Smirnov - A Course of Higher Mathematics - Vol 3-2 - Complex Variables Special Functions
Smirnov - A Course of Higher Mathematics - Vol 3-2 - Complex Variables Special Functions
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
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
CHAPTER I
C H A P T E R II
CONFORMAL TRANSFORMATION
AND THE TWO-DIMENSIONAL FIELD 120
C H A P T E R III
C H A P T E R IV
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
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
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
] 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:
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 __
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 = -------- ,
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.
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
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
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
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\
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:
v y
^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£/(C*) (zfc—Zfc-x) =
k=l
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.
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:
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:
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:
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
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:
/;(* )= 0 .
We can write the following two expressions for the derivative:
11 ' > 0X + 3x dy 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:
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]
= 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
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):
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
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:
(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).
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)]
<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
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
<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):
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
,42)
^ • - - 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
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:
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
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.
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) ’
This series is said to be convergent when the sum of its first n terms
S„ = (a l + a2 + • • •+ a n) + + ^ + + &n) (^ 8 )
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,
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
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)
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-term integration:
z z
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:
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
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 >
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
f { z ) = m + I ^ L { z ~ b ) + J l ^ L { z - b Y + ... (63)
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:
i.e. given the above assumptions, we can add these infinite series like
ordinary polynomials.
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:
■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):
where
i r ___ m n
ak 27ii J iz' _ ,k+i ■dz'.
( * ' - by
IST
o c*.
JiA— d2' = 0 - 1( z - 6 ) - 1 + a_2(3- 6 ) - 2 + . . . ,
where
/ ( *) = ^ “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
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:
sm z — —2 S3 , 2s
gj- + -gj ... (74)
c o s z = l - — + -^- —. . . (75)
- ^ = 1 + * + =? + . . .
serves as an example of a series with a circle of convergence | z \ < 1.
16] exam ples 65
? < 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
(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):
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
T f T " = 1 — z2 + z4 — . . .
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
^ = 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
/(*) = 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
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,
<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,
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.
/(* )= 2 ak ( z - b ) k, ( 88 )
fc=—m
60 TH E BASIS OP TH E THEORY OP FUNCTIONS OP A COMPLEX VARIABLE [17
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)
positive powers of z:
f{z) = a_mzm+ a _ m+1 zm~1+ ... +o_! z + a 0+ — +
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.
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.
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 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
/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
two circles. Notice that the series (96) can be obtained from the
series (95) as follows: we rewrite the series (95) in the form:
fa (2 ) = 2 aW - b^ ’
k= 0
1
(z — b2)k .
2
k=0
(1 - bt)k+1
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
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
/ (z) = ± z ( l - ~ ) 2 (l - } ) 2-
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
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
/ (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
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:
/(*)= 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:
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:
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:
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:
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.
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 ^ ] .
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
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
— f— - l )
94 T H E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [2 3
{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'\
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
/i+L
0 = j2m
_ r /, (o dz'.
J z' —z
l.+L
h +l.
Similarly, by taking the point z in the domain B2, we obtain:
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
F i g . 24
(138)
2 a k (z -
k=0
b )k
100 TH E BASIS OF TH E THEORY OF FUNCTIONS OF A COMPLEX VARIABLE [25
2 a kek^ (139)
k=0
or
00
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)* '
(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:
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
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
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
(148)
/ ( x ) = j T = l T d(’
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
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
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
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 ,
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
e2
® + e’ + 7) = a: + e/i’ ({) -f — /t" (£ + 02e) (0 < 0, and 02 < 1) ,
whence:
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” .
f (f) = j 7 — | - d r - (153)
L
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:
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;
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:
{-*
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:
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):
A£
[co(t) — oi(£ + A£)] <
( ! - { ) ( ( - { - A£)
\A£ | k M* 1
1—a
1
Mr
Mr
a
1
w
9
1
1
1
1
1
t- £
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,)
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:
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
(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 .
