we can integrate any uniformly
convergent series term by term, the integral of the limit of a uni­
formly convergent sequence of functions is the limit of the integral;
hence taking the limit in the equations (8) we see that the limit
functions u and v satisfy the equations (5). Assuming in these equations
that z = z0 we see that our functions satisfy the intial conditions (4);
differentiating equations (5) we see that they give the solution of the
system (3).
Let us now return to the equation (1), which is a particular case of
the system (3). We have shown that in any circle, centre z0, which
lies inside the circle | z — z0 | < R, a solution of the equation can be
found which satisfies the conditions (2) for every c0and cx.The functions
p(z) and q(z) can be expanded in the circle \ z — z0 \ < R into power
series
P (z) = a0 + ax (z — z0) + . . . ; q (2) = b0 + (2 — 20) + . . .
The solution so obtained is also a regular function and therefore it
can be expanded into a power series in which, from (2), the first two
coefficients are equal to c„ and cx:
tv = c0 + c1(z — z0) + c2(z — Z0)2 + . . . (13)
Substituting this series in equation (1) we equate to zero the
coefficients of different powers of 2 and this gives us, as we saw in
[II, 45], equations of the type:
2. lc2 -(- a0Ci -(- b0c0 = 0
3. 2ca 2ouc2 -(- ffljc.y 4- bQc1 61c0 = 0,

whence the coefficients ck are determined successively. This shows,


first of all, that there can only be one solution. It also follows from the
95] THE EXPANSION OF A SOLUTION INTO A POWER SEMES 361

above proof that there must be a solution, i.e. that by substituting the
coefficients we obtained above in the series (13) we obtain a series
which converges in every circle inside the circle \z — z0 \ < R, in other
words, it converges in the circle | z — z0 \ < R. We thus obtain the
following theorem.
T h e o r e m 1. I f the coefficients of the equation (1 ) are regular functions
in the circle \z — z0 | < R then there exists a unique solution of the equa­
tion in this circle which satisfies the initial conditions (2) for any given
Cq and' Cj.
By giving c0 and cx definite numerical values we can construct two
solutions wx and w2, which satisfy the initial conditions:

u,i |2-2o = ai; w'i\2=2„ = /3i>


w2 |r»2o = a2> W2Ir.zo = fit ■
If
al a2Pi ^ 0 > (14)
then any solution, regular in the circle | z — z0 \ < R, can be ex­
pressed in terms of v \ and w2 in the form:

w = A 1w1 + A2w2 . (15)

In fact if this solution w is based on the initial conditions (2), then we


have the following system of equations for the constants Ax and A2:

A i cq -f- A 2a2 = cu; A x f}x A 2@2 = cx,

and this system, from (14), determines the values of Ax and A.z. The
constructed solutions for wx and w2, will be linearly-independent
solutions of the equation (1) [II, 24].
Note. The application of the method of succesive approximations in
the system (3) gave us the infinite series (12) for the function u. This
series will not be a power series but its uniform convergence in the
circle \ z — z0 \ < Rx guarantees the existence of a regular solution
in that circle in the form of a power series. We can construct the func­
tion un(z) and the series (12) in any domain in which the coefficients
of the system (3) are regular functions and it can be argued, in the
same way as above, that in any such domain the series (12), and the
analogous series for v, will be uniformly convergent and will give the
solution of the system. The form of these series in certain cases will be
explained below.
362 LINEAR DIFFERENTIAL EQUATIONS [96

96. The analytic continuation of the solution. Let us now suppose


that the coefficients p(z) and q(z) are regular functions in a domain B
of the z plane. Let us take a certain solution of this equation, which
satisfies the initial conditions (2) at a point z0 in B. This solution, as we
know from above, will have the form of a convergent power series in
a circle, centre z0, which lies wholly in B (and perhaps also in a larger
circle). It will be a power series of the form (13). Let us now take a
fixed point z in the circle of convergence of this series and rewrite the
series in powers of (z — zx), as we did in the section on the analytic
continuation of a function. We thus obtain a new series

2 d k{ z - Zl)k . (16)
k=0
Its sum will coincide with w in the common domain of the circles of
convergence of the series (13) and (16). Consequently in this common
domain the sum f(z) of the series (16) will be the solution of the equation
(1); in other words, when substituting w = f(z) in the left-hand side
of equation (1) the latter will vanish in a certain part of the circle of
convergence of the series (16). But in that case, as a result of the funda­
mental principle of analytic continuation, it will also vanish in the
part of the circle which belongs to B; the series (16) will thus give a
solution of our equation. This solution will be fully determined from
its initial conditions at the point zv

A *i) = w I*-*,; /'( z i) = w ' |* -n -

where w is determined by the initial series (13).


As a result of the theorem proved in the previous section the
series (16) is bound to converge in a circle, centre zv which belongs
to that domain B in which p(z) and q(z) are regular functions. We
thus arrive at the following theorem.
T h e o r e m II. I f the coefficients p(z) and q(z) are regular functions in a
domain B, then any solution of the equation in the form of a power series
with the centre of convergence in B can be analytically continued in any
direction in B and this analytic continuation gives everywhere a solution
of the equation (1).
Let us make some essential additions to the above theorem. Notice
that if B is a connected domain, then according to the fundamental
principle of analytic continuation [18], w will be a single-valued regular
function in B which, according to the given proof, will be a solution
97] THE NEIGHBOURHOOD OF A SINGULARITY 363

of the equation (1). If, however, B is a multiply-connected domain,


then w will not, in general, be a single-valued function in B.
If wx and w2 are two solutions of equation (1) then we have the
following formula [II, 24]:
d ( w ,\ G -SP&*2
dz (, u\ J w\ 6
(17)

where G is a constant. If C is not zero, then the left-hand side will


never be zero during the analytic continuation, i.e. the analytic con­
tinuation of linearly-independent solutions always gives linearly-
independent solutions, and formula (15) therefore gives the analytic
continuation of any solution in terms of the analytic continuation of
two linearly-independent solutions.
If, for example, the coefficients p(z) and q(z) are rational functions
then any solution of the equation can be analytically continued in
the plane in any direction except through the poles of p(z) or q(z).

97. The neighbourhood of a singularity. Let us now investigate the


behaviour of the solution in the neighbourhood of a singularity of the
coefficients p(z) and q(z). Let us suppose that the point z = z0 is a
pole or an essential singularity of the coefficients p(z) and q(z), so
that these coefficients can be represented by a Laurent series in an
annulus K, centre z„ and zero inner radius:
+ as
P(z)= ak(z 20)fc.
k = — CD
(18)
7(2)= 22 bk(z — z0)k (0 < I z — 20| < B)
fi=~ aa
Any solution of equation (1) can be analytically continued along any
path in the annulus K, but if the path encircles the point z = z0 the
solution w may take new values, i.e. the point 2 = z„, will, in general,
be a branch-point. Let us explain in greater detail the character
of this branch-point. Take any two linearly-independent solutions
wx and w2. If we cut the annulus from the centre along any radius,
then in the connected domain so obtained, our solutions wx and w2will
be regular single-valued functions, but will have different values on
opposite edges of the cut. In other words, after describing the point
z = z0 the functions wl and w2 will become new functions wx and w t ■
They must also be solutions of the equation, and can therefore be
364 UKEAR DIFFERENTIAL EQUATIONS [97

expressed as a linear combination of and w2. Hence the formula,


given below, must hold:
wi ~ an wi + <h2“>2, I (19)
w} - a21w1 -f a22w2, (
where alk are constants. In other words, by describing a singularity,
linearly-independent solutions undergo a linear transformation. It
can readily be seen that
®11 ®22 ®12®21 ^ ®• (20)
In fact, if we have au a22 — a12a21 = 0 then the solutions Wy and w2
differ only in constant terms and are linearly-dependent; however
this cannot be so for we saw earlier that the analytic continuation
of linearly-independent solutions produces linearly-independent solu­
tions. The form of the linear transformation (19) does, of course,
depend on the choice of the solutions Wy and w2.
We shall try to construct a solution which by describing a singu­
larity acquires a constant term, i.e. a solution which undergoes the
simplest possible linear transformation
w+ = h v . (21)
This solution, if it exists, should be a linear combination of the
solutions wy and w2:
w = b1w1 + b2w2,
where the coefficients by and b2 must be found. We have, from (21):
by w{ -f b2w£ - - Z.(by Wy + b2w2) ,
and this, from (19), gives:
by(a,yyWy + dy2U>2) -f b ^ y W y + Ojjj Wjj) = k(by Wy + \ Uljj) .
Comparing the coefficients of the linearly-independent solutions we
obtain a system of homogeneous equations for by and b2:
(Uji X)by-\- a2y b2 = 0 ,
an b\ (U22 — ^) b2 = 0 . (22)
To obtain values for by and b2 which are not zero we must equate
the determinant of the above system to zero
°11 “ C21 = 0. (23)
®12> ®22 ^
97] THE NEIGHBOURHOOD OP A SINGULARITY 366
This is a quadratic equation in A. Taking a zero A = Ax of this
equation and substituting this for the coefficients of the system (22)
we obtain a solution for bx and b2 which is not zero. Hence the zeros
of equation (23) give the possible values of Ain formula (21), i.e. these
zeros are equal to the constants by which the existing solutions of equa­
tion (1) must be multiplied when describing the singularity z0 in the
positive direction. If we take at the start different linearly-independent
solutions, then the linear transformation (19) will be different but the
zeros of equation (23) should remain the same, for they are fully de­
fined and independent of the choice of the fundamental solutions.
Consider, first of all, the case when the quadratic equation has two
different zeros
A= Aj and A= A2.

We then have two solutions which satisfy the conditions

wf = A j = A2w 2 . (24)

These two solutions must be linearly-independent. Otherwise w2lw1


would be a constant and would not alter by describing the singularity;
but, according to (24), this fraction acquires the factor A2/A1in doing so.
Notice also that, from (20), Ax and A2 must not be zero.
Introduce the two constants

Pi = - 2^ log ea = 2^ log Aa, (25)

where the logarithms are arbitrary. Construct two functions:

(z - z0)»i = eei 1ob<*-*>; (z - z0)e» = ee*log <*-*<>>.

By describing the singularity they acquire the factors


60 i 2ni _ e io g _ Xi, e*» 2nl = elog A* = A2.
Hence the relationships
and m,
(z - z0)«‘ and (z - z0)e*
remain single-valued on describing the singularity, i.e. they are regular
single-valued functions in the neighbourhood of the point z = z0, and
they can therefore be represented by a Laurent series in that neigh­
bourhood. Hence the constructed solutions can be represented in the
366 LINEAR DIFFERENTIAL EQUATIONS [97

neighbourhood of the singularity as follows:


+ co
w1 = {z — z0)e<
- £ c'^z - zo)k >
fc=—"
(26)
W2 = ( z - 2 0 ) ea Cfc(Z - Z0)k .
k= —“
Notice that log Ais accurate to the term 2mm, where m is any integer.
Hence, from (25), and q2 accurate to integers. This is in full agree­
ment with the formulae (26), for by multiplying a Laurent series by
(z — z0)m, where m is any integer, we obtain another Laurent series
and therefore and q2 in the formulae (26) are accurate to integers.
Consider now the case when the zeros of the equation (23) are the
same, i.e. A2 = A2. As before, we can construct one solution of the
equation which satisfies the condition
Wy = \yW 1. (27)
Take an arbitrary second solution w2 which is linearly-independent
of wv By describing the singularity it is subjected to the following
linear transformation:
u>Z = a21wx + a22 w2 . (28)
The quadratic equation (23) will have the following form for these
constructed solutions
A, a21 _ q
b, &22 ^
It is given that this equation has a double zero A = A2 and it follows
that a22 = Aj, i.e. formula (28) must have the form
w$ = \ w 2 + SjiWj. (29)
It follows from (27) and (29) that the relationship only acquires
a constant term by describing the singularity

and consequently the difference

-Sr ~ ~ £ k log(z ~ Zo) = ^ r ~ a log(z - 2o)


is single-valued and can be represented by a Laurent series. Hence,
bearing in mind that w1 has the form (26) and that the product of a
Laurent series and w1 has the same form as wlt we can see that, in this
97] THE NEIGHBOURHOOD OP A SINGULARITY 367

case, our solutions can be expressed as follows in the neighbourhood


of the singularity:
+ CD

Wy = {2 — Z0)«1 2 C'k(Z - Zo)k •


k= —a>
+ CO
(30)
w2= (z — z0)ei 2 c"k(z - z0)k + awy log (z - z0) .
fc=—=
We thus arrive at the following theorem.
Theorem III. I f z = z„ is a pole or an essential singularity of the
coefficients p(z) and q(z), then two linearly-independent solutions exist
which can be expressed in the form (26) or (30) in the neighbourhood of that
point. Notice that in the second case, when the zeros of equation (23)
are the same, it may happen that the constant a21 and the dependent
constant
a = 2 °2'iXx

are both zero and we apply formula (26) in the neighbourhood of this
point.
All that was said above refers to the case when z0 lies at a finite
distance. When it lies at infinity we must replace z by a new variable t,
according to the formula

Differentiating with respect to t, instead of z, we have:


jd_ 12 _L
dz dt : dz2 d<2 + 1 dt ’
and equation (1) with the new independent variable takes the form
d2w Aw ,
t1 -r -p
~ar + q w= 0. (31)

For this new equation the former point at infinity will become t = 0
and we shall now investigate the neighbourhood of this point.
Notice that all the arguments used above were purely theoretical.
They do not give a practical method for constructing equation (23)
or for finding the coefficients of the expansions (26) and (30). We shall
now consider a practical method for finding these. We can do this
only in one case, viz. when the expansions of these formulae contain
a finite number of terms with negative powers.
The singularity z = z0 is then known as a regular singularity, i.e.
a pole or an essential singularity of the coefficients in equation (1) is
368 LINEAR DIFFERENTIAL EQUATIONS [9 8

known as a regular singularity of that equation if Laurent's expansion


(26) or (30) contains only a finite number of terms with negative powers.
Replacing pj and q2by an integer we can, as with a regular singularity,
always achieve the fact that the power series in the formulae (26) or
(30) should contain no terms with negative powers and should begin
with a constant term, i.e. for example, instead of (26) we can write

w1 = {z — Zo)* J p c'k(z - z0)k ,


k”° {c'o and cl # 0) (26,)
w2 = ( z - 2 o)- y c"k(z — z0)k .
k=0

Otherwise, if at least one expansion in the formulae (26) or (30)


contains an infinite number of terms with negative powers, the singu­
larity is known aB irregular. We must, first of all, indicate criteria
by which it is possible to determine from the coefficients of the equation
whether the singularity is regular or irregular.

98. Regular singularity. Let us suppose that w2and w2are two linearly
independent analytic functions. It is not difficult to construct a linear
equation for which these functions are solutions. In fact, we should
have:
w{ -f p(z) w[ -f q(z) w1 = 0;
w”
2 + p(z) w'2 + q(z) w2 = 0,
whence the coefficients of the equation can readily be determined
[II, 24]:
wt
p(z) = (32)
w' w1—w>' wt
and
?(z) = - — - P(Z) (33)

Let us suppose that the point z = z0 is a regular singularity and


only consider the case when # g2, since the formulae (30) can be
investigated similarly. In future we shall denote by Pk(z —z0) any series
in positive integral powers of (z — z0), the constant term of which is
not zero. In the case under consideration it is given that the singu­
larity z = z0 is regular and the solution is of the form:

w1 = (z — z0)»i P 2(z — z0) ; w2 = (z - z„)«* P 2(z — z0).


98] REGULAR SENGULARITY 369
Hence
= (z 20)8s—Bl P 3(2),
since the quotient obtained by dividing two power series with constant
terms is also a power series with a constant term. We have further:

A(z) = w'2 w l ~ w[ w 2 = w \ ^ =

= (3 - 30)2ei P 4(3 - Z q) [(3 - 30)'»-«‘ P3 (3)]' ,

or, differentiating the product and taking (2 — 20)C2~ ei-1 outside the
brackets:
d(2) = (3 - So^+o * - 1 P 5 (3 - Z q)

and differentiating with respect to 2 we obtain:


^'(*) = (Pi + e* - 1) (2 - Z0)°i+e*-*P6(z - 20) +
+ (Z - Z q^ + O * - 1 P 'S(Z - Z q) .

Hence:
p(z) =
A'(Z) l-gi-gj , P't(z - Z
q)
A (z) z - z0 ^ P A Z - zo) ’
i.e. p(z) can have a pole at the point 2 = z0of an order not higher than
one.
By differentiating the expression for w1 we find that w[jw1 can have
a pole of an order not higher than one at the point z0, and w"lw1
a pole of an order not higher than two. Formula (33) shows that q(z)
can have a pole of an order not higher than two at the point z0.
We thus arrive at the following theorem.
T heorem I. The necessary condition for the point z0 to be a regular
singularity is that the coefficient p(z) should have at the point z0 a pole
of an order not higher than one and the coefficient q(z) should have at the
point Zq a pole of an order not higher than two, i.e. equation (I) should be of
the following form
w" + - r = i r w' + 7r- h f w = 0 ’
Z Z q {Z Z q) (34>
where p1(2) and qy(z) are regular functions at the point z0.
We will now show that this condition is not only necessary but is
also sufficient for the singularity to be regular. Let us recall that
equations of the form (34) are similar to equations which we con­
sidered in [II, 47] and for which we constructed a formal solution
in the form of a generalized power series. However, before we did not
investigate the convergence of the constructed series. We shall now
370 LINEAR DIFFERENTIAL EQUATIONS [98

investigate this problem fully and show that the formally constructed
series will be convergent and will give a solution of the equation.
To simplify our notation we put z0 = 0.
Multiply equation (34) by z2 and rewrite it in the form:
z- w” -f z(a0 + ax z + . . .) w' + (b0 + bx z + .. .) w = 0 . (35)
We shall seek a solution of this equation in the form:

w = z ? ^ c kzk . (36)
k=0
Substituting this in the left-hand side of equation (35) and equating
the coefficients of like powers of z to zero we obtain equations for the
determination of the coefficients ck. These equations have the form:

CofoiQ) = 0
Ci/o(e + 1) + c 0/i(e) = 0
c i f o{ Q + 2) + c i f i ( 6 + 1) + co f i ( 6 ) — 0 (37)
cnfo(Q + + n — 1) + . . . + c0f„{q) = 0

where, for the sake of simplicity, we introduced the following notation:

/oW — ~ 1) + Aa<> ~r bo
(38)
fkW = Aak + bk
We have said already that we can assume that c0 ^ 0; the first of
the equations (37) gives a quadratic equation for the determination
of g:
fo(d) = Q(Q —1) + 6a o + = 0• (39)
This equation is usually known as the determining equation for the
singularity under consideration. Suppose that is a zero of this equa­
tion which is such that for every whole positive n we have the condition:
/o(e! + » ) # 0 ( » = 1,2, . . . ) . (40)
In this case the equations (37), starting with the second equation,
enable us to determine successively the coefficients cv c2, . . . The first
coefficient c0 remains arbitrary and will play the part of an arbitrary
constant so that we can take, for example, c0 = 1. We must also
show that the constructed series which forms part of formula (36)
will be a convergent series in the neighbourhood of the point z = 0.
98] REGULAR SINGULARITY 371

Let R be the circle of convergence of the series which are the coeffici­
ents of equation (35). If R l is a positive number, smaller than R , then
we obtain the following result for the coefficients ak and bh in these
series [14]:
K l< ^ ; \h \< ^,
where m1and m2 are constants. Hence

and, consequently, taking M sufficiently large, we obtain the result


in the form:
M
Ia* I + |&fr| (41)
The relationship
lei + » _ _____________ lei + n _____________
/o(e + n) (e + n) (e + n - 1) + (e + n) o0 + b0
tends to zero as the integer 7i increases indefinitely, for the numerator
is a polynomial of the first degree in n and the denominator is a poly­
nomial of the second degree. Hence a positive integer N exists such
that
Ifo(e + n ) I > I e I + n when n > N . (42)
We have from the formulae (37):
- _ /i(g + w — I) /z(e + w - 2) _ /n(g) „
n hie + w) " 1 f0(e + n) n 2 '" /0(e +«) c° '
whence
l/i(e + » - D I - I , Ifjie + n - 2 ) 1
I cn I < Ifoie + n) I n"ll+ l/.(e + «)l cn- 2I +

+ ...+ Ifnie) I (43)


Ifoie + n) i 1*1-
We have further:
fk(e + n — k) = bk + (o -f n — k) ak
IA (e + » - * ) | < | 6 * H - ( | e | + » ) | f l * | ( ’ n)
and for this reason also:
Ifk(Q + n - ft) | < (| Q| + ») (| ak | + | h | ) . (44)
We can always select a sufficiently large positive number P, so as
to obtain the following result for the first N coefficients:

kl<l*F (& = 0,1, . . N — 1). (45)


372 LINEAR DIFFERENTIAL EQUATIONS [98

We now recall that we took c0 = 1. We also assumed that P is


chosen so that
P> 1+ M. (46)
For the remaining coefficients, starting with cN, we can use the in­
equality (42). We will show by using it that if the result (45) holds
for all ck, from c0 to c„ exclusively, then it will also hold for cn. In fact,
from (42), (43) and (44) we have:

Icn I < (| a l I + I&l |) Icn - 1 I + (I a i I + I&2I) Ic n - 2 ! + ••• +


+ (|®nl + l&n|) | col >
or from (41)
I I - I I I M. | I , , M I |
ICn I R_ I cn—1 I Icn—2 | “I” • ■• H Bi
Rn" I C0I >
or, assuming that for c0, cv . . . , cn — 1 we have the result (45):

l c" l < - ^ ( f n ~ 1 + f n ~ 2 + - - - + 1) = (4 7 )

We will now show that


- 1) . pn (48)
P -1
In fact, this inequality is equivalent to the following:

Pn+1 - ( l + M ) P n + M > 0
or
Pn[P — (1 + M)] + M > 0 .

This last inequality follows from (46). The inequalities (47) and (48)
give
pn

and our proposition is proved.


The result (45) is therefore valid up to k = N — 1 inclusively as
a result of the choice of P. The inequality (42) holds for later symbols;
by using this inequality we have shown that if the result (45) holds
up to a certain suffix k it will also hold for the succeeding suffix k + 1
We have thus proved that the result (45) holds for every suffix, i.e.
for every n we have:
98] REGULAR SINGULARITY 373

But the series


z"

will converge absolutely in the circle | z | < RJP. Hence in this circle,
the series in formula (36), the moduli of the terms of which are not
greater than the terms of the preceeding series, will also converge
absolutely and this series can be differentiated term by term, like
any other convergent power series.
We have thus shown that formula (36) gives, in fact, a solution
of our equation in the neighbourhood of the point z — 0. We will now
show that the series (36) converges in the whole circle | 2 | < R, where
the series which are the coefficients of the equation (35) converge. Other­
wise the function which is determined in the neighbourhood of z = 0 by
the power series in formula (36), would have a singularity in the circle
| z | < R in the course of the analytic continuation [18] (other than
the point 2 = 0). But this is impossible since the coefficients of
equation (35) are regular functions in the whole circle | z | < R, ex­
cept at the point 2 = 0 and, from [97], the solution cannot have
any singularities during analytic continuation at this point.
If the difference of the zeros of the quadratic equation (39) is not an
integer, then the condition (40) is satisfied for each zero; two linearly-
independent solutions of the form (36) ( ^ p2) can therefore be
constructed.
We shall investigate the case when the quadratic equation (39) has
zeros whose difference is either an integer or zero.
In the second case, using the above method w7ith repeated zero of
the equation, we can construct one solution of the form (36) and a second
solution will have to be found. Consider the first case. Let gx and g2
be the zeros of equation (39) where ^ = g2 + to and to is a positive
integer, i.e. ^ is that zero of the equation, the real part of which is
greater than that of the second zero. Condition (40) is evidently satis­
fied by the zero gx and, by using this zero, a solution can be constructed
in the way described above. When attempting to use the second zero
g2 for the construction of a solution we find the following obstacle:
g2 + to is a zero of the equation (39) and, consequently, if we take
the (m + l)th equation of the system (37)
CmfoiQi + »») + cm-i/i(o 2 + to — 1) + .. . + c0fm{g2) = 0 ,
then in this equation the coefficient f0( g2 + to) of the unknown cm
will be zero. The sum of the remaining terms will, in general not
374 LINEAR DIFFERENTIAL EQUATIONS [98

be zero and this contradicts the equation. Hence even in this case
we have to find the second solution in another way. Notice that if it
should by chance happen that the above sum in the latter equation
is zero, then we could take any value for cm and continue to calcu­
late the successive coefficients cm+v . . . Our previous results show
that the series obtained will converge and we shall therefore also
obtain a second solution in the form (36) for this particular case.
Let us now establish the form of the second solution by assuming
generally that
Pi = 02 + m > (49)
where to is a positive integer or zero. Let us recall that for a linear
equation
w" + p(z) w' + q(z) w = 0
we have a formula which gives a second solution w2 of the equation
when one solution w1 is known [II, 24]:

w2 = C w ^ e - i ^ dz~ , (50)

where C is an arbitrary constant. In this case

p(z) = — + a,i + a2 z + • • •
and
Jp(z) dz = logz°<>+ Cx + axz + —a2z2 + . . . ,
whence
e-!p( 2)d:= z -a, ^ (z ) ,

where, as before, Px(z) is Taylor’s series in powers of z, the constant


term of which is not zero. The constructed solution has the form
w1 = z»‘ P 2(z), (51)
whence
w\ = z2ei P 3(z),
where Ps(z) is a Taylor’s series, the constant term of which is not zero.
The integrand in formula (50) will therefore be:

e - SP(*) ^ = z - a 0-2ei PJz) .


Wf
The numbers gx and g2 are the zeros of the quadratic equation (39)
and therefore:
Pi + &2 — 1 — ao •
98] REGULAR SINGULARITY 375

Hence, from (49):


— — 2px = e2 — Qi ~ 1 = — (1 + m ) ,
i.e. the integrand in formula (50) can be expressed as:

e - S p W d z J . = g—(l+m) p i ( z ) = + . . . + V f - + y0 + YlZ + . . .

( y - a + m ) T6 0) •

Integrating this expression we obtain one logarithmic term log z


and a series which begins with z~m. Multiplying this by the expres­
sion for w-y given by formula (51), we finally obtain the following:

w2 = z~mPh{z) ■& P2(z) + Wy log z ,


and from (49):
w2 = 3®' P 6(z) + y_x Wy log 3, (52)
where P6(z) is a Taylor’s series with a constant term. The expression (52)
has the same form as the second of the expressions (30) and in formula
(52) the Laurent series has no terms with negative powers. Notice
that the constant >,_1 is, in general, not zero but in isolated cases
it can be zero. This will be so in the case described above. Hence in
that case we also obtain a characteristic second solution for a regular
singularity, i.e. we have the following theorem.
Theorem II. In order that the point z = z0 should be a regular singu­
larity it is sufficient that the coefficient p(z) in equation (1) should have
a pole at the point z0 of an order not higher than one and the coefficient
q(z) a pole of an order not higher than two.
The necessity of this condition has been explained above.
Notice that it may sometimes happen that at a regular singularity
neither solution has any peculiarities. This will be so when and q2
are positive integers and when the second solution does not contain a
logarithmic term. Thus, for example, the equation

w" 2 '

22
w= 0

has the following two linearly-independent solutions*


Wy = z; w2 = z2.
Notice also that, when = q2, the constant y_xin formula (52) will
not be zero; this follows from the above calculations when m = 0.
376 LINEAR DIFFERENTIAL EQUATIONS [99

99. Equations of Fuchs’s class. The first systematic investigation of


regular singularities was undertaken in the middle of the nineteenth
century by the German mathematician Fuchs. We shall now investi­
gate equations, all the singularities of which are regular singularities.
Such equations are usually known as equations of Fuchs’s class. Let us
write our equation in the form:
w* + p(z) w' + q(z) w = 0 . (53)
On putting the independent variable
1
* = T>
we obtain, as we saw above, the equation

It is given that the point t = 0 is an essential singularity of this


equation. Bearing in mind that, after dividing by tA, the coefficient
of dwjdt cannot have a pole of an order higher than one at the point
t = 0 , we have for p(l/t) the following expansion:

v (t) = di 1 + d*t2 + •' ' ’


i.e. near z — °° the function p(z) can be expanded as follows:

p(2) = d i ^ - + d 2— + . . . (54)

Similarly, bearing in mind that the coefficient (ljt4)q(ljt) cannot have


a pole of order higher than two at the point t = 0 , we obtain:

® (-) = d'z F-+ d'3t* + . . . ,

and, consequently, we have for q(z) the following expansion near z = °°

q(z) = d'z d'3— + • • ■, (55)

i.e. if the point at infinity z = °° is to be a regular singularity of the


equation (53) it is necessary and sufficient that p(z) should have a
zero at the point z = °o and q(z) a zero of order not less than two.
Notice that if in the expansion (54) dx = 2 and in the expansion
(55) d{ = d'3 = 0, then the point t = 0 is not a singularity of the
equation (531). In this case the equation has the following solution in
99] EQUATIONS OF FUCHS’S CLASS 377

the neighbourhood of 2 = °°:

w ‘C0 + -£- + -fi- + .


where the coefficients c0 and cl are constant.
Let ax....... an be the singularities of our equation within a finite dis­
tance. The function p(z) can have poles of the first order at these points
and, according to (54), it must vanish at infinity, i.e. it will be a
rational fraction of the form
Pi(g)______
p(z) = (z — a ,)...(z — a„)
where the order of the numerator is at least by one unit less than
the order of the denominator. Similarly, from (55), we can see that
q(z) must have the form

qK ) ( z - a lY . . . { z - a nY- ’

where the order of the numerator is at least two units less than
the order of the denominator. Converting the rational fraction into
partial fractions we obtain the following general expressions for the
coefficients of the equations of Fuchs’s class:

(56)

We have from (55):


zq(z) —*■0 as z -> 00 ,
and the second of the expressions (56) shows that the constants Ck
satisfy the condition
Ci + C2 + . . ■+ Cn — 0 . (57)
The expressions (56) and the condition (57) are the necessary and
sufficient conditions for the equation (53) to be an equation of Fuchs’s class.
Let us now construct the determining equation for the singularities
z = ak and for the point z = 00. For the point ak the coefficient of
(2 — ak) - 1 in the expression for p(z) is equal to Ak and the coefficients
of (2 — ak) ~2 in the expression for q(z) is equal to Bk, so that the deter­
mining equation at that point will be
e(e- i) + A e + B k = o (k = , ,..., n).
1 2 (58)
378 LINEAR DIFFERENTIAL EQUATIONS [99

Let us now consider the point z = °°, i.e. the point t = 0 for the
equation (531). The coefficient of t~l in the expression

- L [ 2 0 - P P (± )]

must be determined as follows:

or
lim s3 — -J-p(z)] = 2 - lim zp(z) .

Bearing in mind the first of the equations (56) we obtain the follow­
ing expression for the coefficient:

2 - 2 A k.
k= l

Similarly the coefficient of t~2 in the expression

4 -* f t
will be
1™ -7T <7 (-f) = Jim z2 q(z).

But from (56) and (57):

k____ _

k=1 ' fc =l 1 —
a*

_ -v Bk V f ak^k i akCk _i \.
& 2s + 23
hence
lim z2 q(z) = (Bk + ak Ck)
2—~ k=1

Finally, the determining equation will have the following form when
2 = co;
Q(G — 1) + £?(2 — ^ 4- (Bk + akCk) = 0 . (59)
k=1 fc=l

Bearing in mind (58) and (59) it can readily be shown that the sum
of the zeros of the determining equations at all singularities will be
100] TH E GAUSS EQUATION 379

equal to

k=1 *=1
i.e. this sum will be equal to one less than the number of singularities
within a finite distance.
If we should wish to construct an equation of Fuchs’s class with one
singularity then we can always assume that this point lies at infinity so
that there will be no singularities at all within a finite distance. In the
formulae (56) we would have to assume that all coefficients A and
Ck are equal to zero, when we obtain the uninteresting equation w" = 0 .
Let us now consider an equation of Fuchs’s class with two singu­
larities, one of which can always be assumed to lie at infinity. In that
case the sums in the formulae (56) should consist of single terms and,
according to the condition (57), we obtain:

w” -)---- — w' 4 - (z —
1 2 —a
B a)2
. w= 0 ,

where a is the only singularity at a finite distance.


This equation is Euler’s linear equation [II, 42] and we know that
by a simple substitution
r - log (z - a)
it can be converted into an equation with constant coefficients.
In the following section we shall investigate in detail equations of
Fuchs’s class with three singularities.
Let us recall the Bessel equation [II, 48]
z2 w" + zw' + (z2 — p2) w = 0 .
This equation has a regular singularity at the origin z = 0. The
coefficient of w in the neighbourhood of infinity does not satisfy
the condition (55) and therefore the point z = 00 will be an irregular
singularity of the Bessel equation, i.e. the Bessel equation has two
singularities: the regular singularity z = 0 and the irregular sin­
gularity Z = oo.

100. The Gauss equation. Consider now an equation of Fuchs’s


class with three singularities. Using a bilinear transformation of the
plane of the independent variable we can, without loss of generality,
assume that these singularities lie at the points
z= 0; z= 1 and z = oo .
380 LIN EA R D IFFEREN TIA L EQUATIONS [100

We denote the zeros of the determining equation at these points as


follows:
ai> a2t Pv Pl't 7v Yi ■
We obtain the following expression for the coefficients of the
equation:

where
-t- c = .
2 0 (no)
It is given that the equation
(?((? — 1 ) + A i Q+ -®i = 0
has the zeros oq and a2, whence it follows that
Ai = 1—(<h+ a2); B1 .= cqcq
Similarly, from the determining equation at the point z = I we
obtain
A 2 = 1 (Pi + P2) > B 2 = Pi Pt ■
The determining equation at the point z = °° has the form:
Q(Q — 1) + (ai + a 2 + Pi + P2 ) 6 + ai a 2 + Pi P2 + @2 = 0 •
Substituting one of its zeros p = y1 we find an expression for C2:
C2 = — 7i(yi — 1 ) — (cq + a 2 + Pi + P2) 7i — (ai a 2 + Pi P2) >
And the condition (60) finally gives C1 = —C2. It thus appears that
in the case of three singularities the coefficients of the equation are fully
•determined by the zeros of the determining equations at the singula­
rities. It follows from the above calculations that the zeros at the
points z — 0 and z = 1 can be chosen arbitrarily but at the point
z = 0 0 only one zero can be chosen arbitrarily. The second zero is
fully determined from the condition that the sum of all six zeros is equal
to unity (one less than the number of singularities at a finite distance).
The solutions of the constructed equation are sometimes denoted
by the following symbol:
/ 0 , 1, 0 0 \
P P i , 7i , A (61)
\a2’ P2< 72 /
which was introduced by Riemann.
101] THE HYPERGEOMETRIO SERIES 381

We shall now introduce an elementary transformation of the


function w, so as to simplify the form of this equation. Notice that if we
replace w by a new unknown function u, according to the formula:
w = zp(z — \)qu\ U = Z~~p(z — l)~qW ,

we obtain another equation for the new function with three singulari­
ties: at z = 0 , 2 = 1 and z = » , but instead of the zeros ax and a2
of the determining equation the presence of the factor z~p (z— 1 )“ ?
will give at the point 2 = 0 new zeros al — q and a2 — q. Similarly
at the point 2 = 1 the new zeros of the determining equation will be
/?i — q and 02 — q. By choosing p = cq and q = 0± we can require
one of the zeros of the determining equation to be zero at the points
2 = 0 and z = 1 ; we shall assume this to be so in future.
We shall now introduce some new symbols. We denote by a and 0 the
zeros of the determining equation at the point z = We have one zero
at the point 2 = 0 which is equal to zero and the second zero we
denote by 1 — y. Finally, at the point z = 1 we have one zero equal
to zero and the other zero is determined from the condition that the
sum of all six zeros is unity; hence it will be equal to y — a — 0. Thus
instead of the general symbol (61) we can investigate the particular
case
/ 0, 1, oo \
P 0, 0, a; z (61!)
\ 1 — y, y — a — 0, 0 )
The coefficients of the equation are determined from the above
calculations, if we assume that
ai = 0 ; «2 = 1 - y; ft = 0 ; 0.t = y - a - 0; ft = a; ft = 0.
We thus obtain the following equation:

w" -f -y-Hl + q+ flz , , w = 0. (62)


2(2 - 1) "r 2 (z - 1 )

This equation is known as the hypergeometric differential equation


or the Gatiss equation. We shall now construct its solutions in the
neighbourhood of the singularities.

101. The Hypergeometric series. Let us construct, to start with,


the solutions of the equation (62) in the neighbourhood of the singula­
rity z = 0 . These solutions have the form:
^ (z ); * ~ r p t (z). (63)
382 LINEAR DIFFERENTIAL EQUATIONS [101

where P x(z) and P 2(z) are McLaurin’s series with constant terms. Let
us find, first of all, the first of the above solutions. To do so rewrite
the equation (62) as follows:
z(z — 1) w" + [ — y + (1 + a + P) zl w' + o/Smj = 0 ,
and in its left-hand side put
w1 = 1 + cx z -f c2 z2 + . ..
Using the usual method of undetermined coefficients we finally
obtain a solution in the following form:

+ 1aP
Wx = F(a, 0, y; z) = 1
!y Z +
a (a + 1

21y(y
)

+ 1) z - -
+ 1 ) 2 1

q(a + l)...(q + n — 1)/?(/?+1) .(/> + »- 1)


n \ y ( y + \ ) . . . ( y + n — l) ~l~
(64)

where F(a, P ,y;z) denotes the infinite series on the right-hand side.
Owing to the fact that the singularity nearest the origin is the point
2 = 1 we can maintain that the above series must converge in the
circle | z | < 1 . This series is usually known as the hypergeometric
series. When a = P = y = 1 the terms are in a geometric progression.
We investigated this series earlier in [I, 141]
To find the second solution of (63) we shall use the elementary
transformation of the function described in the previous section. We
replace mi by a new unknown function, according to the formula
1
w = zl~vu\ u = zY~ 1w = 2i-v w (65)

For this new unknown function the zeros 0 and (1 — y) of the deter­
mining equation at the point z = 0 become (y — 1) and zero. The zeros
0 and (y — a — P) at the point z — 1 remain unchanged and, finally,
from the second of the formulae (65), the zeros a and p at the point
2 = oo become (a + 1 — y) and (/? + 1 — y ) respectively.
In fact, before the transformation, the expansions in the neighbour­
hood of z = oo were of the form:

^1 = (|)> 1 (I) and I2= (l)/,p2(i). u

After the transformation these expansions became


101] THE HTPERQ EOMBTRIO SERIES 383
Hence the new unknown function will be described by the follow­
ing symbol:
I 0, 1, °o ^
-H 0 , 0, a + 1 — y; z .
\y — 1, y — a — P, P + 1 — y j
Comparing this Riemann symbol with the symbol (61x) and denoting
the parameters (a, ft, y) which correspond to this new Riemann symbol,
by ax, /3Xand yx we obtain:
l — Yi = Y — l\ ax= a + 1—y; P1 = P + l — y,
i.e.
ax = a + 1 — y; /3x = / 3 + l — y; yx = 2 — y .
The solution of the new equation, which is regular at the origin
2 = 0 , will, therefore, be as follows:
u = F(altpv yx; z) = F{a + 1 — y, /? + 1 — y, 2 — y; z) ,
hence, from (65), we obtain:
w2 - z1-? F(a + 1 — y, P + 1 — y, 2 — y; z ) .
This will, in fact, be the second of the solutions (63).
Let us now try to find the solutions of the equation (62) in the neigh­
bourhood of the singularity z = 1. To do so let us introduce a new
independent variable, according to the formula
z' = 1 —z .
The point 2 = 0 becomes z' = 1 and the point 2 = 1 becomes z' = 0 ;
finally z = °° becomes z' — Thus this new independent variable
also satisfies a Gauss equation and the function w will be described
by the following symbol:
/ 0, oo \ 1,
P 0, a; z '\ ,0,
\y — a — P, 1 - Y , P J
whence we obtain the following values for the parameters (a, P, y):

ai = a I Pi = P\ Y i = l + a + P — Y-
In the neighbourhood of z' = 0 we have two solutions:
F{a,p, 1 + a + P — y; z')\
z'y-n-t y — a, l + y — a — P;z'),
384 LINEAR DIFFERENTIAL EQUATIONS [101

or returning to the former independent variable, we obtain the follow­


ing two solutions in the neighbourhood of z = 1 :
wa = F(a, P, 1 + a + p — y; 1 — z)\
= (1 — zY~a~<>F{y — P, y — a, 1 + y — a — /?; 1 — z).
To construct the solutions in the neighbourhood of 2 = “ we per­
form the transformation of the independent variable:

after which the point 2 = 1 remains in its former place and the positions
of the points 2 = 0 and 2 = °° are interchanged. With this new
variable the function w is described by the following symbol:
(0, 1, oo
P ja , 0, 0; z' .
\P, y — a — P, 1 - y ,
Continuing further the transformation:

w = z'au; u = — w, (65!)

we obtain the Riemann symbol for the function u, which corres­


ponds to a Gauss equation:
/ 0, 1, oo '
PI 0, o. a; z
\P — a, y — a — P, 1 -f- a — y.

The parameters of the Gauss equation will be:


Oi = a; px = 1 + a — y; y2 = 1 + a — P ,
and we have the following two solutions of the function u in the
neighbourhood of z' = 0 :
u1 = F(a, 1 + a — y, 1 + a —/S; z');
u2 = z,fi- a F(P, 1 + P — y, 1 + P — a; z '),
whence, bearing in mind (65!) and the fact that z' = 1jz, we obtain
two solutions of the equation (62) in the neighbourhood of z =

l + a-P; |) ,
( 6 4 3)
^ = (1 )^ (0 , 1+ P 1 + /* - a; - ) ■
101] THE HYPERGEOMETRIC SERIES 385

We thus see that all six solutions obtained in the neighbourhood


of each singularity are expressed by hypergeometric series. We
assumed in all above calculations that the difference of the zeros of the
determining equation is not an integer. Bearing in mind the expansions
in the neighbourhood of the singularities we can maintain that the
formulae (642) holds when | z — 1 | < 1 and the formulae (643)
when | z | > 1 . Notice that the solution (64) also has a meaning when y
is an integer.

In [I, 141] we investigated the convergence of a hypergeometric series when


x = 1 and showed that it will converge provided the following condition is
satisfied:
y — a —p > 0 , (66)
where a, /? and y are real. In this case, according to the second of Abel’s theorems
[I, 149] we have F(a, p, y; x) F(a, p, y; 1) as x 1 — 0 and
q(q + 1) + 1) + ...
F(a,p,y, 1) = 1 + - ^ y +
21 y(y + 1)
We shall prove the formula
F(y) r ( y — a — 0)
F(a,p,y\ 1) = (67)
F(y — a) r (y —P)
Comparing the coefficients of xP it is easy to prove the relationship:
y[y — 1 — (2y — a — p — 1) x] F(a,p,y; x) + (y — q) (y — P )x F( a, p, y + 1; x)
= y(y-l)(l~ x)F (a,P ,y-\;x) (|x |< l)

y[Y — 1 — (2y —a — P — l ) z ] F(a, p,y; x) + (y — a) (y — P) xF(a, p, y + 1; x) =

= y(y - i) [i + 2 (vn - vn-\) *"]» (68)


n=i
where vn is the coefficient of xn in the expansion of F(a, P, y — 1; as). From
the condition (66) we will show th at it follows that vn ->■ 9 as n ->■ oo.
We have [I, 141]:
K l _ j | y-a-P a>„
K +il ^ n ^ ’
where the modulus of a>n is bounded as n -► oo . Let p be a positive integer
which is such th at p(y — a — P) > 1. We can then write:
\ v n \ P | [ ton
K +ilP n n%
where the modulus of cu'n is bounded. From this equation and from the in­
equality p(y — a — P) > 1 it follows th at the series, which is the sum of
the terms | « n |p converges [I, 141] and therefore «„-► 0 as n -<• o o. If in the
386 LINEAR DIFFERENTIAL EQUATIONS [101

formula (68) x tends to unity, then using the second of Abel’s theorems we obtain:
y(a + 0 — y) F(a, P, y; 1) + (y — a) {y — P) F(a,p, y + 1; 1) = 0 ,

*■<«■ft v- v - ft y + 15«■
Using this relationship several times we can write:

I-™ -1 (y — a — i:) (y — P + k) ~|
F(a, P, y: 1) = [ 77 (y + fc) (y _ „ _ p + k) J *(«, p, y + m; 1) . (69)

As m -*■ oo the product in the square brackets has the limit [73]:
F(y) r(y - a - 0)
F(Y — a) T(y — P)
We will now show th at F(a, p, y + m; 1) -► 1 as m ->■ o o . Denoting by
u^a, p, y) the coefficient of xP in the expansion of F{a, P, y; x) we can write:

IF{a, P, y + to; 1) - 11< 2J \ P,Y + ™)i•


n=l

In the numerator of the expression un(a, p, y + m) we replace a and P by


| a | and | p \ and the sum y + m by the difference m — | y |, where m > \ y\
to obtain:

I P, y + ™; i) - 11 < 2 “„(l <»1.1P I-»» - Iy I).


n=l

where the series on the right-hand side is composed of positive terms. Taking
| a | | P | (m — | y |) outside the brackets and replacing m l by (m — 1)! we
have:

| F(a,p,y + m; 1) - 1 | < \ • 2 gU„(\ a I + 1, | P \ + 1, m - | y | + 1).

When m is sufficiently large the arguments al = | a | + l, ^1 = | i8 | - [ - l


and yt = m — | y | + 1 satisfy the condition (66) and the series on the right-
hand side of the above inequality converges while its terms, and therefore the
sum on the whole, decrease as m increases. The first factor | a 11 p \ (to—|y|) 0
as m -*■ oo and, therefore, F(a, p, y + m; 1) 1, as to ->- oo. Formula (69)
finally brings us back to (67).
Using the formula (67) we can express the solution wt in terms of the linoarly-
independent solutions w3 and w t. All these solutions hold in that domain of the
plane common to both the circles, centres z = 0 and z = 1 and unit radii. We
have:
F(a, p, y; x) = Cx F{a, p, 1 + a + P - y; 1 - x) +
-f C2(l —x)Y~a-P F(y — a , y — P, 1 + y —a —P; 1 — x) .
102] TH E LEGENDRE POLYNOMIALS 387

Assuming that a, p and y satisfy the inequalities: l > y > s + /?we can also
assume in this equation that x = 1 and x = 0 and thus determine Ct and C2.
Using the formula (67), the equation (122) from [71] and the formula given
below which can easily be proved
sin na sin .t /5 = sin n(y — a) sin n(y — p) — sin ny sin n(y — a — P) ,
we arrive at the following equation:
r (y — a) r(y — P) T(a) r t f ) F(a, P, y ; x ) =
= U(a) rtf) r(y) r ( y - a - j9) F(a, p , \ + a + p - y - \ - x) +
+ r(y) r (y - P ) r ( a + P - y ) ( 1 - Xy ~ a - f X

X F(y — a, y — p, 1 + y — a — P; I — x ) . (70)

We have proved this formula for 1 > y > a + p. I t can be shown that this
formula will hold in all cases when (y — a — P) is not an integer.

102. The Legendre polynomials. Let us now consider a very impor­


tant form of the hypergeometric series. To start with let us establish
a general transformation for linear equations of the second order.
Let us consider the following equation of the second order:
a(z)w" + b(z)w' + c(z)w = 0 . (71)
We can find a factor f(z) which is such that when the first two terms
on the left-hand side of (71) are multiplied by it they become deriva­
tives of a product, i.e.

a(z) f{z) w" + b{z) f{z) w' = [a{z) f{z) w'] .

Where we must have:


&(*)/(*) = -^ K * ) /(z )].
whence
«(z) /'(z) + I>'(z) - &(z)] /(z) = 0
or
m | °'(2) b(z)
I(Z) "T" 0 (2 ) O(Z)
i.e. we can take
1 I - i-
/(z) (72)
0 (2) 6

from which we have:


r«a-dz r b(z) dz
Pi(z) = a(z) f(z) = e a(z) J (73)
<h(z) = c{z)f{z) = -^ j-e
388 LINEAR DIFFERENTIAL EQUATIONS [102

and the equation (71) takes the form:

— [Pl(2)Ml'] + ^ ( 2 ) ^ = 0. (74)

Performing the above operations with the Gauss equation (62) we


obtain it in the form:

\z'(z — 1)a+e+1-'w '] + a/fe* - 1 (2 — ly+P-rw = 0 . (75)

We shall now try to establish a general formula for the hyper-


geometric series. Differentiating the series (64) n times we obtain:
(„> a(a + l ) . . . ( q + w — ! ) / ? ( / ? + 1 ) . ■■(/? + » — 1)
1 y(y + + » - i)

or
«(g + l)...(g + w - l)fi(P + l)...(P + n - 1)
X
y(y + !)• • -(Y + n — 1)
X F{a + n, p + n , y + n; z) (76)

i.e. the nth derivative of the hypergeometric series (64) is equal to


a factor multiplied by the hypergeometric series with the parameters
a + n, @ n and y + n. Hence the function will satisfy the
equation (75) if we replace a, /? and y in it by a + n, /? + n and
f + n, i.e.

+ (a + n) (0 + n) zv- 1+n[z - i)«+0-h-» = 0 .

Differentiating this identity n times we obtain the new identity:

= - (a + n) (fi + n) — [zy~1+n(z - 1)«+fl-H-n ^n)] _


dz"
We shall write out tliis identity for the values n = 0, 1, 2 , .. k — 1
and multiply term by term the identities so obtained. The left and
right-hand sides of this identity will contain similar factors and after
102] THE LEGENDRE POLYNOMIALS 389

simplification we arrive at the required identity:


_d*
[*2 V + k - 1(2 _ j )■+/,_*+* =

= (— 1 )*a(a + 1 ).. .(a -f- k — 1) x


X /?(/?+ l)...(/5 + k - \ ) z ' ~ \ z - 1 )«+t-rWl (77)
(* = 1, 2, 3, ...) .
Let us recall that in this identity the symbol wx denotes the hyper-
geometric series (64).
Notice that a hypergeometric series generally ends abruptly and
becomes a polynomial when either a or /?, which are symmetrical in
the hypergeometric series, is a negative integer. We shall now consider
one particular case when this is so, viz. we shall take the hypergeo­
metric series
F{k + 1, — k, 1; z) {a = k + 1, /9 = — k, y = 1 ), (78)
where A; is a positive integer or zero. The function (78) will simply
be a polynomial of degree k and the coefficient of the last term zk
of this polynomial will be
(k + 1) (k + 2). ■, 2 k( — k) ( - k + 1 ) . . . ( - 1) _ 2kl
Jfct 1 - 2 . . . k \ l) (jfeip •

On supposing in the formula (77) that wx = F(k + 1 , —k, 1; z), i.e.


a = k + 1, p = —k and y = 1 we obtain coik) = ( —l)fc 2k\jk\
and performing the obvious simplifications we have:

F(k + 1 , - k, 1; z) = - J - [z‘(z - l)k] . (79)


kl dz

Let us now replace z by a new independent variable x according


to the formula
*= ^ ■ (80)

The points z = 0 and z = 1 become x = 1 and x = —1. Put

Pk(x) = F[k + 1 , - k, 1; - 1 ^ ) . (81)

Substituting (80) in (79) we obtain the following expression for the


polynomial Pk(x):
(82 )
390 LINEAR DIFFERENTIAL EQUATIONS [102

These polynomials P*(x) are usually known as the Legendre poly­


nomials. We shall use them in future in connection with spherical
functions.
We shall now explain some of their fundamental properties. The
function (79) satisfies an equation which can be obtained from the
equation (75) by writing
<z = fc + 1; = — Jr, y= 1 , (83)
i.e. the function (79) satisfies the equation

— [2(2 - 1 ) w'] - k { k + l ) w = 0.

Replacing the independent variable, according to the formula (80)


we can see that Legendre’s polynomials Pk(x) are the solutions of the
equation
_d_ [(1 _ x2) i W ] + k{]c + 1)Pk{x) = o . (84)

Let us now consider a more general equation in the form:

+ = <85>
where A is a parameter. Both zeros of this equation are equal to zero
at the singularities x = ± 1 . This can easily be seen from the form
of the equation and it also follows from the condition (83). We have
y — 1 = 0; y — a — /? = 0 .
Hence at the point x = ± 1 we have one regular solution and another
solution containing a logarithm; this latter solution will have the fol­
lowing form at the point x = 1 :
p iix — !) + Pt{* ~ 1) log (x — 1),
where Px{x — 1 ) and P2(x — 1 ) are Taylor’s series with constant
terms. It follows from this that the solution containing the logarithm
will, in any case, become infinite at the corresponding point. Notice
that when both zeros of the determining equation are the same,
the coefficient y_^, mentioned in [98], cannot vanish, i.e. when the
zeros of the determining equation are the same there must be a
solution containing a logarithm.
Let us now return to the equation (85) and take one of its solu­
tions yv which is regular at the point x = —1 . In the process of
analytic continuation of this solution along the line — 1 < x < -[-1
102 J THE LEGENDHE POLYNOMIALS 391

we obtain a solution wliich will, in general, be logarithmic at the


point x = 1 where it becomes infinite. For certain exceptional values-
of the parameter X in the equation (85), the solution, which is regu­
lar at the point x = —1, will also be regular at the point x = + 1 ,
i.e. we obtain a solution of the equation (85) which is finite in the
whole interval (—1, +1), including its ends. Such exceptional values
will be the values given below
Xk = h[k -)- 1), (86)
for which the equation (85) has a solution in the form Pk(x). It can
be shown, but we shall not do so here, that the values (86) include
all values of the 'parameter X, for which the equation (85) has a finite
solution in the interval (—1, +1), including its ends.
We shall now explain some other properties of the Legendre^
polynomials. Let us consider the equations of two different Legendre
polynomials:
-± r [ (l-x * )P 'm(x)] + XmPm(x) = 0
(n ^ m)
A [ ( l - , * ) W ] + U n W = 0.
Multiplying the first of these equations by Pn(x) and the second
by P m(x), subtracting and integrating over the interval (—1, +1) we
obtain:
(Am- K) J P J? ) p n{*) dz =
-1
1
= J {**»>(*) i [(! - - P„(X) ± [ d - **) P'JxK ) dx .

Integrating the first term on the right-hand side by parts we obtain:

J ^ m( * ) ^ [ ( l - * 2) T O ] d z =

X=1 1
= (I - X2) P J x ) P'n(x) X2) P'Jx) P'n{x) dz ,
X= —1 -1
or
1 1
J P J x ) — [(1 - Z2) P'n{x)} dx = - J(1 - z2) P'm[x) P'n(x) d x .
-1 -I
392 LINEAR DIFFERENTIAL EQUATIONS [102

Similarly
1 i
J !»„(*) [(1 - *2) ^ (* )] dx == - J(1 - x*) P'Jx) P'n[x) dx.
-1
We thus obtain:
( ^ - ^ ) f ^ ( x ) P n(x)dx = 0
-1
-or
J Pm(x) P n(x) dx = 0 {m±n) , (87)
-l
i.e. Legendre's polynomials are orthogonal in the interval ( —1, +1).
If we were to attempt the integration of the square of the Legendre
polynomial
I n = J -PnM d * , (88 )

we would find that the integral is not unity. Legendre’s polynomials


form an orthogonal but not normal system of functions. Bearing in
mind (82) and using Leibnitz’s formula we can write:
d ( x + l) * d*_1(x — 1)* 1
dx dx,*-i
We therefore have:
d*"s( x - 1)*
dV = ^ = j fcl and = 0 (« = 1 , 2..........t ) ,
dx* dx*"s x=l

from which follow the equations


P d 1) = 1 • (89)
Let us evaluate the integral Using the formula (82) we can
write
T 1 f d V - 1)" d"(x* - 1)" dx
" (nl)222nJ dx" dx"

Integrating by parts we obtain:


j = — -— r d"-^!* - 1)" d"(x2 - 1)" |x=l
n (»»!)*22" [ dx" dx" x=-i:
1
d"-1 (x2 - 1)" d"+1(x2 - 1)"
dx .
- J dx"-1 ' dx"+1
102] THE LEGENDRE POLYNOMIALS 393

The polynomial (x2 — l)n has zeros x = ± 1 of order n. Differen­


tiating it (n — 1) times we obtain a polynomial which also has zeros
x = ± 1 (of order one). Hence in the above formula the term
outside the integral vanishes. Continuing the integration by parts
we obtain in every case a zero term outside the integral and we thus-
arrive at the formula:

I l)nI Q — i r - ^ T ^ da:.
j> 2(z) da:
(n!)222n J v ' dx2n
—1
We have further
d2n(a:2- 1)" _ d2n
(x2n + . . . ) = 1-2 . . . 2 n = (2n)\
dx2n dx2"
and consequently:
x
n (w + l ) ( w + 2 ) . . . 2 n
f « d * = ( - l ) J ( * 2 - l)n d x .
n\ 22

Putting x = cos q>we obtain:

J" {x2 — l)"dx = (— 1)" J' sin2n+19?d<p = (— l)n 2 f sin2n+1 q>d<p ,

i.e. [I, 100]:

2 • 4... 2n
J (x2 — l)n dx = (— l)n 2 ,(2 n + l) ’

and the above formula finally gives:


i
J>n(*) dx (90),
2n + 1 '

Using the expression (82) and applying Rolle’s theorem it can


easily be shown that all zeros of the polynomial Pn(x) are different
and lie in the interval —1 < x < + 1. In fact, the polynomial
d(x2—l)n/dx of the (2n— l)th degree has zeros x = ± 1 of order (n —1);
according to Rolle’s theorem there is still another zero x = a in the
interval ( —1, +1). There are no more other zeros. The polynomial
d2(x2—l)n/dx2 of the (2n — 2)th degree has the zeros i = ^ 1 of order
(n — 2); also, according to Rolle’s theorem, it has two more real.
394 LINEAR DIFFERENTIAL EQUATIONS [103

zeros, one in the interval (—1, a) and the other in the interval (a, +1).
Continuing in this way we can see that Pn(x) has n different zeros in
the interval (—1, +1).

103. Jacobian polynomials. Legendre’s polynomials are only one


particular example of polynomials obtained when a hypergeometric
series ends abruptly and becomes a polynomial. We shall now
investigate the general case. Let us introduce the following notation:
y — l= p ; a + p —y = q (91)
and suppose that p and q are fixed numbers, greater than (—1);
we shall always assume that the parameters a, ft and y are real
numbers. If a hypergeometric series is to end abruptly and become
a polynomial of degree k we must assume that either a or ft is equal to
(—k). Without loss of generality we can, for example, assume that
ft = —k and determine a and y subsequently from the equation (91).
The polynomial so obtained we denote as follows:
Q[P’q\z) = Ck F(p + q + k + 1, — k, p + 1; z ). (92)
where Ck is an arbitrary constant.
Applying to the above hypergeometric series the formula (64) we
can see that the coefficient of zk in the polynomial (92) is equal to
/_ (P + g + k + 1) (P + g + k + 2). ■. (p + q + 2k) p
K ‘ (P + l)(P + 2)-.-(p + k)
Applying the formula (77) to the polynomial (92) we have:
^!(P + 9' + & + l ) ( p + ? + fc + 2 ) . . .
.. . (p + q + 2k) zP(z - 1)0 Q{P’0\z) =
_/ __ I \k (P + g + ft + l )( P + g + fc + 2).--(P + g + 2k) fc!
' 1 ( p + l ) ( p + 2)...(p + k)

x c k-^ [ z p + k (Z- i r * ] .

The constant Ck is determined from the formula


n _ ( P + 1 ) (p + 2)...(p + fc)

In this case we obtain the following formula for the constructed


polynomial:
zP{z - 1)qQiP’q\z) = (~ 1}- [zP+k{z - l)?+k] .
103] JACOBIAN POLYNOMIALS 395

If we replace 2 by x, according to the formula (80), we obtain poly­


nomials in x which are known as Jacobian polynomials:

(1 - x)P (1 + x f P i ^ \ x ) = [(1 - x)P+" (1 + *)*+*]. (93)

When p — q = 0 these polynomials are the same as Legendre’s


polynomials. When k = 0, P$'q\x ) = 1.
It follows directly from the determination of Ck that the coefficient
of zk in the polynomial (92) is equal to:
I_ 1 (P + g + f e + l ) ( P + g + fc + 2)...(p + g + 2k)
( > *1
and in the polynomial iV ^ ^ x ) the coefficient of xk is equal to:
(P + 9' + f e + l ) ( P + 3 + fc + 2 ) . . . ( p - ( - j + 2k)
ak ~ zTo* •

In the case under consideration we have:


a = p + q+ k+ l, p = — k, y = p + 1,
and the function (92) is a solution of the equation

- ^ - [ 2p+1(z - 1Y+1w'] - k ( p + q + k + 1) zP(z —- 1)«w = 0.

Replacing the independent variable according to (80) we obtain


the following equation for Jacobian polynomials:
d dP<P.«) (x) j +
dx [(1 - x ) P +1(l + Z),+1 dx (94)
+ k(p + q + k + 1) (1 - x)P (1 + xYPtf' «>(x) = 0.
In this case the Jacobian polynomials (93), when p and q > 0, are a
solution of the following limit problem: what is the value of the
parameter X for which the differential equation

^ [(1 - x)P+i (1 + x)*+i - g ] + X(l - x)P { \ + x ) * y = 0 (95)

has a finite solution in the interval (—1, +1), including its ends.
The values of this parameter are
Xk = k(p -(- q + k + 1), (96)
and the corresponding solution is given by Jacobian polynomials.
396 LINEAR DIFFERENTIAL EQUATIONS [103

Using the equation (94) for Jacobian polynomials it can readily be


shown that, as for Legendre polynomials, the following equations
hold:
i
f ( l - x ) P ( l + x ) i p %’Q{z)PK''»(x)dx = 0 (m*n). (97)
-i
The properties (97) can be expressed as follows: Jacobian poly­
nomials are orthogonal in the interval ( —1, +1) with the weight
r(x) = (1 — z)p (l + x)q. (98)
As with Legendre’s polynomials, we can deduce from the formula
(93), that
P[P>o\l) = (P + *> (P + f c - 1 ) •••(? + !)

Let us now evaluate the integral:

Ik = f (1 - x)P (1 + x)0 [P[P- «>(*)]» da:.


-1
Using the formula (93) we can write:

I k = " T ^ r / p<kP’q){x) [(1 ~ x)p+k ( 1 + x)q+k] dx ■


—1
Integrating by parts, as in [102], we obtain:

h = ~ *)r*‘ (1 + *),+‘ d‘ V x> d* ■


-l
since the terms other than the integral vanish when p > —1 and
q > —1. Denoting as before the coefficient of x? in the polynomial
I f ' * ( x ) by ak we obtain:
1
I k = - p - J (1 - a;)p+fc (1 + x)q+k dx .
—1
Introducing a new variable of integration t = (1 - *)/2:
I k = ak 2P+9+fc+i f tP+k( 1 — t)q+k d*,
6
i.e. [72]:
I k = ak 2P+9+k+iB (p + k + 1, «7+ & + l) =
_ a op+g+k+i r(P + fc + 1) A g + h + 1)
* r(p + g + 2 k + 2 )
10 3 ] JACOBIAN POLYNOMIALS 397

or, substituting the above expression for ak and using the formula
(120) from [71]:
2P + q + X r(p + k + \)r(q + k + \)
k p + q + 2k + 2 ’ k<r(p + q + k + l)
i.e. the formula holds:
i
J (1 - x)P (1 + x)« [P<nP. «)(*)]S dx =
-1
2P + ? + 1 r ( n + p + 1) r(n + q + 1)
+ p + q+ 1 n\ r(n + p + q + 1) ( » = 1. 2,. . .). (100)

When n = 0, owing to the fact that r (x + 1) = x 71(x), the latter


expression has the form:
np+ii+i -f(P + 1) f(g + 1)
A p + ff + 2) '
Notice another particular case, viz. when p = q = —1/2. We shall
introduce special symbols for these polynomials.
- 1)
Tk{x) = CkPk 2 2'(x ), (101)
where Ck is a constant.
They are determined, from (93), by the following relationship:

( - l )kCk dk
( I - * 2) 2 T k(x) = k\ 2* [ ( i - * 2) ( 102 )

We shall now introduce another expression for these polynomials


i.e. a differential equation which they must satisfy. These equations can
be obtained from the equation (94) by assuming that p = q = — lj2.
We have for Tk(x) the following equation:

+ = (103)

The zeros of the determining equation at the singularity x = 1


will be z and l/2. The first zero corresponds to a solution in the form
of a polynomial, but the second solution will not be a polynomial.
To find a polynomial in a convenient form which would satisfy the
equation (103) we replace x by a new independent variable <paccording
to the formula
x = cos <p. (104)
Differentiating with respect to <p instead of x we have, according
to the differentiation law for complicated functions:

Substituting this in the equation (103) we obtain

d''Tf ° S(P) + V T k(cosy) = 0 .


The latter equation has the following solutions
cos k<p and sin k f ,
and for the equation (103) we obtain solutions in the form
cos(k arc cos a:) and sin(& arc cos x ).
Using the formula from [I, 174]

cos k<p = cosfccp — cos4-2 <psin2<p+ .. .,

we can see that the first of these solutions is a polynomial in x and,


consequently, except for the arbitrary term, the polynomial which
gives the solution of the equation will be
T k{x) = cos(k arc cos a:), (105)
and this polynomial is known as Tschebyshev’s polynomial. When
<p = 0 we have x = 1 and, consequently, Tk(1) = 1; on the other
hand, from the formula (99):

l-3...(2Jfc-l)
** ^ = 2 •iT772k---
whence the constant term in the formula (101) can easily be deter­
mined.

r » W ° - 1.32.,4 ' ( a - 1) f »(~ T ’ ~ T )W - (106>

104. Conformal transformation and the formula of Gauss. We shall


now try to explain the connection between the equation of Gauss and
a certain problem of conformal transformation; we shall again assume,
as in the previous paragraph, that the coefficients a, /3 and y are
real. We will prove, first of all, that the solution of the equation of
Gauss (62) cannot have multiple zeros in the plane of the complex
variable, except at its singularities. For suppose we have at a point
z = z0 a zero of order higher than one, i.e.
w(z0) = w'(z0) = 0 ,
then it follows from the equation (62) that w"{z0) = 0. Differentiating
the equation (62) and putting z = z0 we obtain w"(z0) = 0, etc.
However, it is well known that if all the derivatives of an analytic
function vanish at a certain point then the function is identically
zero; but we assumed that w gives a solution which does not vanish.
The above proof is also valid for any linear equation of the second
degree with analytic coefficients. The result obtained can also be
deduced directly from the existence and uniqueness theorems [95].
Let us now consider the quotient of two solutions of the equation
of Gauss:
= • (1OT)
When this function is analytically continued it can only have
singularities at the points z = 0, 1 and °°, as well as at points at
which the values of z are zeros of the solution wx(z). These values
of z will be simple poles of the function (107). If u\(z0) — 0. we can
say that w2{z0) ^ 0. In fact, if we had
w2{z0) = 0,
then our two solutions would be determined from the initial conditions
u>i(20) = 0; wj(z0) = a ]
a and /9 ^ 0 ,
“>2(Zo) = 0; wl(z0) - P J
and, according to the existence and uniqueness theorems, we would
have had
a
^ { z ) = —Wi[z),

i.e. the solutions wx(z) and w2(z) would be linearly dependent; we are
given, however, that the numerator and the denominator in the
formula (107) are linearly independent solutions.
Consider the upper half-plane of the complex variable z. In this
connected domain B the analytic functions io^ z) and w2{z) have no
singularities in the course of analytical continuation and, con­
sequently, are single-valued regular functions of z. The function (107)
will also be single-valued in the upper half-plane, where its only
singularities are simple poles. We will now show that the derivative
400 LINEAR DIFFERENTIAL EQUATIONS [104

of the function (107) cannot vanish except at the poles. We have,


from [II, 24], the following expression for this derivative:
cl [ w2(z) ] _ C - SpW dz (108)
dz I w^z) ) ~ w\(z) e
where C is a constant and p(z) is the coefficient of w' in the equa­
tion (62).
Our hypothesis follows directly from the formula (108). Bearing
in mind that conformity is not impaired at a simple pole, we can
say that the function (107) gives a conformal transformation of the
domain B into a new domain Blt in which there are no branch-points.
We shall now determine the contour of the new domain Bv
When the point z, in the upper half-plane approaches a point z0
on the real axis, which is not one of the singularities 0, 1 or the
function (107) tends to a definite limit and, moreover, it will remain
regular at the point z0; it will also be regular on each of the following
three lines:
(— oo, 0); (0, 1); (1, oo) (109)
on the real axis. We will now show that the function (107) will also
tend to a definite limit even though z tends to one of the singularities.
Consider, for example, the point 2 = 0.
To start with let us explain one circumstance which will be
important in what follows. Suppose that instead of wx(z) and w2(z)
we had taken two other independent solutions of the equation, viz.
w*(z) and w\(z). These can be expressed linearly in terms of the
former solutions:
w*(z) = an w^z) + a12w2(z),
w*,{z) = a21 wt(z) + a22w2(z),
where
flil a22 Oj2 ®21 ^ ^ •
Let us construct a new function rj*(z) using these new solutions:
„ * ( Z) = = °il Wi(z) + Oi2 U>2(Z)
' V' wj(2 ) O,! W,(r) -I- o 12w2(z)
or
°21 + aZ2q(g)
if{z)
“ll + “12»?(2) ’
i.e. when choosing different independent solutions in the formula (107)
the corresponding functions T](z) will be simply connected with each
104] CONFORMAL TRANSFORMATION AND THE FORMULA OF GAUSS 401

other by a bilinear transformation, the determinant of which is not


zero.
We shall now investigate the function (107) in the neighbourhood of
the point z = 0. We select the independent solutions as follows:
w^z) = F(a, p, y; z);
( 110)
w2(z) = z1-* F(a + 1 — y, 0 + 1 — y, 2 - y; z ) ,
Also
F(a + 1 —y, P + 1 —y, 2 — y; z)
Tj(z) = Z1—v
F(a, p, y ; z)
(111)

This latter formula must be interpreted as follows: in the neigh­


bourhood of z = 0 the function r\(z) is determined by the formula
(111); in the half-plane B this single-valued function can be determined
by analytic continuation. It follows from the formula (111) that, for
example,
77(2) —>■0, if z —>-0 and y < 1.
If we choose any other solutions, the new r;(z) will be expressed
by (111) as a bilinear transformation, and, consequently, it will also
have a definite limit as z —>- 0.
We will now show that the function (107) transforms the lines (109)
into circular arcs. For example, let us consider, the line (0, 1) and
take a point z0 on this line. We shall determine the solutions w2(z)
and u>2(z) at the point z„ from the initial conditions; we select these
conditions so that w(z0) and w'(z„) are real. Bearing in mind the
fact that the coefficients in the equation of Gauss are also real
we obtain for w+z) and w2(z) Taylor’s series with real coefficients in
the neighbourhood of the point z0. The analytic continuation of
these solutions along the line (0, 1) will evidently also lead to Taylor’s
series, in other words, for this choice of solutions the function
7](z) will have real values on the line (0, 1), i.e. it will transform
this line into another line on the real axis. For any other choice of
solutions the new function r?(z) will be obtained from the former
function by the bilinear transformation which transforms the line
on the real axis into the arc of a circle. Therefore the function (107)
will, in fact, transform each of the lines (109) into the arc of a circle.
Let us again consider the case when the fundamental solutions
wY(z) and w2(z) are real on the line (0, 1). Applying the formula (108)
to this line we can see that the derivative of the function r\{z) does
not change its sign on this line, i.e. the function r\(z) is a monotonic
402 LINEAR DIFFERENTIAL EQUATIONS [104

function of the variable z on that line. In other words, if the point


z moves along the line (0 , 1 ) in a definite direction, the point 77(2)
always moves along the corresponding line in the same direction.
Notice that the point rj(z) can also pass through infinity, so that the
line, described by the point 77(2), can be infinite. In certain cases
this line can overlap itself. In the general case, when the independent
soutions in the formula (107) are chosen arbitrarily, the point 2
moves along the line (0 , 1 ) in a definite direction and the point 77(2)
moves along the arc of a circle always in one direction; in certain
cases it is not an arc of a circle but a full circle overlaping itself which
corresponds to the line (0 , 1 ).
From what was said above we obtain the following result: the function
(107), i.e. the quotient of two independent solutions of the equation of
Gauss, conformally transforms the upper half-plane into a domain bounded
by three circular arcs or, in other words, into a circular triangle which
contains no branch-points. Let us now determine the angles of this
circular triangle. Take the apex A of the triangle which corresponds
to the point 2 = 0. Select the fundamental solutions by using formula
(110) and assume that y < 1 . Turn to the formula (111). In the
neighbourhood of the point 2 = 0 we have 77(2) > 0 , when 2 > 0 and
we assume that arg 2 = 0 . Describing the point 2 = 0 from the upper
half-plane we obtain arg 2 and, consequently, arg zx~v = jr(l — y);
the fraction in formula (111 ), when 2 is close to zero, will be real
and close to unity. Hence assuming that y < 1, we obtain two straight
lines in the plane 77(2); one of these goes from the origin in the direction
of the positive part of the real axis and the other makes an angle of
ji(1 — y) with this direction. If y > 1, then instead of the relationship
(111) we must take the inverse relationship. Thus for the given
choice of fundamental solutions we have in our circular triangle an
angle n | 1 — y | at the apex corresponding to the point 2 = 0. For
any other choice of fundamental solutions we obtain another triangle
which can be obtained from the former by a bilinear transformation.
We know that this transformation conserves the angles and therefore
we obtain, in general, the angle ji | 1 — y | at the apex A. The pro­
cedure is similar when we determine the other two angles of the circular
triangle, which correspond to the points 2 = 1 and 2 = 00. We obtain
then two angles equal to n | y — a — /? | and n | /S — a | respectively.
The direction in which these angles are measured is determined, as
always in conformal transformation, from the fact that when the point
moves along the real axis in the positive direction the corresponding
106] IRREGULAR SINGULARITIES 403

point moves along the contour of the circular triangle so that this
triangle lies to the left of the moving observer.
The above result can be formulated as follows: an angle of the triangle
in the rj(z) plane is equal to the product of n and the modulus of the
difference of the zeros of the determining equation at the corresponding
singularity of the equation (62). We notice, without going into the proof,
that this will still be so when the difference is equal to zero (the
arcs of the circles touch), or to an integer.
It can be shown conversely, that any circular triangle, even one
with several sheets but without branch-points either inside or on
the sides, can be obtained from the upper half-plane as a result of
the conformal transformation by the quotient of two solutions of the
equation of Gauss, when the parameters a, /? and y are suitably
chosen. In particular the usual triangle with straight sides can be
taken: this is a particular case of a triangle bounded by circular
arcs. In this case we can express the function which performs the
conformal transformation more simply, viz. by means of Christoffel’s
formula.

