Anda di halaman 1dari 378

SPECIAL

FUNCTIONS
Earl D. Rainville, Ph.D.
PROFESSOR OF MATHEMATICS
IN THE UNIVERSITY OF MICHIGAN
THE MACMillAN COMPANY
New York
SPECIAL FUNCTIONS
.
.s 0
TilE CmlPANY
:--;1-.\\' YORK. CHICAGO
DALl.AS ATl.A::-.iTA SAS FRA,SnSCO

IN CANADA
LTD.
GALT,
Earl D. Rainvill{' 1960
All rights reservr,d-no part of this book may he ill ally
form without in writing from the except by a
reviewer who wishes to quote brief passages in COllnection with a
review written for inelusion in magazine or newspaper.
First Printing
Library of Conaress cataloa carr! number: 60-5115
The .'\facmillan Company, :\ew York
Brett-.'\lacmillan Ltd., Galt, Ontario
Printed in the United States of America
Preface
I have attempted to 'write this book in such a way that it can be
read not only by professional mathematicians, physicists, engineers,
and chemists, but also by \yell-trained graduate students in those
and closely allied fields. Even the research worker in special functions
may notice, however, some results or techniques with which he is not
already familiar.
Many of the standard concepts and methods which are useful in
the detailed study of special functions are included. The reader will
also find here other tools, such as the Sheffer classification of poly-
nomial sets and Sister Celine's technique for obtaining recurrence
relations, which deserve to become more widely used.
Those who know me will not be surprised to find a certain empha-
sis on generating functions and their w,efulness. That functions of
hypergeometric character pervade the bulk of the book is but n
reflection of their frequent occurrence in the subject itself.
More than fifty special functions appear in this work, some of
them treated extensively, others barely mentioned. There are dozens
of topics, numerous methods, and hundreds of specinl funrtiolls
\vhich could well have been included but which have been omitted.
The temptation to approach the subject on the encyclopedic level
intended by the late Harry Bateman was great. To me it seems that
such an approach \vould have resulted in less, rather than more,
usefulness; the work would never have reached the stage of publi-
cation.
The short bibliography at the end of the book should give the
reader ample material \'lith which to start on a more thorough study
of the field.
This book is based upon the lectures on Special Functions which
I have been giving at The University of Michigan since 1946. The
enthusiastic reception accorded the course here has encouraged me
to present the material in a form which may facilitate the teaching
of similar courses elsewhere.
v
VI PREFACE
I wish to acknowledge the assistance given me in the way of both
corrections and comments OIl the manuscript by Professor Phillip E.
Bedient of Franklin and Marshall College, Professor Jack R.
Britton of The University of Colorado, and Professor Ralph L.
Shively of 'Western Reserve University. I v,as also aided and
encouraged by the late Fred Brafman who was Associatc Professor
of IVlathematics at Thc Lnivcrsity of Oklahoma at the time of his
death. Professor Drafman read the first ten chapters criti('ally and
discussed some of the latcr material with me. Scvcral students now
taking the course have been helpful in catching crrors and pointing
out rough spots in the presentation.
Professor Bedient has also aided me by an indepcndent reading
of the proof sheets.
EAHL D. RAlr\VILLE
Ann Arbor, Michigan
Contents
Chapter 1: INFINITE PRODUCTS
1. Introduction
2. Definition of an infinite product
3. A necessary condition for C'ol1\'crgcnce
4. The associated "cries of logarithms
5. Absolute cOI1\'ergence
6. Uniform convergence
Chapter 2: THE GAMMA AND BETA FUNCTIONS
7. The Euler or l'.Iascheroni constan t I
8. The Gamma function
9, A series for r'(z)/r(z)
10, Evaluation of r(1) and r'(l)
11. The Euler product for r(z)
12. The difference equation r(z + 1) = zr(z)
13. The order symbols 0 and 0
14. Evaluation of certain infinite products
15. Euler's integral for r(z)
16. The Beta function
17. The \'alue of r(z)r(l - z)
18. The factorial function
19. Legendre's duplication formula
20. Gauss' multiplication theorem
21. A summation formula due to Euler
22. The behavior of log r(z) for large ! z I
Chapter 3: ASYMPTOTIC SERIES
23. Definition of an asymptotic expansion
24. Asymptotic expansions about infinity
25. Algebraic properties
26. Term-by-term integration
27. Uniqueness
28. 'Watson's lemma
Chapter 4: THE HYPERGEOMETRIC FUNCTION
29. The function F(a, b; c; z)
30. A simple integral form
Page
1
1
2
2
3
5
8
9
10
10
11
12
12
13
Li
1."
Hl
2:2
28
24
2G
29
33
36
3"
39
40
41
4.')
47
\"11
Vlll CONTENTS
31. F(a, b; C; 1) as a function of the parameters
32. Evaluation of F(a, b; c; 1)
33. The contiguous function relations
34. The hypergeometric differential equation
35. Logarithmic solutions of the hypergeometric equation
36. F(a, b; C; z) as a function of its parameters
37. Elementary series manipUlations
38. Simple transformations
39. Relation between functions of z and 1 - 2
40. A quadratic transformation
41. Other quadratic transformations
42. A theorem due to Kummer
43. Additional properties
Chapter 5: GENERALIZED HYPERGEOMETRIC FUNCTIONS
44. The function pF q
45. The exponential and binomial functions
46. A differential equation
47. Other solutions of the differential equation
48. The contiguous function relations
49. A simple integral
50. The pF q with unit argument
51. Saalschiitz' theorem
52. Whipple's theorem
53. Dixon's theorem
54. Contour integrals of Barnes' type
55. The Barnes integrals and the function pF q
56. A useful integral
Chapter 6: BESSEL FUNCTIONS
57. Remarks
58. Definition of J n(z)
59. Bessel's differential equation
60. Differential recurrence relations
61. A pure recurrence relation
62. A generating function
63. Bessel's integral
64. Index half an odd integer
65. Modified Bessel functions
66. Neumann polynomials
67. Neumann series
Chapter 7: THE CONFLUENT HYPERGEOMETRIC FUNCTION
68. Basic properties of the )F)
69. Kummer's first formula
70. Kummer's second formula
Page
48
48
50
53
54
55
56
58
fi1
63
65
68
68
73
74
74
76
110
85
85
86
88
92
94
98
102
108
108
109
110
111
112
114
114
116
116
119
123
124
125
CONTENTS
Chapter 8: GENERATING FUNCTIONS
71. The generating function concept
72. Generating functions of the form G(2xt - t2)
73. Sets generated by etif;(xt)
74. The generating functions A(t) exp[ -xtl(l - t) 1
75. Another class of generating functions
76. Boas and Buck generating functions
77. An extension
Chapter 9: ORTHOGONAL POLYNOMIALS
78. Simple sets of polynomials
79. Orthogonality
80. An equivalent condition for ortllOgonality
81. Zeros of orthogonal polynorniab
82. Expansion of polynomials
83. The three-term recurrence relation
84. The Christoffel-Darboux formula
85. Normalization; Bessel's inequality
Chapter 10: LEGENDRE POLYNOMIALS
86. A generating function
87. Differential recurrence relations
88. The pure recurrence relation
89. Legendre's differential equation
90. The Rodrigues formula
91. Bateman's generating function
92. Additional generating functions
93. Hypergeometric forms of P n(X)
94. Brafman's generating functions
95. Special properties of Pn(x)
96. More generating functions
97. Laplace's first integral form
98. Some bounds on Pn(x)
99. Orthogonality
100. An expansion theorem
101. Expansion of xn
102. Expansion of analytic functions
Chapter 11: HERMITE POLYNOMIALS
103. Definition of Iln(x)
104. Recurrence relations
105. The Rodrigues formula
106. Other generating functions
107. Integrals
108. The Hermite polynomial as a 2Fo
IX
Page
129
131
132
13.5
13i
1-10
143
14i
14i
14S
149
15t)
1.51
153
155
15i
15S
15U
lGO
161
162
163
16;')
16i
16S
169
IiI
172
1i3
1i6
1i9
lSI
lSi
ISS
lSU
190
190
191
x
109. Orthogonality
110. Expansion of polynomials
111. More generating functions
Chapter 12: LAGUERRE POLYNOMIALS
112. The polynomial Ln(a)(x)
113. Generating function"
114. Recurrence relations
115. The Rodrigues formula
lUi. The differential equation
117. Orthogonality
118. Expansion of polynomials
119. Special properties
120. Other generating functions
121. The simple Laguerre polynomials
Chapter 13: THE SHEFFER CLASSIFICATION
AND RELATED TOPICS
122. Differential operators and polynomial sets
123. Sheffer's A-type classification
124. Polynomials of Sheffer A-type zero
125. An extension of Sheffer's classification
126. Polynomials of o--type zero
CHAPTER 14: PURE RECURRENCE RELATIONS
127. Sister Celine's technique
128. A mild extension
Chapter 15: SYMBOLIC RELATIONS
129. Notation
130. Symbolic relations among classical polynomials
131. Polynomials of symbolic form Ln{ y(x)
Chapter 16: JACOBI POLYNOMIALS
132. The Jacobi polynomials
133. Bateman's generating function
134. The Rodrigues formula
135. Orthogonality
136. Differential recurrence relations
137. The pure recurrence relation
138. Mixed relations
139. Appell's functions of two variables
140. An elementary generating function
141. Brafman's generating functions
142. Expansion in series of polynomials
CONTENTS
Page
191
193
196
200
201
202
203
204
204
206
209
211
213
218
221
222
226
228
233
240
246
247
249
254
256
257
258
261
263
263
265
269
271
272
CONTENTS
Chapter 17: ULTRASPHERICAL AND
GEGENBAUER POLYNOMIALS
143. Definitions
144. The Gegenbauer polynomials
145. The ultraspherical polynomials
Chapter 18: OTHER POLYNOMIAL SETS
146. Bateman's Zn(X)
147. Rice's Hn(t, p, v)
148. Bateman's Fn(z)
149. Sister Celine's polynomials
150. Bessel polynomials
151. Bedient's polynomials
152. Shively's pseudo-Laguerre and other polynomials
153. Bernoulli polynomials
154. Euler polynomials
155. Tchebicheff polynomials
Chapter 19: ELLIPTIC FUNCTIONS
156. Doubly periodic functions
157. Elliptic functions
158. Elementary properties
159. Order of an elliptic function
160. The Weierstrass function P(z)
161. Other elliptic functions
162. A differential equation for P(z)
163. Connection with elliptic integrals
Chapter 20: THETA FUNCTIONS
164. Definitions
165. Elementary properties
166. The basic property table
167. Location of zeros
168. Relations among squares of theta functions
169. Pseudo addition theorems
170. Relation to the heat equation
171. The relation 8
1
' = 8
2
8
3
114
172. Infinite products
173. The value of G
Chapter 21: JACOBIAN ELLIPTIC FUNCTIONS
174. A differential equation involving theta functions
175. The function sn (u)
176. The functions en (u) and dn (u)
177. Relations involving squares
XI
Page
276
277
283
285
287
289
290
293
297
298
299
300
301
305
306
306
308
309
311
311
313
314
315
316
319
322
325
328
329
332
334
339
342
343
344
xii
178. Relations involving derivatives
179. Addition theorems
Bibliography
Index
CONTENTS
Pa!!:e
3,15
3.fi
3,19
359
SPECIAL FUNCTIONS
CHAPTER 1
Infinite
Products
1. Introduction. Two topics, infinite products and asymptotic
series, which are seldom included in standard courses are treated to
some extent in short preliminary chapters.
The variables and parameters encountered are to be considered
complex except \vhere it is specifically stipulated that they are real.
Exercises are included not only to present the reader with an
opportunity to increase his skill but also to make a\'ailable a few
results for which there seemed to be insufficient space in the text.
A short bibliography is included at the end of the hook. All
references are given in a form such as Fasenmyer [2J, meaning item
number two under the listing of references to the work of ~ i s t e r 1\1.
Celine Fasenmyer, or Brafman [1 ;944J, meaning page 944 of item
number one under the listing of references to the work of Fred Braf-
man. In general, specific reference to material a century or more
old is omitted. The work of the giants in the field, Euler, Gauss,
Legendre, etc., is easily located either in standard treatises or in the
collected \yorks of the pertinent mathematician.
2. Definition of an infinite product. The elementary theory of
infinite products closely parallels that of infinite series. Gi\'en a
sequence ak defined for all positive integral k, consider the finite
product
n
(1)
2 INFINITE PRODUCTS [Ch.l
If Lim P n exists and is equal to P ~ 0, we say that the infinite
product
(2)
converges to the value P. If at least one of the factors of the product
(2) is zero, if only a finite number of the factors of (2) are zero, and
if the infinite product ,,-ith the zero factors deleted conyerges to a
value P ~ 0, we say that the infinite produet converges to zero.
If the infinite product is not conYergent, it is said to be divergent.
If that divergence is due not to the failure of Lim P n to exist but
n+
oo
to the fact that the limit is zero, the product is said to diverge to zero.
We make no attempt to treat produets with an infinity of zero
factors.
The peeuliar role which zero plays in multiplication is the reason
for the slight differenee between the definition of conyergence of an
infinite produet and the analogous definition of conyergenee of an
infinite series.
3. A necessary condition for convergence. The general term
of a convergent infinite series must approach zero as the index of
summation approaehes infinity. A similar result will now be ob-
tained for infinite products.
m
THEOREM 1. If II (1 + an) converges,
n=l
Proof: If the product COll\-erges to P ~ 0,
P
1 = P
Lim II (1 + ak)
_ ~ " , _ k ~ ~ ___ _
n-l
Lim II (1 + ak)
n+oc k ~ 1
Lim (1 + a,,).
n+
oo
Renee Lim an = 0, as desired. If the product conyerges to zero,
remove the zero factors and repeat the argument.
4. The associated series of logarithms. Any product without
zero factors has associated with it the series of principal \"alues of
the logarithms of the separate factors in the following sense.
5] ABSOLUTE CONVERGENCE
3
= co
THEOREM 2. If no an = - 1, II (1 + an) and L Log (1 + an)
n=l n=l
converge or diverge together.
Proof: Let the partial product and partial sum be indicated as
follows:
n
Sn = L Log(! + ak).
k=!
Then * exp Sn = P n. 'Ve know from the theory of complex variables
that Lim exp Sn = exp Lim Sn. Therefore P
n
approaches a limit
n+ro n+oo
if and only if Sn approaches a limit, and P n cannot approach zero
because the exponential function cannot take on the value zero.
co
5. Absolute convergence. Assume that the product II (1 + an)
n-l
has had its zero factors, if any, deleted. We define absolute con-
vergence of the product by utilizing the associated series of logarithms.
co
The product II (1 + an), with zero factors deleted, is said to be
n=l 00
absolutely convergent if and only if the series L Log (1 + an) IS
absolutely convergent. n-l
=
THEOREM 3. The product II (1 + an), with zero factors deleted, is
11==1
'n
absolutely convergent if and only 1j L On is absolutely convergent.
Proof: First throw out any an's which are zero; they contribute
only unit factors in the product and zero terms in the sum and thus
have no bearing on conyergence.
We know that if either the series or the product in the theorem
conyerges, Lim an = O. Let us then consider n large enough, n > no,
n+oo
so that I an I < ! for all n > no. 'Ve may now write
(1)
Log (1 + an) _ f (-I)kan
k
--0-;-- - k=O k+---Y-'
from which it follows that
I
Log (I + On) - 11 s f ~ < f -!- = !.
an - k_l k+ 1 k=12k+l 2
We make frequent use of the common notation exp 11 = e
u

4 INFINITE PRODUCTS (Ch. 1
Thus we have
from which
I
Log (1 + an) I <
an 2
and
By the comparison test it follows that the absolute convergence of
m
either of L Log (1 + an) or L an implies the absolute convergcnce
n=!
of the other. We thcn use the definition of absolute comcrgencc
of the product to complete the proof of Theorcm 3.
Because of Theorem 2 it follows at once that an infinite product
which is absolutely convergent is also convergent.
EXAMPLE (a): Show that the following product converges and
find its value:
IT [1 + 1 ]
n-l (n + l)(n + 3)
The series of positive numbers
m 1
L
(n + l)(n + 3)
is known to be convergent. It can easily be tested by the poly-
nomial test or by comparison with the series Hence our
n
product is absolutely convergent by Theorem 3.
The partial products are often useful in evaluating an infinite
product. When the following method is employed, there is no need
for the separate testing for convergence made in the preceding
paragraph. Consider the partial products
n [ 1 ] n (k + 2r
P
n
= g 1 + (k + l)(k + 3) = g (I.; + l)(k +- 3)
n+2 3
[234 .. (n + 1) J[4 .. 5 . 6 .. (n + 3) J = --2- . n + 3'
At once Lim P n =!, from which we conclude both that the in-
n+
m
finite product converges and that its value is
6] UNIFORM CONVERGENCE
EXAMPLE (b):
Lim
n+oo
exists.
Show that if z is not a negatiyc integer,
(n - I)! n'
(z + l)(z + 2)(z + 3) (z + n - 1)
We shall form an infinite product for which the expression
(n - 1)! nz
P
n
=
(z + l)(z + 2)(z + 3) .. (z + 1/ .- 1)
5
is a partial product, proye that thc infinite product COll\"erges, and
thus conclude that Lim P n exists.
Write
n! (n + 1)'
P n+! = (z + 1) (z + 2) .. (z + n)
n! 2z 3' 4' (n+1)-
(z + 1) (z + 2) ... (z + n) '}z' 2 z 3 z ..
= IT [_k .' = IT [(1 + +
z + f, f, k_1 J.. k
Consider no,,", the product*
(2)
Since
+ 1J
. (1 + zi3)-I(l + i3)z- 1 . 1 + /J)' - 1 - zi3
= LIm = LIm
13+0 13
2
1l+0 13"
_ I' z[(l + (3)z-1 - 1] _ I' z(z - 1)(1 + 13),-2 _ 1 ( 1)
- ,Inl 2 - 2 - zZ z - ,
1l+0 13 13+0
we conclude with the aid of the comparison test and the COll\"ergence
of f: 1:.2 that the product (2) conyerges. Therefore Lim P n exists.
n=1 n n+ro
6. Uniform convergence. Let the factors in the product
00
II [1 + an(z)] be dependent upon a complex variable z. Let R
n=1
*\Ye shall find in Chapter 2 that this product has the value zr(z).
6
INFINITE PRODUCTS [Ch. 1
be a closed region in the z-plane. If the product converges in such
a way that, given any EO > 0, there exists an no independent of z for
all z in R such that
.."
for all positive integral p, we say that the product II [1 + an(z)] is
un1formly convergent in the region R-
Again the convergence properties parallel those of infinite series.
We need a Weierstrass M-test.
CD
THEOREM 4. If there exist positive constants M n such that 1: M n
n=1
is convergent and I an(z) I < M n for all z in the closed region R, the
CD
product II [1 + an(z)] 1S uniformly convergent in R.
n ~ l
CD CD
Proof: Since 1: M n is convergent and M n > 0, II (1 + M n) is
,,=1 71=1
convergent and Lim II (1 + M k) exists. Therefore, given any
71+00 k=l
EO > 0, there exists an 110 such that
nQ+p nQ
II (1 + M k) - II (1 + M k) < E
k ~ ! k ~ I
for all posithe integers p. For all z in R, each ak(z) is such that
I ak(z) I < J1,1 k. Hence
I nfi [1 + ak(z)] - g [1 + ak(z)] I
= I g [1 + (hcz)l/ . 1 k:ft [1 + ak(Z)] - 1\
no+p no
< II (1 + M k) - II (1 + M k) < EO,
k=! k=l
which was to be proved.
6] UNIFORM CONVERGENCE
EXERCISES
7
1. Show that the following product converges, and find its value:
m [ 6]
g 1 + (n+l)(2n+9) .
m ( 1) 1
2. Show that II 1 - - =-.
n_' n
2
2
3. Show that g (1 - diwrges to zero.
m
4. Investigate the product II (1 + l") in I z I < 1.
n=O
5. Show that g diverges.
6. Show that IT exp( _l) diverges to zero.
n=l n
21
Am. S
1
Ans. Abs. conv. to -1--'
-z
m ( z')
7. Test II 1 - 2 . Ans. Abs. conv. for all finite z.
n=l n
m [ (-l)n+l]
8. Show that II 1 + --- converges to unity.
>=1 n
9. Test for convergence: IT (1 - for real p ;;c o.
n-2.. n
P
Ans. Conv. for p > l' div. for p 1.
10. Show that IT sin_/(z/n) is absolutely convergent for all finite z with the usual
Z n
convention at z = O. Hint: Show first that
Lim n
2
[sinJ.z/rQ - 1J =
n+'" z/n
11. Show that if c is not a negative integer,
IT [(1 - _z -) exp(!)J
n-l C + n n
is absolutely convergent for all finite z. Hint: Show first that
CHAPTER 2
The Gamma
and Beta Functions
7. The Euler or Mascheroni constant 'Y. At times we need to
use the constant 'Y, defined by
(1) 'Y = Lim (H n - Log n),
n+co
in which, as usual,
(2)
n 1
Hn = 2:-.
hI k
We shall proye that 'Y exists and that 0 ~ 'Y < 1. Actually
'Y = 0.5772, approximately.
Let An = H n - Log n. Then the An form a decreasing sequence
because
An+l - An = Hn+l - Hn - Log (n + 1) + Logn
= n 1 + Log n - ~ l = n ~ 1 + Log (
1
- n J
co 1
= - ~ (k + l)(n + 1) HI < O.
Furthermore, since lit decreases steadily as t increases,
(3)
1 fk dt 1
Tc < k-I t < k - l'
k fE; 2.
We sum the inequalities (3) from k = 2 to k = n and thus obtain
8
81 THE GAMMA FUNCTION
J
2 dt JOdt fn dt
H n - 1 < - + - + ... + - < H n-h
1 t 2 t .. -1 t
or
H n - 1 < Log n < H ,,-1,
from which it follows that
or
-1 < - H n + Log n <
1
1 > An > -'
n
9
Thus we see that the A n decrease steadily, are all positive, and
are less than unity. It follows that 'Y exists and is non-negative
and less than unity.
8. The Gamma function. We follow Weierstrass III defining
the function r(z) by
(1) _1 = ze'Y< IT [(1 +
r(z) n=1 n n
in which 'Y is the Euler constant of Section 7. The product in (1)
is absolutely convergent for all finite z as was seen in Ex. 11, page 7,
the special case c = 0 and z replaced by (-z). That the product
is also uniformly convergent in any closed region in the z-plane is
easily shown by employing the associated series of logarithms.
We shall see in Section 15 that the function r(z) defined by (1)
is identical with that defined by Euler's integral; that is,
r(z) = i""e-tt<-I dt, Re(z) > O.
The right member of (1) is analytic for all finite z. Its only zeros
are simple ones at z = 0 and at each negative integer. We may
therefore conclude that
(a) r(z) is analytic except at z = nonpositive integers and z = co;
(b) r(z) has a simple pole at z = each nonpositive integer, z = 0,
-1, -2, -3", .;
(c) r(z) has an essential singularity at z = co, a point of con-
densation of poles;
(d) r(z) is never zero [because l/r(z) has no poles].
10
THE GAMMA AND BETA FUNCTIONS [Ch.2
9. A series for r/(z)/r(z). By taking logarithms of each mem-
ber of equation (1) of Section 8, we obtain
log r(z) = - Log z - 'YZ - i: [Log(I + -
,,-1 n n
Term-by-term differentiation of the members of the foregoing
equation yields
or
(1)
r/(z)
r(z;
- 'Y - f ( __ 1 -
z Z + n n
r/(z) 1 co z
--- = - 'Y - - + L: ----
r(z) z n(z + n)'
the series on the right being absolutely and uniformly convergent
in any closed region excluding the singular points of r(z), a result
easily deduced by using the Weierstrass M-test and the convergence
10. Evaluation of r(l) and r/(I). In the Weierstrass definition
of r(z) put z = 1 to get
_1 = II [(1 + exp(-
r(l) n n
n [k + 1 ( 1)J
= LIm IT --exp -
n+co h1 k k
= Lim (n + 1) exp( -Hn)
n+
w
= Lim (n + 1) exp( - 'Y - Log n - En),
n+
w
in which En -+ 0 as n -+ co. It follO\vs that
1 . n + 1
-- = LIm = 1
rei) n+w n '
so that r(1) = 1.
We know from the series for r'(z)/r(z) obtained in Section 9 that
r'(l) 00 1
reI:) = - 'Y - 1 + nen + 1)'
so that
11] THE EULER PRODUCT FOR r(z) 11
ro (1 1)
r'(l) = -'Y - 1 + L - - ---
n n + 1
= - 'Y - 1 + Lim (1 -
n+ro n + 1
since the series involved telescopes. Thus we find tha t r' (1) = - 'Y.
11. The Euler product for r(z). From the Weierstrass product
definition of r(z) we obtain
zr(z) = exp( - ,),z) ,
11 [(
1
+ ex
p
( ]
so that
(1) zr(z) = exp(-')'z) g [(1 + Ir1exp(i)}
But
'Y = Lim (Hn - Log n) = Lim [Hn - Log (n + 1)]
n+oo
[
n k + 1J
= Lim Hn -
n+ro I"
Hence
[
n k + 1J
exp( -,),z) = exp -zHn + z {; Log -k-
n [(I{; + 1)' ( z)J
= LIm II exp --: .
n+ro k k
Therefore (1) can be written
zr(z) = DI [(
1
+ i)'ex
p
( -I)(
1
+ IrIexp(i) J,
from which it follows that
(2)
which IS Euler's product for r(z). Note that for real x > 0,
rex) > o.
Refer now to Example (b), page 5, to conclude that
(3)
r(z) = Lim (n - I)! n
z
n+'" z(z + 1) (z + 2) .. (z + n - 1)
It will be of value to us later to note that, since
12 THE GAMMA AND BETA FUNCTIONS [Ch.2
Lim (n --tJl: = 1
n+c:o n
z
'
we can equally well write the result (3) in the form
. n! n
z
(4) r(z) = z(z + l)(z + 2) .. (z + n)
12. The difference equation r(z + 1) = zr(z). From Euler's
product for r(z) we obtain
r(z + 1)
r(z)
Therefore
(1)
00 [( I)Z+I( Z + 1)-IJ
II 1+- 1+--
z n n
z + 1 co [( I)Z( Z)--IJ
II 1+- 1+-
"_I n n
= IT [(1 + .!.)(1 + + z + --.!)-IJ
Z + 1 n n n
z . n(k+l k+Z)
= z + 1 g --1--' k + z +1
= _z _ Lim n +.-!. . 1 + z = z.
z + 1 n+ro 1 n + z + 1
r(z + 1) = zr(z)
for all finite z except for the poles of r( z).
If z = m, a positive integer, iterated use of the equation (1) yields
rem + 1) = m!. Since r(l) = 1, this is another of the many reasons
we define O! = 1.
13. The order symbols 0 and O. Let R be a region in the com-
plex z-plane. If and only if
we write
Lim [(z) = 0
z+c in R g(Z) ,
fez) = o[g(z)], as z -+ c in R.
If and only if I I is bounded as z -+ c in R, we write
fez) = O[g(z)], as z -+ c in R.
14] EVALUATION, CERTAIN INFINITE PRODUCTS 13
It is common practice to omit the qualifying expressions such
as liZ ~ c in R" whenever the surrounding text is deemed to make
such qualification unnecessary to a trained reader. The point
z = c may on occasion be the point at infinity. Also, the symbols
o and 0 are sometimes used when the \"ariablp z is real, the approach
is along the real axis, and e\"en when z takes on only integral values.
EXAMPLE (a):
l" L' sin
2
z 0 .
mnce lIn ---- = ,we may wnte
z+o z
sin
2
z = o(z), as z --> 0,
noting that in this instance the manner of approach is immaterial.
EXAMPLE (b): For real x, leos xl ~ 1, from whieh it is easy to
conclude that
cos x - 4x = O(x), as x --> 00, x real.
EXAMPLE (c): In Chapter 3 we shall show that if
Sn(X) = L k! x
k
,
k=O
li
"'e-
t
dt I
- - ~ - Sn(X)
. a I-xl
~ (n + I)! Ixjn+
1
, for Reex) ~ O .
From the preceding inequality we may eonclude that, for fixed n,
i
"'e-
t
dt
---- - Sn(X) = o(xn),
o I-xl
as x --> 0 in Re(x) ~ O.
14. Evaluation of certain infinite products. The Weierstrass
infinite product for r(z) yields a simple eyaluation of all infinite
products whose factors are rational functions of the index n. The
most general such product must take the form
(1)
because convergence requires that the nth factor approach unity as
n ~ 00, whieh in turn forces the numerator and denominator poly-
14 THE GAMMA AND BETA FUNCTIONS [Ch.2
nomials to be of the same degree and to have equal leading coeffi-
cients. Now the nth factor in the right member of (1) may be put
in the form
so that we must also insist, to obtain conyergence, that
3
(2)
L: ak = L: bk
k=!
If (2) is not satisfied, the product in (1) diverges; we get absolute
convergence or no convergence.
We now have an absolutely convergent product (1) in which the
a's and b's satisfy the condition (2).
Since
we may, without changing the yalue of the product (1), insert the
appropriate exponential factors to write
(3)
The Weierstrass product, page 9, for l/r(z) yields
g [(
1
+ exp( -;i) J = z exp( = -r(z + IJ exp(
Thus we obtain from (3) the result
P _ IT 1'(1 + bk ) exp(
- k=l r(1+ (h) expl
THEOREM 5.
integer,
8 8
If L: a k = L: b k, and 1f no a k or b k is a negative
k=l k=l
15] EULER'S INTEGRAL FOR r(z)
IT (n+aI)(n+a2)' .. (n+a.) = r(1 +bI) r(1 +b2) .. r(1 +b.)
,,=1 (n+b
I
)(n+b2) (n+b.) r(1+aI) r(I +a2)" . r(I+a
8
)
15
If one or more of the ak is a negative integer, the product on the
left is zero, which agrees with the existence of one or more poles of
the denominator factors on the right.
EXAMPLE: Evaluate
co (c - a + n - 1)(c - b + n - 1)
IT (c + n - 1) (c - a - b + ~ n - 1)'
Since (c - a - 1) + (c - b - 1) = (c - 1) + (c - a - b - 1),
we may employ Theorem 5 if no one of the quantities c, c - a,
c - b, c - a - b is either zero or a negative integer. With those
restrictions we obtain
(4)
IT (c - a + n - 1)(c - b + n - 1) = r(c)r(c - a - b).
n=I (c + n - 1)(c - a - b + n - 1) r(c - a) r(c - b)
15. Euler's integral for r(z). Elementary treatments of the
Gamma function are usually based on an integral definition.
Theorem 6 connects the function r(z) defined by the Weierstrass
product with that defined by Euler's integral.
THEOREM 6. If Re(z) > 0,
(1) r(z) = iCOe-ttz-l dt.
We shall establish four lemmas intended to break the proof of
Theorem 6 into simple parts.
Lemma 1. If 0 ~ a < 1, 1 + a ~ exp(a) ~ (1 - a)-I.
Proof; Compare the three series
co
l+a=l+a, exp(a)=I+a+ : a ~ ,
n=2n.
(1-a)-1= 1+a+ L: an.
71=2
Lemma 2. If 0 ~ a < 1, (1 - a)" ~ 1 - na, for n a positive
integer.
Proof: For n = 1, 1 - a = 1 - 1 . a, as desired. Next assume
that
(1 - a)k ~ 1 - ka,
and multiply each member by (1 - a) to obtain
(1 - a)k+I ~ (1 - a)(1 - ka) = 1 - (k + 1)a + ka
2
,
16 THE GAMMA AND BETA FUNCTIONS [Ch. B
so that
(1 - O')k+! 1 - (k + 1)0'.
Lemma 2 now follows by induction.
Lemma 3. If 0 t < n, n a positive ?'nteger,
Proof: Use a = tin in Lemma 1 to get
t (t) ( t)-l
1 + n exp n 1 - n
from which
(2)
(
t)n (t)-"
1 + n e
t
1 - n
or
(
t)-n (t)"
1 + n e- t 1 - n '
so that
(3)
(
t)n
e-
t
- 1 - n o.
But also
and, by (2), e
t
(1 + Hence
(4)
(
t)n [ ( t2)n]
e- t - 1 - n e- t 1 - 1 -
Now Lemma 2 with a = F/n
2
yields
(
t2)" t2
1-- ;;:::1--
n
2
- n
which may be used in (4) to obtain
(5) e-
t
- (1 - e-
t
[1 - 1 + =
The inequalities (3) and (5) constitute the rewlt stated m
Lemma 3.
15] EULER'S INTEGRAL FOR rez) 17
Lemma 4. If n is integral and Re(z) > 0,
(6)
r(z) = Lim {n(l - dt.
n+co Jo n
Proof: In the integral on the right in (6) put t = n{3 and
thus obtain
An integration by parts us the reduction formula
iteration of which yields
i
l n(n-I)(n-2)1 il
o (1 - (3)n(3z-1 d{3 = z(z + l)(z + 2) ... (z + n _ 1) 0 (3z+n-1 d{3
11!
z(z + l)(z + 2) (z + n)
Now (7) becomes
i
"( t)n n!n
z
O
1 - -n t
z
-
I
elt = --:---:-.,--:- --;--:---:-
z(z + 1Hz + 2) ... (z + n)
so that
Lim (n(l _ elt = Lim n!n
z
= r(z)
n+co Jo n n+co z(z + 1) .. (z + n)
by equation (4), page 12.
We are now in a position to prove Theorem 6, which states that
(8) Re(z) > O.
The integral on the right in (8) conyerges for Re(z) > O. With
the aid of Lemma 4, write
18 THE GAMMA AND BETA FUNCTIONS lCh.2
From the convergence of the integral on the right in (8) it follows
that
Hence
But, by Lemma 3 and the fact that Itzi = tHeC,),
i
"tZe- t
~ ----. tH.e(z)-1 dt
o n
lin
- e-tlHe(zHI dt.
n 0
Now i""e- ttRe(zl+1 dt converges, so ine- ttRe(z)+l dt is bounded. There-
fore
~ i ~ iTe-
t
- (1 - ~ ) fiJtt-1 dt = 0,
and we may conclude from equation (9) that (8) is valid.
16. The Beta function. We define the Beta function B(p, q) by
(1) B(p,q) = i1tP-1(1 - t) q-I dt,
Re(p) > 0, Re(q) > 0.
Another useful form for this function can be obtained by putting
t = sin2 cp, thus arriving at
(2) B(p,q) = 2 i!'I" sin
2p
-
1
cp COS
2
q-1cp dcp, Re(p) > 0, Re(q) > 0.
The Beta function is intimately related to the Gamma function.
Consider the product
(3) rep) r(q) = i""e-tt
p
-
1
dt . i""e-Vv
q
-
I
dv.
In (3) use t = X
Z
and v = y2 to obtain
17] THE VALUE OF r(z)r(1 - z)
19
r(p)r(q) = 4.ro iooexp( -x
2
- y2)X
2
p-ly2q-l dx dy.
Next turn to polar coordinates for the iterated integration over the
first quadrant in the xy-plane. Using x = r cos 0, y = r sin 0, we
may write
i
OOi'" rep) r(q) = 4 0 0 exp( _r
2
)r
2P
+2q-2 cos
2
p-1e sin
2q
-
1
e rdedr
Now put r = Vt and e = !7r - <p to obtain
rep) r(q) = iOOe-ttP+q-l dt 2 i
l
.. sin
2p
-
1
<p COS
2
'l-1<p d<p,
from which it follows that
THEOREM 7.
(4)
r(p)r(q) = rep + q)B(p, q).
If Re(p) > 0 and Re(q) > 0,
B(p q) = rep) r(q).
, r(p + q)
By (4), B(p,q) = B(q,p), a result just as easily obtained directly
from (1) or (2).
Equations (2) and (4) yield a generalization of 'Yallis' formula
of elementary calculus. In (2) put 2p - 1 = m, 2q - 1 = n, and
use (4) to write
r(m t ) r ( ~ + ~ )
2r(m +; + 2)
(5)
valid for Re(m) > -1, Re(n) > -1.
17. The value of r(z)r(1 - z). The important relation (4) of
Section 16 suggests that the product of two Gamma functions whose
arguments have the sum unity may possess some pleasant property,
since if p + q = 1, rep + q) = reI) = 1.
If z is such that 0 < Re(z) < 1, both z and (1 - z) have real part
positive, and we may use (4) of Section 16 to write
20
THE GAMMA AND BETA FUNCTIONS [Ch.2
r(z) r(1 - z) = B(z, 1 - z) = llt
z
-
I
(1 - t)-z dt
= (l(_t )Zdt.
)0 1 - t t
Now put t/(1 - t) = y to arrive at
(1)
(OOy,-1 dy
r(z)r(l - z) =)0 1 + y'
o < RC(z) < 1.
The integral on the right in (1) call be e\'aluatl'd with the aid of
contour integration in an a-plane where Re(a) = y. The contour
IX-plane
R
Figure 1
C in Figure 1 encircles a single simple pole a = -1 of the integrand
III
f
a,-1 da
c 1 + a
so that the residue theory at once yields
f
z-1 d
(2) c a
1
+ : = 27ri( _1)z-1 = 27ri exp[7ri(z - 1)].
17] THE VALUE OF r(z)r(1 - z) 21
The integral on the left in (2) may be split into four parts, as indi-
cated in the figure. In detail \ve use
(a) a = Re
i8
, e from 0 to 27r;
(b) a = ye
27ri
, y from R to 8;
(c) a = 8e
i8
, e from 27r to 0;
(d) a = yeO
i
, y from 0 to R.
Thus (2) can be written in the form
(3) triRz exp(1ze) de + ('y
Z
-
1
exp(27r1z) dy
J
o
1 + R exp(ie) JR 1 + y
+i
o
io
z
expUze) .dO +JR
YZ
-
1
exp(Oiz) dy = 27ri ex [7r'i(z-I)J.
21r 1 + 0 exp(le) 0 1 + y p
Now let 8 ----> 0 and R ----> (X) and use 0 < Re(z) < 1 to conclude that
the first and third integrals on the left in (3) approach zero. Then
the limiting form of (3) is
. fO yz-l dy (ooyz-l dy . .
exp(27rlZ) 00 1 + y + J
o
1 + Y = - 27rL exp(7rl.Z),
from which \ye obtain
i
ooYZ
-
1
dy = 27ri exp( 7riz)
o 1 + y exp(27rlz) - 1 exp( 7rIZ) - exp( - -rriz)
We have thus shown that, for 0 < Re(z) < 1,
(4)
r(z) r(I _ z) = (ooyz-l dy = _. _7r_.
J 0 1 + Y SIn 7rZ
But each member of (4) is analytic for all nonintegral z, and the
theory of analytic continuation permits us to come to the useful
conclusion of Theorem 8.
THEOREM 8. If z is nonintegral,
r(z) r(I - z)
7r
= ~
SIn 7rZ
Our first, and extremely simple, application of Theorem 8 is the
evaluation of r(!). Use z = ! to get
r(!)r(!) = 7r,
which, since r( ) > 0, yields
(5) r(!) = -yI;.
22
THE GAMMA AND BETA FUNCTIONS [Ch.13
18. The factorial function. Throughout this book we make fre-
quent use of the common notation
(1) (a)n = II (a + k - 1)

= a(a + 1)(a + 2) .. (a + n - 1), n 1,
(a)o = 1, a O.
The function (a)n is called the factorial function. It is an immediate
generalization of the elementary factorial, since n! = (l)n.
In manipulations with (a)n it is important to keep in mind that
(a)n is a product of n factors, starting with a and with each factor
larger by unity than the preceding factor.
Lemma 5. (a)2n =
Proof: In the product
(a)2n = a(a + l)(a + 2)(a + 3) .. (a + 2n - I),
group alternate factors, factor 2 out of each factor on the right,
and thus conclude that
? (a) (a + 1)
(a)2n = 2
n
"2 n --2- n'
Lemma 6. If k is a positive integer and n a non-negative integer,
(2) (ahn = (a ... .
k n k" Ie n
The proof of Lemma 6 is like that of Lemma 5 except that the
factors of (a)kn are grouped into k sets of n factors each, and then
k is factored out of each factor to obtain (2).
Other properties of (a)n will be introduced when needed, particu-
larly in series manipulations involving functions of hypergeometric
character. At present we are concerned only with the relation of
(a)" to the Gamma function.
We know that r(1 + z) = zr(z). It follows that, for n a posi-
tive integer,
r(a + n) = (a + n - l)r(a + n - I)
= (a + n - I)(a + n - 2)r(a + n - 2)
= (a + n - I)(a + n - 2) ar(a).
19] LEGENDRE'S DUPLICATION FORMULA 23
THEOREM 9. If 01 is neither zero nor a negative 1'nteger,
(3) (
01) = r(OI_+ n).
n r(OI)
We ha\'e already had, in equation (3), page 11, the result
() L
' (n-1)!nz
rz = 1m ,
n+w z(z + l)(z + 2) (z + n - 1)
which can now be written in the form
(4) r() L
' (n-1)!nz
z = 1m .
n+w (z) n
Equation (4), reinterpreted in the light of Theorem 9, yields a
result of \'alue to us in the subsequent two sections.
Lemma 7. If n is integral and z is not a negative integer,
(5) L
(n - 1)! n
z
I
Jm =
n+w r(z + n) .
19. Legendre's duplication formula. Let us turn to Lemma 5,
page 22, and use 01 = 2z. We thus obtain
In view of Theorem 9 \ve may re\\Tite the above a;.;
or
r(2z + 2n)
-----
r(2z)
r(2z)
r(z)r(z+!)
22nr(z + n) rez + ~ + n)
r(z) r(z + )
r(2z + 2n)
22nr(z + n) r(z + ~ + n)'
which, since the left member is independent of n, also implies
(1) __ IJ2z) __ = Lim r(2z + 2n) .
r(z) r(z + !) n+w 22n r(z + n) r(z + ! + n)
We next insert in the right member of (1) the appropriate factor;.;
to permit us to make use of the result in Lemma 7. From (1)
we write
r(2z)
r(z) r(z + !)
= Lim r(2z+2n) (n-I) ln
z
(n-I) ln
zH
n+w (2n-I)!(2n)2z r(z+n) r(z+!+n)
22Z(2n-1)!
22nnl[(n -1)!)2'
24 THE GAMMA AND BETA FUNCTIONS [Ch.2
which, because of Lemma 7, becomes
_IJ2z) __ = Lim 22z(2n - I)! .
r(z) r(z + !) n+", 22nnl[(n - 1) !J2
It follows that
r(2z)
22zr(z) r(z + !)
= c,
in 'which c is independent of z. To e\'aluate c we use z = ! and
find that
r(I) 1
c = ---- = --'
2r(t) reI) 2\1;
We have thus discovered an expression for r(2z) in terms of r(z)
and r(z + !). It is Legendre's duplication formula,
(2) \I; r(2z) = 2
2z
-
1
r(z) r(z + D.
20. Gauss' multiplication theorem. Following the technique
used to discover and prove Legendre's duplication formula, we
readily move on to a theorem of Gauss involving the product of k
Gamma functions.
Lemma 6, page 22, can be written
(a)nk = knk IT (a + s. - 1)
s=l I.. n
and by Theorem 9, page 23, (a)n = rea + n)/r(a). We thus
obtain
(
a + s - 1 )
r(a_- nk) = knk IT r ----,,-. -- + n .
rea) r( ;. - 1)
(1)
In (1) put a = kz and rearrange the members of the equation to
arrive at
(2)
r(kz) r(kz + kn)
k ( S - 1)
g r z + -k-
k ( S - 1)
knk II r z + n +
/,
L
' r (/(Z + /;11)
= k ( S - 1)
knk II r z + n + -,-
k
20] GAUSS' MULTIPLICATION THEOREM 25
By Lemma 7, page 23, we know that
0_1
z+--
L
' (n - 1)! n k
Im--------------
n+
oo
r( z + n + S __ - I )
1,
and
Lim (11k - I)! (nk)kz = 1.
n+ro r(kz + I-.:n)
We now use the foregoing two limits in conjunction with (2) to
obtain
r(kz)
IT r(z + ~ - - - . ! )
.=1 k ,-I
= Lim r(kz+kn) (nk)kz(nk-l)! IT (n-l)!r/+
k
_____ 1 ___ _
n+ro(nk-1)!(nk)kz knk ,=1 ( S-l) z+'-1
r z+n+-
k
- (n-l)!n k
L
' (nkyz(nk - 1)! Ilk 1
= 1m
n+oo knk .,=1 z+,-1
(n - 1)! n k
= Lim (nk)kz(nk - 1)!
n+oo knk[(n - 1)I]k
n
kz+!(k-l)
Therefore,
r( kz)
--------- = C,
k ( S - 1)
kkz II r z + -,-
8_1 k
in which c is independent of z, To determine c, we put z = 11k,
use the fact that 1'(1) = 1, and obtain
- = II r ~ = II r ~ _ S
1 k-l () k-l (l )
kc s=1 k 8_1 k
Then
1 k-l () (k ) k-l
_ = II r ~ r ~ - S = II _7r_
k
2
c
2
8=1 k k ,=1 sin 7rS'
k
or
(3)
26 THE GAMMA AND BETA FUNCTIONS [Ch.2
We can obtain c once we know the yalue of the product on the
right in (3).
k-I k
L 8 If k
> 2 II . 7rS - _.
emma. =, 8_1 SIn k - 2k-1
Proof: Let a = exp(27ri/k) be a primitive kth root of unity.
Then for all x,
k-I
x
k
- 1 = (x - 1) II (x - a'),
8=1
from which, by differentiation of both members, we get
k-I
(4) kX
k
-
1
= II (x - a') + (x - l)g(x),

in which g(x) is a polynomial in x. Put x = 1 in (4) to obtain
k-I
k = II (1 - a').
8_1
But

(7rlS)[ (7riS) (-7riS)]