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:
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:
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
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:
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
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 ■ '
| T - 11 = 2 sin I y ~ y° I > ■
— I cp - <Pv | ( I <P ~ <P» I < at) ,
28] INTEGRALS OF CAUCHY’S TYPE 115
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
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
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
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
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
(O(T)
>(£) = dr. (173)
m Jf -t
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:
— (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
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
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.
'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-—^ ,
<1 6 >
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).
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
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)
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
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.
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 ’ ' '
[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
w = exey'
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)
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
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 >
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.
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
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:
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----- + - + ■• •
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) + @
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
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)
1 I 03 1 c*
t* T 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).*
dr
z= A + B'.
7 Yt(t —1)
0
37] INDIVIDUAL CASES 166
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
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)
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
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)
t = + . (63)
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)
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*
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
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
we obtain the following system of equations of the first order for the required
coefficients ak of the polynomial (71):
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.
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:
+ c*
2 2J aij+i atj'+i I .*iP9
P” —“ L,j - j = p J
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:
+ [a| + a| + . . . + AT 0 (a2y+1)]2 + ■• •
(84)
• ~ ~ ^ ^ 1 (a 2]+l)
ui
• ^(a2y+i)
ai
-
+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
z2 + z2
zz — A 1,
2
<*! + <*! + °! + “ ? + a? + a J i + . . .
a, a , + ag a 6 + a 5 a, + a 7 a 9 -f- a 9 a n +
eo = ; pi = ; p2 = • (89 )
al a\
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:
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:
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, + • • • =
al a13 + • • • =
= y [ - ("T") (a. a®+ y al) ~ y (“i “,)2+ («i “i3+ “6“,) (y “2) + -••]
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:
1.0607a,,= 0.
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' ^ 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
whence a.™ = — 1/8, aj.2’ = 1/2 and the remaining a^’+1 are equal to zero, i.e.
22 = -y e '* + y e ^ .
and, finally, substituting in (95) and replacing e'k<p by t * we obtain the approxi
mate expression for the unknown conformal transformation:
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 .
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
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 _
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 \,
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
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 :
]»'<*) “ 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 =
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 ;
= | ‘ 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 - + . . . ,
n*> = « + T C + - ? - + - - -
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:
/ = -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 )
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
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
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
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“’,
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)
£ — 2 71 a / t*+ 1
188 CONFORMAL TRANSFORMATION AND THE TWO-DIMENSIONAL FIELD [47
= = {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.:
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)
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:
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
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 )
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)
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
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:
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.).
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
^ 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.
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
where (0 ,(95) are the limiting values of the imaginary part of the function (162):
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«
•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:
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:
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
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:
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
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:
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):
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.
( 1 9 7 >
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
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
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)
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
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)
£2 + V2 = p - (212)
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
(214)
63] THE FUNDAMENTAL THEOREM 211
<- 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)
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
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:
( 220)
*[/(•>'*)] = 1
« [ / ( o'')] = - l - — <<p<
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)
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
/(<")= — VM 2.
71
-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
/ ( 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:
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:
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
t - e t x -Y b * - d l y - Y a * - 6 \ y a = 0. (233)
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
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
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 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:
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.
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
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)
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
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)
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
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
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
cos mx
J X2-j- o2 dx (a > 0 ; m > 0 ).
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.
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
x sin mx _ nm _ ma
J (x2 + a 2) 2 d x ~ ~ l5" e
(21 >
J / (*) d *
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
~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
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.
Jf xa*sin+ mx
a» ^, (o,> .
0 ; m > 0).
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
Fio. 58
i29'
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 = ° <*«»•
* f-dz (30)
2m J z
-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:
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
where the contour of integration is the real axis, which surrounds the pole
z = a of the integrand. In this case
the residue is
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.
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
I J (-* )a-1«(*)da|<
Ca
< 2nR • Ba _ 1 max | Q (z) | =
on Ca
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:
J x a~ 1Q (x) dx = r> (4 0 )
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
[/ 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:
^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:
l | ^ + 2 T + 5 - dZ = 2^
hence the formula (45) gives us the final value of our integral (43):
____________
J = \ (48)
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:
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»=0 =js4trr-
xJ ^ A(z) (57)
- 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
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.
x (1 2) ~ —0 3 iz
R — z2+ z + 2
- (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.