105. Irregular singularities. We shall now deal with the problem


of obtaining a solution in the neighbourhood of an irregular singul­
arity. By performing a bilinear transformation of the independent
variable we can always achieve the fact that this singularity should
lie at infinity and we shall assume in future that this is so. Consider
the equation
w" + p(z) w' + q(z) w = 0 .

If p(z) and q(z) have the following expansions near infinity:

= = (112>
*=1 * k=2 2

then, as we know from [99], this point will be a regular singularity.


We now assume that the coefficients cannot be expanded in accordance
with (112), but that the expansions of p(z) and q(z) near infinity do
not contain positive powers of z, i.e. we consider an equation of
the type:

^ + ^ o + ^ L + -5- + . - - ) ^ + ( & o + 4 + l | - + - - - ) M, = 0> (113>


404 LINEAR DIFFERENTIAL EQUATIONS [105

where at least one of the coefficients a0, b0 and is not zero. This
equation must be formally satisfied by an expression of the form:

2®(c0 + Ci— + • • •) , (114)

where c0# 0 ; substituting this in the left-hand side of the equation


we obtain a single term which contains 2®, viz. a term of the form
&0c„2®. Hence it will not be possible to satisfy our equation formally
by an expression of the form (114) if b0 ^ 0. To eliminate the coefficient
b0 we replace w by a new function u, according to the formula
w = e azu .
Hence
w' = eaz u' + aeaZ/u; w” = e“2u” + 2aeazu' + a2eaz u ,

and, substituting in the equation we obtain a new equation

u" + |2a + a0 H— -f- . . . j u' +

+ («2 + ao0 + b0+ a°l + b' + aa>+ b'-. + . . . ) « = 0 .

The constant a now remains to be chosen from the condition


a2 + aa0 + 6„ = 0 . (115)
As a result we obtain an equation of the form:

u" + (2a + a0 + — + (-^- + .ju. = 0 (116)

(b'k = aak + bk) ,

where a is a root of the equation (115). This equation can be formally


satisfied by an expression of the form (114). Let us suppose that
u= 2 ®v ,
hence
u' = 2®l/ + g2®-1 V\ u" = 2®u" + 2g2®-1 v' + q(q — 1) 2«-2 V .
Substituting in (116) we have for the new function v the following
equation
v ” + Pi(2) »' + ?i(z) v = 0 . (117)
105] IRREGULAR SINGULARITIES 405-

where
Pi(z) = 2a + a0 + 2e + 0< + ,
?^ _ (2a + Qq) e + K e(e —i) + Qi e + b't
(118)'
, Qg g + , a 3 Q + fe« I
"T 23 "I" 24 -t-
We will now determine q from the condition that the coefficient
<h(z) must not contain a term in z-1, i.e.
(2a + a0) q + b[ = 0; Q= — ' (H 9>
If we assume that the equation (115) has different zeros, it follows
that 2a + a0 ^ 0.
The new equation for v will be:

v" + (2a + a0 + 2e + ai + . . . ) » ' +

+ ( g2 + (a| 721} 9 + bl + °*e^ + . ..) v = 0 . (120)


This equation can be formally satisfied by the series

v = c0 + ^ - + ^ r + . . . (121)

Differentiating, substituting in the left-hand side of the equation


and equating coefficients of like powers of a; to zero we obtain a
system of equations from which cv c2, ■. ., can be determined suc­
cessively and where c0 will be an arbitrary factor. Let us write the
first of these equations
— (2a + a0) c± + [p2 + [d1 — 1) Q+ b’i) = 0,
Hence
„ _ e2+ (“i —1) e + aa2+ &2 „
- 2a+a0 - c«
We finally obtain for equation (113) a solution of the form:

w = e°*#{cQ+ - ^ + -^r + (123)

If the quadratic equation (115) has different zeros then by using


every zero we can construct by the above method two expressions
of the form (123). It appears, however, that the infinite series in
the expression (123) will, in general, be a divergent series for every
value of z.
406 LINEAR DIFFERENTIAL EQUATIONS [105

We will show this to be true in one particular case. Consider the


equation
W + [a0 + -^-)w ’ + -^-u, = 0. (124)

In this case we can assume that a = q = 0 and, substituting in


the left-hand side of the equation (124) a series of the form (121),
we obtain the following formulae for the determination of the coef­
ficients:
[n{n + 1) — nax + b2] cn — (n + 1) a0cn+1 = 0 .

Consider the relationship of two successive terms of the series (121).


Using the above formula we obtain the following formula for this
relationship:
c„+i . _£n_ _ n(n + 1) - wo, + b2 1_
zn+1 zn (n + 1 ) o0 z

from which it follows that for any given z the above relationship
tends to infinity together with n and, consequently, the constructed
infinite series cannot converge for any value of z.
The divergence of the series in the expression (123) at first sight
appears to make the above relationship devoid of meaning. It becomes
evident, however, that this expression can be used to give the solution
of the equation (113). To explain this circumstance we must introduce
a new concept, viz. the concept of the asymptotic expansion of a
function.
We shall lastly consider the case when the equation (115) has a
double zero. In this case 2a + a0 = 0 and the equation (116) becomes:

* ' + (— + ^ r + •••)«' + ( 4 + - ? - + •■•)“ = °-


Replacing z by a new independent variable t = |/z we obtain the
equation:
d2u ( 2a, — 1 . 2az 2a3 'j du .
1 + { I r ~1?~ + T + • ■■J ' d T +

+ (4&i + ^ - + i ^ + . . . ) « = 0. (125)

For this differential equation the quadratic equation (116) becomes


a 2 + 4b[ = 0 and, provided b[ # 0, we have for the equation (125)
the case of the different zeros which we considered above. If, however,
b[ = 0 then t = °° is a regular singularity of the equation (125).
106] ASYMPTOTIC EXPANSION 407

106. Asymptotic expansion. Let us suppose th at we are given an


infinite series
c» + - F + l i - + --- (126)
and denote by 8n(z) the sum of its first n terms:

S n(z) = c0 + - ^ + .
2 Z" 1
The convergence of this series is equivalent to the existence of a
limit Sn(z), as n increases indefinitely. We shall now consider it diffe­
rently viz.: we suppose that n is fixed and z tends to infinity along a
definite straight line L. In future we shall assume that this straight
line is the positive part of the real axis, i.e. z > 0.
We now suppose that a function /(z) is determined on L, so that for
any fixed n the difference
/(*) - S„{e)
as z-*- °o, represents an infinitely small quantity of an order
greater than (l/z)n~ l, i.e. the difference /(z) — 8n(z) is an infinitely
small quantity of an order higher than the last of the terms in the
expression Sn(z). The above conditions can be written as follows
lim [/(z) - <STn(z)] z"-1 = 0 (on L ) . (127)
r —►cd

We usually say that the series (126) gives the asymptotic expansion
of the function f(z) on L and it can be written as follows:

+ + + (on L). (128)

Bearing in mind that (c„/zn)zn *- >0 as z - > o o we can, instead of


the condition (127), write the following equivalent condition:

lim z"-1 = 0. (129)

Consider, for example, the function which is determined when x > 0, by


the integral
f(x) = j' t 1 e* *df . (130)

Performing successive integrations by parts we can write:

(-1 )" -1(n —1)1


/( * ) = - - -L + - 5 - _ . •+ + (-l)"»lJ. dt.
X2 X3 x" f n+ 1
408 LIN EA R D IFFERENTIAL EQUATIONS [106

Construct the series


1 1 | 2! A +... (131)
X X2 X3 X * ^

On considering the relationship of two successive terms it can readily be shown


th at this series will diverge for every value of x. We will show th at it gives the
asymptotic expression of the function (130). We have in this case:

f ( x ) - S n+i(x) = ( - 1 )™n ' J ~^i+T d* -


X

whence, since t > x, the factor ex~1lies between zero and unity, and we obtain:

I/<*) - s n+,(*) I < m j — - = (» —i)! — ,


X

from which the condition (129) follows directly and, consequently:

1_________ 21____3I_
J / - 1e*-'dt X X 2 ' X 3 X4
(132)
X

Suppose that we are given the asymptotic expansion (128). When


n = 1, the condition (127) gives
lim [f(z) — c0] = 0 ,
i.e.
c0 = lim f(z) .

Furthermore the same condition, when n = 2, gives

j
lim f(z) — c0 — 2 = 0 ,
whence
cx = lim [/(z) — c0] z,
Z—►«

and in general we have:

C„ = l t a [ / ( * ) - ( C„ + - i + . . . + ^ ) (133)

These formulae determine in a unique way the coefficients of the


asymptotic expansion when this expansion exists. It follows that
a given function can only have one asymptotic expansion.
106] ASYMPTOTIC EXPANSION 409

Consider the function e x on the straight line x > 0. We know


that for every n we have:
lim e~x xn = 0 ,
X - ►«

i.e. the asymptotic representation of the function e~x on the straight


line x > 0 will be e-x -> 0 . Hence, for example, if a function f(z)
can have an asymptotic expansion along the straight line x > 0 ,
then the function f(x) + e-x will have the same asymptotic expansion.
We can see in this case that the addition to the function f{x) of
the term e“x which decreases more rapidly than any whole negative
power of x, does not change the asymptotic expansion of the function.
From the definition of the asymptotic expansion the laws of term
by term multiplication and term by term integration of asymptotic
expansions can be proved, viz. when:

f(z) ~ ■“*" and


k =o 2 fc=o
then
f{z) <p{z) ~ J p e*d»+ c*=ld«t+--- + cod* .
fc=o 2
Similarly, when
/(*)
k =2 z
then

J
f /(z)dz~2
k=2
ck
( k - l ) z k~ l ‘

We shall not give the proofs of these laws, which follow directly
from the definition of the asymptotic expansion.
It can be shown that the infinite series in the expression (123)
gives the asymptotic expansion of a certain function, viz. a solution
of the equation (113) exists for which the following asymptotic
expansion holds on the straight line z > 0 :

w{z) e - az z~e ~ c„ + -f -^§- + . . .

We shall prove this for a particular case of the equation (113) viz.
for the case when both ak and bk vanish for k > 2 . We shall use
in this proof a special method for integrating the equation (113),
410 LIN EA B DrFFBBKNTLLL EQUATIONS [107

viz. we shall give the solution of this equation in the form of a con­
tour integral. To start with let us try to solve the problem of inte­
grating the equation by means of a contour integral.
Above we found that the condition (127) is satisfied when z lies
at infinity with reference to a straight line L. If this condition is
satisfied when z moves to infinity along a certain sector then it is
said that the asymptotic representation (128) takes place in this
sector.

107. The Laplace transformation. Consider the equation


W + [o0 + -y -) W + (&o + —-) w = 0

or, multiplying by z
zw" -f (a0 z + a x) w' + (b0z + 6 X) w = 0 . (134)
We shall try to find the solution of this equation in the form:
J
w(z) = v(z') e22' d z ', (135)
i
where v{z’) is the required function z' and I is the required path of
integration which is independent of z. Differentiating with respect
to z we have:
to'(z) = J v(z') z' ea' dz'; w"{z) = f
w(z') z'2 ezz>d z '. (136)
i i
Multiplying by z and integrating by parts we obtain:

zw(z) = J v(z') de22’ = [u(z') e^']/ - J ^ P ~ d z ',


i i
where the symbol
to m
denotes the increment in the function <p(z'), when z' has described
the contour I. Similarly we have
zw'{z) = \y{z')z' ea’], - J g>] e22' dz'
i
and
zwn(z) = I>(z') z'2 e22'], - J dNgy /2]- e**' d z '.
i
We must, first of all, make the condition that
\v(z') (z'2 + a0z’ + 6„) e22'], = 0 . (137)
107] TH E LAPLACE TRANSFORMATION 411
Substituting the above expressions in the left-hand side of the
equation (134), the terms outside the integral will vanish, as a result
of (137), and we can write this equation in the form:
rfd[® (z')z'*] , _ d|> (*')*'] , a de(z')
J i ---- dP----- + a»— d ^ ~ + bo~dF~

J
— ax z' v(z') — bx v(z') ea ' dz' = 0 .

This equation will certainly be satisfied if we determine the function


v(z') from the equation

d M E M . + a0 W-2 - + b° - ai z' VW ~ h W = 0 . (138)


Consider the quadratic equation
z'2 + a0z '+ 60 = 0, (139)
which is the same as the equation (115), and assume that it has two
different zeros ax and a2. The equation (138) gives:
1 d« _ (a, — 2 ) z' 4 - (2>t — a 0)
v dz' (z' — at) (z' — a2)

or, converting the fraction into partial fractions:


1 dv p —l q—l
v dz' z' — dj z' — aj ’
(140)
where
(o, — 2 ) a, + (6 t — o0) + (a, — a2) .
V= at a2 ’
(ot — 2 ) a 2 + (6 i — a„) -f (a2 — a,)
“ 2 — at

On the other hand, we obtain from the quadratic equation (139):

al + a2 — — a0 i
and the above expressions for p and q can be transformed as follows:
_ °i °i + . _ 0102+ 61 (141)
y 2at + o0 ’ a 2a2+ a0
Comparing this with the formula (119) we can see that

p = ~ e 1; g = ~ e 2 > (142 )
where gx and g2 are two different values of g which correspond to the
two different zeros ax and of the equation (115).
412 U N B A R D IFFERENTIAL EQUATIONS [108

On solving the equation (140) we have


v(z') = 0(2' - a,)?-1 (2' - a ^ - 1, (143)
and, consequently, the solution of the equation (134) will he
w(z) = C J (z’ - aJP-1 (2' - a2)i~l e«' d 2' , (144)
I
where C is an arbitrary constant and the contour I, as a result of (137)
and (143), must satisfy the condition
[(2 ' - ai)P (2' - a2)<e“ '], = 0 . (145)
108. The choice of solutions. On choosing the contour I, which
satisfies the equation (145) in different ways we can obtain
different solutions for the equation (134). This equation, like the
Bessel equation, has a regular singularity at z = 0 and an irregular
singularity at z = The determining equation at the singularity
2 = 0 will be
Q{Q ~ 1) + aiQ = 0 -
and its zeros will be = 0 and p2 = 1 — <q; we assume, for simplicity,
that 1 — cq is not a positive integer. Hence one of the solutions of
the equation (134) will be a regular function at the point 2 = 0 and
this solution can be represented in the z plane by the series:
1+ Cj 2 + C2 2 2 + ... (146)
We shall first of all, indicate a choice of the path of integration I for
which the formula (144) gives this solution which is regular at the origin.
The integrand in the expression (144) has singularities at the points
z' = cq and z' — a2, which will, in general, be branch-points, since
p and q will, on the whole, not be integers. When circumscribing
the point z’ = <q in the positive direction the above integrand will
gain the factor e*p-1*27" = ep2m and when circumscribing the point
z' = a2 it will gain the factor e(9~1)2n‘ = eq2nl. In future we shall
assume that the numbers p and q are fractions.
Consider a finite point z0other than cq or a2in the plane and denote
by lx and l2 closed contours which originate at z0 and encircle the
points <q and a2.
We denote the symbol by (IJL, l2) the contour which consists of the
following successive circuits: the circuit along l2 in the positive
direction, the circuit along l2 in the positive direction, the circuit
along l2in the negative direction and the circuit along l2in the negative
direction. On completing the first circuit the function (145) gains
108] THE CHOICE OF SOLUTIONS 413

the factor ep2nl. After the second circuit is completed it gains the
factor e?2m; after the third circuit the factor e _p2,,/ and, finally,
after the fourth circuit the factor Thus when we finally return
to the point z0 we have on the left-hand side of (145) the same branch
as the one we took by starting out from z0. Thus by taking for I the
contour (lv l2) we satisfy the condition (145) and formula (44) gives
the solution of the equation. Notice that had we taken for the contour
I a closed contour which did not encircle the singularities cq and a2
of the integrand then this function would, of course, also have returned
to the initial value but, according to Cauchy’s theorem, the integral
round this closed contour would not have been equal to zero and
we would have obtained the solution of the equation (134). In this
case we have chosen the contour (lv l2) so that by describing the
singularities we have, nevertheless, returned to the initial branch
of the function.
We therefore have the solution
w{z) = C f (z' - a^P -1 (z' - a ^ - 1e22' d z '. (147)
(/;, i.)
The variable z' lies on the contour consisting entirely of finite
points and we can therefore write the expansion of the series
a zz> _— "S — z'k
~ 2A =,0 k\ ’

which is uniformly convergent on the contour of integration. Sub­


stituting this series and integrating term by term we can write our
solution in the form:

w0(z) = C 2 i r \ 2'fc(2' - ai)P_1 (*' - 1dz' - (148)


k=°
where C is an arbitrary constant, i.e. the constructed solution happens
to he a solution which is regular (at the origin). It must, however, be
remembered that this solution is not identically zero; this can occur
only in exceptional cases, when p and q are positive integers.
It can readily be seen that the value of the integral in formula
(148) is independent of the choice of the initial point z0. This can
be shown, for example, by applying Cauchy’s theorem but it must
be remembered that the complete contour (lv l2) must be closed, for
when describing it fully we return to the starting branch of the function
and, therefore, the use of Cauchy’s theorem is permissible.
414 LINEAR D IFFERENTIAL EQUATIONS [108

We shall now consider the particular case when the real part of the numbers
p and q is greater than zero. We shall assume that the point z 0 lies on the straight
line at tij near the point a,, th at lt is a small circle, centre a„ and th at lt consists
of a straight line z 0 z, and a small circle, centre a2, where the above straight
line must be described twice. We will show th at when the radii of the above
circles become indefinitely small the integrals round these circles tend to zero.
Consider, for example, the circle, centre a1( and assume, for simplicity, th at
p is a real number which, from the given conditions, must be greater than zero.
Let e be the radius of the circle. On this circumference we have the following
inequality for the integrand:
| z'k(z' - a,)?"1 (z' - a^ - 11= | z' - a, I '- 11z'*(z' - a*)*"11< ep~1 M ,
where M is a positive constant. For the whole integral over the above circum­
ference we have the inequality:
| J z 'k (z' - aI)p—1 (z' - a / -1 dz' | < ep-1 M2ne = ep 2nM,
from which it follows th at the integral tends to zero together with e. For the
complex power p = p, + ip2, where p x > 0, we have:
|(2 / _ a 1) p“ I I = | e l(Pl — *) + l0E - ■<».) | = e (P. - 1) lo g |z ' — a ,| - p .a r g f z ' - « J

or
|(z' - a1)p- ‘ = eP‘~ l ■
and the result will be the same.
Hence using the above method of integration we can, in the limit, neglect
the integration round the cirole; we then have the path of integration l2 and
we must integrate along the straight line at a2 encircling the point a2 and
returning to the point cq along that same line.
Bearing in mind the factors which the integrand gains by describing the
points a, and a2 we obtain the following formula for the solution (147):

u>„ (z) = Oe”2nl (1 - e«!"') J (z' - a ,)" -1 (z' - a2)»_1 ezz*dz' +


<>i
+ C7(e«2m- 1) J (z' - a1)p—1 (z' - a r f - 1 e“ ' dz'

wa{z) = - 0(<ePinl - 1) (e«2"' - 1 ) J (z' - a,)'’" 1 (z' - a ^ - 1e“ ' dz'.


<»1
I f we suppose th at p and q are not integers and reject the constant factor
we can, in this case, write the solution of the equation (134) which is regular
at the origin simply in the form of on integral along the line a, a2:

u>„(z) = C £ (z' - a1)p“ 1 (z' - a2)«-1 e22' dz'. (1 4 9 )


10 8 ] THE CHOICE OP SOLUTIONS 415

This last result can also be obtained directly. If the real parts of p and q
are greater than zero, the expression (145) must vanish when z' = a, and
z' = a2, so th at we can simply take the line at a2 as the path of integration I. In
this case we are not using the fact that p and q are not integers.

Let us now return to the general case. Notice that if we take for
the path of integration only one contour or l2 then the value of the
integral will, in general, depend on the initial point z0 and we
shall not obtain a solution of our equation. However, the point z0
can be so chosen that we can, in
fact, obtain a solution of the y
equation.
We shall assume in future that
z is a positive number and notice
that the expression
( * ' - a 1H z ' - a 2)«e“' (150)
tends to zero as s' tends to
infinity, so that the real part of z'
tends to (—°°) and its imaginary
part remains bounded. We shall say
that z' tends to (-«>). If we chose
for the contour l\ a contour the
ends of which are at the point (—°°) and which circumscribes ax,
then at the ends of this contour the expression (150) will vanish
and the condition (145) will be satisfied; consequently the integral
round this contour will give a solution of the equation (134). We
obtain a second solution similarly by taking for the path of integra­
tion V2 a contour originating at (-«>), which circumscribes the point
a2in the positive direction. We thus obtain two solutions for the equa­
tion (134):
w\ (2) = j (z' — a1)p_1 (z' — a2)i—1e“ ' dz',
V (151)
w,2(z) = f (z' - a1)p_1 (z' - dj)*-1e*' dz'.

The integrand has branch points at z' = ax and z' = Gq. To make
this function single-valued we must cut the plane from these points
to (—oo) and, assuming that the imaginary parts of tq and Oj
are different, we cut the plane along straight lines parallel to the
real axis (Fig. 68). In the cut plane we select that branch of the
integrand for which arg (z' — aj) = 0, when z' — dj > 0, i.e. the
416 LINEAR DIFFEREN TIA L EQUATIONS [108

continuation of the first cut, and arg (z' — a2) = 0, when z’ — a2 > 0.
The contours and l2 are situated as shown in Fig. 68. The
above conditions fully determine the solution (151) when z > 0.
Notice that the exponential function e“' tends to zero as z' — — °o
not only for positive values of z but for any value of z, the amplitude
of which lies within the limits
— - f - < a r g z < - |- . (152)
In fact, assuming that z = x + yi we have, at the same time, x > 0.
Also z' = x' + y'i, where x' — —°° and | y' | remains bounded.
Thus the real part of the product
y zz', which is equal to xx' — yy',
will also, in this case, tend to
ri (_oo) and the function (150) will
vanish at the ends of l[ and V2.
— - - X . Hence formula (151) gives the
unique solution for all z’s in the
sector (152).
We shall now establish the con­
0
nection between the solution (151)
and the solution (134) which is re­
F i g . 69
gular at the origin. Bearing in mind
further applications of the Bessel
equation we shall restrict ourselves
to the case when p = q. To obtain a solution regular at the origin
we can take as the contour of integration not the contour (lv l2)
which we mentioned above, but a simpler contour, viz. a contour
from a point z0, which describes the point cq in the positive direction
and the point a2 in the negative direction. On completing the first
circuit the integrand gains the factor ep2’1' and on completing the
second circuit the factor e~q2nl = e~p2nl, so that we return to the
initial value and the condition (145) is satisfied. As before the con­
structed solution is independent of the choice of the point z0. Let
us remove this point, without affecting either of the points ax or
a2, to (■—oo) for example, along the lower edge of the cut r2 which
goes to the point cq (Fig. 69). The completion of a circuit around
the point tq will then give the solution wq. On completing this circuit
we shall find ourselves on the upper edge of the above cut and we
shall then have to complete a circuit round the point a2 in the negative
direction. Had we completed this circuit from the lower edge of the
10 9 ] THE ASYMPTOTIC REPRESENTATION OP SOLUTIONS 417

cut t v we would have obtained the solution ( —w2). However, by


coming onto the upper edge of the cut, from where the circuit round
the point a2 is being described, the integrand acquired the factor
eipnl , and therefore the completion of a circuit round in the
negative direction gives (—ep2nl w2). We finally obtain the following
rule: when p = q the regular integral obtained by integrating round the
contour shown in Fig. 69 can be expressed in terms of the solution (151)
in the form:
wx (z) — ep2niw2 (2 ). (153)
109. The asymptotic representation of solutions. We shall now
deduce the asymptotic expansions for the solutions (151) for large
positive values of z. Let us recall that we have determined these
solutions for values of 2 in the sector (152). We begin with the first
solution. Replace z' by a new variable of integration t, according
to the formula
z’ - a x = t, (154)
and assume, to simplify the notation, that p = ax — a2. The first solu­
tion will then have the form:
wx (2) = J tP-1 (t + 0)*-1e^“1+,)d2, (155)
1.

where l0 is a contour which goes from t = — 00 and surrounds the


origin. The points t = 0 and t = —/? are branch-points of the
integrand function. Instead of the cuts shown in Fig. 68 we cut
the 2-plane in two places, viz. from ( —°°) to the points t = 0 and
t = —/? = a2 — av respectively; here arg t = 0 when t > 0, while
arg (t -+- /9) = 0 when t-Jr fS> 0, i.e. on the continuations of these cuts.
Using Newton’s binomial expansion we have, when 11 KI I0
(t + P )i-'= p - 1 fl + — V-1 = 2 dktk, (156)
V H} k=0
where
dk = p i-1p - k (d0 = p*-1) . (157)
From the above condition for the amplitude of (t + P) in the 2-plane
with the cut from (~ °°) to (—/?), we can assume that in the expression
Pq~1 which is equal to the function (156) when 2= 0, the amplitude of
P lies within the limits
— 7T< arg P < n , (158)
and we assume that /? is not a real negative number.
418 LINEAR DIFFERENTIAL EQUATIONS [109

When 1t | > | /J | we cannot use the formula (156) and in this


case we can simply write:
{t + P)q~ 1 = d0 + t + . . • + dn tn + Rn (t),
where
Rn (t) = (t + - (do + d, t + . . . + dnn . (159)
Using the above formulae we can write:

Wl (2) = e“lZ£ dk J e2' «P+fc- 1 d* + ea‘z J e2f p - 1 Rn (t) dt. (160)


k=0 /, /,
Consider the sum on the right-hand side. Replacing ( by a new
variable of integration r, according to the formula
zt = — r = e~nl r.
we can obtain the integral in the form:
f e2' tp+k~1 dt = e~*pi (— l)kz-P~k f e ^ r P + ^ d r .
il 1
where the path of integration Ais along a cut from r = + °° and around
r = 0 in the positive direction. Owing to the fact that zt = e~*ix
we have x = ze”11, where we assume that z > 0 and arg z = 0,
i.e. the r-plane is obtained from the i-plane by rotation about the
origin by an angle n, so that the lower edge of the cut l0in the t plane,
where arg t = —n, becomes the upper edge of the cut Ain the r plane;
as a result of the above formula we must assume on this upper edge
that arg r = 0.
Remembering the connection between the above contour integral
and the function r(z) we have [74]:
£ ezt tP+k~1d< = e~*pi (— 1)*z-P-*(e<P+'‘>2" '— l)T (p + k),

and formula (160) gives:

Wl (z) = e ai22-P(e2*P' - l)e-*P' ^ l)kdk r ( p + k)z~ k +


k=-o
+ e“lZj e2' tP-1 Rn (t) dt (161)
Jo
or
e - a»2gP wl (2) = e “"P' (e2l,P‘ — 1) ^ ( — 1)k dk T(p + k) z~k +
k^O
+ zp f e2/<P-1i?„(<)d<. ( 162 )
i.
109] THE ASYMPTOTIC REPRESENTATION OP SOLUTIONS 419

We will show that the infinite series

e ~npi (ez*p _ !) j g ( _ i)*_T(P+t fc) d* (163)


k =0 z

represents the asymptotic expansion of the function e~a'zzpw1(z)


when z > 0.
To prove this we must show that the product of z" and the last
term of the formula (162) tends to zero, i.e.
lim zn+P f e2( tP-1Rn (<) dt = 0.
X,
Consider the following path of integration: the line ( —°°, —r) of the
real axis, a circle, centre t = 0 and radius r and the line (—r, -o o )
of the real axis, where r is a positive number.
We will show, to start with, that the expression

z" + p j V f P - 1£ „ ( < ) & (164)

tends to zero as z —>- + °°- The same thing must also apply to
the integral along the line (—r, —°°), since only the factor
is added on describing the point t = 0.
Returning to the formula (159) we can see that a sufficiently large
positive number N can be found such that

*■(0 as t —> — oo,

and therefore the modulus of the quotient Rn(t)ltN will be bounded


along the whole path of integration, and we can write the inequality

\Rn (0| < m \ (IN (— oo < t < — r), (165)

where m is a fixed positive number. Let e be a small positive number.


Remembering that an exponential function increases more rapidly
than any power function and bearing in mind (165) we can write
I**- 1R„ (t) . ,
------- —> 0 as £—»- — oo
e -rf
or | tp—1Rn (t) | < m1e- *' (— oo < t < — r),
where m1 is a positive constant.
420 LIN EA R D IFFERENTIAL EQUATIONS [109

We therefore have the following inequality for the expression (164):

\zn+p Y e z,tP -'R n ( t) d t\< \zn+p\ J r m,e<2- £>'d< (z > 0),


— OO — 09

i.e., integrating
—r l*n + P |
I2"+P J* eZt ^ | < z _ J mi e _ r*
— CO

It follows that the expression (164) does, in fact, tend to zero


as z - > + ° ° - This will be so for every fixed r. It now remains
to be shown that the expression given below also tends to zero
zn+p $e*tP~1Rn {t) dt,
c
where the path of integration C is the circle, centre the origin and
radius r. We assume that r is very small, for example , r < (1/2) | /3 |.
We can then use Newton’s binomial expansion (156) on the circum­
ference 11 | = r.
According to Cauchy’s inequality we have the following inequality
for the coefficients of the expansion dk

where m2 is a positive number. We now assume that | /? | — e


is equal to, say, q = (l/2) | /? |. We have further:
Rn (0 = ^n+l tn+1 + ^n+2 ^"+2 + • • •
The above inequalities give

141 |<| = r < - i - |/ i |


and, consequently,

\Bn (0 I < K+111t\n+1 + \dn+21\ t r 2 + • • ■< 0n ™ ? ^ e) ■ (166)


where

This inequahty for Rn(t) is also valid when \ t\ < r, i.e. when t lies
in C. We again replace t by a new variable of integration r, according
to the formula zt = —r, and we obtain the following expression:

zn*pj e zttp- l Rn (t)dt = ( - \)p zn \ e - zrp- l Rn ( - - ) d t , (167)


c c'
109] THE ASYMPTOTIC REPRESENTATION OF SOLUTIONS 421

where the path of integration G' is a circle, centre the origin and
radius rz. According to Cauchy’s theorem we can deform this contour
and take for the contour of integration any closed contour, which
originates at the point rz of the real axis, encircles the origin and
lies in C’. In this case the corresponding contour in the t plane will
lie in C and the inequality (166) will be valid. We can, for example,
take for the contou of integration C" the following path: the segment
of the real axis from rz to a point c which lies to the right of the origin,
a circle, centre the origin and radius c and, again, the part (c, rz)
of the real axis, where c is a fixed positive number which is indepen­
dent of z.
We suppose at first that p is real. Using (166) to find an upper
bound of the expression (167) we obtain [4]:
M n+P
( - l ) f 2" J V * T - f ) dr „n+i ( 1 - d«,
6) 1
c- C'

where ds is the differential of the arc of the contour. We will show


that the coefficient of 1jz will remain bounded as z increases inde­
finitely. In fact, integration round the circle with radius c gives an
expression which is independent of z. Let us now also consider the
integral along the line (c, rz). This gives the following coefficient of z~l :
12
m2 rn+Pdr.
en+1(i - e)

As z increases indefinitely this last integral tends to the finite


limit:
J e~r rn+pdr,

where the existence of the above integral is fully assured by the


factor e~Tin the integrand. Hence our proposition is proved for a real
p. When p is complex, i.e. p = px + ip2, we only have to use the
usual property of a complex power, viz.
TP _ e (Pi + ip 2)logT _ e (Pi + ip 2)(log|T | +Iargi)^

whence
|r P | = | t |P i • e~PlZrST.
422 LINTSA_R DIPTBBENTIAL EQUATIONS [110

We can therefore say that the series (163) gives the asymptotic
representation of the function e -aiV ,«;1(2) when z > 0:

e- “i*zPiit^z) ~ e~”P‘ (e^P‘ - 1) y ( - 1)* r ^ ± k) d* , (168)


k5) 2
where
d k = ( a x - a2) 4 - i- f t ( g - D ( g - j2 ).-.(g ^ fcl (169)

(do = (cq - a2)i-1; — n < arg (cq — cq) < rc).


We will not consider the case when in the formula (168) p is an
integer.
Similarly, for the second of the solutions (151) we obtain the asymptotic
representation :

e - ^ ^ u U z ^ e - ^ e 2**' - 1) y (— l)k r (‘7 + *><**., ( i7o)


iSo 2
where
d'k = (a 2 - aJP-1-* ( p - l ) ( P - 2 ) - ( P - t ) . (171)

(do = (a2 — a1)P-1; — w < arg (a2 — cq) < n).


For the powers ?? and zq it must be assumed that arg 2 = 0 when
z > 0.

110. Comparison of results. We now return to the calculations


in [105]. In th at section we constructed the following solution of
the equation (113):

e“ *(co + c1-i- + ---), (172)


which formally satisfied this equation. Let us compare the expression
(172) which we obtained at the time with the expression determined
by the asymptotic formula (168), i.e. with the expression:

eai*z~P e~”pi (e2”P‘ — 1) (- l)k F(p \ k) dk ,(173)


k=o 2
and let us show that these expressions are exactly alike except for
the constant factor which enters formula (172) in the form of the
arbitrary c0.
Comparing the equation (134), for which we deduced the asymptotic
formula (168), with the equation (113), we can see, first of all, that
110] COMPARISON OP RESULTS 423

we must assume that ofc = = 0 when k > 2. The exponential


and power factors coincide in the formulae (172) and (173) since the
equation (139), from which we obtained cq, coincides with the equation
(115); also, from (142), we have p = —gv where px is the value of
q determined by the formula (119) when a = eq. The equivalence of
the power series in the formulae (172) and (173) remains to be shown;
for this purpose it is sufficient to show that the coefficients of these
series satisfy the same relationships from which they can be deter­
mined.
The series in the formula (172), obtained as the formal solution
of the equation (120) satisfies the following differential equation for
ak = bk = 0, when k > 2:

u" + ^2cq + ou + -2gl °l j u' + gl ^a'z2~ ^ u = 0. (174)

Remembering the form of the equation (115) we can write:

al “t" = — ®oi -f- Oq = al — a 2 » 2cq “1“ ®0 = ®2 — al>


whence
~ ____ a i ° i 4*1*1 a i a i ~b^i . ° t a i~ b ^ i g 2 a i- b ^ i
" 2<q a0 a, — a, ’ ”2 2 a a — <x0 a , — cq ’

and therefore

ei + ai = — P2! gi + K — l ) S i = — QiQz-Qv (175)


The equation (174) has the same form as (124) and we obtain the
following relationship for the coefficients cn:
[7i(» + 1) - n(26l + ®i) + el + K - l)gi] c„ =
= (71 + l)(2ax + a0)^n+1
or, from the equation oq + cq — —a0 and from (175):
[71(71 + 1) - 7i(ex - e2) — 6i ft - 0i] cn = {n + l) (cq - a2) cn+1. (176)

If we denote by c„ the coefficients of the infinite series in the


formula (173):
c„ = ( - i r d n p ( p + l ) . . . ( p + n - l ) r ( p ) = ( - i r d n r ( p + n ) t

whence
Cn-n _ dnii(P + n)
424 LINEAB DIFFERENTIAL EQUATIONS [111

or from the formula (169):


CfiM _ _ (g - n - 1) (P + n)
c„ (n + 1) (a, — a 2)
or
( » + l - ? ) (n + p)cn = (n + 1 ) (cq - a2) cn+1.
Bearing in mind that p = — and q = —q2 we can see that the
above relationship is the same as (176). We can thus see that formal
solutions of the equation (134) which are constructed by the methods
given in [105] give the asymptotic representation when z-»--f<=°, of
solutions determined by the formulae (151) except for the constant term.

111. The Bessel equation. Let us apply the above theory to the
Bessel equation [II, 48]:
z2w" + zw' + (z2 — n2) w = 0. (177)
Replacing w by a new unknown function u, defined by the formula
w = znu,
we can obtain the equation (177) in the form:
zu" + (2n +1) u' + zu = 0, (178)
and this happens to be an equation of exactly the same type as the
one we have considered above. In this case we have:
an = 0; a 1 = 2 rc + l; 60 = 1 : ^ = 0.
The quadratic equation (139) will be z'2 + 1 = 0, so that we
obtain:
cq = i; a2 = — i,
and, similarly, from the formulae (141) we have:
+ 1 . 2n + 1
2 ’ ? = — 2“
The final solution (151) will, in this case, have the following form:
2 n -l 2 n -l
tt1 = J (z '* + l ) 2 ezz’ dz’; u2 = f (z'2 + 1) 2 ezz'dz', (179)
i[ I*
where the contours lx and l2 originate at the point —°° and surround
the points z" = i and z' — —i
Ill] THE BESSEL EQUATION 425

These solutions are given by the formulae (179) where


—jr/2 + e < arg z < n/2 — e. From the condition in [108] arg (z' + i) = 0,
when z' + * > 0, and arg (z' — 1) = 0, when z' — i > 0. It follows
that arg (z'2 + 1) = arg (z" + i) + arg (z' — i) = 0 when z' is real.
For the first of the solutions (179) we have, from (168):

n+ I
3 -'zZ V ~ e
-"(n+IV
V 2' (e»(2n + l ) * _ l )

> -(_
-I1fn + *]
1 ) * _ ^ -------- ± -------- i — .
2
Remembering that
e*(2n+i)i _ 1 = _ (1 4- e2-1"')

and the expression (157) for dk


n ——— k /_ i\ n in —
dk = (2*) (? fc1) = ( 2 i )

w < arg 2i < :r or arg 2i = — j ,


i.e.

d* = 2 2 e *-*| 2
k
we obtain the following asymptotic representation:
i
e - , ,z
z "+2
z m, ~ e (1 _|_ e 2 ,,n ,' ) 2 x

X j l l " t » ) r ( » + - r + * } ( s ) ‘ - (180)

Similarly, analogous calculations give the following formula for


the second solution (179):

elzz M2~ e 2 4 (1 + e2ltni) 2 2x

(,81)
where the only difference is in the coefficients d \ which are expressed
by the formula

dk = ( ~ 2 i ) n 2 ^ 2 j (a rg (- 2 < )= - - J - ) .
426 LINEAR DIFFERENTIAL EQUATIONS [111

Let us recall that the symbol (£) for an integer k > 0 denotes the
following:

and 0 -1 .

Remembering the expression (153) for a solution of the equation


(178), which is regular at the origin, where the letter w replaces the
letter u , we can replace the solution u 2 by a new solution u *:
u$ = e p2nl u 2 = e ( 2n+1)!' 1 u 2 .

For this new solution we obtain the asymptotic representation


n+ -
J z r,
e“z 2 m* . (1 + e2”nl) 2 2X

x i ( ” J ) r ( » + 4 - + * )( -■ £ • )'• m

The corresponding solution of the equation (177) can be obtained


from w = zn u, by adding the factor zn.
The solution (179) is sometimes written in a slightly different
form, viz. z' is replaced by a new variable of integration r, according
to the formula: z’ = ir = en^2 r; this corresponds to
the rotation of the z ' plane through an angle (— 71/ 2 ):
_1
uy = i J (1 — t 2) 2eiu dr,
(183)
w2 = i j' (1 — r 2) 2e,ztdr,

where Xy and \ are contours which originate at the


point r — -j- i 0 0 and surround the points r = + 1
Fio. 70 and r = —1 (Fig. 70); arg (1 — r 2) = 0 when r is
purely imaginary, which corresponds to a real z’ or,
which comes to the same thing, arg (1 — r2) = n when r > 1. Assum­
ing th at 1 — r 2 = enl (t2 — 1), we obtain in place of (183):
n—-1
u, = e 1("-*)'•
i J (r2 — 1) 2e<2Tdr,
(184)
n—-1
u,. =e ' i J (r2 — 1) 2e,ZTdr,
Ill] THE BESSEL EQUATION 427

where
arg (t2 — 1) = 0 for r > 1. (185)
The corresponding solutions of the equation (177) will be:

(186)

and the asymptotic representation of these solutions for large positive


values of z will be obtained from the above formulae by the addition
of the factor zn to the right-hand sides. Replace the second solution
by w\ = e^2n+1^li w2, so that

= e ' ' izn J ( t2 — 1) *e'"dr. (187)


4.
The difference — u* gives the solution of the equation (178)
which is regular at the origin 2 = 0 and, similarly, the difference
— w* gives the solution of the Bessel equation which has the
following form near the origin:

k -0

We know already that this solution of the Bessel equation is given


by the following series [II, 48]:
z*
___f------- L
2(2n + 2) T 2 ■4 • (2n + 2) • (2n + 4)

When n is a positive integer or zero then, as we said before, we


select the constant C equal to 1/2” n ! where, as always, 0 1 = 1;
with this choice of the constant we obtain the Bessel function of the
first kind:
( - i)k fj_ y + * k
Jn{z) = k2= 0 kl (n + k)l V ^ )

or, using the function r(z):


(—1A) 2 'in+Zlt
(Z) = j Z r{k + 1) T(n + Jb + 1) 2“J
428 LIN EA R D IFFERENTIAL EQUATIONS [112

When n is not an integer the constant C is equal to


n _____ j;____
2”r(n + 1)
and we arrive, similarly, at the solution

(-1)* 2 \n+2k
fcl(n + fc) (n + fc —l)...(n + l).T(n + l) (t )

or, from the fundamental property of the function r(z):


2 \n+2k
^n (z) 22 p (k + 1}( r ( n + k + 1 ) ( 2 ) (188)

The Bessel function of any variable is thus determined. The


difference wl — w* will not give the exact value of the Bessel function;
it will differ from it by a constant factor which we shall now attempt
to find. We must, therefore, take in place of the solution (186) a new
solution which differs from the one above only by a constant term;
the latter is so chosen that the difference of the new solutions should
give the exact value of the Bessel function J n(z). Introducing the
minus sign in the second solution we shall look for a constant a from
the condition, that half the sum of the solutions

H ™ (z )= foffi = fits'1J (t2 — 1) 2 e izr dr,


1 i
H<n (*) = - bw% =-- - ae^n+1^ ‘zn f (r2 - 1)" 2e,ZI dr (189)
1.

gives the Bessel function.


In all the above calculations we have assumed that (n — 1/2) is
not equal to an integer m > 0. This latter case we shall investigate
in the detailed treatment of the Bessel function.

112. The Hankel function. In the above method of choosing the


constant a the formulae (189) gave two solutions for the equation (177)
which are known as the Hankel functions; they are denoted in the
same way as in the formulae (189). We saw earlier [108], that by
adding the solutions (189) we obtain one integral along the path of
integration C which will have the form of the figure eight, as shown
112] THE HANKEL FUNCTION 429

in Fig. 71. Let us recall that this figure can be obtained from Fig. 69,
when cij = i and a2 = —i, by rotation about the origin through a
right angle in the clockwise direction.
Bearing in mind the fact that half the sum of the functions (189)
should give the Bessel function (188) we have:

( - D *
n+ik
1) rdr — .(190)
fc=0 r(k+ !)/> + * + 1 )

Simplifying both sides by dividing by zn and assuming subsequently


that 2 = 0 we arrive at an equation which gives a:

■ W <
1) 2dr = (191)
t 2" r ( n + 1) ’

and the integral on the left-hand side only re­


mains to be evaluated. Assuming that n is real and
that (n — 1/2) is greater than (—1) we can, as in
[108], transform the path of integration C so that
it runs along the straight line (—1, +1); we must
integrate along the lower edge of the line from F i g . 71
(—1) to (+1) and along the upper edge of the
line from (+ 1) to (—1). We said above that arg (r2 — 1) = 0 when
r > 1, from which follows that arg (r2 — 1) = n on the upper edge
of the line ( —1, +1) and arg (r2 — 1) = —n on the lower edge of this
line, i.e.
n- j In ( n - 1) n- ^
0T2- 1) 2= e V 2)
(1 — T2) 2 (on the upper edge),

5)
(r2 - 1) 2= e (1 — r 2) 2 (on the lower edge);

and, finally, adding the integrals, we obtain:

n—1s
J V - 1) dr = — 2i sin | n ---- n |” (1 — r 2) 2dr,
-1
where
1 (n - 1 ) lo g (l— t*)
(1 - r=)n~ 2 = e (1 —r 2> 0).
430 LIN EA R D IFFERENTIAL EQUATIONS [1 1 2

Bearing in mind the fact that the integrand is an even function


we can write
_1 l _i
J(t2— 1) *dr = — 4i sin ---------------- —t2) *dr,
c o

or, substituting r by a new variable of integration x, according to the


formula r 2 = x:

n -i 1-i n- i
J (t2 — 1) * dr = — 2i sin 5(1 — x) *dx.
c o
But we saw earlier that

r(p)r(g)
r(p q)+
so that the integral in the equation (191) will be

J V - i ) dr = —2i sin (n ---- n T(n+1)

= 2i 6in + -y- T(n+1)


But we had earlier:

r{z)r(i-z) = -£ -r and r (4 -) = ifc,


whence

(192)
^ ( B + " r ) Bin(B + T ')w =
so that finally

2;i® i
J(^ -D *dr = -
r ( - i — » )r(» + i)

We deduced this formula on the assumption that n is real and


that n — l/2 > —1. Bearing in mind the fact that both sides are
112] THE HANKEL FUNCTION 431

analytic functions of n, we can say that this formula will hold


for every value of n. Formula (191) thus gives us the following value
for the constant a:

2"n2i
Substituting this value in the formula (189) we obtain an expression
for the Hankel function:

■f f~9 W
1 / _ ■.n r n—;
(z) = -J (— ) J (t* - 1)2e'« dr,
n *■
(193)
r [-z ----n) / _ ■,/» r n—i
H P (2) = ------Lf------L e(2"+i)xi j_L.j J (T2 _ 1 ) 2e'« dr.
IT*S

We assume in both integrals that arg (r2 — 1) = 0 when r > 1.


If we assume in the second integral that arg (r2 — 1) = when
r > 1, we can write:

HP (*) = --^ 3 ---- [ \] J (t2 - 1) e'» dr,


_2 „t •
71
Xt
(193,)
r{-« M) / - \n (• n—j
H P (z) = ------ ^ ( - | - ) J (T* - 1) 2ef« dr.
_2 ■ it

When t - > +z °o the real part of iz x tends to (—°°) provided the


real part of z is greater than zero; formula (193) thus determines the
Hankel function to the right of the imaginary axis. Let us remind
you that we have assumed that n — l/2 is not a positive integer.
The Hankel function HP(z), as given by the formula (193), differs
from the first of the functions (186) by the factor

3
2™ji2
432 LINEAR DIFFERENTIAL EQUATIONS [112

Taking the asymptotic expansion (180) and bearing in mind that


w = zn u, we obtain after performing elementary transformations:
1 n 3n
25e-S",+ T r ( 4 " - “)■ > " (» + 4 ')”
e~,z z (z) ~ —

fc=»V k I
or, from (192):
1 1 _ . I’m n\

w ~ (4-)* 0 4 ) I (* - j ) r (• + 4 - + *) ( g ;

We can write this as follows:


1 ./ nn n\

i ( * - 7 ) r ( . + -} -+ . ) ( 4 . ) - .
r (n + 4") k
(195)
Similarly, we obtain:
1 ./ nn n\

* * <’>~ (w )' \ t . ‘n k (” ”7 ) r (• + t + *) ( - w )
~) ' (196)
(1
The latter formulae can be written as follows:
1 .( nn n\
,, ( 2 \2 o ' 2 *'

Hi)
p- i ,
x
^W» +i H (sf + , (195,)

,, , 2J

(”+ - f )

X § h ^ ) 7’(’, + - r + i ) ( - w ) , + 0 <W-') (196,)


113] TH E BESSEL FUNCTION 433

where 0(| z |“ fc) denotes a number which is such that the product
| 2 |* 0 ( |z |- k) remains bounded as | z | increases indefinitely. In the
expression (2jnz)112 it must be assumed that arg 2 = 0, i.e. the
positive radical must be taken.
We proved the above asymptotic formulae for the ray z > 0.
It can be shown that these formulae will be valid in certain sectors,
viz. formula (195!) holds in the sector:
—n e < arg z < 2ti — s,
and formula (196!) in the sector:
— 2tt + e < arg z < n — e,

where e is any small positive number.

113. The Bessel function. Substituting the expression we found


for a in the formula (190) we can represent the Bessel function J„(z)
by the integral
4— )
Jni*) = 3 ("f) I { r 'J ~ 9 5e‘2Idr- (197)
C
2 7^ i

This formula, like formula (193), holds for all values of n except
for values of the form m + 1/2, where m is an integer > 0.
If the real part of n is greater than ( —1/2) then integration in the
above formula can be reduced to integration along the double line
(—1, +1) and, using the same arguments as above, we arrive at the
formula:

J "{z ) = ("9 I ( 1 " T’’ ' e“' dT U<«>

(« M > - •
If we put r — sin <p we obtain:

2
1_____
J n (z) Hr) Jncos2n<Px
~2

X [cos(z sin q>) + i sin (z sin 95)] d99,


434 LINEAR DIFFERENTIAL EQUATIONS [113

or, bearing in mind that the coefficient of i is odd:


n
2
1
JnW J cos2" <p• cos (z sin <p) dip (199)
^ r (w+4") 71
2

which can also be written as follows:

_____ 2______
Jn W cos2" q>cos (z sin cp) dq? ( 200 )

Taking half the sum of the asymptotic expansions (195) and (19G)
we obtain the asymptotic representation for the Bessel function.
For simplicity we shall only consider the ray z > 0. In this case
the modulus of the factor e ±IZ is unity. Taking half the sum of the
expressions (195x) and (196!) and bearing in mind that

, , J
(ir) M 0(z-°, = 0 (2 ■) fz > 0),
K 4 -)
we obtain:
j , (2) = — [/;(■) (2, + a«> (!,)] =

7 - n X
(2*)“
r + "2"J k=0V k

/( _ i\2
i ) c o s (^ _ _nn _ _ nj \
+ 0 (z ” 5), (201)
fc±i
/ i \ 2 • ( 717Z 7t
[(-1 ) sin |z ^ 4"J

where the first expression in the braces refers to the even k and the
second to the odd k.
114] THE LAPLACE TRANSFORMATION IN MORE GENERAL OASES 436
Considering only the first terms of the asymptotic representation
we can write:
i / __ n n ti\
-

m - ( i ] +
( 202 )
1
- / nn n\

= e~ 2 5 ti + o tN " 1)]
and for the Bessel function
I
J n (2) = (— ) COS [z — — £") + °(2 2) (Z > 0 ). (203)

The difference in the asymptotic expressions for the Hankel and


Bessel functions plays an essential part in solving problems of mathe­
matical physics in infinite domains which include the point at infinity;
we shall deal with this later.

114. The Laplace transformation in more general cases. The Laplace trans­
formation can be applied to equations of a more general type than the equation
(134). Consider, for example, an equation the coefficients of which are poly­
nomial expressions of the second degree:
(o02 2 + o,z + Oj) w" -f- (60z2 + bt s + &2) w' + (c0z2 + c, z + c2) w = 0, (204)
where we assume th at o0 ^ 0. If we divide the above equation by the coefficient
of w", then the coefficients of w' and w will lie in the neighbourhood of z = oo,
and they will have the same form as in the equation (113). We shall again try
to find the solution of the equation (204) in the form:
w(z) = J r(z') e7Z‘ dz'. (205)
I
Using the same arguments as in [107] we obtain for v(z') a differential equa­
tion of the second order in the form:
/12ii
K 2'2 + & 0 2' + co) + P (z ') - f e T + 3(2') V = 0, (206)

where the coefficients which are omitted are polynomials, the degree of which is
not greater than two. Construct the quadratic equation
a „ a!!H -b „ a + Co = 0 <207)

and assume th at it has different zeros a = at and a = a2. The equation (206)
has regular singularities at the points z' = at and z' = a... At each of these
points the determining equation has one zero equal to zero. Denote by (p — 1)
and (q— 1) the second zeros of the determining equation at the above points;
436 LINEAR DIFFERENTIAL EQUATIONS [1X5

we assume th at p and q are not integers. At each of the above singularities


there is one regular solution and the second integral has the form:
vi (z') — (z' — ai)p _ 1 (z’)t v2(z') = ( z ' - a 2)<-1'p2(z’), (208)
where q>k(z') is regular at the point z' = ak (k = 1, 2). The contour I in formula
(205) should be so chosen th at the increment of the term outside the integral
vanishes on integration by parts, as in [107]), round I. This will be so if the
following conditions are satisfied:
dn (vz’m) c„ ,j Q ( n = 0, 1 ^
(209)
dz'n Ji (m = 0, 1, 2)
Take for v(z') the integral vk(z') and for I the contour lk shown in [108].
We thus obtain, as in [108], two linearly independent solutions of the equa­
tion (204):
wk (2) =I vk (2') e“'dz' (s > 0) (k = 1, 2).

These solutions will have the following asymptotic representation as z -*■ + oo:

(z) = e“‘zz p (e0 +

w2 (z) = ea*z2 “ « (d0 + + . . . j,

which are the same as the expansion in [105] except for the constant terms.
The Laplace transformation can be applied in cases when the coefficients of
the equation are polynomials of degree m. We then obtain for v(z') an equation
of the with order, the coefficients of which are polynomials of the second degree.
As before this equation for v(z') has at the singularities zl = at and z' = a2
(the zeros of the coefficient of dmv/dzm) unique solutions of the form (208).
The remaining solutions are regular at the above singularities. Otherwise the
arguments used are exactly the same as before.

115. The generalized Laguerre polynomials. Investigations of the


condition of an electron in the Coulomb field and some other problems
of modern physics lead us to a linear differential equation of the second
order in the following form:

w" + ± - u f + (& + 4 - - ^ - ) w = 0. (210)

Here s is a given real positive number and e is a real parameter.


The problem involves the finding of those values of the parameter
for which the equation (210) has a solution which remains bounded
along the whole line 0 < z < +<» on the real axis.
11 5 ] THE GENERALIZED LAGUERRE POLYNOMIALS 437

Let us consider, to start with, the region of negative values of the


parameter e and replace z by a new independent variable x, given
by the formula:
X = z ]/ — 8e ( 211 )

and replace the parameter e by a new positive parameter A given


by the formula:
1
( 212 )
Y— 2c

After these transformations the equation (210) will have the form:

d* w . dw
dx2 ' dx + ( - - r + A - - £ r ) M’ = (213)

This equation has a regular singularity at the point x — 0 and


the determining equation at this point will be of the form:

a(a - 1) + a - - J - = 0.

This latter equation has zeros a = ±s/2. Bearing in mind the


condition that the solutions must be bounded at the origin we
take the zero a ~ s/2, i.e. we must isolate the factor X s ' 2 , and our
solution will have the following form near the origin:

w = x2J r bhxk (b0 # 0). (214)


k=0

At infinity we must, according to [105], try to satisfy formally


the equation (213) by an expression

(c0# o ).
k=0 **
The quadratic equation for a will be

It gives the values a = i 1/2 and the corresponding values of the


constant q will be [from (119)]:
438 LINEAR DIFFERENTIAL EQUATIONS [115
Bearing in mind the condition that the solution must be bounded,
we must take that solution which has the following asymptotic repre­
sentation at infinity:

2 - %^ r - (215)
k=0
The problem thus involves the determination of those values of X
for which a solution in the form (214), after analytic continuation
along the line (0, + °°), has the form (125) at infinity.
The above considerations naturally bring us to the conclusion
that w should be replaced by a new function y, according to the
formula
X 3
w=e (216)
Substituting in the equation (213) we obtain for y the equation

a:- S - + (5 + 1 - a:) £ + (;i- 1T 1 ) ^ ==0- <217>


This equation happens to have the same form as the equation
(134) which we investigated before. Bearing in mind our former consid­
erations we should be able to find a solution of the equation (217),
which is regular at the origin and is of the order at infinity.
If we suppose, for briefness, that
— X = p, (218)
then the equation (217) can be written as follows:

j:l f + (! + 1 " I ) ¥ “ R, = a (219)


We shall attempt to find a solution of this equation, which is regular
at the origin, in the form of the usual power series
y = l + b1x + b2xi + . . .
Substituting in the equation (219) and using the usual method of
undetermined coefficients we obtain the solution in the form of a
series, very similar to the hypergeometric series; thus if we denote

F « ,^ ) = l + f - L + - ^ 4 + ..., (220)
then the solution of the equation (219), which is regular at the origin,
will be
y = CF(p,s + 1; x), ( 221 )
115] THE GENERALIZED LAQUERRE POLYNOMIALS 439

where C is an arbitrary constant. Notice that the series (220) conver­


ges for every value of x; this follows from the form of the equation
(219) and can readily be proved by the d’Alembert test. It is evi­
dent that the series (200) ends abruptly when a is either zero or a
negative integer and in this case our solution will satisfy the neces­
sary condition at infinity. We thus obtain from (218) the following
equation for the determination of the parameter X:

^ - ~ - — Xn = — n {n = 0 , 1 , 2 , . . . ) ,
whence
A„ = ^ ^ - + 7 1 (n = 0 , 1 , 2 , . . . ) . (222)

For this value of the parameter the required solution of the equation
(219) will be

<?„(*) = Cn F ( - M + l ; * ) = 0 B[ l - i 7 J T +
n ( n —1)
+ 2! (s + ! ) ( • + 2) + •■• + ( - 1)" ( s + l ) ( s + 2 ) . . . ( s + n)

To eliminate the letter s from the denominator we choose the con­


stant Cn as follows:
I\a + n + l )
@n — (* + 1) is + 2) ■• •(s + n) r{s + 1 )
which finally gives us the following solution for the equation (217)
in the form of a polynomial in x and s:

Q(f)(x) = ^ ,(t ^ + I) F ( - n , s + l ; x ) , (223)


or
£ns) (*) = ( — 1 )" [ z n — - y - (s + n ) X " - 1 - f

+ - - 2f I) (5 + n) {s + n — 1)X"-2 + . . . +

+ ( - 1)" {s + n) (s + n - 1). . .(a + 1)] . (224)

These polynomials are known as the generalized Laguerre polynomials.


We shall deal with them in greater detail later.
It can be shown that formula (222) gives all the values of the
parameter for which our problem has a solution which satisfies the
given conditions at the points x = 0 and x = + ° ° .
440 LINEAR DIFFERENTIAL EQUATIONS [115

Using the same arguments as in [102] we can deduce a simple


expression for the generalized Laguerre polynomials. The series (220)
is the solution of the equation:

* + (y - *) ~ *y = °- (225)

If we differentiate this series m times we obtain the series:


a(a + l)...( q + m — 1)
m, y + m\ x).
'AY + !)• ■■(!' + »»— 1) '

and, writing F(a, y; x) = yy we can see that the derivative of the


mth order:
y(m) = F m (a> y; x) (226)

is the solution of the equation (225) in which a and y are replaced


by (a + ttc) and (y + m), i.e.
d2y(m) dyf»
x dxz + (y + m — x) da: (a m) = 0.

Multiplying both sides of this equation by x,y+m-l e —x we can


rewrite it in the following form [102]:

_d_ ^y + m e -x . d^ ° j _ (a + m ) xY+m-l e - x y(m) = Q .


dx
Differentiating this identity m times we have:
dm+i
- \xy+m e~x dyim) 1 = (a + m) — [ x ^ 1" - 1e - xv<m>].
dxm+1 [ di J ' di™
On constructing similar equations for m = 0, 1, .. -,k — 1, multiply­
ing them term by term and simplifying the results we have:

A _ (a-y+A-i e -* ^*)) = a ( a + 1 ) . . . ( « + 4 — l j r ' e - ' y , . (227)

Let us suppose that a is a negative integer a = —k, so that the


series (220) is a polynomial of the 4th degree and y[k) is a constant, then
- k ( - k + 1 ) ( - k + 2 ) . . . ( — k + k — 1)
F w (— k, y;x) y(y + l)(y + 2 ) ...( y + k - l )

(— l)k ------------ —------------- .


' ’ Y(Y + !)■■■(? + * - i)
116] POSITIVE VALUES OF TH E PARAMETER 441

Formula (227) gives


k\ (Xy+fc-! e-x) =
(-1 )* V(Y+!)..■(?+ k - \ ) dxk
= ( — 1)kk \x v~l e~x F (— k, y; x),
so that we finally obtain:
r»—
P ( - * . * * ) = r(t. + i , . , r + t -_ „ <*»»

We thus have, from (223), the following expression for the generalized
Laguerre polynomials (y = s 4* 1; h = n):
<3<f>(x) = x - sex— (xs+ne - J). (229)
dx"

116. Positive values of the parameter. Let us now consider the


equation (210) for positive values of the parameter e. For this purpose
we replace 2 by a new independent variable xx according to the formula
Xj = 2 l/8e
and the parameter e by a new parameter A,:

The equation (210) thus becomes:

* .-3 -+ ^

This latter equation is obtained from equation (213) by the follow­


ing change of independent variable and parameter:
x = ix1; A= — iAj.
We thus replace w by a new function yv according to the formula
I X| S

w= e 2x*y^ (231)
and we obtain for yl the equation

+ (s + 1 - + [Ai - - H 5 + !)]»i = °- (232)


Comparing this with the equation (134) we can see that in this case

a0 = — i; a, = s + 1; b0 = 0; — {s + 1).
442 LINEAR DIFFERENTIAL EQUATIONS [116

The quadratic equation for a will have the form:


a2 — ia = 0,
this has two zeros
cq = 0; a2 = i,
and the corresponding values of p and q are [107]:

P — — — (s + 1) +

i(S + 1) + K ----- («+ 1) 1


9 = ---------- 2T= i-------------= t ( * + 1 ) -
In the case under consideration we therefore have the following two
solutions for the equation (232):

2/p) = C1J z ' 2 (Z' + i ) exM'd z'


•i (233)
5 (s—1)+Mi M - h,
!/P = C2j z ' 2 (z' - i) ex‘2' dz',

where l[ and V%are contours, with ends at the point z' = — which
surround the points z' = 0 and z' = i. According to the formulae
in [109] these solutions will have the following asymptotic expres­
sions for large positive values of aq:
-2<S+1)-M, Fk
C3 x1 ■
fc=0 1
—»(■+1)+M1 - '
C4 e,Xl aq 2
k=0 '

where C3 and C4 are constants. Using formula (231) also we can see
that the corresponding solutions, wx and w2 of the equation (230)
tend to zero as a q -> + ° ° . Consequently the 6ame can be said
about every solution of the equation (230) and in particular about
the solution which, near aq = 0, is represented in the form

w = xi (fco^o),
fc=0
i.e. for every real the solution of the equation (230) vanishes at both
ends of the interval (0, + °°).
11 7 ] TH E DEGENERATION OF THE EQUATION OF GAUSS 443
117. The degeneration of the equation of Gauss. Let us consider
the general case of an equation, the coefficients of which are poly­
nomials of the first degree:

(Po t + Vi) - J r + (?o t + gx) — + (r01 + r2) u = 0, (234)

where we assume that p0 ^ 0. We will show that this equation can


be obtained in the form (225). Replacing 1 by a new independent
variable z = p 01 + px we can obtain the equation (234) in the form
(134):
z ~ S ' + (ac2 + ai)" S ' + (&o* + &i)“ = 0. (235)

If we now assume that u = eazzpy and replace z by a new variable


x = kz, then with a suitable choice of the constants p and k we
arrive at the equation (225). We will now show that this latter equation
can be obtained by a limit transition from the equation of Gauss:

z(z- ]) + [- y + (i + a + £ ) z] - J - + a P y = °-

Replacing z by a new variable x = az we can rewrite this equation


in the form:

If we assume in the above equation that a tends to infinity we obtain


the equation (225) which, as we have shown, is connected with the
equation (234) by a simple replacement of variables. Ab a result of
the above limit transition, two of the regular singularities of the
equation of Gauss coincide and form one irregular singularity at
infinity while the other regular singularity remains.
Connected with the above is Whittaker’s equation:

d*u>
dz2 + i-^ + 4-+ \w = 0. (236)

If we replace w by a new function u, where w = 2m+1/2 u, we


obtain an equation in the form (225):

d 2u
+ o - 2”» - t + ( u = 0.
dz2"
444 LINEAR DIFFERENTIAL EQUATIONS [1 1 7

Constructing a solution in the form of (151) for this equation, we have:

I
where I is a contour which goes from ( —°°) and surrounds the point
z ' = —1/2. Replacing the variable of integration as follows:

* ' = _ J ____- ,
2 z
we obtain:
—-1z
r
i .
m - - —k ,
i f
, xm—s +fc
w = Ct e zfcJ ( — Z) ^1 + — j e"'d(,
I.
where Z0 is a contour which originates at ( + °°) and surrounds the
point t = 0 in the positive direction; we are assuming that the
point t = —z lies outside this contour. Choosing the constant Cy in a
deliberate manner we obtain the functions deduced by Whittaker:
w k,m («) = (237)

e ' d/.

In this formula it is assumed that z is not a negative, that arg z has


principal values, that | arg(—t) | < rt and arg (1 + t/2) tend to zero
as t —>0 and when they lie inside the contour l0. Formula (237)
becomes devoid of meaning when (k— lj2 — m) is a negative integer.
When the real part of (k —1/2 —m) is not positive and (k— 1/2 —m)
is not an integer, the expression (237) can be transformed to:
i ,i
2„ - z* »
rr — , 1 ,
~ k—
K~ -« +m
* m ,( t, \\
k —- +m
2*
W*.m(*) = j/ 2 (l+^-) o-'dZ, (238)
? ^ ----k + m ) o
and this formula also determines wk<m(z) when (k — 1/2 — m) is a
negative integer.
Using the results from [109] it is easy to write the asymptotic
expression for the function wkim(z):
1
(2) = e z*X (238,)
__ 3^
[mi - (fc - 4 - ) l h 2- (* ___ 2
X 1+ n! s™
n=l
I18J EQUATION'S WITH PERIODIC COEFFICIENTS 445

It holds in the sector ] arg z | < n — e, where e is an arbitrary positive


number.
Equation (236) does not change when k and z are simultaneously
replaced by (—k) and (—2) and, as a result of this the function
— will be a second solution of the equation (236), other than
the solution (237). The linear independence of the two constructed
solutions follows from the asymptotic expression (238:).

118. Equations with periodic coefficients. Consider a linear differen­


tial equation of the second order, the coefficients of which are periodic
functions of the independent variable. The theory of such equations
resembles in many instances the theory of equations with analytic
coefficients which we described above. Let us suppose, for the moment,
that both the coefficients and the independent variables are real. We
are given the equation
y" (*) + p(x) y' {*) + <i{x) yix) = o, (2 3 9 )
where p(x) and q(x) are real continuous functions of the real variable x
with the real period to, i.e.
p(x + co)=p(x); q ( x + w ) = q{x). (240)
The continuity of the coefficients guarantees that any solution of
the equation (239), determined by some initial conditions, exists for
all real values of x. Let yx(x) be one solution of the equation, i.e. we
have the identity
y'i (x) + p(x) yx (x) + q(x) y x (x) = 0.
Replacing x by x + co, we can write:
y"i (* + «) + p(x + (o) y[ (x + co) -f q(x + co) y x (x + co) = 0
or from (240)
y\ (x + w) + p{x) y[ (x + co) + q(x) y x (x + w)= 0.
It follows that yx(x + co) will also be a solution ofthis equation.
Let us now consider any two linearly independent solutions ^(x) and
y2(x) of the equation. The functions yx(x + co) and y2(x + co) must
also be solutions of the equation (239) and, consequently, they can be
expressed linearly in terms of yx(x) and y2(x), i.e.
y x (x + co) = a11y 1 (x) + oia y2 (x),
(241)
y 2 (x + o>) = a.n y x (x) + a22y 2 (x).
446 LINEAR DIFFERENTIAL EQUATIONS [118

where aik are constants. We can see that if we take two linearly inde­
pendent solutions of the equation (239) and add to the argument the
period, this will he equivalent to the linear transformation (241). Similarly,
when considering equations with analytic coefficients we saw that
by encircling a singularity, the linearly independent solutions are
subject to a linear transformation and we can use the same arguments
as in [97]. Let us give the results. The table of the constants aik
depends on the choice of the linearly independent solutions, but the
coefficients of the quadratic equation in g:
\ai l ~ G ’ °12 242)
i
I ®21> ®22 Q
will be the same for every choice of solution. If the equation (242)
has two different zeros g1 and g2 then two linearly independent
solutions exist which will be multiplied by gt and g2 when x is re­
placed by x a>, i.e. denoting these solutions by r\k(x):
rl l (x+a>)=Q1r)1(x); rj2 {x + co) = g2r}2 (x). (243)
If the equation (242) has equal zeros, i.e. = g2 then, in general
only one solution exists which acquires the factor gj whenx is re­
placed by x -(- co, and in that case we have instead of (243) the linear
transformation:
Tj1(x + (o )= g 1r]l(x); r)2 (x + co) = a2l rll (x) -f ^ rj 2{x). (244)
Let us also recall you that the equation (242) cannot have a
zero which is zero, i.e. a determinant composed of the numbers aik
will not be equal to zero.
Having recalled these results, we shall now try to determine
the form of the solutions in different cases. Consider, first of all,
the case (243). Let us look at the two functions

z l0»>* t. l°se«
(?i = e gt = e
where we take definite values for log ^ and log g2. When replacing
x by x + co these functions acquire the factors and g, and there­
fore the quotients %(x) : glla> and rj2(x) : g2'm must be periodic
functions with a period co; consequently, in the case (243) we can
write
X X

v A x ) = ei<Pi (*); y2(*) = q% (245)


118] EQUATIONS W ITH PEHIODIC COEFFICIENTS 447

where (p^x) and qr>2(x) are periodic functions with a period co. In the
case (244) we have a similar expression for rj^x). To investigate
rj2(x) we shall consider the quotient %(x) : 'ij1(x). We have, from (244):
Vi (x + “>) _ Vt (») I , a»i ~)
%(* + «) vA*) ^ I ei ) ’
i.e. the quotient acquires the term c when x is replaced by x -f- co.
The elementary function (c/co) x acquires a similar term and, conse­
quently, the difference i)z(x)l rj^x) — (c/co) x is a periodic function
rp2(x). Thus, in this case, taking into account the expression for
%(x), we have:
X

V i (*) = mQi <Pi (*); V2 (*) = x V i (*) + v»i(*) ’l l (*)


or
X X

»?i (*) = Qi <Pi (*); »?2 (*) = e“ l>2 (*) + (*)]. (246)
where cp^x), <p2{x) and <p3(x) are periodic functions. If the constant
happens to be zero, then the second solution will also have the form
(245).
In this case we do not, strictly speaking, have a general method
for constructing the quadratic equation (242). We will, nevertheless,
note some of the properties of this equation and of its zeros. Let us
determine the linearly independent solutions from the following
simple initial conditions:
Vi (0) = 1; ri(0) = 0; y2 = 0; ^ (0 ) = 1. (247)
Owing to the fact that the initial conditions and the coefficients of
the equation (239) are real these solutions will be real when x is real.
Assuming in the identity (241) that x = 0 and bearing in mind the
initial conditions (247) we obtain an = yv(co) and c21 = y2(w). Thus
for the given choice of the linearly independent solutions the quadratic
equation (242) can be written in the form:

yi M — e. y[ M (248)
y 2 (°>). y 'z M -e
and it follows that the coefficients of this equation are real.
Let us investigate in greater detail the particular case when the
term y'(x) is absent in the equation (239), i.e. when the equation has
the form:
y" (x) + q{x) y{x) = 0. (249)
448 LIN EA R DIFFERENTIAL EQUATIONS [118

Consider the Wronski determinant


d(x) = y 1 (x) y'2 (x) - y2 (x) y [ (x).
We obtained for it the following formula [II, 24]:

/!(« )_ J(0 )e -!W ,“


and therefore in this case, when p(x) is identically zero we have
A(z) = C,

where C is a constant. When the solutions satisfy the initial conditions


(247) this constant must be unity. Let us now turn to the quadratic
equation (248). The constant in this equation is equal to the Wronski
determinant when x = co, i.e. to unity. Therefore, finally, if we take
independent solutions of the equation (249) which satisfy the initial
conditions (247) then the quadratic equation for g must be of the
form:
g2 - 2A g + 1 = 0, (250)
where
2 A = y 1 ( c o ) + y ! 1 (co). (251)
If the real number A satisfies the condition | A \ > 1, then the
equation has different zeros, the product of which is unity, i.e. the
modulus of one of the zeros will be greater than unity and that of the
second smaller than unity. When [ + | < 1 the equation (250) will
have conjugate complex zeros, the moduli of which are unity.
Finally, when A = + 1, the equation (250) has a double zero +1.
The values of A have a decisive influence upon the behaviour of the
solutions when the variable x increases indefinitely. We shall now
investigate the different cases mentioned above.
In the expressions (245) the factors fi(x) and <p2(x) are periodic
functions and therefore remain bounded as x increases indefinitely;
the character of the solution depends essentially on the first few
factors:
X X
r. lose>
ei - e 92
11}
(252)
The real part of log g is, as we know, equal to log | g | and, con­
sequently, when | A | > 1, this real part will be positive for one of
the zeros, say for g t , and negative for the other zero; hence the
modulus of the first of the functions (252) will increase indefinitely
as x —>■+ °° while the modulus of the other will tend to zero.
118] EQUATIONS W ITH PERIODIC COEFFICIENTS 449
Returning to the solution (245) we can say that the first solution
will not remain bounded as x—►+ 00 but the second will tend to zero.
In this case the general solution of the equation:
Cl Vi (*) + c 2 r)2 (x) (253)
(Cl ^ 0) will, in general, remain bounded (the case of non-equilib­
rium). When | A | < 1 the real parts of log gx and log g2 are equal
to zero and the moduli of the functions (252) will be unity for all
real values of x. In this case both solutions (245) and the general solu­
tion (253) remain bounded as x —>- + °°. If the initial conditions
j/(0) = a; y' (0) = b
are determined by the numbers a and b, the moduli of which are suffi­
ciently small, then the constants C2andC2will also be small and there­
fore the modulus of the solution will remain small for every small
positive x (the case of equilibrium).
The particular case when A = ± 1 and the equation (250) has mul­
tiple zeros remains to be investigated. Suppose, to start with, that
.4 = 1, i.e. gj = g2 = 1. We can, in this case, take solutions which
have the following form
Vi (x) = <Pi (x)-, v 2 (x) = <p2 (*) + x<p3 (x), (254)
where <pk(x) are periodic functions. The first of the above solutions
will be purely periodic and the second will, in general, be unbounded
owing to the factor x. Only in the exceptional case, when <p3(x) is
identically zero, will the second solution also be purely periodic.
Finally, when A = —1, i.e. gT = g2 = —1 we can take log gx = ni
and we then have in place of (254)
, nx .nx
I — I —

Vi (x) = ® “ <Pi (x)l rj2 (x) = e “ [<p2 (x) -f x<p3 (x)].