1 - a' = 1 - exp k = -exp k exp k - exp --r-
2
. 7rS
= - t exp k sm -i:;'
Hence
k-I k-I
k = (-2i)k-1 exp[!7ri(k -1)J II sin = 2
k
-
1
II sin 7rS,
k k
which yields the desired result.
With the aid of Lemma 8, equation (3) can be written
k
k
2
c
2
7r
k
-
1
= --'
2
k
-
1
The constant c is positive because the Gamma function is positive
for positiYe argument. Hence
c = (27r)-!(k-!)k-!.
This completes the proof of the Gauss multiplication theorem.
THEOREM 10 IT r(z + S - 1) = (27r)W-I)/,;I-
kz
r(kz)
-1 k
21. A summation formula due to Euler. Let
P(x) = x - [xJ -
21] A SUMMATION FORMULA DUE TO EULER 27
in which [x] means the greatest integer ~ x, a notation also used
frequently in later chapters. The function P(x) is periodic,
P(x + 1) = P(x),
and is represented graphically in Figure 2.
/
-2 -1
1
:I
P(x)
Figure 2
2
3 x
/
Euler employed P(x) in obtaining some useful summation for-
mulas, of which we use only that in Theorem II.
THEOREM 11. If f'(x) is continuous for x ~ 0,
f(k) = if(X) dx + yeO) + yen) + inp(x)f'(x) dx,
k_O 0 0
in which P(x) = x - [x] - !.
Proof. First write
inp(x)f'(x) dx = fk P(x)f'(x) dx.
o k_l k-l
Now
fk P(x)f'(x) dx = fk (x - k + !)!'(x) dx,
k-l k-I
and we integrate by parts to obtain
i:l P(x)f'(x) dx = [(X - k + !)f(x) I_I -i:l f(x) dx
= Y(k) + U(k - 1) - i ~ l f(x) dx.
28 THE GAMMA AND BETA FUNCTIONS [Ch.2
We may therefore write
= ! I: f(k) + - flex) dx
k_1 k_O 0
= I:f(le) - !leO) - H(n) - ii'cx) dx,
0
which is a simple rearrangement of the result in Theorem 11.
Lemma 9. For I arg z i ;;;; 7r - 0, where 0 > 0,
"
L Log(z + Ie) = (z + n + Log(z + n)

1 fnp(x) dx
- n - (z - 2") Log z + 0 x .
Proof: Lemma 9 follows at once by applying Theorem 11 to the
function f(x) = Log (z + x).
Let us next turn to the result
(1) r(z) = Lim (n -=--.!)! n
z
71";':.0 (z) 1't '
established on page 23. In (1) put z = and shift from n to
(n + 1) to obtain
Limn! (C':+ 1)! =
n+co Z)n+l
which may be put in the form
. n! (n + 1) In! 22n -
LIm e (1) 22n = V7r.
n+co 2) ( n n
Now Lemma 6, page 22, yields
2
2n
(Dn(l)n = (2)2n = (2n + I)!.
Therefore we ha,"e
(2)
It is legitimate to take logarithms of each member of (2) and
thus write
22] THE BEHAVIOR OF LOG r(z) FOR LARGE Izl 29
(3) [(2n + 1) Log 2 - Log(1 + n) + 2r; Log(l + k)
- Log(1 + k)] = ! Log 7T.
We shall apply the formula of Lemma 9 to thc sums invoh'cd
on the left in equation (3). Thc choice z = 1 in the result in
Lemma 9 yields
(4) Log(1 + k) = + n) Log(l + n) - n + :x.
By Lemma 9, with z = 1 and 11 rcplaced by 2n, we gct
2n (3) i2np( ) d
t; Log(l + k) = 2" + 2n LogO + 2n) - 2n + 0 1 xX.
Equation (3) can now be put in the form
+ 1) Log 2 + (2n + D Log 11.}2
n
n
+ 2fnp(x) dx _ i
zn
P(x) dX]
ol+x ol+x
1
= 2 Log 7T,
which leads* to
. [ 1 (3)' 2 + 2n]
- 2 Log 2 + 2n + 2 Log 1 + 2n
+
l
oopex) dx - ! L
o 1 + x - 2 og 7T.
But
[( 211 Log i = 1.
Therefore we arrive at thc evaluation
(5)
22. The behavior of log r(z) for large Iz/. From formula (3),
page 11, it follows that
( ) L
. (n + I)! (n + 1).-1
rz = un ,
"+00 (Z),,+l
*For proof of convergence of l'" a:, sce the eXE'r('ises at the end of this chapter.
30 THE GAMMA AND BETA FUNCTIONS [Ch.2
and so also that
(1) log r(z)
= C1+k)+(z-1) Log (1+n)-r;LOg (Z+k)].
Using equation (4) and Lemma 9 of the preceding section, we
may now conclude that, if I arg(z) I 7f - 0, 0 > 0,
(2) log r(z) = [(z + n + {Log(l + n) - Log(z + n) I
+ (z - 1) Log z +J:n P(x) dx _ (n!J!'L dX].
2 0 1 + x Jo z + x
The elementary limit
= 1-z,
together with equation (5) of Section 21, permits us to put (2) in
the form
log r(z) = 1 - z + (z - 1) Log z - 1 + 1 Log (27f) _ (COP(x)dx.
2 2 Jo z + X
THEOREM 12. If I arg(z) I 7f - 0, where 0 > 0,
(3) log r(z) = (z - Log z - z + Log (27f) _ :x,
in which P(x) = x - [xJ - as in Section 21.
Let us next consider the integral on the right in (3). Since
f P(x) dx = P2(X) + c,
we may use c = - -h and integrate by parts to find that
(roP(x) dx = 1 [P2(X) - + 1 ('"'[P(x) - -l:d dx
J
o
z+x 2 z+x 0 2J
o
(Z+X)2
= __ 1_ + 1 (CO[P(x) - -l2"J dx.
12z 2J
o
(z + X)2
Now the maximum \'alue of [P2(X) - 1
1
2J is i and, in the region
I arg z I 7f - 0, 0 > 0,
Iz + Xl2 X2 + IZ!2, for Re(z) 0,
22] THE BEHAVIOR OF LOG r(z) FOR LARGE Izl 31
Iz + Xl2 [x + Re(z)J2 + Izl2 sin
2
0,
It follows that
for Re(z) < o.
("'l
p2
(x) - dx _ 0(_1 )
J
o
(z + X)2 - Izl'
as I z I co in I arg z I 7r - 0, 0 > O.
We have shown that as Izl co in larg zl 7r - 0,0 > 0,
(4) log r(z) = (z - !) Log z - z + ! Log(27r) + 0(1).
Indeed we showed a little more than that, but (4) is itself more
precise than is needed later in this book.
From (4) we obtain at once the actual result to be employed in
Chapter 5.
THEOREM 13. As I z I ro 1'n the region where I arg z I 7r -
and I arg(z + a) I 7r - 0, 0 > 0,
(5) log r(z + a) = (z + a - !) Log z - z + 0(1).
EXERCISES
. r'(z) 1 ro( 1 1)
1. Start WIth - = -'Y - - - L -- - - ,
r(z) z Z + n n
prove that
2r'(2z) r'(Z) f'(z + 1)
r(2z) - fez) - 1'(z +1) = 2 Log 2,
and thus derive Legendre's duplication formula, page 24.
2. Show that f'C!) = - ('Y + 2 Log 2h/;':.
3. Use Euler's integral form fez) = dt to show that f(z + 1) = zf(z).
4. Show that fez) = Lim n'B(z, n).
n+
ro
5. Derive the following properties of the Beta function:
(a) pB(p, q + 1) = qB(p + 1, q);
(b) B(p, q) = B(p + 1, q) + B(p, q + 1);
(e) (p + q)B(p, q + 1) = qB(p, q);
(d) B(p, q)B(p + q, r) = B(q, r)B(q + r, pl.
6. Show that for positive integral n, B(p, n + 1) = n!/(P)n+1'
7. Evaluate1:(1 + x)p-I(1 - X)q-I dx.
Ans. 2
p
+
q
-
1
B(p, q).
32 THE GAMMA AND BETA FUNCTIONS [Ch.:8
8. Show that for 0 k n
() _ (-I)k(a)n
a n-k - (l - a - nh
Note particularly the special case a = l.
9. Show that if a is not an integer,
r(l-a-n) (_l)n
-r(I=-a) - = Taj-;:-'
In Exs. 10-14, the function P(x) is that of Section 21.
10. Evaluate ('P(y) dy.
Jo
Ans. - i.
11. Use integration by parts and the rcsult of Ex. 10 to show that

< _.
n 1 + x = 8(1 + n)
12. With the aid of Ex. 11 prove the convergence of :x .
13. Show that
(ro P(x) dx = f= fn+lp(X) dx = (I(y - dy.
Jo 1 + x n 1 + x n_O Jo 1 + n + y
Then prove that
Lim n2 (\y - dJL =
n+ro Jo 1 + n + y 12
(ro P(x) dx .
and thus conclude that Jo l-+x IS convergent.
14. Apply Theorem 11, page 27, to the function f(x) = (1 + X)-I; let n --> 00
and thus conclude that
'Y = - iroy-2P(Y) dy.
15. Use the relation f(z)f(l - z) = 1r/sin 1rZ and the elementary result
sin x sin y = [cos (x - y) - cos (x + y)]
to prove that
1 _ f(c)f(1 - c)f(c - a - b)f(a + b + 1 - c)
f(c - a)f(a + 1 - c)f(c - b)l'(b + 1 - c)
1'(2 - c)f(c - l)f(c - a - b)f(a + b + 1 - c)
f(a)f(1 - a)f(b)f(l - b)
CHAPTER 3
Asymptotic
Series
23. Definition of an asymptotic expansion. Let us first recall
the sense in ,vhich a convergent power series expansion represents
the function being expanded. When a function F(z), analytic at
Z= 0, is expanded in a power series about z = 0, we write
(1)
Izl < r.
Define a partial sum of the series by
Sn(Z) = L CkZ
k

k ~ O
Then the series on the right in (1) represents F(z) in the sense that
(2)
Lim [F(z) - Sn(Z) ] =
n+
co
for each z in the region I z I < r. That is, for each fixed z the series
in (1) can be made to approximate F(z) as closely as desired by
taking a sufficiently large number of terms of the series.
We now define an asymptotic power series representation of a func-
tionf(z) as z ~ in some region R. We ,,,rite
co
(3) fez) ~ L anz
n
,
z ~ in R,
Tl_O
if and only if
33
34
(4) Lim
.+0 in R
ASYMPTOTIC SERIES [Ch.3
Ifez) - Snez) I
Izi n = 0,
for each fixed n, with
(5) Sn(Z) = L akz
k


By employing the order symbol defined in 13, we may
write the condition (4) in the form
(6) fez) - Sn(Z) = o(zn), as Z ----> 0 in R.
Here we see that the series in (3) represents the function fez) in
the sense that for each fixed n, the sum of the terms out to the
term anz
n
can be made to approximate fCz) more closely than
I z I n approximates zero, in the sense of C 4), by choosing Z sufficiently
close to zero in the region R.
It is particularly noteworthy that in the definition of an asymp-
totic expansion, there is no requirement that the series converge.
Indeed some authors include the additional restriction that the
series in (3) diverge. Most asymptotic expansions do di verge, but
it seems artificial to insist upon that behavior.
Asymptotic series are of great value in many computations. They
play an important role in the solution of linear differential equations
about irregular singular points. Such series were used by astronomers
more than a century ago, long before the pertinent mathematical
theory was developed.
EXAMPLE: Show that
(7)
i
""e-
t
dt ""
1 _ t "-' L n!xn,
o x n=O
x ----> 0 in Re(x) O.
Let us put
Sn(X) = L k!x
k


In the region Re(x) 0, the integral on the left in (7) is abso-
lutely and uniformly cOll\'ergent. To see this, note that t 0 so
that Re(l - xt) 1. Hence 11 - xt I 1, and we have
I J:""e-
t
dt = 1.
For k a non-negative integer,
23] DEFINITION OF AN ASYMPTOTIC EXPANSION 35
(8) J:"'e-tt
k
dt = r(k + 1) = kL
Hence
f
"" e-t dt f"" e-
t
dt n f""
-::--- - Sn(X) = --- - L e-ttkx
k
dt
o 1 - xt 0 1 - xt k _0 0
= J""e-t[-_l- - t (xt)k]dt.
o 1 - xt
From elementary algebra we haye
r =;6. 1.
Therefore
f
"" e-t dt f""e-t(xt) n+l dt
-- - S (x) = ----
o 1 - xt n 0 1 - xt '
from which, since 11 - xt I 1, we obtain
I - Sn(X) I
I x I n+li""e-'t
n
+
1
dt, in Re(x) o.
We may conclude that
(9) 1.["" - Sn(X) I
(n + I)! I x I n+ 1, in Re(x) o.
From (9) it follows at once that the condition (4), page 34, is
satisfied, which concludes the proof. Actually (9) more infor-
mation than that. Let En(x) be the error made in computing
the sum function by discarding all terms after the term n!x
n

Then I En(x) I is the left member of (9), and the inequality (9)
shows that I En(x) I is smaller than the magnitude of the first term
omitted. This property, although not possessed by all asymptotic
series, is one of frequent occurrence.
The preceding example gi\"es little indication of methods for
obtaining asymptotic expansions. Later we shall exhibit two com-
mon methods, successi\"e integration by parts and term-by-term
integration of power series.
Extension of the concept of an asymptotic expansion to one in
which the variable approaches any specific point in the finite plane
is direct. For finite Zo we say that
""
fCz) '" L an(z - zo)n, as z --> Zo in R,
n_O
36 ASYMPTOTIC SERIES [Ch.3
if and only if, for each fixed n,
j(z) - Sn(Z) = o([z - Zo]n), as Z ---+ Zo in R,
in which
24. Asymptotic expansions about infinity. Asymptotic senes
are often used for large 1 z I. We say that
ro
(1) J(z) ,-...., L anz-
n
, as z ---+ 00 in R,
n=O
if and only if, for each fixed n,
(2) J(Z) - Sn(Z) = o(z-n), as Z ---+ 00 in R,
in which
n
(3) Sn(Z) = L akz-
k


At times, as in the subsequent example, we wish to work only
along the axis of reals. 'Ve then use (1), (2), and (3) for a real
variable x, with the region R replaced by a direction along the real
aXiS.
One last extension of the term asymptotic expansion follows. It
may be that J(z) itself has no asymptotic expansion in the sense of
the foregoing definitions. We do, however, write
ro
(4) J(z) '"" h(z) + g(z) L anz-
n
, as Z ---+ 00 in R,
n=O
if and only if
(5) as z ---+ 00 in R,
and similarly for asymptotic expansions about a point in the finite
plane.
Obtain, for real x, as x ---+ 00, an asymptotic expansion
of the error function
(6)
2 LX
erf(x) = ---= exp( _t2) dt.
V7r 0
From the fact that rCD = vi:;', it follows at once that
24] ASYMPTOTIC EXPANSIONS ABOUT INFINITY 37
Lim erf(x) = 1.
z+ro
Let us write
2 rro 2 fro
erf(x) = ---=In exp(-[2) dt - ---= cxp(-t!) dt
V
7r
0 V7r x
2 fro
= 1 - -:= exp( - t2) dt.
V7r z
N ow consider the function
B(x) = fooex
p
( _t2) dt
and integrate by parts to get
Iteration of the integration by parts soon yields
B(x) =
[
1 1 13 135 (-1)n135 .. (2n-1)]
exp( _X2) 2x - 2Zx3 + 23x5 - 24
X
7 + ... + 2n+lxZn+l
(-1)n+l13 5 .. (2n+ l)jro
+ t-
2n
-
2
exp( - t2) dt
2n+1 z . ,
or
(7)
+ (_l)n+l(!) f"'t-Zn-Zexp( - t
Z
) dt.
2 n+l z
Let
Then, from (7),
exp(x
2
)B(x) - Sn(X) = (_l)n+la)n+l exp(x
Z
) j"'t-
Zn
-
2
exp( _t2) dt.
x
38 ASYMPTOTIC SERIES [Ch.3
The variable of integration is never less than x. We replace the
factor t-
2n
-
2
in the integrand by tx-
2n
-
3
and thus obtain
r
exp (x
2
)B(x) - Sn(X) I < ( ~ ) n + 1 ex
p
(x
2
)f"'t exp( _t2) dt
X
Zn
+
3
" '
from which it follows that
(8)
Hence
I exp(x2)B(x) - Sn(X) I < (nn+l.
I 2x
2n
+3
exp(x2)B(x) - Sn(X) = 0(X-
2n
-
2
) , as x ----; co,
which permits us to write the asymptotic expansion
exp(x2)B(x) '" i: (-l)n(!)n
n ~ O 2x
2n
+
1
,
x ----; co.
But erf(x) = 1 - 2_ B(x). Hence
v
7T
(9)
1 '" (_l)n(l)
erf(x) '" 1 - _ exp( _X2) " 2 n
. j-L..J x2n+1 ,
'V ~ 71=0
Note also the useful bound in (8).
x ----; CO
25. Algebraic properties. Asymptotic expansions behave like
convergent power series in many ways. We shall treat only ex-
pansions as z ----; co in some region R. The reader can easily extend
the results to theorems in which z ----; Zo in the finite plane.
THEOREM 14. If, as z ----; en in R,
=
(1)
f(z) '" L anz-
n
n_O
and
CD
(2) g(z) ~ L bnz-
n
,
71=0
then
00
(3) fez) + g(z) ~ L (an + bn)z-n
n=O
and
'"
n
(4)
f(z)g(z) '" L L akbn_kz-
n

n =u k=O
26] TERM-BY-TERM INTEGRATION
Prooj: Let
Sn(Z) = L akz-\ Tn(Z) = L bkz-
k

k=O k=O
From (1) and (2) we know that
(5)
(6)
from which
yielding (3).
j(z) - Sn(Z) = o(z-n),
g(z) - T n(Z) = o(z-n),
j(Z) + g(z) - [Sn(Z) + Tn(Z)] = o(z-n),
To prove the validity of (4), first put
n k
39
which is the "nth partial sum" of the series on the right in (4). By
direct multiplication,
Sn(Z) T n(Z) = Qn(Z) + o(z-n),
and by (5) and (6),
j(z)g(z) = Sn(z)Tn(z) + o(z-n).
Hence
j(z)g(z) = Qn(Z) + o(z-n),
which shows the yalidity of (4).
The right member in (4) is the ordinary Cauchy product of the
series (1) and (2).
26. Term-by-term integration. Suppose that for real x we haye
(1) X ->0:>.
n=O
Surely we are interested here in large x, so that an integral which it
is natural to consider is fCCj(x) dx. But f"'a
o
dx and f"'a1x-
1
dx do
v v v
not exist. Therefore we restrict ourselves to the consideration of
an expanSIOn
co
(2) g(x) "-' L anx-
n
,
n=2
40 ASYMPTOTIC SERIES [Ch.3
and seek j""g(x) dx. Of course g(x) = f(x) - ao - alx-
1

u
Let
Snex) = L akx-
k

k-2
Then
g(x) - Snex) = o(x-
n
), X ---'> co,
and
~ f""l g(x) - Sn(X) I dx
"
< f'"'1 o(x-n) I dx
y
= o(y-n+l).
But
Hence
(3)
the desired result.
27. Uniqueness. Since e-
X
= O(Xk), as x ---'> co, for any real k,
whole classes of functions have the same asymptotic expansion.
Surely if
ro
f(x) ro../ L Anx-
n
,
n_O
then also
""
f(x) + ce-
X
ro../ L Anx-
n
,
71=0
and numerous similar examplcs arc casily concocted.
On the other hand a gi\"cn function cannot have more than one
asymptotic expansion as z --? Zo, finitc or infinite. Let us use Z --? co
in a region R as a representative example.
THEOREM 15. If
co
(1) fez) ro../ L Anz-n, Z ---'> co in R,
n=O
28] WATSON'S LEMMA
41
and
ro
(2) fez) "-' L Bnz-n,
z - co in R,
n_O
then An = Bn.
Proof: From (1) and (2) we have
fez) - L Akz-k = o(z-"),
k-O
fez) - L BkZ-k = oCz-"),

from which it follows that
L (Ak - Bk)Z-k = o(z-,,) ,

or its equivalent
n
L (Ak - Bk)zn-k = 0(1), z _00 in R,

for each n. Therefore A k = B k for each k. The expansion (1)
associated with z - 00 in a particular region R is unique. The
function fez) may, of course, ha\>e a different asymptotic expansion
as z - 00 in some region other than R.
28. Watson's lemma. The following useful result due to
Watson [1 ;236J gives conditions under which the term-by-term
Laplace transform of a series yields an asymptotic representation
for the transform of the sum of the series. For details on Laplace
transforms see Churchill [1].
Since relatively complicated exponents appear in the following
few pages, we shall simplify the printing by the introduction of a
notation similar to the common one, exp u = e
u
The symbol
is defined by
expxCm) = xm.
Watson's Lemma. Let F(t) satisfy the following conditions:
(1) FCt) = fan expt('/1, - 1), in itl :;;:; a + 0, with a, 0, r > 0;
n=! r
(2) There exist positive constants K and b such that
IF(t) \ < Ke
bt
, for t a.
42 ASYMPTOTIC SERIES [Ch.3
Then
(3)
as 's' -4 co in the region 'arg s I 3 h - for arbitrarily small
positive
Note that (1) implies that F(t) is either analytic at t = or has
at most a certain type of branch point there.
Proof: It is not difficult to show (Exs. 1 and 2 at the end of this
chapter) that under the conditions of Watson's lemma, there exist
positive constants c and (3 such that for all t 0, whether t 3 a or
t > a,
(4) IF(t) - - 1)/ < - l)e
pt

We know also the Laplace transform of a pmver of t,
m > -1, Re(s) > 0.
In order to deriye (3), we need to show that for each fixed n
If(S) - t akr( 1 sl niT = 0(1),
as I s I -4 co in I arg s! 3 - > 0.
Now
f(s) - i: = fcoe-'t[F(t) - i: ak - l)J dt.
k=l r Jo k=l r
Hence, with the aid of (4),
Islnlrlf(S) - t
(
n + 1) (n+l)
< clslnlrr --r:- [Re(s) - (3r --;-,
if Re(s) > (3. In the region I arg s I 3 - > 0, Re(s) > {3
28] WATSON'S LEMMA 43
as soon as we choose I S I > (J(sin Ll)-l. Therefore, as I s I ---+ (X) ill
the region I arg s I !11" - Ll,
Isln'rlf(S) - 1; = 0(1),
as desired.
EXAMPLE: Obtain an asymptotic expansion of
(CO
e
-
XI
dt
f(x) = J
o
I + t2'
I x I ---+ co in I arg x I ! 11" - Ll, Ll > O.
Note that the result will be valid in particular for real x -> co.
We shall apply Watson's lemma with F(t) = 1/(1 + f2). Then
co co
F(t) = L (-1)nt2n = L (_1)n+ll2n-2,
It I < 1,
n=O n=l
so that we may write
n=l
in I t I
in which aZ
n
= 0, aZn_l = (-l)n+l, and we have chosen r = 1,
a = !, 0 = i in the notation of Watson's lemma.
For t ?; !, e
t
> 1 and 1/(1 + t2) < 1, from which
F(t) = 1 ! t2 < e
t

We may therefore conclude from Watson's lemma that
or
f"'e-
xt
dt co (-1)n+lr(2n - 1)
J
o
I + t
2
"""" f; X
2n
-
1
,
and finally that
(6)
{"'e-xtdt co (-1)n(2n)!
J
o
1 + r"""" ?; X
2n
+
1
,
as Ixl ---+ (X) III larg xl !11" - Ll, Ll > O.
EXERCISES
1. With the assumptions of Watson's lemma, page 41, show, with the aid of the
convergence of the series in (1), that for 0 t a, there exists a positive constant
Cl such that
44 ASYMPTOTIC SERIES lCh.3
[
F(t) - i:. ak - 1) I < c, - 1).
k=l r I r
2. With the assumptions of Watson's lemma, page 41, show that for t > a,
there exist positive constants Cz and (3 such that
3. Derive the aHymptotic expansion (6) immediately preceding these exercises
by applying Watson's lemma to the function
f'ex) = _ ("'e-
xt
dt
Jo 1 + t2
and then integrating the resultant expansion term by term.
4. Establish (6), page 4:3, directly, first shO\\'ing that
n fWe-"t2n+2 dt
f(x) - L (-l)k(2k) !X-
2H
= (_l)n+' -'----'2-,
k_O .01+t
and thus obtain not only (6) but also a bound on the error made in computing
with the series involved.
5. Use integration by parts to establish that for rcal x ...... 00,
6. Let the Hermite polynomials Hn(x) be defined by
co H n(X)tn
exp(2xt - (2) = L - -,-
n=O n.
for all x and t, as in Chapter 11. Also let the complementary error function
erfc x be defined by
2 fm
erfc x = 1 - erf x = --.:: e::.:p( - (32) d{3.
,\/7r I
Apply Watson's lemma to the function F(t) = exp(2xt - t2); obtain
exp(x - exp( - (32) d{3 "-' E-/! n(X)s-n-', s -> 00,
and thus arrive at the result
!t-'Y; exp[W-l - x)2]erfcW-l - x) "-' f: Hn(x)t
n
, t -> 0+.
n_O
7. Use integration by parts to show that if Re(a) > 0, and if x is real,
{ooe-tt-a dt "-' xl-ae-I f: (-l)n(a)n
J;c n=O x
n
+
1
,
x -> 00,
of whieh Ex. 5 is the special case ex = 1.
CHAPTER 4
The Hypergeometric
Function
29. The function F(a, b; c; z). In the study of second-order
linear differential equations with three mgular singular points, there
arises the function
(1)
00 (a)n(b)n
zn
F(a, b; c; z) = 1 + L -(-)--,-,
71=1 C nn.
for c neither zero nor a negatiye integer. In (1) the notation
(2) (a)n = a(a + 1)(a + 2) .. (a + n - 1), n ~ 1,
(a)o = 1, a ~ 0,
of Section 18 is used. We are here concerned with yarious properties
of the special functions under consideration; that (1) satisfies a
certain differential equation is, for us, only one among many facts
of interest.
Since
Lim I (a)n+l(b)n+lZn+l ~ n n ! I
n+oo (C)n+l(n + I)! (a)n(b)nzn
= Lim I (a + n)(b + n)zl = I I
n+w , (c + n)(n + 1) z ,
so long as none of a, b, c is zero or a negati\'e integer, the series in
(1) has the circle I z I < 1 as its circle of conyergence. If either or
both of a and b is zero or a negatiye integer, the series terminates,
and conyergence does not enter the discussion.
45
46 THE HYPERGEOMETRIC FUNCTION [Ch.4
On the boundary I z I = 1 of the region of convergence, a sufficient
condition for absolute convergence of the series is Re(c-a-b) > o.
To prove this, let
o = ! Re(c - a - b) > 0,
and compare terms of the series
(3)
with corresponding terms of the series
(4)
known to be convergent. Since I z I = 1 and
Lim I
n+co (C)nn.
. I (a)n (b)n (n - l)!n
c
(n - 1)!
n1H
I
= (n - 1) !n
a
. (n - 1) !n
b
--cc3-n -. n!n
c
-
a
-
b
= I rta) . rtb) . ric) I I nc-Lb-ol = 0,
because Re(c - a - b - 0) = 20 - 0 > 0, the series in (1) is ab-
solutely convergent on Izi = 1 when Re(c - a - b) > o.
A mild variation of the notation F(a, b; c; z) is often used; it is
(5)
F[a, b;
c ,
which is sometimes more convenient for printing and which has the
advantage of exhibiting the numerator parameters a and b above the
denominator parameter c, thus making it easy to remember the
respective roles of a, b, and c. \Vhen we come to the generalized
hypergeometric functions, we shall frequently use a notation like
that in (5).
The series on the right in (1) or in
(6)
F[a, b;
C" ,
30] A SIMPLE INTEGRAL FORM 47
is called the hypergeometric series. The special case a = c, b = 1
co
yields the elementary geometric series L zn; hence the term hyper-
n=O
geometric. The function in (6) or in (1) is correspondingly called the
hypergeometric function. Although Euler obtained many properties
of the function F(a, b; c; z), we owe much of our knowledge of the
subject to the more systematic and detailed study made by Gauss.
30. A simple integral form. If n is a non-llegative integer,
(b)n reb + n) r(c)
(C)n = r(c + n) reb)
r(c)
reb) r(c - b)
r(b + n) r(c - b)
r(c + n)
If Re(c) > Re(b) > 0, we know from Theorem 7, page 19, and the
integral definition of the Beta function, that
reb + n) r(c - b) = (ItHn-I(1 _ t)c-b-l dt.
r(c + n) Jo
Therefore, for I z I < 1,
F(a, b; c; z) = r(c) f (It
Hn
-
1
(1 - t)c-b-Izn dt
reb) r(c - b) n! J
o
r(c) (Itb-l(1 _ t)c-b-l dt.
reb) r(c - b)J
o
n!
The binomial theorem states that
co (-a)(-a - 1)(-a - 2) (-a - n + 1)(-I)nyn
(1 - y)-a= L , '
n.
which may be written
(1 _ y)-a = f a(a + 1)(a + 2)',' (a + n -
n.
Therefore, in factorial function notation,
(1 - y)-a = f
n.
which we use with y = zt to obtain the following result.
THEOREM 16. If Izl < 1 and if Re(c) > Re(b) > 0,
F(a b' c z) = r(c) (Itb-l(1 - t)c-b-l(l - tz)-a dt
, " r(b)r(c - b)J
o
.
48 THE HYPERGEOMETRIC FUNCTION [Ch.4
31. F(a, b; c; 1) as a function of the parameters. We kno\\"
already that if c is neither zero nor a negative integer and if
Re(c - a - b) > 0, the series
(1) F(a, b; c; 1)
is absolutely convergent.
Let 0 be any positive number. We shall show that in the region
Re(c - a - b) ;;:; 20 > 0, the series (1) for F(a, b; c; 1) is uni-
formly convergent. To fix the ideas, it may be desirable to think
of Re(c - a - b) ;;:; 20 > as a region in the c-plane, \vith a and
b chosen first. It is not necessary to look on the region in that way.
The series of positive constants
(2)
is convergent because 0 > 0. We show that for n sufficiently large
and for all a, b, c in the region Re(c - a - b) ;;:; 20 > 0, with c
neither zero nor a negative integer,
(3)
Now (see page 46)
Lim I I
,,+00 (C)nn.
= I r( c) I Lim 1 __ 1 1 =
rea) reb) n+
oo
n
c
-
a
-
b
-
o
,
since Re(c - a - b - 0) ;;:; 20 - 0 = 0 > 0. Hence (3) is true
for n sufficiently large, and the Weierstrass M-test can be applied
to the series in equation (1).
THEOREM 17. If c is neither zero nor a negative 7nteger and
Re(c - a - b) > 0, F(a, b; c; 1) is an analytic function of a, b, c.
32. Evaluation of F(a, b; c; 1). If Re(c - a - b) > 0,
Theorem 17 permits us to extend the integral form for F(a, b; c; z),
page 47, to the point z = 1 in the following manner. Since
Re(c - a - b) > 0, we may write
F(a, b; c; 1) = i:
(c)nn.
If we also stipulate that Re(c) > Rc(b) > 0, it follows by the
technique of Section 30 that
32] EVALUATION OF F(a, b; c; 1) 49
F(a, b; c; 1) = r(c) f - t)c-b-l dt
reb) r(c - b) n! 0
= r(c) i
1tH
(1 _ t)C-H(1 _ t)-a dt.
reb) r(c - b) 0
Therefore, if Re(c - a - b) > 0, if Re(c) > Re(b) > 0, and since
c is neither zero nor a negative integer,
F(a b c 1) = --. __ -- -f.
l
/
b
-
l
(1 - t)- c-a-b-l dt
, " I'(b)r(c - b) "
r(c) reb) r(c - a - b)
= r(b)f(c - b)
r(c)r(c - a - b)
= r(c-"=--a)r(c =--b)'
We now resort to Theorem 17 and analytic continuation to con-
clude that the foregoing evaluation of F(a, b; c; 1) is valid without
the condition Re(c) > Re(b) > 0.
THEOREM 18. If Re(c - a - b) > 0 and If c is neither zero nor
a negative integer,
r(c)r(c - a - b)
F(a, b; c; 1) = r(c--=:-a) r(c _ b)'
The value of F(a, b; c; 1) will playa \'ital role in many of the
results to be obtained in this and later chapters. Theorem 18 can
be proved without the aid of the integral in Theorem 16. For
such a proof see -Whittaker and Watson [1;281-282].
EXAMPLE: Show that if Re(b) > 0 and if n is a non-negative
integer,
[
-!n, -!n + L
F
b + t;
By Theorem 18 we get
[
-1.n -1.n +1..
2, 2 2,
F
b +!;
1] = 2
n
(b)n.
(2b)n
reb + !)r(b + n)
--'----,-
reb + !n) reb + + !)
(b)n reb) reb + D
reb + !n) reb + +
50 THE HYPERGEOMETRIC FUNCTION [Ch. 4
Legendre's duplication formula for the Gamma function, page 24,
yields
reb) reb + ~ ) = 2
1
-
2b
\/;r(2b),
reb + ~ n ) reb + ~ n + ) = 2
1
-
2b
-
n
y'; r(2b + n).
Therefore
F[-!n, -!n + ~ ;
b +!;
as desired.
33. The contiguous function relations. Gauss defined as con-
tiguous to F(a, b; c; z) each of the six functions obtained by increas-
ing or decreasing one of the parameters by unity. For simplicity in
printing, we use the notations
(1)
(2)
(3)
F = F(a, b; c; z),
F(a+) = F(a + 1, b; c; z),
F(a-) = F(a - 1, b; c; z),
together with similar notations F(b+), F(b-), F(c+), F(c-) for
the other four of the six functions contiguous to F.
Gauss proved, and we shall follow his technique, that between F
and any two of its contiguous functions, there exists a linear relation
with coefficients at most linear in z. The proof is one of remarkable
directness; we prove that the relations exist by obtaining them.
There are, of course, fifteen (six things taken two at a time) such
relations.
Put
so that
(4)
and
F(a+) = i: t ~ ) n On.
, , ~ O (a)n
Since a(a + l)n = (a + n) (a)n, we may write the SIX functions
contiguous to F in the form
33] THE CONTIGUOUS FUNCTION RELA TIONS 51
(5)
en a-I
F(a-) = L ----- On,
a-I + n
en b - 1
F(b-) = L ---- On,
b - 1 + n
co c
F(c+) = L --+ On,
C n

F(c-) = ---- On.
II-,(j C - 1
We also employ the differential operator 0 = operntor
has the particularly pleasant property that OZ" = nz", thus making
it handy to use on power series.
Since
co
(6) (0 + a)F = L (a + n) On,
1<..=.0
it can be seen with the aid of (5) that
(7)
(8)
(9)
(0 + a)F = aF(a+),
(0 + b)F = bF(b+),
(0 + c - l)F = (c - l)F(c-).
From (7), (8), and (9), it follows at once that
(10)
and
(a - b)F = aF(a+) - bF(b+),
(11) (a - c + l)F = aF(a+) - (c - l)F(c -),
two of the simplest of the contiguous function relations.
K ext consider
of = f n(at:Cb)nzn = z f (a)n+l(b)n+IZ
n
n=! (C)nn! n=O (C)n+ln! '
from which
(12)
OF = z f (a + n)(b + n) on.
n=O C + n
Now
(a + _n)jb + n) = n + (a + b _ c) + (c - a)(c - b).
c+n c+n
Hence equation (12) yields
52 THE HYPERGEOMETRIC FUNCTION [Ch.4
00 en (c-a)(c-b)zOO c
of = z L: n On + (a + b - c)z L: On + L: -+ On,
n_O C n=OC n
or
(13) (1 - z)oF = (a + b - c)zF + C-l(C - a)(c - b)zF(c+).
From (7) we obtain
(1 - z)eF = -a(I - z)F + a(1 - z)F(a+),
which combines with (I3) to yield another of the contiguous function
relations,
(14) [(I-z)a+(a+b-c)z]F = a(1-z)F(a+)
-cl(c-a)(c-b)zF(c+ ).
The coefficient of F on the left in (14) is in a form desirable for
certain later developments. Equation (14) may be replaced by
(15) [a+(b-c)zJF = a(I-z)F(a+) -cl(c-a)(c-b)zF(c+).
Next let us operate with e on the series defining F(a-). We thus
obtain
or
(16)
But
OF(a-) = f: (a( )-( = f: (a - I)+l(b),n+IZ
n
+
l
,
C n n - 1 . (c n+ln.
COb+n
eF(a-) = (a - I)z On.
C + n
b+n_
1
c-b
c + n - - c + n'
so that (16) becomes
eF(a-) = (a - I)z f: On - (a - I)(c - b)z f: _C_ On,
n=O C n=O C + n
which, in view of (5), yields
(17) OF(a-) = (a - l)zF - c
1
(a - l)(c - b)zF(c+).
We return to (7) and replace a by (a - 1) in it to get
(18) eF(a - ) = (a - I)F - (a - I)F(a -).
From (17) and (18) it follows that
(19) (1 - z)F = F(a -) - CI(C - b)zF(c+).
34] DIFFERENTIAL EQUATION 53
Similarly, since a and b may be interchanged without affecting the
hypergeometric series, we may write
(20) (1 - z)F = F(b-) - el(e - a)zFCe+).
We now have five contiguous function relations, (9) and (20)
together with
(10)
(11)
and
(a - b)F = aF(a+) - bF(b+),
(a - e + l)F = aF(a+) - (e - l)F(c-),
(15) [a+ (b-e)zJF = a(l-z)F(a+) -c-
1
(e-a) (e-b)zF(e+).
From these five relations the remaining ten may be obtained by
performing suitable eliminations. See Ex. 21 at the end of this
chapter.
34. The hypergeometric differential equation. The operator
8 = z(!), already used in the derivation of the contiguous function
relations, is helpful in deriving a differential equation tiatisfied by
(1)
From (1) we obtain
(
+
1)
n(n + c - l)(a)n(b)nz"
00 c- W=L- I
(C)nn.
A shift of index yields
= 1: (a)n(b)n
zn
.
(C)n-l(n - I)!
0(0 + C - l)w = i: (a)n+l(b)n+Izn+I
(C)nn!
= Z i: (a + n)(b + n)(a)n(b)n
zn
n=O (C)nn!
= z(O + a)(O + b)w.
We have shown that w = F(a, b; c; z) is a solution of the differen-
tial equation
(2) [8(0 + C - 1) - z(o + a)(O + b)]w = O.
d
0= z-'
dz
54 THE HYPERGEOMETRIC FUNCTION [Ch.4
Equation (2) is easily put in the form
(3) z(l - z)w" + [c - (a + b + l)z]w' - abw = 0
by employing the relations Ow = zw' and o( 0 - l)w = Z2W".
The second-order linear differential equation (3) is treated in
many texts* and therefore we omit details here. In order to avoid
tedious repetition, \ve shall, in this chapter only, refer to the text
mentioned in the footnote as IDE.
In IDE, pages 144-148, it is shown that if c is nonintegral, two
linearly independent solutions of (3) in I z I < 1 are
(4) WI = F(a, b; c; z)
and
(5) W2 = zI-cF(a + 1 - c, b + 1 - c; 2 - c; z).
We shall also make free use of Kummer's 24 solutions of equation
(3) as listed in IDE, pages 157-158. In any specific instance,
however, previous knowledge of Kummer's 24 solutions is not
necessary; the desired solution ean be obtained directly with the
aid of simple changes of variable performed on the differential
equation (3). See Ex. 12 at the end of this chapter.
35. Logarithmic solutions of the hypergeometric equation. If c
is not an integer, the hypergeometric equation
(1) z(l - z)w" + [c - (a + b + l)z]w' - abw = 0
always has in I z I < 1 the two power series solutions (4) and (5)
of the preceding section. If c is an integer, one solution mayor
may not, depending on the values of a and b, become logarithmic.
In this book we are primarily interested in the functions rather
than in the differential equations. We shall, whenever it is feasible,
avoid discussion of logarithmic solution::;.
If c is a positive integer and neither a nor b is an integer, two
linearly independent solutions of (1) are as listed below. These
solutions may be obtained by standard elementary techniques (see
Rainville [lJ, Sections 92 and 94), and the details are therefore
omitted here.
If c = 1 and neither a nor b is zero or a negative integer, two
linearly independent solutions of (1) valid in 0 < I z I < 1 are
*See, for example, Chapter 6 of Rainville [21.
36] F(a,b;e;z) AS A FUNCTION OF ITS PARAMETERS 55
(2)
(3)
in which
(4)
WI = F(a, b; 1; z),
W3 = F(a, b; 1; z)log z
ro (a)n(b)n
zn
+ (nT)2-1H(a, n) + H(b, n) - 2H(I, n) },
n 1
II (a, n) = L 1 '
a + k -
including the ordinary harmonic sum H (1, 11,) = H n'
If e is an integer> 1 and neither a nor b is an integer, two linearly
independent solutions of (1) yalid in < I z I < 1 are
(5)
and
(6)
WI = F(a, b; e; z)
W4 = F(a, b; e; z)log z
ro (a)n(b)n
zn
+ (e)nn! lH(a, n) + H(b, n) - H(e, n) - H(I, n) l
0-2 "(1')
_ L n. - ( ,,+1
(1 - a),,+1(1 - b),,+IZ
n
+
1
If e is an integer, e 0, equation (1) may be transformed by
using W = zI-cy into a hypergeometric equation for y ,,,ith new
parameters a' = a + 1 - e, b' = b + 1 - e, and e' = 2 - c. If
neither a' nor b' is an integer, the y-equation can be solved by using
(5) and (6).
36. F(a, b; c; z) as a function of its parameters. We han
already noted that the series in
(1) F(a b' e' z) = f
, " (e)nn!
is absolutely comugent (ratio test) for I z I < 1, independent of
the choice of a, b, e as long as e is neither zero nor a negati"e in-
teger. Recall that (e)n = r(e + n) free). Consider the function
(2)
F(a,- b; e; il = i: (a),,(b)nz"
r(e) r(e + n)n!'
in which the possibility of zero denominators on the right has been
removed. In any closed region in the finite a, b, and e planes,
56 THE HYPERGEOMETRIC FUNCTION [Ch. 4-
L' \ (a)n (b)n (n - 1) Inc zln I
n ! ~ (n - 1) Ina (n - 1) !n
b
r(c + n) n
1
+
c
-
a
-
b
1 . I z!n I
= f(a)r(b) ~ ! ~ nl+c-a-b = 0,
for I z I < 1.
Therefore, for any fixed z in I z I < 1, there exists a constant K
independent of a, b, c and such that
ro
I
(a)n(b)nzn I < K Izl In.
r(c + n)n!
Since L K Iz I In is convergent, the series on the right in (2) is ab-
n=O
solutely and uniformly convergent for all finite a, b, c as long as
Izl < 1.
We know the location and character of the singular points of
r(c) and are now able to stipulate the behavior, with regard to
analyticity, of the hypergeometric function with z fixed, I z I < 1,
and a, b, c as variables.
THEOREM 19. For I z I < 1 the function F(a, b; c; z) is analytic
in a, b, and cfor all finite a, b, and c except for simple poles at c = zero
and c = each negative 1"n teger.
Reference to Theorem 19 will enable us to simplify many proofs
in later work.
37. Elementary series manipulations. A common tool to be
used in much of our later work is the rearrangement of terms in
iterated series. Here \ve prove two basic lemmas of the kind
needed. "When convergent pmver series are involved, the infinite
rearrangements can be justified in the elementary sense. In our
study of generating functions of sets of polynomials, we sometimes
deal with divergent power series. For such series the identities of
this section may be considered as purely formal, but we shall find
that the manipulative techniques are fully as useful as when applied
to convergent series.
Lemma 10.
ro n
(1)
L LA(k, n) L L A(k, n - k),
n=O k=O Tt=f) 1,:=0
37] ELEMENTARY SERIES MANIPULATIONS 57
and
co n
(2) L L B(k, n) L L B(k, n + k).
n=O k=O " ~ O k ~ O
Proof: Consider the series
co co
(3) L L A(k, n)tnH
n ~ O k ~ O
in which t,,+k has been inserted for convenience and will be removed
later by placing t = 1. Let UR collect the powers of t in (3) . We
introduce new indices of summation j and m by
(4) k = j, n = m -),
so that the exponent (n + k) in (3) becomes m. The old indices
nand k in (3) are restricted, as indicated in the summation symbol,
by the inequalities
(5) n ~ 0, k ~ o.
Because of (4) the inequalities (5) become
m - j ~ 0, j ~ 0,
or 0 ~ j ~ m with m, the exponent on t, restricted only in that it
must be a non-negati\"e integer. Thus we arrive at
m m
(6) L L A(k, n)tn+k = L L A(j, m - j)t
m
,
n=O k=O m=O )=0
and the identity (1) of Lemma 10 follows by putting t = 1 and
replacing the dummy indices j and m on the right by dummy
indices k and n.
There is no need to use k and n for indices in both members of
(1), but neither is there harm in it once a small degree of sophistica-
tion is acquired. We frequently employ many parameters and
prefer to keep to a minimum the number of different symbols used.
In Lemma 10, equation (2) is merely (1) written in reverse; it
needs no separate deri\"ation.
Lemma 11.
00 [n /2)
(7) L L A(k, n) L LA (k, n - 2k),
n ~ O k_O n=U J...=O
and
co [n /2)
(8) L L B(k, n) L L B(k, n + 2k).
n=O k ~ O ,/"",,1) A. =0
58 THE HYPERGEOMETRIC FUNCTION [Ch. 4-
Proof: Consider the series
(9) L LA(k, n)tn+2k
TI=(J k=O
and in it collect pO'wers of t, introducing new indices by
(10) k = j, n = m - 2j,
so that n + 2k = m. Since
(11) n 0, k 0,
we conclude that
m - 2.1 0, j 0,
from which 2j m and m 0. Since .i and j is
integral, the index j runs from to the greatest integer in !m. Thus
we obtain
'Xl
(12) L L A(k, n)t
n
+2k = L LA (j, m - 2j)t
m
,
n=O k=O m"",Q }-'-'O
from which (7) follO\vs by placing t = 1 and mnking the proper
change of letters for the dummy indices on the right in (12). Equa-
tion (8) is (7) written in reverse order.
There is no bound to the number of such identities. The reader
should now be able to obtain whatever lemmas he needs along these
lines.
Kote also that a combination of Lemmas 10 and 11 yields
co n ro [n/2]
(13) L L C(k, n) = L L C(k, n - k).
n=O k=O n=O k=O
38. Simple transformations. It is convenient for us to write
the ordinary binomial expansion with the factorial function notation,
(1)
(1 - z)-a = f
1/=0 n.
and to recall the result of Ex. 8, page 32,
(2) o k n.
In particular a = 1 in (2) yields
(3) Cn - k)! = (-I)k
n
!,
(-n) k
o k n.
38] SIMPLE TRANSFORMATIONS
Consider now the product
[
a, c - b;
(1 - z)-aF
c;
With the aid of (1) we may write
59
(l_z)-a
F
[a, c-b; _-z] = f f (a)k(c-b)k(a+'")n( -l)kznH.
1 - Z h 0 (c) k k! n!
c;
Now (a)k(a + k)n = (a)"+k, so that
--a "[a, c - b; -z] -0 m (c - b)k(a)n+k( -l)kznH
(1 - z) F ---- = L L ----------
1 - Z hO (c)kk!n!
c ,
and we collect powers of z to obtain
by (3). The inner sum on the right is a terminating hypergeometric
senes. Hence
(1 _ z)_aF[a, c - b; = f F[-n, c - b;
1 - z n!
c; c;
Since F( -n, c - b; c; 1) terminates, we may write
(1 _ z)_aF[a, c - b; = f r(c) reb + n)(a)"zn
1 - z r(c + n) r(b)n!
c;
= F(a, b; c; z),
a result valid where both JzJ < 1 and Jzj(l - z) J < 1 (for which
see Figure 3, page 60).
60
THE HYPERGEOMETRIC FUNCTION (Ch . .4
THEOREM 20. If [z[ < 1 and [z/(l - z) [ < 1,
(4) F[a, b;
c' ,
] [
a, c - b;
z = (1 - z) -a F c;
-z ]
1 - z
F[a, c - b;
c' ,
I
: z-plane
I
I
Figure 3
The roles of a and b may be in-
terchanged in (4).
The type of series manipula-
tions in vol ved above in arriving
at the identity (4) will be used
frequently throughout the re-
mainder of this book, and such
steps will be taken hereafter
without detailed explanation.
Let us use Theorem 20 on
the hypergeometric function on
the right in (4). Put
-z
y=--'
1 - z
Then
] [
c - a c - b'
y = (1 - y)-cHF ' ,
C" ,
-y ]
1 - y
But 1 - y = (1 - Z)-l and -y/(l - y) = z. Hence
[
a c - b' ] [C - a c - b'
, , -z ' ,
F -- = (1 - z)c-bF
1 - z
c; c;
which combines with Theorem 20 to yield the following result due
to Euler.
THEOREM 21. If [z [ < 1,
(5) F(a, b; c; z) = (1 - z)c-a-bF(c - a, c - b; c; z).
The identities in Theorems 20 and 21 arc statements of equality
among certain of the 24 Kummer solutions of the hypergeometric
differential equation. In the terminology of IDE, pages 1;')7-158,
we have shown that IlIa = 'Va = Illb. Alternate proofs of
Theorems 20 and 21, making use of the differential equation, are
left as exercises.
39] RELATION BETWEEN FUNCTIONS OF Z, l-Z 61
39. Relation between functions of z and 1 - z. The hyper-
geometric differential equation
(1) z(1 - z)w" + [c - (a + b + l)zJw' - abw = 0
has, in 11 - z I < I, the solution
(2) w = F(a, b; a + b + 1 - c; 1 - z),
as indicated in IDE, page 157, formula IVa. The solution (2) can
be obtained independently by placing z = I - !J in the differential
equation (1) and observing that the transformed equation is also
hypergeometric \vith parameters a' = a, b' = b, c' = a + b + 1 - c,
and argument y = 1 - z.
We already know that in I z I < I, the equation (1) has the linearly
independent solutions
(3) WI = F(a, b; c; z),
(4) W2 = zl-cF(a + 1 - c, b + 1 - c; 2 - c; z).
Then there must exist constants A and B such that
(5) F(a, b; a + b + 1 - c; 1 - z) = AF(a, b; c; z)
+Bzl-cF(a + 1 - c, b + 1 - c; 2 - c; z)
z-plane
2
Figure 4
is an identity in the region (Figure' 4) where both I z I < 1 and
11 - zl < 1. If we insist that Re(1 - c) > 0 and let z ~ O f r o m
within the pertinent region, (5) yields
F(a, b; a + b + 1 - c; 1) = A 1 + B . 0,
62 THE HYPERGEOMETRIC FUNCTION [Ch.4
from which, by Theorem 18, page 49,
(6)
A = rea + b + 1 - c)r(1 - c).
rea + 1 - c) reb + 1 - c)
Again from (5) if we let z - 1 from inside the region and insist that
Re(c - a - b) > 0, we obtain
1 = AF(a, b; c; 1) + BF(a + 1 - c, b + 1 - c; 2 - c; 1),
or, by Theorem 18,
(7)
r(c) r(c - a - b)A + r(2 - cU'(c - a _-=b)B = 1
r(c - a) r(c - b) r(I - a) r(1 - b) .
Employing (6) in (7) leads to
(8)
r(2 - c)r(c - a - b)B
r(I - a)r(I - b)
= 1 _ rea + b
rea + 1 - c) r(b + 1 - c) r(c - a) r(c - b)
In Ex. 15, page 32, we showed that the right member of (8) IS
equal to
(9)
r(2 - c) r(c - 1) r(c - a - b) rea + b + 1 - c)
rea) r(1 - a) reb) r(I - b)
From (8) and (9) we obtain
(10)
B = rea + b + 1 - c) r(c - 1)
rea) reb) ,
which completes the proof of the follO\ving statement.
THEOREM 22. If Izl < 1 and 11 - zl < 1, 1f Re(c) < 1 and
Re (c - a - b) > 0, and if none of a, b, c, c - a, c - b, c - a - b
is an integer,
(11) F(a, b; a + b + 1 - c; 1 - z)
rea + b + 1 - c) r(I - c)
= rea + 1 - c) reb + 1 _ c) . F(a, b; c; z)
rea + b + 1 - c) r(c - 1) 1- .
+ rea) reb) . z cF(a + 1 - c, b + 1 - c, 2 - c, z).
The restrictions on a, b, c can be relaxed somewhat, if desired.
The expression of F(a, b; c; z) as a linear combination of functions
40J A QUADRATIC TRANSFORMATION 63
of (1 - z) is left as an exercise; see Ex. 15 at the end of this chapter.
Theorems 20, 21, and 22 exhibit only three of the numerous rela-
tions among the 24 Kummer solutions. For other such relations
see volume one of the Bateman Manuscript Project work, Erdelyi
[1; 106-109].
40. A quadratic transformation. In any detailed study of the
hypergeometric differential equation
(1) z(I - z)w" + [c - (a + b + I)zJw' - abw = 0,
the derivation of the 24 Kummer solutions is a natural result of the
study of transformations of equation (1) into itself under linear
fractional transformations on the independent variable. It is rea-
sonable to attempt to use a quadratic transformation on the in-
dependent \'ariable for the same purpose. Such a study shows that
the parameters a, b, c need to be related for the new equation to be
of hypergeometric character. Since differential equations are not
our primary interest, we bypass the fairly simple determination of
all such quadratic transformations and corresponding relations
among a, b, and c. Here we move directly to the particular change
of variables which leads to the relation we need for our later work.
In equation (1) put c = 2b to get
(2) z(1 - z)w" + [2b - (a + b + l)zJw' - abw = 0,
of which one solution is w = F(a, b; 2b; z). Next let
4x
(3) z = U+-X)2 '
and obtain, after the usual labor invoh'ed in changing independent
variables, the equation
(4)
.dw
x(I - x)(1 + x)2dx2 + 2(1 + x)(b - 2ax + bx- - x-)cTx
- 4ab(1 - x)w = 0,
of which one solution is therefore
(5)
[
a, b;
w=F
2b;
4x ]
(1 + X)2
In (4) put w = (1 + x)2ay to obtain the equation
(6) x(1-x
2
)y" +2[b- (2a-b+ - 2ax(1 +2a- 2b)y = 0,
64 THE HYPERGEOMETRIC FUNCTION (Ch.4
of which one solution is
(7)
[
a, b;
Y = (1 + x)-2aF
2b;
4x ]
(1 + X)2
The differential equation (6) is im'ariant under a change from x
to (-x). Hence we introduce a new independent yariable v = x
2