<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
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
''"/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)
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.
----- 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
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
7i cot nz = ---- f- ( 81 . )
z
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
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:
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° ( 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
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)
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
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
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
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
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
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
F ( Z) = 77 “ A W
fc=l
= W ■u 2 ( 2 ) • • • (9 9 )
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:
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
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
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)
= L _ f ^ r +,_ 1
k~k, L m ' m+1 {ak ) J
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):
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
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.
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
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
J f{t,z)dt
a
w '{z)=\*t^-dt.
a
a>'n { z ) = ] ^ ^ - d t .
a
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
71. Euler’s integral of the second class. Consider the function given
by Euler’s integral of the second class:
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
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:
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
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:
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:
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 ) ;
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
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
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
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
1
r(z)r(-z)
73] THE INFINITE PRODUCT OF THE FUNCTION f/■(?)]-1 277
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 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
J0 [ ° " =
no n
-/[•" - ( ‘ - f ) “ ] ‘" l d ‘ + f [•" ' - ( ‘ d‘
0 no
an d from (138):
n no n
0<J ['" ~ ~ +JV v “
0 0 no
°< 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)
G z + z 2 k{z + k )- (141)
k= 1 v 1 '
74] THE REPRESENTATION OP AO DT A CONTOUR INTEGRAL 279
and from this follows formula (143). In exactly the same way as
above the following, more general formula, can be proved:
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
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
n n n
( n + \ ) 1 { n ) - 2 f ( k ) = J } J /' (x) dx. (155)
k-o k-ok
75] STIRLING’S FORMULA 283
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:
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:
+ (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]:
f - r £ * + . Jf f1 ■+<*X> Cx
= (2 • z) log z —
-■ > + ( 4 -
I) o
75] STIRLING'S FORMULA 285
or [73]:
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
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:
or
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 ’
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 <
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
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
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 i m (o ) = 1 i p ™,x(o)=o.
Bm
Am(0) = V (174)
(2 m) 1
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:
when z > 0:
_1_ l)m- ] 1
•6 z* (2m - 1) 2m 22m-i
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
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
- 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
U tl “ 1 (178)
- 1 + - = 2u22 '
eu — 1 2 h-i 4kz n*-{-u2
We can write:
— - 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
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)
< 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:
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
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:
(194)
298 THE APPLICATION OP THE THEORY OP RESIDUES, FRACTIONAL FUNCTIONS [79
According to formula (191) integral (188) can be broken up into two integrals:
= 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
*(<■>+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
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)
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
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:
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:
Therefore
J [ ( l + 2) e - z]"d2 = 0 [ ( - - | ]. (204)
1
The integral
l
J[(l+*)e-*]"d! (205)
-l
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):
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
_1
a
where
_i
2
« -
J
f —~
2 + 10
e" (2*-2*) dz.
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
( 210 )
z + ia z + io
Also
where
I“ i I < • -nz«dz
fj-
a*+ ^r i
and
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 ^ )-
C M - J ’ - S ? ! d(
d/(r)
dy = I (y) -
80] EXAMPLES 305
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)
(216,)
—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 )
( 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)
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,)
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
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
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
/ = 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!)*
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 =
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 <
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,)
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
/( z i. z 2) (1)
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
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.
Let us suppose that the series, constructed from the moduli of the
terms of the series
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
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
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:
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
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
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
| “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:
and construct the Taylor’s series for the function f(zv z2) in integral
powers of the differences (z2 — cx) and (z2 — c2)
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 :
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.
{*2},* = 2 { x U i x U
s= 1
and, in general
{ * ")« = , , 2 , M I U - P U - P U
JuJnt • •
°(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
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
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)
2 aDd 2 bmXm>
m =0 m=0
(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 ) £ -!.