The first solution will, in this case, be purely periodic with a
period 2u> while the other solution will, in general, be unbounded,
as before.
Let us consider as an elementary example the equation with a
constant coefficient
y" (*) + qy (x) = o. (255)
This constant coefficient q can be regarded as a periodic function
with an arbitrary period a>. Suppose that the constant q is positive.
450 LIN EA R DIFFERENTIAL EQUATIONS [1 1 8

Writing q = k2 we obtain the following two solutions of the equation:

t]1( x ) = e ikx; t]2 (x) = e~ikx.

On replacing x by x + w these solutions acquire the factors


^ = eika and g2 = e~‘ka>, the moduli of which are unity. This corres­
ponds to the case when | A | < 1. If the constant q in the equation
(255) is negative then writing q = —k2, we obtain the following two
solutions of the equation:

rjl { x )= e kx; rj2(x) = e - kx.

On replacing x by x -j- co these solutions acquire real positive


factors = eka and q2 — e~ka and this corresponds to the case
when | A | > 1. Analogous circumstances also hold when the coeffi­
cient q(x) in the equation (249) depends on x but does not change its
sign. Suppose of all first that q(x) < 0. Let yx(x) be the solution of our
equation which satisfies the initial conditions ^(0) = 1 and ^'(O) = 0.
Integrating the equation (249) and remembering the initial conditions
we can write
y'i (*) = — J q{x) Vi (*) dz • (256)
o
When the values of x are positive and close to zero and when
yx(x) is close to unity, since q(x) < 0, yi(x) > 0, i.e. y2(x) increases.
It follows from the relationship (256) that y\(x) could only become
negative after y2(x) has become negative. But, on the other hand,
if y^(x) is to become negative it must decrease to start with, i.e.
this is necessary if yx'(x) is to become negative. We thus arrive at
a contradiction and we can say that for every x > 0, y'(x) > 0
and yx(x) > 1 and, in particular, yx(a>) > 1. Let us now consider
the solution y2(x) which satisfies the initial conditions y2(0) = 0
and y2'{0) = L Integrating the equation (249) we have:

Vi(*) = l - J q(x)y2(x)dx. (257)


o
For values of x close to zero y^x) will be close to unity and it will
therefore be positive; for this reason y2(x) increases and must be grea­
ter than zero, since y2(0) = 0. Formula (257) shows that y ’z{x) can
become negative only after y2(x) has become negative. But, on the other
hand, y2(x) can become negative only if it decreases to start with,
119] TH E CASE OF ANALYTIC COEFFICIENTS 461

i.e. only after y2'(x) has become negative. This contradiction shows
that for every positive x we have y2(x) > 0 and y ’^ x) > 1 and, in
particular, that y^co) > 1. The inequalities for y^co) and y'2(o>) give

2A = y x (co) + 3/2 (0 1 ) > 2 ,

and we thus arrive at the following theorem; when in the equation


(249) q(x) < 0 then A > 1 and, consequently, the numbers q2 and q2
are different and positive.
To correct somewhat the above arguments we could have sub­
stituted the condition q(x) < 0 by the condition q(x) < 0 where it is,
of course, understood that q[x) is not identically zero.
The case when q(x) > 0 presents greater difficulties. We shall only
give the result here, the proof of which can be found in the work of
A. M. Liapunov: The General Problem of The Equilibrium of Movement.
When q(x) > 0 and when the function also satisfies the condition
ai
c0 J q(x) dx < 4, (258)
0

then gj and q2 are complex conjugate numbers the moduli of which


are unity. This theorem gives the necessary conditions to show that
Ml<i.
A detailed and extensive investigation of linear (and non-linear)
equations with periodic coefficients was undertaken in the above
mentioned work by A. M. Liapunov and in a series of his later works.
With reference to the equation (249) we can mention his work: “ One
series in the theory of linear differential equations of the second
order with periodical coefficients” (Notes of the Academy of Science,
physico-mathematical section, 8th series, 1902, vol. XIII).

1 1 9 . T h e c a s e o f a n a l y t i c c o e f f i c i e n t s . Let us suppose that p{x) and


q(x), with a real period co, are regular functions of the complex variable
x in a band which contains the real axis of the x-plane. Assuming that
x = x2 + ix2 we can maintain that this band is defined by the
inequality —h < x2 < +A. We can divide it into similar rectangles,
co in width, by straight lines parallel to the imaginary axis. In every
rectangle the values of p(x) and q(x) will be similar owing to their perio­
dicity. Take, for example, the rectangle defined by the inequalities:

0 < xt < co; — h < x2 < h.


452 LINEAR D IFFERENTIAL EQUATIONS [119

We now replace x by a new variable z according to the formula

z (259)

In the z-plane, instead of a rectangle, we obtain a circular annulus


bounded by the circles, centre the origin and radii e2nA'“ and Q~2nhla‘-)
this annulus will be cut along the radius in the direction of the real
axis; the opposite edges of the cut correspond to the sides xx = 0
and xl = co of the above rectangle. Owing to their periodicity our func­
tions have equal values along the cut and, consequently, the same
can be said about their derivatives of all orders.
In other words, the functions p(x) and q(x), which are functions of z,
will be regular and single-valued in the above annulus where they can
be expanded into a Laurent series:
+oe
P(x ) = a , 2*- ?(*) - M s-
S = — qo s= —»

We have from (259):


d _ .2 n d d2 in 2 , d2 4ji2 d
dx 2 to ^ dz ’ dx2 to2 ^ ds2 to* ^ d z’

and we obtain instead of the equation (239) the following equation:


in 2 „ d2u . f . 2ji . in'1 1 <\y
------r
to2
2 ds2
+ L
»—
01 2 ■“^ a.* is*-----to2r z J\ ds
+

+ 2 b, zSy = ° - (260)

The change in x along the section co of the real axis corresponds


to the completion of a circuit inside the annulus in the z-plane.
In the course of this the solution of the equation (260) undergoes a
linear transformation. If p(x) and q(x) are integral functions of x,
which frequently happens in practice, then the Laurent series in the
coefficients of the equation (260) will converge for every finite z
except, of course, z = 0. But in this case the equation (260) has,
in general, an irregular singularity at z = 0 since the above Laurent
series contain terms with negative powers of z.
Returning to the equation (239) we notice that, as a result of
the regularity of its coefficients at the point x = 0 we can construct
120] SYSTEMS OF LINEAR D IFFERENTIAL EQUATIONS 453

its solutions yx(x) and y2(x) in the form of power series to satisfy
the initial conditions (247):
yx(x) = 1 -f- a2x2 + a3x3 + .. ., y2(x) = x + /?2a;2 + £, x3 + . . .
These series will converge when [ x \ < h and, when p(x) and q(x)
are integral functions, they will converge for every value of x. When
h > to, we can use these series for the evaluation of y*(co) and y k(M)
in the quadratic equation (248).

120. Systems of linear differential equations. Previously we considered one


linear differential equation of the second order. This is a particular case of
a system of two linear equations of the first order [95]. In general, one linear
equation of the nth order can be represented by a system of linear equations
of the first order if the derivatives are accepted as the new unknown functions.
We shall now deal with the general case of systems of linear equations of the
first order of the form:
Vi = Pu (x) Vi + P2, (x ) y, + . . . + pm (x ) yn \
1/2 = Pi2(x)yl +p„i (x)y. + . .. + pn2 (x) ( (26I)
y'n = Pin (x) Vi + Pin (x) + p„„ (x) yn, I

where y, are the unknown functions, y] are their derivatives and are the
table of the given coefficients where, in contrast to the former notation [93],
we assume that the first symbol indicates the unknown function of which it is
the given coefficient and the second symbol indicates the equation in which
this coefficient appears. We shall apply the method of successive approximations
described in [95] to the above system and, consequently, all corollaries obtained
as a result of using this method will also hold. Let us recall these corrolaries.
I f all the coefficients Pjk(x) are regular in a circle \x — a \ < r then the system
(261) has a unique solution which satisfies at the point x = a any previously assumed
initial conditions:
Vi (a) = a,;- • •; yn (°) = an>
and this solution will he regular in the above circle \ x — a | < r. This solution
can be analytically continued in any direction except through the singularities of
the singularities of the coefficients Pik(x) and, in the course of this continuation,
it always remains a solution.
The solution of the system (261) consists of n functions. Suppose that we have
n solutions for the system (261). These solutions form a square table of functions
?/ll» V\t* • • ** V\n
Y = i/2l» J/i2’* ’ *» Vzn

i/nl’ Vnv' ** Vnn


where the first symbol denotes the number of the solution and the second
the number of the function entering the solution. We can now say that the
454 LIK E AH DIFFEREN TIA L EQUATIONS [120

solution of the system is a square table of the above type consisting of n solu­
tions; we shall denote by P a table consisting of the coefficients p ik(x) and by
F a table determining the solution. Using the multiplication law for matrices
we can write the system of linear equations as follows in the same way as we
did in [93]:
dF
fir = YP. (262)

Notice th at in this case we have used a different notation for the symbols
han in [93] and for this reason we have obtained a different sequence of factors
on the righthand side of the formula (262). Denoting, as usual, the determinant of
the matrix A by D(A) we can prove the following equation for the determinant
D(Y) of the solution Y:

/ [ P i . (* ) + P i. (* ) + . . . + Pm M l d *
D(Y) = D(Y) x -b e » (263)
where 6 is an ordinary point for the system (261), i.e. a point at which all the
coefficients p ik{x) are regular. Formula (263) is usually known as Jacobi’s
formula and is the generalization of the formula we obtained earlier for the Van
der Monde determinant.
Bearing in mind the fundamental definition of a determinant as the sum
of the products of its elements, we can say th at when differentiating a deter­
minant it is sufficient to differentiate separately every column and to add sub­
sequently all the determinants so obtained, i.e.

d P (F ) d 2/n> 2/u
dx
V1 1 . Viz
+ 2/u. V12
d® V211 Vzz I/21. Viz I/21. I/22 ’
where, to simplify notation, we have assumed that n = 2. Replacing the deriv­
atives by their expressions from the equations of the system we have:

d P (F ) Pll l/ll _~l~P2t I/121 I/12 1 l/u> PlzVll PzzVlZ


dx PnVz\.~\~ Pz\Vzi’ Vzt Vzu P 12 2/n P 22 2/22
Expanding the determinants into the sum of determinants and taking
p ik outside we can see th at certain terms consist of zero determinants since
they have similar columns, so th at the former formula gives

d D (F ) 2/n> 2/12 2/n> 2/12


dx = Pn + P22
2/ 21. V22 2/ 21 . y»z
or
dP (F )
dx = (Pu + P 22) D(Y),

from which Jacobi’s formula (263) follows. This formula shows th at if at a point
x = b the determinant P ( F) is not zero then it will not be zero for any x which
is an ordinary point of the system (261), i.e. at a point where all the coef­
ficients of the system are regular. If this is so we say that the solution F is
SYSTEMS OP U N B A R DIFFERENTIAL EQUATIONS
455
120]

the general solution corresponding to n linearly independent solutions which


form Y. We can then also consider the inverse matrix Y 1 where, as we ow

whence, from (262), we can see that it satisfies the system

d ^ 1 = _ p Y - i' (264)
da:
Let Z be a solution of our system (262), i.e.
d Z = ZP. (265)
dx
Construct the matrix
A = ZY-K
On using the usual law for differentiating a product [93], and also the
equations (265) and (264) this gives:

i.e. the matrix A is a constant matrix C, the elements of which are independent
of x. Hence
Z = CY
or, in other words, any solution of the system can be obtained from the general
solution by multiplying, on the left, by a constant matrix. Conversely it lollows
from the form of the equation (262) that by multiplying the so ution, o
left, by any constant matrix we can also obtain the solution. Bearing in mm
that
D(Z) = D(C) D(Y),

we can see th at D(Z) # 0 and this wiU be so then, and only then, when D(G) * 0,
i.e. multiplying on the left the general solution Y, by the constant ma rix
C we obtain the general solution when, and only when D(C) t - a °
follows from the formula (263) that in the course of analytic continue ion o
the general solution Y it always remains the general solution; we have a e y
mentioned this above in the definition of the general solution. Notice that
with the system of notation used in [93] we would have had to multiply by a
constant matrix not on the left but on the right so as to obtain ano
solution. ,
Let us suppose th at x = a is a point in the plane which is either a pole or
an essential singularity of the coefficient p ,&(#)■ When describing t is poin e
coefficients will return to their former values but the solution Y, in 6ene™ -
will become a new solution in the course of analytic continuation, w c can e
obtained from the former solutions by multiplying, on the left, by a cons an
matrix V:
Y + = V Y.
Let us call this matrix V an integral matrix when describing the point x = a.
Bearing in mind that
L>(T+) = D{V)D{Y)
and that in the course of analytic continuation a general solution always
remains a general solution we can say that the determinant of the matrix V will
not be zero. The matrix V depends on the particular solution Y considered.
If, instead of Y, we take another general solution Z = CY, where C is a cons­
tant matrix, the determinant of which is not zero, we have:
ZJ = C V Y = CVC~'Z,
i.e. the integral matrix of the new solution will be a matrix, similar to the matrix
V. In other words, different complete solutions have similar integral matrices.

121. Regular singularities. We shall now consider a singularity of a system


which is a pole of the coefficients of an order not higher than the first. To
simplify our notation we assume that this point is at the origin x = 0, when
we can write our system as follows:

x y\ = 9 u ( * ) Vi + 9 s i ( x ) 1/2 + • ■ ■+ 9 m ( * ) Vn
xy'i = 9i: (x ) Vi + 922 ( x ) y , + . . . + q„, (x) y„ (266)
xy'n = 9 in (x ) Vi + 92 n (x ) Mi + • ’ • + 9 n n (x ) 9 m .

where qgfx) are functions regular at the point x = 0:

Qlk (x) —atk + aik x + aik x* + • ■• (267)


We shall try to find the solution of the system (266) in the form:

Vi = *®(4° - f c(‘) x + . . .). (268)


Substituting in the system (266) and comparing the coefficients of x» we
obtain a system of homogeneous equations for the determination of the coef­
ficients Cj*:
(ou — q) c*1) + a21 c<2>+ . . . + anl c<"> = 0
°12 4 ^ + (°2 2 - 9) 4 2> + ■ • ■+ a „ 2 c 0n ) = 0 /OfiCH

°in 4 ° + °2 n 4 2) + • • ■+ K n ~ 6) = 0 .
and, subsequently, comparing the coefficients of xa+k we obtain a system of
equations for the coefficients cj.() when the former coefficients c<
f|l), where
m < k, are known:

(“n — e — *) 4 ° + ° 2 i 42) + • • • + <*nl 4n) = H i k


° 1 2 4 ° + (° 2 2 — e — *) 42) + • • • + °n2 4n) = n 2 k

«ln 4 ° + «2n 42>+ ■• • + Kn ~ 9 ~ k) 4^ = H nk


Vhere H jk are linear homogeneous lunctions oi tug UW llluon uo
uumiiucirno m*v±v
m < k. All these calculations are exactly the same, as in [98]. Denote by / ( g)
the determinant of the homogeneous system (269):
° n Q> °2( » . . ., O nI

°I2> °22 Q* • ■ •, o n2
(271)
/ ( B) =

°IIJ> °2 n< ■> ° n n — B

To obtain a solution of the system (269), other than a zero solution, we


must equate this determinant to zero

/(?) = 0; (272)
and during the further solution of the non-homogeneous systems (270) it is
necessary that the determinant of these systems should not be zero. This deter­
minant can be obtained from the determinant of the system (269) by replacing g
by e + i-e- *9 equal to / ( g + k). Let gj be a zero of the equation (272) so that
the numbers gt + k, where k is any positive integer, should no longer be zeros
of the equation (272). At the same time our earlier calculations will be formally
satisfied and we can construct the series (268) which formally satisfies the system
(266). I t can be shown, as in [98], that these series will converge in the circle
| a; | < r in which the series (267) converge.
When the zeros of the equation (272) are different and provided they do not
differ from each other by an integer, then the above method enables us to
construct n linearly independent solutions of the system (266). Otherwise, as
in [98], we shall also have, in general, one solution containing log x as well
as the solutions (268).
Let us write the system (266) in the form of a matrix
dF
X — 3—
dec
= YQ,

where Q is a matrix consisting of functions qik(x), regular at the point x = 0.


We can write this matrix in the form of a series in positive integral powers of x:

Q= -h A t + A z x2 +
X

where A s are matrices with constant elements: the matrix A 0consists of elements
ajk, the matrix A l of elements a'jk etc. The system (266) can be written in the
form:
= Y (A t + A, x + A 2x * + . . . ) . (273)

We shall try to find a solution in the following form for this system:
Y =' xw (1 + Gxx + C2xs + ...),
where W and Cs are the unknown matrices. We have:
dF
= W x w_1 (1 + C ,x + C ,x 2 + . . . ) + x w (Ol +2C„x + . . .).
dx
458 LINEAR D IFFERENTIAL EQUATIONS [121

Substituting in the equation (273) and multiplying on the left by x~w we


obtain:
W( l- \- Cl x-{-Ot x2-{-. ..)■)■ x (Pi "h 2C72x =
= (1 + C X* + 0 , * > + ...) (4 , + A lX + ...) .
A comparison of the constant terms gives
W = A0
and also, comparing the coefficients of z* we obtain a system of equations of
matrices for the successive determination of the matrices Ck:
A<,Ck + kCk = Ck-Ao + @k-i +■ • - + C\ -^*-1 + A k
or
■Ao@ k ~ C k A o + k ^ k = + ■■- + C \ - 4 * -i + A k.

Without attempting to investigate this system in general we shall only deal


with th at particular case when the matrix A 0 can be obtained in the form of
a diagonal matrix, i.e. a matrix S exists, which has constant elements and a
determinant, which is not zero, such that
S A „ s-i = [g„ e2.......e„],
where gs are the zeros of the equation (272).
Replace Y by the new unknown matrix according to the formula
Y = Y l S. (274)
Substituting in the equation (273) and multiplying, on the right, by S ~1
we obtain the following system for the matrix Y r

* - ^ - = * , ( £ , + B xx + + ...) , (276)
where
B k = S A k S~i,
and in particular
-B0 = [ei. &>•••• Gn]* (276)
As before we are trying to find a solution of the system (275) in the form:
Y l = x w' ( l + D l x + Dt x* + ...).
On substitution we obtain Wt = B 0 and the other coefficients are determined
from equations of the form
B 0 Dk — Dk B 0 -|- kDk = Ek, (277)
where Ek is a matrix expressed in terms of the proceeding matrioes Dm, where
m < 1c. Bearing in mind th at B 0 is a diagonal matrix (276) we obtain, from
(277), the following equation for the elements of the matrix Dk:

Ql { D k } iJ — { D k } ij 6 j + H & k } lj = { E k } lj>
i.e.

= et - o j + k
122] REGULAR SYSTEMS 459

If the difference of the zeros (p, — gj) in the equation (272) is not an integer
it is possible to determine all the coefficients. Notice that when there are equal
zeros among the zeros of the equation (272) but the matrix A 0can be obtained in
the diagonal form (it has simple elementary dividers) then the above calculations
remain valid.
In our arguments we did not touch the problem of convergence; this, as
we have said already, can be done as in [98], Notice also th at we assumed
above th at the constant term in the unknown solution of the equation (273):
Y = x w (I + (7, x C2 x- + . . -)
is a unit matrix. This, however, is insignificant. I t is only important that the
matrix should have a determinant which is not zero. In fact, let
Y = x ^ (C0 -t- Ci x -|“ C'2 x~ -(-...),
where D(C'0) 0. Consider a new solution
Co-1 Y — Co-1 x ^ Co Co"1(Co -(- Ci x + C2 x2 ).
But we know th at for any analytic function of a matrix
C '"1 = f(C'0-' W'C'o),
so that, for example;
Co-1 ew’Cq = ew (W =C o"‘ W'Oo)
and, consequently, the new solution will be:
Co”1 Y = x ^ (I -|- C[ x -|- C2xs -(-...) (C„ = Co"1Cfc).
Similar arguments can be used also in the solution of the equation (275).

122. Regular systems. Consider a system of simple equations the coefficients


of which are rational functions with poles of the first order at a finite distance
and zero poles at infinity. Let x = ajbe a certain pole of the first order of the
coefficients. Each of the coefficients Pik(x) has at this pole a residue w$(<)
and these residues form a square table Uj. We can therefore write our system
as follows:
dr Uj (278)
dx Y 2 x — aj ’
7-1
where Uj are matrices consisting of constant elements. We shall try to find
a solution for the system (278) so that at a point x = b, which is not one of the
points aj, it becomes a unit matrix; let us denote this solution by
F(6; x).
Bearing in mind this initial condition we can rewrite the system (278) as
an integral:

Y{b; (279)
x — aj
of its elements.
We shall now use, as usual, the method of successive approximations viz.
we assume th at Y 0 = 1 and make the successive approximations according
to the usual formula:
x m u ■
Y n (x) = I + f F n_, (x) 2 1 dx. (280)
J j —l •C aj
b

We have from the method of successive approximations:

Y(b; x) = F 0 + (F , (x) - Y„) + ( Y 2 ( x ) - F , (x)) + . . .

or, assuming for the sake of briefness that

Zn (X ) = Y n (X ) - F„_, (X ) (Z0 = 1),

where we have, from (280),

r TL
™ U,i
u
Z „ ( x ) = ^ Zn.A (x)
x) 2 T — dx, (281)
k
b J **
we can write:
Y(b- x) = I + Z i ( x ) + Z 2{ x ) + . . . (282)

We shall determine the first few terms of this expansion by using the general
formula (281). Introducing the notation

r / . f* dx x — ah
Lb{aji’ = l0« T ^ j ; ’
b
we have
X m U m
Z i (*) = f 2 - z r — A* = 2 Uh L b(aji; x).
J j = l x ~ aJ h=l
Similarly, introducing the notation

r /
L b (aji> ah \ x) = j
> f Lt>(°j.; x) ,
dx,

we have
r ™ m U 4,
: (*) = 2 Uh L„ (ah ; x) £ dx,
i !.=■> x ah
or
1,..., m
z« ( X ) = 2 Uh Uh L b (ah ,ah \ x) ,
iu j.
m . Continuing this further and introducing the formulae:

(283)

■which successively determine the coefficients L b(ajv ajv\x), we obtain:


..., m

J.... U
where the summation includes all letters shown under the symbol of summation
and every letter, independently of the others, runs through all the integers,
from 1 to m. Finally, from (282), we obtain the following representation for
our solution in the form of a power series of the matrix U j:
60
» fTl
I , , , . , Ill

Y(b\ x) = I + 2 2 v h .. . Ujv L b (ajv . . . , aJv; x), (284)


*= 1 j ..... > jv

where the coefficients of this series are determined by the recurrent relationship
(283).
The solution Y(b\x) can be analytically continued in any direction except
through the singularities aj, and the series (284) gives this solution in the whole
domain of its existence, i.e. for every analytic continuation. In fact, we shall
show, to start with, that the series (284) converges for every analytic continua­
tion of the coefficients Lb(ajv . . ., aj; x). Let I be a curve which originates at
the point x = b and remains at a finite distance from the points a-. Let d
be the shortest distance from the points to the curve I and a the length of
the arc of this curve from the point b. Using the usual inequality for an integral
round the contour I we obtain the following inequality for the coefficients of
the series (284) on I [4]:

0
whence
s

0 0
and generally on I:

But the power series


462 LINEAR D IFFERENTIAL EQUATIONS [122

converges ior every z and we can therefore say that the series (284) is absolutely
convergent for every matrix Uj and for every analytic continuation of its coef­
ficients [96], I t also follows from the above inequality that the convergence
will be uniform in every finite domain (in general, in a domain with several
sheets) which lies at a distance greater than zero from the point aj. Finally,
differentiating the series (284) term by term with respect to x it can readily
be shown th at it also satisfies the system (278). In fact we can rewrite it as
follows by isolating one of the summations:

m
Y(b\ x) = I + £ Uj Lb (ay x) +
j=i
— m
+ 2 2 2 u h . . . U j v U j L b (ah , ■ • •> a j v aj'< x).
'=1 J = 1 j „ — ,J v

Differentiating with respect to x and bearing in mind that from the definition

d-kf, {ay x) 1
dx x —a i
and
di-n (aj,.......ajv, a]\ x) L b (ajt,. . . , ajv‘, x )
dii X ^ Ot

we obtain by differentiating

'th -
dF(6;x) U ™ j U, r“ . b - - " 1 m Ui
ox = j2=1 —
x — aj7 + „=1
2 jU" 1 . . a Jv; x )j 2=i —
2 j v UJl. . . U Jjv L b (aj1. x, a};

dY(b;x) r ” rr r r ,1 £"> U,
UJ
=LI + ,fi j Z Jv uh- ■■U+L»(“h • • • aio’ X>J f i x l ^ T j

(b;x)
d F (6;x) Uj
= y(b,

Finally it becomes clear th at the constructed solution (284) becomes a unit


matrix when x = b because, from the definition (283), the coefficients of the
series (284) vanish when x = b. The above considerations give rise to the fol­
lowing theorem.
T h e o r e m . A eolation of the system (278), which becomes a unit matrix
when x = b,is determined with respect to x by the series (284) in the whole domain
of its existence and for every matrix Uj.
122] REGULAR SYSTEMS 403

If we make cuts lj in the x-plane from the points aj to infinity, so that the
cuts do not intersect, then in this cut plane, which will be a connected domain,
the solution (284) wall be a single-valued function of x, but on opposite edges of
the cut it will have different values, viz. when encircling each of the points
dj in the positive direction our solution will be multiplied, on the left, by a con­
stant matrix V, which we called an integral matrix corresponding to the sin­
gularity aj. We shall now deduce an expression for the integral matrices Vj
in terms of the matrices Uj which are part of the coefficients of the given system.
At the initial point x = b our solution is equal to I, i.e. it becomes a unit matrix
and, consequently, to obtain an integral substitution Vj we must determine
the value of our solution during analytic continuation round the closed con­
tour lj, which encircles the points aj and returns to the point b.
This value can be obtained directly from formula (284) when the formulae
(283) are integrals round the above closed contour lj and, in this case, the coef­
ficients obtained will obviously no longer depend on x.
We shall introduce the following notation:

P j(aj; b ) = -h e n / = / ,
J J J x — ajx |0 when j ^
b
and

Pj (ayi, . .., Oy„, b) = f Lb dx. (286)


j •E—ajp

We thus obtain Vj in the form of a power series of the matrices Uj and this
series converges absolutely for every choice of these matrices:

- i... m
Vj= 1 + 2 2 Ujl. . . U j vPj{ajl...,Oj„;b). (287)
J .... J v

T h e o r e m . The integral substitutions Vj are integral f unctions of the matrices


XJj as determined by the series (287), the coefficients of which are determined by
the formulae (285) and (286).
Instead of the formulae (286) the following formulae can be proved which
connect the values of Pj for adjacent values of v:

b
Pj (aji'- • ajv-i> b) Pj(ajl, . . . , a jv; b) -i
■fj (°/i>• • •>ajvi b) — J"
b ajv b Oyj J
(288)
We shall not give the proof of this formula.
If we analytically continue the constructed solution round any contour which
originates at and returns to a point x then this closed contour is, in the analytic
continuation sense, equivalent to several circuits round the points ay in the
positive or negative directions. Consequently, on returning to the point x
464 LINEAR D IFFERENTIAL EQUATION'S [122

our solution is multiplied, on the left, by a constant matrix which is given as


a product of factors Fy or V~l. In these circumstances it is usually said that
the integral matrices Fy form a group of the equation (278).
Let us explain this by a simple example. In the figure the singular­
ities a,, o2 and o3 are marked and the continuous line indicates the path of
the analytic continuation I which consists of
a number of circuits round the points ay. The
dotted line indicates a contour, equivalent to
the contour of the analytic continuation, and
it is assumed th at x = b.
The first circuit encircles the point al and on
completing this circuit we arrive at the point b
with the solution Vt Y(b\ x). The next circuit
encircles the point a3 and after completing this
circuit the constant matrix F, remains un­
altered and the matrix Y(b\x) is multiplied, on
the left, by F3, i.e. after completing the second
circuit we arrive at the point 6 with the solution
F ,F 3Y(6;
x finally, after completing the third circuit, we return to the point
);

b with the solution


F , F 3 F j 1 Y(6; x).
Any solution Y(x) of the system (278) differs from the solution Y(6, x)
by the constant matrix
Y(x) =OY (b; x),
and its integral substitutions will be as follows [120]:
CVjC-K
Consider now the matrix Yfbjx)-1 which is the inverse of the matrix
Y(b; x). This matrix, as we know from above, satisfies a system of linear equa­
tions
dY (b; x)-1 m V,
Y(b; x)">.
dx
Applying to this system the method of successive approximations we obtain
the following representation in terms of a power series of the matrix Uj:

Y(b; x)- 1 + 2 2
»=lfl....jv
u. ' . Cf jii Lb (aj\ ( ;t-» (289)

the coefficients of which are determined by the formulae:

L i (aj ■. X ) = - | = - log x ~ aU (290)


b — Oy,
b 11
and
x
' 0>jv] x)
Lb (O/i»■■•i&jvi x) = — dx. (291)
X — Oy.
b
122] REGULAR SYSTEUS 465

The expansion (289) is absolutely convergent for every matrix Uj and for
every analytic continuation with respect to the variable x. The results are
obtained in exactly the same way as above. Bearing in mind that

IVj Y(b; x)]-‘ = Y(b;x)-‘ V J \

we can see that by encircling the singularity a;- the matrix Y(b; x)~ 1 is multi­
plied, on the right, by the matrix Vj~l and it is thus possible to obtain an
expression for V j 1 in terms of a power series of the matrix Uj by using the series
(289) and by analytically continuing its coefficients round the closed
contour lj which surrounds the points ay. This gives us the series

VJ1 = 1 + 2 2 Uj,... Ujv P*j (ay,, . . . , a ]v;b), (292)


»-i J.... iv
in which the coefficients are successively determined by the formulae:

(293)

Lb (°y2>• • ■> a jv> x )


P* (a Ji ............“ajv;
7v> b)
d.r.
- S

Notice one particular case when the system (278) can be solved in final form,
viz. we assume that the matrices Uj commute in pairs, i.e. we have for any sym­
bols i and j :
U, Uj = U j Uj.

We will show that, in this case, the solution Y(b\x) of the system (278)
can be written in its final form as follows:

It can readily be seen th at the above function becomes a unit matrix when
x = 6. We will show that it satisfies the system of equations (278). Differentia­
ting according to the usual laws for differentiating products and bearing in
mind that

d d u' logi b ^ ( x - a j \Ui Uj


(295)
dx \ b — aj ) ~ dx e b — ay ) x — aj
we obtain
dF(6; x) ™ (x-a j-A Uj I x - + \ u>
da: Vb — a, J "' [ b — aj_t J x — aj V6 —ayJ

f x - am \ V™
U —onl I
466 LINEAB D IFFERENTIAL EQUATIONS [123
Since the matrix ZJj commutes with the matrix Ut it will also commute with
every other function /(C7;) which is given as a power series of Ut. We can
therefore write the above formula as follows:
d T (b;x) ™ f x - a , -\u, ( x - a m \ Um Vj
dx j Zi Vb — a i ) — x —a j '

whence d Y{b; x) m U,
dx j=l x aj
i.e. the matrix (294) does, in fact, satisfy the system (278). The formula (294)
can be obtained from the system (278) if we perform a purely formal separation
of the variables in this system without taking into account the fact that we are
dealing with matrices and not with numerical variables. In this case this is
possible owing to the fact th at the matrices Uj commute in pairs. The right
hand side of formula (294) represents the sum of the series (284) on the assump­
tion th at the matrices Uj commute in pairs. I t also follows from the formula
(294) th at in the case under consideration the matrix Y(b\x) acquires on the
left, the constant factor shown below, on encircling the point at
e2*' u U j .
This follows directly if we write the formula

and use the known many-valuedness of the logarithm.


Notice also th at the relative position of the factors on the right-hand side
of the formula (295) is unimportant, for both factors only contain one ma­
trix Uj and do therefore commute.

123. The form of the solution in the neighbourhood of a singularity. We shall


now consider the logarithms of integral matrices with an additional numerical
factor

ifrir. <»>

We have taken the principal value of the logarithm to be given by a power


series which converges when the matrix Vj is sufficiently near to a unit matrix.
I t follows from the formula (287) that this condition will certainly be satisfied
if the matrices Uj are close to a zero matrix and we shall assume this to be so
in future. Substituting in the series (296) the expression for (Vj — 1) from the
formula (287) and grouping similar terms together we obtain an expression
for Wj in the form of a power series of the matrices Us which converges
when these matrices are sufficiently close to a zero matrix:
- l,...,m
Wj = Z 2 Uh . . . UjvQh (ah .......ajv;b). (297)
Jlf-Jv
123] THB FORM OF THE SOLUTION IN TH E NEIGHBOURHOOD OF A SINGULARITY 467

We shall not evaluate the coefficients of this expansion since it can easily
be done by the direct substitution of our series into another series. Consider now
the elementary function

l x - a j \wt ^logjE-g
(298)
I b — a] )
Taking the corresponding values of the logarithm which vanish when x = b
we can see th at the function (298) becomes a unit matrix when x = b and,
by encircling aj, the logarithm acquires the term 2ni and the function (298)
becomes the new function

( x — a] \ w i TT ( x ~ ay
w' H +logH ] )
= e2*1 — aj) \ b — a.j)

the relative order of terms in this expression is unimportant for both are power
series of one and the same matrix Wj and, consequently, commute with each
other. We can therefore see that the elementary function (298), after encircling
aj, acquires on the left the same factor Vj as the solution Y(b; x) and it also
becomes a unit matrix when x = b. We can therefore write:

Y(b;x) = (299)

where yb>(6; x) is a matrix equal to a unit matrix when x = 6 and which is


single-valued in the neighbourhood of the point x = aj. We will now show that
it will not only be single-valued in the neighbourhood of the point x = b but that
it will also be regular at the point x = 6, i.e. the factor [(x — a})j{b — aj)] contains
not only the branch-point of our solution but also all the characteristics of the
solution at the point aj as was the case with regular singularities of equations of
the second order.
We have from (299):

(300)
y 0 ( b : x ) =l T ^ - J r(6;x),
whence, differentiating with respect to x:
d yU>(ft; x) TVs — r x — ca i\d Y ( b ; x )
A t.
------- — k ---- ^
x — aj \ b — aj J
T(6;x) + \ -r---- L\
b — aj J Hr

or, using the equations (278) and (300):


dy ^ b jx ) W ^ I x — as — W f m v.
dx aj Y ( b ; x ) + \\-0r —
—Oj i- ' y(6;x) v
5=1

i.e. the matrix (6; x) is a solution of the system of equations

d YU>(b;z) ~m b' y Us Wj Y U) (b ; x)
(301)
dx ' ’ ' sTi x — as x — aj
4G8 LINEAR D IFFERENTIAL EQUATIONS [123

Looking at the right-hand side of formula (300) we can see that both factors
are power series of the matrix Us and therefore their product is too; if all the
matrices Us are equal to zero then Wj will also be zero and the first factor to
the left on the right-hand side of (300) becomes a unit matrix. Similarly for
Y(b;x) and therefore also for yU\b; x). I t is thus possible to seek Y^(6; x)
in the form of the following power series:

(6 ; x) = 1 + J - 1 "■£ Uh . . . Ujv W (oy, .......aJv ; x ). (302)


'~ l J l ....... J v

Substituting the series (297) and (302) in equation (301) and comparing
the coefficients of the product C7, . . . Ujy we have:

d L ^ (flji> . . . . UyV; x) iJjj'* (ojl i ■... *^) _


dx x — ajv
1 V •
- - — — 2 Qj(ah " " > aik; b) LbJ) »■*- ah>; x)
'T ™
“aj k= 1
and, in particular,
dLi/\cij l ; x) _ 1 _ Qj (a J i i b )
dx x —cijl x — Oj
Notice that in the above sum with respect to k the second factor becomes
devoid of meaning when k = v and, in this case, it should be replaced by unity.
In future we shall often deal with analogous sums in which the factors of border
terms become devoid of meaning when the accepted method of notation is used;
they should then be replaced by unity.
We have already mentioned above that the matrix Y ^\b; x) becomes a
unit matrix when x = b for any matrix E7y sufficiently close to zero, i.e. all
the coefficients in the expansion (302) must vanish when x = b. Bearing this
in mind and using the above formulae we can write the following formula which
successively determines the coefficients in the expansion (302):
X

U j) (oy,, , cij„ ; x) = I' H)i) (Oh »• ■• . °;V-1 ; x)


x — ajv
*6

I Qj(*j . ah b) W . ; *)] dx. (303)


•*J
In particular, when v = 1, we have:

Li/) (ah ; x) = Qj (ah ; b) 1 dx. (304)


x-aj \

The coefficients in the expansion (302) must be single-valued functions in the


neighbourhood of x = aj, for we know from above that the sum of the series
is a single-valued function. I t follows th at the residue at the pole x = Oy
124] CANONICAL SOLUTIONS 469

under the integral in the expression (304) vanishes and therefore the func­
tion (304) also regular at the point x = aj. We shall now continue the proof
from (v — 1) to v. Let us suppose that all the functions

7}J\ah ........ aj, ; x) (305)


are regular at the point x = Oy when s < v. We will show that the functions
(305) will have the same property when s = v. We have the formula (303)
for these functions. As a result of the regularity of the function (305) when
s < v the integrand in formula (303) can only have a pole of order one at the
point x = Oy. However, if the residue at this pole had been other than zero then
the function (303) would have been many-valued in the neighbourhood of the
point x = Oy, which is impossible. I t therefore follows that the integrand func­
tion in formula (303) and the integral itself will be regular at the point x = Oy.
We shall not give a more detailed determination of the coefficients in the
expansion (302) here.
All the above arguments applied only to the case when the matrices Us were
sufficiently close to zero. Later we shall give a representation of the matrix
Wj and of connected matrices which will hold for any matrix; it will then be
evident that the singularities in this representation will be those matrices
Us among the characteristic zeros of which are zeros which differ from one other
by an integer other than zero.

124. Canonical solutions. The solution F(ft; x) depends on the choice of Ihe
point b at which we approximate the matrix to a unit matrix. For this reason
the matrix F(6; x) is known as a matrix (solution), normal at the point x = 6.
This latter point must not be one of the singularities Oy. We obviously cannot
make any initial conditions at the singularity x = oy but we can attem pt to
construct a solution which would have the simplest form in the neighbourhood
of this singularity; this can be done in exactly the same way as in the construc­
tion of a solution in the neighbourhood of a regular singularity of an equation
of the second order. We shall now construct this solution and call it the canonical
solution at the singularity x = aj.
We can write:

Y (b ; x) = ? U) (&!*) = ( * - aj)wt (b - oy)~ WI Y«> (ft ; x) ,

where the relative position of the first two factors on the right-hand side is
unimportant for both factors contain only the one matrix Wj. Combining the
factor (b — with the factor Y^\b; x) we can write:

Y (b ; x) = (x - aj)wi Y (j>(6 ; x ), (306)


where
Yu) (ft ; x) = (ft - a j)~ wi F y) (6 ; x)
is a matrix, regular at the point x = ay. If all matrices Us are zero matrices
then Y^\b; x) becomes a unit matrix and, consequently, the determinant of
this matrix is not zero provided that Us is sufficiently close to zero. The deter-
470 LIN EA R DIFFERENTIAL EQUATIONS [124

minant of the matrix (6 — aj)~wj = e~wj lo« “d is not zero for it is the deter­
minant of an exponential function of a matrix [93] and, consequently, the
determinant of the matrix Y ^ \ b \ x) is not zero at the point x = Oy if all
matrioes Us are close to zero, i.e. in this case the matrix Y^^(b; x)~ 1 will be
regular at the point x = a y . Every solution of our system differs from the solu­
tion Y(b; x) by the constant factor C (the matrix on the left):
Y (x) = C Y (b ; x) (307)
and we assume that the determinant of C is not zero so as to obtain a comp­
lete solution. In place of formula (307) we can write:

Y (x) = C (x — a j ^ C - ' C Y ^ (b ; x) ;
but from [121]:
C (x — aj)wt C~l = (x —a,j)w'i,
where
Wj = CWj C~1. (308)
We now choose a matrix C equal to
0 = [yW>(6; o y )]"1, (309)
so that we have:
CYW (b ; x) = 1 when x= oy.

We thus obtain a solution which we denote by 0 y (x ) and which we call


canonical at the point x = ay. This solution has the form:

0y (x) = (x — a})w'i Bj (x),


where Wy(x) is a matrix, regular at the point x = Oy, which becomes a unit matrix
at th at point. We will now show that in this canonical solution the matrix IF'.
coincides with the matrix Uj.
Notice, first of all, th at all the matrices constructed above are given as power
series of the matrix Us provided the latter matrices are sufficiently close to
zero. In this case the matrix Wj, like the matrix Wj, should have no constant
term in its expansion, i.e. we should have the following expansion

W j= £ "Z Uh . . . Ujv J j (ah , . . . , oy.). (310)


"=1 j i. - Jv

Differentiating with respect to x the formula


0y (x) = (x —aj)~Wi 9j (x),
wo obtain, as in the previous section, the following system of equations for
the elements of the matrix 0y(x):

Us m Wj 6j (x)
d0y (x)
dx = Oj (x) Z (311)
j=I X — a. CC “ Clj
124] CANONICAL SOLUTIONS 471

If all the Us are equal to zero then ~Sj(x) becomes a constant matrix and,
as a result of the condition at the point x = ay, it must be a unit matrix, i.e.
vce must have the following expansion:

6j ( x ) = 1 + 2 ’ ' "2 uh . . . U j v N j (ah . . . a J v ; x ) . (312)


*■=1 J h - J v

All coefficients in this expansion must be regular at the point <jy and must
vanish at that point since the sum of the series, for any Us, should become a
unit matrix at the point x = ay. Substituting the expansions (310) and (312)
in the equation (311) we obtain the following equation:

N j (a j i ...................a J v - i ; x)
, ajv; x) =

n
Nj (ah ,
x — ajv
°y
i v 1
- —— — 2 »• • • »%) N j (aJk+l . . . ajv ; x) d x , (313)
x ~ aj k=\ J
and, in particular
A
N, (ah ; i) = [ dx.
X “ Q 1.

Owing to the regularity of the left-hand side, this latter equation shows
that
j , , f 1 when ji = 7 ,
(314)
* ( 0 when ;t # j.
Let us consider the equation (313) when v = 2:

r \ N, ( a h -,x) 1 T i
Nj (aJl, a}i ; x) = J ( ^ - x _~aJ IJ j (ah) N (aJt ; x ) + J j (aJlt aJt)]j dx.

I t is given th at we must have Nj(ajs;aj) = 0 and, consequently, the first


term of the integrand has no pole at the point x = a-j. I t follows that the
second term does not have a pole at that point either and that the square
bracket vanishes when x = aj\ therefore all the coefficients must
be zero. Similarly, considering the equation (313) when v = 3 we can see that
all the coefficients J(oJi, oJa, Oy3) are zero etc. As a result of this and (314), the
expansion (310) becomes the simple equation W'j = Uj and we have the fol­
lowing representation for the solution canonical at the point x = Oy:
Bj (x) = (x —a,j)ui 9j (x). (315)
Formula (313) enables us to evaluate successively the coefficients in the
expansion (312). Bearing in mind that
when j l = j
when 7\ ^ j ;
J ] ( ah ......... ai») = 0 when v > 2 ,
472 LINEAR DIFFEREN TIA L EQUATIONS [126

we have:

Nj(ah - x ) = f [----1-----------S
± L - dx,
J n J I X ---(ljx X —Oj

r r N j ( a h . . . a Jv_ , ; x) SN N j ( a Jt . . . a j v ; * ) ]
Nj iaji y ■ ■ ■ i a jv ; X) = J [ — ----------- ^ - - - - - J d*.
where 6pll = 1 when p = q, and 8pq = 0 when # g.
By describing a circuit round the point ay the solution (315) acquires a
factor e2nlUt on the left. Any other solution, as we know, will have an integral
matrix, similar to the matrix e2”tUf, i.e. by describing a circuit round the sing­
ularity aj any solution of the system acquires on the left a factor which is a matrix
similar to the matrix e2!t,Ui.
Let us now return to the formula (315). The second factor, as we said before,
is regular at the point x = aj. The inverse matrix

Oj(x)-1
will, therefore, also be regular at the point x = Oy for the determinant of the
matrix Oj(x) is equal to unity at the point x = aj. In general, if any solution
Y(x) can be represented in the neighbourhood of the point aj in the form:

Y ( x ) = ( x - a j ) w] Y { x ) ,
where the matrix Y(x) is regular at the point aj and its determinant at that
point is not zero, then the matrix WJ is said to be an exponential matrix of
the given solution. I t can be shown th at this matrix is determined in a unique
way for a given solution when the matrices Us are close to zero. In particular,
for the solution, canonical a t the point oy this will be the matrix Uj itself, and,
generally, for any solution, it will be a matrix similar to the matrix Uj.
Note. In all the above arguments we have used the fact that the representation
of a function of a matrix in the form of a power series of these matrices is unique.
This uniqueness theorem forms the basis of the method for the comparison of
coefficients; we used this method above by substituting a series with unknown
coefficients into both sides of the equation and comparing the coefficients of
similar terms. The hypothesis th at the sum of a power series of a matrix Us is
a single-valued function of x near x = oy and th at all the coefficients of this
series must be single-valued, is also based on this theorem of uniqueness.
We said earlier in [94] th at the uniqueness theorem holds if the sums of
the power series coincide for matrices of every order. In all our arguments the
order of the matrices was immaterial and it was therefore permissible to use the
uniqueness theorem.

125. The connection with regular solutions of Fuchs’s type. Let us now
consider the canonical solution at the singularity x = Oy:
Qj (x) = (x — aj)Ul 6j (x).
125] TH E CONNECTION W ITH REGULAR SOLUTIONS OP FUCHS’S TYPE 473

For the sake of simplicity we assume that the order of the matrix n = 2,
i.e. th at we have a system of two equations with two unknown functions.
Let Sj be a matrix which converts Uj to the diagonal form:
Sj U j S f 1 = [<?!, e2].
Consider the integral matrix:
Zj (x) = Sj 8j (x) = (x - aj)slu>si 1Sj Bj (x) p
0r _
Zj (x) = (X — Oy)t?1,e,l Z j ( x ) ,
where
Zj (x) = S j Bj (x)
is regular at the point x — aj. Denoting by Z^(x) the elements of this latter
matrix:
( x ) , z[»(*)
Zj (x) =
2S5>(*),zS2(*)
where Z^(x) are functions which are regular when x = aj and bearing in
mind th at
(x - aj)e', 0
(x — =
0, (x — aj)e‘
we have:
(x —ajf', 0 (*), Z[» (x)
Zj (x) =
0, (x - a , r Z[Ji (x) , Z$ ( x )
= (X - Oy)Cl Z « > (X ) , (X - a jr z['} (x )

(x - a / * Z(H (X ), (x - a j f ' Z<£ (x)


Every line of this matrix contains a solution of the above system [120].
We thus have two solutions of the system which have the same form as the
solutions of one regular equation in the theorem of Fuchs [99]:

Y „ (x) = (x - a,)5’ Z(H (x) ; Y 12 (x) = (x - ay)®‘ Z(£ (x) ;

Y n (x) = (x - ajr Z<H (x) ; Y „ (x) = (x - a/* Z $ (x).


In these formulae the first subscript of Y ( x ) gives the number of the solution
and the second the number of the function. Notice also that from the definition
of Z j ( x ) and the fact that Vj(dj) = 1 we have

Z&(aj), Z $(aj)
Z j(aj) = = Sj>
Z $ (a j), zJJhaj)
where Sj is a matrix, the determinant of which is not zero. The number Z^^(aj)
must be the constant term in the expansion of Z(J)(x) into the Taylor’s series in
powers of (x — aj).
474 LIN EA B DIFFEREN TIA L EQUATIONS [126
The numbers g, and g2 which in [98] were the zeros of the determining equa­
tion are, in this case, determined from the characteristic equation of the matrix
Uj. In the works of I. A. Lappo-Danilevski the integral matrix Oj(x) is known
not as the canonical matrix but as the metacanonical matrix at the singularity
x — ct]. In this terminology the matrix Zj(x) can be called canonical at the
point x = aj.

126. The case of the arbitrary Us. The formula (297) in [123] determines an
exponential substitution Wj for an integral matrix Y(b\x) in the form of a
power series of Us which converges only when the U's are close to zero matrices.
Similarly, formula (312) in [124] gives the analogous representation for the
regular factor of a canonical matrix 0s{x). We shall now examine this repre­
sentation for arbitrary matrices Us.
By definition we have for the Us which are close to a zero matrix [123]

r-1
Let us denote by g,, {?», . . . , gn the characteristic zeros of the matrix Uj. We
know from [124] th at the matrix Vj is similar to the matrix &niui and there­
fore the characteristic zeros of the matrix Vj will be:
V\ = ean,e‘; rjz = e2"*®*;. . . ; = e2*'®..
If we suppose th at the T]k are all different and use Sylvester’s formula we can
write:
W = JL V (FJ~ *>• •~(F1~ <*0~ ~ lng r„_.
1 2 n i * fi <r] k - r h ) . . . ( r i k - T i k _ l )(Tik - r lk+l ) . - . ( T) i t - V n )

In future, for the sake of simplicity, we shall only consider the case when
n = 2. Replacing ijk by the expression containing gk we have:
Vj — e2"'®* Vj — e2n'ei
~ e2nie, — e2nlei e2n'e2 — e2n,Cl ^2’
or
--- ------ 1----- « _|----- St —Si y
=1----=1--- (316)
e2nie, _ e2ii(si e2»/ei _ 02j«ei 1
When gj = g2 this formula becomes
1 (317)
Wj = ___ Vi.
2me2n'«‘ 1
Above we obtained a representation of Vj in the form of a power series in
Us for any Us. Similarly formula (316) gives a representation of Wj for any
Vs. This formula becomes devoid of meaning when pt and g2differ by an integer
other than zero, for then the denominator on the right-hand side of (316)
vanishes while the numerators are other than zero. Thus for Wj as functions
of Us, those matrices Uj will be singularities, the characteristic zeros of which
differ by an integer other than zero. With regard to the remaining matrices
12 6 ] TH E CASE OF TH E ARBITRARY V, 475

XJS the function W j has no other singularities. The existence of these singularities
* causes the series (297) to converge only when the U s ’e are close to zero matrices.
We will show how the series (297) can be used for obtaining W j in the form
of a quotient of two power series which converge for any U s . Let us construct
a numerical function of U j , i.e. a function which for a given U j has a definite
numerical value:
e2nl'e, _ 02J«'pi sin n (e, - e2)
A ( U j ) = e “ " ' (e‘ + e ,) (3 1 8 )
2m (g, — g2) ” (ei - e2)
We can represent it in the form of a power series which converges for any
g, and g 2 :

(319)
Denoting by {Uj}pq the elements of the matrix Uj we can write the quadratic
equation which is satisfied by gj and g2-
{ U j }ii ~ C?> {Uj } 12 _q
{ U j } 2i> { U j}i2 6

We have further:
(ei - e2)J = (Qi + Qz)1 ~ *QiQ2>
and, bearing in mind the property of the sum and product of the zeros of a
quadratic equation, we obtain an for expression ( — p2)* in terms of the ele­
ments of the matrix Uj-.
(C i - e 2) 2 = ({U j}n + {U j)„)* - 4 ( ( U j ) n { U j ) 21 - { U J) l l { U j ) 2i ) .

Substituting in (319) we obtain an expression A(Uj) in terms of the elements


of the matrix Uj\

A {Uj) = J o (2 7 + 1 )1 ^ [({C7J'}l1 + ~ 4 « ? /> » ~

and this series converges for every U j , i.e. it is an integral function of the
elements of the matrix U j .
For the sake of briefness we denote the terms of the above sum by 6 v ( U j ) :

^ (Uj) = 2 * . ( U j ) , (320)
>.=0
where S0(Uj) = 1 and, when v > 0, Sv(Uj) is a homogeneous polynomial of
degree v in the elements of Uj. I t follows from the formulae (316) and (318)
that the elements of the product A(Uj)Wj are integral functions of the elements
of Uj and, in general, they are integral functions of the elements of all the
matrices Us. This function can be expanded into homogeneous polynomials of
the elements of Us [83]. Bearing in mind the expansions (297) and (320) we
can write the expansion into homogeneous polynomials as follows:

^ (U j) W j = 2 V ( 'Y ™ U j r - U j M * r S ( U j ) Q j ( a j l . . . a , s ; b) )
»=Is=l \j„...Js J
476 LINEAR D IFFERENTIAL EQUATIONS [126

The above series converges for every U s . We thus obtain an expression for
Wj in the form of a quotient of two integral functions of the elements of V s :

2 2 [ 2' Uh . . . U j s dv_s (Uj) Qj (ah . . .aJa; b)


Ty [ Jv - J*_________________________________ (321)
i^ (^ )
v= 0

Notice th at the numerical terms in the series in the denominator depend only
on the elements of the matrix U j . Arguing in the same way as before we can
show th at the products
A (U j) (x — a j ) w i and A (U j) {x — aj)~Wi
are integral functions of the elements of U s . I t follows from the formula (306)
that
A ( U j ) Y ^ ( b ; x ) - 1 = Y (b ; x ) ~ ' A (Uj) (x — aj)wt .

The matrices Y(b;x) and Y(6; x)~*, as we know, are integral functions of
the matrices Us and, consequently, the product A(Uj)Tii)(b; x)~1 is an integral
function of the elements of Us. The canonical matrix 8j(x) can be written in the
form [124]:
flj(x) = r«>(6; Oj)—1 Yp(b ; x) ,
and, consequently, A ( U j ) 8 j ( x ) is an integral function of the elements of U s.
The same can be said about the product:

A (U j)8j(x) = (x - a j)~ ui A ( U j ) 8j (x ) ,
since (x — O j ) ~ Ul is an integral function of U j . Using the expansion (312) we
can also write the canonical matrix 6j(x) as a quotient of two integral functions
of the elements of U s :
,, *> » n ....... m -v
( x - a j ) ui 2J 2 \ 2 Vh - - - U],Sv_s ( U j ) N j ( a ] i . . . a j , ; i )
Bj (x) = ____ »-o »-o Wi i>________________________________________ i .(322)
2M U j)
p-0

Notice th at in all the above formulae, A(Uj) can commute with every matrix.
The numerators in the formulae (321) and (322) contain series the terms of
which are matrices which because of the factors Uj%and the numerical factor
dv_s(Uj) depend on the elements of Us .
The formulae (312) and (322) give the canonical matrix in the form of a
power series or as a quotient of two power series, in terms of the elements of the
matrix U s . At the same time the coefficients N j ( a . j v . . . , a j t ; x) depend on x .
We can, conversely, write O j ( x ) in the form of a Taylor’s series in powers of
( x — c t j ) . The coefficients of this series will also depend on the elements of U s .
This series will converge in the circle \ x — | < R which contains no sing­
ularities except x = a j.
EXPANSION IN TH E NEIGHBOURHOOD OP AN IRREGULAR SINGULARITY 477

We had the equation (311) for Oy(x) and we have shown that W'j = Uj,

d6jW_ =- J- _ U s ------UjQj(x)_ _
dx 1 s-i x —as x — dj
Substituting this in the equation

fly (x) = 1 + i 1 A j(p) (x - n7)P, (323)


P-i
where are the unknown matrices which are independent of x, and equating
the coefficients of equal powers of (x — Oy) we obtain an equation for the
successive determination of the matrix A ^ :
p—1 j( .DTJ
V } A j( p) + P a Y> - Af Uj = - y 2 -------- ~ (P = 1, 2, . .. ) • (324)
« -» (o A — Oj)p 9

We have already met similar systems in [121]. We are not giving here the
solution of the equation (324) or the proof of the convergence of the series
(323). We can use for this the same method of proof as in [98]. Notice only that
the product A (U j)A ^ is an integral function of the elements of the matrix
Us and that the following formula holds for this product:
T
A (U j) A j(p) = 2 2 Sp-kiUj) f ( - D * * l U j T p U j - xj .
p-o L*"® a- o A1 (fc — A) 1

where T p denotes the right-hand side of the equation (313).

127. Expansion in the neighbourhood of an irregular singularity. We shall


now consider a system of linear equations, the coefficients of which have a
pole of any order a t the point x = 0 and, for the sake of simplicity, we assume
that the coefficients of this system are quotients obtained by dividing a certain
polynomial by a whole positive power of the variable x. Using the matrix
method of notation we can write this system as follows:

— = Y i T p xp, (325)
p— s
where T p are the given matrices. The point x = 0 will, in general, be an
irregular singularity of the above system but we can nevertheless apply the
method of successive approximations; we thus obtain the solution in a finite
form which will hold for every analytic continuation with respect to x. This
solution, as always, will be given as a power series of the matrices T p which
occur in the coefficients of the system. Let us take a point 6, other than x = 0
and construct the solution F(6; x) which becomes a unit matrix when x = 6.
For this solution we can write the usual integral equation

Y (b ; x) = 1 + J2 Y (6 ; x)
h
£
'
P -----S
T p xP dx.
478 LIN EA R DIFFEREN TIA L EQUATIONS [127

Assuming th at F 0 = 1 and

Y„ (x) = 1 + f Yj,., (x) £ T p (x)P dx, (326)


b P— s
we have
Y (b ; x) = F 0 + [Y ! (x) - F 0] + [F 2 ( x ) - Y t (x)] + . ..
Writing, for brevity:
2„(x) = F l, ( x ) - F v_1(x) (Z0 = 1),
we can, from (326), write

2 , (x) = f 2,,., (x) 2 T p xPdx. (327)


6 P—

Let us introduce functions of x which are determined by the following recur­


rent relationships:
x x
LPl (b ; x) = ^ xp*dx ; Lp,,. . . ,p^ (6 ; x) = j* LPl.......Pv_^ (b ; x) afv dx , (328)

On evaluating by the method of successive approximations we have, from


(327):

r 1 1
z i (2) = j I 2 T PixPl d x = 2 T PlL Pl( b ; x ) ,
b Pi— s Pi— s

? < ( '/
2 2 (x) = J 2 7Tp1_Lp1(6 ; x) 2 T PaxPl dx = ^ T PlT PlLpiP2 ( b x) ,
b P i “ —5 —s Pi» Pa " —5

and in general

£ „ (* )= S T p i . . . T p L p i ...pv ( b - , x ) .
P it . . . »Pp" — s

Thus the required solution is obtained as a power series of matrices

Y ( b ; x) = l + 2 2 T Pl. . . T p vLPl...Pv(b-, x). (329)


►-1 Pi...... pv----s
We can, as in [122], prove th at the above series converges absolutely and
uniformly and th at it gives the required solution of the system. In this case we
can perform the squaring shown in the formulae (328) and we can therefore
write the coefficients of the series (329) in a definite form.
Instead of the functions (328) we shall consider, first of all, the above squares,
but only after using a different method for determining the arbitrary constant
x
M pi (*) = J xP‘ dx ; • • • ;

m pi - p„ (x ) = J m p i -■■p , - j (*)xPv A x -
127 J EXPANSION IN TH E NEIGHBOURHOOD OP AN IRREG U LA R 8INGULARITT 479

Wo shall find the arbitrary constants in these equations as follows: when


pt + • • • -\- Pv + v T4 0, the function M pi ___^ (x) must be of the form:

2Vj ° pi- •P. log'' X ,


EP l + . . , + P +»
AT,Pi - Pv (x) (330)
/i-0
where a^jl. Py are numerical coefficients. If, however, p x -)- . . . -f- p r + v = 0
then the constant of integration will remain arbitrary. We will show, first of
all, that this determination of the arbitrary constant is possible. When v = 1
we have:
x xPl+1 • ^ ^ , when p x + 1 ^ 0
Mpx (x) = J xp‘ dx =
ap°i + log x , when p x + 1 = 0 ,
where is the arbitrary constant. We shall now assume th at formula (330)
holds for every M pi (x), when A< v, and investigate the function-8fPl. ,.p (x):

ifp ,...p (* ) = J x P ‘+ - " 'V + ” 2


p-0
a<Pi...pv l o g " x - x pv+i d x .

There are two cases. When p x + . . . + Pv+i + v + 1 ^ 0, integrating by


parts we have:
xp‘+---+pv+i'‘+1 log'1x —
M Pi- ■Pv+,
Pi + • ■•+ Pi-n + p + 1
r xp‘+---+pv+i+r
2 Va(p i...p v log" 'xdx.
J p l + - - - + p„+l + p + 1 M-l
Continuing to integrate by parts wo finally obtain the following expression:

M p i - • -p v + x (*) = * Pl+ ■- + p™ + v+ 1 2 l “ PL - .PV+I l o g " * .


/J-l
where the coefficients ...P(,+, can be expressed linearly in terms of
the a*11) p and will contain no other arbitrary constants except the constants
* <>■:;/
If, however, p , + . . . + p v+l + r + l = 0 then

TP' + - . - + PV+ v X pv+ i =

and we have
q(")
M,Pi- “P l-P y <p ) log" X
(*) ,(° >
aPi-- ■P-+. + log"+1x = 2,’ Pi..
P-0 A*+ 1 p-o
where apl>i. . . Pv+i is a new arbitrary constant. We have thus shown that it is
possible to determine the constants so that formula (330) holds. I t also
follows directly from the above that the arbitrariness of the coefficients is due to
the arbitrary choice of the coefficient a when p x + . .. + p, + v = 0.
480 LIN EA R DIFFEREN TIA L EQUATIONS [127

We shall now write down relationships which make it possible to evaluate


the aj") successively. When v = 1 our earlier calculations give:

(°) f---V"T when p, + 1 # 0 , f 0 when p, + 1 ^ 0 ,


a' 1 = Pl + 1 “* “ ( 1 when p. + 1 = 0 .
( arbitrary when p t + 1 = 0 ,

I t also follows from the definition of M Pl Pr(x) that :


M Pi - - - p v ( x ) = M Pi - - P v - t (x ) x P v

and from (330)

» fl. „ log" 1* =
(Pi + ■• • + Pv + v) 2 a{p l . .p log" x + 2 iia
p-0 p-1

= p-0
2
whence
(Pi + • ■• + Pv + v) aPi-- Pv — 0 >

(Pi + • • • + Pv + v) a<Pi---Pv + (P + 1) a<Pi*^P„ = a<p‘i- -Pv_I


(ft = v — l, v — 2........1, 0).
Consider first the case when p l + . . . + p , + v # 0. We then have:
a(')
Pi---P„ ' ° i• „<"> . = ________ I_____
aPl'—Pp P. + • • • + p„+ + “ (/i + 1) api+-1>p-1-
The successive application of this latter formula when p = v — 1, v — 2, ...
gives:
n(»—1> - (’—0
aPi-- Pv - Pi + • • • + Pv + v aPi-Pv-i -

( V_ j ) _
1_______ L l'-s ) — v- 1
api- p" - P , + •••+?,+»’ L pi—p*-‘ Pl + • • • + Pv + 7 aF i - J ’
and, in general
„(") _ P+1 „0*+l> -I-
“Pi •••/>„—-----------
Pi + - - - + P p + f [aPi- Pl + + Pv + V
*t *

I (p + 1) (p + 2 ) (p +2) _ I
(Pl+ +Pv + V)* Pl- P*-'
, , ix v -p -l (P + 1 ) . . . ( v - 1) _ (,_ !) ]

When p t + . . . + Pr + v = 0 the above formula gives:


]2 7 ] EXPANSION IN TH E NEIGHBOURHOOD OP AN IRREGULAR SINGULARITY 481

and p remains arbitrary. Summarising all that has been said above we
obtain the following formulae for the determination of the coefficients a:

dpi = ----—-p when Pi + 1 # 0


" Pi + 1

(l) f 0 when + 1# 0
api { 1 when p, + 1 = 0
lPi- ■Pv — ° (P i + • • • + Pv + v ■

1
n " 'Pv P l+ ---P v + V
P + 1

Pl + + Pv + V Pl -Pu.
____ (/* + !)(/* + 2)
I (H+2> i
(Pi + ••• + P V+ '’Y Pl" Pv-'
(331)
, , (P+ 1) ••• O’— 1) _(»—1) 1
+ ~ aPi- -pv- l
( P l + - - - + P v + v) p J
[p = v — l, v — 2, , 1,0; p l + . . . + p v + vjL0) ,
n0‘) _ _(P—1>
aFl--pv — ap l - p l^ l

(fi = v,v I, • • 2 , 1 ; p t + . . . + p„ + v = 0).


To find an expression for the functions Lpi,,,Pv(b; *) in terms of M pi Pw(x)
certain new functions M pi p^(x) have to be introduced which can be determined
as single-valued functions with the aid of the following successive equations:

M*Pl (x ) = - M Pl (x),
(332)
Mp1...pv (x ) = — 2/ Mpi-'-Pp ^Pp+i' -Pv

In the above sum in the term corresponding to p = 0 the factor M pi p^ (x)


becomes devoid of meaning when p = 0 and must be replaced by unity.
We will now show that

L fr-P , ( b ; x ) = £ M*Pl...p (6) M p Pv (x), (333)


p-o
where, when p = 0, the first factor is replaced by unity and when p = v
the second factor must be replaced by unity.
When v = 1 this formula is self-explanatory since we have from the evalua­
tion of our functions:
LPl (b ; x) = M pi (x) - M Pl (b) = M pi (x) + M*Pl (6).
To prove the validity of the formula (333) for every v we shall prove th at if
it is true for v then it is true for v + 1.
482 LINEAR DIFFEREN TIA L EQUATIONS [127
On supposing th at formula (333) is valid for all Lpi px (x), when A < Vf
we have:

uPi■ ((6 ; x) = J LPl' ,.Pv(b; x) xK+x dx =


— j 2 M p i- - P a (&) M p P , ( X ')P v + t ^ r*
b P-°
But from the definition of M,Pi ■■Py (x):
X
M Py.+l---Pv (X) xPv+1 — MPfL+x - Pvy, ^ M rPp+f'Pv+l
n (b)
' ’
!
and consequently:

h i . . .pv+l <b > = Z M Pi ••-pM<*) \ M P.. ....-P...< * > - M ppyx -•-PV+1 ■


p-a 7 M -» V +l

or, bearing in mind the usual condition for border terms in the above sums
and also (332) we can write:
v+i
h , . . p v^ b • x ) — Z_ M*i...pa (b) Mpll+,...pv+^ x ) >
/J-0
i.e. formula (333) appears to be valid for L p Pp+l (6; x), also and we can
therefore say th at this formula will also hold in the general case. I t follows
directly from (332) th at M*i p^ (x) has the same form as M pi Pf{x) but that
coefficients are different

M *i-..Pv (x ) = xp^ - . - + p^
j? lo g P x . (334)
M-0
To simplify the construction of the relationships which permit the evaluation
of the coefficients ap^M
} Pp, we shall prove the validity of the following formulae
f” M Pi...Py-
M Pl (x) = — J xPl dx ; M*1...Pv(x) = - (x)dx. (335)
The constants of integration are, in this case, be so chosen that the
formulae (334) holds. This will, as before, determine the OpJ^.p when
Pi + . . . + Pv + v ^ 0. We shall consider below the choice of a*<°) _p , when
Pi + . . . + Pv+ v = 0. When v = 1 we have:

±;M *Pi (x ) = - ^ M pA x ) = - tP>


and, consequently:
x
M Pl(x) = — J xPldx.
Let us now suppose that the equation
12 7 ] EXPANSION IN TH E NEIGHBOURHOOD OF AN IRREGULAR SINGULARITY 483

holds for A< v —1. We will now show th at it is also valid when A= v. We have
from (332):

dx M *1'' py (X) ~ “ 2 0 • Pv (x) d i M Pi--Pp.(*) +

+ M ti...Py, ( * ) - g j M P ^ . . - P V ( * ) ]

or from (336) and the definition of M pi Pf(x)\

f a r M Pi-- Pv (x) = xPl 2 Mpt-'-Pp (x) 1-••?»(*) ~

- * 2 M p x . . . Pp. W M p ,p ( * ) xPv.
p-o *
But according to (332):

2 ^Pi-- Pu. ^Pii+f-Pv-, ~ 0


p -«
and
P2 M P>-Pa (*) M P ^ - P V (*) = - MPa-P„ (*) ■
/‘-l
and therefore
- g j M Pi...Pv (x ) = - x P l M Pr---Pv (* ) •

which we wanted to prove. Using the formulae (335) we can introduce relation­
ships for ap^ p analogous with those deduced above for a^j) p^. The proof will
be exactly the same and we shall only give the final result:
1
*<°> —
a Pi
- - when Pi + 1 # 0
P1 + 1
0 when p. + 1 / 0
°„*<i)
pi- f 1 when P. + 1 = 0
,*w
'p i - p
= 0
u (Pi + • • ■+ P r + v ^O )
—1 ______ L * ( " ) ____________ p + 1________ a (p + i) i
'P\--P„
Pi + ■■ ■+ P,iw+ v [ P i + - - . + P v + v Pi " Pv

i (P + 1) (P + 2) _*bn2) , ,
+ (Pi+ ••• + P„ + v)2 p * - p* + — +

v-A -l (P + 1) . . . (V - !)
1) „<—i) 1 (337)
+ ( - 1) Z ^ ~ [ aP*-Pv \
(Pi + ••• + P V + * ) '
( / ! ■ » - ! , v — 2, . 1, 0 ; Pi + . .. +Pi> + , , 7^0)>
aP*<p)
i - P v
aP*{t0-—1)
P v
P
(P = v, v — 1...........2, 1 ; p, + -•. + pv + v = 0).
484 LINE AH DIFFERENTIAL EQUATIONS [127

From these relationships all the a*, except <**pf ) .Pv>can be determined, when
p, + .. . + P„ + v = 0. As a result of the single-valued determination of
M pi..,pv(x) from the formulae (332), these latter coefficients can also be
expressed in a definite way in terms of the known coefficients a and a*. To
find these expressions we replace M pi ,,Pv(%) by M pi,.,Pv(x) hi formula (332)
using their expressions in the formulae (330) and (334). After substitution in
both sides of the equations we obtain polynomials in log x:

i “p l . p j o g 5* = - ”2 ( 2 aPl(S)V ° g SX) ( 2 aV r . f , 1° 8 'I l


s-o p-o Vs—0 ^ ) Vs-o/
Comparing terms which do not contain log x in this formula we obtain the
relationship
2 a T . . P ^ (p l , . . P v = 0 (338)
P -o

or, separating the first and last terms:

n(Q) 1 „*(o) -4- ft


a Pl--Pv + 2 a Pl- • P „ a Pp+i- -Pt, + a Pl---P„ — u - vd J J i
P -1

These latter relationships enable us to determine the a*^ p^ when


Pi + ••• + Py + r = 0-
Lastly, if we replace M pi p^(x) and M pi p^(x) in the expression (333)
by their expressions from (330) and (334), then we obtain a definite expression
for the coefficients of the series (329):

LPl...Pv (b; x) = £ bPi+.-.+Pp+P^Pp+i+.-.+Pv+’-P x


v p-o

X 2 aT..PPlog' b * 2 o « . log*' *.
A-o k- 0
Introducing these expressions into the formula (326) we obtain the final
expression for the solution:
Y (b ; x) =

= 1 + 2 2 T P1- PV 2 b^+---+pP+l‘ x’,P ^ +- +Pv+v->‘ X



-1 Pl-..p - - S P-0

From the above considerations we arrive a t to the following theorem.


T h e o r e m . The solution of the system (325), which becomes a unit matrix
when x = b, can be determined for every analytic continuation with respect to
x and for any matrix T p given in the form of the series (337); the coefficients a
are determined from the relationships (331) and when p l + . . . + p w+ r = 0,
°Pi Pv’ rema^n arbitrary. The coefficients a* are determined from the relation­
ships (337) and (338).
127] EXPANSION IN THE NEIGHBOURHOOD OF AN IRREGULAR SINGULARITY 486

Formula (340) determines the use of the method of successive approximations


in its application to the system (316).
By describing a circuit round x = 0 the solution Y(b;x) is multiplied on
the left by a constant matrix V. I t is not difficult to obtain a formula for the
any positive integral power Vm. If m is an integer, either positive or negative,
we can obtain Vm if we take the value of Y(b; x) at the point x = b, obtained
as a result of completing a circuit round the point x = 0, | m | times in the posi­
tive direction when m > 0, and in the negative direction when m < 0. After
completing this circuit the initial value of log x becomes log x + 2mm and
we have:

v m = i + 2 2 T p l - - - T PV h P l + - +p^ v I S ^ . p i o i f b x
* - l Pu - ,P V ---- - /J-OA-O

x A-o
*2 ap L - ■-p, (los 6 + 2mni)k- (341)

The value of log b in the inside sum is the same as log b in the outside sum
and the coefficients in the above power series are polynomials in log b. I t can
be shown, but we will not do so here, th at the coefficients of all terms con­
taining log 6 will cancel each other, so that instead of formula (341) we can
use the simpler formula

v m= i+2 V=I P,,...,ft,=-s


2 T*- -t p, • 6P,+- +'v fr 2 «#°U x 0

* "M aZ +l.....(342)
A-0

Let us now return to the fundamental expansion (340). I t can readily be


seen th at the series on the right-hand side can be formally obtained as a product
of two series, viz. it can bo represented as a product of two power series of
matrices as follows:

[i + 2 2 t p>- ■■TPvbP'+- p»+r 2 ap ^ p , log" b] *


L » -l Pl...pv — S p -0 J

x [i + 2 2 T P f - - TPV*P'+■■• 2 “ft -■Pv ,0g" X1 • (343)


i ' - I P l....... P„----s #*-• j

The first factor does not contain x and is a constant matrix. If we cancel
this factor, in other words, if we multiply the product (343) on the left by
a constant matrix, which is the reciprocal of the first factor, then we are left as
a result with the series:

1 + 2 2 T P t . . . T P v xP>+ - + pv + ” 2 4 t . p v '°e'x (344)


<’- 1 Pl..-..P„--S /"-0
486 LINEAR D IFFERENTIAL EQUATIONS [128

which should also be a solution of the system (325). The above considerations
are only formal in character but it can be proved that the series (344) does,
in fact, converge and give a solution of the system when the matrices T p are
close to zero. This solution (344) no longer depends on the choice of the point
b at which we approximated the solution to a unit matrix. We are not going
to investigate in greater detail the solution given by the series (344). A detailed
investigation of the system (325) can be found in the original works of I. A.
Lappo- Danilevski.