The equation in y and v is found to be
(8) v(l-v) =
dv
2
2 2 dv 2'
which has, in I v I < 1, the general solution
(9) Y = AF v + BvHF v .
[
a, a - b + ] [ a - b + a + 1 - 2b; ]
b + - b;
We now haye the following situation. The differential equation
(6) has a solution (7) valid in
! 4:c I
< 1
! (1 + X)2
as long as 2b is neither zero nor a negatiye integer. At the same
time, equation (6) has the general solution* (9) with v = this
solution valid in I x I < l.
Therefore, if both Ix I < 1 and i (1 x)2i < 1 and if 2b is neither
zero nor a negath'e integer, there exist constants A and B such that
(10)
[
a, b;
(1 + x)-2aF
2b;
4x ] [a, a - b +
----- = AF
(1 + xF b + 1.
2,
[
a - b +
+ BX
l
-
2b
F
La + 1 - 2b;
- b;
In (10) the left member and the first term on the right are analytic
at x = 0, but the last term is not analytic at x = 0 because of the
*If 2b is a positive integer, the second term on the right in (9) mayor may not need
to be replaced by a logarithmie solution. If such a 10j.(arithmie solution is involved in
(9), reasoning parallel to that follo\\"inj.( equation (10) sho\\"s aj.(ain that B = O.
11 1 OTHER QUADRATIC TRANSFORMATIONS 65
factor Xl-Zb. Hence B = 0 and A is easily determined by using
x = 0 in the resultant identity
(11)
(1 + x)_zaF[a, b; = AF[a, a - b Xz].
2b; (1 + x) b + L
Thus A = 1, and we are led to the following result due to Gauss.
THEOREM 23. If 2b is neither zero nor a negative integer, and 1j
both Ixl < 1 and 14x(1 + X)-2! < 1,
(12) (1 + x)_zaF[a, b; = F[a, a - b X
z
].
?b . (
1
+ x) b + 1
- , 2,
41. Other quadratic transformations. For Yaripty of technique
we shall now proye the following theorem without reeourse to the
differential equation.
THEOREM 24. If 2b is neither zero nor a negative 1'/lteger and 1j
I y I < ! and I y / (1 - y) I < 1,
(1)
[
!a, + 2] [a, b; ]
(1 - y)-aF __ JL_--;, = F 2y .
b+1. (l-y)- 2b'
2, ,
Proof: Let if; denote the left member of (1). Then
+ (a)2kYZk
if; = 6 (b + - y)a+Zk = BJ 2
2
-
k
(b - y)a+Zk
with the aid of Lemma 5, page 22. Also
(1 - y)-a-Zk = f (a + 2,k)n
yn
n=O n.
Using Lemma 11, page 57, we may collect powers of y and obtain
00 [n/2] (a)nyn
if; = ?; {;o 2Zk(b + - 2k)1
We know that (n - 2k)! = nl/( -n)2k and that
66 THE HYPERGEOMETRIC FUNCTION [Ch . .4
(-n)2k = 22k( - k( - + k.
Therefore we have
[
- J_
n
-!n + !.
CD 2, 2 2,
= 'LF
b + !;
In the example on page 49 we found that the terminating hyper-
geometric function above has the value 2 nCb) ,,/ (2b) n' Hence
m ') (b) ( ) [a, b;
if; = 'L _n n a nyn = F
(2b)nn! 2b-
,
which completes the proof of Theorem 24.
In Theorem 24 put y = 2x/(1 + X)2. Then
1 + X2
1 - y = (1 -+ x) 2'

1 - Y
2x
and we may write
(1+x
2
)-a(1+x)2aF = F

42] [a,b; 4X]


(I+x) 2b; (1+X)2
In view of Theorem 23 we may now conclude that
2 -a [!a,!a + !; 4X2] [a, a - b + !; X2].
(1 + x) F (1 + X2)2 F
b+L b+L
Now put X2 = z and replace b by (! + a - b) to obtain
(1 + + 4Z] = F[ a, b; z].
(1 + z)2
1 + a - b; 1 + a - b;
(2)
By Theorem 20, page GO, with appropriate sub:,;titutions for the
a, b, c and z of the theorem,
(1--
4
:),]-
41] OTHER QUADRATIC TRANSFORMATIONS 67
We may therefore rewrite (2) in the form
[
!a, ! + !a - b; -4Z] [ a, b; ]
(3) (1 - z)-aF (1 _ Z)2 = F z ,
1+a-b; 1+a-b;
which will be useful in Section 42.
Let us next return to the difTerential equation to establish one
more relation involving a quadratic transformation.
THEOREM 25. If a + b + ! is neither zero nor a negative 1:nteger,
and if Ixl < 1 and 14x(1 - x) I < 1,
[
a,b; ] [ '2a,2b;
F 4x(1 - x) = F
a + b + a + b + L
(4)
The function
[
a, b;
y=F
a + b + L
(5)
is a solution of the differential equation
d
2
y 1 dy
(6) z(1 - z) dz
2
+ [a + b + 2 - (a + b + l)z] dz - aby = O.
In (6) put z = 4x(1 - x), and with some labor thus obtain
(7) x(l-x)y" + - (2a+2b+ l)x]y' -4aby = O.
Equation (7) is hypergeometric in character and has the general
solution
[
2a, 2b; ] [t + a - b, ! + b - a; ]
y = AF x + x ,
a + b + L - a - b;
as well as the solution
[
a, b;
y=F
a + b + L
4x(1 xl]
from (5) above. By the usual argument it is easy to conclude the
validity of (4).
68 THE HYPERGEOMETRIC FUNCTION [Ch.4
42. A theorem due to Kummer. Let us return to equation (3)
of the preceding section and let z ---> -1. Thc result is
[
.!a .! + Ia - b'
2 , 2 2 ,
2-
a
F
1 + a - b;
a, b;
+ a - b;
We can sum the series on the left and thus obtain
(1)
{
a,b;
1 1 + a _ b;
]
1'(1 + a -
-1
Legendre's duplication formula, page 24, yields
1'(!)1'(1 + a) = 2arO + +!a),
which may be used on the right in (1).
THEOREM 26. If (1 + a - b) 7'S neither zero nor a negative
integer, and Re(b) < 1 for convergence,
[
a, b;
(2) F
1 + a - b;
-1] = 1'(1 + a - b) 1'(1 +
1'(1 + - b)1'(l + a)
43. Additional properties. Further results applying to special
hypergeometric functions appear in later chapters, where we shall
find that the polynomials of Legendre, .Jacobi, Gegenbauer, and
others are terminating hypergeometric series.
\Ve now obtain one more identity as an example of those resulting
from combinations of the theorems proyed earlier in this chapter.
In the identity of Theorem 25, page G7, replace a by (!c - and
b by (!c + - D to get
[
.!c - .!a .!c + .!a - J. ] [C - a c + a-I' ] 2 2,2 2 2, "
F 4x (1 - x) = F x .
c; c;
Theorem 21, page GO, yields
F[C - a, c + a-I;
c ,
] [
a, 1 - a;
x = (1 - x)l-cF c;
which leads to the desired result.
431 ADDITIONAL PROPERTIES 69
THEOREM 27. If c is neither zero nor a negative integer, and 1f both
Ixl < 1 and 14x(1 - x)! < 1,
F[
a, 1 - a; ] {!c - !a, !c + !a - !;
x = (1 - x)c-1F
c; c;
EXERCISES
1. Show that
d [a, b;
-F
dx
c ,
x] = a: F[a + 1, b + 1;
c + 1;
2. Show that
F[ 2a,2b;
a + b+ t
3. Show that
[
a 1 - a' , ,
F
c ,
4. Obtain the result
F[-n, b;
c ,
5. Obtain the result
6. Show that
F[-n, a + n;
c ,
F
[
-n 1 -b - no , ,
a-,
IJ rca + b + ) I ' ( ~ )
2 = rca + 1YRb + !).
1 = ----------.
]
( - 1 )"(1 + a - C)n
(e)"
1 = - .. - .. -. .- ._-.
J
(ft + Ii -- Ih"
(o),.(a + I, - 1)"
4xil - xl J.
7. Prove that if (fn = F( -n, a; 1 + a - n; 1) anu a iii not an integer, then
gn = 0 for n ~ 1, go = 1.
8. ShO\v that
d
n
dxJxa-1+nF(a, b; c; x)] = (a)nxa-1F(a + n, b; c; x).
9. Lse equation (2), page 66, with z = -x, b = -n, ill whieh n i ~ a nOll-
negatiYe integer, to conclude that
70
THE HYPERGEOMETRIC FUNCTION [Ch. 4
[
-n, a;
F 1 + a + n;
]
+ !;
-x = (1 - x)-aF
1 + a + n;
10. In Theorem 23, page 65, put b = 1', a = I' + L 4x(1 + x)-Z = z and thus
prove that
J
[
J
2"(_1
Z = (1 - z)-!
1+'\I
1
-
z
and further tha t
[
1.
F 1',1' - 2,
21';
11. Use Theorem 27, pagc 69, to show that
[
a,l - a; ] - 1
C
- 4 ( )]
(1 - x)l-CF x = (1 - 2x)a-cF -.:: x -..!.- .
(l - 2x)2
c; c;
12. In the differential equation (3), page 54, for
w = F(a, b; c; z)
introduce a new dependent yariablc '11 by '11' = (1 - z)-au, thus obtaining
z(1 - Z)2U" + (I - z)[c + (a - b - l)zJu' + a(e - b)u = 0.
l\ext change the independent variable to x by putting x = -z/(1 - z). Show
that the equation for '11 in tcrms of x is
d
2
u du
x(l - x) dx
2
+ [c - (a + c - b + l)x] dx - a(c - b)u = 0,
and thus derive the solution
[
a, c - b;
w = (1 - z)-aF
c-,
13. Use the result of Ex. 12 and thc method of Section 40 to pro\'C Theorem 20,
pagc 60.
14. Proye Theorem 21, pagc 60, by the mcthod by Exs. 12 ane! 13,
15. LEe the method of Section 39 to prove tha t if both I z I < 1 and 11 - z I < 1,
ane! if a, b, c are suitably re;;tricted,
F[a, b;
c ,
z] = b) F[
r(c - a)I'(e - b) a + b + 1
a, b;
- c;
1 - z].
43] ADDITIONAL PROPERTIES
16. In a common notation for the Laplace transform
L-ll!(S) I = F(t).
Show that
1
1 [ a, b;
L-l -F
s s + 1;
z(1 -- e
l
) J
17. With the notation of Ex. 16 show that
. al'(n + 2) [1 + +
Llt
n
sm atl = -----F
sn+2 .a .
2 ,
18. Obtain the results
Log (1 + x) = xF(I, 1; 2; -x),
Arcsin x = xFO, L X2),
Arctan x = xFO, _X2).
19. The complete elliptic integral of the first kind is
Show that K = !7l'-F(!, !; 1; k
2
).
20. The complete elliptic integral of the seconr-I kind is
E = .f" vr-=-k
2
sm
2
8 dO.
Show that E = !7l'-F(!, -!; 1; k
2
).
21. From the contiguous function relations
(1) (a - b)F = aF(a+) - bF(b+),
(2) (a - c + l)F = aF(a+) - (c - l)F(e-),
(3) [a + (b - c)zJF = aO - z)F(a+) - e-I(c - a)(c - b)zF(e+),
(4) (1 - z)F = F(a-) - e-1(e - b)zF(e+),
(5) (l - z)F = F(b-) - cl(e - a)zF(e+), derived in Seetion 33,
obtain the remaining ten such rela tions:
(6) [2a - e + (b - a)zJF = a(l - z)F(a+) - (e - a)F(a-),
(7) (a + b - e)F = aO - z)F(a+) - (c - b)F(b-),
(8) (c - a - b)F = (c - a)F(a-) - b(l - z)F(b+),
(9) (b - a)(1 - z)F = (e - a)F(a-) - (e - b)F(b-),
(10) [1 - a + (e - b - l)zJF = (e - a)F(a-) - (c - 1)0 - z)F(e-),
(11) [2b - e + (a - b)zJF = b(l - z)F(b+) - (e - b)F(b-),
(12) [b + (a - e)z]F = b(1 - z)F(b+) - cl(e - a)(e - b)zF(e+),
(13) (b - c + I)F = bF(b+) - (e - l)F(c-),
(14) [1 - b + (e - a - l)zJF = (c - b)F(b-) - (c - 1)(1 - z)F(c-),
71
(15) [c-l+(a+b+ 1-2c)z'F = (c-1)(1- z)F(c-) - - n(r- b\zF(c+i.
72 THE HYPERGEOlvlETRIC FUNCTION [Ch.4
22. The notation used in Ex. 21 and in Section 33 is often extended as in the
examples
F(a-, b+) = F(a - 1, b + 1; c; z),
F(b+, c+) = F(a, b + 1; c + 1; z).
U ~ e the relations (4) and (5) of Ex. 21 to obtain
F(a-) - F(b-) + c-1(b - a)zF(c+) = 0
and from it, by changing b to (b + 1) arrive at
F = F(a-, b+) + c1(b + 1 - a)zF(b+, c+),
a relation we wish to usc in Chapter 16.
23. In equation (9) of Ex. 21 shift b to (b + 1) to obtain the relation
(c - 1 - b)F = (c - a)F(a-, b+) + (a - 1 - b)(l - z)F(b+),
or
(c - 1 - b)F(a, b; c; z)
= (c - a)F(a - 1, b + 1; c; z) + (a - 1 - b)(l - z)F(a, b + 1; c; z),
another relation we wish to use in Chapter 16.
CHAPTER 5
Generalized
Hypergeometric
Functions
44. The functionpF q' The hypergeometric function
(1)
F[a, b;
c ,
studied in Chapter 4 has two numerator parameters, a and b, and
one denominator parameter, c. It is a natural generalization to
move from the definition (1) to a similar function with any number
of numerator and denominator parameters.
We define a generalized hypergeometric function by
"
en II (ai)n zn
(2)
1 + ~ - I - - 'I'
71=1 q n.
II ({3j)n
)= 1
in which no denominator parameter (3j is allowed to be zero or a
negative integer. If any numerator parameter O'i in (2) is zero or
a negative integer, the series terminates.
An application of the elementary ratio test to the power series
on the right in (2) shows at once that:
(a) If p ~ q, the series converges for all finite z;
(b) If p = q + 1, the series con verges for I z I < 1 and diverges
for I z I > 1;
73
74 GENERALIZED HYPERGEOll1ETRIC FUSCTIOSS [Ch.:j
(c) If p > q + 1, the series di \uges for z O. If the series
terminates, there is no question of convergence, and the conclusions
(b) and (c) do not apply.
If p = q + 1, the series in (2) is absolutely convergent on the
circle I z I = 1 if
The proof Can be made to parallel that used in Section 29 for the
corresponding result on the ordinary hypergeometric series. The
proof is left as an exercise.
\Vhen we wish to indicate the number of numerator parameters
and of denominator parameters but not to specify them, we use the
notation pFq. For instance, the ordinary hypergeometric function
of Chapter 4 is a 2FI
\Ve permit p or q, or both, to be zero. The absence of parameters
is emphasized by a dash. The most general OFI, for example, is
en zrt
(3) OFl( -; b; z) =
\Ve shall see later that thc function in (3) is essentially a Bessel
function.
For more results on the pFq see [1; 182-247J and Bailey
[1]. Additional references may also be found on pages 103-108 of
Bailey [lJ and on pages 199-201 and 246-247 of Erdelyi [1].
45. The exponential and binomial functions. Two elementary
instances of the pFq follow. If no numerator or denominator para-
meters are present, the result is
m zn
(1) oFo( -; -,. z) = L = exp(z).
TI=f; Il.
If we use one numerator parameter and no denominator para-
meter, we obtain
(2)
Hence, by the argument on page 47,
(3) IFo(a; - ; z) = (1 - z)-a.
46. A differential equation. Recall that the ordinary hyper-
geometric function F(a, b; c; z) satisfies the differential equation
46] A DIFFERENTIAL EQUATION 75
(1)
d
2
w dw
z(l - z)- + [c - (a + b + l)zJ-- - abw = 0
dz
2
dz'
or, III terms of the differential operator 0 == z :z' the differential
equation
(2) [0(0 + c - 1) - z(o + a)(O + b)Jw = O.
With the suggestion of equation (2) before us, we can proceed
as follows. Let
w = F = f (al )k(a2)k"
p q hO (b
l
)k(b
2
)k" (bq)k k!
Since OZk = kzk, it follows that
q
o II (0 + bj -
j=1
Now we replace k by (k + 1) and have
1=1
p
= zIT (0 + ai)w.
1=1
Thus we have shown that w = pFq is a solution of the differential
equation
(3) [0 D (0 + bj - 1) - z n Co + a,) ] w = 0,
76 GENERALIZED HYPERGEOM FU.Y(TIO.YS [CII. 5
when no b
j
is a nonpositive integer. The solution is yalid for all
finite z when p q. If p = q + 1, the solution is \"alid in ! z! < 1.
We are not concerning ourselyes with the caRe p > q + 1, when the
series for pF q has a zero radius of convergence unless the series
terminates.
47. Other solutions of the differential equation. If p q + 1,
the equation (3) of the preceding section is a linear differential
equation of order (q + 1). We have, in the neighborhood of the
origin, one solution,
Naturally we wish to obtain q other solutions near z = O.
Let us turn for the moment to the one case, the 2Ft , about which
we already have quite a bit of information. The differential equa-
tion for p = 2, q = 1, is
(1) [0(0 + b
I
- 1) - z(O + al)(O + a2)]w = O.
We know that if b
I
is not integral, the general solution of (1) In
the neighborhood of z = 0 is
W = Awo + BW1,
where A and B are arbitrary constants and Wo and WI may be taken
to be
Wo = F(aI' a2; b1 ; z),
WI = zI-b,F(aI - b
l
+ 1, a2 - b
I
+ 1; 2 - b
l
; z).
When we have determined just what it is that makes the WI
sa tisfy equation (1), we shall be in a better position to mo\"e on to
the generalized case.
Now
so that
0(0 + bI - l)WI
f (1;+ I-b
l
)(k+ -b
1
+ 1) Ja2-b
I
+ l)k Zk+
I
-
b
,
hO (2-b
t
)k k!
f k(k + 1 - bI)(aI - b
I
+ 1)k(a2 - b
I
+ l)k /+I-b,
(2 - b
1
) k --y.r-.
4-7] SOLUTIONS OF DIFFERENTIAL EQUATION 77
But the last factor in (2 - b
l
) k is 2 - b
l
+ k - 1 = k + 1 - b
l
,
so that
8(8 + b
1
- 1)wI = i: (al - bl + I)k(a2 - bl + I)k l+l-b,
(2 - bl)k_1 (k - I)!
= i: - bl + l)k+l Zk+2-b,.
(2 - bl)k k!
The last factor in (al - b
1
+ l)k+1 is al - b1 + 1 + k + 1 - 1 =
k + 1 - bl + al. lIence
e(e + b
l
- l)wI
= i: (k+ 1-b
l
+al) (k+ 1-bl
+a2) (al -b
l
+ 1) k(a2 -bl + 1) k
(2 - blh k!
= (0 + )(0 + ) (al - bl + 1)k(a2 - bl + l)k
Z al a2 L..J (2 _ b ) k I
I k
= z(e + al)(O + a2)wI'
Therefore WI is a solution of equation (1).
Notice the way in which the operator e introduced a factor
(k + 1 - b
l
) which canceled the last factor in (2 - b
1
) k and the
way in which the operator (0 + b
l
- 1) introduced a factor k
which canceled the factor k in k!. If we were dealing with the
generalized case, other operators in the form (0 + bj - 1) would
occur. Such an operator would introduce a factor
k + 1 - b1 + bj - 1 = bj - b] + 1 + k - 1,
which is the last factor in the product (b, - b] + 1) k. Hence, in
the solutions we seek, a denominator parameter b
j
in the pFq should
be replaced by the parameter (b
j
- b
l
+ 1) for the :oolution cor-
responding to the parameter b
1
in the original pFq. Of cour:oe eaeh
of the b
j
should playa role like that of b
l
, eaeh yielding a different
solution of the differential equation.
Thus we are led to the de::sired generalization. That i:o, the
generalized hypergeometric equation,
(2) [0 g (0 + bj - 1) - Z (0 + aJJ 10 = 0; p q + 1,
has, when no b
j
is a nonpositi\'e integer and no two b
J
'::s differ by an
integer, the solution
78 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
(3)
where the em are arbitrary constants, where
and where, for rn = 1, 2, ... , q,
(4)
Wheneyer, in addition to the aboye restrictions, no b
j
is a positive
integer, then the linear combination (3) is the general solution of
equation (2) around z = o. Note abo that if p ~ q, then the series
for W
m
; rn = 0, 1, 2, ... , q, conyerges for all finite z and that for
p = q + 1, the series for Wm conwrges for I z I < 1.
Tt is important to realize that the procedure we are now using is
not necessary. All (q + 1) solutions of the differential equation (2)
can be obtained by the standard series methods, the method of
Frobenius, keeping in mind that, for p ~ q + 1, the equation (2)
has at worst a regular singular point at the origin. The whole
point to the method of the present section is an attempt to gain
some insight into the way in which the parameters in pFq must be
altered to preserye its property of being a solution of equation (2)
and, perhaps most of all, the method is an attempt to obtain some
small practice in the technique of manipulating generalized hyper-
geometric series.
We shall now wrify that Wm satisfies equation (2). Let, as
before, for rn anyone of the integers 1, 2, 3, ... , q,
(5)
47] SOLUTIONS OF DIFFERENTIAL EQUATION 79
Note that the mth factor in the product
is (l)k, which furnishes the factor k! needed to make the right side
of (5) a hypergeometric series.
From (5) we get at once
q
oll (0 + b
i
- I)Wm
i-I
q p
00 (k+I-bm) II (k+I-bm+bj -1) II (a;-bm+I)k
= L ____ ---')'-----=--1 __
q
II (bj - bm + I)k
Now (b
i
- b
m
+ k) is the last factor in the product (b j - bm + 1) k
and (k + 1 - b
m
) is the last factor in the product (2 - bm)k.
Hence
p
q 00 II (a; - bm + l)k Zk+l-bm
0tl (0 + b
j
- I)w
m
= (2 _ b
m
)k-l'
II (bi - bm + I)k-l

where the k = 0 term dropped out because of the numerator factor
k + 1 - b
m
+ b
m
- 1 = k, which was later canceled.
Next, with a shift of index from k to (k + 1), we get
p
q 00 II (a;-b m +1)k+l Zk+2-bm
oll (O+bj-I)w
m
= -c-c--:;---:--
q (2-b
m
)k
II (bj-bm+I)k
Therefore
q p
o II (0 + b
j
- I)wm = Z II (0 + a,)wm,
1=1
so that W
m
; m = 1,2, "', q, satisfies the generalized hypergeometric
equation.
80 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
If the exponents of the differential equation at z = 0 are such that
one or more of their differences are integral, then the presence of
logarithmic terms is to be expected in the general solution. For
p = q + 1 the logarithmic solutions are obtained in F. C. Smith [1].
48. The contiguous function relations. * In this Rection the
parameters are fixed, and the work is concerned only with the func-
tion pFq and its contiguous functions. Hence, we are able to use
an abbreviated notation illustrated hy the following:
(1) F = pFq(al, a2, "', a
1
,; {3l, {32, "', (3q; x),
F(al +) = pFq(al + 1, a2, "', a,,; {3l, {32, "', (3'1; x),
F({3l -) = pFq(al, a2, "', a,,; {31 - 1, {32, "', (3q; x).
There are, of course, (2p+2q) funetions contiguous to F. Corre-
sponding to Gauss' five independent relations in the case of 2F
I
, there
is for F a set of (2p+q) linearly independent relations, which we shall
obtain. The canonical form into which we put this basic set may be
described as follows:
First, there are (p+q-l) relations, each containing F and two of
its contiguous functions. These will be called the simple relations.
Each simple relation connects F, F(al+), and F(a k+) for k = 2,3,
.. " p, or it conneets F, F(al +), and F({3j-) for j = 1, 2, "', q.
The simple relations are immediate extensions of two of Gauss'
five relations and are not novel in any way.
Second, there are (p+ 1) less simple relations, each containing F
and (q+ 1) of its contiguous functions. In our canonical form we
shall select these so that one of them connects F, F(al+) and all the
functions F({3j+) for j = 1, 2, "', q. Each of the other p relations
will contain F, all the functions F(3j+); j = 1, 2, "', q, and one of
the functions F(a k-) for k = 1,2, "', p. The less simple relations
are generalizations of three of Gauss' five relations but differ from
them in one essential aspect in that each relation contains F and
(q + 1) of its contiguous functions. For Gauss' case q + 1 = 2,
and the less simple relations contain the same number of contiguous
functions as do the simple ones.
Since we shall actually exhibit the (2p+q) relations, it will be
evident upon looking at them that, just as in the case of the ordinary
hypergeometric function, the coefficients are polynomials at most
linear in x.
*This section is taken from an article (Rainville [3]) published in the Bulletin of the
American Mathematical Society. Reproduced here by permission.
48] THE CONTIGUOUS FUNCTION RELATIONS 81
It is convenient to use the following notations:
k-1
(2)
IT As = IT As' IT As,
s-l,(k)
a symbol denoting a product \yith a particular factor deleted,
p
IT (a" - {3J
(3)
U
j
= _
q
{3; II ((3, - (3;)
,(J)
(4)
(3) II ((3. - (3j)

(al),,(a2),,' .. (ap)n
en = -- ,
((31),,((32),,' .. ((3q)n
(5)
(6)
S = (a1 + n)(a2 + rz) .. (al' + n)
n ((31 + n)((32 + n) .. ((3q + n)'
(7)
(8)
q
(9) B = L (3 .
8=1
An examination of (5) and (6) shows that
(10)
The relation a(a+1)n = (a+n)(a)", together with the definitions
of the contiguous functions, yields the formulas:
F = 00-':
1I! '
82 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
Using the operator (J = x(d/dx) , we see that
(fJ + ak)F = i: (ak + n) ~ " . ~ ~ .
n ~ O n.
Hence, with the aid of (11),
(12) k = 1,2, .. ', p.
Similarly, it follows that
(13) (fJ + 13k - l)F = (13k - l)F({3k -); k = 1, 2, ' .. , q.
The (p+q) equations (12) and (13) lead at once by plimination of
fJF to (p+q-l) liIlear algebraic relations between F and pairs of its
contiguous functions. Let us use F(CXl+) a:-; an element in each
equation. The result i:-; the set of simple relations,
(14) k = 2,3, "', p,
and
(15) (al - 13k + I)F = a1F(cxl +) - (13k - I)F({3k -);
k = 1, 2, ... , q.
From
it follows that
Thus, because of (10) and (G),
where
S _ (al + n)(0:2 + n) '(ap + n).
n - ({31 + n)({32 + n) .. ({3q + n)
Now, if p < q, then the degree of the numerator of S n is lower than
the degree of the denominator, and the elementary theory of rational
fraction expansions yields, for no two {3's equal,
p < q,
in which the U
j
is as defined in (3).
48] THE CONTIGUOUS FUNCTION RELATIONS 83
Therefore,
ro q /3U "
of = x L: L: j j
n_O 1-1 /3j + n n.
which in view of (11) becomes
q
(16) of = xL: UjF(/3j +).
The elimination of of, using (10) and the case k = 1 of (12), leads to
q
(17) (XIF = (XIP( (XI +) - x L: U);'(/31 +), p < q.
If p = q, the degree of the numerator of Sn equal:,; that of the de-
nominator. However, when p """' q,
p q
II (a, + n) - II (/3, + n)
Sn = 1 + _8---'=1=--_____ ,'_=_1 ---
q
II (/3. + n)
8=1
in which the fraction on the right has the desired property that its
numerator is of lower degree than its denominator. Thus
Sn = 1 + P = q,
]d /3] + n
and it is easy to see that in this case
q
of = xF + x L: UjF(/3j +),
so that (17) is replaced by the relation
q
(18) (0'1 + x)F = alF(al +) - x L: Uj F(/3j +), p = q.
]=1
If p = q+ 1, then with the notation of (8) and (9) we may write
p q
II (a, + n) - (n + A - B) II (/3. + n)
Sn = n + A - B + 8=1
q
II (/3. + n)
in which the fraction on the right has the desired rational fraction
expansion, so that
Sn = n + A - B + t
/3j + n
84 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
Thus we conclude that, when p=q+1,
q
of = xoF + (A - B)xF + x L UjF((3j +),
so that (17) is this time to be replaced by the relation
[(1 - X)al + (A - B)xJF = (1 - l;)aJ,'(al +)
'I
(19) - x L U
1
F((3j +);
)_1
From equation (11) we obtain
,,> 1 c n
OF(ak -) = L _O:i; - 12 .
-I O:k + 11 - 1 n! '
or
Now, with the notation of (7),
Cn +l
-+-- = CnTn,k,
O:k n
p=q+1.
where Tn,k has, for p ~ q, its numerator of lower degree than its de-
nominator. Thus
where
~ (3jWj ,k
Tn k = ~ - -
, }=1 (3j+n
in which the Wj,k is as defined in (4).
\Ve now have
q
OF(ak -) = (ak - l)x L Wj ,kF(3j +); p ~ q; k = 1,2, . " p.
j=1
But, from (12),
OF(a k - ) = (0: k - l)[F - F(o: k - )].
The elimination of of(a k-) from the preceding two formulas yields
the p relations:
q
(20) F = F(ak -) + x L W
j
,kF(3] +); p ~ q; k = 1,2" . " p.
When p=q+l, the fraction Tn,k has its numerator and denomina-
50] THE pFq WITH UNIT ARGUMENT 85
tor of equal degree. Then we write
p q
II (a, + n) - II ({3. + n)
Tn.k = 1 + 8=1.( k) s =1 ____ _
q
II ({3, + n)
})=1
and conclude in the same manner as before that
q
eF(ak -) = (ak - l)xF + (ak - l)x L lVi,kF({3j +).
) 1
Therefore, for p=q+l, (20) is to be replaced by
q
(21) (1 - x)F = F(ak -) + x L lVi,kF({3j +); Ie = 1,2"", p.
)-1
We have shown that for
in which no two {3's are equal and no (3 is a nonpositive integer, a
canonical set of (2p+q) contiguous function relations is as described
below.
If p<q, (14), (I5), (17), and (20) hold.
If p =q, the relations are (14), (15), and (20) together with (18)
to replace (17).
If p=q+l, the relations are (14), (15), (19), and (21).
49. A simple integral. It is an easy matter to extend the work
of Section 30 to obtain the following result.
THEOREM 28. If P ~ q + 1, if Re(b1) > Re(aJ) > 0, 'f no one
of b
J
, b
2
"', b
q
is zero or a negative integer, and if I z I < 1,
[
aI, a2,' . " a
p
; ]
pFq z
b1, b2,' " bq ;
(b)
1:
1 [a2, ... ,ap;
= r 1 t
a
'-\1 _ t)b,-a,-l
p
_IF q-l
r(al)r(b1 - al) 0 b ... b'
2, ,q,
If p ~ q, the condition I z I < 1 may be omitted.
50. The pFq with unit argument. We haye already found much
use for the evaluation of 2Fl (a, b; c; 1). We know that if no denom-
inator parameter is zero or a negative integer, the series
86 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
(1)
is absolutely convergent whenever p ;;:;; q and for p = q + 1 IS
absolutely convergent if
Re( b] - f ai) > O.
)=1 1c:l
It is natural to seek the value of (1), at least for p = q + 1. For
p ;;:;; q, the point z = 1 is not a singular point of the differential
equation for pFq, and there is less reason to hope that (1) wiII have
a relatively simple value. Compare Theorem 2R with Theorem 16,
page 47, to see why the method of Section 32 docs not carryover
to the more general case. In several special instances the desired
evaluation has been accomplished. This book contains a few of
the more widely used theorems along these lines.
S1. Saalschiitz' theorem. From Theorem 21, page 60, we see
that
(1) F(c - a, c - b; c; z) = (1 - z)aib-rF(a, b; c; z).
Equation (1) may be interpreted as an identity invohoing three
power senes:
We know that
which we now use both for ex = C - a - b and ex = 1. Thus we
find that
i: (c-a)n(c-b)n
zn
= i: ' (a)k(b)k( -n)k (c-a-b)n
zn
n=O (C)nn! , , ~ O Adl lC)k(l-c+a+b-n)kh:! n!
co [ -n, a, b;
= L 3F2
n=O
c,l-c+a+b-n;
51] SAALSCHUTZ' THEOREM 87
THEOREM 29. If n is a non-negative integer and if a, b, care
independent of n,
[
-n, a, b;
3F2
C, 1 - c + a + b - /1,;
Thc 3F2 of Thcorem 29 hac; thc propcrty that the c;um of its
denominator parametcrs exceeds the sum of its numerator para-
meters by unity. Any qtiJi'Q with that property is taIled Saalschutzian.
The proof given abo\'c for Theorem 29 requircs that a, b, c be
independent of 11. Bailcy [1; 21] gi\'es a general form for the
Saalschutz theorem. Here we need only the simple special eases
exhibited in the preceding Theorem 29 and the subsequent Theorem
30. Sec, howcver, the discussion at the end of this seetion.
THEOREM 30. If n is a non-negative 'integer and if a and bare
independent of n,
[
-n, a + n, + }a - b;
3F2
1 + a - b, +
Proof: On page G7 wc showcd that
[
a, b; ] + - b;
2FJ Z = (1 - z)-a 2FJ
1 + a - b; 1 + a - b;
Then
:t (a)n(b)n
zn
n=O n!(l + a - b)n
-4z ]
(1 - Z)2
f + 1a
k!n!(l -t--a--- b)k
f + - b)k(a)n+u(-1)k2
2k
z
nH
n,k,l) - --- ---k!n!(l + a - -bfXab
t (1, + - b)k(a),,+2k(-l)kzn+k
l-!n!(l + a - +
= i: + 1a - b)!.(a)n+kZn
ndl b'l U(n - k)!(l + a - b)kOa + Dk
Now (a)n+k = (a)n(a + n)k and (-l)k/(n - k)! = (-n)dn!'
Therefore \ve may write
88 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
f (a)n(b)n
zn
= i::t (-nh(a + n)k(! + !a - b)k. (a)nzn
n=on!(l + a - b)n ,,=0 ,=0 k!(l + a - b)k(!a + !)k n!
,r, [-n, a + n, ! + !a - b;
L oF2
n -, 1 + a - b, + L
1] (a)nzn
n! '
from which Theorem 30 follows.
Results such as those in Theorems 29 and may be extended
in the following way. Suppose that we have established the identity
(2) lea, n) = g(a, n)
for an arbitrary a ,vhich is independent of n and for every non-
negative integral n.
l\ow consider an a which depends upon n, a = hen). Equation
(2) yields
lea, 0) = g(a, 0)
for arbitrary a. Therefore
f[hCO), 0] = g[h(O), 0].
Again from (2) we obtain
f(a, 1) = g(a, 1),
from which
f[h(1), 1] = g[h(l), 1],
etc. It follows readily that (2) leads to
(3) f[h(n), n] = g[h(I1), n],
so that (2) is also true for a dependcnt upon n.
52. Whipple's theorem. Latcr \ve shall need the following
theorem due to Whipple[l].
31. If n is a non-negative integer and If band care
independent of n,
[
-n, b, c;
3Fz
1 - b - n, 1 - c - n;
[
_In -In + 1 1 - b - c - n
2, 2 2, ,
= (1 - x)n 3Fz
1 - b - n, 1 - c - n;
-4x ]
(1-- X)2 .
52] WHIPPLE'S THEOREM
Proof: Consider the ordinary hypergeometric function
[
b, c;
2Fl
b + c;
t(l - X + xt) ]
= ' (b)n(c)n[(l - x) + xtJnt
n
n!(b + C)n
co n (b)n(c)nxk(l _ x)n-kt
nH
= [; t; k!(n - k)!(b + C)n
= f['f (b)>;_k(c)n_kxk(l - x)n-Zktn
bO k!(n - 2k)!(b + C)n_k
= f[I! (-n)2k(b)n(c)n(1 - b - C - n)k( -l)kxk(1 - x)n-
2k
t
n
n=O k!n!(l - b - n)k(1 - C - rz)k(b + C)n
89
co [-!n, I-b-c-n; -4:r ]
= L (I-x)n 3Fz ---.2 -,---'
I-b-n I-c-n' (I-x) n.(b+c)n
, ,
We next expand the same hypergeometric function in powers of
t in another way:
2Fl
[
b, c;
b + c
t(l - x + xt) ]
= f: (b)n(c)n[l - x(1 - t)Jnt
n
n=O n!(b + C)n
co n (b)n(c)n( -1)kxk(1 - t)ktn
= [; t; k!(n - k)!(b + C)n
= f: (b)n+k(C)n+k( -1)kxk(1 - t)ktnH
n,k=O k!n!(b + C)n+k
= f: ' (b + k)"(c + k)"t
n
(b)k(C)k( -l)kxk(l - t)ktk
k=O "=0 n!(b + C + k)n /,'!(b + C)k
co [b + k, C + k;
= LzFJ
k=O b + C + k;
90 GENERALIZED HYPERGEOMETRIC FUNCTIONS [CII. 5
co [C' b;
= L (1 - t)-k 2FJ
k_O b+c+h;
co [ -n, b, c;
= L3F2
n-O 1 - b - n 1 - c - n-
, ,
Theorem 31 now follows by equating coefficients of tn in the preced-
ing two expansions.
In Whipple's work [1] the 3F
2
's inyolved are not necessarily
terminating. We now prove Whipple's theorem for nontcrminating
series.
THEOREM 32. If neither (a - b) nor (a - c) nor a is a negative
integer,
(1) 3F2[ a, b, c; x]
1 + a - b, 1 + a - c;
[
~ a , ~ a + ~ , 1 + a - b - c;
= (1 - x)-a 3F2
1 + a - b, 1 + a - c;
Proof:
[
a, b, c;
3
F
2
1 + a - b, 1 + a - c;
-4x ]
(1 - X)2
52] WHIPPLE'S THEORE1'r1
= 1'(1 + a - b)r(1 + a - c) f (a)k(b)k(C)k
Xk
.
1'(1 + a)r(1 + a - b - c) bO k!(1 + a)2k
91
1'(1 + a + 2/;) 1'(1 + a - b - c)
1'(1 + a - b + Ie) 1'(1 + a - c + k)
r(1 + a - b) 1'(1 + a _ c) 00 ,[b + k, c + k; ]
= ---- L of 1
1'(1 + a) 1'(1 + a - b - c) k_O - J 1 + a + 2k;
(a) k (b) k (c) kX k
--/';10:-+ a)2k
= 1'(1 + a - b) 1'(1 + a - c) f
1'(1 + a) 1'(1 + a - b - c) k!n!(1 + a)n+2k
= 1'(1 + a - b) 1'(1 + a - c) f i:: (a)k(b)"(c),,x
k
1'(1 + a) 1'(1 + a - b - c) k!(n - k) !(I + a)n+k
=
1'(1 + a - b)r(1 + a - c)
r(1 + a) 1'(1 + a - b - c)
00 [ -n, a; -x]
L 2FJ I)
1 + a + n; n.(l + (il n
By Ex. 9, page 69, we know that
[
-n, a;] [!a, !a + L
2FJ -x = (1 - x)-a 2FJ
1 + a + n; 1 + a + n;
Therefore we may write
[
a, b, C;
aF2
1 + a - b, 1 + a - c;
-4x ]
(1 - X)2
1'(1 + a - b) 1'(1 + a - c) co [!a, !a + !; 4]
= 1'(1 + a) 1'(1 + a - b - c) 2FJ 1 + a + n; (1-- :)2 .
(b)"(c),,
n!Cl + a)n(1 - x)a
= 1'(1 + a - b) r(1 + a - c) f C -1)k(a)2k(b),,(c)nXk
1'(1 + a) 1'(1 + a - b - c) k!n!(1 + a)n+k(1 - x)a+Zk
92 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
r(I + a - b) r(I + a - c) f: 2Fl[ b, c; 1]'
r(I + a) r(I + a - b - c) k=O 1 + a + Ie
( -I)k(a)2kxk
k!(1 + a)k(I - x)a+2h
r(I + a - b) r(I + a - c)
r(I + a) r(1 + a - b - c)
co r (1 + a + k) r (1 + a - b - c + ( - 1) k (a)z kX k
r(I + a - b + k) reI + a - c + k)k!(l + a)k(1 - x)a+2k