Y = f 2(X)f1(X).
{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
(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
F = e x = l + -IT + — + . . .
enables us to determine the logarithm of the matrix in the form of
the power series
= (44)
|F-1|< n
The equation of the matrix
ex = Y (46)
332 FUNCTIONS OF SEVERAL VARIABLES AND MATRIX FUNCTIONS [88
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)
(49)
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
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
\< 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:
/, (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
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) ’
h{ z - , z v 21) = H z 1) + L ^ 1 f ' { z 1) -
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
i.e. simply the zero A0. The values of the function are determined in
the original domain (6 6 ) by the formula
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:
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!
S[±^T, ± (87)
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
_ 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
xn = + a n2X2 + • * • 1
" ” >
x = x0 + Ax0 + A2x0 + .. •
y + x% . (98)
= 2 { X U T U
$=1
whence
d (x r> ,A = ^ d ( X } ls
dt ^ dt {YU+
1 s=l
A (z r z ) = ^ F Z + A ^ z + z r - ^ . . (9 9 )
X X ! = /.
93] SYSTEMS OF LINEAR EQUATIONS W ITH CONSTANT COEFFICIENTS 351
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:
and formula (92) can be rewritten in the form of the following matrix:
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)
— = - I - M Z I - 1 = - X - 'A . (109)
dt dt
where, from (108) and (109)
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.
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
a»+ 2 a(m) i
m=1 j , ..........
«0 + 2 2 a h ............... j - X h • • • X Jm =
m“ 1jit •*'tjt»—1
= \ bh, , jmX Jl . . . X jm
m = l ht ■• - *jm~ 1
and show that this system has a regular solution in the circle
| z — z„ | < R, which satisfies all initial conditions:
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 )
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,
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:
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
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
wf = A j = A2w 2 . (24)
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
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
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)
A(z) = w'2 w l ~ w[ w 2 = w \ ^ =
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)
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
/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
l c" l < - ^ ( f n ~ 1 + f n ~ 2 + - - - + 1) = (4 7 )
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
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)
p(z) = — + a,i + a2 z + • • •
and
Jp(z) dz = logz°<>+ Cx + axz + —a2z2 + . . . ,
whence
e-!p( 2)d:= z -a, ^ (z ) ,
e - S p W d z J . = g—(l+m) p i ( z ) = + . . . + V f - + y0 + YlZ + . . .
( y - a + m ) T6 0) •
w" 2 '
—
22
w= 0
p(2) = d i ^ - + d 2— + . . . (54)
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)
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 (± )]
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
4 -* f t
will be
1™ -7T <7 (-f) = Jim z2 q(z).
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
-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 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:
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
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:
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
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!)
l + a-P; |) ,
( 6 4 3)
^ = (1 )^ (0 , 1+ P 1 + /* - a; - ) ■
101] THE HYPERGEOMETRIC SERIES 385
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:
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:
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.
— [Pl(2)Ml'] + ^ ( 2 ) ^ = 0. (74)
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)
— [2(2 - 1 ) w'] - k { k + l ) w = 0.
+ = <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
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 )
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
2 • 4... 2n
J (x2 — l)n dx = (— l)n 2 ,(2 n + l) ’
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).
x c k-^ [ z p + k (Z- i r * ] .
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
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)
( - l )kCk dk
( I - * 2) 2 T k(x) = k\ 2* [ ( i - * 2) ( 102 )
+ = (103)
l-3...(2Jfc-l)
** ^ = 2 •iT772k---
whence the constant term in the formula (101) can easily be deter
mined.
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
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.
= = (112>
*=1 * k=2 2
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:
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 = c0 + ^ - + ^ r + . . . (121)
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:
+ (4&i + ^ - + i ^ + . . . ) « = 0. (125)
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:
whence, since t > x, the factor ex~1lies between zero and unity, and we obtain:
1_________ 21____3I_
J / - 1e*-'dt X X 2 ' X 3 X4
(132)
X
j
lim f(z) — c0 — 2 = 0 ,
whence
cx = lim [/(z) — c0] z,
Z—►«
C„ = l t a [ / ( * ) - ( C„ + - i + . . . + ^ ) (133)
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 :
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.