128. Expansions into uniformly convergent series. The series constructed


above were uniformly convergent in a closed domain without singularities.
In the neighbourhood of a regular singularity, after the isolation of this sin­
gularity, we obtained a Taylor’s series, which converged uniformly in the neigh­
bourhood of the singularity. We shall now describe the construction of series
which converge uniformly on the real axis in the neighbourhood of an irregular
singularity which we place at infinity. This leads us to a consideration of the
asymptotic representation. Let us consider a system of two equations of the
first order of the form:

(345)

where T k are constant matrices. When T 0 = 0, then x = oo is a regular sin­


gularity. Assuming that Y = Y 1S, where S is a constant matrix, we obtain a
similar system for Y lt but the coefficients will be T'k = S T k S ~ 1 and we can
choose S so th at the matrix T'k is given in canonical form. We shall assume
in what follows th at in the system (345) the matrix T„ is already in the canoni­
cal form. Let us consider the case when T„ has a purely diagonal form:
T 0 = [0 (, o„], and the real parts of a t and o2 are different. Without loss of
generality we can assume that
<S(o,) > H (o2) , (346)
where Ol(z) is the symbol of the real part of the complex number z.
Let us now introduce the elements hk of the matrix T:

tu = « / + 2 $ - r S *« = 2 $ - r {i * k) (347)
s-l X s-1 X-

and assume th at
T = P 0 + P, (348)
where P„ is the diagonal matrix:

(349)

The elements P ik of the matrix P are:


128] EXPANSIONS INTO UNIFORMLY CONVERGENT SERIES 487

We replace Y by a new unknown matrix Z according to the formula


X

y _elP*dX z = ai*+l22 logx-a.]% (351)


On substituting in (345) we obtain an equation for Z-.

(352)

or, introducing the parameter A, we have:

^ = z r , - r j + xzr. (353)

We shall try to find a solution of this equation in the form of the series

z = 2 Zmr . (354)
m-o
Substituting in (353) we obtain:

^ - = ZmPB- P 0Zm + Zm. lP (m = 1 , 2 , 3 , . . . ) , (355)

or, for the elements of the matrix z m-.

= e V + 'S l°sx-ak zp _ e^+ 'H logx~a<z\f> + 2 Psk, (356)


dx s-i
and it can be assumed that Z a = 1. The above equation can easily be solved and,
as a result, a formula for the successive evaluation of the zP will be obtained:

z/fc0 = e- f “ f er“ 2 Psk , (357)


J S-l
where
rlk = (“/ —Vk)x + (-Hi* — Pm! ) log x = aik X + pik log x . (358)
In formula (357) the interval of integration is (x0, x) when i > k, and ( oo, x)
when i < k, where x 0 is a sufficiently large real value of x, and x > x 0. We
assume, in any case, th at x 0 > 1. For the elements zik of the matrix Z, which
must satisfy the equation (352), we obtain the series:

z/A = 2 2/t'* (i¥=k),


m-1
(359)
*« = i + 2 W -
m-l
If we can prove the uniform convergence of these series in the infinite interval
x0 < x < oo, then it follows from the formulae (357), that the series composed
of the derivatives:
(i, A: — 1, 2),
488 LIN EA R D IFFER EN TIA !, EQUATIONS [128

converge uniformly in every finite part of the above interval and that the matrix
Z satisfies the equation (352) while the matrix Y, given by the formula (351),
satisfies the equation (345).
We shall now prove the uniform convergence of the series (350). I t fol­
lows from (350) that:

\ V i k \ < ~ (*'#&); (36°)

where a is a positive constant. Also, using the rule of l’Hopital it is easy


to obtain the inequalities
e-"'* f — dx <-^1- ; f dx < , (361)
J X X J X2 X

where r'jk is the real part of rjk and o, is a positive constant. In these formulae
and in the formulae below, the integrals should be understood in the same
sense as above. Notice th at as x 0 increases the value of the constant a, re­
mains unaltered.
We have from (355) and the fact th at Z 0 = 1:

2/i* = e-r« J eru p ikdx (i # k) ; zty = § ptl dx , (362)


and from (360) and (361) we obtain:

(ifc#Z); | 4 J > | < - S - 063)

Also from (355), (363) and the condition x > x 0 > 1:

Mi>l<e"r:tJer'‘ 2 9 -d x< ^ -e-^ Jera|-d x< ^ ---5 -,

I^l< j2^-dx< 2 a .A ,

| z\$ | < 2a, e~'a J er“ 2 dx < •J - ,

Iz(?l < J J M?51Ips* I dx < | ^ + 2a, dx ,

whence

' x„ x
Further inequalities are of the form:

| 4 m)| <
(2a,)2m_1 a
z\}m) | < (2a- xm-i

(2a,)2 2FTl
1 4 r +i) I
a \ (364)
x
128] EXPANSIONS INTO UNIFORMLY CONVERGENT SERIES 489

They can readily be proved, as above, by induction from 2m to (2m -f- 1) and
from (2m + 1) to (2m + 2).
It follows from these inequalities th a t if we take x B> (2a!)2 then the series

i I4m)i
m- 0
will converge uniformly in the infinite interval x 0< x < oo, when either i ^ k
or i = k.
Hence, bearing in mind the formula (351), we obtain the following solution
for the system (345):
<i)
Vik = ° a‘X x “ z ik (i> & = 2 ) , (36 5 )

where, as always, i is the number of the solution and k denotes the number of
the function. I t follows from the formula (359) and the inequalities (364) that

z ik = 0 ( ^ - ) ( i # * ) i z ii = 1 + 0 ( ^ r ) > (3 6 6 )

and, from (346), this shows th at the solutions (365) are linearly independent.
When substituting the expansions (360) in the integral (357) we obtain the
following integral:
X X

9lo g JC f eax+0 lo g x ds an(j eax+P lo g * f e«-/9 lo g * (367)


J Xn J xn

where 02(a) > 0 and n > 1.


Integrating by parts and assuming th at Ios x = x^u we can write the
asymptotic expansion of these integrals in powers of \jx.
Using the expansions (350), the expansions of the integrals (367) into asymp­
totic series and calculating the upper bounds of the firstand last terms of the ex­
pansions by evaluating the inequalities (364) more accurately, we obtain the
asymptotic expansion of zifc in powers of l/x; this, according to (365), will give
us the asymptotic expansions of y,*. Let us separate the first terms in the
asymptotio expansions of z,^:
We have from (362):

, ,(i) ,(s)
gill __ e- au*-0tt,oex | eaux+ft*iogx I 1'* i l,k
■J- (4+ 4-+ ^ )* “ <*>■

2$ = e—aux+0i*logx I en«*-0u log * + ijlL + i!* ) dx (i > k) ,


J \ X X2 X2 J
490 LINEAR DIFFERENTIAL EQUATIONS [128

where eik and e*/ 0 as x -*• co. Integrating by parts we obtain from this:

*$ = — + — + -(4t-*0 as z — ; i, k = \, 2). (368)


X X* X*

Substituting these expressions in the formulae (357) when m = 2 we have:

2<2) _ 0 - “U X-Pik log * J 0 a rt X + P tl log X


i ( ^ - + a |_ + M x
s - x I. ® 3J X2 )

x ( « + i ? + i»K
[ X X2 X2 )

When i ^ k this gives the asymptotic representations

(a) , £ik " n (as x->-oo) (i # k). (369)


Z‘K = + ~ ^ i ~ 0
Also

from which it follows that

.(a) bil’ , b\f , £// (370)


z ii = ^ r + -zi~ + -zr-

From (369) and (370) we have the inequalities:

| |< — ; I I< ~ (b is a constant). (371)

From this, (360) and the condition x > x B> 1 we obtain:

I I > 2o6 e-r“ J*er“ dz ,

I I < 2 a 6 J - ^ r da:-

Ineqalities, analogous with (361), can readily be obtained:

e~ 'a I er<l-4 -d z : f —1- dx < (6. being a constant) ,


I x2 x2 J x3 x2

from which we obtain the inequalities:

I2 $ | <o6(2b,) - L ( i , k = 1,2).
X*
128] EXPANSIONS INTO UNIFORMLY CONVERGENT SERIES 491

Substituting this in the formula (367) when m = 4 and bearing in mind that
x > *0 > 1 we have:
I (4) | abb,22 _/■;
z» I < — - — e ®J|V « -x2
Lda:<ab-^^l- — >
a;2

12 $^ | < abb,22J dr < ub (2b,)2 • — •

Continuing in this way we obtain the general inequalities:

(2b,)2m-a 1 . | _<2m) | „ (26,)2m“ 2 1


“ iS=i---- > l*« 1< ab -

W ? r i) \< a b ^ . r - 1 1 . I .(am+i) | ^ , (2b|)tm~ 1 1


2;m—, X2 ' 1 jrjj1-!2**

from which it follows that:

I A m) | + I4 T l> i < °b (1 + 2b,) A . ,


(371,)
I z i r I + I 4 r +l) I < ab (*0 + 2b,) J j- .

Using the series (359) and the formulae (368), (370) and (371) we obtain:

zlk; ■= # + M ; Z/, = 1 + ^ ± ^ _ + (372)

where Tjik and t)u -<■ 0 as x — co. Substituting in (365) we obtain the asymp­
totic expansion for yjk. In the same way as before it is possible to separate
the following terms in the asymptotic expansion zlk. The above method with­
out modifications can also be applied to a system of n equations provided
that the real parts of the characteristic zeros of the matrix T 0 are different.
Let us now assume that the numbers o, and <x2, which form part of the
diagonal matrix T 0 = fo,, au], have the same real part a. Replacing 7 by a
new unknown matrix T, = eax Y we obtain an equation for Y in the form
(345) in which T 0 is the diagonal matrix [a,i, a3i] with purely imaginary terms
on the main diagonal. We will suppose that the equation (345) already has this
property. If, at the same time, the matrix T, is equal to zero, then without
making any modifications, we can use the above method to obtain a uniformly
convergent series and the asymptotic representation. When T 0 = (a, a], then
the substitution F, = e~ax y gives a system with a regular singularity at
infinity.
Let us apply the results we have obtained to a linear differential equation
of the form:

+ ^ + £ + + + (373)
492 LINEAR DIFFERENTIAL EQUATIONS [128

The usual separation of the functions yl = y and yz = y ' gives a system in


which

and the characteristic equation of this matrix has the form:


A* + a0A+ &0 = 0. (374)
When the real parts of the zeros of this equation are different we can use
the above method for the construction of the series. This method of successive
approximations was first worked out by N. P. Erugin in his paper Transform­
able Systems (1946).
In the works of V.V. Khoroshilov {Proc. Acad. Sci. SSSR., 1949) it was
worked out for this case when the real parts of the zeros of the characteristic
equation for the matrix T 0 were different.
C H A P T E R VI

SPECIAL FUNCTIONS

§ 1. Spherical functions
129. The determination of spherical functions. In this chapter we
shall study certain special classes of functions which are met with in
the solution of equations in mathematical physics. All these
functions are usually determined as solutions of certain linear equations
with variable coefficients. For example, in problems connected with
the vibration of a cord we met trigonometric functions and in problems
connected with the vibration of a round membrane we met the Bessel
functions.
We shall begin with the study of the so called spherical functions
which are closely connected with the Laplace equation. We have
already encountered this equation. In the Cartesian system of coor­
dinates it has the form:
8H7 dW_ d2U
A17 =
dx2 + 3y2 dz2 = 0. (1)

We shall seek a solution of this equation in the form of homogeneous


polynomials in the variables x, y and z.
Let us consider the simpler cases. The only homogeneous polynomial
of zero degree is an arbitrary constant a which evidently satisfies
the equation (1). The general form of homogeneous polynomials of
the first degree will be:
U l = ax + hy + cz.
This polynomial also satisfies the equation (1) for every choice of
the coefficients a, b and c. In other words, we have here three linearly-
independent solutions of the equation (1), viz. x, y and z and their
linear combination with arbitrary constant coefficients gives a
general solution of the equation (1) in the form of a homogeneous
polynomial of the first degree. Consider a homogeneous polynomial
of the second degree
U2 = a,x2 + by2 -+- cz2 + dxy + eyz + fz x .
493
494 SPECIAL FUNCTIONS [129

Substituting in the equation (1) we obtain a relationship for the


coefficients, viz. a + b + c = 0. We can, for example, put
c = —a — b and, consequently, the general form of homogeneous
polynomials of the second degree, which satisfy the equation (1),
will be
U 2 = a (x2 — z-) + b (y 2 — z2) + dxy + eyz + fzx ■
Here we have five linearly-independent solutions of the equation:
viz. x2 — z2, y2 — z2, xy, yz and zx and the linear combination of
these solutions with arbitrary constant coefficients gives the general
solution of the equation in the form of a homogeneous polynomial
of the second degree.
Let us consider a homogeneous polynomial of the third degree:
U 3 = ax? -j- by3 + cz3 + dx-y + ex2z + fy 2x + gy2z + hz2x + kz2y + Ixyz.
Substituting in equation (1) we have:
6 (ax + by + cz) + 2dy + 2ez + 2fx + 2gx + 2/ia; + 2ky = 0.
Equating to zero the coefficientsof x, y, z we obtain three equations
which connect these coefficients:
3a + /-|-A = 0 a = — jj~(f h) >

3&+ d + &= 0 or 6 = — i-(d + &),

3c + e + gr = 0 c = — — (e + g ),
so that the general solution of the equation (1) in the form of a poly­
nomial of the third degree will be

U 3 = d \x2y ~ — y3) + e [x2z — — z3) + / \y2x — — z3) +

+ g \y 2z — — z3) + h [z2x — — re3) -f k [ z 2y — —y3] + Ixyz.

In this case we have seven linearly-independent solutions for the


equation.
We will now show that, in general, there are (2n + 1) linearly-
independent homogeneous polynomials of the n-th degree which satisfy
the equation (1). We shall now count the number of coefficients in
a homogeneous polynomial and the number of equations which
they must satisfy. A homogeneous polynomial of the nth. degree
in two variables
a0xn + a ^ " -1y + . . . + a^yn
129] TH E DETERMINATION OP SPHERICAL FUNCTIONS 495

baa (» + 1) coefficients. A homogeneous polynomial of the nth


degree in three variables can be written in the form:
« 03" + <Pl (x >y) z " '1 + ■• • + <Pn- 1 (*. y) Z + <Pn (*. y) ’ (2)
where <pk(x, y) are homogeneous polynomials of degree k. Consequently
the total number of coefficients in the homogeneous polynomial (2)
will be:
1 + 2 + . . . + n + (n + l ) = (p + y + 2)..

Substituting the polynomial (2) in the left-hand side of the equa­


tion (1) we obtain a homogeneous polynomial of degree (n — 2)
which has, in all, (n— 1) nj2 terms. In this way (n + 1) (n + 2)/2
coefficients of the polynomial (2) will be connected by (n — 1) nj2
homogeneous equations. If these equations are independent then the
number of arbitrary coefficients will be:
(n+lHn + 2) _ frzL*)H = 2 n + l t

which is what we had to prove. It is still not quite clear whether the
above equations will, in fact, be independent. We shall therefore give
another complete proof of the above proposition. We can write the
polynomial (2) as follows:
Un = 2 apqr xpy«zr,
p+q+r=n
, _ 1 dp + q + rUn
where (3)
p<>r pi q\ rl dxp 0i/?0zr
Equation (1) can be rewritten in the form
8*17 8*17 8*17
82 * 8 x* dy2
By using this equation we can eliminate in the expressions (3) any
differentiation with respect to the variable z higher than the first
order; we can therefore write:
&>U _ _ 8* / 8*17 8 * 1 7 -j _
dx dy 02 4 dx dy dz2 ( dx2 dy2 J
— 8< f , 8*17 84 f 8*17 8*17 'j _
dx2dy I. dx2 ' dy2 j ' dx dy3 I. dy2 ' 8a:* J
_ 8°U d*U , dBU
dxady 8i3 dy3 dx dy3
In this way only those coefficients apqr remain arbitrary which
either cannot be differentiated with respect to z or which can only be
496 SPECIAL FUNCTIONS [130
differentiated once. These coefficients are as follows: apqo (p + q = %)
or apql (p q — n — 1) and their total number is equal to (2n -f l)i
which is what we had to prove.

130. The definite expression for spherical functions. We shall now


establish the definite expression for homogeneous polynomials men­
tioned in the previous section. Introducing spherical coordinates
we have
x = r sin 9 cos cp; y = r sin 6 sin cp ; z = r cos 9 . (4 )
In this case a harmonic homogeneous polynomial of the nth degree
can be represented in the form
Un (x, y, z) = rn Y n {9, cp) . (5)
This polynomial, which is a solution of the equation (1), is usually
known as a volume spherical function and the factor Y„(9, cp), which
is a polynomial in cos 6, sin 6, cos cp and sin <p, is known as a sur­
face spherical function or simply as a spherical function of order n.
It is our aim to find (2n + 1) linearly-independent spherical
functions.
We begin by observing one simple fact about the solution of the
equation (1). Let us write the following integral which depends on
the parameters x, y and z:
7t
U (x, y, z) = J" / (z -f- ix cos t + iy sin t, t) d<, (6)
—71
where we assume that above integral can be differentiated under
the integral sign with respect to x, y and z. After differentiation we
can readily see that the function U(x, y, z ) satisfies the equation (1)
for every choice of the function f(r, t) as long as the above differentia­
tion is permissible. In fact:
71
AU (x, y, z) = J (1 — cos21 — sin21) f" (z + ix cos t + iy sin t, t) dt ,
—71
where by f"(x, t) we denote the second derivative of f(x, t) with respect
to x. Notice that this argument is a complex quantity. Subsequently
by using formula (6), (2n + 1) homogeneous polynomials of the
wth degree, which satisfy the equation (1), can readily be
constructed.
130] THE DEFINITE EXPRESSION FOR SPHERICAL FUNCTIONS 497

They can be written as follows:


n
J (z + ix cos t -f- iy sin t)n cos mt di (to = 0, 1, 2, . . . , n ) , (7)
—a
n
J (z + ix cos t + iy sin t)n sin mt dt (to = 1, 2, . . . . to) . (8)
—n
Introducing spherical coordinates into the integral (7) we obtain the
following expression for the spherical functions:
n
J [cos 0 -\- i sin 0 cos (t — g?)]n cos mt d< =

71—(p
= J (cos 0 + i sin 0 cosy)" cosm (<p + ip) dy.
—71—tp

Bearing in mind the fact that the integrand on the right-hand side has
a period of with respect to y>we can take any interval of integra­
tion, 2ji in length [II, 142]. Hence the above integral can be rewritten
as follows:
n
J (cos0 + isin0cosv>)ncosTO(9> +
—71

Expanding cos m{<p + ip) and bearing in mind that the function
sin to ip is odd we can rewrite these spherical functions in the form:
*
J
cosm<p (cos d + i sin 0 cos y>)n cos mip dy (to = 0, 1, 2, . . . , n ). (9)
—71

Similarly integral (8) gives the following n spherical functions:


71
sin mg? J (cos 0 + i sin 0 cos ip)n cos mip dy (m = 1, 2, . . . , n ) . (10)
—71

The linear independence of all the (2n + 1) functions (9) and (10)
is directly due to the fact that the dependence of these functions on
<p is due to the factors cos m<p and sin m<p\ these functions cannot
be linearly dependent for they are orthogonal in the interval {— n, + ti)
[II, 142]. We have thus constructed all the (2n + 1) spherical func­
tions of order n. The coefficients of cos m<p and sin m<p in the expres­
sions (9) and (10) are functions of 0. We can express them in terms
of the Legendre polynomials.
498 SPECIAL FUNCTIONS [130

We have the following expressions for the Legendre polynomial


[ 102 ]:

<»>
Let us also introduce the functions Pn,m(x), which are expressed in
terms of the Legendre polynomials as follows:
m
p _ ii rx2 \™ dmPn (x) (1 —x 2)2 dn+nl r, „
) > — — ---------- n l -2„ da;„+m l(*
i\nl
! ) J- (12)

We can now introduce different expressions for Pn(x) and Pn,m{x).


According to Cauchy’s formula we can write

where £7 is a closed contour which is described in the counter-clockwise


direction and which contains the point z = x. From this and from
(11) we have:

’ 2n+1m J ( z - x ) n+1 ' 1J


c
Take as the contour C a circle, centre z = x and radius | x2 — 1 \]l2
(we assume that x # ± l ) . The variable z is given by the formula:
i
z = x -j- (x2 — l)2 e‘v.,
where the choice of the value of (x2 — 1)1/2 is immaterial and where
it can be assumed that ip varies from (—n) to (+ ji). Replacing the
variables in the integral (13) we have:
n
[x - 1 + (s* - l)2 elv,1 [x + I + (a1— l)2 eiv]
Pn ( ^ = ~2n J
dip.
2 {xz — l)2 e‘‘v

Performing elementary calculations and remembering that the


integrand is an even function we have:
* i
Pn = J [* + (v2 - !)2 cos v>]n d v =

x + (x2 — l)acos y]n dip. (14)


130] TH E DEFINITE EXPRESSION FOR SPHERICAL FUNCTIONS 499

Let us carry out analogous calculations for Pn>m(x). In place of


(13) we have:
m
( 1 — J 2)« (n + l ) ( n + 2 ) . . . ( n + m) f (g» — 1)"
dz,
P n,m (*)
2n+1m
' '
J (z — x)n+m+1
c

and performing the above change of variables we obtain:

P n.m (*) = — ) (w + 22J + j[x +


—n
( l)2cosy]ne-fm*’dy>,
x2—

or, bearing in mind that sin my? is odd,


Pn,m (z) =

_ _ i^ n + (n • ~(n + m) J [a; -|- (a;2 — l) 2cos rp\n cosm yd^. (15)

If we put x = cos 6 in either the integral (14) or (15) then the


integrals which appear in the formulae (9) and (10) will be obtained.
Bearing in mind that the constant factor of a harmonic polynomial
or of a spherical function is irrelevant we arrive at the following
conclusion: (2n + 1) spherical functions of the n-th order can be written
in the form :
Pn (cos0); P n>m(cos0) cos m<p ; P n>m(cos 6) sinmip (16)
(m = 1, 2........ n ) ,
where Pn(x) are Legendre polynomials as defined by the formula (11)
andPn,m(x) is given by the formula (12). Notice that the factor ( l —£2)m/2
becomes sinm 6 when x = cos 9. Multiplying the solutions (16) by
arbitrary constants and adding we obtain the general form of a
spherical function of the nth order:
n
Yn(0>?)=aoP n{c° Bd) + v (®mcos m<p + bm sin m<p) P n>m(cos 6). (17)
m=1
In place of trigonometric functions we could, by constructing linear
combinations of the solutions (16), use exponential functions, so
that instead of taking the spherical functions (16) of the nth order
we would use the following system of spherical functions of the
nth order:
P„(cos0), P nm (cos0) e,m?’, f >nim(cos0)e-fm?’ ( m = l , 2........ n). (18)
600 SPEOIAIj FUNCTIONS [131

It follows from this construction that the general form of homogeneous


polynomials of the nth order of the variables (x , y, z) which satisfy the
Laplace equation, will be rn Y„(0, <p), where Y n(6, <p) is given by the
formula (17).

131. The orthogonal properties. We shall now prove the orthogonal


properties of spherical functions (16) on a unit sphere and evaluate
the integral of the square of these functions on the unit sphere.
We begin by evaluating the integrals:

Im= S
-1
From the definition of these functions we have:

-l -l
and when m = 0 we obtain the integral of the square of the Legendre
polynomial:
l
7o = I

We have shown earlier [102] that

(19)
-l
At the end of this section we shall again give the proof of this
formula but at present we shall evaluate the integral I m by using
formula (19).
Integrating by parts we can write:
dmP„(x) dm- 1P„(x) *-+1
/ m = (i - x * r
dxm d r171-1 x— l

or

(20 )
dxm 1 dx
-i
131] THE ORTHOGONAL PROPERTIES 501

But the function


_ dm~ 1 P„ (x) 1 cy + m - l (j .2 — i)rt
~ " d a :"1 -1 2" n l d a :n + m “ 1

as can readily be assertained by using the equation (84) from [102],


satisfies the equation

(1 _ ^ # 1 - 2m, +
dxm+1 dxm
dw- 1P „ (a!)
+ (n + m) (n — m -+- 1) = 0.
da'7'- 1
Multiplying by (1 — x2)m 1, we can rewrite it in the form:

_ ^2)m^ £ nm(- ] = - (n+ m ) (n - m + 11 ( l - x 2)"-1— **<x) ■

Substituting in formula (20) we have

r dm_1 P„ (x) d m_1 P n (x) j


I m = (n + m ) ( n - m + 1 ) [ ( 1 - x 2) " 1- 1 — - - - - - - - - - - - - — dx
-l
or
Im= {n + m )(n — m + 1) I m_ j.
Using this as a reduction formula, we obtain
I m = in + m) (w — tn + 1) {n + m — 1) (n — m + 2) I m_2 = . . . =
= (n + m) (n — m + 1) (n + m — 1) (n — m + 2) . . . (n -f 1) n l 0 =

= (n + m ) (7 i + m — 1 ) (n + m — 2 ) . . . (re — m + 1 ) 7 0 = I 0-

This, with (19), gives the following final expression for the integ­
rals of the squares of the functions Pnm(x):

I [p <*)]’ dx = 2^TT T » ~= 3 t • <21>


-1
This result makes it possible to evaluate the integral of the
square of a spherical function. The spherical functions Y n(0, <p)
can be regarded as being given on the surface of a sphere of unit
radius; 0 and <p are the usual geographical coordinates of points on
the surface: <p = const, are the meridians and 6 = const, the parallels.
602 SPECIAL FUNCTIONS [131

With this choice of coordinates an element of surface are, as is well


known, is expressed by the following formula [II, 59]:
d<r = sin 0 d0 d9 . (22)
We shall prove, first of all, that two different spherical functions
Y p(9, <p) and Y q(Q, q>) of different orders, i.e. when p ^ q, will be
orthogonal on the surface s of a unit sphere, i.e.
J J Y p (0, <p) Y v (0, (p)da = 0. (23)
5

Let v be the volume of this sphere and s be its surface. Applying


Green’s formula [II, 193] to the harmonic functions:
Up = r”Y p (0, <p) and Uq = Y q (0, <p) (24)
we obtain

S V

where A Up = A Uq = 0.
In this case, differentiation along the normal coincides with dif­
ferentiation along the radius r, so that the last formula and (24) give:
J I [qYp (0, 9 ) Y q (0, 9) - p Y q (0, 9) Yp (0, ?)] da = 0 ,
S

from which formula (23) follows directly.


We shall show that the spherical functions (16) whioh correspond
to one and the same value of n, will also be orthogonal. In fact,
integration over a unit sphere involves, among other things, integration
with respect to 9 in the interval (0, 2 ). But the functions (16) contain
71

the following factors which depend on 9 :


1, cos 9 , sin 9 , cos 2ip, sin 297, . . . . cos 7197 , sin 119,
and the product of any two of these factors, when integrated over
the interval (0, 2n), is zero [II, 142], It can be shown similarly that
the functions (18) also form an orthogonal system.
Lastly we can evaluate the integral of the square of each of the
constructed functions. Let us consider the spherical function Pn(cos 0)
which does not depend on 9 , and construct the integral of its square
on the surface of a unit sphere
n 2n
J J P% (cos 0) sin 0 d0 dg?.
0 0
131] TH E OBTHOGONAL PROPERTIES 603

On introducing the new variable of integration x = cos 0 and recall­


ing formula (19) we have:
n 2n 1
4n
J J P% (cos 0) sin 0 d0 dip = 2n J (x) dx 2n + 1
0 0 -1

Similarly for other functions:


n 2n
I I [^n.m (cos 0)]2sin2uvp sin 0 d0 dy = J [^n>m(*)]2dx.
o o

The latter formula and (21) finally give:

(']'[/>„ (cos 6 ) ] « d « = ^ - r

2tt (71 4- m) I
J J > - m(cos 0) cos mcpY d<r 271+1 (n — m) I ’ (25)
5
2 71 (n + m ) !
J / K m(cos 0) sin 7/195]2da 2t»+1 (71 — m)l
5

We shall use these formulae in future in connection with the


expansion of an arbitrary function which is given on the surface of
a sphere, in terms of spherical functions.
We shall now prove formula (19). Using the definition (11) of the
Legendre polynomials we can write:

/ - 1 f d "(x 2 - l ) n dn ( x * - l ) n ^
0 22n(n!)J J dx" dx"
Integrating by parts we have:
J __ I r d"- 1 ^ - 1)" dn (x* — l) n ~|*°+1
0_ 2“ (»l)* [ d x ""1 ’ dx" J* = -i ~

_ 1 f dn+1 (x» - l)n _ d"-1 (x* - l)n da.


2an (nl)1 J dxn+1 dx"-1

The polynomial (x2 — l)n has the zeros x = ± 1 each of order


n. Its derivative of order (n — 1) has the same zeros which, however,
are only of order one [I, 186], and, consequently, the first term on the
604 SPECIAL PUKOTIONB [132

right-hand side in the above equation is equal to zero. Continuing to


integrate by parts we obtain:
1
( - i)n r d2" (s» — i)n
h •\ x 2 — l ) n d x .
22n (n!)1 dx,n
But
d2n ( x 2 — l)n djn
(x2n + ■■•) = (271.)!,
dx2n dx2n
therefore:
T _ / i \n (n + l ) ( n + 2) . . . 2 n
i0~ I L> ... n ^(x 2 — l)ndx.

Introducing the new variable of integration cp according to the


formula x = cos <p we obtain:

J = (n + l)(n + 2) . . . 2» f gin2n+i d
0 nl22n J
0
T T
71
= (2t + 1)(" + 2) . . . 2n . 2 f gjn2n+l fp d^j
n \ 22n J0 Y Y

and, using the formula (28) from [I, 100] we obtain the formula (19).

132. The Legendre polynomials. We shall now study the Legendre


polynomials in greater detail. Notice, first of all, that if we use
the definition (11) and apply the Leibniz formula for a derivative
of the nth order to the product (x2 — l)n = ( x + l)n ( x — l)n we obtain:

dn (x — l)n n d (x + l)ndn~1( a ;- l)n


+ ••• +
dx" 1 dx dxn 1

d" (x + l)n
(x -1 )"
dx"
Bearing in mind that
dn (x — l)n d* (x — l)n
n! and = 0 when k <n,
dx" dx* X=1

we obtain directly from the previous formula:

P n ( 1) = 1- ( 26 )
132] TH E LEGENDBE POr.YNOMTAT.fi 505

We shall now use a special method viz. the method of the domina­
ting function, in the study of further properties of the Legendre
polynomials. We shall also use this method in future in the study
of other special functions.
Place at the North pole N of a unit sphere a positive charge ( + 1)
and let M be a variable point, the spherical coordinates of which
are r, 6 and <p. The Coulomb field created by this charge will have
the following potential at the point M :

d ^1 —2rcos0 + rJ ’

where d is the distance from the charge to the variable point M .


The function (27) will be a regular function of the variable r at
the point r = 0 and we can expand it in positive integral powers of r :

± = a 0 (d) + a 1 ( d ) r + a 2 ( d ) r * + . . . , (28)

where the coefficients in this expansion are polynomials in cos 0.


We could calculate these coefficients exactly by applying to the
function
-i- = [l + (r2— 2r cos 0)] 2 .

Newton’s binomial expansion and by subsequently collecting together


terms with equal powers of r . However we shall approach the
problem somewhat differently.
The function (27) in the usual system of coordinates can be expressed
as follows
j = [i + (*! + y2 + 28- 2 j ) p . (29)

We can obtain the series (28) if we apply Newton’s binomial ex­


pansion to the function (29) and subsequently collect together, in
the infinite series so obtained, all terms of equal value with respect
to x , y and z , i.e. all terms of the series (28) which are homogeneous
polynomials in x , y and z . As we know, the function 1j d is a solution
of the Laplace equation [II, 119] and consequently, the same can be
said about the individual terms of the series (28) i.e. the terms of this
series must be volume spherical functions. These functions do not
depend on the angle <p and, consequently, each term of this series
506 SPEO U Ji FUNCTIONS [132

must appear in the form of a product c„ P „ (cos 6 ) , where cn is a


constant which has to be found. We thus have:

■A 9 1
Y1 — 2r cos 6 + r2
= c 0 + ciP i (cos e ) r + c2p 2 (cos 0 ) r 2 + ...

If we take 9 = 0 we obtain, since ,Pn(l) = 1:

T ~ T = C0 + Cj T + C2T 2 + - . . ,

from which it follows that c n = 1 for all n ; we thus obtain the follow­
ing final expansion of our elementary potential in powers of r:

3 = = = = = = = - = 1 + ^ 1 (cos 6) t + P 2 (cos d ) r 2 + ... (30)


K1 — 2r cos 8 + r*
On replacing cos 0 by x and r by z we can write:
1
= 2 P n (x)z» (31)
/ l — 2xz + z2 n=0
This formula can be used as a definition for the Legendre poly­
nomials viz.: we can say that t h e L e g e n d r e p o l y n o m i a l P n( x ) i s th e
co efficien t o f z n i n th e e x p a n s i o n o f the fu n c tio n

1
(32)
— 2x5 + z2
i n p o s itiv e in teg ra l, p o w e r s o f z. In other words, the f u n c tio n (32) i s th e
d o m in a tin g fu n c tio n o f the L e g e n d r e p o ly n o m ia ls .
We shall now determine the radius of convergence of the power
series (31). Values of z for which the expression under the radical
vanishes will be the singularities of the function (32). Solving the
corresponding quadratic equation we obtain the following zeros:
Z= x ± Yx2 — 1 = x ± y i - x 2i. (33)
Since x = cos 6 we can assume that x is real and that it lies in the
interval —1 < * < + 1. In this case the zeros (33) will be conjugate
complex zeros and the square of the modulus of each zero will be:

a 2 + ( F 1 - x 2) 2 = 1.

When x = ± 1 the zeros (33) coincide and are both equal to ±1.
Thus when —1 < x < + 1 the singularities of the function (32) will
lie at a unit distance from the origin and, consequently, the series
(31) will converge when | z | < 1. Thus the expansion (30) will be
132] THE LEGENDRE POLYNOMIALS 507

valid when r < 1, i.e. for all points inside a unit sphere. For points
outside a unit sphere we have another expansion. In fact, when r > 1,
the function (27) can be rewritten as follows:
_______1_________ _ _L__________ 1__________
Vl - 2r cose + r* “ r 1 _ 2 1 cos e +

In this case 1/r < 1, so we can apply the former expansion and
we finally obtain the following formula for the potential (27) outside
a unit sphere:
__ 1 . = y p » (coa (3 4 1
][\ — 2 rc o s 0 -fr2 ” 0 rn+1

No term of this sum has singularities outside the sphere and each
term vanishes at infinity.
Until now we have considered a sphere of unit radius. For a sphere
of any radius R we have, by taking R 2 or r 2 outside the radical:

, = y p n (cos 0) — — (r< R ), (35)


YR2 - 2r R cos 0 + r2 " Rn+1
I « pn
- = y p n (cos 0) ——— (r> R ). (36)
YR2 — 2r R cos 0 + r2 r

From formula (31) the fundamental properties of the Legendre


polynomials can easily be deduced. Differentiating this formula with
respect to z and multiplying subsequently by (1 — 2x z -f- z 2) we
obtain
X—2
= (1 — 2x z + z2) j y n P n (x) zn_1
^1 — 2lZ + Z2 n=0
or

(* - 2) 2 Fn (*) 2" = (1 - 2xz + Z2) 2 nPn (*) ^ •


n=0 n=l

Comparing the coefficients of like powers of z we obtain a relation­


ship for successive Legendre polynomials:
(» + 1) -Pn+i (*) — (2» + 1) x P n (*) + nP n_i (a;) = 0 (37)
(n — 1, 2, 3, . . . )
P 1 (x) — x P 0 (x) = 0.
508 SPECIAL FUNCTIONS [132
Similarly, differentiating formula (31) with respect to x and multiply,
ing it subsequently by (1 — 2 x z + z2) we have;

P n ( x ) = dP™
(x) j dPn_, (x )
2x
dPn(a:) (38)
da: 1 da: dx
or, substituting P n+1( x ) from (37):
dP„(a:) dP„_, (x ) = n P n (x).
dx dx (39)

Eliminating x d P n( x ) j d x from (38) and (39) we obtain:


dPn+i (x )
da: ~ = ( 2" + 1)f n(a?)- (40)

This formula remains valid when n = 0 if we assume thatP_j(x) = 0.


On supposing in formula (40) that the symbol n is equal to 0 , 1, . . . , n
in succession and adding we obtain the new relationship:

P 0 (x) + 3 P 1 (*) + . . . + (2n + 1) P n (*) = dFX ( ~ + <41>

Let us write formula (40) replacing n by n — 2Jc + 1:


(g)____ dPn-zk (g) (2n — Ik + 3) P n- 2k+1 (*).
dx dx
Adding with respect to k from k = 1 to k — N, where N = (l/ 2)ra when
nis even, and N = ( 1/ 2) ( n -f- 1) when n is odd, we obtain the formula:

= J ( 2 7 l - 4 i + 3 ) P „ _ 2fc+1(x). (42)
fc= i

It follows from formula (11) that P n(x ) contains only even powers
of x when n is even, and only odd powers of x when n is odd. Similarly,
it also follows from this formula that:

P zn ( 0 ) = ( - i ) n - '4 r 4L; - 2 — : p *n+i ( ° ) = 0

Pn ( - ! ) = ( - ! ) " • (43)
Applying Newton’s binomial expansion we can write:
1 1 1
)^1—2rcos0 + r* / l —e'®r —e—
*®r
-V" 1 •3. ■.(2m —1) im9 m
~ \U=o
2 2- 4. . . 2 n 0 T 2
m=0
2 - 4 . ..2m 6 T
133] THE EXPANSION IN TEEMS OF 8PHEBICAL FUNCTIONS 609

where, when n = 0 and m = 0 , the terms of the series must be


assu m edto be equal to unity. Multiplying the series and comparing
the coefficients of like powers of r we obtain the following expression
for the Legendre polynomials:
P n (cos 0) = a 0a n cos n d + a 1o„_i cos ( n — 2 ) 0 + . . . -f-
+ a„a0 cos 7i 0 , (44)
where all the coefficients ak are positive and are given by the formulae

a0 = l ; = (k = 1, 2, .. .). (45)

From this it also follows that


I P n (cos 0) | < a 0a n + a 2 o„_r + • ■• + a na 0 = Pn (1) = 1. (46)
The formulae (37) make it possible to evaluate the Legendre poly­
nomials successively. Let us write the first five of these polynomials:

P 0 (x) = l) P 1 (x) = x ; P 2 (x ) = — (3x2 — 1) ;


1 1 (47)
P 3 (*) = - (5x3 - 3x); P A (x) = - (35x4 - 30x2 + 3).

When /(x) is a certain function given in the interval (—1, +1),


the question of representing it by a series of Legendre polynomials
arises
/ (*) = a 0 + c i i P i (*) + cl2P 2 (*) + . . . (48)
Using the fact that P n( x ) is orthogonal and also the formula (19)
we can see, as in the theory of trigonometric series, that the coeffici­
ents a n are given by the formulae
l
an + (49)
-I
It can be shown that for this choice of coefficients the series (48)
converges in the interval ( —1, + 1) and that its sum is equal to f ( x )
provided this latter function satisfies certain very general conditions.

133. The expansion in terms of spherical functions. Any function


given on the surface of a sphere of any radius, is a function with
geographical coordinates 0 and (p so that we can denote it by /( 0 , <p).
Let us suppose that it can be expanded in terms of spherical functions,
i.e. that it can be represented on the sphere in the form of a series
610 SPECIAL FUNCTIONS [133

analogous with the Fourier series:

/ (A, <p) = 4°) + 2 K ° p " <C08 0) +


n=l

+ ^ cosmq> + bM sin m<p) P n>m (cos 0)}. (50)


m —1

Using the property of orthogonality of spherical functions together


with the formulae (25), we obtain the following expression for the
coefficients of the series in the same way as for a Fourier series

o-n> = j si / {e’ ?,) P n 'm (cos 9) cos m<p d a ’

b" = 9)Pn.m ^ 6 ) ^ n m < p d a


(51)
S

(dm = 2 when m — 0 and 8m = 1 when m > 0 ; P„ „ (x) = P n (x ))

Speaking more strictly these considerations are only preliminaries


in the determination of the coefficients of the series (50). We must
subsequently substitute the values of the coefficients obtained from
the formulae (51) in the series (50) and prove that with certain assump­
tions made with regard to the function f ( 6 , <p) this series will converge
and its sum will be equal to f ( 6 , <p). We shall prove thiB in the next
section.
As a preliminary we shall explain certain integral relationships
which must be satisfied by spherical functions. Let S R be the surface
of a sphere of radius R and ¥ „ ( 6 , <p) — a spherical function of
order n. The function
U „ { M ) = Tr>Yn (d, 9 )
is a harmonic function and we can apply Green’s formula [II, 193]
to it:

a ,. (52)
S s

where d is the distance from the variable point M ' on the sphere S R
to the point M which lies inside the sphere, ds is an element of the
surface area of the sphere and v is the direction of the outside normal
to the sphere S r s o that, in this case, djdv = djdR. We therefore
have
__________ 1_________
d ]fR* — 2Rr c o b y + r 2 ’
j 33] t h e e x p a n s io n i n t e e m s o f s p h e r ic a l f u n c t io n s 511

ftnd also from (36)

i = J 'P ^ c o s y ) — (r< R ),
k=0 n
gO th&t

and
= n R n~ 1 Y n (0, V ) .

In these formulae y is the angle between the radii-vectors O M and


O M ' - Substituting all this in formula (52) and assuming that the radius
R is unity we obtain:

n Y „ ( 9 ' , <p') 2 P k (cos y ) rk +


fc=0

+ Y n (6\ <p') 2 (* + 1) p u (cos y ) rk da,


ft=0

where 9 ' and q>' denote the geographical coordinates of the variable
point M ' on the unit sphere. Since r < 1, the above series converge
uniformly with respect to 9 ' and <p' and the Legendre polynomials
satisfy the inequality (46). Integrating these series term by term we
have
rnY n (9, <P) + + x) Y n (9',<p') Pk (cosy)da.
ft—0 i
It follows directly from this sum that all its terms vanish except
the term corresponding to k = n ; this gives us the following integral
formulae which are important in the applications of spherical func­
tions:
Jf r„(« > -) p , (cosy)dc = 0 when m # 7i. (53)
S

J J Y n (d', < p ' ) P n ( c o s y ) d o = Y n (d, q>). (54)

We now introduce a formula which expresses cos y in terms of


trigonometric functions of the angles 9 , ip , 9 ' and <p'. We draw for
this purpose two radii O M " and O M ' of the unit sphere, the ends of
612 SPECIAL FUNCTIONS [133

which have coordinates ( 6 , <p) and (O', cp') respectively. The projections
of these radii on the axes of coordinates will be:
sin 0 cos q>, sin 6 sin <p, cos 0 and sin 6 ' cos <p', sin 0 ' sin <p', cos O',
and the cosine of the angle between these two radii will be expressed
as the sum of the products of these projections, i.e. we obtain the
following formula for cos y :
cos y = sin 0 sin 0' cos (<p — q>') + cos 0 cos O’. (55)
Let us return to the series (50) again. If this series is uniformly
convergent and its sum is equal to f ( 0 , <p) then we have the formulae
(51) for its coefficients, in the same way as we did in the theory of
trigonometric series. We can now unite into one term those terms
of the series (50) which are spherical functions of any given order
n , i.e. put

/ (0. 9>) = (0, <p). (56)


n= 0

On replacing 0 and <p by 0 ' and q>' in this expansion, multiplying


by P n (cos y ) and integrating with respect to the variables 0 ' and <p'
we have, from (53) and (54), the following formula for the terms of
the series (56):
r„(0, <p) = 2n4t 1 J { / ( 0/’ 9>')^n (cosy)da. (57)
S

This formula gives the sum of those terms of the series (50) which
stand under the symbol of summation with respect to n and which
refer to the given value of n .

Substituting the values of the coefficients (51) in a separate term of the sum
(50) we have:
n (ji_| 2n 4- 1 I Cr
y„ (0. <p) = 2 (w + m)l ~ 2 3^ - 1c o s m ( p J J f (0,>^ cos m(p’ P n ‘m (cos 0,) dff +

+ sinm<p J J/ (O', <p') sinnwp' P„,m (cos O') da p n,m (c ° 3 0


s
or
Y„(B, <p)

= J J / ( 0 ,.7 ’') 2 p T + ot)I ^ 2v T Fn’m (c o s P n -m (c° 3 0) 008 m ~ d<T~


9 (58)
134] PROOF OF CONVERGENCE 513

A comparison of the formulae (57) and (58) gives


y (n — to) 1 2
JJ/(6 '. <P') ( cos y) - ^=o (« + m)l (5m X

X P, (cos 0') P nfn (cos 0) cos to (y' — q>) do = 0. (69)

Strictly speaking we deduced this formula on the assumption that /(0, <p)
ia the sum of the uniformly convergent series (50). In particular this formula
will certainly be valid if the sum of the series (50) can be obtained in a finite
form. Notice th at the angle y is one of the geographical coordinates (latitude) if
we take one of the points with geographical coordinates 0, <pfor the North pole.
Hence rn P n(cos y) is a homogeneous harmonic polynomial of the nth degree and,
consequently, P„(cos y) is a spherical function of the nth order of the variables
S' and q>'. We thus see th at the square bracket in formula (59) is a finite sum
of spherical functions and it is therefore correct to assume that /(0 ', <p') is equal
to this finite sum of spherical functions. Hence with this choice of functions
we find th at the integral of the square of the above square bracket is equal to
zero and therefore the whole expression in the square bracket must also be
equal to zero:

P„ (cos y) = £ P„, m (cos 0') P n>m(cos 0) cos to (<p' - g>). (60)

This formula is usually known as the addition theorem for the Legendre poly
nomials.

134. Proof of convergence. We will now show that the arbitrary function
/(0, <p) which is given on the surface of a sphere and which satisfies certain
conditions, can be expanded into the series (56) in terms of spherical functions.
Bearing formula (57) in mind we obtain the following expression for the
sum of the first (n + 1) terms of the series (166):

s„ = i* n Jf JI".f (B’’
k -o
2 {2k + 2) p k(cos y) d<T-
5
Let us introduce new geometrical coordinates y and fi, plaoing the North
pole at a point with former geographical coordinates 0 and <p. At the same time
the function /(0 ', rp') will, in this new system of coordinates, become a new
function F(y, ft) and we have
n
= j F (y> P) 2 ^ (2& + 1) P* (cos y) sin ydydfi. (61)
0 0
We now introduce a new function <P(y) which ia the mean value of the func­
tions F(y, ft) on different parallels of the new system of coordinates
j w
®(y) = 2^" J F(v> P)&P- (62)
0
614 SPECIAL FUNCTIONS [134
We now introduce a new variable x = cos y and put
4> (y)= Y (x). (63)
Integrating the formula (61) with respect to /S we can rewrite it in the form
n
s„ = f 0 M 2 (2* + !) p k (°°3 y) sin V dy
2i *-•
or
1
s„ = - fv ( X) 2 (2k + 1) Pk (x) dx ,
z J k-Q
“1
i.e. from (41)

“I
We assume th at the function/(0, <p) is such th at yj(x) has a continuous deri­
vative in the interval ( —1, +1). Integrating by parts we have
i
S„ = y [ ¥ ( X) (P„+1 (X) + P„ (X))]*:1^ - y J [P„Tl (X) + P„ (x) ] «P' (x) dx ,

or bearing in mind that


P» (1) = -Pnu(1) = 1! pn(- 1) = ~ P„+t (1) = (- 1)".
we have
i
S n = y (1) - - J [Pntl (x) + P„ ( X ) ] V (x) dx. (64)
-1
Let us now explain the meaning of the first term tf'(l) on the right-hand side.
We have from (62) and (63)
2JI
sr(1) = ' s r J F(0,/,)djS- (65)
0
But, when y = 0 and /9 is arbitrary, this point is the North pole of the sphere
or, which comes to the same thing, the point with the former geographical
coordinates 0 and q>. In other words P(0, /S) = /(0 , <p) does not depend on §
and formula (66) gives
'F(i) = f(e,<P).
We can therefore rewrite the formula (61) in the form
1

s n = / (0, <P) - y J [p n„ (x) + P„] ¥" (x) dx. (66)

We have to prove that


lim S„ = / (0, p) ,
135] T E E CONNECTION BETWEEN SPHERICAL FUNCTIONS AND LIMIT PROBLEMS 515

i.e. we have to prove th at the integral in the formula (66) tends to zero as
n increases indefinitely. Let M be the maximum value of the continuous func­
tion V'(x) in the interval (—1, + 1). The modulus of the above integral will be
amaller than the following expression:
1 i
■ ^ J l P„» ( * ) | d * + - ^ - J | P „ ( * ) | d * .
-1 “I
We therefore only have to show that the integral

/ I P n (*) Id* (67)


-l
tends to zero as n increases. Using Buniakowski’s inequality [in ,, 29], we
have:
( I 2 i i i
J \P„ (X) I d r < J P5(*)dx J i*d* = 2 J P8(*)dte
-1 -1 -t -1
or from (19):
i

from whence it follows that the integral (67) tends to zero as n -* oo.

The above method of proof of the expansion theorem in terms of


spherical functions is taken from the book D i f f e r e n t i a l E q u a t i o n s i n
P a r t i a l D e r i v a t i v e s i n M a t h e m a t i c a l P h y s i c s by Webster-Sage. The
fact that an arbitrary function which satisfies the above general
conditions [ P ( x ) has a continuous derivative] can be expanded in
terms of spherical functions shows that spherical functions form a
closed system [II, 155] on the surface of a unit sphere. The fact that
this system is closed was first proved by A. M. Liapunov (1899).

135. The connection between spherical functions and limit prob) ms.
We will now show the connection between the theory of sph< rical
functions and certain limit problems of differential equations. Ix;t us
write the Laplace equation in spherical coordinates [II, 119]:
9 [-2 dU \ . 1 3 ( . a dU \ , 1 3’ U n /AQV
3r V 3r j + sin 0 30 (Sm0 00 J+ sin20 3p2 °'

We will seek its solution in the form of a product of functions of r ,


and 0 and <p respectively:
U = f ( r ) Y (0 . cp).
616 SPECIAL FUNOTION6 [135

Substituting in the equation (68 )

y (9 , 9 ) ['■ / ' w ] + / w w h 9 +

■ 1 &Y (fl.y) 1 n .
sin* 6 J ’
on separating the variables this can be rewritten as follows

A fr, r(r)1 j- 92r_


dr I ‘ y 'I sine 96 I, 96 ) ^ sin20 9?>2
Hr) ~ Y
The left-hand side contains only the variable r , the right-hand side
only 6 and <p and both sides will therefore be equal to one and the same
constant. Denoting this constant by X we obtain two equations:
r* /'(r) + 2 r /'( r ) - V ( * ’) = 0 (69)
and
A1 F + A F = 0, (70)
where, for the sake of briefness, we put

The function f ( r ) we know already, viz. from (5) it should be equal


to r n and therefore we will pay greater attention to equation (70).
The function Y ( 6 , tp), as we have seen above, is a trigonometric
polynomial and, consequently, it must be finite and continuous
on the whole sphere, i.e. for any choice of the angles 6 and cp and,
in particular, wheni 0 = 0 and 0 = n and sin 0 vanishes. We thus
come to the followng limit problem: f i n d v a l u e s o f t h e p a r a m e t e r X,
f o r w h i c h t h e e q u a t i o n (70) h a s c o n t i n u o u s s o l u t i o n s o n t h e w h o l e u n i t
s p h e r e a n d c o n s t r u c t t h e s e s o l u t i o n s . The first part of the problem
presents no difficulties for we know that f ( r ) is equal to r n and, sub­
stituting this in the equation (69), we obtain an infinite number of
solutions for the parameter X viz.:
Xn = n (n + 1) (» = 0 , 1, 2 , . . . ) . (72)
At the same time the equation
T* f n ( r) + 2rf'n (r) - » (n + l)/„ (r) = 0 (73)
will have one solution f n( r) — r n and another solution f n( r) = r ~ n~ 1.
Substituting X = n ( n + 1) in the equation (70) we obtain an equation
136] TH E CONNECTION BETWEEN SPHEBICAD FUNCTIONS AND LIMIT PROBLEMS 517

for spherical functions:

I C T [ w ( ™ flT ) + l S T ^ L] + »(» + 1) Ir" = 0 ' <74>


In this case the individual value of An = 71(71 + 1) corresponds to
(2n + 1) individual functions. These will be spherical functions of
order n . Owing to the fact that spherical functions form a closed
system on a unit sphere these functions give all the values of the
individual functions in the equation (70). Substituting the expressions
(16) in the equation (74) and putting x = cos 6 we obtain for P n>m{x )
the following differential equation of the second order:

_5L [ (1 _ * .) (*) = 0 : (75)

when m = 0 an equation for the Legendre polynomials P n( x ) is


obtained. The individual values and the corresponding individual
functions P n,m(x) solve the following limit problem. Find values of
for which the solution of the equation (75) remains finite in
the interval —1 < x < + 1 including its ends. Notice that at the
singularities x = ± 1 the equation (75) has a determining equation
g ( g — 1) + Q — m2/4 = 0 the zeros of which are g - ± 771/ 2 .
The solution which corresponds to the zero g = —m/2 becomes
infinite at the corresponding singularity.
The above problem involves the finding of those values Xn for
which the solution corresponding to the zero g = m/2 at the point
x = — 1 will still correspond to this zero at the point x = + 1 .
The values Xn = n ( n + 1) give the solutions of this problem and
the corresponding individual functions are determined from for­
mula (12).
The orthogonality of spherical functions is directly connected with
the fact that these functions solve the above limit problem for the
equation (70). Similarly the functions P n>m(x) are orthogonal on the
line ( - 1 , +1):
1

J P p , m W p < ,.m {x )te = 0 when p jtq . (76)


-1

This can be proved by using the equation (75) in exactly the same
way as we did in [102] for the Legendre polynomials. Notice also one
other fact connected with the theory of spherical functions. If we
use the solution f n(r) = r n of the equation (73) we obtain the solu-
618 6PECIAL FUNCTIONS [136

tion r n Y n ( 9 , <p) of the Laplace equation. This is a harmonic poly­


nomial of the n t h degree. If we use the second solution f n(r) = r- " ' 1
of the equation (73) we arrive at the following conclusion: t h e f u n c t i o n
Y„ (0, <
p)
n+1 (77)
T

w h e r e Y n{Q, tp) i s a s p h e r i c a l f u n c t i o n o f o r d e r n , i s a s o l u t i o n o f the


This solution becomes infinite when
L a p la c e eq u a tio n . r = 0 and it
is obviously not a polynomial in x , y , z.

136. The Dirichlet and Neumann problems. Spherical functions are


used in problems of mathematical physics connected with the Laplace
equation with reference to a sphere. As an example consider the
Dirichlet and Neumann problems which we mentioned earlier [II, 192]
in connection with a sphere. It is necessary to determine a harmonic
function inside a sphere of radius R when its limit values are given
on the surface of that sphere (the inside Dirichlet problem). We ex­
pand the given limit values in terms of spherical functions:

f ( e , < p ) = 2 Y n(0.<P)' (78)


n=0
and construct a new series by multiplying the ttth term of the
above series by ( r j R ) n , where r is the distance of a variable point
from the centre of the sphere:

U ( r , Q, <p) = ^ (0 , <p) (-£■)" (r < R ). (79)


n=0
Bearing in mind th a t( l j R f i) Y n {9 ,( p ) r n is a harmonic polynomial we
can see that the function (79) is harmonic inside the sphere and also
that, when r = R , the series (79) becomes the series (78) so that this
harmonic function satisfies the necessary limit conditions.
Consider now the e x t e r i o r D i r i c h l e t p r o b l e m , i.e. assume that it is
necessary to determine a function, harmonic outside a sphere, which
becomes zero at infinity [II, 192], from its limit values (78) on the
surface of the sphere. Bearing in mind that Y n( 6 , <p) r ~ n ~ l are har­
monic functions which have no singularities outside the sphere and
which are zero at infinity we obtain the solution of the exterior
Dirichlet problem in the form:
U ( r , e , tp) = J r n (flt <p)(4 ) n+1 (80)
n=0 v J
136] TH E DIRICHLET AND NEUMANN PROBLEMS 619

We ehall now consider the i n t e r i o r N e u m a n n p r o b l e m . Suppose that


it is necessary to determine a harmonic function U ( r , 6 , <p) inside a
sphere from the values of its normal derivative on the surface of the
sphere
917 tta ^
~ W = f ( 6 ’ 9) (t = R ). (81)

We know that the integral of the normal derivative of a harmonic


function vanishes [II, 194]:

i.e. the given function f(0 ,c p ) which appears in the condition (81)
must be such that
J j 7 (0 , cp) do = 0 . (82)

On recalling the formula (57) which determined the spherical functions


obtained in the expansion of f ( 6 , <p) and also that P n (cos y ) is constant
when n = 0 , we can see that the condition (82) is equivalent to the
fact that in the expansion of f ( 9 , <p) in terms of spherical functions
all spherical functions of zero order are absent. We therefore have

m ? ) = 2 Y n{0,9)- (83)
n=1

It can readily be seen that the solution of the Neumann problem


will be given by the following formula:

U (r, 6, ? ) = > ’- Y n (0, 9 ) -^ = r + O, (84)


n=l

where C is an arbitrary constant.


In fact this series determines a harmonic function and differentiation
along the normal coincides, in this case, with differentiation along the
radius r . It can easily be proved that, by differentiating the series
(84) with respect to r and putting r = R , we can obtain the series (83),
i.e. the limit condition (81) will be satisfied. In the case of the e x t e r i o r
N e u m a n n p r o b l e m the function f ( 6 , cp), which appears in the condition
(81) will no longer satisfy the condition (82) and we have an ex­
pansion of the general form (78) for it. It can readily be seen that
520 SPECIAL FUNCTIONS [137

the solution of the exterior Neumann problem will be given in the


form of the series

= l y , , _ ? £ i, (8 5 )
n =0 '
where we assume that the direction of the normal v coincides with
the direction of the radius r.

We will now consider one special case of the exterior Neumann problem.
Let us suppose th at a sphere of radius if moves in a limitless fluid resting at
infinity with a velocity a directed along the Z axis. Take a system of coordina­
tes with the origin as the centre of the sphere which moves with the sphere. In
this case the normal component velocity of the fluid with the surface of the sphere
will be given by the formula:
az
----= a cos 0.
T
If we suppose that the fluid is stationary and has a potential velocity, we
have a problem in which the function U is to be found from the following
conditions: (1) outside the sphere U must be a harmonic function, (2) at in­
finity the component velocities, i.e. the derivatives of the function U with
respect to the coordinates, must vanish and (3) on the surface of the sphere the
function U must satisfy the condition
dU = — a cos 0.
a
or
In this case f(0,<p) = —a cos 0, or, remembering the expression for the
Legendre polynomial we have
/ (0, <p) = —aPl ( cos 0),
i.e. the function f(8, <p) is a spherical function of the first order. The solution
of the problem will be given by the formula
a R3 aR3
U (r, 6,<p) = — P l (cos 0) = ~2 ~T cos e -

137. The potential of voluminous masses. Let us suppose there is a bounded


volume V in space which is filled with a mass of density pfM'J.The potential
of this distribution will be expressed by the treble integral

[ 7 ( M ) = J J J e ( M'± d V , (86)
v
where d is the distance from the variable point M ' of the volume F to the point
M at which the value of the potential is being determined. Let 0 bo the origin
and wo will introduce into our considerations the lengths of the radius-vectors
r= \O M \, r' = \ OM'\
137] TH E POTENTIAL OP VOLUMINOUS MASSES 621

and the angle y made by these vectors. We now consider points M a t a suffici­
ently great distance so that the value of r is greater than the maximum value of
r'. For these points we have the following expansion [132]:
1 1 “ r 'n
— = — — - = ^ P n ( c o s y ) -------------->
d ifr2 — 2rr' cos y + r'2 ~ 0rn+1
which converges uniformly with respect to r' since | P n(cos y)| < 1. Substituting
this in the integral (86) we obtain the expansion of the potential U(M) in nega­
tive integral powers of r:

Y„(8,<p)
U (M ) = ^ rn+1 (87)
n=0
where
= (M’) r'n P„ (cos y) AV. ( 88 )
V

We will now determine the first three terms in the expansion (87). On re­
calling the expression for the first three Legendre polynomials and also the
self explanatory formula:
xx' + y y ' + zz'
cos y =
rr'
we can write
P0 (cos y) = 1; r’ P x (cos y) = ” +V^_ + “

r'2 P, (cos y) = i - +.

Substituting these in the formula (88) we have first of all


Y0(0, 9>)=jne(iW')dF = m,
v

i.e. the coefficient of 1/r in the expansion (87) is equal to the total mass m
of the volume F. We obtain further:
(0, <p) = J J J e {M') r' P, (cos y) d F =
V

= - J J J e (i i P ) * ' d F + | - J | | e ( M ' ) 2/' d F + - l J J J e ( M ' ) 2 ' d F .


V V V

The above integrals express the product of the mass m and the coordinates
of the centre of gravity. We shall assume th at the origin is so chosen as to coincide
with the centre of gravity of the mass. In this case we shall obviously have
Ft(0, <p) = 0. Lastly we shall evaluate Y t{9,<p). To do so we introduce the
moments of inertia of our mass with respect to the axes:
A = j-j- J q {M') (y’2 + *'*) d F ; B = J J J <? (AP) ( z '» + x '2) d F ;
V V

<? = n i< ? (A P ) (x'2 + y'2) d F , (89)


v
622 SF E O IA l FUNCTIONS [138
and also the products of inertia with respect to the axes

D = n Je ( M ‘) y 'z'd V ; E = J J J e {M') z' x ' d V ;


v v

F = SSS<>(M’) x ' y ' a V . (90)


V

I t can be shown, but we shall not do so here, that the system of coordinates
can always be chosen so th at the products of inertia (90) vanish. We will
assume th at the coordinates have been chosen in this way. Substituting the
expression r'1 P 2(cos y) in the formula (88) we can see, as can easily be shown,
th at the following expression for Y Z(B, q>) is obtained:
1 {B + C - 2A) xt -f (C + A - 2B) y1 + (A + B - 2C) 2 *
Y* <«.*) 2 r*

and for the potential U(M) we have up to terms of the order of 1/r3
U (M) =
m 1 (B + C - 2 A ) x * + (C + A - 2 B ) y * + (A + B - 2 C ) z *
~ t + 2 r5 + '

Replacing x, y and z by spherical polar coordinates we can rewrite this ex­


pression as follows:

U (M) = — +

1 (R+C —2^4)cos*ysin!0-|-(C?+-4.—2B)sin*ysin20-(-(^4.-(-B—2 0 ) 008*0


+ ~2 r3 + ‘"
(92)
138. The potential of a spherical shell. Suppose that on the surface
S R of a sphere of radius R a mass with surface density q ( M ' ) is distri­
buted. The potential U ( M ) of this simple layer will be expressed on
the surface of the sphere by the integral

*7(Jf)=JJ-^-ds, (93)
S r

where d is the distance from the point M to a variable point M '


on the surface of the sphere. The expression 1[ d will have different
forms inside and outside the sphere S R .
If we assume, to start with, that r < R , we obtain [132]:

- = v p ^ c o s? )-^ . (94)
n=l
138] TH E POTENTIAL OF A SPHERICAL SHELL 623

where y is the angle between the radius-vectors O M and O M ' . The


density q ( M ' ) is assumed to be a given function f ( 9 ' , <p') of the
geographical coordinates of the sphere.
Substituting the expansion (94) in the integral (93) and remem­
bering that ds = ii^dcr = R 1 sin 0 ' d d r p ' d ' we have:

U(M ) = jg -j— \ <P') P n (cos y) do. (95)


n= 0 Js

The above integrals are directly connected with the expansion of


the function f ( 6 , <p) in terms of spherical functions, viz. if

f(9,<p) = 2 T n i ° ’ V) , (96)
n=0
then, as we know

Y n ( 9, <p) - 2n + 1 J J / ( e'. <P') P n (cos y) do,


S

and, consequently, the expansion (95) can be written as follows:

U(M ) = in ^ (2w +7)i?n-i- Y n ( 0 , 9 ) (r < * ). (97)


n^O ' ' ’

Similarly, using the expansion (36) we have

V (Jf) = i n 2 YAe.f) (r>R). m

By using this expansion we can note certain properties of the


potential of a layer. Notice, first of all, that the expansions (97) and
(98) are the same when the point M falls on the surface of the sphere.
In this case we put r = R and obtain the following result:

U ( M 0) = 4 n R 2 ^ - T Y n ( 6 , <p) , (99)
n=0

where 9 and <p are the geographical coordinates of the point M 0 on


the surface of the sphere. We can thus see that t h e p o t e n t i a l o f a s i m p l e
la yer varies co n tin u o u sly a s the poin t M passes th ro u g h the su rfa c e
This property of the potential of a simple layer holds
of th e sp h e r e .
not only for a sphere but for surfaces generally.
624 SPECIAL FUNCTIONS [138

Let us now investigate the behaviour of the normal derivative of


the potential (the normal component of the force) when the point M
moves across the surface of the sphere. We denote by ( d U ( M Q)jdv)i the
limit of the normal derivative as the point M tends to the point
along the radius from inside the sphere, and by ( d U ( M 0) j dv) e that
same limit as the point M tends to the same point M 0 from outside
the sphere, v denotes the direction of the outward normal to the sphere
at the point M 0. In this case this direction coincides with the radius
O M 0. Differentiating the formulae (97) and (98) with respect to v, i.e.
with respect to r, and putting r = R we obtain the expressions for
the above limits:

n=l

(«.*’>• (ion
s ' n=0 1

This shows that the normal derivative of the potential of a simple


layer has, in general, a discontinuity on passing through the surface.
The following formulae follow directly from the formulae (100)
and ( 101):

(«• *■>■
0

( 3U (iWo) ^ | ( d U { M q) \ . 1 v /fl «\
[ oi Je + 1— § r ~ ) i = - An2n—
0
-w + t 7 "{6’v)’

and from (96) and (99) we can write:

(•E S a ) - ( ^ ) 1— <IM>

I 9v Je + { Qv ), = --------------R— ■ ( 10d)

Formula (102) shows, among other tilings, that the d is c o n tin u ity
i n the n o r m a l d e r i v a t i v e i s e q u a l to th e p r o d u c t o f ( — i n ) a n d th e d e n s i t y
a t th e g iv e n p o i n t o n the su rfa c e.
Let us now explain the meaning of the right-hand side of the for­
mula (103). Denoting, as before, a definite direction by v, viz. the
direction of the radius O M 0, and bearing in mind that in the integral
13 9 ] THE ELECTRON IN A CENTRAL FIELD 626

(93) only the factor \jd depends on the coordinates of the point M,
we have

Sr

But we have also:


8 (n l
"97 l"dj = d2" 003 <U’
where co is the angle made by the radius-vector M ' M and the direc­
tion v. Let us evaluate the integral (104) on the assumption that
the point M lies on the sphere, viz. at the point M 0. In this case
we have d — 2 R cos co and, consequently:
± ( ± - ) = ____ L_
dv U J 2Jta
We thus obtain the following expression for the integral (104):

Sr

Let us denote this by d U ( M 0) / d v . We can then rewrite the formula


(103) as follows:
( dU (M0) \ ( dU (Ma) \ g dU (M0) _
I 9v )e ' I dv )i dv
From this follow the expressions for the limits of the normal
derivative of the potential of a simple layer which are given below:
( dU (M0) ■) dU (M0)
+ 27tQ ( M 0),
I dv Ji ~ dv
(105)

These formulae are also valid for bodies other than spheres.

139. The electron in a central field. When dealing with an electron in a


field created by a positive nucleus we have, according to Schr6dinger’s theorem,
the following equation:
h1 ( 8 ! c 92 y> ( 106)
2/t [ 9a:2 dyz ^ t ) ~ eV (r ) V = Ev>

where h is Planck’s constant, fi is the mass of the electron and e is its charge.
V(r) is the given function which depends only on the distance r from the origin,
and determines the potential of the field, y(x, y, z) is the wave function
and, finally, E is a constant which determines the energy level of the given
5 26 SPECIAL FUNCTIONS [139

physical system. The solution of the equation (106) must be finite in the
whole infinite space and remain bounded at infinity. We shall seek this
solution in the form of a product of functions of r multiplied by functions which
depend only on 0 and (p. Expressing the Laplace operator Ay* in terms of spherical
coordinates we can write:
82y> 2 dtp 1
AV, = “0^“ + T " 0 T + 7 ^ AlV’’
where, as before [136]:
d*tp
AiV = sin 0 00 + dtp*
The equation (106) can be rewritten in the form

^ - [ ^ + T ^ - + ^ A' y] + eF(r)v + £v = 0-
Writing the expression yj = /( r ) Y(8, <p) and separating the variables we
have:

Ai Y
- ^ [ /" ( r) + - r (r)] - eV ( r ) / (r) - Ef (r)
h*
2/1 T 1 f{r)
Both sides of this equation will be equal to one and the same constant
which we denote by A. This gives the equations
A1r - i r = o, (io7)

- ^ [r(r) + -fW + - / ( r ) ] - « V ( r ) / ( r ) - i / ( r ) = 0. (108)

The equation (107) should have a continuous solution on the whole surface
of the sphere. We know already th at in this case the parameter A will be
equal to —1(1 + 1) and the solution will be expressed in terms of spherical
functions T;(0, tp). Substituting the above value of A in the equation (108) we
obtain an equation for the determination of the function of r which we now
denote b y /j(r):

~ f'1' <r >+ ^ fl <r >+ [£ + eV M “ ] f‘ <r) = °- <109)


The values of the parameter E are determined from the condition th at the
solution of the equation (109) is bounded when r — 0, as well as when r tends
to + oo. In general, we can obtain an infinite number of such solutions. They are
usually numbered starting with the integer (I + 1), i.e. they are numbered
in the following order:
w = I + 1, I + 2, I + 3, ...
Hence the values of E depend on two symbols, viz. on the integer I and on
the number n. I is known as azimuthal quantum number and n as the principal
quantum number. When I and n are given we have, in general, a definite function
139] THE ELECTRON IN A CENTRAL FIELD 527

f nl(r) which satisfies the equation (109) as well as the above limit conditions
when r = 0 and r = + oo. As far as the functions Yj(6, <p) are concerned, there
will be (21 + 1) of these:
(m = — 1, - 1 + 1............ 1 - 1 , 1)
and for the full definition of the wave function we must also determine the
value of the third symbol m. This is usually known as the magnetic quantum
number. I t is of importance in disturbance problems of a given physical
system caused by a magnetic field acting along the Z-axis.
Let us now consider the particular case when the potential is the Coulomb
potential
V (r)= — ,

where & is a constant which is equal, for example in the case of a hydrogen
atom, to unity. Substituting the expression of the potential in the equation
(109) we obtain the following equation (h = 1):
A21(1 + 1) 0.
Tt7 /'' (r ) + - 7 /i(r) + [ s + - i - - ^ -] It (r) ( 110 )

We replace r by a new variable z:

h2
and put
Eh2
and a = 21 + 1. (Ill)
fie*
We also replace f e(r) by a new unknown function y:
fi(r) = - ^ y -
V*
Substituting all this in the equation (110) we obtain the equation
d 2y
dz2 + T - i r + (2e + T ~ ^ y J/ = 0-
which we considered in [115].
Let us now consider the negative values of the parameter E. In this case,
as before, we obtain an infinite number of discreet values for the constant
E, viz. having put
= — 1 -,
r^ 2 ?
we obtain the following value for the parameter A:

( P - 0 , 1, 2, ...) ,
whence
1 12 _ i i
^ r= (^ -+ p ) and 2 (P + 1 + 1)!
628 SPECIAL FUNCTIONS [140

and, consequently, from (111) we obtain the following values for the para­
meter E:

Enl = ~ 2h2 {p + 1 + 1)* ~ “ 2/t2na * (U2)


where n is the principal quantum number equal to (p + 1 + 1)-
We thus see th at in the case of a Coulomb field the values of the parameter
E are independent of the azimuthal quantum number 1. I f we fix n, and there­
fore also the value of the parameter E then, since n = p + 1 + 1, we can
give 1 the following values:
I = n — 1, n — 2.............. 0.
Every value of 1 corresponds to (21 + 1) natural functions of ip. Therefore
for the parameter E, as given by (112) we obtain the following total number
of natural functions:
1 + 3 + 5 + . . . + (2n - 1) = n2.
If, instead of Schrfldinger’s equation we had taken the Dirac equation for one
electron then we would have obtained functions analogous with spherical func­
tions. These “Spinning spherical functions” are dealt with in the book The
Origins of Quantum Mechanics by Prof. V. A. Foch.