+ 1 + a - b - c;
= (1 - x)-a SF2
1 + a - b, 1 + a - c;
\vhich completes the proof of Whipple's theorem.
-4x ]
(1 '
53. Dixon's theorem. We shall obtain one more theorem on a
special sF 2. The series
is said to be well poised if
1 + aD = al + h = a2 + b2 = ... = aq + bq
Note that the left members of the relations in Theorems 31 and 32
are well poised sF
2
's.
Dixon [lJ summed the well poised SF2 with unit argument.
TREORE:\{ 33. The jollowing 1'S an identity 1f a, b, and c are so
restricted that each oj the junctions involved exists:
[
a, b, c;
SF2
1 + a - b, 1 + a - c;
r(1 + + a - b)r(1 + a - c)r(1 +!a - b - c).
r(I + a) r(I + - b) n 1 + - c) r(I + a - b - c)
53] DIXON'S THEOREM
Proof: Let
Then
[
a, b, c;
'" = aF2
1 + a - b, 1 + a - c;
if; = f (a)k(b)k(c)kr(I + a - b)r(I + a - c)
bO k! r(l + a - b + k) 1'(1 + a - c + k)
= E(l + a - b)r(I + a - c) f (a)k(b)k(ch .
r(I + a - b - c)r(I + a) k!(1 + 0.)2k
93
r(l + a + 2k) r(1 + a - b - c)
reI +-(i--':'b + k)r(1 + a - c + k)
But
r(I + a + 2k) r(I + a - b - c) = 2Fl[b + k, C + k; IJ
reI + a - b + k) r(I + a - c + k) 1 + a +2k;
Hence
_ (b + k)n(c + k)n.
-.L... ,
n=O n. (1 + a + '2k)"
r(I + a - b)r(I + a - c) (a).k(b)n+k(C)n+k
if; = -----7:,------'---;--'--'---c--'-o-::-----=------:- .L...
r(I + a - b - c) r(I + a) n,I.=O h!n!(1 + 0.)n+2k
r(I + a - b) r(I + a - c) , (a)k(b)n(c)n .
r(I + a - b - c) r(I + a) n=O k=O k!(n - k) !(l + a)n+k
Therefore
1'(1 + a - b) r(I + a - c) ro (-I)k( -n)k(a)k(b)n(c)n
if; = 1'(1 + a - b - c) r(1 + a) ?; bO k!n!(1 + a + n)k(I + a)n
= r(I + a - b) r(l + a - c) , 2Fl[ -n, a; -1] (b)n(c)n .
r(I + a - b - c) r(I + a) n=O 1 + a + n; n!(1 + a)n
By Theorem 26, page 68, with b replaced by (-n),
2Fl ' , -1 = r(l + a + n) r(I + !o.) = (1 + a)n.
[
-n a' J
1 + a + n: r(I + + nJ r(I + a) (1 +
94 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
Therefore we have
r(I + a - b) r(I + a - c) f: (b)n(c)n
1/1 = r(I + a - b - c)r(I + a) n=on!(l+
= r(I + a - b) r(1 + zFl[ b, c; 1]
r(I + a - b - c) r(I + a) 1 + !a;
r(I + a - b)r(l + a - c)r(I +1a)r(1 + ,1a - b - c)
= r(I + -a--=b=cYr;n + a) r(I + - b) r(I + - c)'
which completes the proof of Dixon's theorem.
For another form of Dixon's theorem see Ex. 3 at the end of this
chapter.
54. Contour integrals of Barnes' type. Consider the integral
_1 j[tIr(a
m
+ s)]r(-S)(-Z)8
dS
q ,
B II r(b
j
+ s)
(1)
where B is a Barnes path of integration: that is, B starts at -i en
and runs to +i en in the s-plane, curving if necessary to put the
poles of ream + s); m = 1, 2, "', p, to the left of the path and to
put the poles of r( -s) to the right of the path. The assumption
that there exists such a path precludes the possibility that any am
is zero or a negative integer. A representative Barnes path with
m = 3 is shown in Figure 5.
Let the integrand in (1) be I/I(s),
p
r( -s)( _Z)8 II ream + s)
(2) I/I(s)
m=l
We wish to determine the behavior of I/I(s) as s recedes from the
origin along the upper portion of the pure imaginary axis and along
its lower portion. We know from Theorem 13, page 31, that
(3) logr(a + s) = (s + a - !) Log s - s + 0 (1)
as I s I en in a region in which we have both I arg s I 7r - 0 and
I arg(s + a) I 7r - 0, 0 > O. Also, by (2) above,
54 CONTOUR INTEGRALS OF BARNES' TYPE 95
(4) t(8)
= exp[tllogr(am+8) - logr(bj +8) +logr( -8) +8 loge -z) J.
For the moment let 8 be on the upper imaginary axis. Then
s = it, I s I = t, arg 8 = !7r. -Csing (3) we haye, as t -> ro,
1
"-
"-
"-
it
splane
" -2
XX
" /
/
x x x
J( X H )( oX
3 4 5 6 7
)(XXXXX xx xx;<.xx
B
-it
Figure 5
log r (am + 8) (1t + am - D Log(it) - it + 0 (1)
= (it + am - t + -it + 0 (1).
Since lexp(iy) I = 1 for real y, we are completely uninterested in
any pure imaginary terms involved in the exponent in f(8). Hence
any such terms will hereafter be lumped together and designated as
PI (pure imaginary).
We now have, as t -> ro, 8 = it,
log r (am + 8) = [Re(a m ) - !J Log t - ht + PI + 0 (1).
Then also
96 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
log r (bj + 8) = [Re(bj ) - !J Log t - ht + PI + 0 (1)
and
log r (-8) = (-it - !)(Log t - hi) + it + 0 (1)
= -! Log t - !1!"t + PI + 0 (1).
Finally,
S loge -z) = it [Logizi + i arg (-z)J
= -t arg( -z) + PI.
'Ve are now prepared to examine if/(s) as 8 --> i co. Indeed,
p
log if/(8) = L [Re(a m ) - 1J Log t - iP1!"t + PI + 0(1)

q
- L [Re(b
j
) -}J Log t + + PI + 0(1)

- t Logt - t1!"t - targ(-z) + PI + 0(1).
Let
q
B = L Re(bJ.
m=-l
Then, for 8 = it, t 0'),
(5) log if/(8) = [A -B+Hq-p-1)] Log t+Hq-p-l)1!"t-t arg( -z)
+PI+O(l),
or
(6) if/(s) = O[tA-B+!(q-p-l) exp{ - t arg( -z) + -1) I].
'We turn next to a consideration of the behavior of >/;(8) on the
lower portion of the imaginary axis. Let 8 = -it, 181 = t, and
arg 8 = -!1!". Then as t --> co, 8 = -1't,
log ream + 8) = (-it + am - !)(Log t - hi) + it + 0(1),
so that
log ream + 8) = [Re(am ) - !] Log t - !1!"t + PI + 0(1),
log r(bj + 8) = [Re(b
j
) - Log t - !1!"t + PI + 0(1),
log r( -8) = (it - !)(Log t + hi) - it + 0(1)
= -! Log t - !1!"t + PI + 0(1),
8 log(-z) = -1't[Loglzl + i arg(-z)J
= t arg( -z) + PI.
54J CONTOUR INTEGRALS OF BARNES' TYPE 97
Therefore, as t -'> co, S = -it, we have
(7) log 1/;(s) = [A - B + Hq-p-1)] Log t
+ Hq-p-1)7rt + t arg( -z) + PI + 0(1),
or
(8) I/t(s) = O[tA-B+!(q-p-ll expl t arg( -z) + p-l} J.
We wish to impose conditions whieh will insure that 1/;(s) be
dominated by an exponential with negati\'e exponent. Equations
(6) and (8) show that we need both
(9) arg( -z) + < 0
and
(10) -arg( -z) + 9.7r(q-p-l) < O.
If (9) and (10) are satisfied,
1/;(s) = 0(e-
cI8
!), I s I -'> co on B, c > O.
Then in any z-region in which (9) and (10) are satisfied
11/;(s) ds
represents an analytic function of z.
If p = q + 1, h(q - p - 1) = -7r, and we require only that
larg( -z) I 7r - 0,0 > 0 to obtain (9) and (10).
If p = q, - P - 1) = - h, and we proceed to choof'e
larg( -z) I h - 0,0 > 0 to obtain (9) and (10).
If p < q, h(q - p - 1) ;;;:; 0 (since p and q are integer",), and
there is no region in the z-plane for which both (9) and (10) are true.
THEOREM 34. If P = q + 1 and no am is zero or a negative
integer,
p
(11)
_1_ f r( -s)( -z)81I ream + s) ds
27ri q
B II r(bj + s)
is an analytic function of z in the cut plane I arg( - z) I < 7r. If p = q
and no am I:S zero or a negative 7'nteger, (11) 7'S an analytic function oj z
in the open half-plane I arg( - z) I < !7r [i.e., in Re(z) < 0]. The
contour B is to be a Barnes contour as in Figure 5, page 95.
98 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch. 5
55. The Barnes integrals and the function ]>F
q
\Ve next relate
the Barnes integrals of Theorem 34 with the function pFq for
p = q+ 1 and for p = q.
Once more let
p
r( -8)( -z)s II ream + 8)
(1) f(8)
___ _
Let n be a non-negati\'c intcger, and consider the integral
(2)
2
1_; f f(8) ds
71"1 B.+C.
over the closed path (Bn + C n) shown in Figure 6.
C
n
is defined by 8 = (n + - 0 h.
s-plane
x x
)( x
xxxxxxxxxxx
Figure 6
The semicircle
The path Bn is
one of Barnes' type but is terminated at 8 = (n + and
8 = - (n + As n -> co, Bn -> B, the Barnes path of Figure 5.
55] BARNES INTEGRALS AND FUNCTION pFq 99
The value of the integral (2) is the negative of the sum of the
residues of -.J;(s) at the simple poles s = k, k = 0, 1,2,, n within
the closed contour. Now
r(-s)r(I+s) 1 -'Ir
r( -s) = -f[l+-s)- = reI + s) sin 'IrS'
and it is convenient to write 1/;(8) in the form
p
(3) -.J;(8)
(-z)' II rCam + s)
- 'Ir
= . ----------
SIn 'irS q
r(I + s) II reb} + 8)
1
Since Lim 'Ir( s - k) /sin 'irS = (- 1)\ the negative of the residue of
s+k
1/;(s) at s = k is
P
(-I)k(-z)k II r(a", + k)
(4)
q
r(I + k) II r(bj + k)
i-I
Therefore
(5)

f 1/;(s) ds + f 1/;(s) d8
'lrL B. 'lrL c.
As n -4 en, the first integral in (5) approaches the Barnes integral
of the preceding section, and we now show that the second integral
in (5) approaches zero.
Let so that the path C
n
becomes s=pe
i8
,
Let us determine the behavior of the -.J;(8) of (3) for 8 on C
n
and n
large. Since s = p cos 0 + i p sin 0, we have, as p -> co,
logr(am + 8) = (pcos 0 + ipsin 0 + am - p + 1:0) - pCos 0
- i p sin 0 + 0(1),
or
log ream + 8) = [p cos 0 + Re(a
m
) - !J Log p - p cos e - pO sin 0
+ PI + 0(1).
100 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
Again we use PI to denote all pure imaginary terms. The real
term - OIm(am ) is bounded by the constant hlm(a
m
) and is there-
fore included in 0(1). Of course
logr(bj+s) = [pcos 0+ Re(b
j
) - p- pCos 0- posin o+PI +0(1),
log r(l+s)=[p cos O+I-!]Logp-p cosO-pO sin O+PI+O(I),
and
loge -z)" = p cos 0 Logjzj - p sin 0 arg( -z) + PI.
It is an elementary matter (conflider j sin 'If.'! I -1) to show also that
log-;-1I'-- = -'lfP jflinol +PI+O(I).
fllll 'lfii
For the if;(s) of (3) it follows that
(6) log if;(s) = [A -B-1 +(p cos 0- !)(p-q-l)]Log p
- p( cos 0 + 0 sin 0) (p - q - 1) + p cos 0 Log I z I - p sin 0 arg( - z)
-'lfP I sin 01 +PI +0(1),
in which
m=l
Because of Theorem 34, page 97, we are interested only in
p = q + 1 and in p = q. First let p = q + 1. Then (6) yields
logif;(s) = (A -B-l)Logp+pcosOLoglzl-psinOarg(-z)
-'lfplsin 01 + PI + 0(1),
from which
(7) if;(s) = O[pA-B-l exp{ p cos 0 Log I z 1- p sin 0 arg( -z) -'lfP I sin 011],
Since -!'If cos 0 0. If we choose Izl < 1, Logizi < 0.
For I arg( - z) j 'If - 0, 0 > 0, the term - 'If P I sin 01 , never positive,
dominates the term - p sin 0 arg( -z). Hence for p = q+ 1, I z I < 1,
I arg( -z) I < 'If, if;(s) = O[exp (-cp)], c > 0, as p CX), and there-
fore
(8)

r if; (s) ds as n cx).


'lfl J C.
Then reference to equation (5) leads us to the following result.
THEOREM 35. If Izl < 1, 1j larg( -z) I 'If - 0, 0> 0, and if
no am or b
j
is zero or a negative integer,
55] BARNES INTEGRALS AND FUNCTION pFq
q+!
(9)
_1 I( -z)'r( -8) n ream + s) ds
211"i q
B II r(bj + s)
J" 1
q+!
II ream)
[a" a,, .. , a .. ,;
zJ
m=1
q+!F
q =
q
II r(bJ
b!, b2," " ", b
q
;

in which B 1"S the Barnes path of 1"ntegration, Figure 5, page 95.
Theorem 35 states the equal- I
ity of the two members of z-plane
(9) in the region shown in
Figure 7, a region in which
both members are analytic"
Since the left member of (9)
is analytic in the larger re-
gion, the cut plane with the
non-negative axis of reals de-
leted, that left member of (9)
furnishes an analytic contin-
uation of the right member
of (9).
Let us now return to equa-
tion (6), page 100, and put
p = q. The result is that as p ----> 00, son C
n
,
Figure 7
101
(10) log 1/;(s) = cos 0) Log p + p (cos 0 + 0 sin 0)
+ pCos OLoglzl - psin Oarg(-z)
- 11" P 1 sin 0 1 + PI + 0 (1),
from which
(11) 1/;(s) = O[pA-B-I-p cos 8 exp! p cos 0(1 +Log 1 zl)
+posino - psinOarg(-z) - 1I"plsinollJ.
Kow 1 01 h, and we choose 1 arg( -z) 1 h- 0, 0> 0, in order
to conclude that
pO sin e - psin Oarg(-z) - 1I"plsin 01 -oplsin 01.
102 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
If we also choose I z I < e-t, Log I z I < -1, we may conclude that,
f o r e ~ h,
p cos 0 (1 + Log I z I) < 0.
Since sin2 0 + cos
2
0 = 1, the coefficient of p In our exponent 1S
always negative.
Once more, for s on C\, f(s) = O[exp( -cp)], c > 0, as p ---)0 CD.
Again (8) holds true, and ,ve may let n ---)0 CD in each member of
equation (f, page 00. Since p = q, the right member of (5) ap-
proaches an entire function, a series convergent for all finite z. The
first term on the left approaches a Barnes integral known (Theorem
34) to represent an analytic function for Re(z) < 0. We may no\v
drop the unnecessary restrietion I z I <. e-
l
, which was merely a
tool in our proof, since two analytic functions shown to be identical
in some region are identical throughout their common region of
analyticity.
THEOREM 36. If Re(z) < and if no am or bj is zero or a negative
integer,
q
I
c -z)8r( -s) IT r(a", + s) ds
1 111=1
-- --
'27ri q
H IT rebj + 8)
J=l
in which B is a Barnes path of integration, Figure 5, page 95.
56. A useful integral. Consider the integral
(1) A(t) = llxa-l(t _ x)iJ-I pFq[a
l
"'" a
p
: CXk(t _ X)8]dX
b
l
,' . " b
q
,
in which the parameters are subject to the conditions that
(a) Re(a) > 0, Re(/3) > 0;
(b) k and 8 are to be non-negative integers not both zero;
(c) No b
j
is to be zero or a negative integer;
(d) p ~ q + 1, unless some am is a nonpositive integer, in which
case p may be any positive integer.
56] A USEFUL INTEGRAL 103
We evaluate A (t) by term-by-term integration. Let x = tv.
Then
A (t) = taH-lilva-l(1 _ v) pF q[a
1
' " a
p
: ctH'vk(l - V).] dv
b
1
,' . " b
q
,
p
1 CD II (a
m
)nt
n
(k-t8)c
n
v
a
-lHn(l - v) O-l-l.ndv
= r L md _____ _
Jo q
nl rr (bJn
so that
p
00 rr (am)ncntn(sH) B( a + 1:71, /3 + sn)
(2)
A(t) = t aH-
1
L - -------
n%:;O If
nl rr (bJn
)=1
in which B(zl' Z2) is the Beta function of Chapter 2. Now
B(a + kn /3 + sn) =
, rea + f3 + + sn)
r(a)r(f3) (a)kn(/3)sn
= rCa . (a + (3)(k+s)n
In Lemma 6, page 22, we found that
which permits us to write
(3) B(a + kn, (3 + sn)
rr
k (a + i - 1) II' (/3 + u - 1)
B(a, (3)k
nk
s
sn
------
= _____ oS n.
k+s (a + /3 + LV - 1)
(k + s)(k+
8
)n II -----------
I,. + s n
The use of (3) in the right member of (2) yields a hypergeometric
function with (p + k + s) numerator parameters and (q + k + s)
denominator parameters. The precise result follows.
THEOREM 37. If Re(a) > 0, Re(/3) > 0, and 1.j k and s are posi-
tive integers, then inside the region of convergence of the resultant series
104 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
(4)
cx'(t - Xl'] dx
/3+s-1 a a+I a+k-l f3 f3+1
al'" a --- ... -.. ------ ...
, ,p, If' k' , k 's' s '
s
leks'elk t.
If either, but not both, of k and s is zero, it is a simple matter to
modify the steps in the derivation of (4) and thus arrive at the
pertinent result.
THEOREM 38. If Re(a) > 0, Re(f3) > 0, and 1f k is a positive
integer, then inside the region of convergence of the resultant series
(5)
et
k

a+f3 a+f3+1 a+f3+h-l
b ... b --- ~ - ~ ... -- -- ---'
I, ,q, k' k " /; ,
The reader can easily write the corresponding theorem for k = 0,
s > 0.
EXAMPLE: In equation (4) of Theorem 37 choose p = 0 (no a's),
q = 1 (one b), b
1
= 1, a = 1, /3 = 1, k = 1, s = 1, e = - t. The
result is
] [
1,1;
- tx(t-x) dx = B(I, l)t 2F:,
1, ! , ~ ;
= toFI[-;
3
~ ,
56] A USEFUL INTEGRAL 105
The reader may already know, or will discover in the next chapter,
that Jo(z), the Bessel function of the first kind and of index zero,
is given by
Jo(z) = oFl( - ; 1; - !Z2).
The oFl in the second form of the right member of (6) is elementary.
Indeed
= 2 sin
Therefore (6) may be rewritten in the form
(7) llJoC yxV-- x) dx = 2 sin tt.
EXERCISES
1. Show that
OFI[-; XJoFI[-;
a; b;
] [
la + Ib !a + Ib - J..
2 2' 2 2 2,
X = 2F3
a, b, a + b - 1;
You may use the result in Ex. 6, page 69.
2. Show that
i [
1
t 4' ",
o x!(t - x)-![1 - X2(t - x)2]-idx = "FI 1.
,
3. With the aid of Theorem 8, page 21, show that
and that
r(1 +
r(1 + a)
cos !7rar(1 - a)
r(1 - !a)
Thus put Dixon's theorem, Theorem 33, page 92, in the form
[
a, b, c; ]
3F2 I
I + a - b, 1 + a - c;
cos sin 7r(b - r(l - a)r(b - !a)r(l + a - c)f(l + - b - c)
= -s[; 7r(b=- a)-- . r(I - - c)r(l + a - b - c)
106 GENERALIZED HYPERGEOMETRIC FUNCTIONS [Ch.5
4. Use the result in Ex. 3 to show that if n is a non-negative integer,
[
-2n Ci 1 - '" - 2n-
" fJ ,
Ji'2
1 - Ci - 2n, (3;
5. With the aid of the formula in Ex. 4 prove Ramanujan's theorem:
] [
a; ] [ a, (3 - a;
X IFI -x = 2F.
{:J; {:J, !(3, ;(3 + !;
6. Let'Yn = sF;( -n, I - a - n, 1 - b - n; a, b; 1). Use the result in Ex. 3
to show that 'Y2n+1 = 0 and
(-I)n(2n) !(a + b - 1)3n
'Y2n = -,---------.
n.(a)n(b)n(a + b - Ihn
7. With the aid of the result in Ex. 6 show that
oF
2
( -; a, b; t) oF
2
( -; a, b; -t)
[
!ea + b - 1), i(a + b),!ea + b + 1);
-= J's
a, b, !a, !a + !, !b, !b + !, !(a + b - 1), Ha + b);
8. Prove that
n (_l)n-k('Y- b-
c
)n_kC'Y-b)k(-y-chx,,-k [ -k, b, c; x]
t;o k!(n - k)!('Yh SF2 l--y+b-k,I--y+c-k;
= ('Y-b)n('Y-c),,(l-x)n sF2[-!n, -tn+t, l--y-n; -4X]
n!('Y)n +b (l_X)2 '
1-'1' -n,I-'Y+c-n;
and note the special case -y = b + c, Whipple's theorem, page 88.
Exs. 9-11 below use the notation of the Laplace transform as in Ex. 16, page 71.
9. Show that
10. Show that
11. Show that
56] A USEFUL INTEGRAL 107
12. Show that
13. In Ex. 19, page 71, we found that the complete elliptic integral of the first
kind is given by
Show that
CHAPTER 6
Bessel
Functions
57. Remarks. No other special functions have received such
detailed treatment in readily available treatises* as have the Bessel
functions. Consequently we here present only a brief introduction
to the subject, including those results which will be used in later
chapters of this book. f
For a discussion of orthogonality properties and of zeros of Bessel
functions, see Churchill [2J and Watson [1]. An extremely simple
result on zeros of Bessel functions appears in Ex. 13 at the end of
this chapter.
58. Definition of In(z). We already know that the oFo is an
exponential and that the lFo is a binomial. It is natural to examine
next the most general of l, the only other pF q with less than two
parameters. The function "\ve shall study is not precisely the oFl
but one that has an extra factor as in (1) below.
We define J n(Z), for n not a negative integer,
(1)
(z/2)n ( Z2)
In(z) = r(I + n) of! -; 1 + n; -4 .
If n is a negative integer, we put
*The most strik;'lg example is \'latson's ('xhaustive (804 pages) work; Watson [1].
An exposition sufficiently thorough for most readers will be found in Chapter 17 of
Whittaker and Watson [1].
108
59] BESSEL'S DIFFERENTIAL EQUATION
109
(2) In(z) = (-l)nJ_n(Z).
Equations (1) and (2) together define J n(Z) for all finite z and n.
The function J n(Z) is called "the Bessel function of the first kind of
index n."
Since
of 1( - ; 1 + n; -z4:) = f J:-:
(1 + n) kl .. !
,>; r (1 + n) ( - 1) kZ2 k
t=o ITr(l--+-n--t=-k)2
2k
'
the relation (1) is equivalent to
m (_1)kz2k+n
(3) In(z) = 622k1n1Jr(1 + n + k)
Note also the immediate result
(4)
59. Bessel's differential equation. We know a differential
equation satisfied by any oFJ by specializing the result in Section
46. The equation
(1) [0(0 + b - 1) - yJu = 0;
d
0- y-
- dy'
has u = oFJ( -; b; y) as one solution. Equation (1) can also be
written
d
2
u du
y - + b -- - u = 0.
dyZ dy
(2)
We now put b = 1 + n, y = -zz/4 in (2) to obtain
(3) zu" + (2n + l)u' + zu = 0,
in which primes denote differentiations with respect to z. One
solution of (3) is u = oFJ( -; 1 + n; - z2/4). We seek an equation
satisfied by w = znu. Hence in (3) we now put u = z-nw and arrive
at the differential equation
(4) z
2
w" + zw' + (Z2 - n2)w = 0,
of which one solution is w = znOF
1
(-; 1 + n; -z2/4).
Equation (4) is Bessel's differential equation. If n IS not an
integer, two linearly independent solutions of (4) are
110
(5)
and
(6)
BESSEL FUNCTIONS [Ch.6
WI = J n(Z)
Wz = J _,,(z)
and they are valid for all finite z.
If n is zero or a positive integer, (5) is still a valid solution of
Bessel's equation, but then (6) is not linearly independent of (5).
For integral n, a second solution to accompany J"(z) is logarithmic
in character and can be obtained by standard procedures. * The
result is
n-I (-l)k+I(n _ I)!zzk-n
(7) Wa = Yn(Z) = J n(Z) log Z + k 22k ,1-n/,;1(1 - n)k
1 m (-I)k+l(Hk + H
k
+
n
)z2k+
n
+ "2 6, -- n) ,--
for n > 1. In (7) the common convention IIo = has been used.
For n = 0, the first series on the right in (7) is to be omitted;
for n = 1, the first series on the right in (7) is to be replaced by
the single term (-Z-I).
60. Differential recurrence relations. In the equation
co (-I)k
z
2k+
n
J n(Z) = L --;;k+-;;- .,------- -_.
kdJ 2" ". r(1 + n + J )
(1)
\ye may multiply both members by zn and then differentiate through-
out \vith respect to z to obtain
d en ( - 1) k
z
2k+2n-1
(2) dz [znJ,,(z)] = .6 r(n + Ie)
in which we have canceled the factor (2k + 2n) in numerator and
denominator, using r(I + n + k) = (n + k)r(n + k). Equation
(2) can be rewritten as
d 00 ( -1) kzZk+n-1
dz [znJ n(Z)] = zn (; 22k+n-1k! r(I + n - 1 + k)'
and we thus see that the right member is znJ n_I(Z). We conclude
that
(3)
'In Rainville [1], pertinent techniques are explained on pages 285-291 (for n = 0)
and on pages 299-303 (for n > 0).
61] PURE RECURRENCE RELATION 111
Equation (3) may also be put in the form
(4)
which is called a differential recurrence relation, differenti.al because
of the differentiation recurrence hemuse of the presenec
of different indices nand (n - 1).
Next we return to equation (1), insert the fad.or z-n on each side,
and again differentiate each member with rrsppcL to z to obtain
(5)
d CD (-I)k
z
U-l
dz [z-nJ n(Z)] = t; 22k I ,,-=i(l .. = 1) kY
A shift of index from k to (I .. + 1) yields
d 00 ( _ 1 ) k1 l
Z
2 HI
dz [z-"J,,(z)] = 6 r(1 + n+- i
from which we obtain
(6)
Equation (6) can be expressed equally well as
(7)
In equations (7) and
(4) zJ n'(z) = zJ n_l(Z) - nJ,,(z)
we haye two differential recurrence relations. From them it follows
also that
(8)
61. A pure recurrence relation. Elimination of J ,,'(z) from the
relations (4) and (7) of the preceding section gives us at once the
pure recurrence relation
(1)
It is also instructive to obtain (1) from the sole contiguous function
relation possessed by the oFl function.
Consider the set of Bessel functions J n(Z) for non-negative integral
index. If we write (1) in the form
2(n - 1)
(2) J n(Z) = J n-1(Z) - J n_2(Z),
Z
112 BESSEL FUNCTIONS [Ch.6
we obtain each J n of the set in terms of the two preceding it; J
from J
o
and J
1
, J
3
from J
1
and J
2
, etc. In this \vay we can, for
integral n, write
(3)
The coefficients An(z) and BnCz) are then polynomials in liz. These
are simple special cases of Lommel's polynomials Rm.,(z) which may
be encountered* by applying the same process to (2), with n re-
placed by (II + m) to arrive at the result
(4) J,+m(Z) = Rm.,(z)J,(z) - Rm-I.,+I(Z)J._1(z).
The Lommel polynomial is a 2F3 (sec Watson [1:207]):
(5)
- ~ ] .
Z2
62. A generating function. Our approach (Section 58) to the
function J nCZ) was from the hypergeometric standpoint, which is
natural here because this hook is largely concerned with functions
of hypergeometric character. Some authors approach J n(Z) by
first defining it, for integral n only, by means of a generating func-
tiont relation which we shall now obtain.
Lemma 12. For n ~ 1,
[n/2] [(n-1)/2]
(1) L A (k,rn) = L A(k, n) + L A(n - k, n).
k ~ O k ~ O k ~ O
Proof: First note that for integral n ~ 1,
(2) n = 1 + [!nJ + [!en - l)J,
in which [ ] is the usual greatest integer symbol. Equation (2) is
easily verified separately for n even and for n odd.
Next note that
n [n/2] 1+ [n /2] + [(n-l) /2]
(3) LrA(k, n) = L A(k, n) + L A(k, n).
k=O k ~ O k=J.+[n/2]
In the last summation in (3) replace k by (n - k); that is, k by
1 + [nI2] + [en - 1)/2J - k. Then
*See Watson [1 :294].
tSee Chapter 8 for some detail on the generating function concept.
62] A GENERATING FUNCTION
113
n [n/2]
L A(k, n) = L A(k, n) + L A(n - k, n),

from which Lemma 12 follows by reversing the order of the second
summation on the right.
THEOREM 39. For t T"= 0 and for all finite z,
(4) - DJ = "too J,,(z)t
n

Proof: Let us collect powers of z in the sUlllmation
co ':n
L J ,,(z)t" = L J,,(z)tn + L J ,,(z)tn
Tt=-UJ 'ft.,... -00 11=0
00 00
= L J _n_I(Z)t-
n
-
1
+ L J n(Z)tn.
7l=O n=O
We defined J m(Z) for negative integral m in Section 58. Using that
definition we get
ro ro ro
L J n(Z)t
n
= L (-l)n+IJ n+I(Z)t-
n
-
1
+ L J n(z)ln
n=-ro n=O n=O
ro (_ 1) n+k+ll-n-Izn+Zk+1 ro ( _ 1) ktnzn+Zk
= + nto2"+
2k
k!(n + k)!
ro [n/Z] (_l),,-k+lt-n+Zk-lzn+1 ro [n/Z] (-l)ktn-Zkzn
= ?; {;o 2n+lk!(n + 1 - k)! 2nk!(n - k)!
ro[(n-
l
l/2](_1)n- ktk-(n-kl zn ro[n/2](_1)kt n- k- k zn
- L L . - + 1 + L L . -'
- n-l k_Q k!(n - k)! 2n n-I k!(n - k)! 2"
We now use Lemma 12 to conclude that
ro ro n (_ 1) kln-k-k zn
J,,(z)t
n
= 1 + (; k!(n _ k)! 2"
it'" cxp[iz(t - l)J.
n.2 2 l
See also Ex. 23 at the end of this chapter.
114 BESSEL FUNCTIONS [Ch.6
63. Bessel's integral. Theorem 39 of the preceding section
may be interpreted as giving the Laurent expansion, valid near the
essential singularity t = 0, for the function exp - t-
1
)J. The
Laurent series coefficient is known. Indeed,
(1)
1 f(Ot) [1 ]
In(z) = 27ri u-
n
-
1
exp 2
z
(u - u-
1
) du,
in which the contour (0+) is a simple closed path encircling the
origin u = in the positive direction.
In (1) let us choose the partieular path
u = e
iO
= cos 0 + 'I sin 0,
o running from (- 7r) to 7r. Then u-
I
= cos 0 - i sin 0, and (1)
yields
J ( )
1 f" [ . .. J d
n Z = 2; _ :x
p
- mO + tZ sm 0 0
= 2
1
7r i:cos(no - Z sin 0) do - f; - Z sin 0) do.
In the last two integrals the former has an even function of 0 as
integrand, the latter an odd function of 0 as integrand. Hence
]i"
J n(Z) = - cos (nO - Z sin 0) do,
7r 0
which is Bessel's integral for J n(Z).
THEOREM 40. For integral n,
1 i" (2) J n(Z) = - cos (no - Z sin 0) do.
7r 0
Bessel's integral representation of J n(z) can be* extended to non-
integral n. The result, called Schlajl'i's integral, IS
(3) J n(Z) = 1. ("cos(no-z sin 0) do- n7r ("'exp( -nO-z sinh 0) do,
7rJ 0 7r J 0
valid for Re(z) > O. Equation (3) will not be used in our work.
64. Index half an odd integer. Let us put into hypergeometric
form the elementary expansion
*See Whittaker and Watson [1 :362].
641 INDEX HALF AN ODD INTEGER
ro (_1)k
z
2k+l
(1) sm z = .6 (2k + 1)(
Since (2k + I)! = (2)2k) equation (1) yields
or
(2)
Now
Jl(z) = (z/2) I oFl( _ . ;t. _ lZ2)
) 2, 4
and = !y;. Hence
(3)
Ji(z) = (;zY sin z.
or
In much the same manner the elementary expansion
=
cos z = L -----
(2n)! '
(4) cos
z = F (- . l - lZ2)
o I , 2, 4
leads us to the relation
(5)
(
?)!
J -1(Z) = ;z cos z.
In Section 61 we derived the pure recurrence relation
(6) J n(Z) = 2(n-1)z-IJ n_I(Z) - J n_2(Z).
In (6) replace n by (n + !) to obtain
(7) In+!(z) = (2n - l)z-lJ
n
_
l
(z) -
Let n be a positive integer and iterate (7) to see that
(8)
in which PI and P
2
are polynomials in their arguments.
From equations (3), (5), and (8) it follows that for integral n
In+l(z) = A(z) cos Z + B(z) sin z
115
116 BESSEL FUNCTIONS [Ch.6
in which A(z) and B(z) are polynomials in z-l.
Bessel functions of index half an odd integer are often called
spherical Bessel functions. They, as well as most other I3essel
functions, are encountered in various physical problems. Spherical
Bessel functions led to the definition and study of I3essel polynomials
which we discuss to some extent later in this book.
65. Modified Bessel functions. Many physical problems lead
to the study of Bessel functions of pure imaginary argument. This
in turn leads to the definition of such functions as
. . (z/2) " ( Z2)
(1) I,,(z) = r"J,,(zz) = 1'(1 + 71) uFI -; 1 + 71; 4 '
71 not a negative integer. The function I n(Z) is called a modified
Bessel function of the first kind of index n. A study of J,,(z) for
complex z includes corresponding properties of I ,,(z) by simple
changes of variables. The function I n is related to J n in much the
same way that the hyperbolic functions are related to the trigono-
metric functions. Some elementary properties of I n(Z) will be
found in the exercises below.
66. Neumann polynomials. From Theorem 39, page 113, we
obtain, for w ;;e 0 and for all finite z,
co
(1) - w-
I
)] = L J n(Z)W
n
,
n=-ro
which can equally well (Ex. 2, page 120) be written
co
(2) exp[!z(w - w-
I
)] = Jo(z) + L In(z)[W
n
+ (-l)nw-
n
],
n=l
because J _n(Z) = (-1) nJ n(Z). In (2) put w = t + Vr + 1 and
note that (-w-
1
) = t - Vt2 + 1. The result is
co
(3) e
zt
= Jo(z) + L In(z)[(t + Vr + 1)n + (t - Vt2 + I)n].
7t=1
Let us definefn(t) by
(4) fn(t) = (t + Vt2-.f.l)n + (t - Vt2+I)n, n O.
Then fn(t) is a polynomial in t and (3) now appears as

(5) e
d
= Uo(t)Jo(z) + Lfn(t)Jn(z).
T/=1
66] NEUMANN POLYNOMIALS
117
The Laplace transform (Churchill [1]) of a polynomial in t is a
polynomial in S-1. Let 20n(s) be the Laplace transform of our
f,,(t):
(6)
Then from (5), since LI eztl = (s - Z)-I, we obtain
(7)
The polynomials O,,(s) are called Neumann polynomials.
Let us assume that the series in (7) is sufficiently well behaved
(proved below) that the manipulations to be performed are legit-
imate. Differentiation of (7) yields
m
(8) -(s - Z)-2 = Oo'(s)Jo(z) + 2 L., On'(s)Jn(z)
7/.=1
and
=
(9) (s - Z)-2 = Oo(s)Jo'(z) + 2 L., On(S)J n'(z).
71 0 -1
Now 2J n'(z) = J n_I(Z) - J n+I(Z) and Jo'(z) = -J1(z), so that (9)
may be written as
co '"
(s - Z)-2 = -Oo(s)J
1
(z) + L., On(S)J
n
_
1
(z) - L., On(S)J
n
+
1
(z)
n=l
co (0
= -Oo(s)J
1
(z) + L., On+l(S)J n(Z) - L., O,,_I(S)J ,,(z),
n=O n=2
or
OJ
(10) (s - Z)-2 = 01(S)JO(z) + L., [O,,+I(S) - O,,-I(S)JJ n(Z).
71=1
From (8) and (10) it follmvs that
OJ
[Oo'(s) + 01(S)]JO(z) + L., [20n'(s) + O,,+I(S) - O,,_I(S)JJn(z) = O.
Since for each n the function z-nJ ,,(z) is nonzero at z = 0, it
follows that an expansion of the form
IS umque.
118 BESSEL FUNCTIONS [Ch.6
Hence OI(S) = -Oo'(s) and
(11) On+I(S) = On_I(S) - 20/(s), n;;;:;1.
We know that Oo(s) = g-I and now that 01(S) = g-2. The
Neumann polynomials may now be described as follows:
(12)
(13)
Oo(S) = S-l, 01(8) = g-2,
On(8) = On_2(S) - 20:_
1
(8), n 2.
The Ones) are uniquely determined by the description (12) and (13).
THEOREM 41. The Neumann polynomials defined by (12) and (13)
above are given by CMs) = S-I and
(14)
n [n/2J (n - 1 - k)!(2js)n+I-2k
Ones) = 4 (; k! n 1.
Proof: From (14), OI(S) = t(2/S)2 = g-2. Also, for n ;;;:; 2,
and
[n-IJ
n - 1 2 (n - 2k)(n - 2 - k)!( _2/s2) (2/s)n-I-Zk
- .. . _-
4 k-O k!
n - 1 [n/2] (n - 2k)(n - 2 - k)!(2/s)n+l-2k
= --- L: , .
8 k_O k.
Therefore, for the OneS) of (14),
On_2(8) -
_ 2 [n/2J [en - 2)k + (n - l)(n - 2k)J(n - 2 - k)!(2/8)n+l-2k
- 4 L: /'-'
c.
= On(8),
as desired.
By Theorem 41 the dominating term in is 2
n
-
I
n! 8-
n
-
1
The
dominating term in In(z) is (!z)n/n!. Therefore, as n -> 00,
2
n
-
I
n! zn
On(S)J n(Z) = 2nn! (1 + En)
67] NEUMANN SERIES 119
in which En -> O. For I Z I r, choose lsi R where R > T. Thus
for n sufficiently large,
I On(S)J n(Z) I < c n, C a constant.
Then the expansion (7) is absolutely and uniformly convergent, and
the manipulations performed on it are justified.
For the moment let F denote the right member of (7). Because
of (11) the right members of (8) and (9) have zero as their sum.
lIenee
+ = 0
as az
from which it follows that the right member of (7) is a function of
the single argument (s - z). But at z = 0, that right member is
Oo(s) = S-I. Hence, once again, the left member of (7) is (s - Z)-l.
67. Neumann series. On the basis of the expansion
(1)
1 00
-- = Oo(s)Jo(z) + 2 L On(S)J n(Z)
S - Z
and the Cauchy integral formula
(2) fez) = i ds,
27f'l c S - Z
where C is I s I = r described in the positive direction, we obtain
at once
1 f f(s) ds 1 ro f
fez) = -2 .Jo(z) + L J n(Z) f(s)On(S) ds,
7f'l c S 1r'ln_l C
or
ro
(3) fez) = L aJn(z),

in which ao = f(O) and
(4)
1r'/, c
That is, iff(z) is analytic in Izi r, thenf(z) can be expanded into
the Neumann series (3), with coefficients as described, and the ex-
pansion is valid for i z I < r.
Some of the expansions in the exercises may be obtained by this
method, if it seems desirable.
120
EXERCISES
BESSEL FUNCTIONS [Ch.6
(A)
1. By collecting powers of x in the summation on the left, show that
f: J
2
n+I(x) = t ('Jo(Y) dy.
n"",O Jo
2. Put the equation of Theorem 39, page 113, into the form
co
exp[!z(t - t-I) J = Jo(z) + LJ n(Z)[t
n
+ (-l)"t-n].
n=l
Use equation (A) with t = i to conclude that
00
cos z = Jo(z) + 2 L (-1)kJ
2k
(Z),
k_1
00
sinz = 2 L (-1)kJ
2k
+
l
(z).
k_O
3. Use t = e,9 in equation (A) of Ex. 2 to obtain the results
00
cos (z sin 0) = Jo(z) + 2 L J 2k (z) cos 2kO,
k_1
00
sin (z sin 0) = 2 L J2k+
I
(Z) sin(2k + 1)0.
k_O
4. Use Bessel's integral, page 114, to obtain for integral n the relations
(B)
21"
[1 + (-l)nJJ n(z) = - cos nO cos(z sin 0) dO,
7r 0
(C) [1 - (-l)nJJ,,(z) = ~ i"sin nO sin(z sin 0) dO.
7r 0
With the aid of (B) and (C) show that for integral k,
Ii"
J
2k
(Z) = - cos 2kO cos(z sin II) dll,
7r 0
J 2k+
I
(Z) = ! i"Sin(2k + 1)0 sin(z sin 0) dO,
7r 0
i" cos(2k + 1) 0 cos(z sin 0) dO = 0,
i"sin 2kO sin(z sin 0) dO = O.
5. Expand cos(z sin 0) and sin(z sin 0) in Fourier series over the interval
-7r < 0 < 7r. Thus use Ex. 4 to obtain in another way the expansions in Ex. 3.
6. In the product of e x p [ ~ x ( t - t-I)J by exp[ -tx(t - t-I)J, obtain the co-
efficient of to and thus show that
co
Jo2(x) + 2 L J n
2
(X) = 1.
n_l
For real x conclude that I Jo(x) I ~ 1 and !In(x)! ~ 2-iforn ~ 1.
67] NEUMANN SERIES 121
7. Use Bessel's integral to show that IJ,,(x) I 1 for real x and integral n.
8. By iteration of equation (8), page llI, show that
dm m
2
m
d
mJn(z) = 1: (_l)m-
k
C
m
.Jn+m_2k(Z),
Z k_O
where C
m