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:
J
— ax z' v(z') — bx v(z') ea ' dz' = 0 .
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
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\ ’
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):
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
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
i.e., integrating
—r l*n + P |
I2"+P J* eZt ^ | < z _ J mi e _ r*
— CO
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:
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'
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:
and therefore
whence
Cn-n _ dnii(P + n)
424 LINEAB DIFFERENTIAL EQUATIONS [111
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
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"')
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)
(,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 .
x i ( ” J ) r ( » + 4 - + * )( -■ £ • )'• m
where
arg (t2 — 1) = 0 for r > 1. (185)
The corresponding solutions of the equation (177) will be:
(186)
k -0
(-1)* 2 \n+2k
fcl(n + fc) (n + fc —l)...(n + l).T(n + l) (t )
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 )
■ W <
1) 2dr = (191)
t 2" r ( n + 1) ’
5)
(r2 - 1) 2= e (1 — r 2) 2 (on the lower edge);
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
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
(192)
^ ( B + " r ) Bin(B + T ')w =
so that finally
2;i® i
J(^ -D *dr = -
r ( - i — » )r(» + i)
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
3
2™ji2
432 LINEAR DIFFERENTIAL EQUATIONS [112
fc=»V k I
or, from (192):
1 1 _ . I’m n\
w ~ (4-)* 0 4 ) I (* - j ) r (• + 4 - + *) ( g ;
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 )
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,
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:
(« M > - •
If we put r — sin <p we obtain:
2
1_____
J n (z) Hr) Jncos2n<Px
~2
_____ 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)
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
These solutions will have the following asymptotic representation as z -*■ + oo:
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.
After these transformations the equation (210) will have the form:
d* w . dw
dx2 ' dx + ( - - r + A - - £ r ) M’ = (213)
a(a - 1) + a - - J - = 0.
(c0# o ).
k=0 **
The quadratic equation for a will be
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
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
^ - ~ - — 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)
+ - - 2f I) (5 + n) {s + n — 1)X"-2 + . . . +
* + (y - *) ~ *y = °- (225)
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"
* .-3 -+ ^
w= e 2x*y^ (231)
and we obtain for yl the equation
a0 = — i; a, = s + 1; b0 = 0; — {s + 1).
442 LINEAR DIFFERENTIAL EQUATIONS [116
P — — — (s + 1) +
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:
z(z- ]) + [- y + (i + a + £ ) z] - J - + a P y = °-
d*u>
dz2 + i-^ + 4-+ \w = 0. (236)
d 2u
+ o - 2”» - t + ( u = 0.
dz2"
444 LINEAR DIFFERENTIAL EQUATIONS [1 1 7
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/.
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
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
»?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
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
z (259)
+ 2 b, zSy = ° - (260)
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).
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
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:
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]
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.
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 .
°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:
°I2> °22 Q* • ■ •, o n2
(271)
/ ( B) =
/(?) = 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,
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
* - ^ - = * , ( £ , + 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).
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
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)
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
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:
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,
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
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
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
Y(b; x)- 1 + 2 2
»=lfl....jv
u. ' . Cf jii Lb (aj\ ( ;t-» (289)
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
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
(293)
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
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
ifrir. <»>
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)
(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
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:
Substituting the series (297) and (302) in equation (301) and comparing
the coefficients of the product C7, . . . Ujy we have:
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
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:
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:
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.
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:
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.
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 )
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 ) .
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 :
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
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
— = 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
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
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
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
» 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 >
( 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) _ (,_ !) ]
and p remains arbitrary. Summarising all that has been said above we
obtain the following formulae for the determination of the coefficients a:
(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
M*Pl (x ) = - M Pl (x),
(332)
Mp1...pv (x ) = — 2/ Mpi-'-Pp ^Pp+i' -Pv
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:
holds for A< v —1. We will now show th at it is also valid when A= v. We have
from (332):
+ M ti...Py, ( * ) - g j M P ^ . . - P V ( * ) ]
- * 2 M p x . . . Pp. W M p ,p ( * ) xPv.
p-o *
But according to (332):
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:
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) =
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
* "M aZ +l.....(342)
A-0
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:
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.