140. Spherical functions and the linear representation of rotating groups.


We have already mentioned the fact th at homogeneous polynomials of the
variables (x, y, z) which satisfy the Laplace equation give a linear expression for
a group It of the space rotating about the origin.
We thus see th at a set of spherical functions of order I gives a linear expression
for the group R which is of the (21 + l)th order. Let us consider this problem
in greater detail.
Let us consider spherical functions of order 1 in the same form as given by the
formulae (18) and introduce for them a special notation:
Q|m)(y,e) = e"nvP<im(cos 8) (m = - l , -1 + 1 1 - 1 , I),(113)
where
P ;i_m(c o s 0 )= p im>(cosfl)
Let {a, /?, y} be a certain rotation of the group R with Euler angles a, fi
and y. As a result of this rotation a point on the sphere with coordinates (<p, 0)
will move into a new position (<p', 0')and the function 8') can be expres­
sed linearly in terms of 0).The matrix of this linear transformation will
correspond to the rotation {a, /?, y j of th at linear expression for the group R,
which is given by the functions (113). The simple dependency of these func­
tions on the angle tp shows th at a rotation about the Z axis through an angle
a, i.e. the rotation {a, 0, 0}, corresponds to the diagonal matrix
e - " a. 0, 0, , 0
0, 0, . 0
e -i(/-2 )a
0, 0, , 0 (114)

0, 0, 0, . e"°
140] SPHERICAL FUNCTIONS AND T H E LINEAR REPRESENTATION OF ROTATING GROUPS 529

Denote by the elements of the matrix corresponding to a definite


rotation R„, where the symbols i and k take the values —I, —I -f 1........ I.
Take for R 0 the rotation about the F-axis through an angle fi, as a result of
which points on the sphere with coordinates <p= 0 and 0 move to points q>' = 0
and 0 + f) respectively. Bearing in mind the form of the functions (113) we can
say th at for the given choice of R 0 the matrix D((JJ0) transforms the functions
Rjm>(cos 0) into the functions P I m [cos (0 + /?)], i.e.
1
Pt m [c°s (fl + /*)] = ^ {Di(Ro)}msPl,s(cOS 6) (m = - I - l + l I).
S = — l

Returning to the formulae (12) we can see that P f ](cos 0) vanishes when
$ = 0 if s 0. Putting 0 = 0 in the above formulae we have

1*1, m (COS0) = {A (A)}mO Pl(P) — {A (Ro)}mo-

This shows th at the elements of the column k = 0 of the matrix Dt(R0\


ore, in general, other than zero, i.e. they contain zero elements only for excep­
tional values of /J.
Thus among the matrices Dt(R) are the diagonal matrices (114) with dif­
ferent elements, as well as matrices, all elements of a certain column of which
are other than zero. As we saw in [EH,, 69] the matrices give, in this case, an
irreducible expression for the group R and we can therefore say that the expres­
sion given by the matrices Dt(R) is irreducible. The functions (113) are orthogonal
in pairs but they were not compared to unity for integrals of the square of the
modulus. Ascribing to these functions suitably chosen constant factors, we
can construct comparable functions:

(115)

These functions now give the unitary irreducible expression for the functions
Dj(R) [III,, 63] which is equivalent to Dt(R) and in this new expression the
rotation {a, 0, 0} corresponds to the old matrix (114) since the constant factors
we have chosen do not change the character of the dependency of the functions
(113) on <p.
Multiplying the functions (115) by arbitrary factors the moduli of which
are unity we also obtain a unitary expression with the same matrix (114) which
corresponds to the rotation {a, 0, 0}. One of these representations is exactly
the same as that which we constructed in a different way [EH,, 62],
The individual functions of the Schrodinger equation which we considered
in the previous section, fall into groups in accordance with the values of I, and
each group contains (21 + 1) individual functions (I being the azimuth quantum
number). The functions which form such a group are numbered by m, (m being
the magnetic quantum number) which runs through m = —I, —1 + 1, . . . , i .
I t follows directly from the form of the functions (113) that

(9, 0) = mQ{T ) (<P, 6).


630 SPECIAL FUNCTIONS [141
i.e. the mth function of our group is the individual function of the operator
1 9
^ t ' (116)

and m is the corresponding individual value. Also each function (113), as we


know, satisfies the equation:

- d 1 ( # n) (<p, 6) = I (I + 1) Q(,m) (r, 6),

i.e. each function belonging to the above group of (21 + 1) functions is an


individual function of the operator

and the corresponding individual value is equal to 1(1 + 1). The operator Lt
differs only by the factor h from the operator of the component of the quantity
of movement on the 2-axis. Similarly, the operator (117) differs only by the
factor h1 from the operator of the square of the moment of the quantity of
movement.

141. The Legendre function. Let us consider the Legendre equation:

(l - x2 ) ^ . - 2 x ^ : + n ( n + l ) u = 0
dx (118)

where we assume that a; is a complex variable and where n can be


any number. Both zeros of the determining equation (118) are zero
at the singularities x = ± 1 [102]. Hence at both these points we have
one regular solution and one solution containing a logarithm; this latter
solution will not be bounded in the neighbourhood of the singularity.
We shall try to satisfy the equation (118) by an integral of the form
(13) which is a Legendre polynomial when n is a positive integer:

u (x ) 1 f (*a — l)n df. (119)


2n+1 m J it
(t —x)n+1

Substituting in the equation (118) we have:


da u
(1 - x*) dxa - 2 x - ^ + m (ra + l ) u =

= J (,'‘1 „!,« [- (« + ) «* - 1) + (» + 1)
\n
2 2 t if - x)] di =
_ n+ 1 fr d d [r ((«
(f - p w i i .
2n+17ii J d/ L ((-*)"«] ’
c
141 ] TH E LEGEN D RE FUNCTION 531

which shows that formula (119) is a solution of the equation (118)


provided that by describing the variable point t round the contour C ,
the expression
(t2 - l)n+l
(I — x)"+2 ( 120 )

returns to its initial value. When n is a fraction then the integrand


in (119) has three branch points: t = x and t = ± 1. By describing
the point t = 1 or t = —1 in the counter-clockwise direction the
numerator (t 2 — l )n+1 acquires the factor e(2n+1)271' and by describing
the point t = x the denominator acquires the term e(2n+1)2,“.

Let us cut the i-plane from t = —1 to t = along the real axis and
take as the contour C a closed contour originating at a point A on the
real axis which lies to the right of the point t = 1 and round which
the points t = 1 and t — x are described in the counter clockwise direc­
tion (Fig. 73).
We assume that x does not lie on the cut and that the con­
tour G does not intersect the cut. The original value of the many­
valued integrand function is determined from the conditions that
arg (t — 1) = arg (t -+- 1) = 0 and | arg (t — x ) | < n , when t > 1. As a
result of this the expression (119) returns to its initial value when t de­
scribes the contour C . Notice also that, according to Cauchy’s theorem,
the value of the integral does not depend on the choice of the point A
to the right of t = 1 on the real axis or on the form of the contour.
It is only essential that the contour should not intersect the cut.
We thus obtain the solution of the equation (118):

= < 1 2 1 >
c
where C is the contour we described above. This solution is a regular
function of x in the whole cut plane and, in particular, at the point
532 SPECIAL FUNCTIONS [142

x = 1. But, as we know from [ 102 ], the equation (118) is obtained


from Gauss’Bequation when a = n + 1, /? = — n and y = 1, and when
the independent variable z in the Gauss equation is substituted by
(1 — x ) j 2 . Since the solution (121) is regular when x = 1, i.e. when
z = 0 , it must be the same as the hypergeometric series given below,
except for the constant term:

F n { x) = C F [ n + 1, — n, 1 ; 1 ~-*-j • (122)

To determine C we evaluate Pn (1):


p 1 f (« * -!)" d, _ J ___ f <t + 1)n dt
r n \ 1) — 2n+l ni J (t-l)n+l t-1 ’
C C

and, evaluating the last integral in accordance with the theorem of


residues, we obtain Pn(l) = 1 ; hence the formula (122) when x = 1,
gives C = 1, i.e.:

P n ( x) = F { n + 1, - n , 1; - ^ ) . (123)

When n is a positive integer we obtain a Legendre polynomial.


It also follows from the formula (123) that owing to the fact that
F ( a , f i , y \ z ) does not alter when the positions of a and ft are inter­
changed, we have for any n
P n { x ) = P . n_ l { x ) .

By using the formula ( 121) the relationships (37), (39) and (40) from
[132] can be tested. The function P n( x ) , which is the solution of the
equation (118) has, in general, singularities at x = —1 and x =
Formula (121) defines this function in the whole cut plane.

142. The Legendre functions of the second kind. We constructed one


of the solutions of the equation (118). We shall now try to construct
a second solution. We know that when y x( x ) is one of the solutions
of the equation
y" + p (x) y' + q{x)y = o,
then a second solution can be constructed according to the formula

ya (*) = Cy1 ( * ) J e - ^ WdxEi;J n r , (124)


where C is an arbitrary constant. Consider, first of all, the case when n
is positive. At the singularity x = «> of the equation (118) the deter-
142] THE LEGENDRE FUNCTIONS OF THE SECOND KIND 633

mining equation has the zeros = n + 1 and q2 = —n. The solution


of the equation which corresponds to the first zero vanishes when
X = ° ° - Using the formula (124) we can write this solution as follows:

Qnix) = P n ix) \ (1 - * 2) [P„(x)]2 • (125)


00

The function Q„(x) has singularities at x = ± 1 and is regular in the


2-plane which has a cut from z = - l t o z = l . Formula (125) defines
Qn(x) in the whole'plane. Notice that the zeros Pn(x) lie in the in­
terval (—1, + 1).
Let us express Q n( x ) in terms of a Legendre polynomial and loga­
rithms. To do this we replace u ( x ) by a new function v ( x ) in the
equation (118) according to the formula

«(*) = - P „ ( z ) l o g - |± ! -v {x ). (126)

We obtain the following equation for v(x):

(1 - *2) i s r - 2xl& + n { n + l ) v = 2 P 'n ( x) ,

and from (42) we can rewrite this equation as follows:

U “ **) - £ r ~ 2 a ; + n in + ! ) v =

= 2 J ( 2 7 i - 4 £ - f 3)P„_ 2fc+1 (x), (127)


*=i

where iV = n/2 when 7i is even, and N = (n + l )/2 when n is odd.


Bearing in mind that P n ~ 2k+ i { x ) satisfies the equation:

(1 - 2z P 'n - 2 k+1 ( * ) +
X 2) P"n- 2 k + l ( * ) -

+ {n — 2k -f 1) (n — 2k + 2) P„_2*+1 (*) = 0,

we can see that the equation

C1 “ **) ~ 2X1 & + 71 + X) W = 2 (271 ~ 4k + 3) P n - 2 k + 1 (*)


has a solution
/ \ 2w — 4k -|- 3 D / i
W ~ (24 — 1) (w — A + 1) P n~2k+i ix ) ■
634 SPECIAL FUNCTIONS [142

This, from (126) and (127), gives the following solution of the Legendre
equation (118):
* 2n-4fc + 3 D , .
«o (*) = T P n (x ) log ~ T ^ (21c — 1) (n — k + 1) (* )• (128)
fc=l
It can be expressed in terms of P n( x ) and Q n( x ) :

u0 (x) = C 1 P n (x ) + C 2 Q n (x ) . (129)
From (128) and the obvious expansion
1, 1+ x 1 .1 , 1 ,
(1*1 > !)
it follows th at M0(x)/xn~2 remains bounded as x — On the other
hand on the right-hand side of (129) P„(x) is a polynomial of the nth
degree and Q n( x ) tends to zero like l/(xn+1) a sx -> This being true,
we can say that 0 2 = 0 , i.e.

C1 Qn ( * ) = «o ( * ) = y - P n (*) ^ g — y ~ R n ( X ) , (130)
where R n( x ) is a polynomial of the (n — l)th degree. It follows that
r d f &.(*)! _ 1 , 8„ [ x )
2 dx L-Pnlx) J 1 [P„(*)]» ’
where S n( x ) is a polynomial in x. On the other hand, from (125):
d T<?„(*) I _ 1
d*LP„(*)J (1 - as*) [P„ (*)]* *
Comparing this equation with the one above we have:
______ Cf______ _ 1____ t &n (x )
( l - * * ) [ P fl(*)]* _ l - x J ^ [P„(X)]! ’
whence
C2 = [ P n (x)]2 + ( l - x 2) S n ( x) ,

and putting x = l we obtain C2 = 1, i.e.from (128) and (129) we have


finally:

«» W - T ■
p- w ~ i (a - r n f - t + ! ) f . * . <*> • <131>

The function Q n( x ) is usually known as t h e L e g e n d r e f u n c t i o n o f the


second h in d.
The logarithmic terms are due to the character of the singularities
x = ± 1 of the equation (118). Q n( x ) can easily be written in the form
142] TH E LEGENDRE FUNCTIONS OF TH E SECOND KIND 535

of a definite integral. Notice that when n is a positive integer the


expression (120) vanishes for t = ± 1. Therefore, when constructing
the solution of the equation (118) in the form (119) we can simply take
the line —1 < t < + 1 as the contour G and we have:
i

%(*) = J C
—1
d*’ (132)
where G is an arbitrary constant. This integral tends to zero like
ljxn+1 as x -> °o, and therefore this solution differs only by its con­
stant term from Qn(x). Let us determine the constant C so that the
solution (132) is the same as Qn(x). It follows from the formula ( 11)
that the coefficient of xn in the expansion of Pn(x) is equal to:
2 n (2 n — 1) . . . (n + 1) _ 2n!
n f 2" — (nt)22"
(133)

Returning to the formula (125) we can see that the expansion of


the integrand in whole positive powers of a;-1 begins with the term
(—l /a % x Zn+2) and the expansion of Q n( x ) begins with the term
l/(2ra + 1) a nx n + 1 . Comparing this with the formula (132) we obtain
the following equation for C :
i
C f ( l - * 2)ndt = - - 1 -,
J v ' on(2n + l)
—l
or
2
2C ( sin2n+1 cpd<p = — 1 lT
J
0
r r an (2n + 1)
hence [I, 100]
2ra(2w — 2) . . . 4 -2 _
2C (2n + 1) (2w — 1).. .5 • 3 o„(2n + l) ’

and from (133) we obtain G = l/ 2n+1. Substituting this in the equation


(132) we obtain an expression for Qn(x) in the form of the integral

Qn (*)=-2^kT J (l-ff+l—1
(134)

This expression holds in the whole of the 2 -plane except on the line
— 1 < x < + 1. We shall now express Qn(x) as a hypergeometric series.
To start with let us rewrite the formula (133) in terms of the function
536 SPECIAL FUNCTIONS [142

r(z) and use, for this purpose, the relationship (143) from [73] when
z = n + 1 , and the formula r(2n + 2 ) = (2n + l ) / 1 (2 n + 1):
2"+ ir(n + | )
r(2n + 1)
a" — [ r ( n + l)]22" (135)
( 2 n + l ) / i t / ’ (ii + 1)

Notice also that the substitution t = x1 transforms the Legendre


equation (118) into the equation
d2u 3t — 1 du n (n + 1)
* ( < - ! ) 1 0 2" 2 dT 4 u = 0,

and this is the Gauss equation with parameters a = n/2 + 1/ 2,


/? = — n/2 , y = 1 . Using the first formula when z = t and replacing t
by 7? we obtain the solution of the equation (118):
C
u(x) - - *n+l -2+ ' 1'
1 (1*1 > 1 ). (136)

which at infinity has the same properties as Qn(x) and only differs
from it by the constant term. The constant C must be so chosen that
the solution (136) coincides with Qn(x), the expansion of which
in powers of l/x begins with the term l/[( 2 n + 1)] anxn+1. Hence
C — l/( 2 n + 1 )an and we have:
YnT(n+ 1)
<?„(*) ■+ 1>
2 " + ir( n + -|)

Until now we have only investigated the function Qn(x) when n is a posi­
tive integer. Qn(x) can also be determined as the second solution of the
equation (118) when n takes any value, in the same way as we did with
Pn(x). Consider the integral (134). This integral retains its meaning
when the real part of (n-f-1) is positive and it can be used for the deter­
mination of Qn(x) for these values of n. In general Qn(x) can be deter­
mined by the contour integral (119) but the contour must be suitably
chosen. The expression (137) holds for all values of n except when n
is a negative integer. It must be remembered that when n is not a
positive integer then the point x = °° will be a branch-point for the
function Qn(x). It is determined in a plane cut from x = - o o to z = 1.
When n is a negative integer then, putting n = —m — 1 , where to
is a positive integer or zero, we can see that the equation (118) becomes
143] TH E DETERMINATION OP BESSEL FUNCTIONS 537

and P m(x) and Qm(x) are the solutions of the equation (118). The
formulae (37), (39) and (40) from [132] can readily be tested for Qn(x).

§ 2. Bessel functions

143. The determination of Bessel functions. We first met Bessel


functions in connection with the vibration of a round membrane
[II, 178]. Let us recall the results which we obtained at the time
and establish the connection between the wave equation and Bessel
functions.
The wave equation in two dimensions is as follows:

When dealing with the vibration of a round membrane we intro­


duced plane polar coordinates
x = r cos <p; y = r sin <p,
and found those solutions of the equation (1) which are in the form
of a product of three functions, one of which depends on t, the
second on r and the third on <p. These solutions, as we saw above,
have the following form:
(a cos cot + ft sin cot) (C cos pep -j- D sin pep) Zp (kr), (2 )
where a, ft, C and D are arbitrary constants, while the constants co, k
and a are connected by the relationship:
co2 = k2a2. (3)
Zp(z) in the above formula denotes an arbitrary solution of the
Bessel equation

z ; (z) + ± Z ' p(z) + ( l - £ ) Zp (z) = 0 . (4 )

Notice also that the constant p in the expression (2 ) can have any
value. We have taken p as an integer since we wanted a solution with
a period of 2.-r with respect to the variable <p. We also want our solution
to remain finite when r = 0 and therefore we took as Zp(z) that
solution of the equation (4) which remains finite when z = 0, i.e. we
took the solution J p(z) (p > 0), which is a Bessel function. The value
of the constant k, and hence of cofrom (3) was determined from the limit
538 SPECIAL FUNCTIONS [143

condition. Later we shall deal with the applications of Bessel functions.


At present we shall study the properties of the functions satisfying the
equation (4) and we begin with the above Bessel function.
The Bessel function, except for the constant term, is determined
by an expansion of the form [II, 48]:

Jp (2) = CzP[* — 2(2p + 2) + 2■4 •(2p+ 2) (2p + 4) “ •••]• (5)

When p = n is a positive integer or zero, then the constant factor C


is equal to 1/ 2“ n I and, as always, 0 I = 1 . Hence we have the following
expression for a Bessel function with a positive integral subscript

'.W -life S jrftr- <6>


When p is not an integer then we take the constant term in formula
(5) to be equal to
n ____ [____
2PT(p + 1)
and we obtain the following expression for the Bessel function:

J p (2 ) = 2PT(p + 1) f
1~ I I (P + 1) ( t ) + 2 ! ( p + l ) ( p + 2) ( t ) “ •••]

or, as a result of the fundamental property of the function r(z):

w - l w & n i r '
When p = n is a positive integer the latter formula is the same as
formula (6 ). Consider now formula (7) when p is a negative integer
p — —n. We know that r(z) tends to infinity when z is a negative
integer or zero. Hence in the expansion (7) all terms will vanish in
which the argument of the function r(z) in the denominator is equal
to a negative integer or zero. These terms will correspond to the follow­
ing values of the variable of summation:
— n -f k + 1 < 0, i.e. k < n — 1.

In other words we have to start the summation from k = n:

J ( g ) = ^ ( - 1 )* ( z y n + 2k
^ M T { — n + k + 1) \ 2 I
144] RELATIONSHIPS BETW EEN BESSEL FUNCTIONS 639
Replacing 1c by another variable of summation I = k — re and
taking (—l)n outside the summation symbol we have

•L-. (*) = ( - 1)" j2 ( i + i r l i + i) ( t ) >

i.e.
■/_„ (z) = (— 1)nJ n (2) (re being an integer) (8 )

In other words the Bessel function with a negative integral subscript


(—re) differs only by the factor (—l)n from Bessel a function with a
positive integral subscript.
When p is not an integer then the functions J p(z) and <7_p(z)
will be two linearly independent solutions of the Bessel equation
[II, 48]. The series (7), as we know, converges for all finite z’s.

144. Relationships between Bessel functions. We shall now deter­


mine certain fundamental relationships between Bessel functions with
different subscripts. Differentiating the power series (7) we have:
d Jp(2) d (-1)* z2* “ (—l)k • 2k Z2ft-i
dz zP ~ dz 2 ^ k\T(p + k + 1) 2P+w ~ Jc\ T(p + k + I) 2P+2fc ’

or, substituting the variable of summation k by k + 1 and starting


the summation with k = 0
d Jp(z) _ v (-l)* +12(fc + l) z2k+1
dz ZP (k+l)'.r(k + p + 2) ' 2P+2A+2

d_ Jp(z) = _ _1_ “ (-1)* / z *\p+i+2/c


dz zP zP 2 ^ fc! 71(p+ 1+ fe+ 1) [ 2 J

Comparing this with (7) we obtain the following formula:


d Jp (2 ) J p+i (2 )
dz zP zP (9)
Carrying out the differentiation the fraction we can rewrite this
formula as follows:
pJp (g) ( J '( 2) = _ J l(2)). (10)
J p (2 ) — — J p + 1 (2) + z
540 SPECIAL FUNCTIONS [144

Divide both sides of the formula (9) by 2 :


1 d Jp(z) Jp+1 (z)
Z Az ZP ZP+1

The above relationship can be formulated as follows: the differen­


tiation of the fraction J p(z)lzp and its subsequent division by z is
equivalent to the addition of unity to p and to the change of sign
of the above fraction.
Using this rule several times over we obtain the following formula
which is valid for any positive integer m:

dm Jp (Z) __ ,_ . Jp+m (z)


( z d z)m zP ' ' zP+nx
(11)

This formula can be rewritten as follows:

dm Jp {%) _ / _ 1 \m Jp+m (g) ( 12)


d (zt )m zP ~ ' ’ 2m zP+m ■

Differentiating now the product zpJ p(z) with respect to 2 :

( - l ) * 2 (p + fc) Zlp+lk-l
= k2= 0 *!r(p + * + l) 2P+2fc ’

or, bearing in mind that r(p + k -f- 1 ) = (p + 1c) r (p + Jc) we have

T M - * ^ (-D * f Q p - l +»
2 k'.r(p- 1 +k + 1 ) I 2 J
dzZ Jp{z) —z fc=0

i.e. from (7) we obtain a formula which is analogous with the formula
(9):
A zPJp (z) = zPJp_l (z). (13)

Differentiating the product we can rewrite this formula as follows:

J'p (z) = J p_l{z ) - (14)

Divide both sides of formula (13) by 2 :

— zpJ p {z) = ZP-1 J P- 1 {Z).


144] RELATIONSHIPS BETWEEN BESSEL FUNCTIONS 641

Applying this formula several times over we obtain a formula which


is analogous with formula (11 ):

( 15)
or

In the formulae (11) and (15) we have used the following notation:
dm t d d d ,, «
(z dz)m ' W ~ z d « 2 d2 ‘ ‘ 2 dz ' W '

where the number of differentiations with respect to z and of sub­


sequent divisions by z is equal to to.
Comparing the formulae (10) and (14) we obtain a relationship
between three successive Bessel functions

~ J P+1 (*) = J P- 1 (*) - ^ L


or
= J ^ ) + J p+l (z). (17)

Using the above formulae we will show that Bessel functions, the
subscripts of which are equal to one half of an odd integer, i.e.
± ( 2to+ l )/2 where to is an integer, can be expressed in terms of elemen­
tary functions. To show this we consider formula (7) when p = 1/2:

^ (-D * +
J i(z)
*=® A!r(fc + - |) '-2 -'

Applying the fundamental property of the function r(z) several


times over we have

~ ( * + t ) ( * - t ) - 4 ^ ) = i a + m “ 7,1 1 ' 3 1 r*.


and we therefore obtain
l +ik
J (2) = y ( ~ D* _____ 1/2 ", (—1)*22*+1
£ 2"0 i ! 2 * - l - 3 . . . (2fc + 1 ) ^ I ™ (2fe+ 1)!
642 SPECIAL FUNCTIONS [146

i.e.
■m * ) = y ^ - sinz- (is)
2 r

Applying now formula ( 11 ) we have for any positive integer m:


r— 2m+l

^ w - t - i r ] ( »)

Analogous results are also obtained for negative subscripts. Formula


(7) when p - —1/2 gives

J _ i ( 2) COS2’ (20)
2

and using formula (15) we obtain for any positive integer m


, 2/71 + 1

< 2 1 >

In [II, 48] we have written Bessel functions in an expanded form


when p - ±3/2 and p = ±5/2.

145. T h e o r t h o g o n a l i t y o f B e s s e l f u n c t i o n s a n d t h e i r z e r o s . As we
have already said we have used Bessel functions in connection with the
vibration of a round membrane. At the time we used the usual Fourier
method and, in order 10 satisfy the initial conditions of the problem,
we had to expand the given function into a series of Bessel functions.
We then obtained series, analogous with the Fourier series and found
that Bessel functions have the property of orthogonality [II, 178] in
the usual sense. We shall now consider this problem from a more
general point of view and explain some additional circumstances.
As we know the function J p(kz) satisfies the equation [11, 48]:
dVp(te) , 1 dJp (fcz) sL) j p (kz) = 0,
dz2 + z dz + ( z2
or, multiplying by z we can write this equation as follows:
_d_ dJ p (kz)
= 0 .
dz dz

In future we shall assume that the symbol p is real and also that p > 0.
145] THE ORTHOGONALITY OF BESSEL FUNCTIONS AND THEIR ZEROS 543
We take two different values of k and write the corresponding diffe­
rential equations
4 ^ d j + ^ = 0>

4 [ ’ ^*5^] + ~ ~ )■I , <*.»> - »■•


Multiplying the first equation by J p{k%i) and the second by J p(k1z),
we subtract and integrate over the finite interval (0 , 1):

i \ J , < M 4 [* - j , (t. *> - 4 [* * +


0
I
+ ikl — kl) J zJp {k1z ) J p (k2z)dz = 0.
o
The first integrand represents the complete derivative with respect
to z of the difference
_d_|-2 d J ^ z ) Jp {h z) _ 2 dJ ^ z ) Jp {h g)j (

and the equation can therefore be written ill the form

[» ^ j , <*, >) - ^ j , f t *>j , : i +

+ (* 1 — * 2) I z J p (K Z) J p (* 2 2 ) dz = 0 .
But
^ ± = lcJ'p {kz),
where we write

•W -i-M * ).
and consequently, the above formula can be rewritten in the form
[k1z J 'p {k1z ) J p (k2 z) - k 2z J p, (k2z ) J p {k1z)]zz 'J0 +
1
+ {k i - kD J z Jp i.k i z ) Jp ( h z) d2 = 0 . (22)
0

Let us recall the expansion of Bessel functions


“ (_ l)fc g2k
J P ( 2 ) = zP y , k i r ( p + k + 1) 2 p +m •
(23)
k=0
544 SPECIAL FUNCTIONS tU5
Owing to the fact that p > 0 the term in the square brackets
vanishes when 2 = 0 and we finally arrive at the following formula
which is of fundamental importance to what follows:

I [&1 J p ( ^ 1 1) J p (^ 2 I) — &2 J p (^ 2 Q J p (^ 1 ^)] ""t""


l
+ {M - *1) J z J p (*1 2) Jp {k2z)dz = 0. (24)
0

When I = 1 this formula becomes:

Jp (^i) Jp (^2)^2 ^p (^2) ^p (^1) +


/
+ m — k\) J zJp (*1 z) Jp (h z) dz = 0 . (25)
«
In the above calculations we have assumed that p > 0 . It can readily
be shown that the integrals retain their meaning and the term in the
square brackets in formula (22 ) vanishes when z = 0 when we make
the even wider assumption that p > —1 .
We shall show, first of all, that the Bessel function cannot have
complex zeros. Suppose, to start with, that it has such a zero a -f- ib
where a ^ 0. All the coefficients in the expansion (7) are real and,
consequently, the function J p(z) must have the conjugate zero a — ib
as well as the zero a + ib. If we now suppose that in formula (25)
fcj = a + ib and Jc2 = a — ib, where # A w e have
1
j" zJp {kl z) J p (k2z) dz = 0 .
0

J p(k1z) and J p(kg) will have conjugate complex values and there­
fore, the integrand in the above formula is positive, so that this
formula is contradicted. The case when a — 0 remains to be con­
sidered, i.e. it must be shown that the function J p(z) cannot have
purely imaginary zeros either. In fact, substituting in the
formula (23) we obtain an expansion with positive terms

Jp (ib) = iib)P2 ~k\r(p + k + l) ~2 P+2*" •

This is directly due to the fact that, according to the formula (111)
in [71] the function r(z) is positive when 2 > 0. We thus arrive
at the following result: if p is real and p > — 1 then all zeros of the
function J p(z) are real. Notice also that it follows directly from the ex-
145] THE ORTHOGONALITY OP BESSEL FUNCTIONS AND T H HI H ZEROS 545

pansion (23) which contains even powers only that the zeros of J p(z)
will in pairs have equal absolute values of opposite sign so that it is
sufficient to consider positive zeros only. In future we shall only con­
sider these zeros. Let us write the asymptotic representation for the
Bessel function [113]:

J p (2) = f i cos [z ~ "IT ~ t ) + 0 2)


or

J p (2) = [cos [z ~ -IT ~ t ) + 0 (3-1) ] '


As z moves towards infinity along the positive part of the real axis
the second term in the square bracket tends to zero while the first term
passes from — 1 to + 1 an infinite number of times. It follows that
Jp(z) has an infinite number of real zeros.
If z = kx and z= k 2 are two different positive zeros of the equation
J„(zl) = 0 , (26)
then formula (24) gives directly the following property of orthogonality
of Bessel functions
i
J zJp (fca z) J p (k2 z) dz = 0 . (27)
o
According to Rolle’s theorem, the function J'p(z) must also have an
infinite number of real positive zeros, and if we now denote by kx and
k2 two different positive zeros of the equation
J rP (zl) = 0 , (28)
then, from (24), we obtain exactly the same condition of orthogonality
(27).
Let us now consider a more general equation than the one above,
viz. an equation of the form
a J p (zl)+ p zJ 'p (z') = 0, (29)
where a and /? are given real constants. Let z = kx and z = k2 be two
different zeros of the equation (29), i.e.
a J p (k, I) + 0 k, J ’p (M ) = 0 J a J p (k21) + f$k2J rp (k21) = 0.
It follows directly that
K J'p (k, i) J p (k21) - k2j ; {k21) j p (kx i) = o,
646 SPECIAL FUNCTIONS [146

and, consequently, the term in square brackets in formula (24) will


again vanish, and we have the same condition of orthogonality as
before. The equations (26) and (27) are particular cases of the equa­
tion (29). As before, it follows from the conditions of orthogonality,
that the equation (29) cannot have complex zeros a + ib, where
a ^ 0.
It can be shown similarly that the equation (29) has no purely
imaginary zeros provided a > 0 and /? > 0 .
Let us recall the two relationships

(30)

According to Rolle’s theorem, the first of these shows that between


two successive zeros of J p(z) there can only be one zero of J p+1(z). The
second relationship shows that between two successive zeros of J p+1(z)
there can only be one zero of J p(z). A comparison shows that the posi­
tive zeros of J p(z) and J p+1(z) separate one another, i.e. between two posi­
tive zeros of J p(z) there is one and only one, zero of J p+l(z) and vice versa.
Let a and b be the smallest positive zeros J p(z) and J p fl(z) respec­
tively. Remembering that z = 0 is a zero of zp+1 J p_ , (z) and applying
Rolle’s theorem to the second of the formulae (30) we can see that
J p(z) must have a zero in the interval (0 , b), i.e. a < b .
We thus find that the smallest positive zero of the function J p(z) will
be nearer the origin than the similar zero of J p+l(z). Notice also that
the function z~pJ p(z) is a solution of the equation [111]:

z^ - + (2 p + 1) ^ + zy = 0’
and therefore the functions z~p J p(z) and d/dz \z~pJ p(z)'\ cannot have
common positive zeros [104]. Consequently, from (30), the same can be
said about the functions J p(z) and J p+1{z).
The orthogonality of Bessel functions is of great importance in the
expansion of a given function into Bessel functions as, for example,
in the vibration of a round membrane.
It is also essential to be able to evaluate an integral of the form
i
J" zJf, ( kz) d z,
o
where z = k is a zero of an equation of the form (29). Consider the case
when k is simply a zero of the equation (26). Take the formula (24)
145] TH E ORTHOGONALITY OF BESSEL FUNCTIONS AND T H E IR ZEROS 547

and assume that k2 = k, while k2 is variable and tends to k. We thus


have

(*! + *) \ z J p (K 2) J p m d* = l± J ' o f ^ P } k ^ .
o
When ib2—>-k both the numerator and the denominator of the
fraction vanish since J p(k1l) tends to J p(kl) = 0 . Expanding by the
usual rule we have in the limit:
i
2k 1 2«7j5(kz) dz = I2kJp2 (kl)
o
or
I
\ z J 2( k z ) d z = ^ - J '2(kl). (31)
0
Now let us consider the known relationship
d JP(g) _ _ JP+i (2)
dz zP zP

and put z = kl. AVe have


J'p(kl) = - J p+1(kl),

so that the above formula can also be written as follows:


1
[sJp (kz) d2 = — J 2P+1 (kl) . (32)
6

When z — k is a zero of the equation (28) we obtain similarly

J zJ2 (kz) d 2 = - — j ; (kl) J p (kl) . (33)


0

But we have:

j ; m + — j ' p( ki) + (i - j p (M) =o


and using the equation J p(kl) = 0 we can rewrite the formula (33) as
follows:
/
\ z J 2(kz)te = ^ - ^ ) j 2(kl). (34)
548 SPECIAL FUNCTIONS [146

146. Converting function and integral representation. Consider the


analytic function of the complex variable t
i / _
■ (35)
This function has essential singularities at the points t = 0 and
t=oo and it can therefore be expanded into a Laurent’s series in the
whole Z-plane; the coefficients of this expansion will be functions of
the parameter z in expression (35):

e’Z('~ 7) = +
j? a n (z)t ". (36)

We will now show that these coefficients will be the Bessel func­
tions J n(z). In fact we can represent the coefficients of the expansion
(36) by the follow’ng contour integral [15]:

a n (z ) = ~ 5 r i \ u ~n_1 ~ du’
t.
where Z0 is any simple closed contour which encircles the origin in the
positive direction. Wc now replace i by another variable of integration,
according to the formula u = 2l/z, where z has a fixed value other
than zero. The point u = 0 corresponds to t = 0 and the contour Z0
will be transformed in the Z-plane into another contour which also
encircles the origin in the positive direction. Changing the variables
we obtain the following expression for the coefficients:

On the contour Z„ we can represent the exponential function by a


power series which converges uniformly with respect to t:
*2k
e- 4i =
kl 2lk tk
ami
Substituting this in the previous formula we obtain
t—n—k—l nt dt .
/(=o v ' i
When n + k is a negative integer then the point t — 0 will not
be a singularity of the integrand in the above formula as this will
146] CONVERTING FUNCTION AND INTEGRAL REPRESENTATION 549
be equal to zero. When (n -f- k) is a positive integer or zero then,
remembering the expansion of e*, we can see that the residue of the
integrand at the point / = 0 is equal to lj(n -f k) !. Thus when n
is a positive integer we have
_ ^ ( - 1)* i M n+2fc
an ^ ^ Jfc! (w+ ife)! I 2 ]

i.e. an(z) is, in fact, the same as J„(z). If we replace t by —1ft in


formula (36) then the left-hand side remains unaltered, and this shows
that a_n(z) = (—1 )" an(z), i.e. when n is negative we have from (8 )

°-n (z) = (— l)n iz) = J-n (z) ■


Hence, instead of the formula (36), we can write the following ex­
pansion:
e2z( '~ ) = y j n (Z) tn. (37)
n= —oo

In other words the function (35) is the converting function for Bessel
functions when n is an integer. Formula (37) is convenient for
deriving the properties of Bessel functions when n is an integer. We
shall use this function for deducing the integral representation of
Bessel functions when n is an integer.
On putting t = e,<p in formula (37) we have

g fiz s i n ? __
2 Jn
n=a—OO
or, separating the real and imaginary parts, where we assume that z
and q>are real:

cos (z sin <p) = J 0 (z) + J n (z) cos n<p + Jf? J n (z) cos nq>,
n=1 n - —1
00 —
cs
sin (z sin <p) = ^ J n (z) sin n<p + ^ J n (z) sin n(P<
n= 1 n =—1

or, from (8 ), we have:

cos (z sin <p) = J 0 (z) + 2 J p J ln (z) cos 2n<p,


n=I (38)
OB

sin (zsin<p) = 2 ’y j 2n_1(z) sin (2n — 1) q>.


n=1
550 SPECIAL FUNCTIONS [146

The formulae (38) represent the expansion of the functions into a


Fourier series; applying the usual method for determining the coeffici­
ents we obtain the following integral representation for Bessel functions:
ft

J 2n (z) = -i- Jcos (zsin (p) cos 2n<pd<p (n = 0, 1, ...) ,


o
n (39)
J 2n—\ (2) = ~ Jsin (z sin 9?) sin (2 n — 1 ) <pd<p (n = 1 ,2 , . . . ) .
o

The same method for determining coefficients gives the following


two equations:
ft

-i-J cos (zain<p) cos (2 n — 1) <pd(p = 0 ,


o
n
n J sin (z sin <p) sin 2719?d^ 0 .
0

The formulae (39) can be combined into one formula which will be
valid whether n is even or odd. Consider in connection with this the
integral

z sin 93) d 9>=


ft ft

= — J cos (z sin 93) cos n<pdi<p sin (z sin 93) sin n<pArp.
0 b
When n is even, the first term on the right-hand side is J„(z) and the
second is zero so that the sum is equal to J n(z)- When n is odd, the
first term will be zero and the second term will be J n(z) so that for
any positive integer n we have the integral representation
71
J n (z) = — cos (n<p — z sin 93) d93 (n = 0, 1, 2, . . .). (40)
b

Strictly speaking the proof of the above equation is only valid


when z is real. As a result of the principle of analytic continuation we
can maintain that it is also valid for every complex z. Bearing in mind
146] CONVERTING FUNCTION AND INTEGRA! REPRESENTATION 551

the fact that the integrand is even we can write the above formula as
follows:
n
(z) = J cos (nip — z sin 93) . (41)'
—71
This equation can also be written in the form:
71
J n (z) = — [ e,<"*’- zs,n*’>d9>. (42)
— Jt

In fact, applying Euler’s formula to the exponential function we


obtain two terms, one of which is equal to the integral (41) and the
other being zero since the integrand function is odd.

Notice th at the formula (40) no longer applies when n is not an integer.


In this ease we have a more complicated formula, viz.:
n ««*
J p (z) = — J cos (7up —z sin <p) — Sin Vn J*e—p?—z sin h <pdgj. (4 3 ),
0 0

and this formula is valid for values of z which lie to the right of the imaginary-
axis. Notice also the formula for sinh <p:
eV-e-f
sinh <p = -----g----- •

The proof of this formula will be given in [151].


Applying the formula (37) and using the obvious equality

we have

2 J n (a + b)tn = 2 Jk( a) t k- 2
n*» —— k = —<~ ft-*—*

Multiplying the power series on the right and collecting terms in f 1 we


have
+°°
J„{a + b ) = 2 J k ( a ) J n - k ( b )- (4 4 )

This formula expresses the addition theorem for Bessel functions when n is
an integer.
When n is zero a more general addition theorem must be applied, viz.:

Jo (/o* + 62 + 2a6 cos a ) = J 0 (a) J 0 (6 ) + 2 2 Jk (°) Jk (&) cos ka. (45)


k= 1
550 SPECIAL FUNCTIONS [146

The formulae (38) represent the expansion of the functions into a


Fourier series; applying the usual method for determining the coeffici­
ents we obtain the following integral representation for Bessel functions:

J 2„ (z) = -i- j"cos (z sin <p) cos 2n<pdq> (n = 0, 1, ...) ,


'o
n (39)
J 2n—i (z) = -i-Jsin (z sin <p) sin (2 n — 1) <pd<p {n = 1, 2 , . .
o

The same method for determining coefficients gives the following


two equations:
71
-i-J cos (z sin (p) cos (2n — 1 ) q>d<p = 0 ,
o
n
n J sin (z sin <p) sin 2n<p dip 0 .
o

The formulae (39) can be combined into one formula which will be
valid whether n is even or odd. Consider in connection with this the
integral
n
j" cos (mp — z sin <p) dip =
o

When n is even, the first term on the right-hand side is J„(z) and the
second is zero so that the sum is equal to J n(z). When n is odd, the
first term will be zero and the second term will be J n(z) so that for
any positive integer n we have the integral representation
n
J n (z) —' ~ [ cos (n<p — z sin <p) dip (n = 0, 1, 2, . . .). (40)
o

Strictly speaking the proof of the above equation is only valid


when z is real. As a result of the principle of analytic continuation we
can maintain that it is also valid for every complex z. Bearing in mind
146] CONVERTING FUNCTION AND INTEGRA!. REPRESENTATION 651'

the fact that the integrand is even we can write the above formula as
follows:
n
J „ (2) = J cos (n<p — 2 sin <p) dep. (41)'
— 71

This equation can also be written in the form:


71
J n (2) = — [ e'<n*-*,l,,*>d9>. (42)
— 71

In fact, applying Euler’s formula to the exponential function we


obtain two terms, one of which is equal to the integral (41) and the
other being zero since the integrand function is odd.

Notice that the formula (40) no longer applies when n is not an integer.
In this case we have a more complicated formula, viz.:
n «o
Jp (z) = -i- J cos (nip — 2 sin <p) dip — ~ ~~ ~ J e-P?- *sinh <pdp. (43}
0 0

and this formula is valid for values of z which lie to the right of the imaginary
axis. Notice also the formula for sinh <p:
e* —e- 9
sinh ip = --------- .
it

The proof of this formula will be given in [151].


Applying the formula (37) and using the obvious equality

>K)
we have
2 J n (a + b)tn = 2 2 J k (b)tk.
n= —— k= —- fc=—-

Multiplying the power series on the right and collecting terms in tf1 we
have
+°°
J n (a + b) = 2 Jk(a)Jn-k(V- (44)
k= -~

This formula expresses the addition theorem for Bessel functions when n is
an integer.
When n is zero a more general addition theorem must be applied, viz.:

J a (/a 2 + b2 + 2ab cos a ) = J a (a) J 0 (b) + 2 2 Jk (“) Jk (6) cos ka. (45>
552 SPECIAL FUNCTIONS [147
147. The Fourier—Bessel formula. Arbitrary functions which are defined in
the interval (0, oo) and which satisfy an additional condition in this interval
can be represented by an integral, analogous with the Fourier integral but
containing Bessel functions instead of trigonometric functions, viz.: if
J(q) is continuous in the interval (0, oo) and satisfies the Dirichlet condition
[II, 143] in any finite interval and if the integral given below also exists

J"e 1/ (e) |de.


0

then for any integer n and q> 0 the following formula applies:

flQ) — S*J n (6Q) d* I */(0 J n (8t) & * (46)


0 0

We shall give the formal proof of the relationship (46) without going into
greater detail. Supposing th at q is the radius-vector, we introduce polar co­
ordinates and apply to the function
X = Q COS < f
9 (■*. y) = f (e) 0,r"p (47)
y = Q & m < p

the Fourier formula [II, 160], changing the order of the inside integrals:

+ "»+•• +«“ +«*»


g (x, y) = J J e'(“x+ty) du dt> J J g (£, ij) df dt).

We replace the variables (u, v) and (£, tj) by the polar coordinates
f = 8 cos a ; u = t cos /J;
?; = s s m a ; « = tsin/S.
Using the formula (47) we can write:
■* 71 • A
f (g) elnip — —jjj- J t di j* e '# 'C03^ - W d ^ j 8f(8) ds J e'»«e-Wcos<a- « d a .
0 —7i 0 —n

Replacing ^ by a new variable of integration /S' according to the formula

0 - = - + 0 ',
we obtain
n
(fe)einv = - ± ^ j t d t J e- '« ,sln*'d)3'Js/(*)ds J efnae " /S' C0S(a“ <P“ iJ'" ^ d a .
0 3* 0 —ji
9
148] TH E HANKEL AND NEUMANN FUNCTIONS 553

Bearing in mind the periodicity of trigonometric functions wo can change


the interval of integration to the former interval ( —jt, +Jt). Similarly, replacing
a by a new variable a', according to the formula
“ — <P— P' = a',
yjQ obtain
oo 7% » n
f(e) einv = 0
J e - ,efsin/9'+‘^'d/3'Js/(5)d« J e- ,s,slna'+'na'da',
—n 0 —7i
which, with (42), gives the formula (46).
When a function is given in a finite interval (0,1) then instead of the formula
(46) we can consider an expansion into a series, analogous with the Fourier series,
in terms of orthogonal functions with which we dealt in the previous section.
Notice th at formula (46) can be proved for all real n ’s greater than (— 1/2)
and also when less strict assumptions are made with regard to the function
/((?)•

148. The Hankel and Neumann functions. We obtained in [ 112 ] two


solutions for the Bessel equation:

4 5 -+ f£ + (i-& )— <«>
which were given by the following formulae
r [ — — p\ !
(z) = — ^ ------ ( | ) Pj V ~ l)P" 2 e,zr dr,
n2i
1 (49>

Hf{z) = - ( | ) p j(T* - l)p-^ e'»dr.


ji2 i
In these formulae the integrand is single-valued in the t-plane
cut parallel to the imaginary axis from r = ± l t o r = +*' °° viz.
wo assume in the first of these formulae that arg (t 2 — 1 ) = 0 when
t > 1 and in the second that arg (t 2 — 1 ) = 2n when r > 1 . On going
from the line ( 1 , + ) on the real axis to the line (—00, —1 ) through the
lower half-plane, and thus describing half the circuit round the points
t = ± 1 while omitting the cuts, the amplitude in the expression
(r2 — 1 ) = (t — 1) (r + 1) increases by ( —2n), in other words, in
the second of the formulae (49) we can assume that arg (r 2 — 1 ) = 0
when x < —1 . Formula (49) gives the Hankel functions for values
of z to the right of the imaginary axis, i.e. the real part of which is
greater than zero. Notice also that the integrand in the integrals (49)
■564 SPECIAL FUNCTIONS [148

is an integral function of the parameter p for a fixed value of z


and, bearing in mind its rapid decrease at infinity, we can maintain
that the Hankel functions H ^(z) are also integral functions of the
parameter p for fixed values of z. It follows directly from the asymp­
totic expressions for Hankel functions [112] that these functions are
two linearly-independent solutions of the Bessel equation. We also saw
that the Bessel function is equal to half the sum of the Hankel functions
7 f l (p1) (*) + H {?>(*)
J p \ z ) — 2 (50)

There is a close connection between the Bessel equation (48) and


"the equation
J g L + p2w = 0, (51)

which is satisfied by the usual trigonometric functions cos pz and sin pz.
The Hankel functions are in this case analogous with the solutions
eipz and e~ipz while the Bessel functions J p(z) are analogous with the
solution cos pz of the equation (51). Let us also consider the solution
of the equation (48) which is equal to the difference of the Hankel
functions divided by 2i:
H $ \ z ) - H f (z)
N P(*) = (52)
2i
This solution, usually known as the Neumann function, is analogous
with the solution sin pz of the equation (51). From the formulae (50)
and (52) we obtain directly the following expressions for the Hankel
functions in terms of the Bessel and Neumann functions:
{z) = J p (z) + iN p ( z ) ; H f (z) = J p (z) - iN p (z). (53)
This shows that the functions J p(z) and Np(z) determine two linearly
independent solutions of the equation (48).
The Hankel functions have the following asymptotic representations:

-
2 A z - ip - d [1 + 0 (*-!)],
(54)

(z) = |/ [1 + 0 (*-!)],
which we have proved for z > 0. Using formula (50) we can, as in
[113], obtain the asymptotic representation for the Bessel function

Jp(z) = f ^ - [ c o s ( z - ^ p + 0 ( z - 1)], (55)


148] TH E HANKEL AMD NEUMANN FUNCTIONS 655

and similarly, using formula (52) we can obtain the asymptotic repre­
sentation for the Neuman function when z > 0

n p (*) = f i [sin (z ~ ^ ~ t) + 0 • (56>


In all the above formulae it must be assumed that z > 0 and the
positive radical must be taken.
We will now introduce a formula which expresses the Neumann func­
tion in terms of the Bessel function. To start with we consider the case
when p is not an integer. We know that the equation (48) then has two
linearly independent solutions J p(z) and J - P(z). The second of these
can be expressed linearly in terms of the solutions J p(z) and N p(z),
which as we have already said above, are also linearly independent
solutions, i.e. we can write
J _ p (z) = C1J p (z) + C2N p (z), (57)
where C±and C2are constant coefficients which we shall now determine.
Bearing in mind the asymptotic expressions (55) and (56) we can write

cos (z + - - ) = ClC0S [z - - -) +

+ C2sin(z — - { J + O j O (z-1) + C20 (z-i).

Notice that the product of a constant, or of any bounded function,


and 0(z~1) of the order 1/z also gives 0(z-1) of the order 1/z. We thus
obtain

= 0 , cos (2 - -E- - A) + 0 , Bin(2 - - f - - -J) + 0 (2- 1) . (58)

From this the values of the constants can be deduced by comparing


the principal terms in the above expansions. In fact, let us suppose that
Ct = cos pn — ; C2 = — sin pn — A 2,
where Ax and A2 are the new unknown constants. Substituting in the
formula (58) we have

cos(z + - -J) = cos (z + - | ) - A.coa (z - - -J) -

- A 2sin (z - - —) + 0 (z-1)
656 SPECIAL FUNCTIONS [148

or
Ai cos (z - - —j + A 2 sin (z - - —j = 0 (z~l ),

i.e. the left-hand side of the above equation which is a periodic function
of period 2n must tend to zero as z —*■+ ° ° . It follows directly that
Ax = A2 = 0, i.e.
C1 = cos p n ; C2 = — sin p n .
Substituting these constants in the formula (57) and solving it with
respect to N p(z) we arrive at the required formula which expresses the
Neumann function in terms of the Bessel functions:
Jp(z) COS pjl— J_p (2 )
N p {z) sin pn (59)

The Neumann and Hankel functions are integral functions of the


parameter p. Formula (59) is valid as long as p is not an integer.
When p is an integer the denominator in formula (59) vanishes. But,
from (8), the numerator will also be zero. Hence to obtain the value
of the fraction (59) when p is an integer we simply have to abolish the
indefiniteness and replace the numerator and the denominator by their
derivatives with respect to the parameter p and assume subsequently
that p is equal to an integer n:
9J p (z) . .. .
— ' — cos pn — nJ„ (z) sin vti -------—
9J _ p (s)
op H op
N n (z) = n cos pn p=.n
We thus obtain the following expression for the Neumann function
with an integral n:

(60)

Substituting the expression (59) in the formulae (53) we obtain


formulae in which the Hankel functions are expressed in terms of
Bessel functions in which p is not an integer:
J p (z)e lp7l — J - p (z)
B$>(z) = sin pn
(61)
H(p2) (z) = ^Jp(z) elpn —J-p (z)
sinpjr
This leads us directly to the following relationship between Hankel
functions, the symbols of which differ only by the symbol
H<J) (z) = e'pjI (z); Hty (z) = e-"* <?>(z). (62)
148] TH E HANKEL AND NEUMANN FUNCTIONS 557

Strictly speaking this formula can be proved on the assumption that p


is not an integer. But the left and right-hand sides of the formulae (62)
are integral functions of p and therefore the formula holds for any p.
When p is an integer both the numerator and the denominator in the
formulae (61) vanish. Abolishing the indefiniteness, as before, we can
also obtain a formula when p — n is an integer.
Consider lastly the case when p has the form p = (2m -f- l)/2, where
m is a positive integer or zero. If we substitute this expression for p
in the formulae (49) which determine the Hankel functions, then the
integrand will be regular in the whole plane including the points
r = ± 1 and therefore the integrals will be equal to zero. But at the
same time the factor /'(1/2 —p) becomes infinity and the formulae (49)
will be devoid of meaning. Instead of these formulae we take the
expansions (195) and (196) from [112]. These expansions are, in
general, divergent but they do formally satisfied equations as we
have proved before. In the case under consideration they will not
only be convergent but they will simply become finite sums and give
Hankel functions in the finite form. Consider, for example, the first
Hankel function with the value of p = (2to + l)/2 :
__ i (,_<»!+«;
•Him+l (z) = f r (to + f)

or

(Z_ q n ^ )
2 m ( m — I) . . . (m — fc -f-l)
nz ml 2 Fl (”* + *)’ ( i )
fc=0
This shows that all terms corresponding to k > m + 1 will vanish
and we obtain the following formula for the Hankel functions
(m + l)n
----
9. bl(2
V )
ml I r a < " .+ * > '& )* • («)

Similarly, for the second Hankel function we obtain the following


finite formula:

H?2
2m + l
2
9 » ' Z /
Ira (M
>
558 SPECIAL FUHOTIONS [148

The formulae (59), (61) and (62) remain valid for the values
p = (2m l)/2. Notice that formula (61) determines the Hankel
functions when p = (2m + l)/2; from (19) and (21) we have:

so that we can finally write

H su1w = ( - i r + H ^ z ‘ -j^ ir(-V ) (•»)

and similarly
# — 2m +l —/?

The expansions (63) and (64) can also be obtained from this. Using
(61) when p = 1/2 and also the expressions

J i (*) = sm
z; J i ( 2) = f c c°s z .
2

wo have
H P (z) = - •• f ~ e '* ; H f (z) = i jf-

A number of relationships which we proved earlier for Bessel func­


tions can also be easily proved for Hankel functions. We give here
some of these relationships:
dm , i \mH*p+m (z). ( H $ h z)\ H $ m (z)
(z dz)mv zP J ' ’ zP+m ’ (z dz)m( zP J ' ' zP+m ’

— -^ (2) = (*)+ # f t i (2); — H P (z) = H fl, (z) + H % (z).


It follows from the definition of J p(z) that J p(z) and N p(z) are
real while //f t (z) and Z/ft (z) are complex conjugates when p and z
are real.
149] TH E EXPANSION OP TH E NEUMANN FUNCTION W ITH INTEGER SUBSCRIPT 559

149. The expansion of the Neumann functionn with an integer subscript. When
the subscript is an integer the solutions J n(z) and «7_n(z) will be linearly depend­
ent and we can take N n(z) for the second linearly-independent solution. I t is
therefore interesting to deduce an expansion for this solution which will hold in
the whole plane. According to the general theorem of Fuchs this expansion,
apart from integral powers of z, will also contain terms in log z.
To start with we shall explain certain formulae which refer to the function
r(z). We obtained for this function the following infinite Weierstrass product

((7 = 0.57 ...),


F(z)

where C is Euler’s constant. We know from [68] that the logarithmic derivative
of this product can be obtained in the same way as for a finite product. Hence
r'(z)
r(z) - T + c + i ( r i r - : 9 ;
and putting z = re, where re is a positive number, we have

r (n) 1 I____
r (re) re I. re+ k kJ

= _ ¥ “ c + ( T " ^ ) + ( T “ ^ T 2 ') + ( i ~ ¥ T l ) +

-T'(re) 1 . .. + 1 - C (re = 2,3, ...),


r (re) re — 1 ‘ re — 2

We also have r(re) = (re — 1)! and therefore:


d 1 r'(t) 1 ( 1 1 , , (6 7 )
dtr(t) (<-i)!^-i+ t _ 2 + +1 °)
also, when t = 1, we have F (l) = 1 and .T'(l) = —C, and therefore
d 1
dt r(t) ■c ( t = i). ( 68 )

Let us now consider the case when t is a negative integer or zero. We know
that r(z) has a pole of order one with a residue ( —1)n/n ! at the point z= — re,
i.e. in the neighbourhood of this point we have the expansion

F(z) =
(-1)" + “o+ °i (s + w) + ...
re! (z + re)
1 z + re
1)" re!
F(z) '(- *' l + f t ( z + ») + ft ( * + « ) » + . . .
By performing a simple differentiation we obtain directly
d 1
= (—!)" re! (re = 0, 1,2, ...). (69)
dt r (t)
5 60 SPECIAL FUNCTIONS [160

L e t us now tr y to fin d th e expansion o f th e solution N n(z) w hich is given by


th e form ula (60). W e h av e

'-M-GTi-Wr fc-0 ' ' ■ n ± p + * + 1)

and, differentiating w ith respect to th e p ara m ete r p , we o b tain :

3Jp (z) . z 7 / \ i C2 Y ^ l)k ( 2 \2k( d 1 )


dp ~ log 2 J P (z) + { 2 ) 2 k\ 1 2 J ( d t 7’ (t)J<-p + * + r

i *=pJzL= —w —j (Z)- (±.)-p > ( - i ) hf z H d _


3P l ° e 2 J - p (z) U J -f# *i I 2 J Id t p+*+i-

W e now p u t p = n a n d we ob tain

3Jp (z)
dp „ - l ° 8l J"(z>+ ( T ) \ i T ( T r ( T T W ) , - « +. t l
an d

3J-p (z)
9p - - k* T - r~ <■>- ( t )'" i T ( i f ( w ? k ) . - fl+A+l

S u b stitu tin g in form ula (60) a n d using th e form ulae (67) an d (68) we obtain
finally, w hen n > 1:

” (t ) (^ :+ + • - • + J) —

- (!)" 2
k=l
k\ (ri + it)! & T (^ T F + + • •' + 1 +

(7 0 )
+ T + ’ ■ 3 T + ••• + 1)-
an d w hen n = 0:

(z) - 2 J 0 (z) (lo g ± + c ) - 2 2 -^ k ip - ( y ) k ( t + t = T + - • ■+ 1)- (7 1 )

150. The case of the purely imaginary argument. If Zp{z) is a


solution of the Bessel equation then, as we know, Z p(kz) is a solution
of the equation [II, 4 9 ]
d 2u> 1 dm nL w = 0. (72)
dz2 2 dz z- )
150] TH E CASE OF THE PU RELY IMAGINARY ARGUMENT 561

If we put that k = i we can see that the function Zp(iz) is a


solution of the equation
d2u) 1 du>
d¥ z d2
(73)

Let us suppose, to start with, that Zp{z) is equal to J p(z):


CO
2_AP+2A 1
J p (iz) = Z, (-l)W *
k\T(p+k + 1) ( I ) '" '1*
& fci r(P+ k + 1 ) T )
To obtain a solution of the equation (73) which is real when p and
z > 0 we multiply the above solution by the constant i~p = e~(1/2)pi".
We then obtain the following solution of the equation (73):

-lp ”i 1 f z Ap+2/c
I p {z) = e J P(*o = 2 o *!r(j» + fc + i) I t J (74)
k
The function I - P(z) is also a solution of the equation (73) and, when
p is not an integer, I p(z) and 7_p(z) are two linearly independent solu­
tions of the equation (73).
If we now take Zp(z) to be equal to the first Hankel function Hp^(z)
then, by adding another constant factor, we arrive at the following
solution of the equation (73):
1 1pni
K P (2) = y nie~ • (75)

From (62) we can rewrite this formula as follows:

K p( z ) = \ n i e ~ ]iPnt HW{iz). (76)

Using the first of the formulae (61) we can express K p(z) in terms of
I +P(z). In fact this formula gives:

K p (z) = - Y n^ PM ^ {%Z) B~ipn~


sin p n J-p (i2)

or, using (74):


- ±pm
(z)e *
K Pp (z)
' '
= - 2
JP(g)e *" sin- Jp n- p

and finally:
K Az\ = L a £=B. (77)
p (iz)
' '= 2 Q,Tl T
sin pl7
nT
.
562 SPECIAL FUNCTIONS [160

The functions I p(z) and K p{z) satisfy relationships analogous to


the relationships deduced for J p(z) in [144].
Using (74) and the fact that J-n(z) = ( — 1)n J„(z) for Bessel func­
tions with integral subscripts, it can readily be shown that
I-n (z) = I n (z ) • (78
An expression for the function K n(z) with an integral subscript c a n
be obtained from (77) by taking the limit as p-»- n and eliminating
the indefiniteness by taking the differentials:

As we said in [112] the asymptotic formula:

is valid when —n + e < arg z < n — e, and therefore we can replace


z by iz, where z is real and positive and arg (iz) = nj'2. Using formula
(75) we obtain an asymptotic expression for Kp(z) when z > 0:

K p[z) = ± m £ p* jf ' e '( 'z ^ i) [1 + 0 (z-1)]


or
(z>0) (80)

i.e. the function K p(z) decreases exponentially as z —*■ + = ° .


The equation (73) is often met in mathematical physics and there-
fore the solution K p{z) with its exponential decrease is of great im-
portance in applications to problems in physics.
In some cases the symbol K p(z) denotes a function which, in our
notation, is equal to cos pn Kpn(z).
If k is replaced by ik in the equation (72) then we find that the
functions I p(kz) and K p(Tcz) are solutions of the equation:

(81)

These solutions will be linearly independent in the same way as


J p(z) and H $\z) are independent in the Bessel equation.
Numerous tables for Bessel functions are available. We mention,
for example, the book by Prof. Kuzmin Bessel Functions in which
tables are given.
161] INTEGRAL REPRESENTATION' 66a
151. Integral representation. To explain certain properties of Bessel functions
It is convenient to use integral representations which differ from those we
considered earlier. These representations can be obtained by superimposing
two-dimensional waves (Frank and Mizes, Equations of Mathematical Physics)
0r by the method of integral transformations (Courant-Gilbert, Methods of
Mathematical Physics), or, finally by the method of linear transformation of
Bessel functions which we introduced above. We shall pursue the third method.
Replacing 1/T(p + k + 1) in formula (7) by its expression in terms of a con­
tour integral [74], i.e.
1 — fe TT-(',+*+1)dT,
r(p + k + 1) 2m J
v

where V is a oontour which encircles the negative part of the imaginary axis,
we obtain

_ lU7.1
i'*
k=0 r

As a result of the uniform convergence of this latter series the transposition


of summation and integration is permissible. The summation gives:

Jp(2)= 2 ^ J ( y ) PT" <',+1)e ‘lTr<iT-


r
We assume th at the complex number z
satisfies the condition

I arg 2 1< y . (82)


and we replace the variables according to
the formula r = (1/2) zt. We then have

^ ( z ) = 2 s r J r P ‘ l e i ^ <’ 1^ df’ (83)


where we can take the former looped contour I' as the contour of integration-
Formula (83) was developed by N. Ia. Sonin (in 1870).
Take for I a contour consisting of the lower edge of the cut along the negative
part of the real axis, the circle | 1 1 = 1 and the upper edge of the same cut.
If we take the new variable of integration given by the formula t = ew, the con­
tour of integration I will be transformed into the contour C0 shown in Fig. 74
and the function J p(z) will be given by the following final expression:

(84).
C.
■564 SPECIAIi functions nsi
Notice th at all parts of the contour a t a finite distance from the origin can be
■deformed in any way we please. To obtain further results it is convenient to
transform the integral (84). This can easily be done if we assume that C0 hag
the form shown in Fig. 74 and th at w = <p — m.
Using the relationship sinh (<p + 2ni) = sinh <p the following formula can
readily be obtained (cf. [146]):

Jp (g) — ~ J 003 (py — 2 sin <p) d<p — J e Pf “ '"'■’’dip. (85 )


0 0

Let us now construct an integral representation of the type (84), for the remain­
ing cylindrical functions.
If we use formula (85) and also the relationship [148]
Jp (z) cos pn — J .p (z)
N p (z) = sin pn
we can obtain the equation
n
TtNp (z) = cot pn J cos {p<p — z sin <p) dq> —
0
it w

— sin^ n J*cos + z s*n ?) — J”ef?’_2Si^h,’ d<p-


d

— cos pn j" e-P P -lsinh?’ (jy

N p (2 ) = -i- j'sin (2 sin <p— p<p) d p — i-J'(e p,’ + e p,lcospjt)e 2,lnh?,d9>. (86)
b 0
This formula, together with formula (85), enables us to find an integral repre­
sentation for the Hankel functions:
(z) = J p (z) + iN p (z); H p( * (z) = Jp (z) - iN p (z).
We have:
f l ‘l> (2 ) = - L f e**l'""’-P* dw ,
n% J
Cl (87)
U ‘2) (2 ) --------- L f e 2»inh«'-pi»dtt,
m J
C,
where Cy and C2 are infinite contours connecting (— 00) with the points ( 00,
-\-ni) and ( 00 , —ni) respectively. The extension of the formulae (86) and (87)
when z is arbitrary can be performed by the method of analytic continuation
152] th e asymptotic representation ^ ha nk el functions 666

152. The asymptotic representation of Hankel functions. The integral repre­


sentations (84) and (87) are convenient for finding the approximate expression
for cylindrical functions when the values of ] z | and | p \ are large. Put

£ =* m
and consider the function
/ (w) = sinh w — t-w. (89)
The integrals in the formulae (84) and (89) will then take the following form
\ e2f(w) du>. (90)

We shall use the method of the steepest descent and assumethat p and z
are positive and real.

tv

It
C •
-c* 0 0(-£ J

F i g . 75 F i g . 76

Before using this method it is important to explain the position of the saddle
points w0 which are determined by the condition
f K ) = cosh w0—f = 0,
and to establish the position of the contours
I m (sinh w — tjw) = l m (sinh w0 — £ia0)
and, lastly, to convince ourselves th at the contours <7„ C2 and C„ can be trans­
formed to the lines of the steepest descent of the function (89).
We shall investigate all these cases and consider three separate instances
depending on the value of f = p/z.
Case 1. £ > 1, g s> 1. The saddle points are at w„ = where a > 0 is
derived from the equation cosh a = (. The equation of the stationary con­
tours which pass through the saddle points cure as follows:
v = 0 and sin v cosh u — v cosh a (w = u + vi). (91)
These stationary contours, situated symmetrically with respect to the axes of
coordinates, are shown in Fig. 76 where the directions in which the real part
of f(w) decreases are shown by arrows. Considering
[/ (ta)] = sinh u cos v — £u,
•566 SPECIAL FUNCTIONS [152

it can readily be seen that if we take for the contours of integration C0, C
and C 2 in the formulae (84) and (87) the stationary contours ( — co, —a, a, B)
( — oo, —a, a, A) and (A, a, B) respectively, then the cylindrical functions for
large values o f the argument z are determined b y integration along small
sections o f the contours in the neighbourhood o f the saddle points. We shall
give details o f these calculations by taking the function H ^ ( z) as an example.
Changing the path o f integration we replace the stationary contour (— oo,
— a, a, B) b y the contour C shown in Fig. 76. We then have

H p( l) (z) = ^ - J e2Au)du;. (92)


c
Choose the value

«- (rd a n j (93)

and assume that z is such that


z sinh a
= N > 8 . (94)
2

N otice that we had to make similar conditions in Exam ple 2 in [80].


It follows from (93) and (94):

N 1 1
z sinh a — —— (z cosh a ) 3 > 3 (z cosh a )3 (95)

and
6 sinh a „ sinh a
e = -rr:---- r— < 0 .7 6 ------r — . (96)
N cosh a cosh a

This will be useful in later results.


Break the integral (92) into the sum o f the following fiv e integrals:

f e2A“)du>+ J e2^"’l du>+ J e2^'*’^dw -\-


—<■» —a —* —a + e
a —9+nl »+nl
+ J e2Aw>du>+ J e2Aw>dw. (97)
a->« a —«+«/

The second integral we considered in [80]. Let us find the upper bounds of all
the remaining integrals in (97). To do so we consider

0 (w) = <TL[ / (u:)] = sinh u cos v — u cosh a. (98)

In the interval — o o < u < — a — e w e have:

■0 (w) = 0 (— a — e) + (w) — 0 (— a — e)\ = 0 (— a — e) —

— [cosh (a + e) — cosh a] | u + a + e | ---- S *I>*1 ^ |u + a + e |2 — . . . <

< 0 (— a — e) — [cosh (a + e) — cosh a] | u + a + e | .


J52] THE ASYMPTOTIC REPRESENTATION OP HANKEL FUNCTIONS 567

But
, , . ,, . ,. . sinh a „ cosh a „
(— a — e) = ] (— a — e) = / (— a ) ------ ^ — £ ---------—— e3 —. . . <
21 3!
, , sinh a , . N
< / ( _ a ) ------ _ £ = = / ( _ a ) - —
and
. . . . , sinh a , cosh a , , sinh a , cosh a .
cosh (a + e) — cosh a = ---- -----e ------^— £ 2 + . . . > ----- ^----e -|------- - — £ 2 >

£2 sinh a 2 , cosh a „ A
‘ N cosh a
> ---- 5---------
2 e -----o— £ > ■
2
---■
z sinh—■
a
Therefore finally
jt . . .. . N AN , , , , cosh a
® (w) < / (— a ) -------------------- \ u + a + e — r— .
' 7 1' ' z z 1 1 sinh a

Using this inequality we find


—a —a
xf(—a) e ^ sinh a (99)
| ezA»>du> < e‘
3A7 cosh a

In the interval —a + e < w < a — £, v = 0 we have:

(5 (u) = / (u) = sinh u — u cosh a


/' (u) = — (cosh a — cosh m) < — [cosh a — cosh (a — e)]
/ ( « ) < / ( — a + £) — [cosh a — cosh (a — e)] (u + a — e).

But from (96)


sinh a „ . cosh a „ sinh a
/ ( —a + £) = /(—a) — 2!
£3— 4! ■£<+...<
3!
, , , sinh a „ , cosh a „
< / ( - u) - - ^ - £*+ - ^ - £3=
. . sinh a . 7 , eco sh a ) ., . n N
= / ( - “) ------ £ U 1- 3 ^ h ^ J < / ( - a)- 0-75V -
and
, , . sinh a cosh a „ , sinh a „
cosh a — cosh (a — e) = — —— £ — ——— E2 -|--------— £ 3 —.. . >

. 7 £ cosh O'!
o'! 5 sinh a „ 2 5A7 cosh a
> £ 8,nhal 1- 2 ^ h ^ J > 8 2 £ £ ^ 3z sinh a ’

using (94) and (96).


We have finally
N N_ 5 cosh a
/ (u) < / ( - a) - 0.75 — - — (u + a - £) .
z
668 SPECIAL FUNCTIONS [162

U sing this inequality we obtain:


a —s
J e zW d w
3 sinh a ezf(—a)-o.7tJV
( 100)
5N cosh a
—a + «

To obtain inequalities for the last tw o integrals in (97) we consider


/ (— a) = — / (a) = (a — tanh a) cosh a.
U sing the expansion
fjS
a = arc tanh n = rj 4- -I— 4- —— (-•••>
J o
we obtain
tanh 3 a sinh 3 a
/ (— a) > cosh a
3 cosh 2 a

If, however, w e use (96) we can write

/ ( - a ) < 54?-
In this case, from the inequality a > e, it is evident that
N3
f (a — e) < < / (— a)
0
64 s
To find an upper bound for the fourth integral in (97) w e notice that in the
interval M = a — £, 0 < u < ji and we have

(w ) = / (a — e) — ( 1 — cos t>) sinh (a — e).


B ut
,1 — cos V > ——
2u2 ,
and from (96),

sinh a sinha
sinh (a — e) . , , sinh a
sm h a — e cosh a + e2 ----»—

sinh a ^
____. 100^
sinh a — e cosh a 25 ’

i.e. sinh (a — e) > (1/4) sinh a. Therefore


tf3
&(w) < f (a — e) — sinh a < / (— a) ■
64 z 2ji2 sinh a.

Using this inequality we obtain


a —e+ n i
7l3 „ \ N3
J e^**') du> < 2z sinh a ( 101)
152] TH E ASYMPTOTIC REPRESENTATION OF HANKEL FUNCTIONS 569

Finally in the interval a — e < u < oo, v = ji and we have



0 (w) = — sinh u —u cosh a < — u cosh a < / (— a) —
542
w cosh a.

Therefore we obtain the following inequality for the last integral in (97):
•»+ n i _N
e
e*/(w) (jjfl < e' ( 102)
2 cosh a
a —8+ ni

Notice that the inequalities ( 1 0 1 ) and ( 1 0 2 ) could easily be made more exact.
Let us now return to the expression (92). Using (97) and also the inequalities
(99), (100), (101) and (102) w e obtain the following expressions:
—a+«
J e A"’l dio
2 to (103)

in which
3 sinh a -0 .7 6 N sinh a
to < e'
6 N cosh a 3N cosh a e '

4.4
N»-
+ Yz sinh a
+• 2 cosh a
64 (104)

The integral in (103) we investigated in [80] where we saw that it can be repre
sented by the formula
—a+ e
- i- f e ^ d w=
Tit J
—a —8

= _ J _ e zA ~ “) ( - - - - - - - V [i _ 1 fi _ 5cosh2al I + a)'l , ( 105)


J L J’
_

fa U sinh a 8 I 3 sinh 2 a ) z sinh a T K 1

where
6
. ,. e~ N f N 2 cosh 2 a 1 f 2 Vf 1 cosh 2 a cosh 3 a 'j
^ ^ Y~n v 62sinh 3 a J ( z sinh a J [ 8 25 sinh 2 a 8 sinh 1 a J ‘
(106)

If the term to in (103) is taken into account then the function H ^ \z ) can be
represented by the right-hand side o f the formula (105) where instead of to
we have to' + to' and where to' satisfies the following condition:

1 /'g s i n h a V f 3 sinh a „ - 0.,5n , sinh a ,


Yn v 2 J L5.A7 cosh a 3N cosh a

4.4
+ Yz sinh a
+ ■ Jsinh
— ),
2a)
(107)
570 SPECIAL FUNCTIONS [152

We can easily find an upper bound for th e right-hand side. To do so it is


sufficient to use the equation
( 2 sinh a y sinh a Yn
{ 2 ) N cosh a (T ’
which follows from (95). W ith the aid o f this equation the values o f the right-
hand sides o f (106) and (107) can be compared. I t thus appears that when N > 8
the value o f (106) exceeds the value o f the right-hand side o f (107). Therefore
when N > 8 the error in the formula for [of the type (105)] will be
determined by the second term in (106). N otice that the condition N > 8 in
our calculations is equivalent to the requirement:
i
z sinh a > 3 (z cosh a )3 , (108)
i.e.
____ a
Yp 2 — z 2 > 3p 3 . (109)

Terms o f an even smaller order can be calculated similarly. We then obtain the
following formulae: (cf. W atson: A Treatise on The Theory of Bessel Functions)

h P ( z) f2— e
-s+parctanh-fP G ( - s )
(H O )

4 !>(z)
' 718
and

Jp (2) ( H I )
2 \ 318
where
s 2 = p 2 — z2,
and G («) = 1 + -i- 5p 2 -V 1•3 f 3 77p! 385p*'i ( 112)
3s=> J + 8 212a2 9a* + 64a* J +
This series does not converge for any a and p. However when a and p are suf­
ficiently large then the terms decrease before beginning to increase again.
The series (112) m ust always be terminated by terms which are still decreasing.
I t can be proved that if the series (112) is discontinued in the way described
above and if the inequality below is satisfied
_____ 2
Yp 2 — z 2 = a > 2.5p3 , (113)

(Note th at this condition can only apply when p > (2.5)a ~ 16), then the right-
hand side o f (111) gives approximate values for Bessel functions to an accu­
racy greater than that o f the last remaining term.
To obtain a clear picture o f the behaviour o f Bessel functions when z < p
the following expansion can be used:
162] THE ASYMPTOTIC REPRESENTATION OP HANKEI, FUNCTIONS 571

It is then apparent that the expression


s
« (p 2 — z 2) 2 .
s + p arc tanh — = ---- ^ ------ + . . .

increases as z tends to zero from values close to p. It can be seen from (110)
and (111) that when the values of z vary in the w ay described above, the Hankel
functions will grow exponentially and the Bessel functions will decrease expo­
nentially. This latter fact is used, for exam ple, for testing the convergence o f
series of the following type:

2 cnJn(e)-
n=0
If I Cn I < M n a (o > 0 ), then the above series will, in any case, converge when
n > e-

F ig . 77

Case 2. £ < 1, z 1. The saddle-points now have coordinates w 0 = ± fli


where cos 0 = £ (/? > 0). The stationary contours, determined from the
equations
sinh u sin v = (v — /?) cos fl + sin |9,
sinh u sin v = (« + /?) cos /S — sin

ore situated symmetrically with respect to the axes o f coordinates and pass
through the saddle-points and oo respectively. These stationary contours
are represented in Fig. 77 where the directions in which ^ [/(w )] decreases are
marked by arrows.
I f we take the stationary curves (a) and (6 ) which connect ( — oo) with the
points (oo, - \-7 ti) and (oo, — n i ) respectively, for the contours o f integration
C, and C 2 in the formulae (87), then the determination o f the principal parts
of the Hankel function will involve integration in the neighbourhood o f the
saddle-points ±/?i. It, is then apparent that the values o f both Hankel functions
are of the same order as their sum. Therefore to obtain the asym ptotic represents-
672 BPEOlAli PUNOTIONS [162
tion for the function J p (z) it is not necessary to perform additional calculations
but it is sufficient to use the formula:

J p ( z ) = - [ H « ( z ) +£?<?>(*)].