k
is the binomial coefficient.
9. Use the result in Ex. 1, page 105, to obtain the product of two Bessel func-
tions of equal argument.
(z/2),,+m [Hn + m + 1), Hn + m + 2);
Ans. J"(z)J,,,(z) = 1'(1 + n)1'(1 + m) 2
F
3
1 + n, 1 + m, I + n + m;
10. Start with the power series for J,,(z) and use the form (2), page 18, of the
Beta function to arrive at the equation
_ 2
Z
" 2n .
2( ')" it ..
In(z) - R!)f(n + !) 0 sm cp COi-;(z cos cp) dcp,
for Re(n) > -!.
11. Use the property
d 1 du
d
-oF,(-;a;u) =-d-oF,(-;a+I;u)
x a x
to obtain the differential recurrence relation (6) of Section 60.
12. Expand
[
-;
of,
1 + a;
in a series of powers of x and thus arrive at the re;:ult
13. Use the relations (3) and (6) of Section 60 to prove that: For real x, be-
tween any two consecutive zeros of x-nJ ,,(x) , there lies one and only one zero of
x-nJn+l(x).
14. For the function I,,(z) of Section 65 obtain the following properties by using
the methods, but not the results, of this chapter:
zln'(z) = Zln_I(Z) - nln(z),
zln'(z) = zln+'(z) + nl,,(z),
2/n'(z) = In_,(z) + I,,+,(z),
2nln(z) = z[ln_,(z) - In+,(z)J.
15. Show that I n(Z) is one solution of the equation
Z2W" + zw' - (Z2 + n
2
)w = O.
16. Show that, for Re(n) >
f
l
1r. 2n
In(z) = 1'mr(n + 8m cp cosh(z cos cp) dcp.
122 BESSEL FUNCTIONS [Ch. 6
17. For negative integral n define In(z) = (-l)"L,,(z), thus completing the
definition in Section 65. Show that 1,,( -z) = (-1)"1 ,,(z) and that
rn
+ t-
I
)] = L: I nCzW.
18. Use the integral evaluated in Section 56 to show that
.flYX(t - x) l"J" (yxCZ-=- x dx = 2-ny;;: t,,+iJn+!W).
19. By the method of Ex. 18 show that
i\./l=X sin (a0) dx = 7ra-IJ2(a),
and, in general, that
i\l - x)C-I x i
n
Jn(a0) dx =
20. Show that
i'ex
P
[ -2x(t - x)]/o[2x(t - x)] dx = i'ex
P
( _(32) d{3.
21. Show that
.f[X(t - x)]-i exp[4x(t - x)] dx = 7r exp(!t
2
)/o(!t2).
22. Obtain Neumann's expansion
(
! )" _ (n + 2k)(n + k - 1) !J,,+2k(Z)
2
z
- k! ' n 1.
23. Prove Theorem 39, page 113, by forming the product of the series for
exp(!zt) and the series for exp( -!zt-I).
CHAPTER 7
The Confluent
Hypergeometric
Function
68. Basic properties of the IFIo The functions oFo (the ex-
ponential) and IFo (the binomial) are elementary. We have devoted
some time to the study of the uFI' a Bessel function, and to the 2F
l
,
the ordinary hypergeometric function. Except for terminating
series, we are interested in the pFq mainly when p ~ q + 1 so that
the series has a region of com"ergence. To complete the introduc-
tion to special properties of the pF q when q = 0, 1, we need only to
consider the IFlo
The series
(1)
in which b ~ zero or a negative integer is convergent for all finite z.
This function is also known as the Pochhammer-Barnes confluent
hypergeometric function. An equation satisfied by the IFI can be
obtained by confluence of singularities from a Fuchsian * equation
with three singular points. Other common notations for the IFI are
(2) <P(a; b; z) = M(a, b, z) = lFI(a; b; z).
There are certain properties of the IFI which follow from the fact
that it is a generalized hypergeometric function; these properties
*Fu('hsian equations and the ('on('ept of confluenee are treated in Chapters 5 and 7
of Rainville [2].
123
124 CONFLUENT HYPERGEOMETRIC FUNCTION [Ch.7
are not peculiar to the IF
1
By specializing results obtained in
Chapter 5, we obtain the following facts.
The function w = IF1(a; b; z) is a solution of the differential
equation
(3) [0(0 + b - 1) - z(O + a)] w = 0;
an equation which may also he written
d
o = z dz'
(4) zw" + (b - z)w' - aw = o.
If b is nonintegral, the general solution of (3) or (4) is
(5) w = A IFI(a; b; z) + BZI-b IF1(a + 1 - b; 2 - b; z),
with A and B as arbitrary constants. If b is intcgral, the general
solution may invoh'c log z in thc usual way, since z = 0 is a regular
singular point of the differcntial equation (3) or (4).
There is a canonical set of three relations between the IFI and
pairs of its contiguous functions. They may he written in the form
(6) (a-b+ 1) IF1(a; b; z) =a IFI(a+ 1; b; z) - (b-1) IFI(a; b-1; z),
(7) b(a+z) IFI(a; b; z) =ab IF1(a+ 1; b; z) - (a-b)z IFI(a; b+ 1; z),
(8) b IF1(a;b;z) =b IF
1
(a-1;b;z)+z IF
I
(a;b+1;z).
If Re(b) > Re(a) > 0,
(9)
F (a' b z) = reb) ileztta-l(1 - t)b-a-l dt
I I " rea) reb - a) 0
If neithcr a nor b is a non positive integer, if Rc(z) < 0, and if the
path of intcgration is one of Barnes' type, page 95,
(10) F ( . b' ) = ~ - J rea + s) r( -s)( -Z)3 ds.
1 1 a, ,z 27rir(a) B reb + s)
69. Kummer's first formula. We next obtain results which are
characteristic of the IFI in the sense that they are not held in common
by all pF/s.
Consider the product
70] KUMMER'S SECOND FORMULA
Since 1/(n - k)! = (-l)k( -n)kln!, we may write
_
~ ~ ( - n) k (a) k ( - 1) "z"
e-
Z
lFl(a; b; z) .t...... .t......
- n ~ O k=O ~ - r r - --nr
,n (-l)nz ..
= L 2F
l
( -n, a; b; 1)--,--
71=0 n.
But we already know (page (9) that
(b - a) ..
(1) J\( -n, a; b; 1) = -(6)--:'--
Then
00 (b - a),,( _z)n
e-
Z
lFl(a; b; z) = L ------, "-_.
n ~ O (b)"n.
THEOREM 42. If b is neither zero nor a negative integer,
(2)
This is Kummer's first formula.
125
70. Kummer's second formula. Examination of Kummer's
first formula soon arouses interest in the special case when the two
IFl functions have the same parameters. This happens when
b - a = a, b = 2a. We then obtain
lFI(a; 2a; z) = e
Z
lFl(a; 2a; -z),
or
(1)
More pleasantly, (1) may be expressed by saying that the function
e-
Z
IFI(a; 2a; 2z)
is an even function of z. Since, by straightforward multiplication,
00 n (a)k(-1),,-k2
k
z"
e-
z
lFl(a; 2a; 2z) L L
= n ~ O k ~ O (2a)kk!(n - k)!
= f: i: (a)k( -n)k2
k
_ (-z)n
n ~ O k ~ O (2a) kk! n!
00 (-z)"
= L 2F
l
( -n, a; 2a; 2)-"
n=O n.
we may conclude, from the fact that the left member is an e\-en
function of z, that for k a non-negati\-e integer
126 CONFLUENT HYPERGEOMETRIC FUNCTION [Ch.7
(2) 2F
I
( -2k - 1, a; 2a; 2) = 0,
and also that
(3)
rn Z2k
e-
Z
IFI(a; 2a; 2z) = 6, 2F
I
( -2k, a; 2a; 2)(2k)!'
Next let us determine a differential equation satisfied by the
functionw = e-
z
IFI(a;2a;2z). We know that y = lF1(a;b;x)isa
solution of the equation
d
2
y dy
(4) x-
2
+ (b - x) - - ay = 0.
dx dx
In (4) put b = 2a, x = 2z, and y = ezw. The result is
(5) zw" + 2aw' - zw = 0,
of which one solution must be w = e-
Z
IF I (a; 2a; 2z).
In equation (5) change the independent yariable to a = tz2 and
thus arrive at the equation
(6)
d
2
w dw
a
2
- + (a + ) a - - - aW =
da2 2 da '
or
(7) [8(8 + a + ! - 1) - aJ w = 0;
d
8 = a-'
ria
Equation (7) is a differential equation for the oFI function with
denominator parameter (a + !) and argument a = 1z2. Hence, if
a +! is nonintegral (that is, if 2a is not an odd integer), the
general solution of (7) is
(8) w = AoFI( -; a + !; 1z2) + B(Z2) i-a OFI( -;! - a; tz2).
But (7) is also satisfied by WI = e-
Z
IF I (a; 2a; 2z). Therefore there
exist constants A and B such that the right member of (8) becomes
e-
z
IF1(a; 2a; 2z). In the usual manner it is easy to see that B =
and A = 1.
THEOREM 43. If 2a is not an odd integer < 0,
(9)
If 2a is an odd positiye integer, the second term on the right in (8)
is replaced by a solution inyolying log z, and the same argument
again shows that B = 0, A = 1, so that (9) holds. Equation (9)
is known as Kummer's second formula.
701 KUMMER'S SECOND FORMULA
We may use (9) and (3) to conclude that
en Z2 k 0", Z2 k
L 2F!( -2k, a; 2a; 2)(2k)' = L 2
2k
k'( + 1_) ,
k=O k=O v. a 2 k
from which it follows that
(10)
2F!( -2k, a; 2a; 2) - __ (Vl'_.
- (a +
127
Theorem 43 may be interpreted as a relation bet ween a particular
IF! and the modified Bessel function of the first kind of index
(a -
Many other properties of the IFI will be found in Chapter 6 of
volume one of the I3ateman Manuscript Project volumes, Erdelyi
[1].
The \Vhittaker functions* lVk,m are expressible as linear com-
binations of IFI's. Two subsidiary solutions of the basic Whittaker
equation
(11)
d
2
W [ 1 k 1 - m2]
-+ W=O
dz
2
4 Z Z2
are, if 2m is not an integer,
Mk,m(z) = zm+ le-
lz
+ m - k; 2m + 1; z)
and Mk._m(z). See also Chapter 8 of Rainville [2] for more detail
on equation (11).
EXERCISES
1. The function
2 1%
erf(x) = -_ exp( -t
Z
) dt
V1l" 0
was defined on page 36. Show that
erf(x) = - x
z
).
2. The incomplete Gamma function may be definrd by the equation
'Y(a, x) = 1"e-tt
a
-
1
dt, Re(a) > O.
Show that
*See Whittaker and Watson [1] and Whittaker [1].
128 CONFLUENT HYPERGEOMETRIC FUNCTION [Ch.7
3. Prove that
[e-ZtFt(a; b; z) ] = (-l)k(b - a)ke-z tFt(a; b + k; z).
You may find it helpful to Kumnwr's first formula, TIH'orem 42.
4. Show that
1 fro
IF1(a; b; z) = f(a)o e-tt
a
-
t
oF1(-; b; zt) dt.
5. Show, with the aid of tlw result in Ex. 4, that
l"'CXp( _t2)t2a-n-tJn(zt) dt = (a; n + 1;
6. If k and n are non-negative integers, show that
F[-k, a + n;
a' ,
for k > n,
for 0 k n.
CHAPTER 8
Generating
Functions
71. The generating function concept. Consider a function
F(x,t) which has a formal (it need not converge) power series ex-
pansion in t:
ro
(1) F(x, t) = 'Lfn(x)t
n

n_O
The coefficient of tn in (1) is, in general, a function of x. We say
that the expansion (1) of F(x,t) has generated the set fn(x) and that
F(x,t) is a general1'ng function for the fn(x), If for some set of
values of x, usually a region in the complex x-plane, the function
F(x,t) is analytic at t = 0, the series in (1) conyerges in some region
around t = 0. Com'ergence is not necessary for the relation (1) to
define the fn(x) and to be useful in obtaining properties of those
functions.
Before proceeding to a discussion of some of the uses of generating
functions, we wish to extend the foregoing definition slightly. Let
en; n = 0, 1, 2, "', be a specified sequence independent of x and t.
We say that G(x,t) is a generating function of the set gn(X) if
ro
(2) G(x, t) = 'L cngn(x)t
n

If the C
n
and gn(X) in (2) are assigned, and we can determine the
sum function G(x,t) as a finite sum of products of a finite number of
known special functions of one argument, \ve say that the generating
function G(x,t) is known.
129
130 GENERATING FUNCTIONS [Ch.8
The question arises as to what is a "known special function." It
is, of course, a matter of opinion or conyention. \Ve consider as
known any function \vhich has receiyed individual attention in at
least one research publication. In this terminology we follow the
late Harry Bateman (1882-1946). Bateman, who probably knew
more about special functions than anyone else, is said to have
known of about a thousand of them.
The necessity for some terminology such as that defined aboye
can be appreciated after examination of certain publications purport-
ing to obtain new generating functions for classical polynomials.
Generating functions will playa large role in our study of poly-
nomial sets. For example, we shall define the Legendre poly-
nomials P n(X) by
en
(3) (1 - 2xt + t2)-\ = 2: Pn(x)t
n
,
n..--()
and the Hermite polynomials H n(X) by
(4)
00 H n(X)tn
exp(2xt - t2) = 2: ---,-'
n.
We shall find (page 201) that the Laguerre polynomials Ln(a)(x)
possess the generating relation
(5) etoF'r( -; 1 + a; -xt) = i:
(1 + a)"
One of our major problems will be the search for generating
functions for known polynomial sets. Certain purely manipulatiye
techniques will be found to accomplish much in this direction, but
more systematic attacks are highly desirable. Unfortunately the
box score to date reveals that no known systematic theory has
produced results comparable to those attained by manipulative
skill. Since the latter usually requires long practice and training,
it is hoped that in the future the tide will swing toward theoretical
developments capable of producing practical new results. Some
start in that direction has been made by Sheffer [1] (see Chapter
13) and Boas and Buck [lJ, the latter to be touched upon in this
chapter. See also Weisner [1].
We shall find that if polynomials !n(x) are generated hy
00
(1) F(x, t) = 2: !n(x)lll,
71=0
72] FUNCTIONS OF FORM G(2xt_t2) 131
certain properties of fn(x) are readily deduced from known prop-
erties of F(x,t). This idea will be used frequently in the study of
specific polynomials in Chapters 10, 11, 12, 16, 17, and 18. In the
present chapter we seek properties held in common by many poly-
nomial sets.
72. Generating functions of the form G(2xt - (2). Each of the
generating functions in (3) and (4) of the preceding section is a
function of the single combination (2xt - t2). By studying the
generating relation
co
(1) G(2xt - t2)
L gn(X)t
n
,
n=O
in which G(u) has a formal power-series expansion, we arrive at
properties held in common by Pn(X) and H,,(x)/n!, where F,,(x) is
the Legendre polynomial and I-l n(X) the Hermite polynomial. Let
(2)
Then
(3)
F = G(2xt - [2).
~ ' ! = 2tG'
ax '
aF
at
(2x - 2t) G',
in which the argument of G' is omitted b e e a u ~ e it remains (2xt - t2)
throughout. From equations (3) we find that the F of (2) satisfies
the partial differential equation
(4)
aF aF
(x - t) -- - t -- = o.
ax at
Since
F = G(2xt - t2)
it follows from (4) that
en = DO
L xg,,'(x)t
n
- L gn'(X)t
n
+! - L ngn(X)t
n
= 0,
n=O n=O n=O
or
co co ro
(5) L xg/(x)t
n
- L ng,,(x)t
n
= L O:'_I(X)t
n

{( -=-" 71=0 II --l
In (5), equate coefficients of tn to obtain the following result.
132 GENERATING FUNCTIONS [Ch.8
THEOREM 44. From
0>
G(2xt - t2) = L gn(X)t
n
n=O
it follows that go' (x) = 0, and for n 1,
(6) xgn'(x) - ngn(X) =
The differential recurrence relation (6) is common to all sets
gn(X) possessing a generating function of the form used in (1). For
the choice G(u) = (1 - u)-l, the gn(X) become the Legendre poly-
nomials P n(X), as stated in (3) of Section 71. Hence the P n(X)
satisfy the relation
(7) xP ,,'(x) - nP n(X) =
For the choice G(u) = exp(u), the g,,(x) become H ,,(x)/n! by
(4) of Section 71. Hence the Hermite polynomials satisfy the
relation
or
(8)
xH n'(x) nH n(X)
-nr-- n!

(n - 1)!'
xH n'(x) - nH ,,(x) = nH
73. Sets generated by etif;(xt). The generating function in (5)
of Section 71 suggests that we consider sets (Tn(X) defined by
0>
(1) etif;(xt) = L (Tn(X)t
n
.
n=O
For a short discussion of these polynomials, see also Rainville [4J.
Let
(2)
Then
(3) =
ax '
aF
(4) at = elif; + xet,p'.
We eliminate if; and y/ from the three equations (2), (3), (4), and
thus obtain
(5)
aF aF
x-- - t-- = -tF.
ax at
73] SETS GENERATED BY e1if;(xt) 133
Since
00
F = ei/;(xt) = L un(x)t
n
,
n=Q
equation (5) yields
00 00 00
L xu..'(x)t
n
- L nun(x)t
n
= - L un(x)t
nH
n_O n=O tt-zO
en
- L Un_l(X)t
n
,
n=l
from which the next theorem follows.
co
THEOREM 45.
and for n ~ 1,
From e1if;(xt) = L un(x)t
n
, it follows that uo'(x) = 0,
n=O
(6)
Next let us assume that the function if; in (1) has the formal
. .
power-senes expanSIOn
'10
(7) if;(u) = L "(nun.
,1=0
Then (1) yields
so that
(8)
N ow consider the sum
00 = n (C)n"{kxkt
n
~ (C)nUn(X)t
n
= , ~ { ; (n -=- Ie)!
134 GENERATING FUNCTIONS [Ch.8
THEOREM 46. From
00
etif;(xt) = L un(x)t
n
, if;(u)

it Jollows that Jor arbitrary c
(9)
in which
'"
(10) F(u) = L (C)n'YnU
n

7t=O
The role of Theorem 46 is as follows: If a set un(x) has a generat-
ing fundion of the form etif;(xt), Theorem 46 yields for un(x) another
generating function of the form exhibited in (9). For instance, if
if;(u) is a specified pFQ' the theorem gives for un(x) a class (c is ar-
bitrary) of generating functions involving a p+IF
Q
Furthermore, if
c is chosen equal to a denominator parameter of the original pFQ'
the second generating function becomes one involving a pF
Q
_
I

Let us now apply Theorems 45 and 46 to Laguerre polynomials.
As stated in Section 71, we shall show in Chapter 12 that the
Laguerre polynomials possess the generating relation
co Ln(a)(x)t
n
(11) e
t
oFI( -; 1 + a; - xt) = l; 0+ a),:
We use Theorem 45 of this section to conclude that Lo(a)(x) is a
constant, and for n 1,
or
( 12)

- 0+ a)n-I'
In applying Theorem 46 to the Laguerre polynomials, note that
un(x) = Ln (a)(x)/(l + a)n and that
00 (-l)nu
n
if;(u) = oFI( -; 1 + a; - u) = L
+ a)n
Then 'Yn = (-l)n/[n!(l + a)n], and
00 (-l)n(c)nU
n
F(u) = L = IFI(c; 1 + a; -u).
11=0 n. a n
74] FUNCTIONS A(t) exp[ -xtj(l- t)j 135
Therefore Theorem 46 yields
(13)
(
__ (C)nLn(a)(x)t
n
,
(1 - t) -c IFI c; 1 + ex; L...
1 - t (1 + ex)n
a class of generating relations for Ln(a)(x).
In (13) the choice C = 1 + ex is appealing. With that choice we
obtain
(14) (1 - t)-I-a exp = L Ln1a)(x)t
n
.
(
t) ,,,
1 - t
74. The generating functions A(t)exp[ -xtj(l - t)1. Equation
(14) suggests that we consider sets Yn(X) generated by
(1)
(
-xt) 00
A(t) exp 1=-t = Yn(X)t
n

From
(2)
(
-xt)
F = A(t) exp 1=-t
it follows that
(3)
aF
(1 - t) ax = -tF.
Hence
00 00
L Yn'(X)t
n
- L Yn'(X)t
n
+
1
= - L Yn(X)t
n
+
l
,
n=O n=O 11=0
which readily yields Yo'(x) = o. and for n 1,
(4)
Now (3) can be rewritten as
(5)
so that we obtain
aF
ax
t
---F
1 - t
yn'(x)t
n
= -(i; Yn(X)t
n
)
00 n
- L L Yk(X)t
n
+
1

m tI-l
- L L Yk(X)t
n

n=1 k=O
136 GENERATING FUNCTIONS [Ch.8
Hence, for n ~ 1,
n-l
(6)
Of course (6) can also be obtained from (4) by iteration and
summation.
THEOREM 47. From
(
t) m
A (t) exp 1 - ~ t = ~ Yn(X)t
n
1f follows that Yo'(x) = 0, and for n ~ 1,
(4)
n-l
(6) Yn'(X) = - L Yk(X).
k ~ ( )
Since, by (14) of the preceding section, the choice A (t) = (1- t)-I-a
yields Yn(X) = Ln (a)(x), we have shown that the Laguerre poly-
nomials satisfy, for n ~ 1,
(7)
and
(8)
In equations (7) above and (12) of Section 73 we have two differential
recurrence relations for Ln (a)(x). These polynomials are com-
pletely determined by the two relations once Lo(a)(x), a constant, is
specified. The value of Lo(a)(x) is easily found by putting t = 0
in a generating relation. Indeed, Lo(a)(x) = 1. Thus ,ve see that
the Laguerre polynomials are essentially determined by the fact
that they have both a generating function of the form
e Iy;(xt)
and a generating function of the form
(
-xt)
A (t) exp 1=-t
without specification of the functions y; and A. For more detail
on Laguerre polynomials see Chapter 12.
75] CLASS OF GENERATING FUNCTIONS 137
75. Another class of generating functions. Later we shall en-
counter many polynomial sets each of which has a generating func-
tion of the form next to be considered. Let -.fCu) have a formal
power-series expansion
co
(1) -.f(u) = L.: "InU
n
, "10 O.
""",,0
Define the polynomials f,,(x) by
(2)
(
-4xt) '"
(1 - t)-'-.f ---2 = L.: f,,(x) tn.
(1 - t)
THEOREM 48. The polynomials fn(x) defined by Cl) and (2) have
the following properties:
(3) f (x) = (C)n f
n n! + '
(4) xn = f (-1) k(C + 21j)/k(x)
22n"ln (n - k) !(C)n+k+l '
(5) xf/(x) - nfn(X) = - (c+n-l)fn-1Cx) -xfLl(X), n 1,
n-l n-l
(6) xfn'(x) - nfn(x) = -c L.:h(x) - 2x L.:f/(x), n 1,

n-l
(7) xfn'(x) - nfn(x) = L.: (-I)n-k(c + 2k)h(x), n 1.

For c = 1, equations (3) and (5) appear in Sister Celine's work,
Fasenmyer [1].
Proof: To obtain (3), consider
co co (_ 4) k"l kxktk
L.:fn(x)t
n
= L.: -(I-=-'--t)c+2k
n=O k=O
=:f: i:
hO (n - k)!(ZC)k(ZC + Z)k
from which (3) follows by equating coefficients of tn.
Next, in (2) put
-4t
(f-=-t)2 = v.
138 GENERATING FUNCTIONS [Ch.8
Then
t 1 2 -v
= - 1 + VI-=-v (1 + Vl=-V)2
and (2) becomes
if; (Xi') = ( 2 -),. f _
1 + V1=-; (1 + vi - 1')2k
or
(
xv) = ..... 2 )C+2k.
if; 0rk ----
1 + ,/1 - v
In Ex. 10, page 70, we found that
(8)
(
_2_)21'-1 _ F[I" l' - i;
1 + VI ....:..Z; - 2 1 21';
The use of (8) with 21' = C + 2k + 1 leads to
cr [HI + C + 2k), He + 2k);
if;(xv) = L 2F1
hO 1 + e + 2k;
Therefore
which yields equation (4).
In order to (5), (6), and (7), put
(9)
Then
( 10)
aF
ax
-4l(1 - t)-c-2if;',
75] CLASS OF GENERATING FUNCTIONS 139
(11)
aF
at = e(1 - t)-c-lif; - 4x(1 + t)(1 - t)-c-3if;'.
Therefore F satisfies the partial differential equation
(12)
aF aF
x(1 + 0ilx - t(1 - i)-at = -etF.
Equation (12) can be put in the forms
(13)
aF aF ,.,rJF aF
x-- - t - = - elf; - 1---- - xl--
ax at at ax'
(14)
aF aF - ct 2:rt aF
x-- - t-- = ----- F - -----
ax at 1 - t 1 - t ax'
( 15)
aF aF -d, 2t
Z
riF
x- - - t = ---J< - ---- __ c
ax at 1 + t 1 + t at
Since
C ~
F = .L: fn(x)t",
n_O
equation (13) yields
co
.L: [X!n'(X) - nfn(x)Jt
n
n=O
co co co
= -c .L:f,,(x)t
n
+
1
- .L: nfn(x)t
n
+
1
- .L: xf,,'(x)t
n
+
1
Tt=1) 7!=IJ 11-,-,11
~ ~
= - .L: (c + n - 1)!n_l(X)t
n
- .L: :rf:_I(x)t",
11=1 n=l
which leads to (.5).
Equation (14) yields
co
.L: [X!n'(X) - nfn(X)Jt
n
n=O
co n m n
= -c.L: .L:fk(X)t
n
+
1
- 2x .L: .L:f;'(X)t"+1
11=01.=0 11=0/,=0
m n-1 CD n-l
-c.L: .L:fk(X)t
n
- 2x .L: .L:f/(x)t
n
,
7l=1 k""Q 1t=1 k=O
which leads to (6).
140 GENERATING FUNCTIONS [Ch.8
From (15) we obtain
co
L: [X!n'(X) - nfn(x)Jt
n
n_O
co n
= - L: L: (-I)n-k(c + 2k)h(x)t
n
+l
n-O
m n-l
= L: L: (-l)n-k(c + 2k)!k(X)t
n
,
n=-<l k=Q
which gives (7).
Equation (7) was obtained by Dickinson [1J by a somewhat
different method.
76. Boas and Buck generating functions. In 1956 Boas and
Buck [lJ studied a large class of generating functions of polynomial
sets. Some of their work appeared also in their earlier mimeo-
graphed reports which are not generally available. A rough state-
ment of one of the main results in Boas and Buck [lJ is that a
necessary and sufficient condition for the polynomials Pn(X) to have
a generating function of the form
co
(1) A(t)>/I(xH(t) = L: pnex)tn
11=0
is that sequences of numbers Cik and fh exist such that, for n 1,
n-I n-I
(2) Xpn'(X) - npn(X) = - L: Cikpn_I_k(X) - X L:
k_O
We now present, with minor variations in notation, that part of
Boas' and Buck's work which we wish to have available for later
chapters.
Let
co
(3) >/I(t) = L: /'nt
n
, /'0 0,
n=O
co
(4) A(t) = L: ant
n
, ao 0,
11=0
co
(5) H(t) = L: h t
n
+
I
n , ho 0.
n=O
76] BOAS AND BUCK FUNCTIONS 141
THEOREM 49. If Pn(X) is defined by (1), with (3), (4), and (5)
holding, Pn(X) is a polynomial in x and Pn(X) is of degree precisely n if
and only 1j "I n O.
Proof: Put
00
(6) Pn(X) L s(k, n)x
k


Then
m
A(t)if;(xIl(t)) = L s(k, n)xktn,
n. k=O
SO that m differentiations with respect to x, followed by our putting
x = 0, yield
00
(7) A(t)[H(t)]mif;(fflJ(O) = L m! sCm, n)tn.
n-O
Because of (3), (4), and (5),
00
(8) A(t)[H(t)]my;(mJ(O) = aohommhmtm + L C(m, n)tn,
in which the precise nature of C(m,n) is not important to us.
Comparison of (7) and (8) leads to
(9)
(10)
sCm, n) = 0 for n < m,
The condition (9) shows that Pn(X) is a polynomial of degree n.
The condition (10), with m replaced by n, shows that pn(X) is of
degree precisely n if and only if "In 0, since aoho 0 by (4) and
(5).
THEOREM 50. For the polynomials Pn(X) defined by (1), with (3),
(4), and (5) holding ,and "In 0, there exist sequences of numbers IXk
and (3k such that,jor n ;?; 1,
n-I n-l
(2) XPn/(X) - npn(x) = - L IXkpn-l-k(X) - X L

Indeed,
tA'(t)
00
(11)
A(t) =
L IXntn+l,
11=0
tH'(t)
00
(12)
B(t)
1 + L {3nt
n
+l.
n_O
142 GENERATING FUNCTIONS [Ch.8
Proof: Put
(13) F =
Then
(14) i!E =
ax
(15)
= +
As usual, we eliminate and with the aid of equations (13),
(14), and (15). The result may be written in the form
(16)
xtH'(t) aF aF tA '(t)
._._--- . -- - t-- = - - -- ---- . F
H(t) ax at A (l) .
If we define a" and fJ" by (11) and (12) and recall that
00
F = L Pn(x)t
n
,
,,-=:O
equation (16) leads us to
[1 + fJntn+l][j; XPn'(X)t
n
] - npn(X)t
n
= - p,,(x)t"J,
or
00
(17) L [xpn'(x) - npn(X)]t
n
n=O
ro n 00 n
- L L akPn_k(X)tn+l - XL L
n=O k=O
00 n-l OJ 1
= - L L akPn_I_k(X)t
n
- X L L
n=1 k=O
from which
n-I n-I
(2) xp,,'(x) - npn(X) = - L akPn-l-k(X) - X L fJkP;.-l-k(X)
k=O
follows at once. It is important that the ak and fJk in (2) are in-
dependent of n.
EXAMPLE: Consider the polynomialsf,,(x) of Section 75 in which
(
-4xt) ro
(18) (1 - "(T-=-t)2 = ;fn(x)tn.
77] AN EXTENSION
The fn(x) fit into the Boas and Buck theory with
-4t
A(t) = (1 - t)-c, H(l) (1--=-t)2'
tA'(l) = f ct n +!
A(t) n ~ O '
tH'(t) en
H(tf = 1 + ; 2tn+1.
Henee an = c, f3n = 2, and the relation (2) hecomes
n-l n-l
143
which is equation (6) of Theorem 48, page 1:37, with the right
member written in reverse order. Anyone of equations (5), (6),
and (7) of Theorem 48 can be obtained from any other. For the
Boas and Buck generating function, the results corresponding to (5)
and (7) of Theorem 48 are complicated and are therefore omitted.
The Boas and Buek work applies to the polynomials considered
in Sections 73, 74, 75 but not to those of Section 72.
77. An extension. Consider the generating relation
'"
(1) A(t)if;,xH(t) + get) = Lfn(x)t
n
n_O
in which
'"
(2) 1/;(t)
= L 'Ynt
n
, 'Yo ~ 0,
n=O
'"
(3) A(t) = L ant
n
, ao ~ 0,
n=O
co
(4) H(t) = L hntn+l, ho ~ 0,
n ~ O
and
'"
(5) get) = L g"t
n
+2.
Tt=O
Note that g( t) is permitted to be identically zero. It is not neces-
sary to require that g'(O) = 0, but this involves no loss of generality,
as can be seen by employing a translation in the x-plane.
THEOREM 51. If fn(x) is defined by (1) with (2), (3), (4), and (5)
holding, fn(x) is a polynomial in x, and fn(x) is of degree precisely n
if and only if 'Yn ~ 0.
144 GENERATING FUNCTIONS [Ch.8
Proof: We parallel the proof of Theorem 49. Put
00
(6) fn(x) = L s(k, n)x
k


Then
=
A(t)if;(xH(t) + L s(k, n)xktn,
n
from which we obtain
en
(7) A(t)[Il(t)]"'1f(m)(g(t = L m!s(m, n)tn.
n_O
Because of (2), (3), (4), and (5),
00
(8) A(I)[H(t)]mif;(m)(g(t = aohommhmt
m
+ L C(m, n)t
n
,
Tl=m+l
in which the nature of C(m,n) is, fortunately, unimportant to us.
Comparison of (7) and (8) leads to
(9) s(m,n) = 0 for n < m,
(10)
from which the conclusions in Theorem 51 follow.
THEOREM 52. For the polynomials fn(x) defined by (1), with (2),
(3), (4), and (5) holding, and 'Yn 0, there exist sequences of numbers
ak, 13k and lh such that, for n 1,
Indeed,
(12)
(13)
(14)
Proof: Put
(15)
Then
n-1
tA'(t) = i: a tn+!
A(t) n ,
n-1
tH'(t) = 1 + i: (3 tn+!
H(t) n ,
tg'(t) = i: 0 in+!
H(t) n
F = A(t)if;(xH(t) +
77] AN EXTENSION
145
(16) = A(t)H(t)1/;',
(17) = A'(t)1/; + A(t)[xH'(t) + g'(t)]1/;'.
Eliminate 1/; and 1/;' from (15), (16), and (17) to obtain
(18) [xtH'(t) + tg'(t)] _ = _ tA 'J!)F
H(t) H(t) ax at A(t)
Since
'"
F = LJ,,(X)t
n
,
fI=(j
it follows from (18) with the aid of (12), (13), and (14) that
(1 + tu xin'(x)t
n
) + 0,,[71
- nfn(x)t
n
= ant'dl)(tufn(X)t
n
).
Therefore,
00
L [xfn'(x) - nin(x)Jt
n
n=O
00 n
= - L L [(X!1k + + a4n_k(X)]t
n
+
1
,
n=O k=O
from which (11) follows after a shift from n to (n - 1) on the right.
The polynomials gn(X) of Section 72 fit into the above scheme
with an = 0, !1n = 0, 00 = -1, and On = for n 1.
EXERCISES
00
1. From etif;(xt) = L <Tn(x)t
n
, show that
n_O
and in particular that
2n<TnOX) = '"
2. Consider the set (called Appell polynomials) an(x) generated by
00
extA(t) = L an(x)t
n
.
n=O
Show that a'o(x) = 0, and that for n 1, a/ex) = an_lex).
146 GENERATING FUNCTIONS [Ch.8
3. Apply Theorem 50, page 141, to the polynomials un(x) of Section 73 and thus
obtain Theorem 45.
4. The polynomials Un (X) of Ex. 3 and Section 73 are defined by
00
(A) etf(xt) = L un(x)t
n
,
n_O
but by equation (9), page 134, they also satisfy
(B)
(
xt) 00
(1 - O-c F 1 _ t =?;u (C)nUn(xW,
for a certain fUllction F. By applying Theorem 50, page Ill, to (B), ('onclude that
the un(x) of (A) satisfy the relation
n--l
(C)n[XUn'(X) - nun(x)] = - L (CMClTk(X) + xu/ex)],
k=O
for arbitrary c.
5. Apply Theorem 50, page 111, to the polynomials Yn(X) defined by (1), page
135. You do not, of course, get Theorem 47, since that theorem depended upon
the specific character of the exponential.
6. Apply Theorem 50 to the Laguerre polynomials through the generating rela-
tion (14), page 135, to get
n-l
xDLn(a)(x) - nLn(")(x) = - L [(1 + ex)Lk(a)(x) + xDLk(a)(x)],

in which D = d/dx. Use the above relation in conjunetion with equation (8),
page 136, to derive the differential equation
xDZLn(a)(x) + (1 + ex - x)DLn(a)(x) + nLn(a)(x) = 0
for the Laguerre polynomials.
7. The Humbert polynomials hn(x) are defined by
00
(1 - 3xt + t
3
)-' = L hn(x)t
n

=0
Use Theorem 52, page 144, to conclude that
xhn'(x) - nhn(x) =
8. For the Yn(X) of Section 74 show that
(
-xt)
F = A(t) exp "l=-i
satisfies the equation
x
ap
_ tap = _tZap _ Q - t)tA'(t)p
ax at at A(t)
and draw what conclusions you can about Yn(x).
9. For polynomials an(x) defined by
(1 - = I: an(x)t
n
1 - t
obtain what results you can parallel to tho,p ()f Thcorpm 4R, page 137.
CHAPTER 9
Orthogonal
Polynom ials
78. Simple sets of polynomials. A set of polynomials {'Pn(X) I ;
n = 0, 1, 2, ... , is called a simple set if 'Pn(X) is of degree precisely
n in x so that the set contains one polynomial of each degree. One
immediate result of the definition of a simple set of polynomials is
that any polynomial can be expressed linearly in terms of the ele-
ments of that simple set.
THEOREM 53. If I 'Pn(X) I is a simple set of polynomials and 1f
P(x) is a polynomial of degree m, there exist constants Ck such that
(1)
The Ck are functions of k and of any parameters involved in P(x).
Proof: Let the highest degree term in P(x) be a",x
m
, and the
highest degree term in 'Pm (x) be bmxm. Note that b
m
~ O. Form
the polynomial
(2) P(x) - Cm'Pm(X)
in which C
m
= am/b
m
The polynominal (2) is of degree at most (m-l).
On this polynomial use the same procedure as was used on P(x),
thus reducing the degree again. Iteration of the process yields (1).
79. Orthogonality. Consider a simple set of real polynomials
'Pn(X). If there exists an interval a < x < b and a function w(x) >0
on that interval, and if
147
148 ORTHOGONAL POLYNOMIALS [Ch.9
fbW(X)tpn(X)cpm(X) dx = 0,
a
m r!' n, (1)
we say that the polynomials tpn(X) are orthogonal with respect to the
weight function w(x) ovcr the intcrval a < x < b. Because we
have taken w(x) > and CPn(X) real, it follows that
fbwex)cpn2(X) dx r!' 0.
a
With due attention to conver!!;ence, either or both endpoints of
the interval of orthogonality may be taken to be infinite. The
concept of orthogonality used hcre has been cxtended in many
directions, but the simple version abo\"c is all we use. A lar!!;e
number of thc sets of polynomials encountcrcd later in thc book arc
orthogonal sets. The limits of integration in (1) are important but
the form in which the interval of orthogonality is stated (opcn or
closed) is not vital.
80. An equivalent condition for orthogonality. The following
theorem is of use in our study of polynomial scts.
THEOREM 54. If the CPn(X) form a simple set of real polynomials
and w(x) > on a < x < b, a necessary and sufficient condition that
the set tpn(X) be orthogonal with respect to w(x) over the interval a< x <b
1'S that
(1) k = 0, 1,2"", (n - 1).
Proof: Suppose (1) is satisficd. Since x
k
forms a simple set,
there exist constants b(k,m) such that
m
(2) CPm(X) = L b(k, m)x
k

k=O
For the moment, let m < n. Then
since m, and thercfore cach k, is lcss than n. If m > n, interchan!!;e
m and n in the above argument. We have shown that if (1) is
satisfied, it follo'ws that
(3) f.bW(X)cpn(X)cpmex) dx = 0, m r!' n.
81] ZEROS OF ORTHOGONAL POLYNOMIALS 149
Now suppose (3) is satisfied. The 'Pn(x) form a simple set, so
there exist constants a(m,k) such that
k
(4)
X
k
= L a(m, k)'Pm(x).
m ~ O
For any k in the range k < n
ibW(X)Xk'Pn(X) dx = to a(m, k) ibW(X)'Pm(X)'Pn(X) dx = 0,
since m ~ k < n so that m ~ n. Therefore (1) follows from (3),
and the proof of Theorem 54 is complete.
From Theorem 54 we obtain at once that the orthogonal set
'Pn(X) has the property that
(5) jbW(X)'Pn(X)P(x) dx = 0,
a
for every polynomial P(x) of degree <no
It is useful to note that since
jbW(X)'Pn2(X) dx ~ 0,
a
it follows that also
(6) jbW(X)xn'Pn(X) dx :;e 0.
a
81. Zeros of orthogonal polynomials. Certain elementary in-
formation about the location of the zeros of any set of real orthogonal
polynomials is easily obtained.
THEOREM 55. If the simple set of real polynomials 'Pn(X) 1'S
orthogonal with respect to w(x) > over the interval a < x < b, the
zeros of 'Pn(X) are distinct and all lie in the open interval a < x < b.
Proof: Since, for n > 0,
ibW(X)'Pn(X) dx = 0,
the integrand must change sign at least once in the open interval
a < x < b. Since w(x) > 0, 'Pn(X) must change sign at least once
in a < x < b. Let the polynomial 'Pn(X) change sign at precisely
the points aJ, a2,"', a. in a < x < b. The a's are the zeros of
odd multiplicity of 'Pn(X) in a < x < b. Since 'Pn(X) is a poly-
nomial, s ~ n. N ow form the polynomial
150 ORTHOGONAL POLYNOMIALS [Ch.9
.
ifi(x) = II (x - ai).
i=l
If 8 < n,
(1)
since ifi(x) is a polynomial of degree less than n. But the integrand
in (1) cannot change sign in a < x < b because 'P,,(x) and ",,(x)
change sign at precisely the same points and w(x) > O. Therefore
s < n is impossible, and we must have ,<; = n. Thus 'Pn(X) has n
roots of odd multiplicity in a < x < b. Since 'Pn(X) is a polynomial
of degree n, it has exactly n roots, multiplicity counted, so that its
roots are distinct and all lie in a < x < b.
82. Expansion of polynomials. Let J(x) and hex) be any two
functions for which the integrals to be involved exist, and let an
interval a < x < b and a weight function w(x) > 0 on that interval
be stipulated. We define the symbol (f, h) by
(1) (J, h) = fbw(x)J(x)h(x) dx.
a
The symbol (f, h) has the properties
(J, h) = (h, J),
(fl + J2, h) = (fl, h) + (f2, h),
(eJ, h) = e(f, h), for constant c,
(fg, h) = (f, gh).
For a simple set of real polynomials 'Pn(X) orthogonal with respect
to w(x) on the interval a < x < b, we already know that
(2) m ~ n,
and
(3)
For convenience let us also define a sequence of numbers gn by
(4)
Theorem 53, page 147, becomes particularly pleasant when the
'Pn(X) form an orthogonal set, for we can then obtain a simple
formula for the coefficients in the expansion.
83] THREE-TERM RECURRENCE RELATION 151
THEOREM 56. Let <Pn(X) be a simple set of real polynomials orthog-
onal with respect to w(x) > 0 over the interval a < x < b, and let
P(x) be a polynomial of degree m. Then
m
(5) P(x) = L: Ck<Pk(X),

in which C" = gk-1(P, <Pk); that is,
lhw(x)p(x)<pk(X) dx
Ck = -- -_. ---
fbW(X) <Pk
2
(X) dx
(6)
Proof: We know the expansion (5) exists (Theorem 53). From
(5) we obtain, for 0 n m,
fbW(X)p(X)<pn(X) dx = 'to CkibW(X) <Pk(X) <pn(X) dx,
or
m
(7)
L: Ck(<Pk, <Pn),

from which, by (2) and (4),
(P, <Pn) = Cn(<Pn, <Pn) = Cngn
.
Thus C
n
= gn-1(P, <Pn) which is equi\'alent to (3),
The statement in equation (5) of Section 80 can now be
by writing
(8) (P, <Pn) = 0
for every polynomial P of degree < n.
83. The three-term recurrence relation. Eyery orthogonal set
of polynomials possesses a three-term recurrence relation of a
simple nature.
THEOREM 57. If <pn(X) is a simple set of real polynomials orthog-
onal with respect to w(x) > 0 on a < x < b, there exist sequences of
numbers An, B
n
, C
n
such that for n 1,
(1) x<pn(X) = An<Pn+l(X) + Bn<pn(x) + Cn<Pn-l(X)
in which An 0 and C
n
O.
152 ORTHOGONAL POLYNOMIALS [Ch.9
Proof: Since X'f'n(X) is a polynomial of degree (n + 1), we know
by Theorem 56 that
n+l
L a(k, n)'f'k(x),

in which
a(k,n) = gk-
1
(X'f'n, 'f'k) = gk-
1
('f'n, X'f'k).
By (8) of Seetion 82 we see that (<f!n, X<f!k) = 0 for k < (n - 1).
Therefore the relation (1) of Theorem 57 exists.
If An were zero for any n, the right member of (1) would be of
degree and the left member of degree (n + 1). lIence An O.
We still must show that Cn o.
It is now convenient to introduce a symbol 7f
m
to denote a poly-
nomial of degree m. The symbol 7f m is not to stand for a specific
polynomial but merely to stipulate that the degree of the poly-
nomial does not exceed the subscript used. When 7f
m
occurs more
than once in a discussion, there is no implication that the poly-
nomials indicated are related in any manner other than the fact
that none of them is of degree > m.
Let h
n
denote the leading coefficient in our 'f'n(X). Then
(2)
and h
n
0 because <f!n(X) is of degree precisely n. Now
(3) X'f'n-l(X) = + 7f n _l,
as can be seen by examining the leading coefficient on each side.
In equation (1) of Theorem 57 we know, by Theorem 56, that
Cn = 'f'n-l) = X'f'n-l)
so that by (3),
C
n
= ('f'n, 0C
1
<f!n) + ('f'n, 7f
n
-l) ] = <f!n)
from which
(4)
Thus Cn 0 for n 1, and the proof of Theorem 57 is complete.
That Bn in Theorem 57 can be zero, even for all n, will be seen in
specific examples in later chapters.
841 CHRISTOFFEL-DARBOUX FORMULA 153
By comparison of leading coefficients in equation (1) we find that
(5)
Now (1) can be written
X'Pn(X) = + Bn'Pn(x) +
n+1 gn-l n
or in even more promising form as
(6)
in which the coefficients of 'Pn! J(x) and 'Pn-I(X) are the same except
for a shift of index.
In our treatment of specific polynomial sets in later chapters we
shall obtain the pure recurrenee relations explicitly and thus have
no direet need for Theorem 57. That theorem is useful in the
general discussions, as in Rection 84, and is a powerful tool for show-
ing that a polynomial set is not an orthogonal set. If a set of
polynomials if;n(x) does not possess a three-term recurrence relation
of the form in Theorem 57, the set if;n(x) is not an orthogonal set.
A widely known theorem of Favard [1J states essentially that
any real polynomial set whieh satisfies a pure recurrence relation of
the type in Theorem 57 is orthogonal with respect to some weight
function over some intenoal with Sticltjes (not necessarily just
Riemann) integration used. Ko method of finding the weight func-
tion and interval is given. Recently Dickinson, Pollak and Wannier
[1J have, for a special subclass of such polynomials, constructively
shown orthogonality over a denumerable set of points.
84. The Christofiel-Darboux formula. In equation (6) of the
preceding section put
(1)
The three-term recurrence relation may then be written
(2) gn-1X'Pn(X) = tn'Pn+I(X) + gn-1Bn'Pn(X) + In-l'Pn-l(X),
Then
gn -lX'Pn(X) 'Pn(Y) = tn'Pn+l (x) 'Pn(Y) +In-I'Pn(Y) 'Pn-l (x) +gn -I Bn'Pn (x) 'Pn (y)
and
154
ORTHOGONAL POLYNOMIALS [eh.9
gn -IY'Pn(X) 'Pn(Y) = tn'Pn+1 (y) 'Pn(X) +tn-I'Pn(X) 'Pn-I (y) + gn -1 Bn'Pn(X) 'Pn(Y)
from which it follows that
gn -ley - X)'Pn(X)'Pn(Y) = tn['Pn+I(Y)'Pn(X) - 'Pn+I(X)'Pn(Y)]
- In-I['Pn(Y)'Pn-I(X) - 'Pn(X)'Pn-I(Y)].
Next put
(3) }n(X, y) = In['Pn+I(Y)'Pn(X) - 'Pn+I(X)'Pn(Y)].
Then
so that
(4) L gk-I(y - X)'Pk(X)'Pk(Y) = }n(X, y) - }n(X, y).
k-I
Now, by (3),
}o(X, y) = to['PI(Y)'PO(X) - 'PI(X)'PO(Y)]
and 'Po(x) = 'Po(Y) = ho, a constant. Also to = hogo-ihi-l. Let
'Pl(X) = hlx + c, in which c is constant. Then
}o(x, y) = hogo-lhI-I[(hIY + c)ho - (hlx + c)h
o
]
= h
o
2
g
o
-
I
(y - x) = go-I(y - X) 'PO(X)'Po(Y)'
We then transfer the }o(x, y) term from the right to the left in equa-
tion (4) and obtain
(5) L gk-I(y - X)'Pk(X)'Pk(y) = jn(X, y).