(345)
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)
(352)
^ = 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:
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:
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:
I^l< j2^-dx< 2 a .A ,
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
, ,(i) ,(s)
gill __ e- au*-0tt,oex | eaux+ft*iogx I 1'* i l,k
■J- (4+ 4-+ ^ )* “ <*>■
where eik and e*/ 0 as x -*• co. Integrating by parts we obtain from this:
x ( « + i ? + i»K
[ X X2 X2 )
I I < 2 a 6 J - ^ r da:-
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
Using the series (359) and the formulae (368), (370) and (371) we obtain:
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
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)
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
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.
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
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
<»>
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)
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
(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
(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:
= (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):
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
(']'[/>„ (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
/ - 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 ~
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).
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
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 ’
± = a 0 (d) + a 1 ( d ) r + a 2 ( d ) r * + . . . , (28)
■A 9 1
Y1 — 2r cos 6 + r2
= c 0 + ciP i (cos e ) r + c2p 2 (cos 0 ) r 2 + ...
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:
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:
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)
= 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:
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
a0 = l ; = (k = 1, 2, .. .). (45)
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
i = J 'P ^ c o s y ) — (r< R ),
k=0 n
gO th&t
and
= n R n~ 1 Y n (0, V ) .
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
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
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 +
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:
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 ,
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
from whence it follows that the integral (67) tends to zero as n -* oo.
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 °'
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
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
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)
m ? ) = 2 Y n{0,9)- (83)
n=1
= 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 -
[ 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 - +.
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
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
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 + '
U (M) = — +
*7(Jf)=JJ-^-ds, (93)
S r
- = v p ^ c o s? )-^ . (94)
n=l
138] TH E POTENTIAL OF A SPHERICAL SHELL 623
f(9,<p) = 2 T n i ° ’ V) , (96)
n=0
then, as we know
U ( M 0) = 4 n R 2 ^ - T Y n ( 6 , <p) , (99)
n=0
n=l
(«.*’>• (ion
s ' n=0 1
(«• *■>■
0
( 3U (iWo) ^ | ( d U { M q) \ . 1 v /fl «\
[ oi Je + 1— § r ~ ) i = - An2n—
0
-w + t 7 "{6’v)’
(•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
Sr
These formulae are also valid for bodies other than spheres.
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)
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 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 )
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:
0, 0, 0, . e"°
140] SPHERICAL FUNCTIONS AND T H E LINEAR REPRESENTATION OF ROTATING GROUPS 529
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
(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
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.
(l - x2 ) ^ . - 2 x ^ : + n ( n + l ) u = 0
dx (118)
= 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
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
F n { x) = C F [ n + 1, — n, 1 ; 1 ~-*-j • (122)
P n ( x) = F { n + 1, - n , 1; - ^ ) . (123)
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.
«(*) = - P „ ( z ) l o g - |± ! -v {x ). (126)
U “ **) - £ r ~ 2 a ; + n in + ! ) v =
(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,
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) ,
«» W - T ■
p- w ~ i (a - r n f - t + ! ) f . * . <*> • <131>
%(*) = 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)
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)
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
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
J p (2 ) = 2PT(p + 1) f
1~ I I (P + 1) ( t ) + 2 ! ( p + l ) ( p + 2) ( t ) “ •••]
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.
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
i.e.
■/_„ (z) = (— 1)nJ n (2) (re being an integer) (8 )
( - l ) * 2 (p + fc) Zlp+lk-l
= k2= 0 *!r(p + * + l) 2P+2fc ’
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)
( 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 '
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 -'
i.e.