The asym ptotic formulae can be constructed by the usual m ethod o f the steepest
descent. W ithout performing these calculations here we give the following
final formulae:

(z) •
■\ P713- a (si) e*',

-< p i
(115)

JP(*) ■ ---- (G, cos © + O, sin p),


713
in which

= P* + «*; G (si) = Gt — 0 2 i; q> = s — p aro tan — — ^ , (116)

and where 0(s) is the series (112). I t can be shown that when
_____ »
j z z — p2 = 8 > 2.5p 3 and a > 6 (117)
and if in the expressions G, G, and G2 only decreasing terms are retained, then
the error in th e formula (115) will not exceed the value o f the last remaining
term. It can readily be shown that the asym ptotic formulae (116) become the
H ankel formulae which we obtained in [ 1 1 2 ] when z g> p.
We can obtain the corresponding formulae on the assumption that in the
series G, Qx and G2 only the first terms are retained. N oting that when z §> p
the following approximate equations are valid
71
s ~ z and arc tan - arc tan —
P Y ’
we obtain from (115):

(118)

C ase 3. { ~ 1 , p 1 . The position o f the stationary contours and of the

saddle-points can, in this case, be established by taking the lim it as £-► 1. It


is then apparent th at the saddle-points lie near the origin and that by moving
along the stationary contours, the value of integrand changes very rapidly. But,
in spite o f this, the above calculations become devoid of meaning since the
condition (113) is no longer satisfied.
152] TH E ASYMPTOTIC REPRESENTATION" OP HANKEL FUNCTIONS 573

The case o f the asym ptotic representation of Bessel functions in which we


are interested corresponds to the conditions
____ 2
}f\ p 2 — Z2 I ~ | p |3 p > 1, (119)
and was first system atically studied b y Academician V. A. Fok (V. A. Fok,
A New Asymptotic Representation of Bessel Functions, Reports o f the Academy
of Science, 1934, vol. 1, No. 3, pp. 97 —99; V. A. Fok, The Diffraction of Radio
Waves Round The Surface of The Earth; V. A. Fok, Tables of Eiry functions).
We explained the result o f his investigations in the example of the function

If in the first of the formulae (87) w is replaced by (—w) we obtain for H ^ \z )


the expression
/ 4 ° (s) = ~ J e -2Slnh W+Imdw, (1 2 0 )
C
in which the path of integration connects (— oo, — ni) with ( + oo).
The saddle-point o f the integrand function in (120) lies very close to the
origin and the stationary contour can be modified to a contour along the
straight line I m(w) = — n, from ( — oo, — ni) to the point uj, = — rc/fS' — ni,
then along the straight line from w x to the origin and, finally from the origin
along the positive part of the real axis. When moving along this path away
from the origin the integrand function decreases very rapidly. Therefore the
value of the integral ( 1 2 0 ) is determined by integration along a small section
of the contour near the origin. Denote this section o f the contour by lt . We
can then write:

\ z ) = — [ J e -2slnh V+P" dw + w, (z, p) , J (1 2 1 )

where the upper bound of w c(z, p) is approximated to in the same way as we


did when f > l.W hen the values o f z are greater the value o f w e appears to be
negligibly small.
Let us suppose that
i

p=2+ (i)s1 (122)


and introduce a new variable o f integration
i
(123)
We thus have

— z sinh w + p w = (r —
T3 z z
-... (124)
T 120 5040
and
a
e —zslnh w+pui __ * j T»+... (126)
574 SPECIAL FUNCTIONS [153

N otice that the expansion on the right-hand side converges rapidly on the
contour le.
To obtain an approximate expression for Hjp(z) the equation (125) can be
substituted in (121). We then obtain the following expression for the integral
along la:

(t ) 3 Jt60,t 9dT+--- (126)

in which small terras are om itted. N otice that the contour se in this formula con­
sists of straight lines connecting the point (z/2 ) ( — n / \ 3 — ni)ejZ with the
origin and the origin w ith the point (z/2 )* e.
Let us consider finally the contour r formed by the ray arg r = 4^/3 with the
positive part o f the real axis. The first integral in formula (126) can then be
written in the form:
f I? r ft-I? r „_*!
e i dr = e a dr — e a dr,
*■« r r-L E

and the same thing applies to the second integral.


The upper bound of the integral along that part o f the contour F which does
not belong to the contour L (this part we denoted by T — L e) can easily be
calculated to be negligibly small when z is large. Therefore when z^> 1 the con­
tour o f integration L e in formula (126) can be replaced by the contour P . As a
result we obtain the following approximate formula:

1 ( Z 3 d 6I<! (t )
(127)
60 IT,) dt*
in which

. , 1 P It -
w ( t) = — e a dr (128)

is the Airy function investigated by V. A. Fok. Tables have been constructed for
this function. We notice in conclusion that formulae for calculating the residue
in formula (127) can be obtained without difficulty in the way described above.
The treatm ent o f this subject and the examples in [80] aro due to Prof.
G. I. Petrashen.

153. Bessel functions and the Laplace equation. The Bessel equation
occurs frequently in problems of mathematical physics. Owing to the
lack of space we are unable to investigate fully the applications of
Bessel functions and we shall only consider basic facts which connect
the Bessel equation with the fundamental equations of mathematical
physics.
153] BESSEL FUNCTIONS AND TH E LAPLACE EQUATION 675

Let us begin with the Laplace equation. Previously we investigated


the Laplace equation in spherical coordinates and we thus arrived
at spherical functions. Similarly, by writing the Laplace equation in
cylindrical coordinates and separating the variables we arrive at Bessel
functions.
The Laplace equation in cylindrical coordinates has the form:
3 ( dU \ . 1 02£7 . 3*U n
3? v 3Q ) Q ® dz2

We shall seek a solution of this equation in the form of a product


of three functions, one of which is a function of g, the second a function
of <p and the third a function of z :
U = R (g)0 (<p) Z (z) .
Substituting in the Laplace equation and separating the variables
we have:
d r d fj(en d2 0 (q>) d*Z(z)
de Lg d'e J 1 dp2 | dz2 0
R(e) e &(?) (z)
Each of the above fractions will be equal to a constant since only
the independent variable g varies in the first fraction while z only
varies in the third fraction. Equating the second fraction to a constant
(—p2) and the third to a constant k2 we obtain the following three
equations:
0" (<p) + P2 (?>) = 0; Z" (z) — k2Z (z) = 0;

- ^ [ e # ( e ) ] - Jy K ( e ) + WeR(Q) = o>
or
R" (g) + j - R ' (g) + (a* - R (e) = o.
We shall, for the moment, assume that the constants g and k are
not zero. The first two equations give
cos pq>
0 (93) = e±,p‘p or 0 (99) =
sin ptp
Z(z) = e±kz.
Finally, the third equation gives Z p(k g), where Zp(z) is an arbitrary
solution of the Bessel equation with a parameter p. If we require a
single-valued solution then we must assume that the constant p is
equal to an integer n.
676 SPECIAL FUNCTIONS [163

We then obtain the solution of the Laplace equation in the following


form:

<i29>
where n is any integer and the constant k is arbitrary.
If k = 0 then in place of Z ( z ) = e ±kz we assume that Z ( z ) = 1 or
Z ( z ) = z and the equation for R ( g ) will give R ( g ) = q ± p . Finally
when p = 0 we assume that @(cp) = A + B <p, and when p = k = o
that R ( q ) = C + D log g. When n = 0 formula (129) gives the
solution in the following form:
e^JC i J 0 ( kg ) + C2W0(fce)], (130)
which does not depend on the angle <p. These equations are of impor­
tance in connection with potentials of masses with an axial symmetry.
If we require a finite solution when p = 0 then we must assume in
formula (130) that the constant C 2 is zero and we then obtain a
solution in the form
e ±kzJ 0 {kg). (131)
When solving Laplace equations of this type it is possible to obtain the
solution 1 /r which is of fundamental importance in the theory o f Newton’s
potentials, viz. the following formula will hold

F e - ta ■/„ (fcg) dfc = *---- = - (z>0), (132)


j y e 2 _ |_ g 2 t

which has numerous applications in the theory o f potentials. To prove this


formula we consider formula (42) which gives
n
e -fe70(fce) = — J e-i“-«»**>*d9>1

and integrating w ith respect to k we obtain:

J[
kz—lkQS\n<p
e - k z Jo (kg) d i = dg>,
) 2n — z — iq sin <p k = 0

or, substituting the limits

J
f e~b J,B(he) dA = f
2n J z + %q sin <p
, -1 --- dip.

0 — n

This latter integral can easily be evaluated by the m ethod given in [67]
whence formula (132) follows directly.
1541 TH E WAVE EQUATION IN CYLINDRICAL COORDINATES 677

If we replace the constant (+fc2) by the constant (—Ik2) then e±fcz


becomes cos kz and sin kz, while J p(k 5 ) and Np(kg) can be replaced
by I p(kg) and K p(kg).

154. The wave equation in cylindrical coordinates. Let us consider


the wave-equation:
d'U
dt2
= a2AZ7, (133)
where
d2U d’-U d2U
AU dx2 +1 dy2 ^ 0s2 ’
we shall seek its solution in the form of a product

U = e~iat V (x , y, z) . (134)

Substituting in the equation (133) we obtain for F a solution in


the form
AF + k2 V = 0 , (135)
where
k^ = — . (136)

The equation (135) is sometimes called the Helmholtz equation.


If we take any solution of this equation, substitute in formula (134)
and separate the real parts then it will give us a real solution of the
wave equation which, in relation to time, represents a harmonic
vibration of frequency. In some cases this solution can represent a
stationary wave and in other cases a propagating wave. We shall ex­
plain this with simple examples. If we examine, for example, the pro­
duct e" 1"1 sin kx then its real part cos cot sin kx defines a stationary
wave. Similarly, the product e~iat cos kx also defines a stationary
wave. If, however, we examine the product e~tat e‘kx then its real
part cos (kx — cot) is the sine wave which moves in the direction of
the X-axis with velocity cojk. When Bessel functions are applied,
cos kx and sin kx can be replaced by J p(kq) and Np(kq) while eikx
and e~,kx should be replaced by (kg) and Hf* (kg).
Let us now return to the equation (138) and write the Laplace
operator in cylindrical coordinates assuming, for the moment, that V
does not depend on z [II, 178]
32F , 1 0F 02F
002 ‘ Q 0g d<p2
+ jk2 F = 0.
578 SPECIAL FUNCTIONS [154

We have already solved such an equation by separating the vari­


ables and we know that it has solutions of the form Zp(kg) t
where Zp(z) is any solution of the Bessel equation with parameter p.
By assuming that p = n is an integer we obtain a single-valued
solution. When taking the Bessel function we obtain the solutions

sin n<p

the real part of which


cos n<p
cos cot J n (kg)
sin nq>

determines a stationary wave. If we take the first Hankel function


as a solution then, bearing in mind the asymptotic representation of
Hankel functions when the value of the argument is large, we have,
by taking the first terms only the following asymptotic representa­
tion:
e- M H W (kg) + 0 ( e - 1)].

i.e. we have a propagating wave at infinity the phase of which moves


to infinity. We say that these solutions satisfy the radiation principle.
If, however, instead of the factor e_'“( we take the factor elat, then
in order to satisfy the principle of radiation we must take the second
Hankel function as the second factor on the left-hand side since,
according to the asymptotic expansion, we have the following
asymptotic equation:

+ 0 ( e - 1)]-

Consider now the general case when the function V depends on


the coordinate z. Equation (135) then becomes [II, 119]:
i 0 ( a n . i a2F . 32 r . , OT7 .
q dg 0e J+ ea V + 0z2 + v °*
We will seek its solution in the form:
V =B{g)0[cp)Z(z).
Separating the variables as usual we obtain the solution of the equa­
tion:
Z P (v r _ *2 e) e ±lhz °?Ssin
p9p(p
, (137)
c'
15 4 ] THE WAVE EQUATION IN CYLINDRICAL COORDINATES 579

where Zp(z) is any solution of the Bessel equation. Putting


]c2 — h2 = X2 and considering single-valued solutions (p = n which is
a positive integer) we obtain the following solutions:
------- cos nq> ------ cos mp
(138) and (139)

The first of these remains finite when g = 0 and describes a station­


ary wave. The second solution satisfies the radiation principle. The
solutions of the first type are generally used when the inner part
of a cylinder containing the axis 9 = 0 is the domain in which the
vibrations occur. Solutions of the second type are used for the space
outside the cylinder. In diffraction problems many-valued solutions
where p is not an integer are also be frequently used.

Let ua consider one particular problem. The equation (135) has the obvious
solution e '* 1 = e'*e cos <p. Multiplying this by e~lml we obtain the solution
e i(foc-iu/) yyhjch represents an elementary flat wave propagated along the X-axis.
Let us suppose that this flat wave exists not in the whole infinite space but only
outside the cylinder g = a, where it m ust satisfy the boundary condition:

F = 0 (when g = a ) .

To satisfy this boundary condition we m ust add to the solution e‘kx of the equation
(135) a certain other solution of this equation (additional disturbance caused
by diffraction) and this additional solution m ust be single-valued and satisfy the
radiation principle. Bearing all this in mind as well as the independence o f z
of the fundamental solution we shall seek the additional solution by using ex ­
ponential functions instead of trigonometric functions, in the form o f a linear
combination o f solutions of the form (139) when A = k:

+2 on f lil>(fce)e'^ (e > o ). (140)


n~ —°°

We only have to determine the coefficients an from the boundary condition.


Remembering formula (37) and putting ( = iotlp and z = kg in this formula,
we can write the given fundamental solution in the form :

elkx = oike cos 9 = +j j i nJ n (kg)elnv. (141)


n= —»

The boundary condition gives:

2 inJn(ka)ein* + +2 ar H i l ) (ka) 6 ^ = 0,
n=»—«• n= —~
680 SPECIAL FUNCTIONS [165

and we thus obtain the following expressions for the coefficients an:

a - :■ J n (ka)
H h )( k a )-

The final solution o f the problem will thus have the following form:

The above problem has applications in certain cases o f diffraction of electro­


magnetic waves from an infinite conducting cylinder. The series obtained above
are of practical use only in cases where the waves are comparatively long.
It is interesting to compare the diffraction of an elementary flat wave with
the vibration of a round membrane [II, 178], Notice, first of all, that in the first
case the number k is given (it is determined by the frequency cu of the falling
wave) whereas in the second case it was determined from the boundary conditions.
In the diffraction problem the coefficients o f the expansion are determined from
the boundary condition, whereas in the second case they are determined from
the initial condition, i.e. from the vibration picture when t = 0. In the diff­
raction problem we have no initial condition since we are not considering the
general diffraction problem with an initial disturbance but only an established
sinusoidal case with a given frequency co with respect to time.

155. The wave equation in spherical coordinates. Consider now the


equation (135) in spherical coordinates. It has the form:
82F
9r2 + f l F + 7 ^ F + *: 2 F = 0 -
We consider a solution in the usual form:
V=f(r)Y(d,<p). (142)
Substituting in the equation and separating the variables we obtain
r(r) . 2 f'(r) 1 A, Y (9, <p)
f{r) r f (r) r% Y (0, <p)
+ &2 = 0,

where \ Y is determined from the formula (71) in [135]. We thus


obtain two equations of the following form:
Y + IY = 0 (143)
and
r (r) + - f ’ (r) + (*2 - - ) f ( r ) = 0. (144)

The equation (143) is the same as the equation we obtained for


Bpherical functions. If we suppose the solution is single-valued and
165] THE WAVE EQUATION IN SPHERICAL COORDINATES 681
continuous we obtain the following possible values for the constant X:
Xn = n(n + 1) (n = 0,1,2, . . . )
and these will correspond to the solutions of the equation (143) which
are the usual spherical functions Y n(9, <p). The equation (144) can be
rewritten as follows

fn (r) + -f'n (r) + ( v - fn M = 0. (145)

We now replace f(r) by a new unknown function R(r) according


to the formula

Substituting in the equation (145) we obtain for R n( r) an equation


of the form
. ( (*+t)')
+ W+ p - -7, -) K W - o,

and therefore R n(r) is equal to Z n + l l 2 ( k r ) , where Z n+1!2 (r ) is the


solution of the Bessel equation with the parameter p = n + l/2, and,
according to (142), we have

V = Zn+^ T)- Y n (6, <p) (» = <>, 1,2, . . . ) • (146)


fr
Notice that here the solutions of the Bessel equation are expressed
in finite form in terms of elementary functions. The choice of the
solution Y n + 1/2 ( k r ) is determined, as in the previous section, from
the physical conditions of the problem. The following three functions
are usually considered:

w (e> = 1 (e): ^ 2) (e) = f i i (e)i


(147)
v>n (e) = f £ J n+1 (<?) = | (<?) + «?> (e)].

where the constant term J/ti/2 is added to make calculations more


convenient. In particular when n = 0 we obtain, from [148]:
sin g
#> (<?) = - e ’ Cia) = *■ e v» (e) e
682 SPECIAL FUNCTIONS [155

Solutions which are independent of q> have the form:


Z t (fcr)

-----P n (oos9),
yr

and when n = 0 we have


^ (kr)
2___
Yr
To obtain solutions of the fundamental equation (133) we
must multiply the solutions (146) by e ±lat or, which comes, to the
same thing, by cos cot and sin cot, where co and h are connected by
the relationship (136). If we separate the variables in the equation
(133), as usual, assuming that U = T(t) V(x,y,z), we obtain the
equation (135) for V, whereas for T(t) we have

T ” (t) + a2k2T (t) = 0 (a2F = co2),

which has the above functions of t as solutions. But we have assumed


until now that k (or co) is not zero. When k = 0 then we must take
T(t) = A + Bt and we simply obtain for V the Laplace equation
AV = 0 . Thus we also obtain a solution in the form:
(A + Bt)rn 7 n {6, <p), (148)
which must be linked with the solutions (146).

Here, as in the above case with cylindrical coordinates, we can pursue to the
end the solution o f the problem o f vibrations inside a sphere with given
initial and boundary conditions, as well as the problem o f the diffraction of
a fla t wave from a sphere.
To start w ith let us suppose that we want to find the solution o f the wave
equation
— = a^U, (149)

which satisfies the initial conditions:


dU
u l/-o = /i (r >0. 9); 91 l-o = h (r, 6 , <p) (r < a) (150)

and the boundary condition


dU
(151)
dr
On referring to the solutions (146) and bearing in mind that our solution is
required to be finite when r = 0 , we take ■Zn+ 1/2 (&r) equal to J,,+ 1y2 (A:r) and we
165] TH E WAVE EQUATION IN SPHERICAL COORDINATES 583

determine the values of k for a given n from the boundary condition:

A
d n+ 2
I
= 0 or J n+ i (ka) — 2 k a J ’n + l [ka) = 0. (152)
dr Yr r- a 2 2

In future we shall denote the positive zeros of this equation by

k m \m = 0 , 1, 2 , ...).

A lso the solutions (148) satisfy the boundary condition (161) when n = 0.
According to Fourier’s theorem we m ust seek the solution o f our problem in the
form:
U = A + Bt +
J n+1(*4nM
+ 2 2 (e>9) 003 ofcm) t + ( 6 . 9>) 3>n o&m'*t ) ] ------- Lp • (153)
n=0m =0 yr

The spherical functions Y t y ( 6, <p) and (0, <p) o f order n remain to be


determined from the initial conditions (150). N otice that the equation (152) is
the form which we considered in [150], and we can determine the above spherical
functions by using the property of orthogonality o f Bessel functions. We shall
not explain this in greater detail.
Let us now consider the diffraction o f a flat wave from a sphere r — a given
by the solution e^kz~mt^ o f the equation (149), when the boundary condition is

|r- . = 0.

In this case we have taken a wave propagated along the 2-axis. Instead o f
the formula (141) the following formula given in spherical coordinates, applies:

e/fa = eiftrcos8 = J 1 (2n + 1 ) i" Vn (kr) P„ (cos 6 ), (154)


n=0

where P n(x) are the usual Legendre polynomials. W e shall not give the proof
of this formula. Bearing in mind the radiation principle we shall seek the ad­
ditional disturbance in the form:

2 an & ^ r) P„ (cos 6). (155)


n= 0

The coefficients an are determined from the condition that the sum of the
solutions (154) and (155) m ust vanish when r — a, and this gives
684 SPECIAL FUNCTIONS [156
§ 3. The Hermitian and Laguerre polynomials
156. The linear oscillator and the Hermitian polynomial. The Schro-
dinger equation, as we know, has the form:

- ^ + { E - V ) f p = 0.

We assume that the function ip is a function of x only and that


the potential V is determined by the formula V = kx2j 2 which refers
to the case of an elastic force / = —kx. We thus obtain the equation:

where the value of the parameter E is determined by the condition


that the solution of the equation must remain finite in the whole
interval —°° < x < + ° ° . Let us introduce two new constants:
a 2 = m*; A = 2m£ (a < 0 ) . (1)

Here a2 is given and Ais a parameter which replaces E . The equation


can then be rewritten in the form:

- £ r + ( * - a 2x-)ip = 0. (2)
x = °° is an irregular singularity of this linear equation. We shall
treat it in the same way as we did in [105], viz. suppose that:
ip = e“(x) u (x)
and determine the function co(x) from the condition that in the coef­
ficient of the unknown function u(x) in the differential equation
should be there no term containing x2. Differentiating and sub­
stituting in the equation (2) we obtain the following equation for
u(x)
u ”{x) + 2(o' (x) u' (x) [a>" (x) -f w'2 (x) + ^ — a2x2] u (x) = 0,
and to eliminate the term a2x2 we take:

a>( x) = — y X 2,

where the minus sign is chosen so as to obtain a decrease as x —> ±°°-


We thus obtain
166] THE LINEAR OSCILLATOR AND TH E HEIIHITLAN POLYNOMIAL 685

where u(x) satisfies the following equation:

^ - 2 a s ^ + ( A - a ) W= 0. (4)

If for a certain parameter X this equation has a solution in the


form of a polynomial then the function <p(x) will decrease at infinity
and, consequently, it will satisfy the necessary boundary conditions.
We shall therefore seek the solution of the equation (4) in the form
of a polynomial. Replacing x by the new independent variable

f = 1lax,
whence
dw da y—- d-w d2w
"daT ~ S f ' a; 'dS2' - IF® ’
and substituting in the equation (4) we obtain the following equation
for u :

For this equation the origin is not a singularity and the solution
can be found in the form of an ordinary power series:

u = fc-0
2 a*P

in which the first two coefficients a0 and are arbitrary. Sub­


stituting in the equation (5) we obtain a reduction formula for the
coefficients:

(k -f- 2) (k -f- 1) aft+a — 2kak -]- — lj = 0,


whence
2k - { — - l)
aft+2= (fc + 2)( fc + l) ak (i = 0 , 1 ,2 , .. .) . (6)

We shall now demonstrate a method for obtaining the solution


of the equation as a polynomial of the nth. degree. We assume
that the parameter X is chosen from the condition


a
— 1 = 2n,
i.e.
Xn = {2n+ l ) a. (7)
586 SPECIAL FUNCTIONS [156

Formula (6) then gives successively


a r .+ 2 = a n+4 = a n+6 = ■• • = 0 . (8)

When 7i is even we must also assume that = 0 and a0 o.


From (6) we have = a3 = ah — . . . = 0 and also all afc’s with even
subscripts to k = n inclusively will not be zero while, according to (8),
the remaining ones will be zero. If, however, n is odd we must assume
conversely that a0 = 0 and al # 0. We then obtain the solution in
the form of polynomials and formula (7) gives the corresponding
individual values of the parameter X. Substituting these in the equation
(5) and denoting by Hn(£) the introduced polynomials we obtain for
them the following differential equation:
H'n (I) - 2m'n (f) + 27lHn (f) = 0. (9)
From (3) we have for the function f n(£)
_ 1 {•
V„(*) = e 2 H n (£). (10)
The polynomials Z?n(!) are usually known as Hermitian polynomials
and the functions (10) as Hermitian functions.
We have equation (2) for the Hermitian functions in which x
must be replaced by f. After substitution the equation takes the
form:

^ 8 ^ + (v - £ 2K ( £ ) = ° ( ^ = 2ra + 1)- (n )
We shall now introduce a simple formula for Hermitian polynomials.
Assume that v = e - '" whence v' = —21 v. Differentiating this
equation (n + 1) times by applying the Leibniz formula to the de­
rivative of the product we have
v(n+2) = _ 2f®(n+i) - ( n + 1 ) 2u<")
or
v ( n+ 2 ) + 2|u<n+1>+ 2 (n + 1) t><n>= 0. (12)
Let us introduce a new function K n( f ) = e^ 7)(n) and show that the
equation (9) is satisfied by it. The function K n(Z) must be a poly­
nomial of the nth degree in |:

Kn (f) = e {2— r ( e - f!). (13)


157] ORTHOGONALITY 587

Substituting the expression


v ^ = e - ^ K n (^)
in the equation (12) we do, in fact, obtain the equation (9) for K n(£).
Hence the Hermitian polynomials which have so far only been de­
termined as far as the constant term are the same as the functions (13).
Notice that the second solution of the equation (9) cannot be a poly­
nomial since the point £ = °° is an irregular singularity of this
equation. To obtain a positive coefficient for the first term we ascribe
the constant factor ( —1)" to the expression (13) and determine the
Hermitian polynomial from the following formula:

H n (€) = ( - ! ) " ef,- ( e - f ‘). (14)

Let us write the first three Hermitian polynomials:


H 0(£) = 1; H 1 (£) = 2£; H t (S) = 4£2 - 2 .
i / n(!) generally contains only even powers of £ when n is even
and odd powers of £ when n is odd. This follows from the method
described above by which the coefficients ak were determined.
It follows from formula (14) that the first coefficient £" in the poly­
nomial Hn(i) is equal to 2”. This is due to the fact that the differentia­
tion of the index ( —I2) gives (—2£).
It can be shown, but we shall not do so here, that Hermitian func­
tions represent the complete set of solutions of the equation (2)
which satisfy the boundary conditions.

157. O r t h o g o n a l i t y . Consider two different Hermitian functions y>n(£)


and rpm(£). The following equations are satisfied by these functions:
d2yn (£)
df2 + (^L- £ 2) ^ ( f) = °-

Multiplying the first function by ym(£) and the second by v’nd).


subtracting and integrating over the interval (-o o , + °°) we obtain,
as always, the equation proving the orthogonality of Hermitian func­
tions:

J V>n (£) Vm (£) df = 0 (n#m), (15)
-- 00
688 SPECIAL EUNOTTONS [157

or, from (10),


+ 09
J 0 s*Hn (£)Hm{£)d£ = 0 (n^m), ( 16)

i.e. we can say that Hermitian polynomials are orthogonal with e-f* in
the interval (— -j-°°). Let us now evaluate the integral (16) when
n = m. According to formula (14) we have:
+» +«
/ „ = J e - ^ ( ! ) d ! = ( - l ) " J ff nf fl— A df,
df"

or, integrating by parts,


qn-x (e- '")
+
df" £= —a
+“
d"-1 (e~p)
df.
+ ( - l ) " +1J ff'nO) dfn—l

The first term on the right-hand side is a product of e-f‘ and a


polynomial and therefore it vanishes when f = Continuing the
integration by parts we obtain

In= j tf<">(f)e-*’df,

or, bearing in mind the fact that the first coefficient of the polynomial
Hn(f) is equal to 2”

I n = 2n nl J e-*Mdf,

we obtain finally [II, 78]:

/ „ = J e -ia^ ( f ) d f = 2 " n ! l ^ . (17)


—ee

Series containing Herm itian polynomials and analogous with Fourier series
can be constructed similarly to those containing Legendre polynomials [132].
In this case instead o f the finite interval (— 1, + 1 ) we have an infinite interval
( — oo, + oo). In this interval we obtain the following expansion

/ (f)= 2 (18)
n=0
158] THE CONVERSION FUNCTION 689

where the coefficients an, as a result o f the above orthogonality and from formula
(17), are determined as follows:
+“
on V - f d f. (19)
2 n ’. y n J
—CB
For the expansion (18) to become valid it is, o f course, essential that the
function /(£ ) should satisfy certain additional conditions.

158. The conversion function. Using (14) and Cauchy’s formula


for the derivative of the function e~z*in the form of a contour integral,
we can write

'f

where l( is an arbitrary closed contour which encircles the point


z = £. Replace z by a new variable of integration according to the
formula
z — | — t.
Replacing the variables in the integral and dividing both sides
by e~f we obtain
1 l r e—e+2ff
-d<,
K
where 1'0 is a simple contour which encircles the origin. It follows
from this formula that ffn(|)/n I is the coefficient of tn in the expansion
of the function
e - ‘*+2« (20)
into a McLaurin’s series, i.e. the function (20) is the conversion function
for Hermitian polynomials multiplied by the constant 1/ra!:

* - ™ ts= 2 ^ r H n{Z)tn- (21)


n=0
From this formula the fundamental relationships for Hermitian
polynomials are readily obtainable. Differentiating the identity (21)
with respect to | we have

e - ™ . 2 < = 2 - k H ' ^ ) tn
n=- 0
or

n=0 Ti—0
590 SPECIAL FUNCTIONS [158

and comparing the coefficients of equal powers of t we obtain the


relationship
H'n (£) = 2nHn_1 (£). (22)
On differentiating the identity (21) with respect to t we obtain:

e - ,,+ a ,{ • ( 2 f - 20 = 2 T ^ T Ijr E n (f) t " - 1


n = l v ''

or
i § H„ (f) c - J A ({) ("« = i (f)
n=0 n=0 n=l
and again comparing the coefficients we also obtain the following
relationship:
H n+1 (f) = 2fff „ (f) - 2nHn. 1 (f). (23)
Finally let us determine the constant term in the Hermitian
polynomial, i.e. Hn(0). When n is odd this term must be zero since an
odd Hermitian polynomial contains only odd powers of £. When ra
is even we have, first of all, H 0(0) = 1. Subsequently formula (23),
with n = 1 and f = 0, gives
H 2 (0) = - 2 H q( 0)=-2.

The same formula when n = 3 and | = 0 gives


Ht (0) = — 2-3H2(0) = 22-l -3.
Further, when n = 5 and | = 0 we have
H6 (0) = — 23• 1 • 3 • 5
and in general
t f 2a(0) = ( - ! ) " ■ 2"-1-3-5 . . . - ( 2 n - l ) . (24)

Notice that if Rolle’s theorem is applied several times to the equa­


tion (14) then it can be shown that all the zeros of Hn({) are real and
different. We used similar arguments in [102] to show that all the
zeros of Pn(:r) are different and lie in the interval (—1, +1).
Sometimes Hermitian polynomials are introduced in a slightly
different form, viz. instead of formula (14) the Hermitian polynomials
are determined from the formula
~ 1L e"2~d"e i!2 .
tfn(f)
nv ' = -n\
The only difference is in the constant terms one of which stands
outside the polynomial while the other refers to the argument £.
159] PARABOLIC COORDINATES AND HERMITIAN FUNCTIONS 591

159. Parabolic coordinates and Hermitian functions. We shall now


give one particular case of substitution of the variables in the
wave equation
9s U 02(7
0x2 + 9y 2 + k2U = 0. (25)

Replace x and y by two other variables £ and r) and let us suppose


that this substitution satisfies the formula
x + iy = /(£) = <p(£, rf) + i\p (£, rj) (£ = £ + «?),
where/(£) is a regular function of the complex variable £. Differentia­
ting in accordance with the law for differentiating complicated functions
we have
0(7 _ 9(7 0<p . 9(7 dip 0(7 __ 0(7 dtp 0(7 dip
0£ 9x 0£ ‘ d y 9£ ’ dy d x d y ' d y dy

and further
d2U d 2U ( d < p \ 2 d 2U dtp dip d 2U f d y > \ 2 dU d2<p dU d 2xp
0£2 — dx2 { 0£ J d x d y 0 | d£ + 0y2 U f J + 9x 0£2 + 0j/ 0 |2 ’

d2U _ d2U <’0y y 92(7 d<p_ 0y 92(7 (0v y dU_ jj2? 9(7 9>
dy2 dx2 V0^7J 0x0t/ 0*7 d y dy2 V0^7J ' 9x dy2 dy dy2‘

Using the Cauchy-Riemann equations


dip dip dtp dip
"0f = W W = _ "9F ’
and also the fact that r/>(£, ?j) and y>(£, t?) satisfy the Laplace equation
we can readily prove the following formula:
d2U . d2U ( d 2U 02(7 m 0 < 7 > y /^ y i
9f2 + dy2 “ t0x2 + 0i/2 J L U f J + U f J J
or
^ _0!^ = (0 !U
9£2 ^ dy2 1, 9x2 ^ dy2 J ' ‘
Consider the particular case when

f ( £ ) = - ( £ + iy)2; /'(£) = £ + «?
or
9(£>y) = y ( P - v 2Yi 2p(£,y) = £y-

The coordinate lines £ = U1 and t] = C 2 represent parabolae


[32] in the (x , y)-plane and therefore the new coordinates £ and rj
592 SPECIAL FUNCTIONS [159

are called parabolic. Transforming the wave-equation in the way


described above we have

and therefore equation (25) with the new coordinates will, in this
case, have the following form:

^ - + ^ r + k2(P + *l2)U = 0. (26)

We shall seek its solution in the form o f a product o f two factors, one of
which depends on f and the other on t j :
C7 = X(f)F(J?).
Substituting in the equation (26) and separating the variables in the usua
way we obtain
Y '(y)
+ k*f 2= - -ifcV-
Y(V)
B oth sides of the above identity will be equal to the same constant which
we denote by (—(I*). We thus arrive at the following tw o equations:

X '( f ) + (fc2! s + |92 ) X ( f ) = 0; F '( ,) + ( t V - R f W = 0. (27)

Let us recall the differential equation (11) which is satisfied by Her-


mitian functions:
^ ( f ) + (2n + l - f 2)Vn(f)=0, (28)
where we have the following formula for the Hermitian function:

_L* a*
V>n (f) = e 2 H n (£) = ( - 1)" e 2 — (e - f> (29)

Consider the first o f the equations (27) and replace f by a new variable
according to the formula

This gives

d f _ K t f c df 1 ’ d £ 2 dfj

and substituting in (27) we obtain the equation:


d2X
f 2) x = o. (30)
d£f
I f we determine the constant /S2 from the equation

= (2 n + 1 ) ik,
160] THB LAGUERBE POLYNOMIALS 593

where n is a positive integer or zero, we can obtain the equation (30) in the form
(28). Hence by using this new variable we can take the following Hermitian
function for X

X n = CnVn{£l) = C „ e ’ #„(£,)
or returning to the old variable we obtain
_ lk(*
X n = Ony>n ( f i k Z ) = C n e a H n ( fik {),
where Cn is an arbitrary constant.
Similarly considering the second o f the equations (27) and replacing rj by
the new variable
t]l = iYik t),
we can also obtain the second solution in the form (28) for the same value o f the
parameter fin. Returning to the old variable we have
lkr,»
Y n = D nV’n(Th ) = D „ & a H n { i f i k i ] ) .
We thus obtain an infinite number o f solutions of the wave-equations (25)
in the following form:

U n = A ny>n (yikS)yrn ( i] f lk v ) (n = 0 , 1 , 2 , . . . ) . (31)


These solutions comprise a full system o f functions and are analogous with
Bessel functions expressed in cylindrical coordinates. Here, as before, functions
analogous with Hankel functions can be constructed and this makes it possible
to solve diffraction problems with reference to a parabolic cylinder.

160. The Laguerre polynomials. We obtained the Laguerre poly­


nomials in a generalized form on solving an equation of the
type [115]
a:- 3 ' + {5+ 1 “ a:)‘S + ^ = 0 - (32)
Bearing in mind the formulae (218), (219) and (222) from [115]
we can say that the solution of the equation (32) will be obtained
in the form of a polynomial of the nth degree if the parameter // is
equal to ji„ = n; in this case, the solution of the equation will be
expressed by Laguerre polynomials for which we obtained the follow­
ing expressions:
<?<?>(a:) = x~s gx — (zs+n e~x) . (33)
da:"
Hence these polynomials are the solution of the equation

xS L+ ( 5 + 1 - a;) - S L + ^ = 0- <3 4 )

We are assume throughout that the number s is real and > —1.
594 SPECIAL FUNCTIONS [160

In (32) notice that the independent variable x only differs by a


constant term from the radius-vector and therefore the principal
interval of variation of this variable is the interval (0, + °°). The
Laguerre polynomials are analogous with Hermitian polynomials but
for them the principal interval is not the interval (— + ° ° ) but
the interval (0, + °°). Formula (216) from [115] gives the Laguerre
functions which are analogous with Hermitian functions:

a/ns) (x) = e~ 2 *2 Qff (x ) = x~^e^ (x*+n e~x) .(35)


dxn

According to (213) and (222) from [115] these functions are


solutions of the equation

where
A„ = ^ ± -L + n. (37)

As before the orthogonality of these functions can easily be deduced:

J co$ (x) (x) dx = 0 (m # n) (38)


o
and from (35)
CD

J Xs e~x (x) Q(„s) (x) da: = 0 {m # n ) . (39)


o
We shall now try to evaluate the integral (39) when m = n. Accord­
ing to the definition of the Laguerre polynomial we have

= J> [Q<?>(*)]* dx = J Qfp (x) (xs+" e~x) d x .


0 0
Integrating by parts we have
X=co
r dQW(x) d ” - 1 (xs+ne ~ x) dT
J dx dx " - 1 X
x=0

where as for Hermitian polynomials, the first term on the right-


hand side vanishes. Integrating by parts several times we finally
obtain the integral

In (- l)"Ja d " ^ (x)


dxn
dx.
160] TH E LAGUERRE POLYNOMIALS 595

But dn Qf?) {x)l<lxn is a product of n ! and the first coefficient of


Q(ns\x). Applying the Leibniz formula to the derivative in formula (33)
we can see that this first coefficient is equal to (—1)" and we can there­
fore write
I n = nl j xs+n e~x dx,
o
or, remembering the definition of the function 77(z), we finally obtain

J xs e~x [Q^s) [x)\2dx = nl /'( s + n + l). (40)


o
In the same way as we did with Hermitian functions we can con­
sider the expansion of an arbitrary function f(x) into a series of
Laguerre polynomials in the interval (0, + °°).
We shall now construct the conversion function for Laguerre
polynomials. According to formula (33) and Cauchy’s theorem which
gives the derivative of order n of the function zs+n e~z, when z = x,
we can write

where lx is a small closed contour encircling the point z = x. Notice


that the function zs+n e~z is regular in the whole plane except at the
point z = 0 where it has a branch-point, provided s is not an integer.
Replacing z by a new variable of integration
2 — * Xtr
t 2 z= 1—t 1 —t + 3,
substituting in the integral and dividing through by Xs e x we obtain

— 0(s) (x) = -
nl ' ' 2m fJ e_ ____-_____—
( j t)s+1 tn+1
where l0 is a small closed contour which encircles the point t = 0.
This shows that Q^(x)ln! are the coefficients in the expansion of
the function
_— i
e i-'
(1 — t)s+1

into a McLaurin’s series in powers of t, i.e.


_*L i
e i-f- (41)
(1 - 1) n=0
596 SPECIAL FUNCTIONS [161

From this formula several simple relationships for Laguerre poly­


nomials can be deduced. Differentiating both sides of (41) with
respect to x we obtain

—e n « y 1 W n U x ) tn
(1 - 0S+S ~ 0 M! dx
or
“ 1 dQ $ { x ) n
Qn+1) ix ) <n+1 n! dx
n—0 n=0
which, by comparing the coefficients of tn, gives:
d<3(„s) (x)
dr = - («) • (42)

Similarly, differentiating both sides of (41) with respect to t, we


obtain the relationship
xQW (x) = {n + s) (*) - (*). (43)
Finally if we multiply both sides of (41) by (1 — t) we also obtain
the following relationship:

QiT1' (*) = (*) - »<?& (*) • (44)


Frequently instead of the polynomials Qn\%) the polynomials
(1/ra 1) Qhs)(%) are considered.
By applying Rolle’s theorem several times to formula (33) it can
be shown that all the zeros of Q^\x) are real and different and lie
in the interval (0, + °°).

161. The connection between Hermitian and Laguerre polynomials. The


Hermitian polynomials can be simply expressed in terms of the Laguerre poly­
nomials These latter are, as we know, the solutions of the equation (34)
when a = — 1 / 2 , i.e.

Replace x by a new variable f according to tho formula

d 1 d . d2 1 d ( 1 d \ I d 2 I d
dx _ 2f d f ' dx2' “ 2f df { 2f d f ) ~ ~4f2 d f 2 4 f2 d f '

Substituting in the equation (45) we obtain the equation

d 2 Vn _ 9t _d?/n + 4ny„ = 0, (46)


d f2 df
162] TH E ASYMPTOTIC EXPRESSION FO E HERMITIAN POLYNOMIALS 597

■which is the same as the equation (9) if we write 2n instead o f n in the latter
equation. As we mentioned before the second solution o f the equation (9) is no
longer a polynomial and we can therefore say that f !) is the same as
H„n(f) except for a constant factor, i.e.

H 2n( Z ) = C n Q \ *'({*).
To determine the constant Cn we compare the coefficients on both sides of the
above equation. On the left-hand side the first coefficient, as we know from
[156] is equal to 22n and on the right-hand side to ( —l ) n C„ [160], so that
Cn = ( —l)n 22n, and therefore

#*„(£) = ( - l ) n22n« /1V 2V ) - (47)


Let us now deduce an analogous formula for f f 2n+i(^)- Th© function Q ^ 2\ x )
satisfies the equation:

* + ( I “ x) “S r + ny»= °-
which can be transformed by the formula x = f* into the following equation:

Replace yn by a new function zn according to the formula


1
Vn — £ z n-

On differentiating with respect to f and substituting in the equation we obtain


an equation for z„:
- ^ - 2f - ^ + (4n + 2)zn = 0.
This equation is the same as (9) if we write (2n + 1) for n in the latter
equation. The above transformations give us directly:

0)
H in+1 tf) = D n iQn'*' ({*).
The comparison o f the first coefficients gives D n = ( —l ) n 22n+1, and there­
fore:

H m tl (f) = ( - ! ) " 2™+ 1 Q ® (£2). (48)

162. The asymptotic expression for Hermitian polynomials. The Hermitian


function:

Vn (*) = e 2* H n (x) = (- l)n — (e"*1) (49)

satisfies the equation (11)


Vn (*) + (2n + 1 — x2) y>„ (x) = 0. (50)
Consider the case when n is even. We then have
Vln (x) + (4n + 1 — x 2) Van (*) = °- (51)
598 SPECIAL FUNCTIONS [162
Also from (24) and the fact that H2n(x) is a polynomial in x2, we obtain the
following initial conditions
V>2n (0) = ( - 1)" 2n 1 .3 • 5 ... (2n - 1); yin (0) = 0. (52)
The equation (51) together with the initial condition (52) makes it possible
to obtain the asymptotic expression for Hermitian polynomials when n is large.
Let us recall th at the solution of the equation
y’ + k2y = f(x), (53)
which satisfies zero initial conditions j/(0) = y'{0) = 0 has the form [II, 28]
*

y= J } (u) sin k (x — u) dti. (54)


o
If instead of the zero initial conditions we have the following initial conditions
y (0) = a; y' (0) = b, (55)
then we must add to the solution (54) a solution of a homogeneous equation
which satisfies the initial conditions (55); the final solution of the equation (53)
which satisfies the initial conditions (55) will then be
X
^ I A
y = a cos kx + -j- sin kx + / (m) sin k (x — u) dti. (56)
0
Take the equation (51) and rewrite it as follows:
Win (x) + (4w + 1) y>„(x) = x 1y>2„ (x).
Let us suppose th at in this case k2 = 4n + 1 and f(x) = x 2 y>2n(x). We
obtain from (56):
x

y>2n (x) = V*2n (0) cos y*n + 1 +


X u2V2n ( u ) sin /4 n + 1 (x — u) d«.
K4n + 1 J
1 0
(57)
I t can be shown th at when the values of n are large the first term on the
right-hand side gives the principal value of the function y>in(x). To prove this
we determine the upper bound of the integral term on the right-hand side
when x > 0. Applying Buniakowski’s inequalityw we obtain from (17):
*
I w2y>2n (u ) sin /4 n + 1 (x — u) dtt <

< j y>l„ (u) dti f ul sin2i&n -f 1 (x — u) dti <


o 6

Xs
< y>l„ (u) du f u* dti =
6 IT
162J THE ASYMPTOTIC EXPRESSION FOB HERMITIAN POLYNOMIALS 599

and substituting in (57) we have


___ 1_ 6
2" /(2m)1 fa x *
Vin (*) = Win (0) cos ifin + l x + (x ),
f t Yin + 1
where 0„(x) is a function of x which satisfies the condition
- 1 < 0„ (x) < 1.
Taking vVi(O) outside the bracket and remembering its value from (52) we
have
4 »
(58)
Win (*) = Via (0)
/ 5 / 4 m + 1 - 1 - 3 . . . ( 2 m —1)
Let us consider in greater detail the coefficient of 0„(x):
4 5

f t * 1 . /l- 2 - 3 ...2 w
4
= ft U
*
2 • 4 • 6 ... 2n
ft 1 • 3 • 5 ... (2n — 1) ft ' 1 1 - 3 - 5 ...( 2 n — 1) ‘
If we put

I k = J sin* x dx,
o
then, as we know [I, 100],
(2n — 1) (2n — 3).. .1 n 2n (2n — 2).. .2
lin ~ 2n(2m — 2 ) . . .2 2 J ssm = (2m + 1)( 2 m — 1 ) . . . 3

where I m+i < I 2n, i.e.


2m (2m — 2).. .2 (2m — 1) (2m — 3 ) ...l n
(2m + 1) (2m - 1).. .3 < 2n ( 2 m - 2 ) . . . 2 T ’

{ 2m (2m 2 ) . . .2 V (2m 4 - 1) —
V (2m — 1) (2m — 3) 1J <
... + 1 , 2 ‘
Whence
________________ 4_
1f 2 - 4 - 6 . . . 2m /m J r — —

and, finally, the coefficient of 0„(x) will be


4
f.In i f 2m + 1 J 1
4— ___ r 4 ^ r + r x ^ z = - 0">
/5050 1 /4 m+ 1

where o < e; < l. Rejecting the factor which is less than unity we can write
the above expression in the form

e;;.
/4 m + 1
600 SPECIAL FUNCTIONS [163
where 0 < 8" < 1. Substituting this expression in formula (58) we obtain the
following asymptotic representation for the Hermitian function with an even
subscript:
_____ 2 1
V>2n (x ) = Yzn (°) cos /4 n + 1 x + x2 —----------6n (x)
/4 n + 1

where —1 < o'^"(x) < + 1 . Hence for a given x the second term in the square
brackets tends to zero as the subscript n increases. The assumption that x > o
as can easily be shown, is not essential. By adding the factor e(1/2)xS we obtain the
asymptotic expression for Hermitian polynomials with an even subscript:

H2„ ( x ) = ( - 1 )" 2n 1 • 3 • 6. . . ( 2 ) i - l ) e 2 cos /4 n + 1 x -f- O

When the subscript is odd we can obtain similarly


n+ -
(x) = ( - ! ) " 2 X
_____
X 1 -3 -5 .. ,(2n - l ) / 2 n + 1 e2 sin /4n + 3 x + 0

A_ 4_ 1_
In these formulae 0 (l//n ) denotes a value such that 4/ra 0 (l//n ) (l/4dn)
remains bounded as n increases provided x varies in an arbitrary finite interval.
Notice th at we can take any argument /4 n + ux for the trigonometric function,
where a is a given real number. In fact we have, for example:
cos /4 n + 1 x — cos J^4n + a x =
„ . ^4n + 1 + K4n + a . /in + a — jin + 1
*I sin ----------- ^------------x s in ----------- x=

_ . /4n + 1 + /4 n + a . a —1
= 2 sin '------ x sm — — x.
2 2(/4w + a + /4 n + l)
4_
When x lies in a finite interval the above product will be equal to 0(l//n)
and therefore cos /4 n -f- lx can be replaced by cos /4« + ax with an accuracy
equal to this value. Using the above calculations it possible to obtain even more
accurate results for the additional terms 0 (l//n ).

163. The asymptotic expression for Legendre polynomials. By using the same
method the asymptotic expressions for the Legendre polynomials P„(x) can be
deduced when n is large. We have the differential equation:
(1 — x2) Pn (x) — 2xP'„ (x) + n (n + 1) P„ (x) = 0.
Let us replace x by the new variable t according to the formula x = cos t and
replace Pn(x) by the new function:
v n (0 = Ks >n f P n ( c o s 0 o r P fi (c°s t ) = -— = = -• (59)
/sin t
163] THE ASYMPTOTIC EXPRESSION FOR LEGENDRE POLYNOMIALS 601

Substituting all this in the equation we obtain after simple operations the
following expression for vn(t):

1 1 2,
T ~ T cost v„ (t) - 0,
vn0) + n ( n + 1) + sin21
which we can rewrite in the form:

x varies in the interval —1 < x < + 1 which corresponds to 0 < t < n. Take
t = ji/2 for the initial value corresponding to x = 0. Consider the case when
the subscript is even:

v2n (t) + [ 2 n + — j v 2n ( 0 = - -4 g j ^ y v 2n (I ) . (6 0 )

Bearing in mind formula (59) and also the fact that

P,n (0) = ( Dn 1 ^ ; - 4 (2W2~ ])- and P 2n (0) = 0,

we have the following initial conditions for v2n(t):

*■ ( f ) ° (- 1)" 1 ' 32 v4!2"2; 1) = <6,»


If in the equation (60) we reject the right-hand side then the homogeneous
equation so obtained will have the general solution:
Cx cos (2 n + y j f + C2 sin ^2n + (62)
T '-
If we select Cl and C2 so that the conditions (61) are satisfied:

Cj cos ( 2 n + y ) y + c 2 sin (2n + y ) y = (y ) >

—Cxsin \2n + y ) y + c 2 003 [2n + y ] y = °.

or, using the formulae cos (nn + tp) = ( —I)'1cos <p and sin (nn + <p) =
= ( —1) "sin p, wo have:

Gx cos y + C2 sin y = (— 1)" v2n ( y j ; — Cx sin y + C2cos y = 0,

whence

C| = = ( ^ v2n ( y ) = 1)n ( y ) Sln T ’


and substituting in (62) we obtain the expression:

( - 1)" *•-!„ ( f ) oos [(2n + y ) * - y ] •


602 SPECIAL FUNCTION'S [163
Hence the solution of the equation (60) with the initial conditions (61) will be:

(<) = ( - 1)" <hn ( y ) cos [ ( 2n + y ) 1 - y ] -


i

- 7“ “ 7 T J T ii W ^ (u) s i n ( 2 n + t ) (t - M) dM’ <6 3 )


(2 n + y j ,

where we assume that 0 < a: < 1 and 0 < t < .t/2. Notice also that the solution
of the equation (53), which satisfies the initial conditions y(a) = y'(a) = 0, has
the form (54) where the lower limit of the integral is not zero but a [II, 28],
If we consider the integral on the right-hand side and use formula (59):
l
K zn = f 4 3in» u S*n (2n + y ) ~ u>
' P*n (cos ^sin u du’
Zt
2

then, applying Buniakowski’s inequality, we obtain:

f. sin2 ( 2 n + | ] ( ( - a ) 2
K \n < V . / ----------- du Pl„ (cos u) sin u du.
J 10 sin* u .1
t t

The first of the factors on the right-hand side will be less than
71

where fi(t) has a finite definite value for a given t and remains bounded when
0 < e, < t < ;r/2, where e, is a given positive number. The second factor is
less than
3 1
J P j n (cos u) sin u du = J P\„ (x) dx = ^ _|_ y ■

We finally obtain the result:

I K ln\ <
K4n+ 1 ’

where a(x) is independent of n and remains bounded when 0 < x < 1 — f,


where e is any given positive number. Substituting in formula (63) we have:

"2n (0 = (— 1V «2n ( y ) cos ^ 2 n + y j —jt- +


y W __
(4n + 1)2
3
163] TH E ASYMPTOTIC EXPRESSION FOR LEGENDRE POLYNOMIALS 603

where y(x) remains bounded in the interval 0 < x < 1 — e as n increases.


Bearing in mind the expression (61) we obtain:
] • 3 . . . (2n — 1)
»2n (0 jeos [ ( 2n + i - ) l - | ] +
2 - 4 ...2 n

+ ( - ! ) " ---------- — y (-K)] •


1 ■3.. .(2n — 1) (4n +1)2
We shall use the inequality:
2 - 4 - 6 ...2 n
1 -3 .5 ( 2 n - 1) < ^2 *2 n + l ’

which we have proved in the previous section. I t gives:

1 -3 ...(2 n — 1)
r 2n (0 2 • 4 ... 2n

1 -3
, (') 2

where d(x) and rj(x) are functions of x which remain bounded when 0 < x >
< 1 — e and as n increases. These functions can be determined even more
accurately by using the above calculations.
Finally, using formula (59) we obtain:

Pin(cos0 = |cos [(2n+ y ) - t ] + 0 (y)| • (64)


Similarly for an odd subscript
1 1 • 3 . . .(2 n — 1)
Pin i ( eos 0 2-4...2 n „ [ ( 2 „ + | ) , - i ] + 0 (i-)|. (65|
/sin I
The result obtained is also valid for negative values of x = cos t and the
symbol 0(l/n) in the above formulae denotes a number which is such th at the
product n0(l/n) remains bounded as n increases independently of x , provided x
lies anywhere in the interval —l + e < a ; < l — e, where e is any fixed small
positive number.
We can write the above formula in a simpler form. To do so we use the Wallis
formula [75]
n 22• 42. .. (2n — 2)2 • 2n
l * - 3 * ...( 2 » - l ) 2 '
It gives:

or
W4 SPECIAL FUNCTIONS [164
where rjn -*■ 0 as n oo, i.e.
1 • 3 . . . (2n — 1) = 1 nn
2 ■4 .. ,2n j rut ' ^2w

After this, formula (64) can be rewritten as follows:

P (OOS l) = f wTsinT {C0S [(2” + T ) *“ T ] + •


Applying the same procedure to formula (65) we can see that the following
formula applies to any subscript

p"(cos () = F (cos [(" + y ) 1~ f ]+ A • (6li>


where % -► 0 as » oo uniformly with respect to t provided that e < t <
< — e where £ > 0.
We shall also give below the asymptotic expressions for Laguerre polynomials
without going into the proof. When x varies in the interval 0 < a < x < b,
where a and b are small arbitrary finite numbers, then the following asymptotic
formula holds:

Q<-i?(x)=n r n r t - n ' . x r i e* jcos ^2 ][n x — + 0 J. (67)

§ 4. Elliptic integrals and elliptic functions


1 6 4 . T h e tr a n s f o rm a tio n o f e llip tic in te g ra ls in to n o r m a l fo rm .
In this section we shall deal with certain functions of a complex
variable which are not connected with linear differential equations
and have a slightly different origin, viz. these functions are connected
with certain integrals which cannot be expressed in finite form, i.e.
with the so called elliptic integrals. We have already mentioned
these integrals earlier [I, 199]. Here we shall consider them in detail.
Previously we considered an integral of the type

Ax, (1)

where E(x, y) is a rational function of its arguments, and P(x) is a


polynomial of the second degree. We saw that such integrals can be
expressed in terms of elementary functions. If, however, P(x) is a
polynomial of the third or fourth degree then an integral of the type ( 1 )
is known as an elliptic integral and it cannot, in general, be ex-
164] THE TRANSFORMATION OF ELLIPTIC INTEGRALS INTO NORMAL FORM 605
pressed in finite form. In exceptional cases it is possible to ex­
press it in terms of elementary functions. Thus, for example, if
we take the integral
f x!!'l+1 dx
J -f- 6a:2 + c ’
where n is an integer, then by introducing the new variable t — x2
we obtain the following integral
_1_ f ___tn dt
2 J Yi2 + b t + c

which, as we know, can be expressed in terms of elementary functions.


If an integral of the type ( 1 ), where P(x) is a polynomial of the third
or the fourth degree, can be expressed in terms of elementary functions
then such an integral is known as pseudoelliptic.
Let us now consider elliptic integrals. We notice, first of all, the
case when P(x) is a polynomial of the third degree and does not differ
fundamentally from a polynomial of the fourth degree. One can be
transformed into the other by a simple replacement of the variable
of integration. In fact, suppose that P(x) is a polynomial of the fourth
degree
P (x ) = ax 4 lx 3 -)- cx2 + dx + e, (2 )
and let x = Xj be one of the zeros of this polynomial. We replace x
by the new variable t given by the formula

x = Xj -f- — . (3)

Substituting in the expression (2 ) we obtain

P (x) = a ^xt + — j +6 ^Xj -f- —j -f- c ^Xj + -j") + d ^ j


+ — -)- e .

Removing the brackets and remembering that x = xl is a zero of the


polynomial (2 ), we have
P (x) = >

where P^t) is a polynomial of the third degree. Hence we can go


from a polynomial of the fourth degree to a polynomial of the third
degree. Notice that the transformation (3) must involve the fact
that one of the zeros of the polynomial (2 ), viz. x = xt is, with
the new variable, transformed into the zero t = <=■=.
606 SPECIAL FUNCTIONS [164

Conversely, if P(x) is a polynomial of the third degree, then, as


a result of the bilinear transformation

x - aytt +
+d
P

we obtain
P (x) = (yt■P2(0
+ 6)* ’
where P2(t) is a polynomial which is, in general, of the fourth degree.
Arguing in the same way as in [I, 199] we can show that the elliptic
integral ( 1 ) can be transformed into integrals of the following types:

\-^ L d x (4 )
J fP (x )
and
dx
(5 )
J ( —a)k VP
x (x )

where <p(x) is a certain polynomial. If we suppose that P(x) is a


polynomial of the third degree we can show that the above integrals
can be transformed into integrals of one of three types. To do so con­
sider an integral of the type
4 = f— dx, (6 )

where k is an integer, either positive or negative. Let us suppose that


P (x) = ax3 + bx- -j- c,x + d .
By differentiating we obtain:

(x™yP (x))' = mxm~x yP (x) + xm 3ox, + ^bx + c


2 yp (x)

_mxm~x (ox1 -j- bx- + c.r + d) xm (3ax2 + 26x -(- c)


+
y p (x) 1 2 y p (x>

which after integration, and by bearing in mind the notation (6 ), gives


xm y P (x) + C = ina,Im+2 + m blm+1 + m clm + m d lm_ x +
3o _
2 ~-I m+ 2 i
b lm+ 1 +

(G being an arbitrary constant) or

-(- -yj I m+2 + b (m + 1) I m+X + c {m, + -yj I m + (7)

+ = xm yP (x) + C.
164] TH E TRANSFORMATION OF ELLIPTIC INTEGRALS INTO NORMAL FORM 607

When m = 0 and m = 1 we have

ya/2+ Wj + y/o = If? (*) + C.


—o/ 3 + 2bl2 + y c/i + <Z/ 0 = X ]/P(x) + C.

The above formulae make it possible to express I 2 and / 3 succes­


sively in terms of 7 0 and I 1. Putting m = 2 in formula (7) we have:

y a/4+ 3b l3 + y cl2 + 2 d l1 = x 2 ][P (x) + C,

whence we can determine I 4 etc. Hence all integrals of the type


(6 ) can be expressed in terms of I 0 and I 4 when k is a positive integer.
The integral (4) obviously has the same property.
Let us now consider the integral (5). Replacing x by the new vari­
able x — a = t we obtain the following integral

71 = J 7 F T » d' m
where P4(t) is a polynomial of the third degree and k is a negative
integer. Putting m = —1 in formula (7) we obtain

- a' i [ - - c 1 - d' il 2 = t-1 + c,

where a', b', c' and d' are coefficients of Px{t).


If we now put m = — 2 we obtain

- j a ' / i - i ' I -1 - y c> IL >- 2 d ' Z13 = r " + C


etc. This shows clearly that all integrals of the type (8 ) can be expressed
in terms of I[, I ’0 and IL 1, i.e. by using our former notation they can
be expressed by the integrals
^ x —a ^ p dx _ p dx
J ]/P (x) ’ J / P (x) ’ (x — a) l/P (x) J
Hence we can finally say that when P(x) is a polynomial of the
third degree, then any elliptic integral can be transformed into an
integral of one of the following three types:
r dx r x dx r dx
(9)
J fpjx) ’ Jypjx) ’ J (x - a) y p jx )
The first of these integrals is known as an elliptic integral of the
first kind, the second as an elliptic integral of the second kind and the
third as an elliptic integral of the third kind.
608 SPECIAL FUNCTIONS [166

Notice that when the initial integral is real then the above calcula­
tions might result in formulae which could contain complex numbers.
Thus when using formula (3) the number xx might be complex if
all four zeros of the polynomial P(x) are complex. Similarly during
the expansion of the rational fraction into partial fractions and
the transformation of the integral into the form (5) it is possible to
obtain complex values for a. We shall not give a detailed explanation
of the way in which these calculations must be carried out so that
real quantities only should result. In future we shall to a certain ex­
tent take this circumstance into account.

165. T h e c o n v e r s i o n o f t h e i n t e g r a l s i n t o a t r i g o n o m e t r i c f o r m . We
shall for the moment only consider elliptic integrals of the first and
second kinds and show that they can be obtained in a new form in
which the integrand is expressed in terms of trigonometric functions.
Let us begin with integrals of the first kind. We can, of course,
assume that the first coefficient of P(x) is equal to ± 1 ,
P (x) = i x 3 + bx2 + cx + d .
Let us suppose, to begin with, that this polynomial with real coeffe-
cients has three real roots: a, /S and y. We must, of course, assumi
that there are no equal roots among them since otherwise the poly­
nomial P(x) would contain the square factor (x — a )2 which could be
taken outside the square radical and we would then have under the
radical a polynomial of the first degree. We now suppose that a is the
smallest zreo when x3has the sign, and the greatest zero when x3has
the — sign. Let us suppose further that {S denotes the middle zero.
Replace x by the new variable <p according to the formula
x = a + (/? — a) sin2 <p. ( 10 )
Substituting this expression into our polynomial
P (x ) = =t + bx2 -f- cx d = d: {x — a) (x — /?) (x — y),
we obtain after simple calculations:
P (x) = \y — a \(fi — a)2 (1 — k2 sin2 <p) sin2 <pcos2 <p,
where
(ID

and we always assume that k > 0. We have from (10):


dx = 2 (/? — a) sin ip cos <pd^j,
165] TH E CONVERSION OF TH E INTEGRALS INTO TRIGONOMETRIC FORM 609

and therefore the expression


dx
( 12 )
WW)
differs only by a constant term from the expression
dy
(13)
/ l — k2sin21p
We can show that it is possible to arrive at the same result when
the polynomial P(x) with real coefficients has only one real zero
x — a. In this case this polynomial can be represented as follows
P (x) = ± (x — a) (x2 + px + q).
where the trinomial (x2 + px + q) with real coefficients has no real
zeros and therefore always remains positive when the values of x
are real. Replace x by a new variable <p according to the formula

x = a ± / a 2 + pa + gtan 2 y . (14)

After substitution we obtain


3 tan 2 -f-
± (* — a) (x2 + p x + q) = (a2 + pa + q)2 (1 — k2 sin2 <p)------ — ,
cos4T
where

(15)
]fa2 + pa + q

We shall show that the value of k2 lies between 0 and 1 . To do


so it is sufficient to show that the modulus of the second term inside
the large bracket in the expression (15) is less than unity, i.e. it is
sufficient to show that the square of the denominator is greater
than the square of the numerator. We obviously have:

a2 + pa + q = (a + + (g - -^p2j .

But the second term on the right-hand side is bound to be positive


since it is given that the trinomial x2+ px + q has imaginary zeros,
and we therefore have

a2 + pa + q > (a + |- ) 2.
610 SPECIAL FUNCTIONS [165

Furthermore, it follows from the expression (14) that

__________ ta n | -
dx = ± Ya 2 -f- pa + q --------dip,
cos’f-
and, therefore, as a result of the substitution the expression ( 1 2 ) only
differs by a constant term from the expression (13).
We thus see that every real integral of the first hind can, by means
of a real substitution, be obtained in the form

l . tr •, • (o < &2 < i ) . (i6)


J f 1 — k 2 sin3<p

Let us now consider integrals of the second kind

Jb Yp= (*)- da:-


Applying one of the above transformations we can obtain this inte­
gral in the form (16) together with one of the following two integrals:
tan3 <P
I ]f 1 — k 1 sin2q> Y
dw and ■
J yVli -—k 'k-- sin2<p
dip. (17)

Adding to the first of these the constant k2 we can write it in the


form
k2 f ain dip = f -- — ----- f y \ — k2sin2 <pdip,
J y 1 —k2sin2 <p J Y l —ifc2 sin3 <p J
and therefore can be converted into the integral (16) and an integral
of the following type:
J
|/l — k 2sin2 <pd^. (18)
Let us show that the second of the integrals (17) can be converted
into the first of the integrals (17). To do so we shall use the following
formula which can easily be proved by simple differentiation

2 d (tan f- f l -* * sin * p ) = [ ( l + tan 2 f-) - 2k2sin2 y] y, _ ^ sin7 ^ ■

Integrating this identity we obtain the required result:


©
tan- —
J / l - k'-
7---- dip =
sin-
ir>2 <
rrp
2 tan
£•
|/l — k2sin2 <p

+ 2/fc2 f
J yl — k- sin- <p J y\ — k'1sin- <p
16 5 J THE CONVERSION OF THE INTEGRALS INTO TRIGONOMETRIC FORM 611

We thus an see that elliptic integral of the second hind can be con­
verted into integrals of the following two types:

f ,,, (Y l - £ 2 s i n 2 <pd<p. (19)


J y i — k2sin* fp J
These integrals are sometimes known as elliptic integrals of the
first and second kind in the Legendre form.
The integrals (19) can be written in a slightly different form if we
replace <p by a new variable t according to the formula:
t = sin <p.
In this case
At

and the first of the integrals (19) can be obtained in the following
form:
r d<
J Y(1 - fl) (1 - k2t2) ‘
Here we have under the radical a polynomial of the fourth degree
of a special kind. We could, by taking an elliptic integral of the
general kind, obtain such an integral if we substitute the independent
variable by performing the general bilinear transformation
= +0
yt + 8 '

Let us write the integrals (19) with a zero lower limit and a variable
upper limit by introducing the special notation:
<p <p
F(h,q>) = [ t T ?! ■ : E (k ’<P) = I Fl - h-sm 2q>dq> (20)
0J VI
* — k 2 s in 2 <p r ftJ
If the upper limit is equal to cp = n/2 then these integrals will be
functions of h alone:

F(fc> f
=
2

In • ~ ;
J \ 1 — k 2 sin - <p
o ’ *
2

(^) = f Kl - fc2 sin2 d(jp,


J
0
(2 1 )

and the}- are usually known as full elliptic integrals of the first and
second hind.
Tables are in existence which give the values of the integrals (2 0 )
and (2 1 ). The Legendre tables, published in 1826, were the first of
this kind. These tables contain among other things the logarithms
612 SPECIAL FUNCTIONS [166

of the values of (2 1 ) for different values of k; it is assumed that


k = sin 0 and values of 0 are given in tenths of every degree. The
fundamental table with a dual approach gives the values of the integral
(2 0 ) where, as before, it is assumed that k — sin 0. This table contains
values for integrals of <p and 0 for every degree from 0 to 90 degrees.
The values of the integrals are given with a nine tenths suffix. We can
also mention the book by Janke and Emde: Tables of Functions with
Formulae And Curves, Which Also Contains Tables of Elliptic Inte­
grals.
166. Examples. 1. The time required for the full vibration of a simple pen­
dulum of length I and amplitude of vibration a is expressed by the formula

T = l/a - r a, ,
( 22)
9 J ycos r — cos a
0
where g is the acceleration due to gravity. I t is not difficult to express the
integral (22) as a full elliptic integral of the first kind. To do so we introduce
the constant k = sin a/2 and replace t by a new variable determined by the
formula sin t/2 = k sin <p. We then obtain

cos t — cos a = 2 ^sin2 — sin2 j = 2 fc 2 cos2<p


and also
t 3 , 2k cos w d®
cos — dr = 2k cos <pd<p or dr = — -=- ,
2 y 1— k2sin2 <p
bearing in mind th at because sin r/2 = sin a/2 sin <p, the variable <p must vary
between 0 and ji/2, this gives us finally

d?)
T = 2 = 2 ]f j F W .
YT k2sin2<p

2. Consider the elliptic integral

d <p
j Yi — k2 sin2 <p

where k2 > 1 and assume that the upper limit <fa lies in the interval (0, a),
where a is determined by the equation sin a = 1fk. Replace <p by the new
variable y> according to the formula sin y>= k sin <p. y>can only vary in the
interval (0, v>o) where sin y>0 = k sin rpQ, and after elementary transformation
we have:
d?> dy
/ l — k2 sin2<p 1
1— sin2y
k2
ICG] EXAMPLES 613

whence
I Vain*? k F ( k ’ *•)■

If the upper limit <pa = a then y'0 = jr/2 and we obtain, according to the
notation (21):

f e (—1.
J / I — k2sin2<p ^ \k )

Similarly we can consider the integral


?•
J l T — k2sin2tp dip
where k* > 1. In this case we obtain

J Y . - »■ I n - » d * - i F ( i ) + kE ( i ) - W ( - L ) .
3. Consider the integral
f dx
J Vl — x* '
Putting x = cos <p we can obtain the integral in the normal form:
f dx _ _1_ r df
J /l - x* ~ /2 J
sin2 9*
Similarly by using the same substitution we obtain
r x2dx _ I f d9* Y2. f l / l - i - s i n 2^ ^ -
J / i —x« ~~ V2 J |/in1 - yr s i n 2f>
4. Everything th at was said in [166] applies to the integral:
f dx
J l,x3 + l
and replacing x by a new variable <p, according to the formula:

x = — 1 + /3 tan 2y ,
we obtain:
r dx _ 1 f dp
JysTT-^J i/7 q rp j$ = ^ '
5. I t is more difficult to obtain the integral
f dx
(23)
J T ^ rT
614 SPECIAL FUNCTIONS [167
in a simpler form. In this case the polynomial under the radical can be written
as a product of two real polynomial of the second degree
x* + 1 = (x2 + l)2 - (/2x)2 = (x2 + /2 x + 1) (x2 - /2x + 1).
In general, when a polynomial of the fourth degree under a radical has only
imaginary zeros and factorizes into real polynomial of the second degree
P (x) = (x2 + px + q) (x2 + p'x + q’),
then the substitution of the variable must be performed in accordance with
the formula
A+ fim tan p
1 + m tan p
where the numbers A and p must be determined from the formulae

(p - p') A = q — q' — Y{q — q’f + (p — p') {pq' — qp‘),

(p — p')ft = q - g ’ + Y(q - q'Y + (p - v ’) (pq ' - qp ').


and m is the smaller of the two numbers
1f V - p X + q ttn d 1 / A2 — p ' X + q '
) H2 — p p + q I P* - p'p + q'
In this case the substitution of the variables takes the form
_ tan p — (1 + /2)
tan <p+ (1 + /2)
and as a result of this transformation the integral (23) becomes
dx f (2 + /2 ) dp
J / Y3?
x * 4+- l* J ]/sin*
l / a i r p + 6 (3 + 2 /2) sin2 p cos2p + (3 + 2 /2 )2 cos1p

I t can readily be shown that the expression under the radical is a product of
4 /2
sin2p + (3 + 2 /2 )2cos2p = (3 + 2 /2 )2 1 — = sin2p
3 + 2 /2
and sin2 p + cos2p, and we finally obtain the following formula in which the
integral (23) appears in the normal form:
( 2 - / 2 ) dp
f i = f
J /x * + 1 J 4 /2
sin- <p
3 + 2 /2
1 6 7 . T h e c o n v e r s i o n o f e l l i p t i c i n t e g r a l s . Having explained the
concept of an elliptic integral we shall now proceed to explain elliptic
functions. In some respects these functions are similar to the known
trigonometric functions and are so to speak their generalized form.
We shall explain, first of all, the fact that the fundamental trigono-
107] THE CONVERSION OF ELLIPTIC INTEGRALS 615

metric functions, for example x = sin u, can be obtained by trans­


forming integrals. Consider the elementary integral
X
f da;
u= , = arc sin x . (24)
oJ KT=p
The value of the integral u is a function of the upper limit of
integration x. Let us consider the inverse function, i.e. let us consider
the upper limit x as a function of the value of the integral u. We
thus obtain a single-valued, regular periodic function x = sin u. It is
said that this function is obtained by converting the integral (24).
Similarly, if we take an elliptic integral of the first kind

u=
f dx
]/P(x)

then we obtain an analytic single-valued function x = f(u) as a result


of its conversion. This function will no longer be an integral function.
It will be a fractional function and will have not one but two essentially
different periods. We shall explain this problem in greater detail
later. At the moment we shall consider an elliptic integral of the first
kind of the Legendre form
r dz
(25)

and assume that k is real and satisfies the inequality 0 < k < 1 .
We have already met the integral (25) in connection with the
conformal transformation of the upper half-plane z into a rect­
angle in the w-plane [37]. We will recall the relevant results
obtained at the time, but we shall slightly modify our notation.
Formula (25) gives the conformal transformation of the upper half­
plane z into a rectangle ABCD in the M-plane. Its side A B lies on
the real axis and the coordinates of A and B are as follows

dz
±K = (26)
^ ( 1 — Z * ) ( 1 - £ 2 Z2)
and
dz
length A B = 2 J = 2 K. (27)
016 SPECIAL FUNCTIONS [167

The length of the side BC is determined by the formula:


1
k
length BC = f dg — .
J Y(z* — 1) (1 —k 2Z2)
In this formula if we replace 2 by a new variable of integration
2= l i y i — k' 2x 2, where k '2 = l —k2, then after substitution we obtain
a new expression for the length of the side BC which, like the length
AB, will be expressed by a full elliptic integral of the first kind:
i
length BC = f . ** . = = K ', (28 )
6 J ^ ( l - x l)(l-Jfc'1®*) v ’
where the number k'2 is determined by the formula
Jfc* + k'2 = 1. (29)
The number k is usually known as the modulus of the integral (25),
the number k' as the additional modulus, and the two are connected
by formula (29).
We shall now analytically continue the function (25). If, for example,
we perform the analytic continuation from the upper half-plane into
the lower half-plane across the line ( 1 , 1fk) on the real axis then the
resulting function will transform the lower half-plane into another
rectangle, obtained from the original rectangle ABCD by reflection
in the side BC, which is obtained from the above line on the real
axis. Similarly other analytic continuations of one half-plane into
the other give values of u in the form of a rectangle in the w-plane,
obtained from the initial rectangle by reflection in that side of the
rectangle which corresponds to the section of the real axis across
which the analytic continuation is performed. Hence all the possible
analytic continuations of the function (25) in the 2 -plane result in
a net of similar rectangles in the w-plane which fill the entire n-plane
without overlapping. Each of these rectangles corresponds either to
the lower or to the upper half-plane z. This net is shown in Fig. 78
where the clear rectangles correspond to the upper half-plane and
the shaded rectangles to the lower half-plane. Conversely, by
performing the analytic continuation of the function 2 = f(u), obtained
as a result of the conversion (25), across a line I, we must only pay
attention to those sides of the rectangle which are intersected by
this line when we obtain in the 2 -plane transitions from one half­
plane into the other across the corresponding sections of the real
axis. If, for example, we encircle one apex of our net of rectangles
167] THE CONVERSION OP ELLIPTIC INTEGRALS 617

in the w-plane we obtain the former values of 2 in the 2 -plane. We can


thus see that the function f(u) is a single-valued analytic function
in the whole w-plane.
The point u = iK ', which lies in the middle of the side CD of the
initial rectangle corresponds to the value z = ° ° [37], where the
\y ////////y

H i

j____
1 Dm il e

:
D C

m uA ■
B A g —^-------
1V /////////A in m V ////////A 4K

F ig . 78 F ig . 79

one-sheeted neighbourhood of the point u = iK ' is transformed into


the one-sheeted neighbourhood of the point z = and this shows
[23] that our function f(u) has a simple pole at the point iK '. There
are analogous points in every rectangle of our net, i.e. f(u) is a fractional
function.
We shall show lastly that the function f(u) has a real period equal
to 4K and a purely imaginary period equal to i2 K '. Let us take our
net of rectangles and construct from it another net of larger rectangles,
D C__________ D '

uo- -------------------------- o U+4K

2K B Tk ’*'
F i g . 80

by linking together four rectangles sharing a common apex (Fig. 79).