Using (1) and (3) in (5), we arrive at the cle:;;ired result, the Christof-
fel-Darboux formula, equation (6) of the following; theorem.
THEOREM 58. Let 'Pn(X) be a simple set of real polynomials
orthogonal with respect to w(x) > 0 on a < x < b. Let hn be the
leading coefficient in 'Pn (x) so that
and let
Then
(6) t gk-1'Pk(X)'Pk(Y) h
n
'Pn+I(Y)'Pn(X) - 'Pn+l(X)'Pn(Y)
hO = g"h,= . ----- --
85] NORMALIZATION; BESSEL'S INEQUALITY 155
85. Normalization; Bessel's inequality. For theoretical dis-
cussions it is convenient to replace the orthogonal polynomials
<pn(X) by polynomials
(1) if.tn(X) = gn - !<Pn(X)
which are called orthonormal polynomials. Note that
(if.tn, if.tn) = (g.- !<Pn, gn -l<pn) = g"-I(<p,,, <Pn) = 1.
The usc of (1) increases the neatness of many formulas. When
specific polynomials are being used, the normalization process is of
little help. We shall concentrate on specific polynomials in most
of the later chapters.
Let if.tn(X) be an orthonormal polynomial set over the interval
(a, b) with weight function w(x) > O. Let
(2)
where
(3) Ck = (f, if.tk) = fbW(Y)fCY)if.tkCY) dy,
a
fey) as yet unrestricted except that the integrals involved exist.
Consider
(4) fbW(X)[J(x) - Sn(X)J2 dx
a
Now
ibw(x)feX)Sn(X) dx = ckibW(X)f(X)if.tk(X) dx = Ck
2
and, since the if.tnex) form an orthonormal set,
ibW(X)Sn2(X) dx = ckif.tkeX) JdX
It follows that equation (4) becomes
(5)
f
b fb n
a WeX) [j(x) - Sn(x)J
2
dx = a w(x)f2ex) dx - 6 Ck
2

156 ORTHOGONAL POLYNOMIALS [Ch.9
Since the left member of (5) is never negative,
(6)
which is Bessel's inequality. The right member of (6) is inde-
'"
pendent of n. Hence L c
n
2
converges, and it follows that Cn ~ 0
nsO
as n ----t 00.
THEOREM 59. If the polynom'ials t/;n(X) form an orthonormal set
on the 'interval (a, b) with respect to the wwight function w(x) > 0 and
ifibW(X)p(X) dx exists,
(7) ~ l ~ ibwex)fex)t/;n(X) dx = O.
If we wish to state the result corresponding to (7) for orthogonal
polynomials 'Pn(X) , not necessarily normalized, we need merely to
define
gn = ibW(X)'Pn2(X) dx
and then replace (7) by
(8) Lim gn -!fbw(x)feX)'Pnex) dx = O.
n+co a
For much additional material on general orthogonal polynomials
the reader should consult the following: Szego [lJ, Chapter 10 of
ErdeIyi [2J, Jackson [lJ, and Shohat [1].
CHAPTER 10
Legendre
Polynomials
86. A generating function. We define the Legendre polynom-
ials P n(X) by the generating relation
00
(1) (1 - 2xt + = L P n(X)t
n
,
in which (1 - 2xt + t
2
)-, denotes the particular branch which
-+1 as t -+ O. We shall first show that P n(X) is a polynomial of
degree precisely n.
Since (1 - z)-a = IFo(a; - ; z), we may write
= " (Dn(-I)k(2x)n-ktnH
= 6 "'!(n - I.:)!
= i=[' (-I)kO)n_k(2x)n-
2k
i
n
I.:!(n - 21.:)! '
by equation (13), page 58. We thus obtain
(n/2) ( -1) n_k(2x)n-2k
(2) Pn(X) = 6 I.:!(n - 2k)! '
from which it follows that P n(X) is a polynomial of degree pre-
cisely n in x. Equation (2) also yields
157
158 LEGENDRE POLYNOMIALS [Ch.10
(3)
in which 7rn_2 is a polynomial of degree (n - 2) in x.
If in (1) we replace x by (-x) and t by (-t), the left member
does not change. Hence
(4)
so that P n(X) is an odd function of x for n odd, an even function of
x for n even. Equation (4) follows just as easily from (2).
In equation (1) put x = 1 to obtain
en
(1 - t)-l = L Pn(l)t
n
,
n"" 0
from which
(5)
which combines with (4) to give
(6)
From (1) with x = 0, we get
co
(1 + t2)-! = L: p,JO)t
n

11=0
But
Hence
(7)
results just as easily obtained directly from (2).
Equation (2) yields
(8)
and from (8) it follows that
(9) P
' (0) 0 P' (0) _ (_l)n2(!)n+l _ (_1)n(!)n.
2n =, 2n+1 - n! - n!
87. Differential recurrence relations. We already know from
Section 72 that the generating relation
88] THE PURE RECURRENCE RELATION 159
00
(1) (1 - 2xt + = L: P n(X)t
n
n=O
implies the differential recurrence relation
(2)
From (1) it follows by the usual method (differentiation) that
m
(3) (1 - 2xt + t
2
)-i = L: p,,'(x)ln-I,
n",,1
'n
(4) (x - t)(l - 2xt + = L: nPn(x)t
n
-
l

11=1
Since 1 - t2 - 2t(x - t) = (1 - 2xt + [2), we may multiply the
left member of (3) by (1 - t2), the left member of (4) by 2t, sub-
tract and obtain the left member of (1). In this way we find that
00 00 ro m
L: Pn'(X)t
n
-
1
- L: Pn'(x)t
n
+
1
- L: 2nP
n
(x)t" = L: Pn(X)t",
n=1 n=1 n=1
or
00 00 00
L: - L: P;,_I(X)t
n
= L: (2n + I)Pn(x)t
n

n=O n=2 n=O
We thus obtain another differential recurrence relation
(5)
Equations (2) and (5) are independent differential recurrence
relations. From (2) and (5) other relations may be obtained, each
useful in various ways. By combining (2) and (.5), we find that
(6)
Next in (6) shift index from n to (n - 1) and sub:-;titute the re:-iult-
ing expression for P Ll(X) into (2) to obtain
(7) (x
2
- 1) Pn'(x) = nxPn(x) - nPn_1(x).
88. The pure recurrence relation. The relation (7) abo\"e per-
mits us to eliminate derivati\'es from other recurrence relations,
Equation (2) of Section 87 yields
(1) x(x
2
- l)P ,,'(x) = n(x
2
- l)P " (x) + (X2 -
and we may now substitute for (x
2
-1)Pn'(x) and for
from (7) of Section 87 to arrive at the identity
160 LEGENDRE POLYNOMIALS [Ch. 10
x[nxPn(X) - nP"_l(x)] = n(x
2
- I)P
n
(x)
+ (n - l)xP
n
_
l
(x)-(n - 1)P
n
_
2
(x).
Collect terms in the above equation to obtain the pure recurrence
relation
(2) nP n(X) = (2n - l)xP n-I(X) - (n - 1)P n-2(X), n ?; 2.
Equation (2), with index shifted, is of the character of the pure
recurrence relation for an orthogonal set of polynomials. See
Theorem 57, page 151, and note that here we have an example in
which En = O. We shall show in Section 99 that the P n(X) form
an orthogonal set.
Equation (2) furnishes a fairly easy method for computing suc-
cessive Legendre polynomials. From the relation (2) of Section 86
we easily find that
Po(X) = 1, PJX) = x.
Then the pure recurrence relation may be used to obtain
P (x) = -
3 2 2 '
P
6
(X) = 11
6
(231X
6
- 315x
4
+ 105x2 - 5),
etc.
89. Legendre's differential equation. We have already ob-
tained the relations
(1)
(2)
XPn'(X) = nPn(X) +
XPn'(X) = - (n + l)Pnex).
We now wish to eliminate the differences in subscript to find a
relation involving only P n(X) and its derivatives.
In (2) replace n by (n - 1) to get
(3) = Pn'(x) - nPn_1(x),
from which also, by differentiation, we have
(4)
Both and can be obtained from (1) and put into (4)
to yield
90j THE RODRIGUES FORMULA 161
x[xP n"(X) + P n'(x) -nP n'(X)] =P n"(x) - (n+ 1) [xP n'(X) -nP n(X)].
A rearrangement of terms in the above equation gives us Legendre's
differential equation for P n (x),
(5) (1 - X2)P n"(x) - 2xP n'(x) + n(n + l)P n(X) = O.
We shall return to this differential equation later when we take
up the matter of orthogonality of the set of Legendre polynomials.
Equation (5) is one of many natural starting points for the study
of Legendre functions, which are solutions of (5) for nonintegral n.
90. The Rodrigues formula. In Section 86 we establi"hed that
[n/2] (-1)k(Dn_k(2x)n-2k
Pn(X) = L !t'C _ 2- /:)-,-'
k-_O n h
(1)
We also know that (2m)! = 22m(!)mm! and therefore that
(2) 2
2n-2k(1) = (2n =- 21.:)!.
2 n-k (n - k)!
Employillg (2) on the right in equation (1), we obtain
(3)
[n/2J (-1)k(2n _ 2k)!xn-2k
Pn(X) = (; 2nk!(n - k)!(n - 2k)(
d
If D = dx' we know that
(4)
m!x
m
-.
D'x
m
= - : - - - - ~
(m - s)!
The expression (2n - 2k)! x
n
-
2k
/(n - 2k)! in equation (3) suggests
the use of (4). Indeed, by (4),
(2n - 2k) !X
n
-
2k
Dnx2 n-2 k = - ' - - - - - c - - - ~ ~ -
(n - 2k)! '
so that (3) may be rewritten as
(5)
Of course the k! and (n - k)! in (5) remind us of the binomial
coefficient
n!
Cn,k = k'( - k)'
It. n \;.
Since n is independent of k, equation (5) can now be put in the form
162
LEGENDRE POLYNOMIALS [Ch.1O
(6)
For [n/2J < k n, 0 2n - 2k < n, so that for those values
of k, Dnx2n-2k = O. Hence the summation on the right in equation
(6) can be extended to the range k = 0 to = n. Thus we have
or
(7)
1
P,,(x) = _Dn(x2 - l)n
2nn! '
which is called Rodrigues' formula. We shall use (7) to obtain
another formula for P n(X),
Leibnitz' rule for the nth deri\'ati\'e of a product is
(8) Dn(uv) = L Cn.k(Dku)(Dn-kv),

in which D = d/dx and u and v are to be functions of x. The
validity of (8) is easily shown by induction.
Since, by Rodrigues' formula,
1
Pn(X) = 2nn!Dn[(x - l)n(x + 1)nJ,
the application of (8) with u = (x - 1) n, v = (x + 1) n, leads to
P (x) = __ 1_ C
n 2n ! L.... ". k ( /.) I J.f'
n. n - , . Iv.
or the beautifully symmetric result
(9)
" ( l)n-k( + l)k
Pn(X) = 6 ;
91. Bateman's generating function. Equation (9), when put
in the form
(
1) P ( ) = (n!)2[(x - l)Jn-k[1(x + 1)Jk
n X 6 [en - k)!)2(k!)2 ,
reminds us of the Cauchy product of two power series,
(2)
Hence we multiply each member of equation (1) by t
n
/(n!)2 and
92] ADDITIONAL GENERATING FUNCTIONS 163
sum from n = 0 to co to obtain
ro P n(x)tn ro n [!(x _ l)Jn-k[!(x + l)Jktn
{; (n!)2 = {; t:o [en - k) !]2(k!)2
= OFI(-; 1; - 1) OFI(-; 1; + 1).
Thus from the Rodrigues formula we have obtained a second
generating function:*
(3) OFI( -; 1; H(x - 1) oF1( -; 1; !l(x + 1) = ffJ:r}/:
92. Additional generating functions. The generating function
(1 - 2xt + t2)-! used to define the Legendre polynomials can be
expanded in pO\vers of t in new ways, thus yielding additional
results. For instance,
(1 - 2xt + (2)-! = [(1 - xt)2 - t2(X2 - l)J-!
Therefore
Now
[
t2(X2 - 1)J-1
= (1 - xt)-l 1 - ----
(1 - xt)2
(
(2(X2 - 1)) ro
(1 - xt)-l IFo -; (1 _ xt)2 = {; P n(X)t
n

(1 - Xt)-l IFo(!; -;
on 0) kt2 k (x
2
- 1) k
= t; k!(l - xt)2k+!
ro [n/2Jn!(x2 _ 1)k
x
n-2kt n
= L L 2
2k
(/:.')2(n - 2k) ,.
11::>0 k=O \.-.
*This is a special case of a result published by Harry Bateman in 1905. See
Bateman [11.
164 LEGENDRE POLYNOMIALS [Ch.1O
Hence we obtain a new form for P ,,(x):
[n/2J n!(x2 _ 1)kxn- 2k
(1) P n(X) = (; 22k(k!)2(n _ 2k) '"
Let us employ (1) to discover new generating functions for P ,,(x).
Consider, for arbitrary c, the sum
00 (c)"Pn(x)t
n
00 [n/2J (C)n(X2 _ 1)kxn-2ktn
~ -----n!-- = ~ ? ; 22k(k!)2(n - 2k)!
[
!C, !c + !;
= (1 - xt)-c 2F,
1 .
,
t2(X2 =-12J
(I - xt)2 .
We have thus discovered* the family of generating fundions:
[
!C, !c + !;
(2) (1 - xt)-c 2F,
1 ;
t2(X2 - 1)] _ m (C)nJ:n(x)ln
(l - Xl)2 - ~ n! '
in which c may be any complex number. If c is unity, (2) de-
generates into the generating relation used to define P ,,(x) at the
start of this chapter. If c is taken to be zero or a negative integer,
both members of (2) terminate, and only a finite set of Legendre
polynomials is then generated by (2).
With the aid of Ex. 11, page 70, it is a simple matter to transform
the left member of (2) into the form shown in (3) below. Let
p = (1 - 2xt + (2) i
then an equivalent form for (2) is
*Special cases of (2) have been known for a long time, but the gpneral formula may
have been first published by Brafman in 1951. See Brafman [1].
93] HYPERGEOMETRIC FORMS OF Pn(X)
(3)
!(1 - xl)l f: (c),p ... (x)t'.
p J n.
Let us now return to (1) and consider the sum
co Pn(x)t
n
'" {n/2J (X2 _ l)kxn--Zktn
L-T- = L L;Zk ., 2 -'--:--j"
n. 2 (k.) (n - 21.).
= ex t of I ( - ; 1; l2 (X2 - 1).
We thus find another generating relation,
(4)
165
which can equally well be written in terms of a Bessel function ... s
(5)
The relation (5) was being used at the beginning of this century.
'Ve haye not been able to determine when or by whom it was first
discoyered.
93. Hypergeometric forms of P n(X). Return once more to the
original definition of Pn(X):
m
(1) (1 - 2xl + t2)-! = L P n(X)t
n

n=O
This time note that
(1 - 2xt + t2)-! = [(1 - t)2 - 2t(x - 1)J-}
[
2t(x - I)]-t
= (1 - t)-1 1 - ,
which permits us to write
co en O)k2ktk(X _ I)k
[; Pn(x)t
n
=.8 /';!(1 _ t)2k+l
co (!)k(2k + 1)n2k(x - 1)ktnH
= L k'n' .
n. k:z;;;. 0 "l.
166 LEGENDRE POLYNOMIALS [Ch.10
Thus we have
00 co 00 (t)k2k(n + 2k)!(x _ l)ktnH
t; Pn(X)t
n
= t;?; k!(2k)!n!
= f f (n + 2k)!(x - 1)ktnH
n ~ O k=O k!2
k
l.:!n!
Therefore
(2)
Since P
n
( -x) = (-l)np
n
(x), it follows from (2) that also
(3)
Various formulas of Chapter 4 may now be applied to equation (2)
to obtain other expressions for P n(X). It is interesting also to con-
vert into hypergeometric form the results already derived in this
chapter.
Equation (2) of Section 86, page 157, is
= [ ~ J (-1)kO)n_k(2x)n-2k.
Pn(x) - L.J
k ~ O k!(n - 21.:)!
Hence we may write
or
(4)
[n/2J O)n( -n)2k(2x)n-2k
Pn(X) = 6 k!(t - n)k
n
!
94] BRAFMAN'S GENERATING FUNCTIONS 167
Equation (1) of Section 91, page 162, is
n (n!)2[Hx - I)Jn-k[Hx + I)Jk
Pn(X) = 6 (k!)2[(n - k)!J2 '
from which it follows that
Pn(X) = ' (-n)k(
(k!)2
Therefore
(5)
or, by reversing the order of summation,
(6)
In Section 92, equation (1), page 164, is
from which
or
(7)
P,,(X)
[
-!n, -!n + !;
P n(X) = x" 2F\
1 ;
x +-.!]
x-I '
See also Exs. 14 and 15 at the end of this chapter.
94. Brafman's generating functions. Brafman [IJ obtained a
new* class of generating functions for Legendre polynomials as an
incidental result of his work on Jacobi polynomials. In Chapter
16 we shall prove a theorem which contains as a special case the
following result:
Let p = (1 - 2xt + t2) \ denote that branch for which p ----> 1 as
t ----> O. For arbitrary c,
For negative integral c the generating relation (1) had been known for a long time.
168
(1)
[
C' 1 - c;
J'l
l' ,
LEGENDRE POLYNOMIALS [Ch.10
1 - t-
2
p] [C,l - c;
2Fl
1 .
,
f (C)n(1 -
(n!)2
For proof of (1), put a = /3 = 0 in the derivation of equation (2),
page 272.
95. Special properties of P n(X). We have already shown that
(1)
In (1) first put x = cos a, t = V sin {3; and seeond put x = cos /3,
t = v sin a to obtain the two relations
(2) exp(v cos a sin (3)J
o
(v sin (3 sin a)
en p (cos /3)V" sinn a
(3) exp(v cos (3 sm a)Jo(v sm a sm (3) = L ____ , ___ '
n.
Since sin ({3 - a) = sin /3 cos a - cos (3 sin a,
(4) exp(v cos a sin /3) = exp[v sin ({3 - a)] exp(v cos (3 sin a).
Now combine (2), (3), and (4) to arrive at the identity
Pn(COS a)v
n
sinn /3 [. ( )] Pn(cos (3)v
n
sinn a
..... ---"----'-,;-----'- = exp V sm (3 - a ..... -'--'-:-, ---
n. n.
= i: i:: sin
n
-
k
(/3- a)sink a Pk(cos /3)v
n
k!(n - k)! '
from which it follows that
sinn (3 Pn(COS a) = L Cn,k sin
n
-
k
(/3 - a)sink a Pk(cos /3),

in which C n,k is the binominal coefficient. This last equation can
be written in the form (Rainville [5J)
(5)
Pn(COS a) = i:: Cn,k[sin (.{3 - a)lJn-kpk(COS (3).
sm /3 k=O sm a
Equation (5) relates P n (cos a) to a sum involving P
k
(cos (3) with
a and /3 arbitrary. We make use of (5) later.
96] MORE GENERATING FUNCTIONS
169
Let us return to the original definition of P n(X) and for convenience
use p = (1 - 2xt + t
2
),. We know that
ro
(6) L P n(X)t
n
= p-l.
n=O
In (6) replace x by (x - t)1 p and t by vi p to get
~ P (x - t) _ [1 2(x - t)v + V 2 J - ~
~ n -- P nvTl. ~ ~ 2 2
n_O P P P
We may now write
= [1 - 2x(t + v) + (t + v)2]-!,
which by (6) yields
CD (x t) co
LPn -=- p-n-1v
n
= L Pn(X)(t + v)n
n=O P n ~ O
ro n nIP n(x)fkvn-k
= ~ 6 k!(n - k)!
= f (n + k)!Pn+k(x)tkv
n

n,k_O kIn!
Equating coefficients of v
n
in the above, we find that
(7) -n-Ip (x - t) = ~ (n + k)!Pn+k(x)ik
p n p 6 kIn! '
in which p = (1 - 2xt + t
2
)'. Equation (7) can be used to trans-
form identities involving Legendre polynomials and sometimes leads
in that way to additional results. See Bedient [lJ.
96. More generating functions. As an example of the use of
equation (7) of the preceding section, we shall apply (7) to the
generating relation
(1)
00 P (x)tn
ext oFr( - ; 1; H2(X2 - 1) = L _n ,-,
7t=(J n.
obtained in Section 92. In (1) replace x by (x - t)1 p, t by -tyl p,
and multiply each member by p-l, where p = (1 - 2xt + t2) i, to
170 LEGENDRE POLYNOMIALS [Ch.10
= f : (-1)n(n + A:)!Pn+k(x)tn+kyn
1., (r' ')2 '
tl=O).;=O h. CI.
in which we have used equation (7) of Section 95.
Collect powers of t on the right in the last summation to see that
-1 [-ty(x - t)] F [-;
p exp 2 0 I
P l'
,
ro
= L: IF
I
( -n; 1; y)P,,(x)t
n
.
11=0
In Chapter 12 we shall encounter the simple Laguerre poly-
nomials
(2) Ln(X) = IF
I
( -n; 1; x).
Lsing the notation in (2) we may now write the generating relation *
2' [ty(t - x) ]
(3) (1 - 2xt + t )-, exp 1 _ 2xt + t2 .
[
-; y2t2(X2 _ 1)] ro
oFl 4(1 _ 2xt + t2)2 = ~ Ln(y)Pn(x)t
n
,
1 ;
which we shall eall a bilateral generating function. The relation
(3) may be used to generate either Ln(Y) or Pn(X).
*Equation (3) was first obtained by Weisner [1] by a method different from that used
here. We use the method introduced by Bedient [1].
97] LAPLACE'S FIRST IiVTEGRAL FORM 171
In terms of a Bessel function, equation (3) may be written com-
pactlyas

(4) p-l exp[ty(t - x)p-Z]Jo(lyyI - X2p-2) = L Ln(y)Pn(x)t
n
u=o
in which p = (1 - 2xt + t2) I and in which x, y, and t are inde-
pendent of each other. The procedure used to get (3) or (4) can be
used to obtain further generating functions as indicated in the
exercises at the end of this chapter.
97. Laplace's first integral form. In Section 92 we obtained
the expansion
(1)
which may be written as
(2)
Pn(x) _ [n/2[ I)k.
- (; k!(2k)!(n - 2k)!
Now
(!h r(! + k) = rO)r(! + Ie) = !B(l 1 + k)
Ie! r(!) r(I + k) 7rr(k + 1) 7r 2, 2
2i
1
.. 1 i"
= - cos
2k
<p d<p = - Cm-;Zk <p d<p.
7r 0 7r 0
Hence (2) leads us to
(3)
_ ! [n/2] n!x
n
-
Zk
(x
2
- I)ki" 2k
P n(X) - L (2/0) I ( _ 2k) I cos <p d<p.
7r ,. n . 0
Since i" cosm<p d<p = 0 for odd m, we may replace 2k by k in
the summation on the right in equation (3). Thus equation (3)
leads to the relation
(4)
in which each term involving an odd Ie is zero. Equation (4) III
turn gives us Laplace's first integral for P n(X),
(5)
172 LEGENDRE POLYNOMIALS [Ch.10
98. Some bounds on P n(X). Equation (5) of the preceding
section yields certain simple properties of P n (x). In the range
-l<x<l,
Ix + (X2 - 1)1 cos <pI = Vx
2
+ (1 - X2)COS
2
<p
so that
(1) Ix + (x
2
- l)!cos <pI = VI - (1 - x2)sin2 <p, -1 < x < 1.
From (1) it follows that, except at <p = 0 and <p = 'Jr,
I x + (x
2
- I)! cos <p I < 1,
which leads to
THEOREM 60. For -1 < x < 1, IPn(x) I < 1.
The integral form
(2) 1 i" P n(X) = - [x + (x
2
- 1)!cos <pJn dip
7r 0
and (1) combine to yield, for -1 < x < 1,
or
in which \ve use the fact that 1 - y < exp( -y) for y > o. We
may now write
IPn(x) I < il"ex
p
[- 2n<p2(1 - X2)] dip
7r 0 7r
2
99] ORTHOGONALITY 173
then introduce a new variable of integration {3 = (cp/n-)[2n(1 -
to obtain
I
!
2 7r 1
00
y;
Pn(X) < - . -[2 (-1----;-)]1 exp( - {32) d{3 = .
7r n - x 0 y2n(I _ X2)
THEOREM 61. If -1 < x < 1 and 1f n 1:S any positive integer
(3) I P n(X) I < Ln(I
7r
_ X
2
)l
99. Orthogonality. We know that the polynomial
P n(X) satisfies the differential equation
(1) (1 - x
2
)Pn"(x) - 2xP
n
'(x) + n(n + l)P
n
(x) = O.
Equation (1) can equally well be written
(2) [(1 - x
2
)P
n
'(x)], + n(n + l)P
n
(x) = O.
We are now interested in obtaining integrals involving the product
Pn(x)P m(X) of two Legendre polynomials. Hence we combine (2)
with
(3) [(1 - X2)P m'(X)J' + m(m + l)P m(X) = 0
to get
(4) Pm(x)[(1 - x
2
)Pn'(x)], - Pn(x)[(l - x
2
)P
m
'(x)]'
+ [n(n + 1) - m(m + l)J P n(X)P m(X) = O.
But
[(1 - X2) {Pm(x)Pn'(x) - Pm'(X)Pn(x)}]'
=(1 - x
2
)Pm'(x)Pn'(x) + Pm(X) [(1 - x
2
)Pn'(x)],
- (1 - x
2
)P
m
'(x)Pn'(x) - Pn(X) [(1 - x
2
)P
m
'(x)]',
so that (4) becomes
[(1 - x
2
){P
m
(X)P
n
'(x) - Pm'(X)Pn(x)}]'
+ (n2 - m2 + n - m)Pn(x)Pm(x) = 0,
or
(5) (n - m)(n + m + I)Pn(x)Pm(x)
= [(1 - X2) I P m'(X)P n(X) - P m(X)P n'(x) } J'.
174 LEGENDRE POLYNOMIALS [Ch.1O
From (5) it follows that for any finite limits of integration
(6) (n - m)(n + m + 1) fb
p
,,(x)P m(X) dx
a
= [(1 - XZ)/Pm'(x)P,,(x) - Pm(X)Pn'ex)II
Since (1 - X2) vanishcs at x = 1 and x = -], wc conclude that
(n - rn)(n + m + 1) ill P ,,(x)P ",(x) dx = 0.
Now m and n are non-negativc intcgcrs, so n + m + 1 r'= 0. If
also n r'= m, n - m ~ 0, and wc may conclude that
(7) i: P,,(x)Pm(x) dx = 0,
m :;6 n.
In the terminology of Chapter 9, equation (7) means that the set of
polynomials P n(X) is orthogonal with respcet to the weight function
unity on the inten"al -1 < x < L Thc Legendre polynomials
therefore possess the properties held by all orthogonal polynomials.
THEOREM 62. The zeros of P1l(x) are distinct, and all lie in the
open interval - 1 < x < 1.
THEOREM 63. For k = 0, 1,2, .. ', (n - 1),
(8) i ~ x
k
P ,,(x) dx = 0.
We already know (Section 88) the three-term recurrence relation
(9) nPn(x) = (2n - l)xPn-l(x) - (n - 1)P
n
_
2
(x),
whose existence follows from the orthogonality property.
Latcr wc shall need the value of II P n"(X) dx.
-I
From
we obtain
I
I dx co m II
(10) -1 1 _ 2xt + t2 = ~ o ~ o -I Pk(X)Pm_k(x) dx tm.
99] ORTHOGONALITY 175
In (10) the integral on the left is elementary; on the right each
integral vanishes except when k = m - k. Therefore the only
nonzero terms in the series are those for which m is even, m = 2n,
and k = n. Hence
(11)
[
1 Jl CD II
- - Log(l - 2xt + t2) = L Pn
2
(X) dx t2".
2t _I ,,_0_1
But the left member of (11) is
1 (1 - t)2
- 2t Log(1 + t)2 = t-t[Log(I + t) - Log(1 - t)]
m (_ 1) ml m '" t m
= L----- + L-----:-
m ~ O m + 1 " ' ~ O m + 1
= f _ 2t2,,-
n ~ O 2n + l'
since each term for odd m drops out. We thus obtain from (11)
the desired result:
(12)
I
I 2
Pn
2
(X) dx = --_.
-1 2n + 1
Since, by equation (3), page 158,
n!
xn = 2n(!)nPn(x) + 1r,,_2
the result (12) is easily converted into the form
(13)
As an application of (6) let us evaluate
(14) IIp n(X) P m(X) dx.
By (6), page 174,
(15) (n - m)(n + m + 1) IIp n(X)P m(X) dx
= Pm(O)Pn'(O) - Pm'(O)P,,(O).
If m and n are both odd or both even, the right member of (15) is
zero by (7) and (9) of Section 86, page 1.58, and the integral (14)
vanishes unless m = n. If m = n, it follows from (12) and the
176 LEGENDRE POLYNOMIALS [Ch. 10
fact that P n
2
(X) is an even function of x, that
i
l 1
o Pn
2
(X) dx = 2n + 1
We are left with the need to evaluate (14) when one of m and n
is odd and the other even. Let n = 2k, m = 2s + 1. Then by
(15),
(16) (2k - 28 - 1)(2k + 28 + 2) .f
pZk
(x)P2.,+I(X) dx
= P 2s+l(0)Pfk(O) - PL+I(O)P2k (O).
In Section 86, equations (7) and (9), page 158, we found that
P (0)
= (-I)kO)k
2k k!'
(
Pf'+1 (0) = I .
s.
Therefore it follows from (16) that
i
l (-l)k+O)k(D.+I
(17) 0 P
2k
(X)P
2s
+
I
(X) dx = (k + s + 1)(28 + 1 - 2k)k!s(
100. An expansion theorem. We now seek an expansion of
the form
ro
(1) f(x) = L: anPn(x), -1<x<1.
n=O
From (1) we obtain an in a purely formal manner. With that value
for an we then proceed to prove that the series on the right in (1)
actually converges to f(x), providing f(x) is sufficiently well be-
haved.
From (1) it follows formally that
I
I m II
f(x)Pm(x) dx = L: an Pm(X)Pn(x) dx.
-I n=O -1
(2)
All the integrals on the right in (2) vanish except for the single
term for which n = m. Therefore
I
I fl 2
(3) _J(x)P m(X) dx = am _I P m
2
(X) dx = 2m a+ 1
In (3) replace m by n and x by y to obtain
(4) an = (n + !) f(y)Pn(y) dy.
100] AN EXPANSION THEOREM 177
N ow that we know what an to use, we proceed to prove the
desired result. *
THEOREM 64. If on -1 ~ x ~ 1 f(x) is continuous except for a
finite number of .finite discontinuities, 1j on -1 ~ x ~ 1 f'(x) exists
where f(x) is continuous and the right hand and left hand derivatives of
f(x) exist at the discontinuities, and if
(4) a" = (n + ) fl f(y)Pn(y) dy,
-I
then
ro
(5) L a"P,,(x) = f(x), -1 < x < 1,
n=O
at the pm'nts of continU'l'ty of f(x).
The series on the left in (5) con verges to the mean value
Hf(x + 0) + f(x - O)J at the points of discontinuity of f(x), hut
we omit proof of that portion of the theorem.
Proof: Consider the left member of (5) with the an from (4):
~ anPn(X) = ~ (n + !)i:f(Y)PnCy)Pn(X) dy
The series on the left in (5) will converge to the sumf(x) if and only if
(6) ~ l r ; ; ill f(y)Kn(x, y) dy = f(x),
in which
n
(7) Kn(x, y) = L (k + !)Pk(X)Pk(y).
k=O
To sum the series on the right in (7), we turn to the Christoffel-
Darboux formula of Theorem 58, page 154,
(8)
~ -I () () _ h" <Pn+I(y)<p,,(x) - <p"+I(X)<p"(y) .
.4..J gk <Pk X <Pk Y - --
k=O g"hn +1 y - x
*Theorem 64 is a most elementary form of an expansion theorem. The aim here is
to exhibit the underlying ideas with as little complication as possible.
178 LEGENDRE POLYNOMIALS [Ch. 10
I
I 2 1
gk = -1 P
k
2(X) dx = 21e + 1 = Ie + r
and the leading coefficient in Pn(X) is hn = Hence
hn/(gnh
n
+
l
) = Hn + 1), and we conclude from (8) that
(9) Kn(x y) = . Pn+I(X)Pn(vJ - Pn(X)Pn+l(y).
'2 x - y
The condition ((i) may be rewritten as
(10) Lim [fC:C) -fl f(y)IC(x, y) ely] = O.
n+oo -1
Since
for Ie > 0,
we may write
ill Kn(X, y) dy = II, t;, (Ie + DPk(x)Pk(y) dy
= ill + (k + dy
= II = l.
-I
Now the condition (10) may be put in the form
Lim II [f(x) - f(y)JKn(x, y) dy = 0,
n+ro -I
or because of (9),
(11)
. n+1fl f(x)-f(y)
Lm1-
2
- ---'-[Pn+leX)Pn(Y)-Pn(X)Pn+I(Y)] dy=O.
n+ro -I x-y
By Theorem 59, page 13G, it follows that
(12)
for any g(y) such thatfl g2(y) dyexists.
-I
101] EXPANSION OF xn 179
At a point of continuity of fey) we also assumed thatf'(y) exists.
Hence, \vith
g(y) = =
ill g2(y) dy does exists. It follows from (12) that both
( 13)
Lim (n + t)tfl fex) - fCy)Pn(y) dy = 0,
,,-reo -I x - Y
and
(14)
Lim(n + lex) -f(y)P,,+I(y) dy = O.
n-reo 2 -I X - Y
Then (11) will be satisfied at a point of continuity of f(x) 111
-1 < x < 1 if we can show that each of
is bounded as n -> 00.
We use Theorem 61, page 173, to see that, on -1 < x < 1,
I
n + 1( 1)-1 I' [(n + 1)2 7r J!
-2-- n + 2 Pn+1(x) < 4(n + . 2(n + 1)(1 - X2)
[
7r(n + 1) ]1 [ 7r J!
< 4(2n + 1)(1 - X2) < 4(1 - X2)
and
I
n + 1( 3)-1 I [ (n + 1)2 7r J!
-2- n + 2 Pn(x) < 4(n + !) . 2n(1 - X2)
[
7r(n2 + 2n + 1) ]1 [ 7r J!
< 4(2n2 + 3n) (1 - X2) < 4(1 - X2) .
We haye thus shown the validity of equation (5) of Theorem 64
at points of continuity of f(x).
101. Expansion of xn. We already know from Theorem .53,
page 147, that any polynomial can be expanded in a series of
Legendre polynomials merely because the P n(X) form a simple set.
The orthogonality of the set Pn(x) plays a role only in the deter-
180 LEGENDRE POLYNOMIALS [Ch.10
mination of coefficients; Theorem 64 of Section 100 has no bearing
on expansion of polynomials.
The expansion of xn in a series of Legendre polynomials is useful.
Consider
00 [ 2xt J-1
L Pn(X)t
n
= (1 - 2xt + t
2
)-, = (1 + t
2
)-, 1 - ---2
n_O 1 + t
CD
= n!(1 +i2);-+t'
from whi('h we get
(1)
= (1 + (2)! P (x)ln
L. 1(1 + [2)" L. " .
11=0 n. !!::EO
Kow put
2v
t = --- ---- - -- --.--
1 + vI_'=-4v2
from which
1 + [2 = 2 ,
1 + VI - 4v
2
[
1+(2 = v.
Equation (1) becomes
(2) i: = i: Pn(x)Vn( 2 _. )n+t.
n_O n. n_O 1 + VI - 4v
2
By Ex. 10, page 70,
or
(3)
(
2 )n+1 _ [Hn + !), Hn + !);
-----_ - 2Fl
1 + VI - 4v
2
n + !;
= i: (n +
k=O k! (n + !)k
CD (1) (3.) 2k
= L 2 n+2k 2 nV
k-O (1) kl(3)
- "2 n 3f n+k
102] EXPANSION OF ANALYTIC FUNCTIONS 181
We may now conclude from (2) and (3) that
f CDn(2x)nv" = f: (2n + 1)(!)"+2kPn(X)vn+2k
n! n ,k_O k'(;l)
2 n+k
Comparison of coefficients of un in the preceding equation yields
the following result.
THEOREM 65. For non-negaUve integral n,
(4)
,,1'1/2
1
(,), 4'''+J)P ()
x" = rt. " ,- __ ____ .
2
" L..J
k
2 n-
In later chapters we shall use Theorem 65 to find expansions of
various polynomials in series of Legendre polynomials. See also
Ex. 17 at the end of this chapter.
102. Expansion of analytic functions. We use Theorem 65 of
the preceding section to find explicit expressions for the coefficients
in the expansion of analytic functions in series of Legendre poly-
nomials. The theory of such expansions is treated in seyeral
places. See Whittaker and Watson [1; 321-323J and Szego [1]. For
a general discussion of expansion of analytic functions in series of
polynomials, see Boas and Buck [2].
If we have
(1)
f(x) = f
1/.=0 n.
application of Theorem 65 yields
f(x) = fl'f a
n
(2n - 4k + 1)P
n
_
2k
(x)
2nk!G)n_k
which is the desired expansion.
THEOREM 66. If I x I is sufficiently small and if
(1)
f(x) = f
'1_0 n.
182
then
(2)
in which
(3)
LEGENDRE POLYNOMIALS [Ch. 10
00
f(x) = L: bnP ,,(x),

b .. = i: (2n + I)a
n
+2k.
k_O (3.)
2 n+k
The region of convergence of (2) is the interior of an ellipse with
center at x = O.
EXAMPLE: Expand (t - X)-l in a series of Legendre poly-
nomials.
If I x I < t, then
= X"
(t - X)-l = t-I(I - xt-I)-I = L:--'
n t,,+1
By Theorem 66,
in which
or
(4)
co
(t - X)-I = L: (2n + l)Q,,(t)P,,(x),
n_O
n! i: (1 + n) 2kt-
2k
,
+ n)k
It will be found that the Qn(t) of (4) is a second solution of the
differential equation for P net). See Ex. 3 below.
EXERCISES
1. Start with the defining relation for Pn(X) at the beginning of this chapter.
Use the fact that
(1 - 2xt + t2)-! = [1 - (x + YXZ=l)tJ-I[l - (x - vx' --=-l)tJ-1
and thus derive the result
102] EXPANSION OF ANALYTIC FUNCTIONS 183
P (x) - i:, + - yxCl)k
n - k_D k!(n _ k)! '
in which it is to be noted that x - = (x +
2. Use the result in Ex. 1 to show that
3. In Section 93, equation (4), page lOG, is
(
1) ('I .. J.,)n [-In, - !n + !;
Pn(X) = "--"--j 2ft',
Tl. I.
'2 - n,
We know from Section 34 that the 2FI equation has two linearly independent
solutions:
b; c; z)
and
zl-CF(a + 1 - c, b + 1 - c; 2 - c; z).
Combine these facts to conclude that the differential equation
(1 - t2)y" - 2ty' + n(n + l)y = 0
has the two linearly independent solutions YI = Pn(t) and Y2 = Qn(t) , where
Qn(t) is as given in equation (4) of Section 102.
4. Show that
ro 3
L [xPn'(x) - nPn(x)]t
n
= t2(1 - 2xt + t
2
)-2
_0
and
ro [n/2J
L L (2n - 4k + I)Pn- 2k (xW = (1 - 2xt + t2)-!.
n=O k::.:O
Thus conclude that
I (n-2) /2]
xPn'(x) - nPn(X) L (271 - 4k - 3)P"--22k(X).

5. Use Bateman's generating fUllction (3), page W:3, with x = 0, t = 2y to
conclude that
oFI(-; I; y) of)(-; 1; -y) = OF3(-; 1, 1, L _!y2).
6. Use Brafman's generating function, page lGS, to conclude that
1 - t - [C, 1 - c;
---7
2
-'-. 2F,
1 ;
= 4
F
3
[
' 1 + 1 1 , 1 I.
'Ie, ':Ie 'I, 2 - 'Ie, - 'Ie,
1, 1,
184 LEGENDRE POLYNOMIALS [Ch. 10
7. Use equation (5), page 168, to obtain the results
sin" (3 P n(sin (:3) = L (-1) kG", k cos
k
(3 P k(COS (3),

P n(X) = L (-1) kGn, k(2x)n-kP
k
(X),
k_0
P,,(l - 2x
2
) = L (-2X)kG
n
,kPk(X).
k,,-_d)
8. Use the technique of Section 96 to derive other (see also, Weisner [1])
generating function relations for P,,(x). For instance, obtain the results
co
L lF
2
( -n; 1, 1; y)P n(X)t
n
"..()
-yt(x - t + p)]
---2
p
2--
in which p = (1 - 2xt + t2)!, and
co
L 2F
l
(-n, c; 1; y)P,,(x)t"
,,-0
[
!C, +
= p2C-l(p2 + xy! - yt
2
)-C 2Fl
1 .
Also sum the series '
00
L 3F2(-n, c, 1 - c; 1, 1; y)Pn(x)t
n

n_O
g, With p = (1 - 2xt + t
2
) \, show that
10, With the aid of the result in Ex, 7, page 3], show that
(I [-n,n+ 1,(3;
LP + x)'>-l(1 - x)f3-
I
P
n
(x) dx = 2
a
+
f3
-
1
B(a,{:3)
1, a + (3;
Investigate the three special cases a = 1, (3 = 1, a + fJ = n + 1.
11. Obtain from equation (5), page 168, the result
(1 + x)!npn( ; x) = 2-
1n
to G",kPk(X)
and use it to evaluate (Bhonsle [1]) the integral
L: (l + x)ln
p
n ( x)Pm(x) dx,
EXPANSION OF ANALYTIC FUNCTIONS 185
12. Evaluate
and check your result by means of Theorem 65, page 181. Thus show that
and, equivalently, that
I
I nHkp ( ) d __
x n X X - 'hd2k,",(3)
o It. 2 n+1c
13. Use formula (5), page 104, to obtain the result
where Qn(t) is the function given in (4), page 182.
14. Show that
p ... (x) = 2
n
(!).(x,- l)n F[-n, -n;
n. -2n- 1 - x .
,
15. Show that
Pn(X) = 2
n
(!)nC!, +---.Jl" F[-n, -n;
n. -2n;
16. Show that for I t I sufficiently small
00 a
L (2n + I)P
n
(x)t
n
= (1 - t2)(1 - 2xt + t
2
)-2.
-.0
17. Use Theorem 48, page 137, with c = 1, x replaced by HI - x) and
"Yn = (!)n/n! to arrive at
(1 - x)" = 2n(n!)2 f.
M(n - k)!(n + k + 1)1
18. Use Theorem 48, page 137, to show that
(1 - x)Pn'(x) + nPn(X) = nPn_l(x) - (1 -
n-1 "-1
= L Pk(x) - 2(1 - x) L Pk'(x)
k_O
... -1
= L (_1)n-H1(1 + 2k)P
k
(x).