■m * ) = y ^ - sinz- (is)
2 r
^ w - t - i r ] ( »)
J _ i ( 2) COS2’ (20)
2
< 2 1 >
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>
+ (* 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
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
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]:
(30)
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
(*! + *) \ 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
But we have:
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:
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
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
= — 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
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
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----- •
we have
2 J n (a + b)tn = 2 Jk( a) t k- 2
n*» —— k = —<~ ft-*—*
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.:
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
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
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
>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
then for any integer n and q> 0 the following formula applies:
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:
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
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
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>
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
and similarly, using formula (52) we can obtain the asymptotic repre
sentation for the Neuman function when z > 0
cos (z + - - ) = ClC0S [z - - -) +
- 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)
(60)
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 < " .+ * > '& )* • («)
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:
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-
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
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 ) +
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
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:
-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)
Using the first of the formulae (61) we can express K p(z) in terms of
I +P(z). In fact this formula gives:
and finally:
K Az\ = L a £=B. (77)
p (iz)
' '= 2 Q,Tl T
sin pl7
nT
.
562 SPECIAL FUNCTIONS [160
(81)
where V is a oontour which encircles the negative part of the imaginary axis,
we obtain
_ lU7.1
i'*
k=0 r
(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]):
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
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
£ =* 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
«- (rd a n j (93)
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
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
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
. 7 £ cosh O'!
o'! 5 sinh a „ 2 5A7 cosh a
> £ 8,nhal 1- 2 ^ h ^ J > 8 2 £ £ ^ 3z sinh a ’
/ ( - 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
sinh a sinha
sinh (a — e) . , , sinh a
sm h a — e cosh a + e2 ----»—
sinh a ^
____. 100^
sinh a — e cosh a 25 ’
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
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:
4.4
+ Yz sinh a
+ ■ Jsinh
— ),
2a)
(107)
570 SPECIAL FUNCTIONS [152
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
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
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)
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)
— 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:
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
1 ( Z 3 d 6I<! (t )
(127)
60 IT,) dt*
in which
T®
. , 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
- ^ [ 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
<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
J[
kz—lkQS\n<p
e - k z Jo (kg) d i = dg>,
) 2n — z — iq sin <p k = 0
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
U = e~iat V (x , y, z) . (134)
sin n<p
+ 0 ( e - 1)]-
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 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:
-----P n (oos9),
yr
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)
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
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
|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:
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:
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.
- £ 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,
^ - 2 a s ^ + ( A - a ) W= 0. (4)
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
—
a
— 1 = 2n,
i.e.
Xn = {2n+ l ) a. (7)
586 SPECIAL FUNCTIONS [156
^ 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 |:
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, 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,
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.
'f
e - ™ . 2 < = 2 - k H ' ^ ) tn
n=- 0
or
n=0 Ti—0
590 SPECIAL FUNCTIONS [158
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.
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‘
f ( £ ) = - ( £ + iy)2; /'(£) = £ + «?
or
9(£>y) = y ( P - v 2Yi 2p(£,y) = £y-
and therefore equation (25) with the new coordinates will, in this
case, have the following form:
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:
_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
= (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:
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
where
A„ = ^ ± -L + n. (37)
— 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
—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)
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 '
■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
* + ( I “ x) “S r + ny»= °-
which can be transformed by the formula x = f* into the following equation:
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:
Xs
< y>l„ (u) du f u* dti =
6 IT
162J THE ASYMPTOTIC EXPRESSION FOB HERMITIAN POLYNOMIALS 599
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
{ 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 — —
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:
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 )
or, using the formulae cos (nn + tp) = ( —I)'1cos <p and sin (nn + <p) =
= ( —1) "sin p, wo have:
whence
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
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 ■
I K ln\ <
K4n+ 1 ’
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:
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
Ax, (1)
x = Xj -f- — . (3)
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 )
+ = xm yP (x) + C.
164] TH E TRANSFORMATION OF ELLIPTIC INTEGRALS INTO NORMAL FORM 607
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
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
x = a ± / a 2 + pa + gtan 2 y . (14)
(15)
]fa2 + pa + q
a2 + pa + q = (a + + (g - -^p2j .
a2 + pa + q > (a + |- ) 2.