This large rectangle has a side 4K in length parallel to the real axis
and a side 2 K ' in length parallel to the imaginary axis.
The transition from u to u + 4K or from u to u + i 2K ' is equi­
valent to the geometric transition into the neighbouring rectangle,
and the value of f(u) does not alter during this transition. For example
(Fig. 80) the transition from u to u + 4K is equivalent to successive
reflections in the straight lines BC and A 'D ' which give two reflections
in the real axis in the 2 -plane as well as former value of z. Hence the
618 SPECIAL FUNCTION'S [168

function f(u) does, in fact, have a dual periodicity which is expressed


by the following formulae:
/ ( “ + 4K) = f (u);
f( u + i2K') = f( u ) .
The single-valued function obtained is usually denoted as follows
owing to its analogy with sin u:
z = sin (u).
We shall meet this function again later. As a result of the con­
version of other elliptic integrals of the first kind we shall obtain
other fractional functions with dual periodicity. We shall now deal
with the general theory of such functions and of connected functions,
and for this purpose we shall somewhat modify our earlier notation.
1 6 8 . G e n e r a l p r o p e r t i e s o f e l l i p t i c f u n c t i o n s . Let and « 2 be two
arbitrary complex numbers, the ratio of which is not equal to
a real number. The function f(u) is known as an elliptic function
when f(u) is a fractional function
with two periods a>i and co2, i.e.
f(u + oq) = /( « ) ;
f (U + w2) = / (u) (30)
are identical for any u. In other words
the addition of co1 or ox, to the argu­
ment does not alter the value of the
function. From formula (30) follows
the more general formula
f(u + m 1(o1 + = f(u), (31)
where m1 and m2 are any integers,
positive or negative.
Let us now explain the geometric
aspect of dual periodicity. Draw
from a point A in the u-plane two vectors A B and AD to correspond
to the complex numbers &>1 and a>2. It is given that the relationship
tu2 : w1 is not real and therefore these vectors lie on different straight
lines and therefore we can construct the parallelogram A BCD ', by ma­
king the parallel transition of this parallelogram onto the vectors col
and co2 we can cover the whole plane with a net of similar parallelograms
(Fig. 81). The transition from any parallelogram to the neighbouring
168] GENERAL PROPERTIES OF ELLIPTIC FUNCTIONS 619

parallelogram is equivalent to the transition from u to u ± coL or


u -td jj and, as a result of dual periodicity, the values of f(u) will be
equal at corresponding points of the constructed parallelograms.
Each parallelogram is known as a parallelogram of periods of the
function f(u). Notice that the choice of the fundamental apex A can
be quite arbitrary. If, for example, we take the origin 0 as the funda­
mental apex then the apexes of our net of parallelograms will have

complex coordinates m1(ol + m2co2, i.e. these apexes give the set of
periods of the function f(u) as shown by formula (31) (Fig. 82). If we
take any point M in the w-plane and draw straight lines through it
parallel to the vectors and a>2, then the radius vector from 0 to M
gives the geometric sum of two vectors, one of which is parallel to
co1 and the other to co2; hence any complex number can be represented
in a unique way in the following form
u = kw1 -f- lio2
where k and I are real numbers. These numbers are the curvilinear
coordinates of the point u if we adopt the vectors corresponding to
the complex numbers col and a>2 as the axes of coordinates. Above we
used the phrase the corresponding points of two parallelograms of the net.
These are points, the difference of the complex coordinates of which is
equal to a period, i.e. they are expressed by the numbers 7ra1oi1 + m2ai2.
where and m2 are integers. In this sense any point in the M-plane
corresponds to a point of the fundamental parallelogram of the net.
620 SPECIAL FUNCTIONS [168

If we take, for example, the net shown in Fig. 82 in which the origin
is the fundamental apex, we can give the coordinates of any point u
in the form
u — (kx <M
1 + k2 co2) + m2 co2,
where kx and k2 are real numbers satisfying the conditions 0 < kx < i
and 0 < k2 < 1, and m 1 and m2 are integers. Notice that we ascribe
to each parallelogram one apex, and two sides which originate at this
apex. The remaining sides and apexes are obtained by the addition
of a period as mentioned above.
We shall now explain the fundamental properties of elliptic functions.
Differentiating the identity (31) n times we obtain:
/(" > (u + mx oj1 + m2(o2) = /<"> ( u ),
i.e. the derivatives of an elliptic function are also elliptic functions
with the same periods. Let us suppose that f(u) has no poles, i.e. it
is essentially not a fractional but an integral function. Its parallelogram
of periods is a bounded domain of the plane and in this parallelogram,
including its contour, it is regular and therefore also continuous and
bounded, i.e. a positive number N exists such that in the fundamental
parallelogram of periods the inequality | f(u) | < N is satisfied.
In other parallelograms of the net the values of f(u) are repeated,
and therefore the above inequality is satisfied in the whole plane,
i.e. f{u) is an integral function bounded in the whole plane. According
to the theorem of Liouville we can say that such a function must
be constant, i.e. we have the following theorem.
T h e o r e m I. I f f(u) is an integral function of dual periodicity then
f(u) is constant.
This theorem is very important owing to its two lemmas which we
shall now establish. Let f x(u) and f2(u) be two elliptic functions with
equal periods cox and co2. Assume that they have equal poles in the
parallelogram of periods, with equal infinite parts. The difference
f2(u) — f x(u) will then be a function of dual periodicity without poles,
i.e. this difference will be an integral function of dual periodicity and
it follows from the theorem that this difference must be constant.
We therefore have:
L e m m a 1 . I f two elliptic functions f x{u) and f 2(u) with equal periods
have the same poles in the parallelogram of periods and their infinite
parts are also equal, then these functions differ only by a constant term.
Let us now suppose that fx(u) and f2(u) have the same poles and
zeros of the same order in the parallelogram of periods. In this case
1G8J GENERAL PROPERTIES OF ELLIPTIC FUNCTIONS 621

the relationship f2(u) : fx(u) will have neither poles nor zeros in the
parallelogram of periods and it should therefore be constant, i.e. we
have:
Lemma 2. I f two elliptic functions fx(u) and f2(u) with equal periods
have equal poles and zeros of the same order in the parallelogram of
periods then these functions differ only by a constant term.
Place the parallelograms of periods of the function f(u) in such a
way that the poles of this function do not lie on the sides and con­
sider the integral of this function over the contour of the parallel­
ogram
f f {u) dw = J f (u) dw + j'/(w)dw-)- j f(u) dw 4 - J / (v) dw. (32)
ABCD AB BC CD DA

Let us consider the integral along the side CD and replace u by the
new variable of integration u = v + Li this case the side CD be­
comes the side BA in the o-plane and, as a result of the periodicity
of the function, we have
J / (u) du = J f (v + w2) di; = \ f (v) dv = — \ f (v ) dr,
CD BA BA AB

i.e. on the right-hand side of formula (32) the sum of the first and
third term is zero. The same can be said of the sum of the second and
fourth terms and we therefore have:
J /(w )d« = 0 , (33)
ABCD

i.e. if the poles of the elliptic function f(u) do not lie on the contour
of the parallelogram, then the integral of this function round the
contour of the parallelogram is zero.
Consider the complex number a which is such that the equation
f(u) — a = 0 has no zeros on the contour of the parallelogram. Apply­
ing the above result to the elliptic function

<p(u) = r («)
/(« ) — «
we obtain
Jf 7f ^ - d « = o.
flu) — a
ABCD

The above integral expresses, as we know, the difference between


the number of zeros and poles of the function f(u) — a [2 2 ] and we
can therefore say that the number of zeros of the equation f(u) = a
622 SPECIAL FUNCTIONS [168
is equal to the number of poles of f(u) — a, or, which comes to the
same thing, to the number of poles of f(u), i.e. the function f(u) takes
the value of a and infinity an equal number of times inside the
parallelogram.
This was proved for an arbitrary a on the assumption that the
equation f(u) = a had no zeros on the contour of the parallelogram.
If this is not so then we can slightly displace our parallelogram so
that the zeros of the above equation do fall inside the parallelogram
and the poles remain inside as before. The above result will be valid
for the displaced parallelogram. It can easily be seen that it will also
be valid for the initial parallelogram, provided that when counting
the number of zeros of the equation we ascribe one apex and two
sides, originating at this apex, to the parallelogram. Notice, that when
the equation f(u) = a has the zero u = u 0 and the following expansion
applies in this neighbourhood

f (u) = a + ck ( u — u0)k + ck+1 (u — u0)k+1 + . . . (c* # 0 ),

then this zero must be regarded as a zero of order k of the function


f(u) — a or of the above equation. In all such cases we obtain, from
the above considerations, the following theorem.
T h e o r e m II. The elliptic function takes any value (finite or infinite)
an equal number of times in the parallelogram of periods.
If f{u) takes any value m times in the parallelogram of periods,
then it is known as an elliptic function of order m. Such a function
transforms the parallelogram of periods into an m-sheeted Riemann
surface. The conformity of the transformation may be affected only
at points where f'(u) vanishes or where f(u) has multiple poles. These
values of u correspond to branch points on the above Riemann
surface.
We shall now show that the positive integer m cannot be unity.
In fact, it follows directly from formula (33) that the sum of the residues
of an elliptic function at its poles inside the parallelogram of periods
should be zero. If we had m = 1 then the function f(u) would have one
simple pole inside the parallelogram of periods which contradicts the
above result. We thus see that an elliptic function of the first order
does not exist. Later we shall, in fact, construct elliptic functions of
the second order. It can be shown that elliptic functions of the second
order are obtained by converting elliptic integrals of the first kind.
There are, of course, elliptic functions of higher orders.
169] FUNDAMENTAL LEMMA 629

1 6 9 . F u n d a m e n t a l l e m m a . Consider the elementary function sin u.


This is an integral function with simple zeros at the points u = lcn
(fc = 0 , ± 1 , . . . ) on the real axis, which are at a distance n from
each other. Two other fundamental functions
(sin u)' cos u I
cot u and — (coti/)' (34)
sin u sin* u
have simple and double poles at these points. The function sin u
can be represented by an infinite product and we had previously an
expansion into partial fractions for the functions (34). In the next
section we shall by an analogous process construct an integral func­
tion with simple zeros at the points
OTjCD, + m2co2, (35)
where co2 and oj2 are complex numbers, the ratio between which is not
equal to a real number, and m1and m2 are any integers. The points (35)
are the apexes of the net of parallelograms shown in Fig. 82. To
construct this integral function we shall use the Weierstrass formula
which gives an integral function in the form of an infinite product. To
be able to apply this formula we have to find a number p which is
such that the series
^ I ropu, + rn2co2 |P (®®)
mlt m.
converges. The apostrophe near the symbol of summation shows
that summation includes all integral values of m1 and m2 except
m1 = m2 — 0 . Similar conditions will apply in future in all cases
where there is an apostrophe near the symbol of a sum or product.
We must rewrite the sum (36) in the form
i
2 ' <5m (37)
m|| fii| £,i, /Tig
where bm m is the distance from the origin to that apex of the net
in Fig. 82 which corresponds to the complex coordinate m ycol m2 co2.
Let 26 be the shortest distance between any apex of the net, other
than the origin, to the origin. This number 26 will obviously also be the
shortest distance between two apexes of the net. Imagine that two
circles are drawn in the plane in Fig. 82, centres the origin and radii n
and (n + 1 ) respectively, where n is an integer which satisfies the condi­
tion n > d. Let K n be the annulus between these two circles. Let us
find the approximate number of apexes of the net in the annulus K n.
624 SPECIAL FUNCTIONS [170

Let this number be tn. Draw circles of radius 8 with centres at the
apexes of the net which lie inside the annulus K„. It follows from the
definition of 8 that these circles will not overlap and their common
surface area 7i82 tn will be less than the surface area of the annulus
with inside radius (n — 8) and outside radius (w + 1 + 8), i.e.
n (n + 1 + <5)2 — n (n — <5)2 > 7id2tn
or, after elementary calculations,

t„<A,n+A, ( Ai =

For any apex inside the annulus K n the distance 8mifn,t will not
be less than n and therefore the sum of the terms of the series (37 ),
which corresponds to the apexes in the annulus K n will, in any case,
be less than
d |n -|- A 2 A, A 2
tip nP—1nP

This result will be valid for terms of the sum (37) for which 8m^mt
is sufficiently large with respect to for example > <5 + 1.
In this case the corresponding points are bound to lie inside the
annulus K n when n > 8. Rejecting a finite number of terms of the
series (37) and replacing the remaining terms by greater terms, we
obtain a domineering series

As we know from [I, 122], this series converges when p > 2 and,
in particular, when p = 3 and we can therefore draw the conclusion:
F u n d a m e n t a l Lemma. The series (36) converges when p > 2 and,
in particular, when p = 3.

1 7 0 . T h e W e i e r s t r a s s f u n c t i o n . To simplify the notation let us put


w = m1w1 + m2co2. (38)
Bearing in mind the fundamental lemma we can construct directly
an integral function with simple zeros at the points (38). This function
will be given by the following formula [69]:
170] THE WEIEBSTKAE8 FUNCTION 625

where the infinite product includes all pairs of integral values of mx


and m 2, both positive and negative, except = m2 = 0 .
As we know from [6 8 ] we can evaluate the logarithmic derivative
of this product in the same way as of a finite product; the logarithmic
derivative of an individual factor of this product will be
l tt _1_
w ' w2 w - J—- W+
U —
W + —
W2
1 - —
w

We thus obtain a second function

O ' (w )
C («) = u- + 2 f \u
— — id+ —w (40)
o (W)

which has simple poles at the points (38). This function is obtained
from o(u) in the same way as cot u is obtained from sin u. Bearing
in mind the convergence of the series

^2 —
14,i la ’
fflu '
it can readily be shown that the series (40) converges uniformly in
any finite domain if a finite number of terms with poles in this domain
are rejected. Differentiating the function (40) and changing the sign
we obtain a new function

P («) f ' (“ ) u2 + ^ [(« - w ) 2 w 2] •


(41)

This new function is obtained from £(w)in the same way as ( 1/sin2 u)
is obtained from cot u. It has double poles at the points w. The series
(41) also converges uniformly in domains of the kind mentioned
above [ 1 2 ].
We shall now explain some fundamental properties of the introduced
functions. Let us write the expression for a(—u):
_ U_ _1_/JJ\2
o(-to = - « Zm„ m,7 '(' i + - K
1
" + 2U ■
Owing to the fact that the product includes all pairs of integers
m1 and m2, except m1 = m2 = 0 , we can interchange the signs of
m1 and m2, i.e. change the sign of w, whence we obtain
U 1 /U\2
a ( - u ) = - u ] J (l - - ^ ) e - + ^ W = -<T(M),
m,, mt
626 SPECIAL FUNCTIONS [170
i.e. a(u) is an odd function. It can be shown similarly that £(u) is
also an odd function and that p(u) is an even function. This can, in fact,
be obtained directly from the formulae

f (“ ) = > P («) = - £' («)• (42)

since the differentiation of an odd function gives an even function


and vice versa. It also follows directly from the formulae which
define the introduced functions that
g(«) (u)
u
UL, w2|P(«) u=0= I . (43)

The functions a(u) and £(tt) cannot have periods cox and co2 since
the first is an integral function and the second has one simple pole
in the parallelogram. We shall show that the function p(u) has
periods to, and co2. To do so construct, to begin with,

I? (U ) ~ — ^ (M — u > )3

or
Iff' _ _ 2 'V 1 - 2 V _______ 1______
® ( u — id )3 (m — m ,( o , — m j i o . ) 3 1
f77|, rH| ni|, mt

where summation includes all integer values for m1 and m2 without


exception. Hence
___________1________
P' (u -j- aij) 2
2
/71|, /7la
(u, + a>, — 171,(0, —Wj(o2)3

_________ i__________
= - 2 2
/rij, m.
[ w — ( m , — l ) ( o , — m »(o 2] 8 '

If m1 runs through all integral values then the same can be said
of (mj — 1). We thus have
P' (« + o»i) = jp' («) -
and it can be shown similarly that p'(u + co2) = p'(u). Hence

P’ (« + <ok) = jP' («) (* =1- 2). (44)


Let us now investigate the variation of the function (p(w) when
w1 and co2 are added to the argument. Integrating (44) we have
fP(u cok) — p (u) -j- Ck,
170] TH E WEIF.RSTRASS FUNCTION 627

where Ck is a constant. Let us suppose in this identity that


u = —oj^/2 , where tofc/2 is not a pole of jp(u):

Owing to the fact that the function jp(u) is even we have


j()(—cofr/ 2 ) = ty(oikj2) and therefore Ck = 0 , i.e.

jj>(tt + a,Jk) = t»(«) (*=1,2). (45)

Let us now investigate the variation of the function L,{u) when


tofc is added to the argument. We have from (45) and (42)

r (u + cok) = c' («).


Integrating this we obtain
C(u + 0}k) = c («) + % (* = 1. 2 ), (46)

where rjk is a constant, i.e. the function £(w) acquires a constant term
rjk when the number cok is added to the argument. A more general formula
also follows from formula (46):

C(u + m Ltol + m2o>2) = C(u) + mxr)x + m 2t)2, (47)

where mx and m2 are arbitrary integers.


The number T]k can be defined as a particular value of the function
C(«), viz. assuming in formula (46) that u = —o>fr/2 and remembering
that the function f(«) is odd, we obtain

Vk = 2 C (^ ) (*=1,2). (48)

Let us now turn to the function a(u). As a result of (46) and (42)
we can write
o' (u + wk) _ o' (u) |
o(u + iok) o(u) ■ lk'

Integrating this we obtain


log a (u + cok) = loga(u) + t]k u + Dk
or
a (u + <ok) = Cke w a(u),
628 SPECIAL FUNCTIONS [170

where Ck = eD ia a constant. To determine this constant we suppose


in the above identity that u = —<ufc/ 2 :
_ Vk*k
a [ ^ - ) = Cke 2 ff(--f-).

Remembering that the function a(u) is odd and simplifying by


the factor a{a}k/2) which is not zero we obtain

ck = - e 2 ,
and finally

a (u + <ok) = — e’1*(U+ 2 ^a(u) (k = 1, 2). (49)


Conclusion: the function o(u) is multiplied by an exponential factor
when wk is added to the argument. Instead of formula (49) a more general
formula can be obtained, which is similar to formula (47), viz.:

i (“ + -j-)
cr (it. + u>) = £ e v a{u) , (50)
where
w = mjW; + m2co2, y = m1y l + m2y2
and s = + 1 or e = —1 , depending on whether both integers mx and
m2are even or not. In the last case the relationship (50) follows directly
from (47) in the same way as (40) follows from (46). We shall only use
this case in future when m1 = m 2 = 1 .
In conclusion of this section, we shall introduce a relationship
connecting the constants cok and rjk. To begin with we shall introduce
a rule for the notation of the periods co1 and tu2. Consider the
fundamental parallelogram ABCD (Fig. 81). One of its sides AD
makes a positive angle smaller than n with its other side A B . We shall
always assume that corresponds to that side AB , from which
the angle is measured and (o2 to that side AD of the parallelogram
to which the positive angle, smaller than n is measured; the amplitude
of the fraction co2la>1 will then be equal to an angle between 0 and
7i, i.e. the imaginary part of this fraction must be positive. The
imaginary part of the reciprocal fraction cojcoj will evidently be
negative. Hence we shall always denote the numbers a>k in such a way
that the relationship c o j^ has a positive imaginary part. Let us now
construct a parallelogram with the principal apex at A(u = u 0) so that
the pole C(u) lies inside it. This single pole has, from (40), a residue of
unity and, as a result of the fundamental theorem of residues, the
171] TH E D IFFERENTIAL EQUATION FOR fy ( u ) 629

integral of the function £(a) round the contour of the parallelogram


will be equal to 2 m, i.e.
U , + a ), u 0 + o>, + m , U . + O'l

J £ (w) d a + j £ (a) d a + J £ («) d a +


U« U0 + ®1 Up + <°1 + ®i

Up
+ J C (a) da = 2m.
U . + <0,

Replacing the variable a in the second of the above integrals by


a = Vi + aij and the variable a in the third integral by the new vari­
able a = v2 + co2, we obtain
U» -f a>i u 0 + «>t u#

J £ (a) da + j £K + wj -f J £ (*>2 + wi) +


U9 Up Up + 0>a

-)- fr" £ (a) da = 2m,


U , + CO,

where we integrate along straight lines or by changing the sign of


the variable of integration
U0 + <02 Up + <0,

I [f(« + Wj) — £ (a)] da — J [£ (a -f a>2) — £ (a)] da = 2m .


Uo U,

This, from (46), gives the required relationship between cok and rik:

V ^ 2 — rl2a)l = 2 7 li- (51)


This relationship is usually known as the Legendre relationship.
The functions <x(a), £(a) and p(u) were first introduced by Weier-
strass. It is evident from their definition that any two complex
numbers coJL and co2 can be used for their construction, the only
necessary condition being that their ratio is not be real; these Weier-
strass functions are not only functions of the argument a but also of
the complex parameters and co2 which we just mentioned. For this
reason they are sometimes denoted as follows:
<r(a; cov co2); £ (a; co1( co2); p(a; mv co2). (52)

1 7 1 . T h e d i f f e r e n t i a l e q u a t i o n f o r $>(a). Having established the


fundamental properties of the Weierstrass functions we shall now
study in greater detail the function p ( u ) and, in particular, deduce
a differential equation of the first order for this function. First of
all we shall derive the expansion of the function j?(a) near the point
630 SPECIAL FUNCTIONS [171
u = 0 which is a pole of the second order of this function. To do so
let us turn to the fundamental formula (41). We have near u = 0 :
_ L_ = _L JL + . . . ..
w —u w w* ' wn^~i
and differentiating with respect to u we obtain
___1___ = J _ , 2 ^ (« + 1)m" ,
(w — u )z w1 ' ufl r ■• • i tvn +2 *

This, from (41), gives the following expansion for )?(«) near the
origin
1
P(u ) = ~& + 2 1( n + l ) un TTl\,
2 TT\%

When n is odd, the summation in terms of m 1 and m2contains terms


which are equal in value and opposite in sign in pairs so that we
have
P («) = ITT + r2 « 2 + cau* + . + cnu2n- 2 + . . . , (53)
where
I'5
w

(54)
II

cn = (2 n - )2 ' i
3

Let us now establish the form of the expansion for ty'2(u) and
$3(u). We evidently have

P' (u) = — + 2 c2u + 4cau3 + . . .

where in the last two expressions terms in positive powers of u


are omitted. Hence
jp'2 (u) — 4jp3 (u) + 20c2p (u) = — 28ca + . . . , (65)
where terms in positive powers of u are again omitted. Therefore
the point u = 0 will no longer be a pole of the expression on the
left-hand side. This expression will be an elliptic function without
poles in the parallelogram of periods since the only poles of the
function ty(u) are at the point u = 0 and at the corresponding apexes
of other parallelograms. For this reason the expression (55) (an elliptic
171J Till! DIFFERENTIAL EQUATION FOR p (u ) C31

function without poles) will be constant. But the right-hand side


becomes —28c3, i.e.
P'- (M) = 4fc>3 («) — 20c$ (u) — 28ca.
Let us introduce the notation:

g2 = 20c2 = 60 2 ' 9* = 28c3 = 140 2 ' i • (56)


TTl\f /77j 7771, 77J|

The above calculations give rise to the following theorem.


T h eo rem . The function p{u) satisfies the differential equation

P"1 (») = 4fP3 M ~ ffzP iu ) ~ 03 • (57)


The numbers g.z and g3 are known as invariants of the function p{u).
The function p(u) has only one double pole in the parallelogram
of periods with its main apex at the point u = 0 . The remaining
apexes should no longer be ascribed to this parallelogram and there­
fore p(u) is an elliptic function of the second order; any equation
p(u) = a for any given value of the complex number a will have two
zeros in the parallelogram of periods.
When p{u0) — a and p '(u 3) = 0 then the zero u = u 0 must, in any
case, be a double zero. It can not, however, be of an order greater
than a double zero since p(u) is a function of the second order and
therefore it is equal to a only at one point u — u 0in the parallelogram.
If p'(u0) # 0 then the equation p(u) = a will have two different simple
zeros in the parallelogram. Let us now see where those values of u
for which p'(u) = 0 lie. Assuming in the identities
P' {u + ««*) = P' («) or p’ (u + Wi + a>2) = p' (u)
that u = — cok/2 or u = — (ai1 + co2) 2 we obtain, owing to the fact
that p'{u) is odd
jp '(^ )= 0 ( * = 1 ,2 ) and p 'p l+ ^ ) = 0, (58)

i.e. p'(u) vanishes in the middle of the sides and in the centre of the
diagonal of the parallelogram with the main apex at u = 0 . Consider
the values of the function p{u) at these points

•’ B r ) = I V p (t )-* > - (» )
Each equation p(u) = ek has a double zero at the corresponding
point. Bearing in mind that p{u) is a function of the second order we
can say that the numbers ek are different.
632 SPECIAL FUNCTIONS [172

Let us now turn to formula (57). The right-hand side of this formula
is a polynomial of the third degree in §(u). Assuming that either
u=cokl2 or u = (co1-+- C'j)/2 we can see that this polynomial vanishes
when $(u) = ek since for the given substitution the left-hand side of the
equation (57) vanishes and §{u) becomes ek. Factorizing the polyno­
mial we can rewrite formula (57) as follows:

V'2 («) = 4 (P (u) - e2) (<p(u)-e2) («) - e3) . (60)


Comparing the right-hand sides of the formulae (57) and (60) we
obtain formulae connecting the numbers ek and the invariants g2
and g3:

6k 6^ 63 ~ 0, “1“ ^ 2 ^ 3 + ^3 ^ 1 — £7 2 ’ = ^ 3 * (6 1 )

If we put x = §{u) then the equation (57) can be rewritten in the


form
(£ )’= 4 3 * - ^ 3 .
Bearing in mind that when u = 0 we have x = ° ° , we separate the
variables and integrate to obtain
X

u V 4x»- g2x -=rj3 (62)


i.e. the function ty{u) is obtained as a result of the conversion of an elliptic
integral of the first kind (62). It can be shown conversely that by choos­
ing the constants g2 and g3 in such a way that the polynomial under
the radical has no square zero, the conversion of the integral (62) leads
to the Weierstrass function §(u).
It can be shown further that any elliptic function with periods w1
and co2is a rational function of §{u) and p'(u) so that the set of rational
functions of fy'(u) and fp(w) represents the full set of elliptic functions
with periods cox and co2.

172. The functions ak(u). It follows directly from formula (60) that
the product on the right-hand side is the square of a single-valued
analytic function §'{u). It appears that the same can be said about
each of the factors §{u) — ek. The same thing also applies to trigono­
metric functions
(cos u)’2 = sin2 u = (1 — cos «) (1 -j- cos u),
172] THE FUNCTIONS a^u) 633

for each factor on the right-hand side is the square of aj single-valued


analytic function

1 — cos u = 2 sin2 and 1 + cos u = 2 cos2 .

To prove our hypothesis for the difference p(u) — ek we shall deduce


an auxiliary formula. Consider the difference
P (u )-p (v ) (63)
as being a function of the argument u. In the parallelogram with the
main apex at u = 0 it has a double pole u = 0 . Owing to the fact
that the function p(u) is even, the zeros of the function (63) lie at the two
points of the parallelogram which correspond to the complex numbers
u = diu>i.e. strictly speaking, the zeros lie at points of the parallelo­
gram which differ by a period from If such a point of the parellelo-
gram proves to be a period then the above two points coincide and
form a double point in the way explained above. Together with the
function (63) let us also consider the function

t w = ' {uS n + ’) ■ <M >


Let us prove, first of all, that this latter function also has the periods
£o1 and co2. We have from (49)
t i „. \ _ 6 (“ - v + co4) a (u + v + o)k) _
t (u + cok) = ----------ff*(u + t0fc)--------------

i)*(u - o + ^y2 /<T(w


©V ) , —tt)e \ « (“ \ + v + -y2 ' <r (u + v ) _a (u — v) a (u + v)
— o^Tm) = /(«)
e 2it (\ « * / +a* (u ).

Therefore the function (64) does, in fact, have periods a>k and co2.
It follows directly from formula (64) that it has a double zero at u = 0 in
the fundamental parallelogram and two zeros at points of the parallelo­
gram which differ by a period from ± v . All this follows from the
position of the zeros of the function'a(M), viz. it is due to the fact that
these zeros are simple zeros equal to w = -f- m2 co2. Therefore
the functions (63) and (64) both with periods cij and co2 have the
same poles and the same zeros of the same order in the fundamental
parallelogram. We can therefore say that these functions differ only by
a constant term [168]:
634 SPECIAL FUNCTIONS [1 7 2

To determine the constant C we multiply both sides by v? and sub­


sequently put u = 0
Ca (u — v)a(u + v)
un-p (u) — u2p (u)
u= 0 m u= 0
We have from (43)
1 = Co (— v) a (v) = — Co2 (v),
and we finally obtain the required formula

To investigate the difference ty(u) — ek we only have to assume in


formula (65) that
lUt tOi + to2
V= T : *= 2
and v —~ .

Thus, for example,

i>(«) = f (*) - f ( v ) --------S p la2— J • (66)


(u)

or, bearing in mind that we have from (49)

<* + -y-j = o [u — (o1j = — e” 1 (u r + t ) a[u - ,


i.e.
(67)
°(u + ^-) = - e’,,U(T (M- -T-) 1

instead of (6 6 ) we can write

{(>(u) — e1 = e’’lU » ( - * )
. f i 1-) «* (“)
or

P («) - ei =
(^)e(u)

Two other differences can be analysed similarly. We therefore obtain


the following representations for fy(u) — ek as the square of the quotient
172] THE FUNCTIONS <r*(u) 635

of two integral functions:


f <” > - < * = [ ? $ ] ' , <“ >
where we have used the following notation:

Let us now establish certain properties of the function a(u). These


functions are obviously, integral functions and, putting u = 0 , we
have
<rfr(0) = l (*=1,2,3). (70)
Rewriting the relationship (67) in the form

<T(i 2 ‘ ~ “) = e“''lU£r(i i L + w) ’
we obtain

/ \ — t ,,u I 2 J / \
(«) = e 2 i— r — = <rx ( - «)
e[ 2J
and the same applies to the two other functions ak(u), i.e. the functions
a(u) are even integral functions.
Substituting the expressions (6 8 ) in the right-hand side of formula
(60) and extracting the zero we have

i n ' < * .\ __ I o ffl (M) a 2 (u ) a 3 (u )


” m - ± ^ 5 •

To determine the sign we multiply both sides by u3 and sub­


sequently put u = 0 . Bearing in mind the expansion

P' (u) = ~ ~ r + c « + 4c3u3 + . . . ,


2 2

and also the formulae (70) and (43) we can see that we must take
the — sign in the above formula, i.e.
636 SPECIAL FUNCTIONS [1 7 3

173. The expansion of a periodic integral function. The integral


function a(u) has no periods at all. We shall show later that by mul­
tiplying it by an exponential factor we can obtain a periodic integral
function. At present we shall consider a periodic integral function
and deduce an expansion for such a function in the form of a power
series or a Fourier series [cf. 119].
Let us suppose that the integral function y(u) has a period co, i.e.
for any complex u
<p(u + co) = <p(u). (72)

We draw a vector to through the origin and two lines perpendi­


cular to it through its ends (Fig. 83). The latter lines form the
period band of the function <p(u). The line CD is obtained from the line
A B by the transformation u ' = u + to.
Let us apply the transformation
r = u • 2nij(o to the M-plane. In the plane
r = r! + ir 2 our band will be transfor­
med into a band bounded by the lines
r2 = 0 and r2 = 2 .'t. If we subsequently
perform the transformation
2 n iu

£ = e1 = e " ,
then in the new C-plane our band will
be represented by the whole plane cut
F i g . 83
along the positive part of the real axis £
with the exception of the point £ = 0
[19]. The edges of the cut correspond to the two lines which boun­
ded the original band in the w-plane and the corresponding points
on the two edges refer to values of u connected by the relationship
u ' = u + o). As a result of (72) our function has equal values on
both edges of the cut and therefore the values of the derivatives of
all orders will also be equal, i.e. our function will be single-valued and
regular not only in the cut £-plane but in the whole £-plane except at
the point £ = 0. We can therefore say that it can be expanded in
this plane into a Laurent’s series

+ 08 + CO 2 ti iu

< p{u)= y a„ £ " = 2 ane~"_ n • (73)


t l aa CO n — «»

We thus derive the following theorem.


174] THE NEW NOTATION 637

T h eorem . A ny integral function cp(u) with a period co can be repre­


sented in the whole u-plane by the following series
+ op 2 n iu

?>(«) = ^ a ne ~ n ■ (7 4 )
n - — oo

The above series must converge uniformly in any bounded part


of the plane. If we use Euler’s formula and group together terms with
corresponding values of n, equal in magnitude but opposite in sign,
we obtain an expression for the function <p(u) in the form of the trigo­
nometric series
. . . I f , 2nun , ,, . 2mm)
<P(«) = a0 + £ 009 — + hn sm — - (75)
n=1 V J
where
a'n = an + a_n\ b'n = i (a„ — a_„) (71=1,2,3,...). (76)

174. The new notation. When the theory of elliptic functions is


studied in detail it contains an extensive formal apparatus which is
used in the applications of these functions. Unfortunately not all
authors keep to the same notation. We are giving here the mere basis
of the theory and therefore we shall not give numerous and frequently
evry useful formulae which appear in the theory of elliptic functions.
However below we shall deal with a more complicated formal appa­
ratus than we have used until now. We shall therefore give the nota­
tion due to Jacobi which is systematically explained in the book by
Hurwitz The General Theory of Functions and of Elliptic Functions.
We shall follow the treatment described in this book in the next
few sections.
In future we shall frequently deal with halves of the numbers cox
and 0)2 and therefore, to abolish the use of fractions, we introduce the
following symbols for co^ and co2:
0)1 = 2(o; co2 = 2co'. (77)

In relation to this we also put


Vi = Vi = 2v ‘ ■ (78)
In future the fundamental element in the construction of the
functions will not be the numbers a>1 and co2 themselves, as for ex­
ample in the case of the function fy(u), but their relationship
638 SPECIAL FUNCTIONS [175

or another number directly dependent on this relationship:


h = e,m . (80)
We also replace the argument u by two other arguments

v= ; z = e17"’ = e 2<“ (81)

The above notation affects the symmetry of the numbers co and co',
i.e. these numbers play different parts when the above notation is
used. We always assume, as before, that in the relationship co'/co the
coefficient of i is positive, i.e. if it is assumed that co'/co = r -\- is,
then s > 0 , and therefore
\ h \ = e-*s < 1 . (82)
For these values of col and co2 we had Legendre’s relationship (51)
which, by using the above notation, can be rewritten in the form:

rjco' — 7]'co = — . (83)

Notice certain consequences due to the use of the above notation.


We have from (81):

^ = »+ ^ 2 co
= »+l; e '-M U iz ; e,n(<’+1) = — z,

and similarly
il -4- G )' . T U -4- . (n ( v + — \ 7 — /«■/«•_i__\ 7
—^ r = v + ~2 ~'> 2 to = v + x> e v 2) = hi z; e M(v+I) = hz,

i.e. for example, the addition of co to u is equivalent to the addition


of 1/2 to v or to the multiplication of z by i; the addition of co' to u is
equivalent to the addition of r/2 to v or to the multiplication of z
by h1/2. Notice in conclusion that we shall always denote the powers
of he and 2®by einzQ and e‘Me,

175. The function ^(v). Using the new notation we can express
the fundamental property of the function a(u) as follows:
a (u + 2co) = — e2,?(u+'1’) a (u); a (u + 2co') = — e2,'(u+“') cr [u). (84)
Add to the function a(u) the exponential factor
<p(u ) = eau‘+buo (u) (85)
175] THE FUNCTION «,(r) 639
and select the numbers a and b so that the new function <p{u) has
a period 2co. We have from (84)

(p(u+ 2co) = — ea(u+2“)2+6(u+2“)+2’'(u+<") a (u) =


__ __ g4a«)U +4<i^3+ 2 & a i+ 2 ^ (u + a ))0 a u 2 + tu ^

9> (u + 2cu) _ _ p 2 ( 2 a m + 7 i) ( u + a ) + 2 b a >


( 86 )
<p(u)
and similarly
<p(u + 2 c u ') _ _ _ 2 (2 a m '+ i!-)(u+<u/) + 26<o'
<p(u) (87)

In formula (8 6 ) the index on the right-hand side is a poly­


nomial of the first degree in u. In order that the right-hand side
should be unity for any value of u it is necessary to equate to zero
the coefficient of u in the index while the constant term in the index
should be equated to kni, where k is an odd integer. In relation to this
we assume that
>1 , ni
a = ~ % b =2 ^ '
Substituting in the right-hand side of formula (87) we have from (83):

<p(u + 2oS)_ - £ ( « + *') + » / £ _ y—2


“ *>(«) “ _ e _ „
We thus see that the function
f?ua in u yu*

<p(u) = e 2" 2m a(v) = e 2“ za(u) ( 88 )

satisfies the following equations


<p(u -|- 2 a>) = (p (it), <p(u -(- 2 a/) = — z~2q>(u). (89)
Since <p(u) is an integral function with a period 2 co, we can expand
it in the form [173]
+ sc 2 n iu + no
cp(n)= ^ ane~ton = y a nz2‘

Also the addition of 2co' to u is equivalent to the multiplication


of z by h, i.e.

(p{u + 2 w') = y anh2nz2n,
n =* — <
640 SPECIAL FUNCTIONS [176

and the second of the formulae (89) gives

+2 a > 2nz2" = - +
2 anz*"~\
n = —» n = — oo
or, replacing in the latter sum the variable of summation n by n + 1 ,
+« +®
> ’ anh V = - 2 °»+i!Jin

Comparing the coefficients of like powers of z we obtain

an+1 = - h ^ a n = - + an ,
which can also be written as follows:

( _ !)" + !* -(n+ ^)2an+1 = ( - 1 )"A“ (n“ ^ )an.


We thus see that the expression

( -

must have the same value for all integral values of n. Suppose that

(- 1 )"*"('""*) an = Ci,
where C is a constant. This gives us the following expression for the
coefficients in the expansion of the function <p(u):

an = (— 1 )nl l n~*) Ci,


and therefore:

<P (w) = Ci 2 ( - 1 )nh{n~ T ) z 2n. (90)

Formula (8 8 ) gives, at the same time, the following expression for


the Weierstrass function a(u):

a(u) = e 2<“ 2~V (u ) ■


Here we must introduce a new function

#i [ v ) = i +
2 (- 1 )nh (n~ TJ z 2n- 1, (92)

which is connected with the function a(u) by the following relationship:


rju*
a(u) = e 2a C&1 (v). (93)
175] THE FUNCTION # 1(0) 641

Let us now determine the constant C. Bearing in mind that u = 2cov


which, from (93), gives ^(0) = 0, and that the relationship ^ ( v ) ^ ,
when v = 0 , is equal to #((0 ), we obtain, on dividing both sides of
formula (93) by u and then putting u = 0,

i= -^ (° ),
and finally
7) 11*
2co
o(u) #i (v)• (94)

Let us now transform the power series (92) for the function
into a trigonometric series. To do so we must group together terms
of the above expansion referring to powers of z which are equal in
value but opposite in sign. Denoting by v the odd positive number
11= 2 )1 - 1 (n = 1, 2, 3, . . . ), whence n = —v + 1/2 we can write:

r l , 3 , S, . . . p+1 p* 1, 3, S, . . . -P + 1 P*
(v) = i ^ (— 1 ) 2 h* z”+ J? (— 1 ) h* z~‘
2
I'
where each sum is over the odd positive numbers i.e. v = 1, 3, 5, .. .
Bearing in mind that
V -j- 1 V+l —P+ 1 P-- 1

( - 1) ~ = ( - 1 y ( - = - ( - 1 )- 5 ” = - ■ ( - 1 ) 2

and
zv — z~v = e,wro — = 2 i sin vnv,
we can rewrite the above formula as follows:
1, 35, . . . p— 1 p>
#i (v) = i > ’ (— 1 ) ~ h * (z~v — z’),
or
1 ,3 ,5 ,... v— 1
■&x (d) = 2 ^ (— 1 ) 2 h i sin vnv =
9

1 9 25
= 2 [A4 sin nv — h 4 sin 3 nv h* sin 5nv — . . . ].
The function which is usually known as the first theta function
is an odd integral function of v. To construct this function we only use
one complex number r which, according to the given conditions,
lies in the upper half-plane, i.e. its imaginary part must be positive,
where h = elnT. Therefore the theta-function is sometimes denoted
by ^(v ; x),
642 SPECIAL FUNCTIONS [176
176. The function Above together with the function o(u),
we introduced the three integral functions ak(u). This, naturally,
indicates that we should also introduce three theta-functions together
with the function
Using our new notation we have
a (cu' — u)
a3 (u) = e"'u a (co')
or from (93)
O *l'u + fj (co'2o—
>
tl)*
a 3 ( u )
a (co')
e

Removing the brackets of the power index and replacing uj2co


by v and cd'/co by r we obtain:
TlU* (ij'cu—IJOl') -H.
T
a3 (u) = C3e 2a> e “
)■
where C3 is a new constant. Finally, using the relationship (83) we
obtain an expression for the function o3(u) in terms of the new theta-
function :
— r \
a3 («) = C3e z - 1 i?! . (96)

Similarly we obtain for a2(u):


, v jjUo(a> — u)
= e -V (S T ’
where, for the sake of briefness, we have put
rj = 7) ?]'; to = co -)- co'.
Formula (93) then gives us
(a*—u)>
•/«T»r------
a (cb)

and making calculations analogous with those above we finally obtain


VU*
<t2 (u ) = C2e 2“ z -1 (-g- + —---- v] • (97)

We have similarly

<7i (m) = C i e ^ r ??i(— - u ) . (98)


176] THE FUNCTION ot (i-) 643

Let us now deduce the expansion into a power series from the values
of the theta-functions entering the expressions for the functions crk(u).
We have

But, from (81), the subtraction of 1/2 from v is equivalent to the


multiplication of z by (—i) and, therefore, recalling (92)

Q._ _ i ( - l ) ^ ( n~ i)2(-iz )2"-1=


n = — eo

= (99)
n = — cd
Similarly
^ i ( 4 + - r - v) = — — -I— J -),
and the subtraction of ( 1/2 + */2 ) from v is equivalent to the multipli­
cation of z by —i • h~llz.
It follows that

(— + \ - u) = - i +y; { - 1 )nh(n ~ t ) \ - i h r t z ) 2»-' =


n ■» — «

= r lz h ^ - v 2z2n- 2
n ■=>— oo

or, replacing the variable of summation n by n + 1 .

$ A ± r + Ar - v) = h - T z A»y», (100)
n = —oo
and similarly

-&1 = ATiz +
j? ( - 1 )nhniz2n. (1 0 1 )
n = —oo

We introduce the three new theta-functions


+» / iY
$2 (v) = 2 A l " " ! ) z2n- \
n = —«

03 (t’) = A"2z2", ( 102 )


n = —oo

04(w)= +>' (— l)nAnV n.


644 SPECIAL FUNCTIONS [170

In this case the above formulae for ak(u) can be wrtten in the form
rjU* rjU% iju *
a±(u) = Cl e 2" #2 (v); <r2 (u) = C2e 2m &3 {v); az (u) = C'3e 2“ #4 (v).

where and X}3 are new constants. To determine the constants we put
v = 0 . In this case u = 0 and ak(0 ) = 1 ; hence
1
Cl 0 , (0 ) ’ — 0, (0 ) ’ Ca = K (0 ) ’
and we have finally
m* 02 (v) 03(u) *>“* (V)
<yl (u) = e !“ #, (0 ); a2 (u) = e 2<u 03 (0 ); a3 (u) = e 2" 0t (0 ) ' (103)

Sometimes &0(v) is written instead of # 4(u).


The power series ( 1 0 2 ) for theta-functions can easily be transformed
into trigonometric series in the same way as those of the functions
^ (v). We thus obtain:
_1_ 9_ 25
•&2 (v) = 2h* cos nv + 2h* cos Znv + 2A4 cos 5nv + . . .
# 3 (t>) = 1 -(- 2 h cos 2nv + 2 h* cos 4nv -j- 2 h9 cos 6711; + . . .
(104)
# 4 (v) = 1 — 2 h cos 2nv -f- 2h4 cos — 2 A9 cos 6nv + . . .

In future, to simplify notation, we shall not write the argument


v = 0 , i.e. instead of 0 {(O) we shall simply write •&[ and instead
of #k(0 ). It follows from (95) and (104) that we can write the follow­
ing expansions for these values
_1_ 9_ 2S «
= 2n (h* — 3h* + 5h* — Ih* + . . .)
_1_ 9_ 25 49
&2 = 2 h* + 2 h* + 2h* + 2h* -f . . . (105)
# 3 = 1 + 2h + 2 A4 + 2h9 + . . .

= 1 — 2h + 2h* — 2 A9 + .. .

These series converge very rapidly since it is given that | h \ < 1


and the sum of these series are regular functions of r which are deter­
mined in the upper half-plane.
It is not difficult to establish the connection between the Weierstrass
function ty(it) and the theta-functions. We had earlier

Zfp («) - ek a (u) ’


177] THE PROPERTIES OP THETA-FUNCTIONS 645

and, bearing in mind the expression for the functions a(u) and ak(u)
in terms of theta-functions we obtain

( 1 M >

177. The properties of theta-functions. All theta-functions are


integral functions of the argument v and the fundamental element
used in their construction is the complex number r in the upper half­
plane. To emphasize this fact these functions are sometimes written
as follows: t). As we have said already the function ^(u) is odd
and the remaining functions are even. We shall now investigate the
behaviour of theta-functions when 1/2 is added to the argument v.
Remembering the expansion of theta-functions into trigonometric
series and using the formulae for the conversion of theta-functions,
we obtain immediately

(» + 4 “) = ^ ^ + 4") = ~

#3 (» + 4 ") = ^ + 4 ") = w
(v §3 ■

Let us now investigate the behaviour of theta-functions when t /2


is added to the argument v. This, aB we know, is equivalent to multiply­
ing z by h112. Using the power series for theta-functions we obtain,
for example, from (92)
/ r ’V + 00 / 1\2 2n — l
(i> + — ) = i _v (— i)"A r “ tJ h~r~ z2" - 1 =

= i h ~ T z-1 +
jg ( - 1 )nhn‘z2n,
n = — <=>

or from ( 1 0 2 )
#i (w + - y ) = im&t (v) ,
where

m= h 4 z-1 = h 4 e~im, (107)

and it can be shown similarly that

+ -y ) = m $3 {v)l + -g-) = (v )> ^ 4 + -y ) = i m $ x{ v ) .


646 SPECIAL FUNCTIONS [177

From this more general transformation formulae can be obtained.


Thus, for example:

0 i (v + T) i = 0 ! (*> + \ + - y ) = ih 4 e ,:,^ + + -|-j =

= ih * e ‘ ^ + ih 4 e-1”* ^ (v) = — (0 ),
where
I = h -1z ~~. (108)
The results obtained can be collected in a table as shown below:

u+ l t- + T 0 + 1 +r
•’ + -T V + ~T « +' -2 ^+ 2-
A,'5’,?-'4'

oCD
m03

1
1 1

02 imfti
OT0 j

1
- 0. — i m#4
04 mfr2
1

1
03 m &2

If we want, for example, to express &3(v + 1/2 + r/2 ) in terms of


theta functions of the fundamental argument v, then we must find 03
in the first column and take the expression under v + 1/2 + t/ 2 , in
the corresponding row, i.e.

^3 ( « ' + y + - y ) = mtc0 i (v).

We shall also give a table of the zeros of theta-functions. The


function differs from the function a(u) by an exponential factor
which cannot vanish and, consequently + + ) vanishes when, and only
when <$(«) vanishes; this latter fact takes place when
u = n2<o + n.'2 ft>',
where n and n ' are arbitrary integers. Dividing by 2co we obtain the
following expression for the zeros of the function 'd1(v):
v = n + n ’x.
The zeros of the remaining theta-functions can be obtained by
using the first row of the above table. Thus, for example, we have
i%(v) = m ~ 1'&1(v-\- 1/ 2 + r/2 ), and therefore the zeros of 0 3(w) are
determined by the condition

r +, - y1 +, -Ty = 7l+7lT,
. ,
177j THE PROPERTIES OP THETA-FUNOTIONS 647

since m -1 = hie‘7'v does not vanish, or

* = ( n - - r ) + (n' - - r ) T’

where n and n ’ are arbitrary integers. We thus obtain the following


table for the zeros of theta-functions:

n -f n'x
h n + n'x + - i-

n + nx +
1 + —r
h —

n + n ’x +

Notice also that it follows from the fifth column of the table (106)
that the functions # 3 and # 4 have a unit period, and the functions
and $ 2 have a period of two. The above table shows that different
theta-functions have different zeros.
Theta-functions can be regarded as functions of the two arguments
v and x. For any given r in the upper half-plane they are integral
functions of v, and for any given v they are regular functions of r
in the upper half-plane. This latter circumstance is directly due to the
fact that the series (92) and ( 1 0 2 ) converge uniformly when
| h | < g < 1 . We shall now show that all four theta-functions, by
virtue of being functions of two arguments, satisfy one and the par­
tial same differential equation of the second order:

d20k(v) _ A -fM kW (1 1 1 )
qv2 — 47W Qr

This equation formally resembles the heat conductance equation


with which we dealt earlier [II, 203]. We shall test the equation (111),
for example, for the function #3(«). Differentiating the general term
of the series (104), equal to 2hn* cos 2nnv = 2e,mn‘ cos 2nnv, twice
with respect to v we obtain

— Hn2n2e‘!',n2 cos 2 nnv,


648 SPECIAL FUNCTIONS [178
and the same result is obtained by differentiating once with respect
to x and multiplying the result by in i
4?w (2inn2e,nTn3 cos 2rmv) = — 8 n27i2e,,lTnl cos 2nnv.
This equation can be tested similarly for other theta-functions.

178. An expression for the numbers ek in terms of #s. In the study


of the Weierstrass function P(u) we introduced the numbers ek which,
in our new notation, are defined as follows:
ei = P M : e2 = p (co + (o')-, ea = p{(o’), ( 11 2 )
and we obtained the following fundamental relationship for the
function p(u)
P'2 («) = 4(|? («) - {p («) - e2) {p («) - e3) . (113)
The numbers ek, as we saw, satisfy the condition
ei + e2 + e3 — 0 (114)
and are all different. These numbers are of fundamental importance in
the theory of the function p(u). They can be taken as the basis for the
construction of the function p(u) instead of 2 co or 2 a/. In this case
the function p{u) will be obtained as a result of the conversion of an
elliptic integral of the first kind

u = \-r- ** =■ (H5)
J / 4 (x — e,) (x — e2) (x — e3)

We shall now express the numbers ek in terms of theta-functions,


the argument of which is zero. Take formula (106)

( - £ )
and put u = (o, i.e. v = 1/ 2 , and subsequently, put u = co -f- co', i.e.
v = 1/2 + t / 2 . We thus obtain from (1 1 2 ):

1 ( 2)
K — «* =
2cu^ i * (_Lj ’

]e2 — ek = 2(o &k+l


178] AN EXPRESSION FOR THE NUMBERS ei IN TERMS OF 0, 649

Using the table (109) which gives the conversion formulae for theta-
functions we have
i f --------------■ _1 iV , i if.i
— - 2^-07 0 “ -
lf--------- 1

h i — e°33 :— 2u> 0 , #3

We shall prove below the following important identity


(116)
the use of which makes it possible to rewrite the above formulae in a
very simple form:

= V®! — e3 = —^ = cu
2 (11?)

Let us now prove the identity (116). We have from (106)

yp (2 «w) — e* = 21a> 0/c-Mflih-i («)


0, (i>) ’
whence, expanding the functions and 0fc+1(«) into McLaurin’s
series and bearing in mind that i)y{v) is odd while the remaining func­
tions are even we, obtain
1 + 5S+JJ1+. .
+ **+i 2 +
lty (2 «») — «* = 2iu
, 0 k" ” 3 .
v + Wi ~T +
or, taking the factor v out of the denominator and dividing one series
by the other series

or
J>(21*») - - j± -r [l + - 4- ^-) 4 ) + •' •]'
As we know, the expansion of jj)(w) near u = 0 contains no constant
term and, consequently, squaring the bracket on the right-hand side,
and collecting together all the constant terms we should obtain (—eh),
and so
660 SPECIAL FUNCTIONS [179

This and (114) gives us the following relationship

(119)
*i #2 + #3 ^ #4 '

In all these formulae the dashes above % indicate partial differen­


tiation with respect to v so that, for example, ■&"' is d3ft1(v)[dv3, when
v = 0. The equation (111) when v = 0 gives

*» = 4 ( k = 2,3,4)

and, similarly, assuming in the equation (1 1 1 ), that 1c = 1 , differen­


tiating with respect to v and subsequently putting v = 0 we
obtain
a„ , . .
#1 = ^nl ~dr ‘

By using the last two relationships we can rewrite the formula (119)
as follows
1 _ 1 9#2 1 d»3 1 0fl4
9t 0r ‘ 0r ' 0r

Integrating with respect to x we have

0 J = CVfl 3^4-

where (7 is a constant independent of r, i.e. of A. To determine this


constant we substitute on both sides of the above identity the ex­
pansion (105) writing out only its first terms

2 ti(A 4 — . . . ) = C{2h* + . ..) ( 1 + .. .) ( 1 - .. .)•

Comparing the coefficients of terms containing A14, we obtain C


which gives the identity (116).

179. The elliptic Jacobian functions. Instead of the elliptic Weier-


strass function jp(w) other elliptic functions are frequently used which
have an earlier origin historically and date back to Jacobi. Assume,
as always, that r is an arbitrary number in the upper half-plane,
while co and co' are two numbers, the ratio of which is co'jco = r.
By using these elements we can construct theta-functions. Let us
179] THE ELLIPTIC JACOBIAN FUNCTIONS C61

define three new functions which are the ratios of two integral func­
tions, i.e. which are fractional functions:

sn (u) - a<*(»■)
(u)
3
. o JK («)
0 4 (v)
Qi (u) _ (v)
cn (u ) ( 120 )
03 («) #2 #4 (®) ( - £ )
cl n (m) - (M) _ #4 ^3 I4’)
a3 (u) #3 04 (i>)

According to the known formulae

V P (“ ) — (J(W) ’
these new functions are connected with the Weierstrass function P(u)
by the following three relationships:

/* > (« )-« • = — w ; VP («) - «i = ; VP («) - «2 = S ^ - ( 121)


Eliminating the function {?(«) from these relationships we obtain
two relationships for the new functions:
cn2 (u) + (e1 — e3) sn2 (u) = 1 ; dn2 (u) -j- (e2 — e3) sn2 (u) = 1 . (1 2 2 )
The formulae (117) from the previous paragraph give us

Until now the complex numbers co and co' have remained fully arbi­
trary, the only essential condition being that the relationship co'lco = r
should lie in the upper half-plane. In the theory of the Weierstrass
function these numbers are not subjected to any other limitations.
In the theory of Jacobian functions the number co for a given r is
determined by the condition that the difference ex — e3 is unity.
The second of the relationships (123) then gives co:

o) = —i?2 = — (l -f 2A + 2hl +2h* + ...) 2 (h = elnz), (124)


o £

which is fully defined by this formula for a given r; co' is determined


subsequently from the formula co' — cox. Substituting the expression
(124) into the relationship (123) we obtain:
652 SPECIAL PUNCTIONS [179

where the right-hand sides depend only on r. The relationships (122)


can then be rewritten as follows:
sn2 (u) + cn2 (u) = 1 ; dn2 (u) + k2sn 2 (u) = 1 , (126)
where it is assumed, for the sake of briefness, that

* 2 = f - (127)
The Jacobian functions are constructed by using r alone and there­
fore the following notation is sometimes used
sn (u; t); cn(w; t); dn(«; r).
The number k given by formula (127) is known as the modulus of
the Jacobian function. We shall also introduce the so-called additional
modulus which is defined by the formula

*' *— I ' (128)


Adding the first and third of the relationships (125) we obtain
i 2 + i'2 = l . (129)
The formulae (127) and (128) determine k2 and A'2 as the squares
of certain single-valued functions of r; by taking definite values of
the radicals we can therefore write:
_ • y -1 i (130)

Let us now return to the formulae (120). Factors on the right which
are independent of v can be expressed in terms of k and k'. In fact,
we have from (130)

which, together with (124) and (116), gives

2 o>A = ^ A = A
».
= J_
fk ’
and therefore the formulae ( 1 2 0 ) can be rewritten as follows:

1 fl, (t>)
sn («) = cn (u) (131)
fic 0 * («):

Hi)-
180] THE FUNDAMENTAL PROPERTIES OF JACOBIAN FUNCTIONS 653

180. The fundamental properties of Jacobian functions. The formulae


(131) represent the Jacobian functions as the quotients of two integral
functions. Using the fact that is odd and that the remaining
functions &k(v) are even, we can conclude that sn(u) is an odd func­
tion and cn(w) and dn(w) are even functions.
Also #,(0) = 0 and we have
(») _ 0 i (») = 1
u ti= o 2cov v =o '2cy 11
and the formulae ( 1 2 0 ) give

— |u _ 0 = l; on(0)=dn(0) = l. (132)

Let us now return to the table (109) which gives the conversion
formulae for theta-functions. Bearing in mind the fact that the ad­
dition of 1/2 or t /2 to v is equivalent to the addition of co or co' to u,
and using the fundamental relationships (131) we obtain the following
table of conversion formulae for the Jacobian functions:

U+ CO u + co' u (O co' u + 2co u + 2 co' u + 2co -f 2co'

cn (u) 1 1 1 dn (w)
sn — sn («) sn (u) — sn (u)
do (u) k sn (u) k cn (u)
sn (m) i dn (u) .k' 1
cn — cn (w) —cn(u) cn (u)
dn(u) k sn (u) " k cn(u)
. cn (m)
dn k' dn1(u) ik' cn
Sn (u)
{U) dn (u) —dn(w) — dn(u)
sn (u)

The last three columns of this table show that the function sn(u)
has periods 4co and 2co', the function cn(u) has periods 4co and
2a> + 2co', and finally, the function dn(«) has periods 2co and 4 co'.
The table (110) which gives the zeros of theta-functions leads us
directly to a table giving the zeros and poles of Jacobian functions.
Adding to this the periods shown above we obtain the following table:

Zeros Poles Periods


sn(u ) 2nco+ 2n'co' 2nco+ (2n' -f- 1) co' 4coand2e/
cn(u ) (2w-f 1) co+ 2n’co' 2n<o -j- (2n ' + 1) co' 4coand2co 2co'
dn(u) (2n + 1) co+ (2n ’ + 1) co' 2nco-j-'(2n' + 1) co' 2coand4co'
654 SPECIAI, FUNCTIONS [181

In Fig. 84 below, the parallelograms of periods for Jacobian functions


are given. The circles denote the zeros and the crosses the poles of
the corresponding function. Owing to the fact that both the theta-
functions and the a(u) functions have simple zeros we can say that
the Jacobian functions have simple poles. Below there are two poles in
each parallelogram, i.e. all Jacobian functions are elliptic functions
of the second order with simple poles.

This is directly due to the fact that all these functions can be obtained
by converting certain elliptic integrals of the first kind to a poly­
nomial of the fourth degree under the radical. We shall now try to
explain this circumstance.

181. The differential equation for Jacobian functions. It follows


directly from the formulae (113) and (121) that
2cn(«) dn(u)
r («) = ± sn3(«)
To determine the sign on the right-hand side, we multiply both sides
of the above equation by u3 and then put u — 0. Bearing in mind
the fact that the product u 3 p'(u) equals (—2 ) when u — 0 , and using
the formulae (132) we find that the minus sign must be taken on the
right hand side of the above formula. This sign will remain unchanged
during analytic continuation of the function, i.e. we have
2cn(u) dn(u)
sn3 (u)

On the other hand, differentiating the relationship


1
V (« ) - *3 = sn2 (u) ’
181] THE DIFFERENTIAL EQUATION FOR JACOBIAN FUNCTIONS 655

we obtain
' hA — — 2 (sn
^ ' ' sn* ( h) ’

and comparing these two expressions for ty'(u) we obtain


[sn («)]' = cn (u) dn (u). (135)
Differentiating the equations (126) and using (135) we obtain the
derivatives of two other Jacobian functions:
(cn («))' = — sn (u) dn (u); (dn (w))' = — k2sn (u) cn (u). (136)
Squaring and using (126) we finally obtain the following differential
equations for Jacobian functions:

( d s n ^d y = (1 _ sn2 (»)) (1 - k2sn2 («)),

= ^ - cn 2 («)) ( * '2 + * 2 °n2 («))> (137)

(T T 1 )* = - (1 - dn= («)) ( i '2 - dn2 («)).

Let us investigate the differential equation for the function sn(«)


in greater detail. If we put x = sn(u) we can write

J = f ( l “ *2) (! - fc2*2).
where it must be assumed that x = 0 , when u = 0 , and also that
the radical on the right is equal to unity since sn'(0) = 1 from (132).
Separating the variables and integrating we obtain:

(138)

This shows that the function sn(tt) is obtained as a result of the con­
version of an elliptic integral of the first kind in the Legendre form.
It can be shown, conversely, that by taking arbitrary complex
values other than 0 and 1 for the number k2, we obtain the Jacobian
function sn(w) as a result of the conversion of the integral (138).
Hence together with r the number k can serve as an element in the
construction of the Jacobian function. We investigated the integral
(138) in detail from the point of view of conformal transformation
in the particular case when k2 is real and lies between zero and unity.
We than had one real period, which in [167] we denoted by 4K and
656 SPECtAIi FUNCTIONS [182

another purely imaginary period 2iK '. Comparing this with our new
notation we obtain:

K = co= — &23, »£'= o>' = a>T = -J-0 § T .

182. Addition formulae. Let us consider three functions of the


variable u : <px(u) = sn(«) sn(« + v); <p2(u) = cn(w) cn(u -f- v);
<p3(u) = dn(w) dn (u -+- v) where v is a fixed arbitrary number. By using
the table (133) it can readily be seen that all these functions have
periods 2 co and 2co'. The function <px{u) has simple poles at the points
where sn(«) or sn(w + v) have poles. Using the table (134) we can see
that these points differ from co' or —v + co' by a period, i.e. they differ
by an expression of the form n2co + n'2co', where n and n ' are arbit­
rary integers. In the fundamental parallelogram of periods constructed
on the vectors 2 co and 2 a / there will, consequently, be only two such
points. The same result will also be obtained for the remaining functions
< P k { u ), i.e. all these functions are elliptic functions of the second order

with periods 2 co and 2 co' and have two simple poles in the parallelo­
gram of periods, one of which is equal to co'. The constants A and B
can be so chosen that the two functions
<p2 (u) + A<px (u) and <p3 («) + B<px (u) (139)
have no pole at u = co'. For this choice of constants the functions
(139) will have only one pole of the first order in the parallelogram of
periods, and since no elliptic functions of the first order exist [168]
it follows that these functions are simply constants. We can there­
fore say that with a suitable choice of the constants A and B the follow­
ing relationships apply:
cn (u ) cn (u + v) + A sn (u) sn (u + ») = A x (140)
dn (w) dn {u + v) + B sn (u) sn (u + v) = B t .
The constants A, B, A x and B x are constants with respect to the
argument u but their value depends on the choice of v. Let us deter­
mine these constants. On putting u = 0 in the formulae (140) we
obtain:
.41 = cn(»); £ j = dn(r).
Differentiating the relationships (140) and then putting « = 0 we
obtain from (135), (136) and (132):
(cn («))' + A sn (v) = 0 ; (dn (v))' -f B sn (v) = 0
183] THE CONNECTION BETW EEN TH E FUNCTIONS j)(u) AND snCtO 657

and again, from (136)


A = dn (v) , B = k2cn (v).
Substituting the constants in (140) we finally obtain the following
two relationships:
cn (u) cn (u + v) + dn (v) sn (u) sn (u + v) = cn(t>),
dn (u ) dn (u + v) + k 2 cn (v) sn (u) sn (u + v) = dn (v),
which can be regarded as identities with respect to u and v. Replac­
ing u by (—u) and v by v -f- u we obtain
cn (u ) cn (v) — dn (u v) sn (u) sn (v) = cn (u -f- v),
dn (u) dn (u) — k2cn (u -f v) sn [u) sn (v) = dn (u -|- v) .
The last two formulae enable us to find cn(« + v) and dn(w + v)
and on substitution in the first of the equations (141) we obtain
sn(i/ -f v). We thus arrive at the following addition formulae which
express the Jacobian functions of the sum of two arguments in terms
of Jacobian functions of the individual arguments

sn (u + v) =
sn(u )cn(v) dn(v) -f sn(v)cn(u) dn(u )
1 — k* sn*(u) sn*(v)

cn (zt + v) =
cn(u) cn(«) —sn(u )dn(u) sn(v)dn(v)
1—k *sn*(u) sn*(v) (142)

dn (u + v) =
dn(u) dn(v) —k2 sn(u) cn(u)sn(v )cn(v)
1—kz sn*(u) sn*(v )
The first two formulae resemble the addition formulae for ordinary
trigonometric functions: the sines and the cosines. These latter
functions do, in fact, appear to be special cases of Jacobian functions
when k = 0. Thus if we put k = 0 in the integral (138) then its conver­
sion gives x = sin u; it follows from (126) and (132) that cn(w) be­
comes cos u. Finally, the second of the formulae (126) shows that the
function dn(«) simply becomes unity when k = 0 and therefore it has
no analogous function among the trigonometric functions.

183. The connection between the functions jJ(n) and sn(«). We shall
now establish a direct connection between the functions ty{u) and
sn(w). Let us investigate the function p(u) with periods 2 a> and 2a/.
Consider the number co'jco = x in the upper half-plane and construct
theta-functions and the function sn(w) by using this number and the
first of the formulae (130) and (131). The numbers 2 a>and 2a / which are
658 SPECIAL FUNCTIONS [183

connected with the above Weierstrass function, will not, in general,


satisfy the condition ex — e3 = 1. According to the formulae (117)
quoted above we have the following relationships:

We have new numbers 2co and 2 co' for the function sn(u) instead
of and 2 co' which, as we know, are determined by the conditions:
2<5 = Ji&l; 2(5' = 2u>r. (144
Denoting by X the relationship X = ci>lco = (o'lco', we consider the
function

We have X2co = 2 co and X2co' = 2a>' and, according to the table


(133) , the function f(u) has periods 2 co and 2 co'. We see from the table
(134) that the function f(u) has poles at n2co + n'2co', where n and
n ' are arbitrary integers.
Hence the function f(u), like the function ty(u), has periods 2 co and
2 a/; it also has in the fundamental parallelogram of periods a single
pole of the second order at u = 0 . We shall show that the infinite part
of the function f(u) at this pole will be the same as of that the function
§(u) and will be equal to lfu2. In fact, remembering that the function
sn(tt) is odd and also (132) we have the following expansion near
the point u = 0 :
sn (u ) = u + c3u3 + c5m5 + . .
whence:
l l 1
sn1 (u) m2 (1 + c3u 2 + c4u 1 + . . . ) 2 — + diul + ■• • .
or we have near u = 0 :

which we wanted to prove. Hence the functions f(u) and p(u) have
in their common parallelogram of periods the same poles with equal
infinite parts; it therefore follows that these functions differ only by
their constant terms, i.e.
(145)
184] ELLIPTIC COORDINATES 659

To determine the constant C we put u = co. We have jjl(o)) = e1


and from the table (133)
,, \ ,-v cn (0) ,
sn (*w) = sn (o>) = aJT(6 j‘ = l»
and formula (145) gives:
C = eL- r - . (146)
According to the formulae (143) and (144) we can write
2 <u = 2 co y e1 — e3; 2<*>' = 2co' — e3,
i.e.
. 6> d>' , r------------
a, ~ to' ~~ f'B1 e3’
whence, by using (146) we have C = e3.
Using the equations (143) and (114) we can write the constant G
as follows:
n - (1 + &2)

Hence we finally obtain the following connection between the


functions §(u) and sn(u):
6i ~ ■
P(«) = ------)■ - - ------ h e3 (147)
sn2 ( / e t — e3w)
or
A2 (1 + fc2) A2
P(«) = sn2 (Aw) 3
(A = y ei — e3). (148)

184. Elliptic coordinates. Elliptic functions are frequently used, particularly


in mechanics. Here we shall only deal with fundamental and very simple appli­
cations of these functions. One application is their use in finding elliptic coordi­
nates in space. We have already m et elliptic coordinates earlier [II, 137], Here
we shall repeat what we know already and give some additional properties of
these functions. We shall somewhat m odify our earlier notation, viz. we shall
replace the numbers a2, b2 and c2 by —a2, —b2 and —c2. Let us write the equation
x2 y2 (149)
g —a2 g — b2

This is an equation of the third degree with respect to g. A t any given point
with cartesian coordin ates(ar, y, z) the equation (149) has three real zeros A,
ft and v, which satisfy the inequality
A > a 2 > ft > b2 > v > c2, (150)
and these three numbers are known as the elliptic coordinates of the given p o i n t .
So as not to affect the sign o f the equality we assume that x, y and z are not
660 SPECIAL FUNCTIONS [184

zero and that they are positive, say. If, in the equation (149) we put g = A,
we obtain an ellipsoid which passes through the given point; when Q = y,
this will be a one-sided hyperboloid and when <5 = v a two sided hyper­
boloid. We have seen earlier that the coordinate surfaces A = const
(i = const and v = const are m utually orthogonal, i.e. elliptic coordinates are
orthogonal coordinates. W e shall introduce formulae which give the cartesian
coordinates in terms o f elliptic coordinates. Bringing the left-hand side of the
equation (149) to a common denominator and remembering that the nume­
rator is a polynom ial o f the third degree o f g with zeros A, fi and v, the first
coefficient o f which is ( — 1), we can write an identity for g:

. y2 , z2 , —(e —*) (e — - »)
g — a 2 "1- g — b2 ^ g — c2
m) (e
(g — a2)(g — b2)(g — c2) ' ^ 01>

M ultiplying by ( g — a2) and then putting g = o2 we obtain an expression


for x2 and analogous expressions for y 2 and z2:

,s = (;- ~ a 2) (t* — a 2) (v - a 2)
(a2 — 62) (a2 - c2)
f2 (A- b2) (y - b2)(v - 62)
(b2 — c2) (62 — a2) (152)
,2 = (A — c2) (fi — c2) ( r - c2)
(c2 — o2) (c2 - 62)

We shall now deduce a formula for the square of an elem ent o f arc in
elliptic coordinates. Taking logarithms and differentiating the formulae (162),
we obtain:
2 d# _ dA d/t dv
x A — o2 fi — a 2 v — a2
o dV _ dA d/i dv
y A -62 + ^ - 6 2 + v -6 2 ’
dz _ dA d/t , dv
z A — c2 ' fi — c2 v — c2

Multiplying by x, y, z, squaring eaoh term separately and adding we have


da2 = I ? dA2 + M 2 dy 2 + N 2 dv2. (153)
where, for example,

4£ 2 = 7(A
^—— a 2)2
^
(A — b2)2 +r ■
(A — c2)
(154)

N otice that the right hand side o f formula (153) does not contain products
dA dfi etc. since elliptic coordinates are orthogonal [II, 130]. W e can obtain the
right-hand side o f the formula (154) by differentiating the left-hand side of the
identity (151) w ith respect to g, changing the sign and then putting g = A, i.e.