19. Use Rodrigues' formula, page 162, and successive integrations by parts to
derive the orthogonality property for Pn(X) and to show that
186 LEGENDRE POLYNOMIALS [Ch. 10
I
I 2
-I Pn
2
(X) dx = 2n +1'
20. Show that the polynomial Yn(X) = (n!)-I(l - x
2
)!nP
n
((l - X2)-!) has the
generating relation
co
e' oFI( -; 1; !X2t2) = L Yn(X)t
n
n_O
and that Theorem 45, page 133, is applicable to this Yn(X). Translate the result
into a property of P " (x) , obtaining equation (7), page 159.
21. Let the polynomials w"(x) be defined by
with
Show that
'"
e"'",[t'(x' - 1)] = L wn(x)t",
n-O
co
"'(11,) = L /'nu,n .
_0
co 00 [t2CX2 - l)Jk
L (C)nW"(X)t
n
= (1 - xt)-C L (C)2k/'k -------
n ~ O k-O (1 - xt)'
and thus obtain a result parallel to that in Theorem 46, page 134. Apply your new
theorem to Legendre polynomials to derive erruation (2), page 164.
CHAPTER 11
Hermite
Polynomials
103. Definition of Hn(x). We define the Hermite polynomials
Hn(x) by means of the relation
m H (x)t"
(1) exp(2xt - t2) = L: -"--, -,
n ~ O n.
valid for all finite x and t. Since
exp(2xt - t2) = exp(2xt)exp( - t2)
it follows from (1) that
(2)
_ (CO (2x)ntn)( co (-1)nt2n)
- L: , L: ---,-
71=0 n. n=O n.
co [n/2J (-1)k(2x)n-2ktn
= L:L:-----
71=0 k_O k!(n - 2k)! '
[n/2J (-1)kn!(2x)n-2k
L: k'(n - Ok)' .
k=O ""'.
Examination of equation (2) shows that Hn(x) is a polynomial
of degree precisely n in x and that
(3)
in which lI"n_2(X) is a polynomial of degree (n - 2) in x. From
either (1) or (2) it follows that H 71 (x) is an even function of x for
even n, an odd function of x for odd n:
187
188
(4)
HERMITE POLYNOMIALS [Ch.11
Hn(-x) = (-I)nH
n
(x).
From (2) it follows readily that
Hzn(O) = (-I)n22n(!)n;
Hfn+I(O) = (-I)n22n+IG)n;
H2n+I (0) = 0;
Hfn(O) = o.
104. Recurrence relations. Since the generating function 1Il
Section 103 is of the form G(2xt - [2), the H n(X) must satisfy
(1) xlI/ex) = + nHn(x),
as we saw in Section 72.
Also the relation
(2)
yields at once
(3)
m H '( )tn
2t exp(2xt - t2) = L -"--+-.
n.
From (2) and (3) we get
f 2H n(X)t
n
+
1
= t H n'(x)t
n
n! n!
which, with a shift of index on the left, yields Ho'(x) = 0, and for
n 1,
(4) Hn'(x) = 2nH
n
_
l
(x).
Iteration of (4) gives us
(5) D
sH ( ) = 2
s
n!H
n
_.(x).
nX (n-s)!'
Combination of (1) and (4) yields
d
D == dx
(6) H n(X) = 2xH n-I(X) -
We use (4) and (6) as our pair of independent differential recur-
rence relations. From this pair of equations we at once obtain both
the pure recurrence relation
(7)
and Hermite's differential equation
(8) - 2xHn'(x) + 2nHn(X) = O.
Later equation (8) will be used to obtain an orthogonality property
105] THE RODRIGUES FORMULA 189
for Hn(x). The differential equation is a natural starting point for
the study of Hermite functions with unrestricted n.
The first few Hermite polynomials* are listed here for reference
purposes:
Ho(x) = 1, H1(x) = 2x, H
2
(x) = 4x
2
- 2,
Ha(x) = 8x
3
- 12x,
H,,(x) = 32x
O
-160x"+ 120x,
H,(x) = 16x
4
- 48x
2
+ 12,
H
6
(x) =()4x
fi
-480x
4
+720x
2
-120.
105. The Rodrigues formula. Examination of the defining
relation
(1)
ro H n(X)t"
exp(2xt - t2) = L ----
n!
in the light of Maclaurin's theorem gives us at once
Hn(x) = [!t_ exp (2xt - (
2
)J .
dl
n

The function exp( _X2) is independent of t, so we may write
exp( -x2)H 'I (x) = exp! - (x - 0
2
11=0'
Now put x - t = w. Then
exp( -x
2
)Hn(x) = (-1)n[ d
n
n
exp( -w
2
)]
dw
But it is ridiculous to differentiate with respect to w a function of w
alone and afterward to put w = x. The w is superfluous. There-
fore we write
or
(2)
d
D ==-'
dx'
a formula of the same nature as Rodrigues' formula for Legendre
polynomials.
* As is true for many special functions, the literature contains more than one notation.
For Hermite polynomials the two most common are the one we use here and one which is
employed widely in statistics. The latter notation is often distinguished from ours by
the use of a script yf. The polynomials .Jf,.(x) may be defined by
( t
let') _ yfn(X)t
n
exp x -, - L..J -_.
n_O n!
This is the notation used by Jackson [1].
190 HERMITE POLYNOMIALS [eh.11
106. Other generating functions. Consider the sum
f (C),Hn(X)t
n
= 0 1)k(C)n(2x)n-
2k
t
n
n=O n! n=O k=O k!(n - 2lc)!
_ a>
- 61c!(1 - 2xt)c+2k
Using the fact that (cb = 2
2k
OC)k(!C + we thus arrive at the
(divergent) generating function
(1)

!c + !;
(1 - 2xt)-c 2Fo _;
-4[2 ] f (c)}2!i n(X)t
n
(1 - 2xt)2 - n=O n! '
published by Brafman [1 J. The special case c = 1 was given by
Truesdell [lJ with the left member in a different form.
Brafman also summed the series
(2)
f (c) [n/2JHn (x)t
n
n=O [n/2]!(!)[nf2)'
in which [J is our customary greatest integer symbol, thus obtaining
a class (one for each c) of peculiar generating functions for the
Hermite polynomials. In the special instance c = !, the series (2)
had already been summed by Doetsch.
107. Integrals. Directly from the initial generating function
(1)
it follows by the usual theorem on the coefficient in a Taylor series
that
(2)
n' f(o+)
H n(X) = 2;i u-
n
-
1
exp(2xu - u
2
) du,
where the contour of integration encircles the origin of the u-plane
in the positive direction. From (2) we get, by using the contour
u = exp(i8), the real integral representation
1091 ORTHOGONALITY
(3) Hn(X) =
191
n. exp(2x cos 8 - cos 28) cos(2x sin 8 - sin 20 - nO) do.
'i"
7r 0
There are numerous interesting relations connecting the Hermite
polynomials and the Legendre polynomials. We quote two fairly
simple integral relations.
Curzon [1J obtained many relations between H,,(x) and Pn(X)
with n usually not restricted to be integral. One of the simplest of
his relations, one in which n is to be an integer, is (in our notation)
(4)
Curzon also expresses the Hermite polynomial as a contour integral
involving the Legendre polynominal.
A real integral relation giving Hn(x) in terms of Pn(x) was given
by Rainville [5]. It is
(5) Hn(x) = 2n+l exp(x
2
) f""exp( -t
2
)t
n
t-1P
n
(x/t) dt.
x
Verification of (4) and (5) is left for the exercises.
108. The Hermite polynomial as a 2FO. The formula
[n/21 (-1)k
n
!(2x)n-2k
Hn(x) = 6 k!(n -2k)!
yields at once
or
Hn(X) = (2x)n 2Fo( - ~ n , - ~ n + L -; - ~ 2 }
109. Orthogonality. Equation (8), page 188, is
(1) H n"(X) - 2xH n'(x) + 2nH n(X) = 0,
which may be written
(2) [exp( -x2)H n'(X)J' + 2n exp( -x2)H n(X) = 0.
Along with (2) write
(3) [exp( -x2)H m'(X)J' + 2m exp( -x2)H m(X) = 0.
192 HERMITE POLYNOMIALS [Ch.11
Then (2) and (3) combine to yield
2(n - m)exp( -x2)H n(x)H m(X)
= H n(X) [exp( -x2)H m'(X)]' - H ",(x) [exp( -x2)H n'(X)J'.
= [exp(-x2) I Hn(x)Hm'(x) - Hn'(x)Hm(x)lJ'.
It follows that
(4) 2(n - m).[b
exP
( -x2)H n(x)H m(X) dx
= [ex
P
( -x2){H n(x)H ",'(x) - IL'(x)H ",(x) l T.
Since the product of any polynomial in x by exp( -x
2
) ---> 0 as
x ---> 00 or as x ---> - CJ;) , we may conclude that
(5) m ~ n.
That is, the Hermite polynomials form an orthogonal set oyer the
inten'al (-00, CJ; with weight function exp(-x
2
). Here the
infinite limits cause no trouble because of the factor exp( _X2).
From the definition
we obtain
so that
co H ",(x)tm
exp(2xt - (2) = L: --,-
1"11=0 rn.
(4 2 )
_ ~ ;.. H k(x)Hm_k(X)t
m
exp xt - t2 L.- L.- --------
- m ~ O k ~ O k!(m - k)!
i: exp( -x
2
+ 4xt - 2t2) dx
00 m foo tm
= n:;o6 -00 exp(-x
2
)H
k
(x)H
m
_
k
(x) dx "'!(m - k)(
Because of (5) each term on the right \'anishes except terms for
which k = m - k. Then m must be even, m = 2n, and k = n.
Therefore we haye
Now
110] EXPANSION OF POLYNOMIALS
exp(2t2) f'" exp( - X2 + 4xt - 4t2) dx
-0:>
193
= exp(2t2) i: exp[ - (x - 2t)2] dx
= exp(2t2) f'" exp( _y2) dy = y; exp(2t2).
-00
Therefore
ro fro t
2n
_ _ 00 2nt2n
L exp( -x2)H ,,2(X) dx -12 = Y 7r exp(2t2) = y 7r L --,-
n ~ U -en (n.) ,,000 n.
which yields
(6)
See also Ex. 7 at the end of this chapter for a simpler method of
obtaining equation (6).
We now know that the H n(X) form an orthogonal set over (- CD, CD)
with the weight function exp( -x
2
), and in the notation of Chapter
9, gn = 2nn! V;, and the leading coefficient in H "Cx) is h
n
= 2n.
The theory developed in Chapter 9 yields the following results.
THEOREM 67. For the Hermite polynomials H n(X),
(7) f'" exp( _X2)XkH n(X) dx = 0, k = 0, 1, 2" . " (n - 1);
-'"
(8) The zeros of H n (x) are real and dist1:nct;
(9)
(10) If i: exp( _X2)f2(X) dx exists,
~ l ~ (2
n
n!)-!i: exp( -x
2
)f(x)H n(X) dx = O.
The three-term recurrence relation for H n(X) has already been ob-
tained on page 188.
110. Expansion of polynomials. Any polynomial can be ex-
panded in a series of Hermite polynomials, and the coefficients can
be determined as in the general theory: if
194 HERMITE POLYNOMIALS [Ch.11
(1) P(x) = L c"H k(X),
k
(2) 2
k
k!yI-;Ck = i: exp(-x
2
)P(x)Hk(x) dx.
As with the Legendre polynomials, we find it desirable to bypass (2)
by obtaining the expansion of xn directly from a generating function.
Since
(3)
'" H n(X)t
n
exp(2xt - t2) = L ----,-
n.
it follows that
or
Hence
(4)
'" H (x)t
n
exp(2xt) = exp(t2) L -"-T-'
n=n 'n.
[n/2J 'H ()
xn = L
2nk!(n - 2k)!
Let us employ (4) to expand the Legendre polynomial in a series
of Hermite polynomials. Consider the series
CD CD [n121 (-1)kO)n_k(2x)n-
2k
t"
L P,,(x)t
n
= L L ---------
n_O ,,=0 bO k!(n - 2k)!
= f (-1)k(!)n+k(2x)nt
n
+2k.
n,k-O kIn!
From (4) we have
(5)
(2x) n = H n_2,(X)
n! s!(n - 2s)!
Hence we may write
co co [n/2) ( 1) k( 1) H ()tn+2k
L P n(X)t
n
= L L - ,2 ,n+k
.. _0 k.s.(n - 2s).
= i: (-1)kO)n+k+2sH,,(x)t
n
+
2
k+28
n,k,.=O k!s!n! '
lJO] EXPANSION OF POLYNOMIALS 195
in which 've haye used Lemma 11, page 57. We need to collect
powers of t in the last summation above. By Lemma 10, page 56,
we may write
co co k ( 1) k-8 ( 1 ) H ( ) tn+2 k
L P,,(x)t" = L L --= __
,,=0 sICk - S)!lI!
= f: } + n + k;
kIn!
We use Lemma 11 again to obtain
i: Pn(X)t
n
= -['>Fo( -k, !+n-k; - i 1)(
,,=0 k.(n - 2k).
The final result is
(6)
Pn(X) = [I:
J
2Fo(-k,! + n - k; -;
k!(n - 2k)!
Next let us expand the Hermite polynomial in a series of Legendre
polynomials. By Theorem 65, page 181,
(7)
Now
(2x)n
n!
i: H n(X )t" _ co [n/2] ( -1) k(2x) n-
2k
t"
n=O n! -?; t:o kl(n - 2k)!
- (-1)k(2n + 1)Pn(x)t
n
+2k+
28
.
'k'(lt)
s. . 2 n+8
Again we collect powers of t:
196 HERMITE POLYNOMIALS [Ch. 11
i:
n.
= f t (-l)k-'(2n + 1)Pn (x)t
n
+
2k
8!(k - 8)!0)n+.
= :t t (-1)
8
k! . (-1)k(2n +l)Pn (x)t
n
+2k
81(1.; - 8)10 + n). k10)n
= ['f - k; + n - 2ft'; + l)P n_2k(X)t
n
.
"=0 k1CVn-2k
Therefore,
(8) Hn(x) =
IFJ(-k;! + n - 2k; 1)(-1)kn!(2n - 41.; + l)Pn_2k(x).
k!q)n-2k
The expansion of functions other than polynomials into series of
Hermite polynomials is omitted here. Theorems exist similar to
the ones relative to expansions in series of Legendre polynomials.
111. More generating functions. We wish to obtain for H n(X)
a property similar to that for P n(X) expressed in equation (7), page
169. Consider the series
_ Hn(x)(t + v)n
-L- r
n.
= exp[2x(t + v) - (t + V)2]
= exp(2xt - t2) exp[2(x - t)v - V2]
(2t

= exp x - L- I.. r
k=O L.
By equating coefficients of v
k
/ k!, we obtain
111] MORE GENERATING FUNCTIONS 197
Hn+k(x)t
n
.... ---, - = exp(2xt - t2)H k(X - t).
11.
(1)
As a first example in the use of equation (1) let us derive what is
sometimes called a bilinear generating function. Consider the
serles
00 Hn(x)Hn(y)t
n
00 [n/2) (-1)k(2x)n-ZkH
n
(y)t
n
L ---,--- = L L ---.j - -
n. Ii. (n - '21,).
by (1). Since
= i:
k'!n!
= i: exp( 4xyt - 4XZ[2)H - 2x[) ( -1) k[Zk
k=O k! '
k (-1)s(2k)!(2v)2k-28
H2k(v) = L ---:r(2-k - 2-)-' -,
s. s .
and (2k)! = it follows that
00 (-1)k22s(J) (9y _ 4,,-t)Zkl2k+28
= exp(4xyt - 4x2t2) L '1 k+,.-': --
k, 8 0 s! ( D k
00 DO (! + k)s22st2s (_ 1)ktZk(2y - 4xt)2k
= exp( 4xyt - 4x2t2) L L 2
hO 8! k!
DO (-1) ktZk(2y - 4xl)2k
= exp(4xyt - 4x2t2) .6 k!(l _ 4tz)!+k
[
-4t
2
(y - 2Xt)2]
= (1 - 4tz)-!exp(4xyt - 4xz[2) exp --1 =4t-
z
-
198 HERMITE POLYNOMIALS [eh. 11
The exponential factors may be combined and the preceding
identity written as
(2)
f Hn(x)Hn(y)t
n
= (1 _ 4t2)-i ex
p
[Y2 _ (y - 2xt)2],
n=O n! 1 - 4t2
a generating relation known for about a century.
We can apply equation (1) to any known generating; relation and
sometimes obtain a new result. On page 1DO we obtained the
relation
(3)
[
1C' }c + ;
(1 - 2xt)-c 2Fo _ ;
To (3) we apply (1) in the following manner. Consider the series
Because of (3) it now follmvs that
'" exp(2xt - t2)[1 + 2yt(x - t)J-c
a relation obtained by Brafman [2J with contour integration as the
main tool.
111] MORE GENERATING FUNCTIONS
EXERCISES
1. Use the fact that
exp(2xt - t2) = exp(2xt - x2t2) exp[t2(x
2
+ 1)]
to obtain the expansion
[n12J 'II (1) n-2k( 2 + 1) k
II (x) = L 1l:.: __ =2". __ x x .
n k-O k!(n - 2k)!
2. Use the expansion of x" in a of Hermite polynomials to show that
i: exp( -x")xnlf" n(X) dx = 2-
2k
n!{t
Note in partieular the special case k = O.
199
3. Use the integral evaluation in equation (4), page 1D2 to obtain the result
Looexp( -X
2
)II
2k
(X)H2..+1(x) dx = + I - 2k).
4. By evaluating the integral on the right, using equation (2), page 187, and
term-by-term integration, show that
(A)
2 Lm
Pn(x) = -- exp(-t
2
)tnll"(xt)dt,
n!0 0
which is Curzon's integral for p"(x) , equation (4), page IGI.
5. Let vn(x) denote the right member of equation (A) of Ex. 4. Prove (A) by
showing that
ro
L vn(x)Y" = 0 - 2xy + y2)-i.
n=O
6. Evaluate the integral on the right in
(B) Hn(x) = 2
n
+
1
exp(x
2
) J:':xP( dt
by using
n (:I:) _ [n12J n!(x2 =-f}.'j2x)n-2k
(2t)P
n
t -t;o (k!)2(n-2k)!
derived from equation 0), page 164, and term-by-term integration to prove the
validity of (B), which is equation (5), page 191.
7. Use the Rodrigues formula
and iterated integration by parts to show that
m = n.
CHAPTER 12
Laguerre
Polynomials
112. The polynomial Ln (Q)(x). Let us consider a naturally
terminating IF
I
We define, for n a non-negative integer,
(1) L
(a)( ) - (1 + a)n F (- 1 + . )
n X - I I I n, a, X
n.
The factor (1 + a)n/n! is inserted for convenience only. The
polynomials (1) are called Laguerre, generalized Laguerre, or Sonine
polynomials. The special case a = receives much indi\'idual at-
tention and is known either as the Laguerre or the simple Laguerre
polynomial. When a = 0, a is usually omitted from the symbol:
(2)
We shall work with Ln (a)(x), but for reference purposes a list of
properties of Ln(x) is included at the end of the chapter.
The notation in (1) is quite standard with the one exception that
some authors permit a to depend upon n; others do not. We shall
insist that a be independent of n because for the polynomials (1)
so many properties which are valid for a independent of n fail
(Shively [1]) to be valid for a dependent upon n.
It should be apparent to the reader by the time he has finished
reading this book, if not before, that for a polynomial ~ n ( X ) of
hypergeometric character, the way in which the index n enters the
parameters of the pFq involwd has a vital effect upon the properties
of the polynomial. For a mathematician to use the same name for
200
113] GENERATING FUNCTIONS
the two polynomials
and
(1 a)n . .
-'--------;-,----'---' IF
1
( -n, 1 a, x)
n.
(1 C n)"
--nr---1F1( -n; 1 c n; x),
201
in which a and c are independent of n, is roughly the equivalent of
a layman's using the same name for an eagle and a kitten.
From (1) it follows at once that
(3) L (al(x) - i: _( -I)k(l
n - - k)!(1 + a):
from whieh we see that the I'n (al(x) form a simple set of polynomials,
the coefficient of xn being ( -1) nln!.
From (3) we obtain
Lz(a)(x) = - (2+a)x+tx2,
L
3
(a)(x) = i(1+a)(2+a)(3+a) - H2+a)(3+a)x+H3+a)x
2
- !X3.
113. Generating functions. Direetly from (3) of the preceding
section we obtain
'" Ln(a)(x)tn '" n (-I)kxktn
?; (I a)n = t=o k!(n - k)!(1 ah
- (f f
- n! n!(1 + a)n
Hence the Laguerre polynomials have the generating function in-
dicated in
(1)
Since any of! is a Bessel function, we are led also to write the left
member of (1) in the less pretty form
(2) r(I + a)(xt)-a/ze t J
a
(2yxt) = f
(1 a)n
A set of other generating funetions for these polynomials is easily
found. Let c be arbitrary and proeeed as follows:
202 LAGUERRE POLYNOMIALS [Ch.12
00 (c)nLn(a)(x)tn 00 n (C)n( -x) ktn
(1 + a)n =,?; {; h!(n - k)!(1 + ah
(C) k ( - xtp
= L IFo(c + k; -; t)-Iolro} +--)-
k=(} (. a k
'" (C)k( -xi)k
= =-t)C+k'
We thus arrive at the generating relation (see also pages 134-135)
(3)
Equation (3) is a speC'ial case of a result due to Chaundy [1]. Note
the commonly quoted f'pp(ial case with c = 1 + a:
(4)
1 ( -Xl) ro
exp --- = L Ln(a)(x)t
n
.
(l-t)l+a I-t
114. Recurrence relations. We han already seen in Chapter 8
that the \"ery form of the generating functions (1) and (4) of the
preceding section leads at once to the relations (with D = d/dx)
(1) x D Ln (a)(x) = nLn (a\x) - (a +
(2) D Ln (a)(x) = D -
n-I
(3)
Elimination of the deri\"ati\'es from (1) and (2) yields the pure
recurrence relation
(4) nLn(a\x) = (2n - 1 + a - x)I';,".!.I(x) - (n - 1 +
We already know three (2p + q) contiguous function relations
for the IF
I
From equations (15), (18), and (20) of Section 48,
using p = 1, q = 1, a] = -n, (3, = 1 +a, we obtain
115] THE RODRIGUES FORMULA
(5) (-n - a) IFI(-n; 1 + a; x)
203
= - n IF I ( - n + 1; 1 + a; x) - a IF I ( - n; a; x),
(6) (-n + x) IF
1
( -n; 1 + a; x)
(n + a + l)x
- n IF I ( - n + 1; 1 + a; X) + 1 + a IF 1 ( - n; 2 + a; x),
(7) IF1(-ni 1 + aiX) = IFI(-n - 1; 1 + a;X)
X
+ I + a IF 1 ( - n; 2 + a; x).
Since
n!L,,(a)(x)

equations (5), (6), (7) may be converted into the mixed recurrence
relations
(8) Ln(a)(x) = + L,,(a-l)(x),
(9) (n - x)Ln(a)(x) = (a + - xLn(a+l)(x),
(10) (1 + a + n)Ln (a) (x) = (n + + xLn (a+l)(x).
Next a shift of index in (10) yields
xL
n
_
1
(a+l)(x) = (a + - nLn (a)(x)
= -x D Ln(a)(x),
by (1) above. Hence we have
(11) D Ln(a)(x) = -L,,_I(a+1l(x).
Comparison of (3) and (11) shows that
(12)
L,,(a+1l(x) = L Lkia,(x).
k_O
115. The Rodrigues formula. Let UI-' return to the expanded
form
(1)
Since
we may write
204 LAGUERRE POLYNOMIALS [Ch.12
involving the binomial coefficient C",k. Kow Dke-
z
= (-l)ke-
z
;
so, we may conclude that
eIx-a. no
(2) L,,(a)(x) = -,- ~ C"ADn-kx
n
+
a
] [Dke-
x
].
n. k ~ O
In view of Leibnitz' rule for the nth derh'atiye of a product, equation
(2) yields
(3)
the desired formula of Rodrigues type.
116. The differential equation. Since the Laguerre poly-
nomial is a constant multiple of a IF), we may obtain the differential
equation
(1) x D2 L,,(a)(x) + (1 + a - x) D L,,(a)(x) + nLn(a)(x) = 0
from the general theory. Equation (1) is also easy to deriye by
eliminating L ~ a ~ 1 (x) from the two differential recurrence relations
(1) and (2) of Section 114.
The three-term pure recurrence relation (4), page 202, suggests
that we look for an orthogonality property of the Laguerre poly-
nomials. Either the differential equation or the Rodrigues formula
leads us quickly to the desired result.
117. Orthogonality. The preceding differential equation for
L,,(a)(x) may be put in the form
(1) D[x
a
+1e-
z
DL
n
(a)(x)] + nxae-ZLn(a)(x) = 0;
as is easily verified. Equation (1) together with
d
D == dx'
(2) D[x
a
+1e-
z
DL
m
(a)(x)] + mxae-xLm(a)(x) = 0
leads at once to
(m - n)xae-zL,,(a)(x)Lm(a)(x)
= Lm(a)(x) D [x
a
+1e-
z
D L,,(a)(x)] - Ln(a)(x) D [x
a
+
1
e-
z
D Lm(a)(x)]
= D [x
a
+1e-
z
{L
m
(a)(x) D L,,(a)(x) - /",,(al(x) J) L",(a)(x) l].
117] ORTHOGONALITY 205
Therefore we have
(3) (m - n) fbxae-XLn(a) (x)Lm(a)(x) dx
a
The product of e-
X
and any polynomial in x ---4 0 as x ---4 00.
Furthermore, x
a
+
1
---40 as x ---40 if Re (a) > -1, so equation (3)
yields the orthogonality property
(4) iooxae-XLn(Cl)(x)Lm(a)(x) dx = 0,
m n, Re(a) > -1.
Equation (4) shows that if Re(a) > -1, the polynomials Ln(a)(x)
form an orthogonal set over the interval (0, (0) with weight function
xae-
X
. We now need to e\aluate the integral on the left in (4) for
m = n. For the sake of variety we use the Rodrigues formula
d
(5) Ln(a)(x) = -1z!Dn[e-
x
x
n
+
a
], D = dx'
to evaluate the integral on the left in (4) both for m = nand m n.
Because of (5) we may write
and then integrate by parts n times to obtain
(6)
for Re(a) > -1. At each integration by parts the integrated
portion,
o < k ; n,
vanishes both at x = 0 and as x ---4 00.
Since Lm(a)(x) is of degree m, DnLm(a)(x) = 0 for n > m. There-
fore the integral on the left in (6) \anishes for n > m. Since that
integral is symmetric in nand m, it also vanishes for n < m, which
completes our second proof of (4).
We know that
DnLn(a)(x) = D{( + 1r
n
-
1
] = (-l)n.
206 LAGUERRE POLYNOMIALS [Ch.12
Hence, for m = n, equation (6) yields
or
(7)
L
'" r(1 + a + n)
x"e-
x
[L,,(al(x)]2 dx = -----, ---,
o n.
Re(a) > -1.
In the notation of the theory of orthogonal polynomials, Chapter
9, we ha\'e
r(1 + a + n)
g" = , '
n.
( -1)'1
hn = --,-'
n.
THEOREM 68. If a > - 1, the Laguerre polynomials have the
following properties:
(8) i"'xae-XLncal(X)Xk dx = 0, k = 0 1 ') ", (n - 1)'
, , ... " ,
(9) The zeros of Ln(a)(x) are posit'ive and distinct;
(10) k!Lk (al(x)L
k
(al(y) = I)! (x) (al(y).
(l+a)k (l+a)n x-y ,
(11) If i"'xae-XP(x) dx exists,
Lim [Q_+, a)n]-l r;ac-xf(x)Ln(al(x) dx = O.
n+ro n. J
o
'
The three-term recurrence relation for Ln(al(x) has already been
obtained; it is equation (4), page 202.
118. Expansion of polynomials. Since the Ln(al(x) form an
orthogonal set, the classical technique for expanding a polynomial
by the method indicated in Theorem 56, page 1;')1, is a\'ailable. .-\s
usual we prefer to treat the problem by obtaining first the expansion
of xn and then using generating function techniques whenever we
can.
Equation (1), page 201, yields
(1)
Therefore
118] EXPANSION OF POLYNOMIALS
207
0> (-x)nt
n
0> n (-l)n-kL
k
(a)(x)t"
(1 + a)nn! = {; (n - k) !(1 + ah
from which it follows that
(2)
Let us employ (2) in expanding the Hermite polynomial in a
series of Laguerre polynomials. Consider the series
0> Hn(x)t
n
00 (-1)s(2x)nln+28
L --f- = exp(2xt - t2) = L , ,._-
n. 8.n.
= f -Hn-I); -t] X
... k_O _ Ha+n+k), - Ha+n+k-1);
(- I)k2n+k(1 + a)n+kLk(a)(x)t
nH

ron [-Hn-k),-Hn-k-l); ]
= L L 2F2 -! X
I( ) I( 1)
-'2 a + n ,-'2 a + n - ;
(-1)k2n(1 +
(n - k)!(l + a)k
From the above we may conclude that
n [-Hn-k), -Hn-k-1); -4
1
]X
(3) Hn(x) = 2"(1 + a)n L 2F2
_ Ha+n), - Ha+n-l);
(
(1 + a) k
208 LAGUERRE POLYNOMIALS [Ch.12
N ext let us expand the Legendre polynomial in a series of Laguerre
polynomials. Consider the series
= i: (-1)'( !)n+.(2x) nt
n
+2.
s!n!
= : _C -_1 L"_+_82_n+_k_C ,,_' (a_) C_x ) t n_+_k +_28
n.k . s!n!(l + a)k
00 n [ -Hn-k),-Hn-k-l);]
= L L 2
F
3 t X
n=O k=O ! _ n, - Ha + n), - Ha + n - 1);
We may therefore write
C -1)k2
n
CDn(1 + a)"Lk(a)(x)t
n
(n - k) !(1 + ;r:---'
( -n)kLkC"jx).
(1 + a)k
The Laguerre polynomial can be expanded in series of either
Legendre or Hermite polynomials by employing precisely the tech-
nique used above with the aid of the pertinent expansions of xn
119] SPECIAL PROPERTIES 209
from Chapters 10 and 11. This is left for the reader to do; the
results may be found in Exs. 2 and 3 at the end of this chapter.
119. Special properties. The generating functions of Section
113 lead to certain simple finite sum properties of the Laguerre
polynomials. For instance, from
(1)
(
xt) ro
(1 - t)-J-aexp = L LnCal(X)t
ft
,
I - t n-O
and
(1 - = (I - t)-Ca-/l)(1 -
it follows at once that
(2) Ln(a)(x) = t (a -
k.
for arbitrary a and {3.
From equation (1) and the fact that
(1 - t)-J-aexp( -xt )(1 -
1-t 1-t
it follows that
= (1 _ t)-J-(a+/l+J)exp(=-(X + y)t)
1 - t
n
(3)
Ln (a+Il+!)(x + y) = L Lk

The generating relation
ro LnCal(x)t
n
(4) e
t
aF\( -; 1 + a; -xt) = L (1 + )
n::zO a n
together with the fact that
etoFJ( -; 1 + a; -xyt) = e(l-lil
t
e
lit
oFJ( -; 1 + a; -x(yt)
yields
ro Ln(al(xy)tn _ (ro (1 _ y)ntn)( ro Ln(a)(x)yntn)
L --- - L L --c--'--'--":-
(1 + a)n n! (1 + a)n
from which we get
(5) LnCa)(xy) = t (1 + a)n(1 - y)n-kykLkCa)(x).
(n - 1.:)!(1 + a)k
For (5) see also Ex. 1, page 145.
210 LAGUERRE POLYNOMIALS [Ch. 12
We know that for arbitrary c,
(6)
[
C'
(1 - t)-c IFI '
1 + a;
By Kummer's first formula, page 125, we have
(7) IFl[ c; = exp(.i?t) IFl[1 + a - c;
1 + a; 1 + a;
xt ]
1 - t
Using (6) and (7) we write
i:
,,=0 (1 + a)"
(
t) [1 + a - c;
= (1 - t)-c exp t IF!
1 + a;
xt ]
1 - t
(
t) [1 + a - c;
= (1 - t)-1-(2c-a-2)exp (I - t)-(I+a-c)\F\ 1 + a;
xt ]
1 - t
= [i: Ln(2C-a-2)(x)t n ][ i: - C)n
L
n(a)0
x
)t
n
],
n=O ,,=0 (I + O')n
with the aid of (1) and (6). We conclude that for arbitrary c (not
zero or a negative integer)
(8) Ln(a)(x) = (1 + O')n (I + a - C)kLk(a)(
(C)n k=O (1 + O'h
In equation (8) the two special choices C = 1 + !o' + !m and
c = 1 + a + m, for non-negative integral tn, are particularly rec-
ommended. See Exs. 6 and 7 at the end of this chapter.
We next seek for Laguerre polynomials a relation analogous to
equation (7) of Section 95 on Legendre polynomials and equation
(1) of Section 111 on Hermite polynomials. Consider the series
'" '" (n + '" n nltn-kvkLn(a)(x)
{; S klnl = {; {; kl(n - k)l
'" (-x(V + t)
= L L,,(a)(x)(t + v)n = (1 - t - v)-\-aexp .
n=O 1 - t - v
120] OTHER GENERATING FUNCTIONS 211
We wish to expand the right member above in powers of v in another
way. Now
[
V J-I-a
(1 - t - v)-I-a = (1 - t)-I-a 1 - 1 _ t
and
[
-xCv + t)J ( -xt) [ -xv J
exp = exp exp
1 - t - v 1 - t (1 - t) (1 - t - v)
[
-X 11]
--0--
-xt 1 - t 1 - t
= expC _ t) exp 11'
1---
1 - t
Hence we may write
f f (n +
k_O n_O kIn!
= (1 - t)-I-a f LIc(a)(_X )( __
1 - t k=O 1 - t 1 - t
We find, by comparing coefficients of Vic, that
(9)
(n + k) = (1 _ t)-I-a-k ( -xt)L (a)(_X_)
L.. k I I exp 1 - t k 1 - t '
""",0 .n.
a relation which is useful in discovering generating functions.
120. Other generating functions. Consider the series
f n!Ln(a)(x)Ln(a)(y)t
n
= f i: (-l)kn !ykL
n
(a)(x)t
n
n=O (1 + a)n n=O k!(n - k)!(1 + a)k
For the moment let
n!Ln(a)(x)Ln(a)(y)t
n
.
if; = L..
n=O (1 + a)n
We may now use equation (9) above to conclude that
212 LAGUERRE POLYNOMIALS [Ch.12
-xt 1 - t
( )
co (1 - t)-kLk(al(_X_) ( _yt)k
if; = (1 - t)-I-", exp 1 _ t 6 (1 + a)k
xyt ]
(1 - t)2 .
THEOREM fi9. If I t I < 1 and a is not a negative integer,
[
- (.r + y) t] ,[ - ;
(1 - t)-I-a exp ----y-=--T--- ,,1'1
1 + a;
xyt ]
(I - t)Z
= f ll!i.J
n
(al(x)L
n
(al(y)t
n
.
(1 + a)n
Theorem 69 has been known for a long time.
Applying formula (9) of the preceding section to the relation
(1)
[
C'
(1 - t)-c IFI '
1 + a;
which is (3) of Section 113, yields a result obtained by contour in-
tegral methods by Brafman [2]. We shall perform this transforma-
tion to exhibit the technique. In (1) replace x by x(I-t)-I, t by
(-yt) (1- t)-I, and multiply both sides by (1- t)-I-a exp[ -xt(I- i)-I].
The result is
-1- [ yt ]-c ( -xt) [ c; xyt(I - t)-2 ]
(1 - t) a 1 + 1 _ t exp 1 _ t IFI 1 + yt(I - t)-I
1 + a;
co (c)k(I - t)-I-a-k ex
p
( -xt )Lk(al(_X_)( -I)kyktk
=L: I-t I-t
(1 + ah
= f (C)k(n + -I)kyktk
(1 + a)kk!n!
121] SIMPLE LAGUERRE POLYNOMIALS 213
00 [-n, c;
= L: Ji'1
n=O 1 + a;
We may rearrange the resulting relation into the form
(l-t)-I+<-"(l-t+yt)-c exp IFI
(
t) [ c;
1 t 1 + ex;
xyt ]
(l-t)(l-t+yt)
Y] L.''(x)",
a bilateral generating function involving the Laguerre polynomial
and a certain terminating 2F
1

121. The simple Laguerre polynomials. When a = 0 the re-
sultant polynomial is denoted by Ln(x). It is called the simple
Laguerre polynomial or just the Laguerre polynomial when no con-
fusion with Ln(a)(x) is anticipated. Because Ln(x) arises frequently,
we now list properties of that polynomial for convenience of the
reader. Each of the following results, except for (27), may be
obtained by putting a = 0 in a result already known for Ln(a)(x).
(1)
(2)
(3)
(4)
(4a)
(5)
(6)
Ln(x) = IF
1
( -n; 1; x),
" (-l)kn!xk
L,,(x) = 6 (k!)2(n - k)!'
00 L (x)t"
"n! '
(
-Xl) co
(1 - t)-1 exp -- = L: L,,(x)t",
1 - l n=O
xL,,'(x) = nLn(x) - nL
n
_
1
(x),
214
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(15)
(16)
(17)
(18)
(19)
LAGUERRE POLYNOMIALS [Ch.12
n-1
nLn(x) = (2n - 1 - x)Ln_1(x) - (n - 1)L
n
_
2
(x),
ex d
n
L (x) = - - (xne-
x
)
n n! dxn '
XL""(X) + (1 - x)Ln'(x) + nLn(X) = 0,
f
i>
(m - n) e-xLn(x)Lmex) dx
a
= [xe-x1Lm(X)Ln'(x) - L,,(x)Lm'(x) I I,
m ~ n,
k = 0, 1,2, .. , (n - 1),
The zeros of Ln(x) are positive and distinct,
"
L Lk(X)Lk(Y) = (n+ 1)(x-y)-l[L
n
+1(y)L
n
(x) - Ln+1(x)Ln(y)],
k=O
If i"'e-XP(x) dx exists,
Lim i"'e-Xf(x)Ln(x) dx = 0,
,.+'" 0
n {-!en - k), -!en - k - 1); ]
(20) H,.(x) = 2nn! L 2F -i .
k=O -.In -.l(n - 1)
2, 2 ,
(-n)kLk(x) .
k! '
n> 1,
121] SIMPLE LAGUERRE POLYNOMIALS 215
n [-Hn - k), -!en - k -1); -4
1
J
(21) Pn(X) = 2
n
(t)n L 2F3
1 1 1 ( 1)
z - n, - zn, - z n - ;
n
(24) L,,(xy) = L Cn,k(l - y)n-kykLk(X);
k=O
(25)
-! [-(x+y)t] [-' xyt J _ 00 n
(26) (I-t) exp of! 1; (I-=t)2 - 'foL,,(x)Ln(Y)l,
(27)

= L L,,(y)Pn(x)t
n
,
n-O
(
-xt)
(28) (1 - t)c-!(I - t + yt)-c exp --
1 - t
[
C'
IF! '
l' ,
xyt J [-n, c;
-,-,------,--:-:-='----,- = L..J 2F!
(l-t)(l-t+yt) 1;
y loex)!O,
216
EXERCISES
1. Show that
LAGUERRE POLYNOMIALS [Ch.12
H
2n
(x) = (-l)n22nn!LnH)(x2),
H
2
n+l(X) = (_l)n22n+1n!xLn(!)(x2).
2. Use Theorem 65, page 181, and the method of Section 118 above to derive
the result
1J( -1) 1)1' k(X).
2k(n-k) -+ah
3. Use formula (4), page 194, and the method of Section 118 to derive the result
" [-i(n - k), -!en - k - 1);
Ln(al(x) = L: ZF2
k_o HI -+ Q' + k), H2 + a + k);
4. Use the results in Section 56, page 102, to show that
5. Use the results in Section 56, page 102, to show that
itH",h1x(t - x) dx = (-I)"1l"22n O)"L,,{!t2).
o
6. Show that if m is a non-negative integer and a is not a negative integer,
I (al(x) = __ i -=_
In (1 + !a + k_O (l + Q') k
7. Show that if m is a non-negative integer and a is not a negati\-e integer,
8. Use integration by parts and equation (2), page 202, to show that
I:-YLn(a)(y) dy = e-'[Ln(a)(x) - L;,"l,(x)].
9. Show that
{txa(t _ X)13-1L (a)(x) dx = f(1 + Q')f(iJ2 (I + Q')nt
a
+
13
L
(a+13)(t)
Jo " f(l -+ a + iJ) (1 -+ a -+ (:I)n n
10. Show that the Laplace transform of L,,(t) is
{co 1 ( l)n
Jo c'tLn(t)dt = S 1 - s .
121] SIMPLE LAGUERRE POLYNOMIALS 217
11. Show by the convolution theorem for Laplace transforms, or otherwise, tha.t
I il
Ln(t - x)Lm(x) dx = 0 Lm+n(x) dx = Lm+,,(t) - Lm+n+l(t).
12. Evaluate the integral
IO>x
a
e-
Z
[L,,(a)(x)]2 dx
of (7), page 206, by the following method. From (4), Section 113, show that
f iO>xae-X[Ln(a) (x)]2dx t2n = (1 - t)-2-2a. (""x
a
exp[-x(l + OJ fix
n=O 0 Jo 1 - t
CHAPTER 13
The Sheffer
Classification
and Related Topics
122. Differential operators and polynomial sets. Let <Pn(X);
n = 0, 1,2, .. " be any simple set of polynomials and let D == d/dx.
Let us define the set (not necessarily a simple set) of polynomials
Tn(x), n ~ 0, by
(1) To(x)D<Pl(X) = <Po (x) ,
n-l
(2) Tn(x)Dn+l<pn+l(X) = <Pn(X) - L.,Tk(x)DH1<pn+l(X), n ~ 1.
k ~ 1 )
Because <pn(X) is of degree precisely n for each n, it follows that
T n(X) is uniquely defined and is of degree ~ n. Note that D<Pl is
constant, as is <po, so To(x) is constant. For n ~ 1, each Tn(x) is
defined by (2) in terms of previous elements of the set, Tk(x) for
o k ~ (n - 1). Because Dn+1<Pn+l is constant and the degree of
Tk(x) DH1<pn+l(X) exceeds the degree of Tk(x) by exactly (n - k),
each member of (2) has degree at most n.
THEOREM 70. For the simple set of polynomials <Pn(x) there exists
a unique d1jJerential operator of the form
co
(3) J(x, D) = L., Tk(x) DHI
k ~ O
in which Tk(x) is a polynominal of degree ~ k , for which
(4)
218
n;::;;1.
122] DIFFERENTIAL OPERATORS
219
It is important that J be independent of n.
Proof: The requirement (4) demands that, for n ~ 1,
n-I
L Tk(x) DHIIPn(X) = IPn-I(X),
k_O
which is merely a restatement of (1) and (2). Equations (1) and
(2), as we saw, determine Tk(x) uniquely.
We say that the polynomial set IP,,(x) belongs to the operator J
and that J is the operator assoeiated with the set IPn(X). There is
only one such operator associated with a given IPn(x),but there are
infinitely many sets of polynomials belonging to the same operator.
THEOREM 71. A necessary and sufficient condition that two simpZr
sets of polynomials IPn(X) and if;,,(x) belong to the same operator Jis
that there exists a sequence oj numbers b
k
, 'independent of n, such thal
n
(5) if;n(x) = L bkIPn-k(X).
k ~ O
Proof: Assume (5) to hold. There exists an operator J to which
IPn(X) belongs. That if;n(x) belongs to the same operator follO\vs
from
n n-l
Jif;n(x) = L bkJ IPn-k(X) = L bkIP,,-I-k = if;"-I(X).
k ~ O k ~ O
Next assume that IPn(X) and if;n(x) belong to the same operator J.
We need to show that the b
k
of (5) exist. We know, because IP,,(x)
and if;n(X) are simple sets of polynomials, that there exists the
relation
n
(6) if;n(x) = L A(k, n)IPn-k(X),
k ~ O
but, in general, the coefficients A(k, n) depend upon n as well as
on k. Since IPn(X) and if;n(x) belong to J, we may apply J to each
member of (6) and obtain
n-I
(7) n ~ 1.
Recall that J IPo(x) = 0, which is the reason that the term A (n, n) IPo(x)
dropped out when the operator J was applied to (6). We may
shift index from n to (n + 1) in (7) to get
n
(8) if;n(X) = L A(k, n + l)IPn-k(X); n ~ 0.
k ~ O
220 THE SHEFFER CLASSIFICATION [Ch. 13
Comparing (6) and (8), we see that
A(k, n) = A(k, n + 1)
for all k, n. Then A(k, n) = b
k
, independent of n.
Not every operator of the form (3) is associated with some
polynomial set in the sense we have defined. For the operator J
of the form (3) to be associated with some simple set, it is necessary
and sufficient that J transform every polynomial of degree precisely
n into a polynomial of degree precisely (11 - 1).
EXAMPLE: Determine the operator assoeiated with the set
<p,,(x) = Hn(X)j(n!)2, in which /I,,(x) is the Hermite polynomial.
Here <Po (x) = 1, <Pl(X) = 2x, <P2(X) = X2 - L <P3(X) = - -!rx,
etc. We seek an operator J of the form
co
J = L 7\ (x) D k J

such that J <pn = <pn-I for n 1. Then
or To(x) 2 = 1,
so that To(x) =!. Next we have
[To(x) D + T1(x) D2]<P2(X) = <Pl(X),
or
Then
H2x) + Tl(X)(2) = 2x,
so that T1(x) = !x. In turn
[!D + !x D2 + T
2
(x) - !x) = X2 - t,
or
!C!X
2
- !) + + = X2 - t,
from which T
2
(x) = -}.
If we continue the procedure, we find that T3(X) = 0,
T4(X) = 0, and we begin to suspect that J may terminate. Let us
therefore define
(9)
operate on <p,,(x) with J
1
and see whether the result is <Pn-l(X).
123] SHEFFER'S A-TYPE CLASSIFICATION
Now
J
()
- J Hn(X) - 1 [lH' + 1 H II IH "']
X - 1 (n!)2 - (n!)2 2" n 2"X n - 4: n
- 1 [H '" 2 H " 2H ']
= 4(n!)2 n - X n - n
From Hermite's differential equation
lIn" - 2xHn' + 2nH" = 0
we obtain
H .. '" - 2xH
n
" - 2H,/ + 2nH
n
' = 0
so that we have
-[ -2nH
n
'] Hn'
= -- 4( ')2 = -2 T(-- - 1) I
n. n. n .
But we also know that Hn'(x) = 2nH
n
_
l
(x). lIenee
2nH
n
_
l
(x) Hn_l(x) )
= 2nT(n-=--I)! = [(;;:---=-1)1]2 = ,
221
as desired. Therefore = Hn(X)/(n!)2 belongs to the operator
J
I
of (9).
The example above yields a specific result of interest to us later,
but its simplicity may be misleading. It is wise to keep in mind
that the operator associated with a given simple set of polynomials
need not terminate and, indeed, that there is no reason to think that
the Tk(X) in the operator of equation (3) can be determined other
than successively by iteration of equation (2).
123. Sheffer's A-type classification. Sheffer [1] used the oper-
ators discussed in Section 122 to classify polynomial sets. The
first of his classifications will now be discussed.
DEFINITION: Let be a simple set of polynomials and let
belong to the operator
co
(1) J(X, D) L Tk(x) DHl,

with Tk(X) of degree k, in the sense of Section 122. If the
maximum degree of the coefficients Tk(X) is m, \ve say that the set
is of Sheffer A-type m. If the degree of Tk(x) is unbounded
as k ----> co, we say that is of Sheffer A-type co
222 THE SHEFFER CLASSIFICATION [Ch.13
EXAMPLE: The set 'Pn(X) = H n(x)/(n!)2 is of Sheffer A-type one,
since 'Pn(X) belongs to the operator J
1
of (9), Section 122.
Explicit polynomial sets of type m for each m are obtained, with
corresponding generating functions, by Huff and Rainville [lJ.
They show, among other things, that if Yn(x) is defined by
co
(2) 'P(t) oF m( -; {31, {32,' . " 13m; ([xt) = L Yn(x)t
n
,
n=O
with ([ constant and 'PCt) analytic and not zero at t = 0, then Yn(X)
is of Sheffer A-type m. The ShefTer [1 J paper eontains a study of
polynomials of any A-type, but the most satisfying results are those
bearing on polynomials of type zero.
124. Polynomials of Sheffer A-type zero. Let 'Pn(X) be of
Sheffer A-type zero. Then 'Pn(X) belongs to an operator
en
(1) J(D) = L Ck DHI
k ~ O
in which the Ck are constants. K ote that Co ;6- 0, since J 'Pn = 'Pn-I'
Furthermore, since Ck is independent of x for every k, a function
J(t) exists with the formal power-series expansion
en
(2) J(t) = L CktHI, Co ;6- 0.
k ~ O
Let H(t) be the formal im'erse of J(t); that is,
(3) J(H(t) = H(J(t) = t.
THEOREM 72. A necessary and sufficient condition that 'Pn(X) be
of Sheffer A-type zero is that 'Pn(X) possess the generating function
indicated in
ro
(4) A(t) exp[xH(t)J = L 'P"(x)t",
n=O
in which H (t) and A (t) have (at least the formal) expansions
co
(5) H(t) = L h tn+I
n , ho ;6- 0,
n=O
co
(6) A(t) = L ant
n
, ao ;6- O.
n _0
124] POLYNOMINALS OF SHEFFER A-TYPE ZERO 223
Proof: Assume (4), (5), and (6). Then the <pn(X) form a simple
set of polynomials. The function H(t) has a formal inverse. Call
it J(t), defined by (3),
m
(7) J(t) = L ckt
k
-
H
, Co ~ 0.
k _ ()
Let, as usual, D = dld:r. Then, by (4),
,>.
J (D) L <Pn(X)t
n
= J (D) IA(t) exp[xH(t)]}
1_0
= A(t)J(H(t) exp[xH(t)]
= tACt) exp[xH(t)J.
Hence
00 00
L J (D) <Pn(X)t
n
= L <Pn(X)tn+I
a_O n=O
1i.=1
from which J(D) <po (x) = and J(D) <Pn(X) = <Pn-l(X), n ~ 1. Since'
J(D) is independent of x, <pn(X) is of Sheffer A-type zero.
Next assume that <pn(X) is of Sheffer A-type zero. Let <Pn(X)
belong to the operator J. Then
co
(8) J = J (D) = L ckDk+I, Co ~ 0,
k ~ O
and the Ck are independent of x because <pn(X) is of A-type zero. The
function J(t) has a formal inverse H(t) defined by (3) above. Con-
sider the polynomials if-'n(X) , a simple set, defined by
00
(9) exp[xH(t)] = L if-'n(X)tn.
n=O
We know already that if-'n(X) is of Sheffer A-type zero and that it
belongs to the operator J(D). By Theorem 71, page 219, there
exists a sequence ak, independent of n, such that
n
(10)
<Pn(x) = L akif-'n-k(X).
k ~ O
Define A (t) by
ro
A (t) = L antn.
Tl=O
224 THE SHEFFER CLASSIFICATION [Ch.13
Then
co n
= L: L: Gk'/;n_k(X)t
n
TI ~ O k ,-" n
which is (4).
The generating function in Theorem 72 is one of Boas and Buck
kind (Seetion 7G) with I'n = l/n!' It follows that Theorem 50,
page 141, may be applied to any 'Pn(X) of Sheffer A-type zero.
Sheffer [IJ obtained many properties of polynomial sets of A-type
zero. Here we state (with some modifications in notation) only a
few of his results.
THEOREM 73. Let 'Pn(X) be of Sheffer A-type zero; let 'Pn(X) belong
to the operator J CD) and have the generating function in C 4) of Theorem
72. There exist sequences ak and tk, independent of x and n, such that
forn ~ 1,
n-l
(11) L: (ak + Xtk)JHI CD) 'P,,(x) = ti'P,,(x),
"'_0
(12)
co
(13) L: tktk = R'Cl).
k ~ U
Of course Ek = (k + l)h
k
, in terms of the hk of equation (5), page 222.
THEOREM 74. With the assumptions oj Theorem 73 there exist
sequences IJ.k and Vk, 1'ndependent of x and tI, such that
n-I
(14)
L: (/-Lk + xVk)Dk+I'Pn(x) tI'Pn(X),
k ~ O
( 15)
f /-LktH1 = ?lA'(u)
k_O . ACu) ,
'U = J(t),
co
(16)
L: vktHI = uH'(u), u = J(t).
k ~ U
1241 POLYNOMINALS OF SHEFFER A-TYPE ZERO 225
THEOREM 75. A necessary and sufficient condition that <Pn(X) be
of Sheffer A-type zero is that there exist sequences O!k and Ek, independent
of x and n, such that
n-1
(17) L (O!k + XEk)<Pn-1_k(X) = n<Pn(X).
k=O
The O!k and Ek of Theorem 75 are those of Theorem 73.
THEOREM 76. A necessary and Suffic1"cnt condition that <pn(X) be of
Sheffer A-type zero is that there exists a sequence hk' 1"ndependent of
x and n, such that
n-1
(18) n 1.
Proof of Theorems 73--76: These theorems all follow from Theorem
72. For instance, consider the series
'" a
L n<pn(x)t
n
= t-t {A (l) exp[xH(t)] I
n=O a
= t[A'(l) + :rH'(t)A (t)] exp[xH(t)]
= + xH'(l) ] A (t) exp[.rH(t)J.
The functions A'(t)/A(t) and H'(t) have series expansions; let them
be (12) and (13), thus defining O!k and Ek. Then
t; n<pn(x)t
n
= {; (an + <pn(X)t
n
]
m n
= L L (ak + XEk)<Pn_k(X)t
n
+
1
n=O k=')
en n-l
= L L (ak + XEk)<Pn_1_k(X)tn.