610 SPECIAL FUNCTIONS [165
__________ 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
+ 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:
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
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
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
d?)
T = 2 = 2 ]f j F W .
YT k2sin2<p
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 )
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
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
u=
f dx
]/P(x)
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
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
2K B Tk ’*'
F i g . 80
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
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.
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
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
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
_________ 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
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):
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'
ck = - e 2 ,
and finally
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
Up
+ J C (a) da = 2m.
U . + <0,
This, from (46), gives the required relationship between cok and rik:
This, from (41), gives the following expansion for )?(«) near the
origin
1
P(u ) = ~& + 2 1( n + l ) un TTl\,
2 TT\%
’
(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
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:
6k 6^ 63 ~ 0, “1“ ^ 2 ^ 3 + ^3 ^ 1 — £7 2 ’ = ^ 3 * (6 1 )
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
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
{(>(u) — e1 = e’’lU » ( - * )
. f i 1-) «* (“)
or
P («) - ei =
(^)e(u)
<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
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
£ = 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
?>(«) = ^ a ne ~ n ■ (7 4 )
n - — oo
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:
^ = »+ ^ 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,
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)
+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
an+1 = - h ^ a n = - + an ,
which can also be written as follows:
( -
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):
#i [ v ) = i +
2 (- 1 )nh (n~ TJ z 2n- 1, (92)
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
We have similarly
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
= (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
= r lz h ^ - v 2z2n- 2
n ■=>— oo
$ A ± r + Ar - v) = h - T z A»y», (100)
n = —oo
and similarly
-&1 = ATiz +
j? ( - 1 )nhniz2n. (1 0 1 )
n = —oo
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)
= 1 — 2h + 2h* — 2 A9 + .. .
and, bearing in mind the expression for the functions a(u) and ak(u)
in terms of theta-functions we obtain
( 1 M >
(» + 4 “) = ^ ^ + 4") = ~
#3 (» + 4 ") = ^ + 4 ") = w
(v §3 ■
= i h ~ T z-1 +
jg ( - 1 )nhn‘z2n,
n = — <=>
or from ( 1 0 2 )
#i (w + - y ) = im&t (v) ,
where
= 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
r +, - y1 +, -Ty = 7l+7lT,
. ,
177j THE PROPERTIES OP THETA-FUNOTIONS 647
* = ( n - - r ) + (n' - - r ) T’
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
u = \-r- ** =■ (H5)
J / 4 (x — e,) (x — e2) (x — e3)
( - £ )
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 ’
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
= V®! — e3 = —^ = cu
2 (11?)
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
(119)
*i #2 + #3 ^ #4 '
*» = 4 ( k = 2,3,4)
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
0 J = CVfl 3^4-
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>)
V P (“ ) — (J(W) ’
these new functions are connected with the Weierstrass function P(u)
by the following three relationships:
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:
* 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
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)
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
— |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:
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:
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.
we obtain
' hA — — 2 (sn
^ ' ' sn* ( h) ’
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:
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
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
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
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
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>
,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
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:
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.
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
2 pf dt
=
* J /4 (t - eL) (« — e2) (t — e3)
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,)
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
^ + [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
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:
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)
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
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
sin
0- = r sin
. - 5a-
£t it
s in - L = s i n ^ - s n ( ] f - |- ( ) = ksn ( |f - f t ) ,
(176)
COS - 1 — k2sn 2 ,) = d n ( l f Z t ) ,
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
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
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
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)
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]).
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
< 1 9 >
k= 1 *
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 ,
<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).
/* (* ) = g * ( * ) (z -
~ — k •
~
akYK ( 20 )
We obvionsly have the identity
k= 1
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):
{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
(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.
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
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
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 .
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.
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
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
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
eZ = ei — e3 + ^4
e3 ~ ei — e3 + e5
ei = ei + e2 + e4 — e5
eS = e3
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):
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
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
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