_d_ (g - A) (g - fi) (g - v)
de (e - a2) (g - b2) (g - c2)
185] THE INTRODUCTION OP ELLIPTIC FUNCTIONS 661
We can therefore write the following formula for ds2:
(A — fi) (A - v) (/i — X){fi - v)
4da2= (A — a2) (A — b2) (A — c2)
dA2 +
(H — a2) ( n - b2) (fi - c 2)
d/i2 +
(v - A) (v — /i)
+ (v- o2) (v—b2) (v- c2) dv2. (155)

Knowing the expression for an element of length we can write the Laplace
equation in elliptic coordinates [II, 119]. For the sake o f sim plicity we introduce
the following notation:

1 (e) = (e - »2) (<?-&*) (e-c2).


In the notation of [II, 119] we had:

o u _ l / ( A — M) (A - v) . o ix l/(i“ nFr _ i f (v - A) (v - n)
2Hi~ l —m —’m* ~ l —m —,2fls-F—zoo—
where m ust be positive; it m ust also be remembered that /(A) and f(v) are
positive and f(fi) < 0. The Laplace equation in elliptic coordinates is as follows:

+
\f W f (v) "9A K/ (v) / (A) b/2 I * ^ 0|“ i +
a*- a a = 0, (156)
K/ ( A) / (/ z ) 8 7

where the last tw o terms are obtained from the first as a result o f the cyclic
rearrangement o f the letters A, fi and v.

185. The introduction of elliptic functions. Let us replace the variables


A, fi and v b y new variables a, j) and y, according to the formulae

dA
- da; d^i
= d|3;
dv (157)
Ym W)
i.e. a, |S and y are expressed by elliptic integrals o f the first kind in terms of A,
(i and v and, conversely, the latter are elliptic functions o f the former. Thus we
have, for example, from (157)
_8 _
8 a ’
and we can rewrite the equation (166) as follows:

<’ ■
“ « I ? + (A- M W = °‘ (I58)
Let us now turn to the formulae (152) and show that x, y and z are single­
valued functions of the new variables a, fi and y. In fact, consider the radical

YJW) = /(?- a2) (g—b2) (e - c2),


062 SPECIAL FUNCTIONS [185

which enters the expressions (157). Replace g by a new variable t according


to the formula
Q= P + qt,
where p and q are constants. We obtain
(e - a2) (2 - V-) (q — c2) = q3 (t — e,) (t - et) (t — e3),
where e* are the zeros of the polynomial with respect to t so that we have:
at — p + q e b2 = p + qe2; c* = p + qe3.
Let us select the constant p so that the sum of the should be zero, i.e.
el + f 2 + <3 = 0 ,
hence
a- -j- b- + c-
P = ------- 5-------
The preceeding formulae will determine the numbers accurately except
for q which we assume to be positive and denote by a2. We therefore have:
a 2 + b2 + c2

s2e» = 62— a2 + b2 + c2 (159)


a1 + b2 + c2

It also follows that:


o2 — b2 = a2 (et — e 2); o2 — c2 = a2 (e, —e3); b2 —c2 = a2 (e2 — e3). (160)
Substituting g = p + qt we have:
g —o2= a2(t — e,); g —b2 = a2(t — e2); g —c2= a2(t — e3) .
The polynomial /(g) with the new variable can be written in the form
/ (e) = a» (t — e,) (t — e2) (f — e3).
On putting
A = Q2+b; + - + * 2t, (161)

we can rewrite the first of these formulae in the form:

2 pf dt
=
* J /4 (t - eL) (« — e2) (t — e3)

where the arbitrary constant on the right is omitted since it is of no significance.


Assuming, for the sake of simplicity, th at a = 2 we obtain t = ty(a) as a result
of the conversion of the integral since the polynomial under the radical has the
form given in [178] owing to the fact that e, + e2 + e3 = 0.
Formula (161) gives
A = g, + ^ + C- + 4 p ( a ) . ( 162)
186] THE T.AMfj EQUATION 663

and wo obtain similarly:

a8+ 3 + f i + 4 p (f l; v = ° 2 + fe2 + c 2 + 4 (p(y). (163)

Substituting these expressions in the formulae (152) and taking into account
(159), (162) and (163) we have:
, (P ( f t - « . ) ( P ( f t - « ,) ( P ( f t - « . )
(ei e2) (ei ea)
I/2 = 4 ( P (a ) ~ e«) ( P ( f t ~ « i) ( P ( f t ~ «i) (1641
(e 2 - e3) (e2 - e,)

22= 4 (P (a) - e») (P (ft - «»> (P (ft - **)


(e 3 e l) (e 3 e 2)

As we know from [172] all the differences in the numerators are squares of
single-valued functions of a, fi, y, so that, in fact, the above formulae give x, y
and z as analytic functions of a, /? and y. According to (162) and (163) the
Laplace equation (158) with the new variables will be

(P(ft - P(ft) + (P (ft - P (ft) + (P (ft - P(ft) = °- (165>


186. The Lame equation. Let us separate the variables in the Laplace equa­
tion and find its solution in the form of a product of three functions, one of
which depends only on a, the second only on /9 and third only on y:
U = A (a) B (ft O (y). (166)
Substituting in the equation (165) and dividing by A(a) B(fS) C(y), we have

(P (ft - P (ft) — + (P (ft ~ p (ft) + m ~ ®(a)) = °'


This equation will be satisfied if it is assumed that the factors in the ex­
pression (166) are solutions of an equation of one and the same form viz.:
A" (a)
A (a) — — ° P (ft — b;
where a and b are constants. We thus obtain an equation of the second order
with a coefficient of dual periodicity

^ + [a p (« ) + 6 ] f l( U) = 0 . (167)

Let us determine, first of all, the constant a so that the general solution
of the equation (167) is a single-valued function of u. The coefficient ap(u)-\-b
can be expanded near the point u = 0 as follows:

£ +‘+ -
and therefore the determining equation at this regular point will be
e (e — ft + o = o. (168)
664 SPECIAL FUNCTIONS [186

If the integral is to be single-valued the zeros of this equation must be integers.


The sum of the zeros is equal to 1 and therefore the equation (168) must have
zeros a t —n and n + 1, where n is a positive integer or zero. Hence the constant
a can have the following possible values:

a „ = - n { n - (-1) (n = 0, 1, 2, . . . ). (169)

Strictly speaking we have only shown above that the equation (169) gives
the necessary condition for the general solution to be single-valued. We shall
now show th at this is also the sufficient condition. I t follows from general
theory th at one of the solutions of the equation (167) when a = —n(n -f 1)
can be expanded as follows near the origin:

R (u) = un+1 (c0 + c,u + c2u 2 + . . . ) (c0 0). (170)

The equation (167) remains unaltered if w is replaced by ( —u); hence if we


make the same substitution in formula (170) we should also obtain a solution
which only differs by its constant term from the solution (170) since the second
solution, linearly-independent with (170), has a completely different form near
u = 0. I t follows from these considerations that the power series in formula
(170) contains only even powers of u, i.e.
(m) = wn+1 (c0 + c2m2 + c4u< + . . . ) (c„#0). (171)

The second solution of the equation (167), as we know, can be obtained from
the formula [II, 24]
R2 («) = «!
or
Rt (u) = R l (u) J —^ +2 (c0 + CjM2 + ctu* + • • ■) ' 2 d u .

The integrand can be expanded near the point u — 0 into a series contain­
ing only even powers of u and therefore the term in u~ 1 will be absent while the
second solution of R2(u) will not contain log u. We can thus see th at both
solutions will be single-valued near u = 0. The above arguments can be
repeated word for word for every singularity of the equation (167). Its sin­
gularities lie a t the points u = ml (ot m2 a>2, where col and <u2 are the periods
of y(u) and ml and m2 are arbitrary integers. Hence any solution of the equation
(167) can only have poles at the singularities and therefore it must, in fact,
be a single-valued function of u.
Substituting the value of the constant (169) in the equation (167) we obtain
the equation
^ ^ + [ - n ( n + l)j(J(u) + &]Je(u) = 0, 172)

which is generally known as the Lame equation. The constant b is determined


by the condition th at the equation (172) should have a solution in the form of
a polynomial in {()(«) or in the form of a product of this polynomial and a factor
187] TH E SIMPLE PENDULUM 6C5

of the following form


—e,; \ ' § ( u ) - e 2; fig (u) — e,,
and there can be one, two or three additional factors. I t is apparent that the
constant b can have (2n + 1) values which satisfy this condition. If R 0(u) is a
solution of the equation (172) of the form mentioned above then the product
(a) (P ) * 0 (y).
which is a solution of the Laplace equation, is a polynomial of the nth degree
in the x, y, z coordinates. For a given n there will be (2n + 1) such solutions,
as explained above, and they are generally known as Lam 6 functions. These
polynomials are obviously directly connected with spherical functions with
which we dealt ear her.

187. The simple pendulum. We shall consider a simple pendulum as one of


the simplest applications of Jacobian functions. Assume that a heavy material
point of unit mass moves round a smooth circle. Let the coordinate axes
Z

X and Z be in the plane of this circle and let the Z axis be directed vertically
upwards, let I be the radius of the circle. Suppose th at when t = 0 our point
is set free from the lowest point M 0(z = —I) with an initial velocity v 0. The
increment in kinetic energy is equal the work done by the force of gravity and
we thus obtain the formula

y
1 2— - 1y vo2 = — 0Z — 9h
7
or
** = 2g ( a - z ) (a = - Z + - g - ) . (173)

Suppose th at the line z = a intersects our circle at the points A and A ',
i.e. a < I or v0 < 2 fig. I t follows from formula (173) th at z < a, and therefore
movement takes place along the arc A M ^A ' (Fig. 85) of our circle. We have
666 SPECIAL FUNCTIONS [187

2 = —Zoosfl; let us now introduce an angle a such th at a = —Zcos a


(0 < a < n). The velocity is then given by the formula

v— — 11
V ~ "dT ~ 11 d7
and therefore the equation (173) can be rewritten in the form
d0 V
I2= 2 gl (cos 0 — cos a)

or introducing half-angles
de v
(d f)2=4ff(sin2“T “ sin24 - ) ’
hence
(174)
|/suVy - s u V y

We are supposing th at 0 increases with t. Replace 0 by the new variable


r according to the formula

sin
0- = r sin
. - 5a-
£t it

Differentiating this relationship we readily obtain

2 sin dr 2 sin dr 2 sin -g - dr


d0 = i.e, d 0 =
4 | / i - sin 2 — 1 — sin 2 t2

and therefore, substituting in (174) and bearing in mind that, when t = 0,


we have 0 = r = 0 :
dr
t 2) (1 - k2T2) (k2 = sin2 >
hence
(175)
+ * )•
and using the known property of Jacobian functions we obtain:

s in - L = s i n ^ - s n ( ] f - |- ( ) = ksn ( |f - f t ) ,
(176)
COS - 1 — k2sn 2 ,) = d n ( l f Z t ) ,

and to extract the zero we take into account th at when t = 0, 0 = 0. The


last formulae make it possible to express the coordinates x and z in terms of
single-valued functions of t.
188] AN EXAMPLE OP OONFOHMAL TRANSFORMATION GG7

Let us consider now the case when the constant a in formula (173) is greater
than I. We can rewrite this formula as follows:

I- ) = 2g (o + I cos 6) = 2g -f I — 21 sin2— j ,
or

/2 ( - 5 r ) 2 = 2 f f ( a + Z ) ( 1 “ fc2sinJi ) ’ (1 7 7 )
where
21
k2 = (178)
a+ l’
and, obviously, k2 < 1. Integrating the relationship (177) we obtain:

f dl>
. 4-
Replacing 0 by the new variable r = sin 0/2 we have:

. r 2dr , .0 ('1,1
U -I , whence r = sin ——= sn I -s- At I ,
J /( I — r 2)(l — k2x2) 2 V2 /
o
and similarly

co s“t = f 1 - sni ( 4 - * ) = cn( ^ - ^ ) •


These formulae make it again possible to express the coordinates as single­
valued functions of time.

188- An example of conformal transformation. As we saw above, when


0 < k < 1 the function
dz
M (179)
- z 2) (1 — fc2Z2)

transforms the upper half-plane z into a rectangle in the u-plane and, con­
sequently, the reciprocal function z = sn(tt; k) transforms a rectangle into a
plane. The lengths of the sides of the rectangle are determined by the integrals
[167]:

2f dx- - — and f -- .,
x2) (1 — k°-3?) J /( I - xJ) (1 — k/2x2)
o o
where k2 + k '2 = 1. In this way a rectangle with sides of any length can be
obtained. By adding the constant 1/A to the right-hand side of formula (179),
a rectangle with arbitrary sides can be obtained, and such a rectangle will be
transformed into a half-plane by the function z = sn(Aw; k). We shall now show
that the function which transforms the rectangle into the circle can be simply
expressed in terms of the Weierstrass function a(u). Take a rectangle K x in
the plane, the apexes of which have the following coordinates: (0, 0), (0, a),
668 SPECIAL FUNCTION'S [188

(a, b) and (0, b). Let z =f(u) be a function which transforms K l into a unit circle;
a point (I, tj) in K x becomes the centre of the circle. If we analytically continue
f(u) across the side which connects the apexes (0 , 0 ) and (0 , a) then as a result
of the principle of symmetry, f(u) transforms the rectangle K 2, symmetrical
with K l with respect to the above side, into the outside of the unit circle, i.e.
into the domain | z | > 1 , while the point (£, —rj), symmetrical with the point
(£, rj), becomes the point a t infinity. Bearing in mind the fact th at the reflection
is in one sheet we con say that f(u) has a simple zero at the point i irj and
a simple pole at the point £ — irj. If we construct two more rectangles K 3 and
K 4, symmetrical with K l and K t with respect to the imaginary axis, then one
of these will be transformed into the domain | z | < 1 and the other into the
domain | z | > 1 ; the function f(u) will have a simple zero at the point
z = —£ it] and a simple pole at the point z = —£ + in-
I t can be shown in the same way as in [167] th at f(u) is an elliptic function
with periods 2a and 2bi. The fundamental parallelogram (rectangle) of periods
consists of the above four rectangles and in this parallelogram of periods
has the same zeros and poles as those mentioned above.
If we suppose th at col = 2a and tu2 = 2bi we can construct the Weier-
strass function a(u) and also the new function:
, . _ a (u — £ — it}) a {u + I + in)
(180)
^ W _ u (u — £ + in) a (u + £ — ir\) '
This function has the same simple zeros and poles in the above parallelogram
of periods as our function f{u). We shall show that the function (180) has
periods a>1 and a>2; if this is so then f(u) and <p(u) only differ by a constant term.
Using the property of the function <j{u) as given by the equation (49) we can
write:
Ijfc y ) +>)*(u+ f + /1 + ..
e ' 2/ \ 2/ <j(m—£—tj})ff(M+f+»»7)
<p(u + <ok)
,*(«-«+!,+ f ) + «(u+ e-i,+ f j *(«-*+ *»)*(«+ *-*»)

a (u — g — iri) a (u + £ + irj) . .
(k = 1, 2),
a (u — £ + i-rj) a (u + £ — in) V'
which we had to prove. Hence
,, _ n a(u — S — in) a (u + £ + in)
1 ' ’ a (u — £ + in) a ( u + £ — in)
To determine the constant G we put u = 0 and rewrite the above formula
as follows
/ (0) = O a (— £ — in) ea (£
(I + in)
—in) ‘ (181)
<*(—£ + in)
The definition of the function a(u) gives

mU
18 8 ] AN EXAMPLE OF CONFOEMAL TBAN8FORMATION 669

where w = m1 2a + nh %bi. We assume th at w is real. Owing to the fact that


the product includes all integral values of mx and m2, except ml = m2 = 0,
the factors are conjugate in pairs, i.e. those factors are conjugate which have
the same ml but where m2 has different signs. I f m2 = 0 then the corresponding
factors are real.
Hence in the case under consideration where co1 is real and to2 is purely
imaginary, the function o(u) is real when u is real. As a result of the principle
of symmetry it will have conjugate values for conjugate values of w. I t therefore
follows th at the numerator and the denominator of both fractions on the right-
hand side of formula (181) are conjugate so th at the modulus of each fraction
is unity. Consider the left-hand side. The point u = 0 lies on the contour of
the fundamental rectangle K x (at its apex) and therefore /(0) lies on the unit
circle, i.e. |/(0) | = 1. Hence formula (181) shows that | C | = 1, i.e. C = eia,
where 0 is real. We finally obtain the following formula for the function which
transforms the rectangle K x into a unit circle:
/ (u\ = ~ ~ *7) g (“ + £ + ir>) (182)
f ; o{u — $ + ir))o(u + £ — in)
The choice of 0 is of no importance. When the values of 0 change the unit
circle rotates about its centre.
SUPPLEMENT

THE CONVERSION OF MATRICES


INTO THE CANONICAL FORM

189. Auxiliary hypothesis. I t is the aim of this supplement to prove


the hypothesis which we stated without proof in [IIIj, 27]: if A is a
matrix then a non-singular matrix V can always be found such that
the matrix V A V ~h similar to the matrix A, will have a quasidiagonal
(or diagonal) form

V A V -1 = [ /ei (A,). I ei (;.2).........I ep a ,)], (1)

where the matrices I e(X) are of the form


A, 0 , 0, ... 0, 0

1, A, 0 , . . . , o, 0

0, 1, A, . . . , 0, 0

0, o, 0, . . . , A, 0

0, o, 0, ... , 1, A

The letter g indicates the rank of the matrix and the argument X
gives the value of each element on the main diagonal. When g = 1
then the matrix I x(X) becomes the number X. In the proof of this
hypothesis we shall enlarge upon some essential points.
Let us recall, first of all, the geometric meaning of the trans­
formation to a similar matrix. The matrix A of order n is an op­
erator in an ra-dimensional space in the sense that it effects a given
linear transformation of that space. As we know from [IIIi, 21] the
form of the matrix A depends on the choice of coordinates, i.e. on the
choice of the main axes. If the matrix A gives a linear transformation
for a definite choice of axes and if we transform the coordinates in
the course of which the new components of every vector are given in
terms of former components by means of the transformation V, then
in the new system of coordinates our linear transformation will be
given by the matrix V A V -1. Hence our problem essentially involves
670
189] AUXILIARY HYPOTHESIS 671

the choice of the axes; these are most important for the linear trans­
formation which in the former system of coordinates was effected by
the matrix A, viz. it involves a choice of axes for which our linear
transformation is expressed by a matrix of the form shown on the
right-hand side of equation (1).
Before solving this problem we shall explain some additional hypo­
theses which we shall use later. The majority of these assumptions were
explained in earlier paragraphs but to obtain a complete picture we
shall collect them together here.
First of all we shall explain the concept of the subspace which we
have already met. If xu . . . , xk are k linearly independent vectors in
space, where k < n, then the set of vectors given by the formula
xx + . .. + ckxk, (3)
where cs are arbitrary numbers can be called the subspace of k dimen­
sions formed by the above vectors. When k = n the subspace coin­
cides with the space. Another definition equivalent with the above
definition can be used for defining the subspace viz. the subspace con­
sists of a set of vectors which have the following two properties. If a
vector x belongs to this set then the vector cx, where c is arbitrary,
will also belong to this set, and if two vectors xx and x2 belong to the
set, then their sum xx+ x2 will also belong to this set. In other words by
multiplying or adding vectors of the set we do not depart from the set.
In future we shall use two methods for determining a subspace
which we shall now mention. Let JP be a matrix of order n and x be
an arbitrary vector in an 7i-dimensional space. The set of vectors
defined by the formula
f = Fx, (4)
is evidently a subspace which may coincide with the space. In fact if
a vector £x = Pxxbelongs to the set then the vector cx £x = P(cx x1) also
belongs to the set, and if two vectors f 2= Pxx and f 2= Px^ belong to
our set then the vector £x + f 2 = P(xx + x2) evidently, also belongs
to that set; therefore formula (4) for any arbitrary variable vector x
does, in fact, define a subspace. As we said in [III^ 15] the number
of dimensions of this subspace is equal to the rank of the matrix P.
We shall now give the second method for defining a subspace. Let
Q be a matrix of order n and consider a set of vectors which satisfy
the equation
Qx = 0. (5)
672 T H E CONVERSION OP MATRICES INTO TH E CANONICAL FORM [1 8 9

We can show in the same way as we did above that this set of vectors
forms a subspace. As we saw in [IIIlf 14] this subspace will have Jc
dimensions, where (n — k) is the rank of the matrix.
When we talk about a subspace we are, of course, always assuming
that it is not empty, i.e. that it does, in fact, contain vectors other
than zero vectors. Let us consider the case when formula (4) gives an
empty subspace, i.e. the case when our space formula (4) gives a zero
vector for every x. Bearing in mind the form of the linear transfor­
mation we can see that this will be so if, and only if, the matrix P is
zero, i.e. when all its elements are zeros.
Let Ev . . . , E m be certain subspaces. We say that they form the
complete system of the subspace if every vector x of the space can
be represented uniquely as the sum of the vectors
x= + . . . + fm, (6)
of the above subspaces. Let us explain the condition for this unique
representation. It follows directly from the condition that a zero
vector cannot be represented as the sum (6) which contains terms other
than zero and this, in its turn, is equivalent to the fact that there
can be no linear dependence among vectors in the above subspaces.
As an example consider the usual real three-dimensional space formed
by vectors originating at a point 0. We can define the complete
system of subspaces by means of a plane L which passes through 0
and a line I through 0, not lying in L. Let the first subspace, which is
two-dimensional, be defined by two vectors in L which do not lie on the
same straight line and the second subspace, which is one-dimensional,
by a vector along I, then any vector in our three dimensional space
can be represented in a unique way as the sum of the vectors in
the L-plane and the vector along I.
Let A be a matrix which defines a linear transformation of the
space. Suppose that we succeeded in finding a complete system of
subspaces Ev . . . , E m of dimensions ox, . . . , om, so that each of these
subspaces should be invariant with respect to the linear transformation
defined by the matrix A, in other words, any vector of the sub­
space Es (s = 1, . . . , m) will, as a result of the linear transformation
defined by the matrix A, be transformed into a vector of that same
subspace. In this case we have the following natural choice of axes
for which the matrix A assumes the quasidiagonal structural form,
viz. {pj, . . . , gm}. We take for the first axes any gl linearly-inde-
pendent vectors which form the subspace E{; for the next g2 axes we
189] a u x il i a r y h y p o t h e s is 673

choose any q2 linearly-independent vectors which form the subspace


E2, etc. Since Es forms the complete system of subspaces we have,
evidently, + . . . + gm = ». It can readily be seen that for this
choice of axes our matrix A will, in fact, be of the quasidiagona) form.
We shall investigate this in greater detail but, to simplify our notation,
we shall only deal with the case when m = 2. Let (xx.........x„) be a
certain vector and (x[, . . . , x'n) another vector obtained from the
first as a result of a linear transformation. Since the subspace El is in­
variant and for the given choice of axes, when xei+1 = xe +2 = x„ = 0,
we must have x'i+1 = x'i+2 = . . . = x'n = 0. Similarly, as a result
of the invariance of the subspace E2, when x1 = . . . = * = 0 we must
have x\* = . . . = x'Ci = 0. It follows that for our choice of axes a
linear transformation is effected by a quasidiagonal matrix of the
form
dU, d12, . . ■> aiei> o, 0, . . ., o
d2l, a22, .. •> a2ei’ 0, 0, .. 0

aeii> d’Qi2> • 1• ■> aeiei 0, 0, .. 0


0, 0, .. o, ^11> &12> • • • ’ K ,
0, 0, .. 0, ^21> b22, . . ■>
0, 0, . .., o;
^eal’ • • ’ ^eaea
Notice that the choice of the fundamental axes in each subspace
remains fully arbitrary and in future we shall use this freedom to con­
vert each individual matrix A ' and B', which is part of the quasidia­
gonal matrix (7), into the simplest form in a certain sense.
We shall now make the following assumptions which we shall use
later.
Let f(z) be the polynomial
/ (z) = a0zp + a^P-1 + ap_1z + ap.
Replacing z by a matrix A we obtain a matrix polynomial
/ (-4) = aoAp + a2Ap~x+ . •• + dp-iA + dp. (8)
By performing the operations shown on the right-hand side we ob­
tain another matrix, i.e. any matrix polynomial /(A) is also a matrix.
Notice that the coefficient as of the polynomial is numerical. Owing
to the fact that positive integral powers of one and the same matrix
A/t commute with each other and with all constants, we can say that
both the addition and the multiplication of polynomials of the same
674 TH E CONVERSION OP MATRICES INTO TH E CANONICAL FORM [189

matrix A are done in accordance with the usual laws of algebra in the
same way as with polynomials in numerical arguments. Hence if an
identity connects several polynomials in numerical arguments which
have to be added or multiplied then this identity is still valid if the
argument z is replaced by a matrix A.
The following characteristic equation is of fundamental importance
in the conversion of a matrix into the canonical form:
®11 a12 • > a in
®21> ®22 ^ * * • ) a2n (9)

® nl» a n2' • > ® nn ^


where aik are the elements of the matrix A. This equation can be
written in the form
D{A — A) = 0 , ( 10 )

where the symbol D(U) denotes the determinant of the matrix U.


As we have shown before [90] the following Cayley identity applies:
< P { A ) = 0, ( 11 )

i.e. if in the characteristic polynomial 95(A) of the matrix A the argu­


ment Ais replaced by the matrix A then a zero matrix results.
We shall now state two more simple hypotheses. As we know the
zeros of the equation (9) are known as the characteristic zeros of
the matrix A. We shall now prove the following theorem: if A j , .. .Xn are
the characteristic zeros of the matrix A then the matrix A s, where s is
a positive integer, will have the following characteristic zeros: A J , . . . Xsn.
Remembering that the term of highest degree in the polyno­
mial 95(A) is equal to (—A)n we can write the following identity
for A:
Z) (A — A) = J £ (Xk — A). (1 2 )
k =1
Let e = e2“'/s be the sth zero of unity. We have the obvious
identity [I, 175]:
(z — A) (z — eX) . .. (z — ss-1A) = zs — Xs. (13)
Bearing in mind the fact that the determinant of a product of
matrices is equal to the product of the determinants and also the
identities (12) and (13) we can write

D (As - V ) = f j (Afc - A) I J (Xk - eX) . . . J J (Xk - e^X )


k-1 k=l k ~1
190] TH E CASE OF SIMPLE ZEROS 675

or
D (A° - A*) = / 7 [(A* —A) (A* -eA) .. . (A* —£s-1A)] .
k- 1
As a result of the identity (13), this gives:

D ( A S - A « ) = / / ( 1 | - A ‘ ),
A= I
i.e.
Z > ( A * - /x) = 7 7 ( A j - ^ ) ,
fc= 1
which proves the above theorem
In future we shall have to evaluate a matrix in the quasidiagonal
form
A = [^li» A 2, . . . , Afc].
It can easily be seen that it is equal to the product of the deter­
minants of the matrices A k, i.e.
D (A) = D (AJ D (A2) . . . D [Ak) . (14)

To simplify the notation we shall only deal with the case when
k = 2. The multiplication law gives:
[A, A 1] = [A ,I\ [I, A 2],
hence
D (A )= D ([ A V I])D ([I, A2]).

Using the expansion of a determinant in terms of the elements


of a certain row or column we obtain, for example,

D ([A 1,7]) = D (A 1),

from which formula (14) follows directly.


To conclude this section we recall that similar matrices have similar
characteristic zeros.

190. The case of simple zeros. Above we have investigated in full the
conversion of a matrix into its canonical form when its characteristic
zeros were distinct. Let us now formulate this in a somewhat different
way so that we can subsequently make analogous arguments in the
general case when the characteristic zeros coincide.
676 T H E CONVERSION OP MATRICES INTO TH E CANONICAL FORM [190

Let . . . , Xn be the characteristic zeros of the matrix A which are


all distinct. As we know from above there are, in this case, n linearly-
independent vectors yk which satisfy the equations
Avk = AkVk (it = 1, 2, . . ., n)
or
( A - A k)vfc= 0. (15)
Each one of the vectors \ k gives a one-dimensional subspace Ek and
these subspaces Ek together generate the whole space. Each of the
vectors ck yk, where ck is a constant, must satisfy the equation
Ack yk = Xk ck \ k, i.e. as a result of the transformation A, each vector
is multiplied by Xk. In other words, each of the subspaces Ek is invariant
with respect to the linear transformation effected by the matrix A.
By taking the vectors yk for the axes we convert the matrix A not into
the quasidiagonal form but simply into the diagonal form since each
of the subspaces E k is one-dimensional.
We now consider an equation of the form
(A — Aft)x = 0 (Ic = 1, 2........n ) . (16)
This equation is satisfied by vectors of the subspace E k. It can readily
be seen that it has no other solutions, i.e. that the equation (16) defines
a one-dimensional subspace. In fact, if this equation determined a sub­
space of higher dimension e.g. a two-dimensional subspace, then as we
have shown in [IIIj, 27], each vector of this subspace would be linearly-
independent of the vectors in the remaining subspaces Ek and we would
then obtain (n + 1) linearly-independent vectors in an n -dimensio­
nal space which is impossible. Hence in the case under consideration
the equation (16) defines a one-dimensional subspace Ek.
These subspaces can be defined in a different way. To do so take the
expansion into partial fractions
n
1 a* or 2 a* zff(g)
<p(z) tr, z ~ h k= 1 -X k = 1,

where ak are constants other than zero. Replacing z by the matrix A


we obtain:
y a k A<P(A)

= 1. (17)
k=1
Consider now the subspaces E'k given by the formulae
x. <p(A)
•s ak X ( * = 1 , 2 ......... n). (18)
191] TH E FIRST STAGE OF TH E TRANSFORMATION IN THE CASE OF REPEATED ZEROS 677

where x is an arbitrary vector in space. The constant factor ak in for­


mula (18) is, obviously, of no significance. Formula (17) gives the ex­
pansion for any vector x:

< 1 9 >
k= 1 *

where the terms on the right-hand side belong to the subspaces


E'k. We will show that these subspaces Ek, given by formula (18), are
the same as Ek, which is given by the equation (16). In fact, if g is a
vector in Ek and is given by the formula (18) then, as a result of Cayley’s
formula, we have:
{A — Zk)g = akq>(A)x = 0,

i.e. any vector Ek belongs to Ek. It remains for us to show that, conver­
sely, any vector ?ik in Ek can be obtained from formula (18) provided
xis suitably chosen. To do so we write ^ i n place of x in (19). Owing to
the fact that each polynomial <p(A)j(A — As), when s ^ k contains the
factor (A — Xk) we have, from (16) which determines E k, the
following:

A Vk = ° when s ^ h ,

and therefore by replacing x by y\k we obtain from (19)

<p(A)
^ = ak- ^ j - n k,

i.e. the vector *lk can, in fact, be obtained from formula (18) if we take
the vector ?jk instead of x.
We will now use exactly the same arguments as when above for
the case the zeros of the characteristic equation are repeated. This will
make it possible to divide the space into a complete system of subspaces,
invariant in relation to the transformation effected by the matrix A.
For each of these subspaces all zeros of the characteristic equation are
the same; the second step in our transformation will be to choose
axes in the space which will bring us to the fundamental formula (1).

191. The first stage of the transformation in the case of repeated


zeros. Let us suppose that the characteristic equation (9) has a zero ax
which occurs r1times, a zero a2 which occurs r2times etc. and finally, a
678 TH E CONVULSION OP MATRICES INTO TH E CANONICAL FORM [1 9 1

zero as which occurs rs times. Expanding into partial fractions we


obtain the following formula
_L_= y g*(g)
9>(z) (2 - ak)r* ’
where <7^(2) is a polynomial in 2 of degree not higher than (rk — 1 ),
and gk(ak) # 0 . Consider the polynomials

/* (* ) = g * ( * ) (z -
~ — k •
~
akYK ( 20 )
We obvionsly have the identity

k= 1

or, replacing the argument z by the matrix A

1 = 2 fk(A )-
k=1
Thus any vector x can be represented as the sum of 3 vectors

x = 2 f k{A )x. (2 1 )
fc=1
Let us define certain subspaces Ev . . Es, viz. assume that Ek is
a subspace which is given by the formula
£ = fk(A )x ( * = 1 , 2 ......... s). (2 2 )
We shall see later that none of these subspaces Ek is empty. Denote
by xk any vector in the subspace E k. We shall prove, first of all, the
following two formulae:
fp(A) xq = 0 when p + q and f p(A) xp = xp . (23)
In fact, we have by definition
= fq{A) x ,

where x is any vector in the whole space. We thus obtain from (20):

fp(A )x q = gp{A)gq(A) [<?W]2 I


(A — ap)rp ( A — rg)r.

If p and q are different then the fraction on the right-hand side


of the equation is a polynomial containing <p(A) and therefore, as a
result of Cayley’s identity, this polynomial will be a zero matrix;
this proves the first of the formulae (23). To prove the second
191] TH E FIRST STAGE OF TH E TRANSFORMATION IN T H E CASE OF REPEATED ZEROS 679

formula it is sufficient to assume in formula (21) that x = xp and make


use of the first of the formulae (23). We then obtain directly the second
of these formulae. We will now show that these subspaces form a
complete system of subspaces. Formula (21) shows that any vector
can be represented as the sum of vectors from the subspace Ek. Hence
we have only to show that there can be no linear dependence among
vectors in these subspaces. Assume that this linear dependence does
exist
@i xi + C'2x2 + .. . + Cs xs = 0 , (24)
where the vector xk belongs to the subspace Ek. We must show that if
x k is not zero then the coefficient Ck must be zero. Applying the linear
transformation f k(A) to both sides of the equation (24) we obtain
from (23)
Ck xk = 0,
which proves our hypothesis.
Hence the constructed subspaces Ek do, in fact, form a complete
system of subspaces and the sum of their dimensions must be n, i.e.
the same as the dimension of the complete space.
Each of the subspaces E k can be defined in a different way from
the above definitions, viz. it can be shown that the sub space Ek is
defined by an equation of the form
( A - a * ) r*x = 0, (25)
i.e. it represents a set of vectors which satisfy this equation. In fact,
suppose that we have a vector % given by formula (22) and let us
show that it will satisfy the equation (25). In fact, substituting the
expression
S = f k(A )x
instead of x in the equation (25) we obtain on the left-hand side of this
equation an expression

(A — ak)T*fk(A) x = gk{A) <p{A) x ,

and as a result of Cayley’s identity [(p{A) = 0] the result will, in fact,


be zero. We now have to prove the converse, viz. that any solution ij
of the equation (25) can be obtained from formula (22) for a given
choice of x. Furthermore, we will show that from the equation

{A — ak)r*?i = 0 (26)
680 TH E CONVERSION OF MATRICES INTO TH E CANONICAL FORM [191
it follows that
n = fk{A)n. (27)
In fact, we have from (21)

p~i
But each of the polynomials f p(A), when p ^ k, contains the factor
(A — ak)'1 and therefore, from (26), we have f p(A) y = 0 when p # lc
from which formula (27) follows directly.
Let us now return to [IIIj, 27]. If A = a* is a zero of the characteris­
tic equation then substituting this in place of A in the coefficients of
the system (105) we obtain a homogeneous system with a zero deter­
minant and we can therefore construct for it a solution which is not
zero. This solution yk will satisfy the equation
{A — ak)v k = 0,
and hence the equation (25) will also be satisfied by it, i.e. it forms part
of the subspace Ek which therefore cannot be empty.
It follows from the form of the equation (25) that each of the sub­
spaces Ek will be invariant in relation to the transformation by the
matrix A. In fact, if a vector x satisfies the equation (25) then it is
clear that the vector A x will also satisfy this equation, since
(A — ak)r*Ax = A(A — ak)r*x .
Let qv . . . , qs be the dimensions of the subspaces Ev . . . , Es. Select­
ing the fundamental axes in these subspaces in the way described in
the previous section we obtain instead of the matrix A a similar
matrix in the quasidiagonal form
S, A S r1 = [Av A 2........As] , (28)
and the component matrices will be of rank qk. We will now show that
the numbers qk are the same as the order rk of the zeros of the charac­
teristic equation and that every matrix A k has a single characteristic
zero ak which occurs rk times.
To prove this take any vector § in the subspace Ek. It must satisfy
the equation (25). With the new choice of axes this equation can be
rewritten in the form
SiM - •**)'*# r 1S = o .
But we have, for example
S jHA - aky- 5 f J = S X(A - Oft) S r 1S,(A - ak) S t 1 = (S1 A S i 1 - ak)*,
192] CONVERSION INTO TH E CANONICAL FORM 681

so that the above equation can be rewritten as follows:

(S1A S i1-a „ )r*i = 0,


or
lAi — a*. A 2~ ak, .. As — afc]r*g = 0 . (29)
Consider, for example, the case when k = 1.
In this case all components of the vector g, except the first qv will
be equal to zero and instead of the equation (29) we can write:
A = 0, (30)

where by g' we denote an arbitrary vector in a ^-dimensional space


and the matrix (Ay — a1)fl is of order qv Owing to the fact that the
equation (30) holds for any vector g' we have

(A, - cq)'. = 0 .

It follows that all the characteristic zeros of the matrix (Ay — aL)Tl
must be equal to zero. But they are obtained from the characteristic
zeros of the matrix Ay—cq by raising them to the power of ry, conse­
quently all characteristic zeros of the matrix A1 — cq are equal to zero
and all characteristic zeros of the matrix A are equal to a1. It can be
shown similarly that in the general case all the characteristic zeros of
the matrix A k of rank qk are equal to ak. But the matrix (28), which is
similar to the matrix A, must have the same characteristic zeros as
the matrix A. Its characteristic equation has the form:

A I A — K A 2 - X , . . . , As — X\) = 0
or [189]:
D(Ay - X) D(A2 - X ) . . . D(AS - X ) = 0.

It follows that qk must coincide with rk and that the matrix Ah


has a single characteristic zero ak which occurs rk times.

192. Conversion into the canonical form. We saw that each of the
matrices A k has a single characteristic zero ak which occurs rk times.
To convert this matrix to the canonical form mentioned at the begin­
ning of this chapter it is sufficient to select in a definite manner the
axes in the subspace Ek. We thus have to investigate a particular case,
viz. a matrix with a single characteristic zero. Suppose that a matrix
D of order r has a single characteristic zero a which occurs r times.
682 TH E CONVERSION OP MATRICES INTO TH E CANONICAL FORM [192

The matrix B = D — a will have a single characteristic zero equal


to zero of order r and it is this matrix which we shall now consider.
As a result of Cayley’s identity we have Br = 0, since the left-hand
side of the characteristic equation must be the matrix B of equal
to (—l)r Ar. It may happen that B ! = 0, where I is a positive integer
smaller than r. Take the smallest positive integer I to which following
formula applies
B l = 0. (31)
If, for example, the matrix B itself is equal to zero then I = 1. For the
matrix
0, 0, 0, 0
0, 0 0, 0,
B = 0, 0, 0, 0
1, 0, 0, 0
it can easily be shown that B 2 = 0.
If the matrix B is equal to zero then D = B + a is a diagonal
matrix
D = [a, a, . . ., a] ,
and therefore we have the canonical form already. Hence it is only
important to consider the case when I > 1.
As a result of the condition (31) the equation
B lx = 0
describes the whole space of r dimensions. We shall in future denote
this by co. Let us now construct the equation
B1- 1x = 0 .
Since the matrix Bl~x is not equal to zero this equation gives a
subspace the dimensions of which are smaller than r. We shall generally
definite the sequence of subspaces by the following equations:
B‘x = 0; Bl~1x — 0 ; . . . ; Bx = 0 , (32)
and denote by F m a subspace given by the equation Bm x = 0. Let
rm give the dimensions of this subspace. As we have already said
above, Fl coincides with the whole space co, and rt = r, where T(_x < r;.
If a vector f belongs to the subspace F m, i.e. satisfies the equation
Bm g = 0, then the vector B% satisfies the equation Bm~1 (B%) = 0,
i.e. it forms part of the subspace f m_j. It is also obvious that any vector
192] CONVERSION INTO TH E CANONICAL FORM 683

of the subspace F m also belongs to the subspace F m+1, i.e. the sub­
space F m forms part of the subspace F m+1. We shall see later that the
subspace F m always has smaller dimensions than the subspace F m+1,
i.e. the subspace F m forms a proper part of the subspace F m+1 but does
not coincide with it. We have, for the moment, the following inequali­
ties:
T, > T;—1 > T,_2 > . . . > T lt (33)

and we will show that in each of these equations the strict inequality
holds.
Let us put t ( — r;_j = rt, where rt is a positive integer. In the
subspace Ft (in other words, in the whole space co) we can construct
Ti linearly independent vectors 2jlP . . . , 2jrJ, so that none of their linear
combinations belongs to F ^ v In this case any vector of Fi can be
represented as a linear combination of the vectors glt . . . , §rJ and a
vector of Ft^v To construct these vectors gx, . . . , grj we select
linearly independent vectors in any way we choose, for example,
in the subspace Ft_v In this case the vectors Jq, . . . , £r( will complement
these latter vectors and form a complete system of linearly-independent
vectors in the space a>. Denote similarly r,_1 — r;_2 = r/_1, where
r(-1 is a positive integer, and in the subspace Ft^x construct r(_1
linearly independent vectors so that none of their linear combinations
belongs to the subspace Ft- 2. Denote these vectors and any of their
linear combinations by y. Consider now the vectors

........B l,r (34)


They all belong to the subspace Ft- V Let us show that none of their
linear combinations can belong to the subspace Fj_2. For otherwise
we should have
B'-*(cl B lt + . . . + cr B lr) = 0
or
B ,- 1(c1l 1 + . . . + c,il,) = 0,

i.e. it appears that the linear combination of the vectors ........l r[


belongs to the subspace Bl~x which contradicts the definition of these
vectors. We thus see that the vectors (34) are linearly-independent
vectors belonging to the subspace Ft and that they are the vectors
rt of this subspace, i.e. none of their linear combinations belongs to
the subspace Ft^2. It follows directly that r/_1 > r(. Similarly, denoting
684 TH E CONVERSION OP MATRICES INTO TH E CANONICAL FORM [192

Tj_2 — t ;_3 by r,_2 we obtain r,_2 > and, in general putting


rm— = r m, we have:
0 < r, < r,_! < r,_2 < . . . < rx (r2 = t x) . (35)
It also follows directly that in formula (33) we have everywhere
the sign > , i.e.
r, > Tt_x > . .. > Tj . (36)
The number r, can be called the number of dimensions of the sub-
space Fi in relation to the subspace Ft^x which is contained in it.
More strictly rt is the number of linearly-independent vectors of Ft
which are such that none of their linear combinations belongs to F ^ v
These vectors form a subspace (?, which is part of Ft. Similarly,
gives the dimensions of Ft^x in relation to Ft_2 and we obtain, in the
same way as above, a subspace G(_x wliich is contained in Fi_v
In general, rm gives the dimensions of F m in relation to F m^x and there
are rm linearly-independent vectors of F m having the characteristic
property that none of their linear combinations belongs to F m- X, they
form a subspace Gmwhich belongs to F m. The subspace G1coincides with
Fv If | is a vector of Gm, and therefore also of F m, then !?£ belongs to
F m- V It can, however, no longer belong to F since we would other­
wise have Bm~2(B£) = 0 and, consequently, the vector £ would belong
not only to F m but also to F m_v which contradicts the definition
of the subspace Gm. Hence, by applying the linear transformation B
to the subspace Gm, we obtain part of the subspace Gm_j (or all the
subspace Gm_±) and the linearly-independent vectors in Gm are trans­
formed into other linearly-independent vectors in Gm- V From the
formulae
r m = Tm T m —l (^ l = T l)

and since rt — r, it follows directly that:


ri + '>'1 -1 + -- - + r1 = r,
and the subspaces Gt, . . . , G1 evidently form the complete system of
subspaces.
We finally come to the last stage of the construction, viz. to the
construction of the final subspaces which are invariant in relation
to the linear transformation by the matrix B. Take a vector in Gx
for the first axis and construct (I — 1) more axes according to the
following equations:
I 2=BSi : = ...; £, = -81,-! (££, = £ '£ , = 0).
192] CONVERSION INTO TH E CANONICAL FORM 686

It follows from the above arguments that these axes are linearly-
independent and belong successively to the subspaces Gt, G,_p . . . , Gv
It can readily be seen that they form a subspace, invariant in relation
to the linear transformation B. In fact, it follows from (37) that for
any choice of the constants ck:
-®(C1Si + C2§2 + • • • + Cl £l) == C1£2 + C2§3 + ■• ' + Cl- 1S/ •
It follows directly that the matrix of the linear transformation of
the invariant subspace so formed, when the are taken for the
axes, will be a matrix of order I in canonical form:
0, 0, 0........0, 0
1, 0, 0, . . . . 0, 0
7,(0) = 0, 1, 0, . . ., 0, 0

0, 0, 0, . . ., 1, 0
Using in this way one of the vectors of G, we can take any other
vector ?/1 of G(, which is linearly independent of Jjp and add to it
(I — 1) more vectors, according to the formulae:
*l2 = B tii; i}3 = B t\ 2 \ . . . ; fit = B n t- i .

The I vectors so constructed will be linearly-independent not only


of each other but also of the vectors gk. This is directly due to the
fact that linearly-independent vectors of Gm are transformed into
linearly-independent vectors of Gm_p by the transformation B.
Taking the vectors r)k for the axes we obtain an invariant subspace
corresponding to the linear transformation effected by the matrix
/,(0). Using all rt vectors of the subspace G{ we thus construct r, in­
variant subspaces of I dimensions for each of which the linear trans­
formation is effected by a matrix of the form 7;(0).
We now pass on to the next subspace Gt_x of dimensions, where
ri-r > r,. We have already used r( vectors of this subspace for the
construction of the axes. Using the remaining (r,_x — r2) vectors in the
same way as we did above we construct ( r ^ — r() invariant subspaces,
in each of which our linear transformation for the given choice of
axes, will be effected by a canonical matrix I/_!(0) of order (I — 1).
In general, when we reach the subspace Gmit will contain (rm — rm+l)
unused linearly-independent vectors. Selecting these vectors in any
way we please and applying to each in succession the transformation
B we obtain from each vector (m — 1) additional vectors and, taking
686 TH E CONVERSION OP MATRICES INTO TH E CANONICAL FORM [192

these vectors as the axes, we thus obtain (rm — rm+1) sets of axes,
where each set contains m axes and defines an invariant subspace of m
dimensions corresponding to the transformation due to a canonical
matrix I m(0) of order m.
Finally, when we reach the last subspace Gv it contains (r1 — r2)
linearly-independent vectors satisfying the formula 7?| = 0. Taking
these vectors as axes we obtain {r1 — r2) invariant one-dimensional
subspaces, corresponding to the linear transformation effected by a
zero matrix of the first order. As a consequence of the new choice of
axes we have a linear transformation a of the component vector and
the linear transformation caused by the matrix B, will now be due to a
similar matrix in the quasidiagonal form
o B o-' = [7^(0), I fi2(0), .. ., 7^(0)], (37)
where among the lower suffixes in the square brackets there are r,
equal to I, (r/_j —rx) equal to (I —1) etc. and, finally, (rx — r2) equal
to 1. We obviously have for the matrix D = B + a:
oDo-1 = aBo-1 + aaa-1 = oBa-1 + a ,
i.e. a is added to the diagonal elements and we thus obtain
oDo-1 = [/^(a), I pi(a), IpT{a) ] . (38)
Let us return, finally, to our initial matrix A. In the previous
section, according to formula (28), we represented it as a quasi­
diagonal matrix in which every component matrix Ak had a single
characteristic zero ak which occurred rk times. According to the above
section each matrix A k can be converted to the canonical form (38)
with the aid of a matrix ak of order rk. If we consider the matrix
S 2 — [dj, o2, . . . , ffs] ,
then we have
* *' > ^ sl ^ 2 = [*L -* ^ 1 ^ 1 ^ 2 ^ 2 *77 » • • • » ^ 5 -^ 5 Os ],
and our matrix A is finally obtained in the canonical form
(S28 J A(S2flfx)-i = [7ei(^), 7es(A2), . . . , 7tf(Ap) ] . (39)
The numbers Xj will be equal to ak and the sum of the lower suffixes
of the component matrices I qj(Xj), where Xj = ak must be rk.
Formula (39) completes the conversion of a given matrix to a canoni­
cal form. The question of the uniqueness of such a representation
arises, i.e. it has to be proved that for any method of conversion to the
193] TH E DETERMINATION OF TH E STRUCTURE OF TH E CANONICAL FORM 687

canonical form there will be inside the square bracket on the right-hand
side of formula (39) a definite number of matrices iQj(Xj) for the given
6j and Xj. For example, suppose that the matrix A is converted to the
canonical form in an arbitrary way

Bearing in mind the fact that similar matrices have the same cha­
racteristic equation we can write the characteristic equation of the
matrix A as follows:

or
D (lIei{Xi - X), I &(X2 - X ) , . . . , I eP(Xp - A)]) = 0 ,
which is equivalent to the following [189]:
D il^X , - ;.)) D(Ie,(X2 - X ) ) . . . D(ItP(Xp - X)) = 0 .
but because of the form of the matrix I g(a) it follows that
D[Ie(a)] = ( a- X) ° . (40)
Hence the numbers Xj must be equal to the characteristic zeros
ak of the matrix A and the sum of the symbols Oj, for which Xj = ak,
must equal the order rk of the characteristic zero ak. It
remains to show that all the numbers Qj must have a definite value.
This can be proved by using the same geometric concepts as in the
proof of the conversion of a matrix to the canonical form. In doing
this the consideration of invariant subspaces will be of great importance
We shall not give this proof here but in the following section we
shall indicate an algebraic criterion which determines the values of all
the symbols qj for the given matrix A. This criterion, based on consi­
dering the highest common factor for determinants of a given order of
the matrix (A — A), is given without proof in the first part of this volu­
me [IIIj, 27], It will also establish the uniqueness of the representa­
tion of a given matrix in canonical form.

193. The determination of the structure of the canonical form. As a


preliminary let us prove two auxiliary lemmas.
L e m m a 1. I f A and B are two square matrices of order n and C = A B
is their product, then any determinant of the matrix C of order t, where
t < n, can he represented as the sum of products of certain determinants
of order t of the matrix A and determinants of order t of the matrix B.
688 TH E CONVERSION OF MATRICES INTO TH E CANONICAL FORM [193

This lemma follows directly from the theorem proved in [III^ 6].
C o r o l l a r y . Suppose that the elements of the matrix A(X) are poly­
nomials of X and that the elements of the matrix B do not contain X,
where the determinant of the matrix B is not zero. Denote by dt(X)
the highest common factor of all determinants of order t which belong
to the matrix A(X) and by d't{X) the highest common factor of the
matrix A(X)B. It follows directly from the theorem that dt(X) must
be a factor of d't{X). But we can write
A{X) = {A{X) B ] B - \
and the above lemma also gives us directly that d't{X) must be a
factor of dt(X), i.e. dt(X) and d't(X) are equal. We would obtain the same
result if instead of the matrix A(X)B we had constructed the matrix
BA(X).
It also follows that the highest common factors of the matrix zl(A) and
of the similar matrix BA(X)B~1 will he equal.
Let us explain one more property of the highest common factor
dt(X). To do so we shall require a new definition.
D e f i n i t i o n . B y an elementary transformation of the matrix A(X),
the elements of which are polynomials of X, we understand a trans­
formation of this matrix by means of a finite number of the following
three operations:
(1) the transposition of two rows ( or columns);
(2) the multiplication of all the elements of a row ( or column) by
a certain constant, other than zero;
(3) the addition to the elements of a certain row ( or column) the
corresponding elements of another row ( or column) which are al multi­
plied by a certain constant or bz a certain polynomial of X.
If the matrix A x(X) is obtained from A{X) by means of an elementary
transformation then, evidently, the reverse will also be true and A(X)
can be obtained from Ax(X) by means of an elementary transformation.
If two matrices can be transformed into one another by means of
an elementary transformation they are said to be equivalent.
L e m m a 2 . Equivalent matrices have the same highest common factors
dt{X) (t = 1, 2, . . . , n).
It is sufficient to show that when all the determinants of order t of the
matrix A(X) contain a common factor which is a polynomial of cp{X)
then all the determinants of order t of the equivalent matrix A^X) will
contain the same factor. The first and second of the above three trans­
formations add a numerical factor other than zero to determinants of
193J TH E DETERMINATION OF TH E STRUCTURE OF TH E CANONICAL FORM 689

order t and for these two transformations the lemma is evident. It


remains to show that the common factor <p(X) will also remain in the
third transformation. Suppose, for example, that this transformation
involves the addition to elements of the p row of the corresponding
elements of the qth. row, q ^ p which were previously multiplied by
the polynomial y>(X). All determinants of order t which do not contain
a pth row or which do contain the pth and gth rows will not alter
during this transformation because of the property VI of a determinant
[III1( 3]. Determinants of order t which contain a pth row but which
do not contain a gth row will have the following form after the trans­
formation: A'(X) ± ip(X)A"(X), where A'(X) and A ”(X) are determinants
of the matrix A(X) of order t. It follows from what was said above that
the factor <p(X) of the determinant A(X) of order t will, in fact, be a
factor of all determinants of order t of the matrix A^X).
L e m m a 3 . Any matrix of the form

a —X 0, 0, •, o, 0
1, a — X, o, . o, 0
0, 1, a — X, . o, 0 (41)

0, o, 0, 1, a — X
of order q can be obtained in the form of a diagonal matrix [1,1, . . . , 1,
(a — A)tf] by means of an elementary transformation.
When g = 1 this lemma is trivial. Consider the case when q = 2.
Interchanging the rows and multiplying subsequently the elements
of the first column by —(a — X), adding the products so obtained to
elements of the second column and doing likewise with rows we obtain
the required result after dividing the last column by ( —1)
O

1, a —X
1

1, 0
O

1, a — X,
1

C4
e
1

i, o
- [1, (a — A)2] .
0 , - (a - xy

For a matrix of the third order we obtain the following result after
performing the above transformations:
a — X, 0, 0 1, 0 0
L a — X, 0 o, (a —xy, 0
o, 1, a —X 0, —1, a — X
G90 TH E CONVERSION OF MATRICES INTO TH E CANONICAL FORM [193

Interchanging the second and third rows and performing further


elementary transformations we obtain:
1, 0, 0 1, 0, 0

1
1
0, ( a - x y , 0 ->
0, - 1 , a — A 0, { a - x y , 0
1, 0, 0 1, 0, 0 1

h-I
1
0, - 1, 0 ->

o
0
0, ( a - x y , (a - A)3 o, o, (a - A)31
and after dividing the second column by (—1) weobtain [1,1, (a — A)3].
In this way we can gradually prove our lemma for a matrix of any order.
We shall now prove the algebraic criterion of structure for the cano­
nical form of a matrix A which was given in [IIIj, 27]. The corollary
from Lemma I enables us to find the highest common factors d,(X)
not for the matrix A — X but for a similar matrix
V(A - A)F-i = VAF - i - X= [Jei(Ax- X). I J X 2~ X), . . . . I ep(Xp - X)] .
(42)
Applying Lemma 3 to each matrix in the quasidiagonal form we can,
by choosing the highest common factors dt(X), replace the matrix (42)
by a purely diagonal matrix which has along its main diagonal (gx — 1)
units, (X1 — A )01 units, ( q2 — 1 ) units, (X2 — Xf2units etc. in succession.
Notice also that if, during the construction of a determinant of order t
belonging to this matrix we rule out a 6et of rows, then in the deter­
minant so obtained at least one row and one column will consist of
zeros and, this determinant will be equal to zero. Hence when con­
structing determinants belonging to a diagonal matrix we must
always rule out the same rows and columns which simply involves
the ruling out of diagonal elements, the product of which gives the
value of the determinant.
Consider one zero X = a of the characteristic equation which occurs
It times. The determinant of order n must contain the factor ( A — a)k .
Suppose that the highest common factor of the determinants of
order (n — 1) contains only the factor ( A — o)kl. This means that the
highest power of (A — a) which belongs to the constructed diagonal
matrix is equal to (1c — fcx), i.e. the canonical representation of our
matrix includes a matrix I k-kl(a), and so on. If the highest common
factor of determinants of the order (n — 2) is equal to (A — a)*2
it means that after (A — a)k~kl the highest power of (A — o) belonging
194] EXAMPLES 691

to the constructed diagonal matrix is equal to (A — a)k'~ki, i.e. the


canonical form includes both matrices 7*_kl(a) and 7^_fc2(a). When we
finally arrive at determinants of a certain order of which at least one
does not contain the factor (A — a) then we absorb all the compo­
nent parts of the canonical form A for which A = a. Hence the
algebraic criterion for the canonical structure of a matrix mention­
ed in [IIIlP 27] is proved. Notice that it not only follows from the
above arguments that
k > lc1 > k2 > . . . > km, but also that lx > l2 > . .. > lm > lm+1,
where
l\ = k k±, l2 = k± k2, . . . ; lm = km_l km] lm+1 = km.

194. Examples. We can solve the characteristic equation of a given


matrix A and therefore the canonical form can be directly determined,
for example, by means of the algebraic criterion mentioned in the
previous paragraph. The problem remains of how to construct a
matrix V with a determinant other than zero which would convert
the given matrix A to the canonical form. In deducing the transfor­
mation to the canonical form we successively choose new axes. The
choice of these axes finally converted the given matrix to the canonical
form. But we know from [III!, 21] how, for a given transformation
of axes, we can construct a transformation V to convert the matrix A,
as an operator, to the new form. If T is the linear transformation of
axes then the matrix A will be converted to a new form U A U -1,
where U = T *-1 i.e. to obtain U from T we must interchange the
rowB and columns in T and take the inverse matrix.
Let us now systematize the solution of the problem. First we select
the new axes so as to obtain our matrix in the quasi-diagonal form,
taking into account its various characteristic zeros as described in
[191]. In this case the choice of the new axes involves the solution
of an equation of the first degree in the following form: (A —ak)r*x = 0.
Then we have to convert the matrix B with a single characteristic
zero equal to zero to the canonical form. Here we have, at first, to
determine the smallest number I, such that B l = 0, and then we can
construct a Bystem of equations of the first degree of the form
£ '- 1x = 0.
To determine the order of this system of equations we take those
vectors which do not satisfy it and, on subjecting them to the trans-
692 TH E CONVERSION OF MATRICES INTO TH E CANONICAL FORM [194

formation B, we construct a new series of axes, etc. We thus obtain


a second transformation of axes and, at the same time, a second
transformation of a matrix similar to the fundamental matrix A; this
finally gives us the matrix in the canonical form. We shall explain
these general ideas by a numerical example.
Consider a matrix of order 5:
- 2 , - 1 , - 1, 3, 2
-4 , 1, - 1 , 3, 2
A = 1, 1, o, - 3, — 2
- 4 , - 2 , - 1, 5, 1
4, 1, 1, — 3, 0
Constructing its characteristic equation in accordance with the
usual rules we obtain it in the following form:
(A - 2)3 (A + 1)2 = 0 ,
i.e. this equation has a zero A = 2 of order three and a zero A= —1
of order. Let us now construct the matrices (A — 2)3 and (A + l)2.
The equation (A —2)3x = 0 must give a three-dimensional subspace,
i.e. the matrix {A — 2)3 is of the second rank.
Similarly the matrix (A + l)2 is of the third rank. By an elemen­
tary calculation we find:
-54, o, - 2 7 , 27, 27
-54, o, - 2 7 , 27, 27
[A - 2)s = 27, o, o, - 2 7 , — 27
-54, o, - 2 7 , 27, 27
54, o, 27, - 2 7 , - 27
x = 0 can be written as the
-2 7 :r3 + 27x4 -f- 27*5 = 0
27*4 — 27*4 - 27*s = 0,
where (xv x2, x3, x 4, x5) are the components of the vector x.
We thus have:
*1 = *4
c4 +
T *5
^5) *X^ q --
3 — X
^ 4a X 5i f

where x2, *4 and x3 remain arbitrary. On supposing that one is equal


to unity and the others to zero we obtain three new axes which have
the following components in terms of the former system:
( 0 , 1 , 0 , 0 , 0 ); ( 1, 0 , - 1 , 1 , 0 ); ( 1, 0 , - 1 , 0 , 1) . (43)
194] EXAMPLES 693

Similarly, as a result of elementary transformations we obtain:


0, - 6 , 0, 9, 3
-9 , 3, 0, 9, 3
(^ + 1)2 = o, 6, 0, - 9 , — 3
- 9 , - 12, 0, 18, — 3
9, 6, 0, - 9 , 6
and equation {A + l)2 x = 0 can be written as a system of three equa­
tions:
— 2x2 + 3z4 + x5 = 0
— 3Xj + x2 + 3x4 -f- x6 = 0
3 ^ + 2x2 — 3 x 4 + 2x- = 0 ,
or
x2 = x±; x4 — xx; xs = — x lt
where x1 and x3 remain arbitrary. This gives us two new axes:
( 1 , 1 , 0 , 1 , - 1 ) and (0,0,1,0,0). (44)
The new axes (43) and (44) will be expressed in terms of the former
axes by the formulae:
ei e2

eZ = ei — e3 + ^4

e3 ~ ei — e3 + e5

ei = ei + e2 + e4 — e5

eS = e3

The matrix of this linear transformation has the form


o, 1, 0, 0, 0

1, 0, ■- 1 , 1, 0

1, 0, -- 1 , 0, 1

1, 1, 0, 1, -- 1

o, 0, 1, 0, 0

and by interchanging the rows and columns and taking the inverse
matrix we obtain:
0 , 1 , 1 , 0 — 1 , 1 , 0 , 1
1 , 1 ,

1 , 0 , 0 , 1 , 0 - 1 , 0 , 0 , 2 , 1

S f1 = T w =
x
CQ

H
II

II

0 , - 1 , 0 , 1 1 , 0 , 0 , 0
H

- 1 , - 1 ,

0 , 1 , 0 , 1 , 0 1 , 0 , 0 , — 1
- 1 ,

0 , 0 , 1 , - 1 , 0 0 , 0 , 1 , 1
1 ,
694 TH E CONVERSION OF MATRICES INTO TH E CANONICAL FORM [194
Multiplying the matrices in accordance in the usual way we con­
vert the matrix A to the quasidiagonal form which consists of matrices
of the third and second ranks:

1. 0, - 1 , o, 0
-2, 2, - 2 , o, 0
1, 0, 3, 0, 0 (45)
o, 0, o, - 2 , - 1
o, 0, o, L 0
The matrix of the third rank
1, 0, - 1
A - 2 , 2, - 2
1, 0, 3
has a characteristic zero X= 2 occurring three times. Let us construct
the new matrix:
- 1, 0 , 1
B 1 = D 1 — 2 = - 2, 0, •>
1, o, 1
with the characteristic zero X = 0 occurring three times. By squar­
ing it we obtain Bl = 0. Hence we have, in this case, 1 = 2 and the
system B2x = 0 can be written aB the single equation
+ z3 = 0 .
The subspace F2 in our former notation coincides with a complete
three-dimensional space and the subspace Fx can be formed by the
vectors (1, 0, —1) and (0, 1, 0). We take the vector (1, 0, 0) which
does not form part of Fv It forms the subspace G2. Subjecting this
vector to the operation Bx we obtain
5 1(1,0,0) = ( - 1 , - 2 , 1 ) .
The vectors (1, 0, 0) and (—1, —2, 1) form the first pair of new axes
to which the following canonical matrix corresponds:
0, 0
1, 0
For the third axis we can take any vector Fl which is linearly-
independent of the vector (—1, —2,1). Let us take the vector (0,1,0).
The zero canonical matrix of the first order corresponds to this vector.
194] EXAMPLES 695

The new axes will be expressed in terms of the former axes by the
formulae:
.n —t
ei — ei
e'2 — e[ — 2e2 + e'3
e3 = e'2.
We now take a matrix of the second rank which belongs to the
quasidiagonal matrix (45):

This matrix has a characteristic zero X = —1 occurring twice.


We construct the matrix

B 2 = D2 + 1 = ” lf “ 1
2 2 1 ,1
with a characteristic zero X = 0 occurring twice. Obviously B\ = 0,
as it should be in accordance with Cayley’s formula. The equation
B2 x = 0 is equivalent to x4 + xB= 0, where x4 and xs are the com­
ponents of x in the two-dimensional space under consideration. We
take for the first axis the vector (1, 0), i.e. x4 = 1 and operating by B2
on x6 = 0, which does not satisfy the equation B2 x = 0, we obtain

B2(1,0) = ( - 1 , 1 ) .

Hence the two new axes will be (1, 0) and (—1, 1) so that we obtain
the following expressions which give the new axes in terms of the
former axes:
-ft -/ . J! -t I -t
e 4 — e 4> e 5 — — ei "I" e 5

or, taking into account earlier formulae:


e{ = e[
e2 = — e[ — 2e2 + e3
(46)
e\ = e't
el= - ei + ej.
The cannoical matrix 72(0) will correspond to the last two axes.
696 THB CONVERSION OF MATRICES INTO TH E CANONICAL FORM [194

We must add two to the first two matrices along the main diagonal
and (—1) to the last canonical matrix. The final canonical form for
the matrix A is:
2, 0, o, o, 0
1, 2, o, o, 0
o, o, 2, o, 0 = [I2(2 ),/1(2)172( - 1 ) ] . (47)
o, o, o, - 1, 0
0, o, o, 1, — 1

We finally construct the matrix V which transforms A to the form


(47). As we know from above it is a product of S2 Sv where has
been obtained above and S2 can be determined from the formula
s 2= t [*>-\
where Tx is the matrix of the linear transformation (46). We have:

1, - 1 , 0, 0, 0 1, o, 1, o, 0
0, - 2 , 1, 0, 0 o, o, 1, 0, 0
T{*> = 0, 1, 0, 0, 0 and S 2 - T ^ - 1 = o, 1, 2, 0, 0
0, 0, 0, 1, - 1 o, 0, 0, 1, 1
0, 0, 0, 0, 1 o, 0, o, 0, 1
Multiplying these two matrices we obtain:
0, 1, 0, 0, 1
1, 0, 0 , - 1, 0
V = S2S 1 = 1, 0, 0, 0, 1
1, 0, 1, 0, 0
0, 0, 1, 1, 1
Finally
VAV~ i = [ I 2(2), 7,(2), 72(— 1)].
INDEX

Abel’s theorem for infinite region 29


first 45, 317 for two variables 315
second 101 -R iem ann equation 11
Absolutely convergent series 40 test, existence o f lim it 4
Algebra fundamental theorem o f 89 theorem 21, 22
Analytic on number o f roots 88
continuation 63 Cayley’s identity 339
along curve 67 Characteristic
chain of regions 66 equation 674
o f function o f two variables 323 roots o f m atrix 674
o f m atrix 342 Christoffel’s formula 146
theorem o f uniqueness 69 outside
uniqueness 66 o f polygon 157
function 67 o f square 159
Asym ptotic expansion 407 rectangle 154
regular n-sided polygon 157
triangle 154
Bernoulli numbers 289, 291 Circle o f convergence 46
Bessel Complex number, interpretation o f 3
equation 379, 424 Conformal transformation 16, 120
functions 427, 637 conjugate trigonometric series 164
addition theorem o f 551 Gauss’s equation 398
asym ptotic representation o f 434 o f second kind 123
integral representation o f 550 Conjugate trigonometric series 103
orthogonality of 545 Continuity o f functions 6
of purely imaginary argument Contour
560 o f Cauchy’s typ e 32
relationships o f 539 lim iting values o f 113
Biharmonic integral 17
expression (Hurse’s formula) 201 assessm ents o f modulus 18
functions 200 conditions for independence 19
Bilinear transformation 125 Convergent series o f matrices 325
o f upper half-plane into itself 131 Conversion function
Branch-point 73 Bessel functions 549
order of 81 Herm itian 589
type o f 81 Laguerre polynomials 596
Legendre 506
Cotangent, expansion into simple frac­
Cauchy tions 251
formula 27 Curve in plane 16
697
698 IN D E X

D erivative 7 Flow o f liquid, two-dimensional 172


geometric meaning Fourier-Bessel formula 552
o f argument 15 Fractional function (meromorphic)
modulus 14 248
of matrix 350 construction o f 254
Determining equation 370 expansion into simple fractions 248,
Differential equation o f second order 255
357 Fresnel’s integral 223
analytic continuation o f solutions Full flow round contour 178
636 Function of complex variable 6
determining equation of 370
existence and uniqueness o f 361 Gauss’s equation 381
irregular singularity o f 368, 403
periodic coefficients 445 Hankel functions 428, 553
regular singularity o f 367, 369,
asym ptotic formulae 432
375
Harnak’s theorem 34
solutions, singularities 366—367
Hermitian polynomials 586
Diffraction o f two-dimensional wave
asym ptotic expression of 597
212
orthogonality o f 587
Dirichlet’s problem 518
Hilbert
core 193
formula 193
Elliptic problem 196
coordinates 661 Hurse’s formula 202
function 618 Hypergeometric series 381
o f order m 622
integrals 604 Infinite product 258
complete 611 for bilinear transformations 133
first, second convergent 258
o f Legondre’s form 611 for [ T ( z ) ] - 1 276
and third kind 60? Integral(s)
Jacobian functions 650 calculus, fundamental formula 24
addition formulae o f 656 depending on parameters 264
differential equations o f 654 function 256
modulus o f 652 construction from roots 261
Equations o f Fuch’s class 376, 377 periodic, expansion o f 636
Equivalent matrices 689 indefinite 266
Essential singularity 38, 61, 62 Integration
Euler o f rational fraction 227
constant 276 o f trigonometric expressions 225,
integral 228
o f first kind 272 Interior point 5
o f second kind 267 Interpolation polynomials 337
representation by contour im Invariant functions j(J(w) 631
tegral 279 Irregular singularity, differential equa­
summation formula 290 tion 368, 403
Exponential function 2, 12, 64 Isolated singularity 36, 80
o f matrix 328 Isothermic nets 122
IN D E X 699

Jacobian Picard’s theorem 39


formula 454 Point at infinity 61
polynomials 394 Pole 38
Jordan’s lemma 232 of order m 60
point at infinity 61
Potential
Laguerre o f complex flow 172
function 594 o f spherical shell 622
orthogonality of 594 o f voluminous masses 620
polynomials 439 Power
asym ptotic expressions of 604 function o f matrix 332
Lamd equation 663 series 45, 296
Laplace transformation 410, 435 conversion o f 92
Laurent’s series 50 differentiation and integration of
Legendre 47
function 630 o f m atrix 325
of second kind 634 conversion of 329
expressed as integral 535 m ultiplication 330
polynomials 391, 504 theorem o f uniqueness 328
addition theorem 513 o f tw o variables 316
asym ptotic expression o f 600 Principal value o f integral 103
orthogonality of 392 contour, o f Cauchy’s type 107
relationship 628 in infinite interval 231
Limit Principle o f permanency 69
o f complex variable 4 Pseudoelliptic integral 606
sequences o f matrices 324
Linear transformation 124
Liouville’s theorem 35 Radius o f convergence 45
Lipschitz condition of order 104 Reflection
Logarithm 2, 66, 78 in circle 134
matrix 331, 345
o f elastic waves 216
in straight line 134
Region 6
Magnetic field 188 bounded 6
electrostatic 181 closed 5
Meromorphic see Fractional Function connected 20
Modulus, principle o f 35 positive direction of describing 21
Regular
function (holomorphic) 8
Neighbourhood o f matrix 325 in closed region 20
Neumann o f m atrix 328
function 553 at point at infinity 61
expression in terms o f Bessel of tw o variables 313
functions 655 uniqueness 63
problem singularity of differential equation
exterior 619 367, 369, 375
interior 519 Residue
N ew ton’s binomial 56 calculation of 87
700 IN D E X

o f function constant coefficients 242, 348


at essential singularity 62 existence and uniqueness of
at infinity 62 solution 453
at pole 60 expansion, irregular singularity 477
sum o f rational fraction 246 integral m atrix 456
theorem 85 regular 459
Riemann singularity 456
surface 74
sym bol 380
Taylor
theorem 144
coefficients, assessment o f 49
R oots o f regular function 64
R ouche’s theorem 89 series 49
Theta-functions 644
first 642
Trigonometric functions 2
Schrodinger’s equation 684
Tschebyshev’s polynom ial 398
Schwarz’s formula 189
Sigma function 632
Silvester’s formula 339 Uniform convergence o f series 41
Similarity transformation 126
Singularity 69
Sokhotski, Iu. V., theorem 38 W allis’s formula 287, 288
Spherical function W ave equation 203
o f order n (surface) 496 in cylindrical coordinates 577
definite expression o f 499 fundamental theorem for solutions
integral o f square o f 601 o f 206
orthogonality o f 600 homogeneous solution of 206
o f volume 496 in spherical coordinates 680
Steepest descent m ethod 293 Weierstrass
Subspace 671 function 624
Symmetrical points in relation to infinite product 263
circle 98 theorem 42
Symmetry, principle o f 95 W hittaker
general 99, 133 equation 443
System o f linear differential equations function 444
453
canonical solution 469
comnlete solution 465 Zhukovskii’s, N . E., formula 179

You might also like