Since
Jk+1(D) <pn(X) = Jk(D) <Pn-1(X) = """ = <Pn-1-k(X),
the above argument yields both Theorems 73 and 75.
Theorem 74 is a result of rearrangement of terms in the result
of Theorem 73.
Theorem 76 follows from the fact that
226 THE SHEFFER CLASSIFICATION [Ch.13
'" a
L 'P,.'(x)t
n
= -IA(t) exp[xH(t)] I
n=1 ax
= H(t)A(t) exp[xH(t)]
co n-l
= L L hk'Pn_I_k(X)t
n
,
n _ 1 k= 0
and the fact that from the equality of the first and last series above,
it follows that
en
Y = L 'Pn(X)t
n
n=O
must satisfy the equation Dy = H(t)y.
We shall meet many instances of polynomial sets of Sheffer A-
type zero. In examining a new set of polynomials, it seems de-
sirable always to obtain its Sheffer A-type. For Sheffer's classifica-
tions of B-types and C-types and for more results on the A-type,
see Sheffer [lJ and also the exercises at the end of this chapter.
125. An extension of Sheffer's classification. It can be seen
from the preceding section that much of the value of the A-type
classification lies in the existence of the generating relation
00
A(t) exp[xH(t)] = L 'Pn(x)t
n
1l....;1)
for sets 'Pn(X) of A-type zero. An essential characteristic of the
function y = exp[xH(t)] is that, with D = djdx,
Dy = H(t)y.
It is natural to expect that much can be retained if D is replaced by
some other differential operator (J" and exp(z) by some other function
F(z) such that
(J"F(z) = F(z).
Let D = djdx, () = xD, and define the differential operator (J" by
q
(1) (J" = D IT () + b
i
- 1),
1=1
in which the b
i
are constants and no b
i
is either zero or a negative
integer. Note that application of (J" reduces the degree of any
polynomial by exactly one.
125] EXTENSION OF SHEFFER'S CLASSIFICATION 227
THEOREM 77. Let 'Pn(X) be a simple set of polynomials. There
exists a um'que set of polynomials Tk(x), Tk(x) of degree ~ k , such that
the d1jJerential operator
co
(2) J(x,O') = L Tk(X)O'Hl
k_O
has the property that
(3) J(X, O')'Pn(X) = 'Pn-I(X), n ~ 1 .
It is to be noted that J(x, O')cpo(x) = 0 and that J(x, 0') is independ-
ent of n.
Proof: Given CPn(X), define polynomials Tn(x) successively by
(4)
n-l
(5) Tn(x)O'n+I'Pn+l(X) = 'Pn(x) - L 'Tk(X)aH1cpn+l(X), n ~ 1.
k_O
The definition of Tn(x) is unique and the degree of Tn(x) is ~ n,
by the same arguments as were used in the Sheffer classification.
Furthermore,
n-l
J(X, a)CPn(X) = L Tk(X)ak+lcpn(X) = CPn-I(X)
k-O
by (5) with n replaced by (n - 1).
Again we say that the polynomial set 'Pn(x) belongs to the oper-
ator J(x, 0') and that J(x, a) is the operator associated with CPn(X).
There is only one operator associated with CP,,(X) , hut infinitely
many sets of polynomials belong to a specified permissible (it must
transform every polynomial of degree n into one of degree n - 1)
operator J(x, a).
THEOREM 78. A necessary and sufficient cond1'tion that two sl'mple
sets of polynomials CPn(X) and ""n(X) belong to the same operator J(x, a)
is that there exists a sequence of numbers b
k
, independent of n, such that
n
(6)
""n(x) = L bk'Pn-k(X).
k ~ O
Proof: Parallel the proof of Theorem 71 in every respect.
DEFINITION: Let 'Pn (x) be a simple set of polynomials, and let
'Pn(X) belong to the operator
co
(2) J(x, a) = L TkeX)aHI,
k_O
228 THE SHEFFER CLASSIFICATION [Ch.13
with Tk(x) of degree ;;;2k and J(x, U)ipn(X) = ipn-I(X). If the maxi-
mum degree of the coefficients Tk(x) is m, we say that the set
ipn(X) is of u-type m; if the degree of Tk(x) is unbounded, as k -+ 00,
we say the set ipn(X) is of u-type 00.
126. Polynomials of u-type zero. Let ipn(X) be of u-type zero.
Then ipn(X) belongs to an operator
'"
(1) J(u) = L CkU
Hl
k ~ O
in which the Ck are constants and Co r 0, since J(U)ipn(X) = ipn-l(X).
Here again we are using D = d/dx, 0 = xD,
q
(2) u = D IT (0 + b, - 1),
1==1
as in the preceding section. Since Ck is independent of x, a function
J(t) exists with an inverse H(t):
J(HU) = H(J(l)) = t,
en
(3) J(t) = L: C t
n
+!
n , Co r 0,
Tt=O
~
(4) H(t) = L hntn+l,
ho ~ 0.
n=(}
THEOREM 79. A necessary and sufficient condition that ipn(X) be of
u-type zero, with
q
(5) u = D IT (0 + b, - 1),
1=1
is that ipn(X) possess the generating function in
'"
(6) ACt) oFq( -; b
l
, b2,' ", b
q
; xH(t)) = L ipn(X)t
n
,
n ~ O
in which H(t) is given by (4) and A(t) has the formal expansion
m
(7) ACt) = L ant
n
, ao ~ 0.
n=O
Proof: Assume (6) together with (4), (5), and (7). Then the
ipn(X) form a simple set of polynomials. The function H(t) has a
formal inverse; call it J(t) defined by (3). As we saw in Chapter 5,
page 75, the function
126] POLYNOMIALS OF TYPE ZERO
is a solution of the differential equation
(8) [8. g (8. + b; - 1) - z] y = 0;
d
8. = z_
az
229
Put z = xH(t) and hold t constant. Then 8 = x(d/dx) = 8. and (8)
becomes
q
(9) 8 II (8 + bi - l)y = xH(t)y.
But 8 = xD, so that (5) and (9) combine to yield
(10) <YoFq(-;b1,,bq;xH(l) = H(t)oFi-;b], ... ,bq;xH(t).
We now operate with J(u) on both members of (6):
<X>
J(u) L 'Pn(X)t
n
= J(u)A (t) oFq( - ; b],' . " b
q
: xH(t)
n-U
UJ (};)
= t L 'P,,(x)t
n
= L 'P" ](x)t
n

n=(J 1/= I
Therefore J(u) 'Po (x) = 0 and J(U)'Pn(X) = 'Pn-l(X) for 11 1. Since
J(u) is independent of x, 'Pn(X) is of u-type zero.
Next assume that 'Pn(X) is of u-type zero. Let 'Pn(X) belong
to the operator J(u). The function J(t) has a formal inverse H(t).
Consider the polynomials "'n(X), a simple set, defined by
<X>
(11) oFl-;b1,,bq;xH(t) = L"'n(X)t
n
.

We know that "'n(X) is of u-type zero and that it belongs to the
operator J(u). By Theorem 78, page 227, there exists a sequence
ak, independent of n, such that
n
(12) 'Pn(X) = L ak"'n_k(X),

Define A(t) by (7). Then, using (12), we find that
'Pn(X)t
n
= "'n(X)t
n
)
= A(t) oFl-; b
1
,, b
q
; xH(t),
as desired.
230 THE SHEFFER CLASSIFICATION [Ch.13
It is of interest to note that polynomials of any Sheffer A-type
may be polynomials of u-type zero, if u is properly chosen. It is
also easy to show that a polynomial of Sheffer A-type unity or
larger may not be of u-type zero for any choice of u.
As an example, consider polynomials Yn(X) defined by
m
(13) e' of q( - ; 1, 1" . " 1; - xt) = L Yn(X)t
n
.
n=O
Here H(t) = -t and J(t) = -to Let u = Doq. Then the Yn(X)
are of u-type zero for that u, but are of Sheffer A-type q, because u is
a polynominal in x and D and is of degree precisely q in X.
Because the generating function in Theorem 79 is of the Boas
and Buck kind (Section 76), Theorem 50, page 141, applies to all
polynomials of u-type zero.
THEOREM 80. With u specified as in (5) of Theorem 79, page 228
a necessary and sufficient condition that 'Pn(X) be of u-type zero is that
there exists a sequence hk' independent of x and n, such that
n-I
(14)
Proof: If <Pn(X) is of u-type zero, it follmn; from Theorem 79 that
en
L u<Pn(x)t
n
= uA (I) oFq( -; bl ,' . " b'l; xH(t)
n=O
= H(t)A (t) oFi -; b
l
,' . " bq; xH(t
= (f; <pn(X)t
n
)
0:> It-l
= L L hk<Pn_I_k(X)t
n
,
n=1
so that (14) is satitified. If (14) is satisfied, let
co
y(x, t) = L <Pn(x)t
n

n=O
Then, by (14),
co n-l
uY = L L hk<Pn_I_k(X)t
n
= H(t)y(x, t).
n=1 k=O
Hence y(x, t) is one of the solutions of the oP
q
equation (Chapter 5,
126] POLYNOMIALS OF TYPE ZERO
page 75), with z = xH(t). But
(15) oFi - ; bl ,' . " bq ; xH(t)
231
is the only one of those solutions which is nonzero and analytic at
x = 0, (z = 0). Hence y is a constant (function of t not x) multiple
of (15), which concludes the proof.
Our main interest in both the Sheffer and the (I-type classifications
centers on the generating functions in Theorems 72 and 79. Upon
meeting a new simple set of polynomials, we try to determine its
Sheffer A-type and to see whether the set is of (I-type zero for some
(I. These tools will be used at times in later chapters.
EXERCISES
1. Prove Theorem A (Sheffer): If <p,,(x) is of Sheffer A-type zero, y,,(m, x) =
D"'<P,,+m(X) is also of Sheffer A-type zero and belongs to the same operator as does
<p,,(x).
2. Prove Theorem B: If <p,,(x) is of Sheffer A-type zero,
is of Sheffer A-type m.
3. Show that
",,,(x) = <pn(x{U (1 + P,)n JI
is of Sheffer A-type zero, obtain the associated .T(O, H(O, A(O, and draw
what conclusions you can from Theorems 73-76.
4. Show that L" (al(x) is of Sheffer A-type zero, and proceed as in Ex. 3.
5. Show that the Newtonian polynomials
are of Sheffer A-type zero, and proceed as in Ex. 3.
6. Show that
L,,(al(x)
<Pn(X) = (1
is of Sheffer A-type unity but that with u chosen to be
u = D(B + a)
the polynomials <p,,(x) are of u-type zero.
7. Determine the Sheffer operator associated with the set
L,,(x)
<p,,(x) = (nT)"2 '
and thus show that <p,,(x) is of Sheffer A-type 2.
232 THE SHEFFER CLASSIFICATION [Ch.13
8. Prove that if we know a generating function
co
y(x, t) = L <Pn(X)t
n
n_O
for the simple set of polynomials <Pn(X) belonging to a Sheffer operator J(x, D), no
matter what the A-type of <Pn(X), we can obtain a generating function (sum the
series)
for any other polynomial !fin(X) belonging to the same operator J(x, D).
9. Obtain a theorem analogous to that of Ex. 8 but with Shcffcr A-type re-
placed by <1-typc.
10. Show that if Pn(x) is 01( Legcndrc polynomial,
is a simple set of polynomials of Sheffer A-type zero.
11. Let Pn(X) be the Legendre polynomial. Choose u = DO and show that the
polynomials
<Pn(X) = :=
are of u-type zero for that u.
12. Prove that if the operator J(x, D) is such that a set of polynomials
Jfn(X) belongs to it in the Sheffer sense, then J(x, D) every polynomial
of degree precisely n into a polynomial of degrce precio;ely (n - 1).
CHAPTER 14
Pure
Recurrence
Relations
127. Sister Celine's technique. Years ago it seemed cus-
tomary upon entering the study of a new set of polynomials to seek
recurrence relations, pure or mixed (with or without derivatives in-
\'olved) by essentially a hit-and-miss process. Manipulative skill
was used and, if there was enough of it, some relations emerged;
others might easily haye been lurking around a corner without being
discovered. For polynomials of hypergeometric (pFq) character,
some systemization resulted from the existence of the contiguous
function relations published in 1945. The interesting problem of
the pure recurrence relation for hypergeometric polynomials re-
ceived probably its first systematic attack at the hands of Sister
Mary Celine Fasenmyer in a Michigan thesis in 1945. She in-
troduced the tool in her study of a certain class of hypergeometric
polynomials, for which see Section 149, page 290, or Fasenmyer [1].
In Fasenmyer [2J, Sister Celine illustrated her technique by ob-
taining pure recurrence relations for Bateman's Zn(l), touched upon
in this book in Section 146, page 285, and for one of her own sets of
polynomials (see Section 149, page 290).
As a first example we shall find the pure recurrence relation for
the simplest of Sister Celine's polynomials,
(1)
233
234
(2)
PURE RECURRENCE RELATIONS [Ch. 14
n (-l)k(n + k)!Xk
!,,(x) = t; (kl)2(!)k(n - k)(
For convenience we use 00 hereafter as the upper limit of summa-
tion for series such as that on the right in (2). We shall deal with
hypergeometric polynomials which always terminate naturally.
vVe also use a cornman convention that!n(x) is defined as zero \vhen-
ever the subscript is negati\'e. This convention permits us to avoid
discussing separately the smallest values of n, n = 0, 1, 2 for the
polynomials now to be treated.
First rewrite (2) as
ro
(3) !n(X) = L t(k, n).

Bister eeline's technique is to express In_leX), !n-2(X) , X!n_I(X), etc.,
as series involving t(k, n), and then to find a combination of co-
efficients which vanishes identically. Note that, by (2),
ro (-1)k(n-1+k)lx
k
In-leX) = t; - 1 - f)!'
so that
(4)
In the same manner it follows that
(5)
ro (n - k)(n - k - 1)
!n_2(X) = t; (n + k)(n + k _ 1) t(k, n),
(6)
en (n - k)(n - k - l)(n - k - 2)
!n_a(X) = t; (n + k)(n + k - l)(n + k _ 2) t(k, n).
Let us turn to X!n-I(X), X!n_2(X), etc. We find that
ro (-1)k(n-1+k)!xk+1 ro (-1)k-l(n-2+k)!x
k
xfn-I(X) = t; (kl)2(!)k(n-1-k)! = t; [(k-1)1]20)k_l(n-k)!'
or
_ - k2(! + k - 1) ,
(7) X!n-l(X) - f::o (n + k)(n + k _ 1) t(k, n).
In like manner,
(8)
ro _k2(t + k - l)(n - k)
X!n_2(X) = t; (n + k)(n + k _ l)(n + k _ 2) t(k, n).
127] SISTER CELINE'S TECHNIQUE 235
In equations (3)-(8) the coefficients of E(k, n) have a lowest
common denominator (n + k)(n + k - I)(n + k - 2). When that
denominator is used in each coefficient, the maximum degree with
respect to k of the numerators is four. Then there exist constants
(functions of n but not of k or x)A, B, C, D, E such that
(9) fn(x) + (A + Bx)fn-l(X) + (C + Dx)fn-2(X) + Efn-s(x) =
is an identity. By using equations (3)-(8) and reducing all co-
efficients of E(k, n) to the least common denominator, we see that
(9) is equivalent to the following identity in k:
(10) (n+k)(n+k -I)(n+k -2) + A(n-k) (n+k -1) (n+k -2)
- Bk2(k - !)(n + k - 2) + C(n - Ie)(n - Ie - 1) (n + Ie - 2)
- Dk2(k - !)(n - k) + E(n - k)(n - Ii; - I)(n - k - 2)=0.
The identity (10) easily yields five equations for the determination
of A, B, C, D, E. It is wisest to get those equations by a judicious
mixture of employing specific k values and equating coefTicients of
powers of k to zero. In (10), for instance, the choices k = n,
k = n - 1, coefficients of k4, k = 1 - n, k = 2 - n, yield simple
equations. By that or some other elementary method, it is found
that
A = __ 3n_-_2
n '
B = ~
n'
3n - 4
c=---
n '
D = ~
n'
n-2
E = -----.
n
With the aid of (9) we may conclude that the polynomials !n(X) of
(1) satisfy the relation
(11) nfn(x) - (3n - 2 - 4X)!n_l(X)
+ (3n - 4 + 4x)fn_2(X) - (n - 2)!n_"(X) = o.
As a second example, constder the polynomials
(12)
The <pn(X) interest us because they permit independent verification
in two special cases. If (3 = a, <Pn(X) becomes the simple Laguerre
polynomial Ln(x). If (3 = 0, <pn(X) becomes n!LnCa)(x)/(l + a)n.
For each of those polynomials we already know the pure recurrence
relation.
Now
236 PURE RECURRENCE RELATIONS [Ch.14
Put
(13)
Then
(14)
<Pn(X)
'Yn(X) =
n!
Proceeding as in the earlier example, we find that

(15) 'Y"-l(X) = L (n - k)t(k, n),
k_O
en
(16) 'Yn-2(X) = L (n - k)(n - k - l)t(k,n),
k_O
m
(17) 'Yn-3(X) = L (n - k)(n - k - l)(n - k - 2)t(l.:, n),
k-O
ro _ J,:2(a + k)(n - k)
(19) X'Yn-2(X) = L + k t(k, n).
(3
It follows that there exists a relation
(20) 'Yn(X) + (A + Bxhn_l(X) + (C+ Dxhn_2(X) + E'Yn_3(X) = 0
and that the A, B, C, D, E arc determined by the identity in k:
(21) .B+k+A(3+k)(n-k) - Bk
2
(a+k) +C(3+k)(n-k) (n-k-l)
- Dk
2
(a+k)(n-k) +E((3+k)(n-k)(n-k-1)(n-k-2) = O.
From (21) it is an elementary matter to show that
A = _-_3_n_2 ----'+_3n--::-;-----,-1_--;-a--'-(2_n_-_1-'--)
n2(a + n) ,
C = a + 3n - 3
n2(a + n) ,
Thus the polynomials 'Yn(X) satisfy the relation
E =
n
2
(a + n)
(22) n
2
(a+nhn(x) - [3n2 -3n+ 1 +a(2n-l) - (3+n)Xhn-l(X)
+(a+3n-3-xhn_2(X) -'Yn-3(X) =0.
But, by (13), 'Yn(X) = <Pn(X)/n!. Hence the polynomials
127] SISTER CELINE'S TECHNIQUE 237
(12) <Pn(X) = 2F
2
( -n, 1 + (3; 1, 1 + a; x)
satisfy the recurrence relation
(23) n(a+n) <pn(X) - [3n2 - 3n+ 1 +a(2n -1) - ({3+n)x ] <Pn-l(X)
+ (n-l)(a+3n-3 -X)<Pn_2(X) - (n -1)(n-2)<Pn_a(X) =0.
It is not difficult to rewrite (23) as
(24) (a+n) [n<Pn(X) - (2n -] -X)<Pn_l(X) + (n -1) <Pn-2(X)]
- (n-l) [(n-l)<Pn-l(X) - (2n -3 -X)<Pn_2(X) + (n-2)<Pn_3(X)]
+({3-a)X<Pn-l(X) = 0,
in which form it becomes eyident that if (3 = a, (24) is an iteration
of the known relation
nL,,(x) - (2n - 1 - x)Ln_l(x) + (n - 1)L
n
_
2
(x) = 0,
as it should be, since, when (3 = a, <Pn(x) reduces to Ln(x).
If the relation (23) is rewritten in the form
(25) n[(a+n)<p,,(x) - (2n-1 +a-x)<Pn_l(X) +(n-1)<Pn-2(X)]
- (n -1) [(a+n -1) <Pn-l(X) - (2n -3+a -x) <Pn-2(X) + (n - 2) <P,,-3(X) ]
+ (:iX<p,,-l(X) =0,
it is a simple matter to obtain a check for (3 = 0, for which value
<pn(X) degenerates into n!L(a)(x)/(l + a)".
In both the preceding examples, the polynomials were expressed
in ascending powers of x. That form has nothing to do \vith the
success of the method. As a third example, consider the poly-
nomials
(26)
or
(27)
Put
(28)
Then
(29)
[
-!n, -!n + !, 1. + a;
g,,(x) = (2x)n 3Fl
] + {3;
co (_l)k(1 + a)k(2x)n-2k co
An(X) = L -fl( 1-+--(3)--(----=-2-j.)' = L ~ ( k , n).
k=O \... k n ~ . 1,.=0
238 PURE RECURRENCE RELATIONS [Ch. 14
Form the series
00 (-1) k(1 + a) k(2x)n-2k en
(30) (2X)A
n
_I(X) = t; k!(1 + {3)kCn-1-2k)! = (n - 2k)f(k, n),
or
(31)
m (-1) k(1 + ah(2x)n-Z-Zk
An_Z(X) = 6 k!(l + (3h(n - 2 - 2k)!
= f + a)k_I(2x)n-z; , '
k_l (k - 1) .(1 + (3h-l(n - 2k
00 -k(fJ + k)
An_Z(X) = L: ( + k) E(k, n),
k=O a "
and, in the same manner,
00
(32) (2X)ZA
n
_Z(X) = L: (n - 2k)(n - 2k - 1)f(k, n),
k_O
(33)
00 _ ken - 2k)(fJ + k)
(2X)An _3(X) = (; a + k f(k, n).
It follows from the series (29)-(33) that there exists a relation of
the form
(34) An(X) + A (2x) An-I(X) + [B +C(2X)2] An-z(X) + D(2x)A
n
_3(X) = 0,
in which the constants A, B, C, D are determined by the identity
in k:
(35) a + k + A(n - 2k)(a + k) - Bk({3 + k)
+ C(n - 2k)(n - 2k - 1)(a + k) - Dk(n - 2k)(fJ + k) = O.
The identity (35) yields
2fJ + 2n - 1
A = n) ,
C = 1
n(2fJ + n)'
Hence the An(X) satisfy
B = 2(2a + n)
n(2{3 + n)'
D =
n(2{3 + n)
(36) n(2{3 + n)An(X) - (2{3 + 2n - 1)(2x)A
n
_I(X)
+ [2(2a + n) + (2X)2]A
n
_z(X) - 2(2x)A
n
_3(X) = O.
Since, by (28), An(X) = On(X)/n!, we find that the polynomials
127] SISTER CELINE'S TECHNIQUE
(26)
[
-!n, -!n + !, 1 + a;
Yn(X) = (2x)n 3FI
1 + f3;
satisfy the pure recurrence relation
(37) (2fJ+n)gn(X) - (2fJ+2n-1)(2x)gn_I(X)
239
+(n-l)[2(2a+n)+(2x)2]gn_2(X) -2(n-l)(n-2)(2x)gn_3(X) =0.
When fJ = a the gn(X) degenerates into the Hermite polynomial
Hn(x), for which we already know the relation
(38) IIn(x) - 2xH
n
_
l
(x) + 2(n - l)Hn_z(x) = O.
The relation (37) may be put in the form
(39) (2fJ + n)[gn(X) - 2Xg
n
_I(X) + 2(n - l)gn_z(X)]
- (n - 1)(2x)[gn_I(X) - 2xg
n
_z(x) + 2(n - 2)gn-a(X)]
+ 4(a - fJ)(n - l)gn-z(X) = O.
It is now evident that if fJ = a, (39) is an iteration of (38).
While we are on the topic of the gn(X) of (26), it may be of interest
to express gn(X) in terms of the Hermite polynomials of which
gn(X) is a generalization.
Using (27) we find that
ro gn(X)t
n
ro [n/Z] (-l)k(l + ah(2x)n-
2k
t
n
En! = E k!(l + fJ)k(n - 2k)!
= (i: (2X),n
t
n)(i:
n. n.(1 + fJ)n
so that the gn(X) possess the generating function in
ro g (x)t
n
(40) exp(2xt) IF
1
(1 + a; 1 + f3; -tz) = L _n __ ,_'
n.
By Kummer's first formula, page 125.
IF
1
(1 + a; 1 + fJ; _(2) = exp( _(2)IF
1
(fJ - a; 1 + fJ; (2).
Therefore (40) yields
(41)
Then
240 PURE RECURRENCE RELATIONS [Ch. 14
which yields
(42) )
[ ~ l n!(fJ - a)kH
n
_2k(X)
gn(X = f::o k!(n - 2J,;HTi+ fJ)k'
Shively [1J used Sister Celine's technique to obtain differential
recurrence relations, and Bedient [lJ employed her method to get
mixed recurrence relations. See also Exs. 8-11 at the end of this
chapter.
We already know from Chapter 9 that any orthogonal set of
polynomials necessarily satisfies a particular type of pure recurrence
relation, for which see Section 83. At once we see that the tech-
nique discussed here is a potent method for showing that a specific
polynomial set is not orthogonal with respect to any weight function
over any interval. Sec, for instance, Ex. 12 at the end of this
chapter.
128. A mild extension. Consider the polynomials CTn(X) de-
fined by
(1)
n (-1)kn!Hk(x)
ITn(X) = 6 (k!)2(n - k)!'
in which Hk(X) is the Hermite polynomial of Chapter 11. We shall
encounter the CTn(X) in our study of symbolic relations in the next
chapter. It is not difficult to show that in ascending powers of
x the CTn(X) has the following appearance:
n {-Hn-k), -!(n-k-1);
(2) CTn(X) = L 2F
k ~ O !(1+k), H2+k);
If the reader wishes to attack the problem of finding a pure recur-
rence relation for CTn(X) on the basis of (2) rather than of (1), he is
welcome to the task. Here we prefer to modify Sister Celine's
technique to fit the situation in which a polynomial is expressed
not in a series of powers of x but in a series of other polynomials.
In equation (1) put
(3)
so that
(4)
128] A MILD EXTENSION 241
From the known relation
(5) Hk(x) - 2xH
k
_
1
(X) + 2(k - 1)H
k
_
2
(x) = 0,
we determine a recurrence relation for Vk(X), use that to find a
relation for Sn(X), and finally use (3) to transform our result into a
relation for (Tn(X).
In (5) put Hk(X) = (-1)k(k!)2vk(x), The resulting relation for
Vk(X) is
(6) k2(k - l)vk(x) + 2x(k - l)Vk.ol(X) + 2vk_Z(:r) = O.
First we :set out to use (4) to obtain the t'crie:s
(7)
From (4) we get
(8)
(9) ( )
_ (n - I\,)(n - k - l)vk(x)
Sn_2 X - ( _ k) t '
n.
(10) ( )
_ (n - I\,)(n - k - 1)(n - k - 2)Vk(X).
Sn-3 X - ( _ k)'
k_U n.
The four series (4), (8), (9), and (10) contain numerator coefficients
of not more than third degree in k, the degree of the numerator in
(7). Therefore there exist constants A, B, C, D, such that
o 0 = k2(k - l)lh(X)
(11) ASn(x) + Bsn_l(x) + CSn _2(X) + Dsn _ 3 (x) = L --(- _ 1.-) ,-'
k-O n h.
Indeed (11) implies the identity in k:
(12) A + B(n - k) + C(n - k)(n - k - 1)
+ D(n - k)(n - k - 1)(n - k - 2) = k2(k - 1),
from which it follows readily that
A=n
2
(n-l),
Hence we have
B = - (n-l)(3n-2), C=3n-4, D=-l.
(13) n2(n - l)sn(x) - (n - 1)(3n - 2)Sn_I(X) + (3n - 4)Sn_2(X)
- Sn_3(X) = f .
k=O (n - k).
242 PURE RECURRENCE RELATIONS [Ch.14
N ext we wish to construct, corresponding to the second term in
(6), the series
(14)
f (k - 1)(2x)Vk_I(X)
k=O (n - k)! '
using the convention that V-leX) =:: O.
Now
(15)
Furthermore,
(16) Sn_2(X) = i: ____ ~ ( X ~ _
,.-0 (n - 2 - k)!
= f: _Vk_I(X_) _ = f {n - k)Vk_I(X) .
kdJ(n-1-k)! f." (n-k)!
Hence there exist constants E and F such that
( 17) 2 E ()
') F C) - ~ (Ie- - 1)(2:r)vk_l(x).
x -,S,,_1 X + X Sn-2 X - L.. ( j.)!
k ~ O n - v
From the identity
E + FCn - k) = k - 1,
we obtain E = n - 1, F = - 1, so tha t
~ (Ie - l)C2X)Vk-I(X)
(18) (n - 1)(2x)sn_l(x) - (2X)Sn_2(X) = L.. (_ k)1 .
k=O n .
Finally,
(19)
We now add corresponding members of (13), (18), and (19) to get
n2(n - 1)sn(x) - (n -1) (3n - 2)Sn_I(X) + (3n -4)Sn_2(X) -Sn_3(X)
+ (n - 1)(2x)sn_I(X) - (2X)Sn_2(X) + 2s
n
_ 2(x)
= :t k2(k - l)Vk(X) + (k - 1)(2X)Vk_I(X) + 2Vk_2(X) = 0
k=O (n-k)! '
because of equation (6). Therefore Sn(X) satisfies the equation
(20) n2(n - l)sn(x) - (n - 1)(3n - 2 - 2X)Sn_l(X)
+ (3n - 2 - 2X)Sn_2(X) - Sn_3(X) = O.
128] A MILD EXTENSION
Since, by (3), Sn(X) = ~ n ( ~ ) , we find that the polynomials
n.
(1)
satisfy the pure recurrence relation
(21) nCTn(x) - (3n - 2 - 2:r)(Tn_l(X)
243
+ (3n - 2 - 2X)(T"_2(X) - (n - 2)CT
n
_3(X) = O.
EXERCISES
1. Show that Bateman's polynomial (sec Section 146, page 285)
Zn(t) = 2fi'2(-n, n + 1; 1, 1; t)
has the recurrence relation
nZ(2n - 3)Zn(t) - (2n - 1) [3n
Z
- 6n + 2 - 2(2n - 3)tJZn-l(t)
+ (2n - 3)[3n2 - (in + 2 + 2(2n - 1)tJZn-2(t) - (2n - 1) (n - 2)ZZn_3(t) = O.
2. Show that Sister Celine's polynomial fn(a; -; x), or
fn(x) = 3F2( -n, n + 1, a; 1, ~ ; x)
has the recurrence relation
nfn(X) - [3n - 2 - 4(n - 1 + a)x Jfn-I(X)
+ [3n - 4 - 4(n - 1 - a)xJfn-z(x) - (n - 2)fn-3(X) = O.
3. Show that Rice's polynomial (sec Section 147, page 287)
Hn = Hn(r, p, v) = 3F2( -n, n + 1, r; 1, p; v)
satisfies the relation
n(2n - 3)(p + n - 1)Hn
- (2n-1)[(n-2)(p-n+ 1) +2(n-l)(2n-3) -2(2n-3)(r+n-1)v JHn- 1
+(2n-3)[2(n-l)2- n(p-n+ 1) +2(2n-1)(r-n+ 1)v]Hn_z
+(n-2)(2n-1)(p-n+l)H
n
_
a
= O.
4. Show that the polynomial
fn(x) = IF2( -n; 1 + a, 1 + (3; x),
which is intimately related to Bateman's Jnu . of Section 146, page 287, satisfies
the relation
(a + n)({3 + n)fn(x) - [3n2 - 3n + 1 + (2n - 1)(a + {3) + a{3 - X]fn-l(X)
+ (n - 1)(3n - 3 + a + (3)fn-2(X) - (n - l)(n - 2)fn-3(X) = o.
5. Define the polynomial wn(x) by
244 PURE RECURRENCE RELATIOJVS [Ch.14
in terms of the simple Laguerre polynomial Lk(X), Show that w,.(x) possesses
the pure recurrence relation
n2w,.(x) - [en - 1)(4n - 3) + X]W,._l(X) + (6n
2
- 19n + 16 +
- (n -- 2)(4n - 9)U,._3(X) + (n - 2)(n - 3)W,._4(X) = o.
6. Show that the polynomial w,.(x) of Ex. 5 may be written
7. Define the polynomial (',,(x) by (see Section 131, page 251)
n (_ J) k
n
! P ,J.r)
1',,(X) = I.)',
k_o . n n.
in terms of the Legendre polynomial Pk(X). Show that (,,,(x) satisfies the recur-
rence relation
n
2
v
n
(X) - [4n2 - 5n + 2 - (2n - l)xJv
n
_l(x)
+[6n
2
- 15n + 11 - (In - 5)XJV
n
_2(X) - (n - 2)(4n - 7 - 2X)Vn _3(X)
+ (n - 2)(n - ;)vn_,(x) = O.
8. Show that the vn(x) of Ex, 7 satisfies the relations
(1 - x
2
)vn"(x) - 2xv
n
'(x) + n(n + l)v,.(x) = 2n
2
v,._1(X) - n(n - 1)v
n
_2(x)
and
(1 - x
2
)V,.'(x) + nxv,.(x)
= [(2n - l)x - l]v,._J(x) - (n - l)xv,._2(x) + (1 -
9, Let
'Y(k n) =
, k ! (n - 2k)! '
so that the Legendre polynomial of Chapter 10 may be written
ro
P,.(x) = L 'Y(k, n).
Show that
P () = (n - 2kh(k, n)
x ,,-J x f::o 2n _ 2k - 1 '
ro
xP,,'(x) = L (n - 2kh(k, n),
k_O
ro
= L -2k'Y(k, n), etc.

k-O
ro -2k'Y(k, n)
P,,_2(X) = t;:o 2n - 2k - l'
ro
= L (l + 2n - 2kh(k, n),
k_O
Use Sister Celine's method to discover the various differential recurrence relations
and the pure recurrence relation for p"(x),
10. Apply Sister Celine's method to discover relations satisfied by the Hermite
polynomials of Chapter 11.
1 L Find the various relations of Section 114 on Laguerre polynomials by using
Sister Celine's technique.
128] A MILD EXTENSION 245
12. Consider the pseudo-Laguerre polynomials (Boas and Buck [2; 16]) inC;r;)
defined for nonintegral A by
( - A)n n (- A)n_kXk
In(X) = -,-IF1( -n; 1 + A -n; x) = L -k'(-=--k-)"
n. k--O n .
Show that the polynomials In(X) are not orthogonal with respect to any weight
function over any interval because no relation of the form
is possible. Obtain the pure reeurrence rehitioll
n/n(X) = (x + n - 1 - - x/n-b).
For the polynomials in each of Exs. 13-17, use Sistpr Celine's technique to discover
the pure reeurrence relation whatever mixed rehtions exist.
13. The Bessel polynomials of Section 150.
14. Bedient's polynomiab fln of Section 151.
15. Bedient's polynomiab Un of Section 1.51.
16. Shively's polynomials Rn of Section 152.
17. Shively's polynomials Un of Section 152.
CHAPTER 15
Symbolic
Relations
129. Notation. Many relations involving finite series of poly-
nomials can be put into particularly neat form by use of an old
symbolic notation. Whenever ~ is used to replace =, it is to be
understood that exponents will be lowered to subscripts on any
symbol which is undefined here except with subscripts. For in-
stance, we already know, page 213, that the simple Laguerre poly-
nomial Ln(x) satisfies the relation
(1)
xn = f ( -l)kn !Lk (x) .
n! k ~ O k!(n - h)!
In symbolic notation, since L without a subscript is not defined here,
we may write equation (1) as
(2)
xn
-, ~ {I - L(x)ln.
n.
In Section 126 we obtained a recurrence relation for a polynomial
<Tn(X) defined by
(3)
In symbolic notation (3) is replaced by
(4)
A final example: if Hn(x) is the Hermite polynomial and Pn(x)
246
lS0] CLASSICAL POLYNOMIALS RELATIONS 247
the Legendre polynomial, then the meaning of IH(x) - P(x)Jz is
given by
IH(x) - P(X)}2 H
2
(x)P
O
(x) - 2H
1
(x)P
1
(x) + HO(X)P2(X).
130. Symbolic relations among classical polynomials. In addi-
tion to (2) of Section 129, we have already encountered earlier in
this book a few relations which fit nicely into symbolic form. For
instance,
(1)
(2)
(3)
Ln(xy) {I - y + yL(x) I ",
Pn(X) {2x - P(x) In,
P n(1 - 2X2) 11 - 2xP(x) I ",
p = (1 - 2xt + t2)i.
For the most part the relations with which we deal here are in-
cluded because they are amusing or particularly pretty. It would
be unwise, however, to pass up the subject as one of no other value.
Szego [2J made good use of equation (4). The symbolic notation
also suggests the study of some interesting polynomials which may
not otherwise be noticed.
We shall now prove
(5) H n(P(X)) I H(2x) - 2P(x) I n,
(6) 2
n
L
n
(p(x) IL(x - 1) + L(x + 1) I n,
(7) Hn(H(x)
as examples and leave other relations as exercises for the reader.
To prove (5), consider the series
ro Hn(p(x))t
n
. ro [n/21 (-1)k2n-2kPn_2k(x)tn
?; n! =;= [;6 k!(n - 2k)!
= exp(4xt - t2) exp( -2xt) (PI ( -; 1; f2(x
2
- 1).
248 SYMBOLIC RELATIONS [Ch.15
Then
which is (f.
To pro\"e (fi), consider the series
= e
2t
0/<\( -; 1; -lex - 1) uFI( -; 1; -lex + 1
= etoFI( - ; 1; -lex - 1e
t
oFl( -; 1; -lex + 1
= (f Ln(X 1)tn)
n. 7l.
from which (6) follows by equating coefficients of tn.
Since
[n/2J ( 1) k '2n-2kH ()
H (H( == '""' - n. _ . X
n X L... /., ( _ ') /.) , '
h. n ",h.
we find that
(f Hn(X),(2t)n)
n=O n. II. n.
= exp( - [2) exp( 4xl - 4t2)
= exp( 4xl - 5t2)
131] POLYNOMIALS OF FORM Ln(Y(x))
= ex
p
[ - (t'\l'S)2J
_ f H n(2x/ yS)5
tn
t
n
- n-O n!
which yields equation (7).
249
131. Polynomials of symbolic form Ln(Y(x)). Suppose Yn(X) is
a previously studied simple set of polynomials. We proceed to
start the study of the polynomials Ln(Y(x.
THEOREM 81. A necessary and sufficient crmdtion that
fn(x) =l= Ln(Y(x
be of Sheffer A-type zero is that Yn(X)/n! be of Sheffer A-type zero.
Proof: Consider the series
ffn(x)t n = f t (-
n_O n=O k=O (k.) (n - k).
= f (-l)k(n + k) !Yk(X)t
nH
n.k=O (k!)2n!
By Theorem 72, page 222, if Yn(X)/n! is of Sheffer A-type zero
Ykrr
W
= A(t) exp[xH(t)].
Then also
'tofn(x)t
n
= (1 - f)-lAC __ t ) exp[ xHC __ tt ) ]
= B(t) exp[xh(t)],
so that fn(x) is of Sheffer A-type zero also. Since
250 SYMBOLIC RELATIONS [Ch. 15
the same method of proof works in the other direction.
Because of Theorem 79, page 228, it can be seen that the same
proof \vorks if Sheffer A-type zero is replaced by CT-type zero for any
specified CT.
THEOREM 82. Theorem 81 holds true 1f Sheffer A-type zero is
everywhere replaced by CT-type zero.
Theorems 73-76 and Theorem 80 are then applicable to poly-
nomials of the form Ln(Y(x) for properly selected Yn(X).
Let us consider the polynomials
(1)
for which we ohtained a pure reeurrence relation in Section 128,
"43 Co' l/,,(.r). f <.:'h ff \ . f II h
page L. '. - ,- IS 0 .::1 e er 1 -type zero, It 0 ows t at
77
CTn(X) is of Sheffer A-type zero. Indeed, because
00 H (x)tn
L _n_,_ = exp(2xt - t2),
n_O n.
\ve get
(2)
i: CTn(X)t
n
= (I - t)-I - __ t
2
_J.
1 - t (1 - t)2
In order to apply the theorems of 124, we note that, for
the polynomials CTn(X),
[
_t2 J -2t -t
(3) A(t) = (1 - t)-I exp (1 _ t)2' H(t) = J(t) = 2-=t'
It is easy to show that
(4)
(5)
(6)
(7)
DO [-t
k
, -!k + L
ACt) = L 2F2
h() 1 A'
, 2",
ro
H(t) L (_2)tk+!,

co -1
J(t) = L __ -t
HI
k_
o
2
k
+I '
A/(t) = i: (1 - k - k2)tk
ACt) k-O '
131] POLYNOMIALS OF FORM Ln(Y(x 251
ro
(8) H'(t) = I: [ -2(k + l)]tk,
k=O
(9)
uA'(u) = _It _ lt2 + lt3 ' u = J(t)
A(u) 2 2 4, ,
(10) uH'(u) = t - W; u = J(t).
It follows that, in the notation of Theorems 73-76,
(11) CXk = 1 - k - 1..
2
, fk = -'2(k + 1),
(12) Ilo = -!, 111 = -L Ji.2 = 1, Ilk = 0 for k s; 3,
110 = 1,111 = Ilk = 0 for k s; 2,
(13) hk = -2.
We may apply Theorems n-76 to obtain properties of O",,(x).
For instance, Theorem 76 yields
,,-I
(14) O"n'(X) = -2 I: o-n_I_k(X), n l.
k=O
Theorem 74 yields the differential equation
O"/"(x) - 2(1 + x)o-n"(x) + 2(2x - l)o-n'(x) - 4no- n (x) = O.
N ext let us consider the polynomials
(15)
for which a pure recurrence relation was ohtained in Ex. 7, page
244. \Ye already know one other property, relation (G), page 247,
for these polynomials. Two generating functions for vn(x) arc
easily found. Since
i: vn(x)t
n
= i: t
n=O n=O k=O (Iv.) (n - k).
= i: (-l)k(n + !,')lPk(x)t
nH
n.k=O (kl)2nl
we obtain
252 SYMBOLIC RELATIONS [Ch.15
(
-Xl) [-; l2(X
2
-1)] =
(16) (1 - t)-1 exp 1 _ t oFI 4(1 _ t)2 = vn(x)t
n

1 ;
Next
and we may conclude that
(17) e1oF1C -; 1; - 1 oFIC -; 1; -!t(x + 1 = f
1._0 n.
It is also not difficult to derive such results as
(18) (1 - X
2
)V
n
"(X) - 2xvn'(x) + n(n + l)Vnex)
= 2n
2
v
n
_I(X) -n(n - 1)v
n
_2(X),
(21)
EXERCISES
In Exs. 1-8, Hn(x), Pn(x), Ln(x) denote the Hermite, Legendre, and simple
Laguerre polynomials, respectively. Derive each of the stated symbolic relations.
1. Hn(x + y) [H(x) + 2y]n.
2. [H(x) + H(y)]n 2!nlln(2-!(x + y).
3. Pn(X) n{L(I; x) - J
4. Hn(!P(x) [ll(x) - p(x)Jn.
5. [H(x) - 2P(x)]2n+1 0,
[H(x) - 2P(x) J2n (-I)n22nWnLn(x2 - 1).
6. [H(x) - P(2x) ]2n+I 0,
[H(x) - P(2x)]2n (_l)n22nWnLn(X2 - 1).
131] POLYNOMIALS OF FORM L,.(y(x)) 253
7. H,,(!H(x ~ 2
i
"H,,(2-
i
x).
8. H,,(xy) ~ [H(x) + 2x(y - l)}.
9. Use Laplace's first integral for P,,(x) to derive equation (20), preceding
these exercises.
10. For the vn(x) of equation (15), page 251, evaluate
.:Vm(X)Pk(X) dx,
and use your result to establish equation (21), preceding theKe exercise!>.
11. Show that
12. Define polynomials 'Pn(X, y) by
'Pn(x, y) ~ IIn(xL(y.
Show that
00 (x y)t"
L 'Pn " = exp(2xt - t
2
) OFl( -; 1; - 2xyt).
n=O n"
CHAPTER 16
Jacobi
Polynomials
132. The Jacobi polynomials. The Jacobi polynomial
P n(a{J)(x) may be defined by
[
- n, 1 + a + (3 + n;
(I) p,,(a.{J)(x) = (I +, a)n zFl
n. 1 + a;
When a = fJ = 0, the polynomial in (I) becomes the Legendre
polynomial. From (I) it follows that Pn(a{J)(x) is a polynomial of
degree precisely n and that
Pn(a.{J)(I) = (I +, a)".
n.
In dealing with the Jacobi polynomials, it is natural to make
much use of our knowledge of the 2Fl of Chapter 4. An application
of Theorem 20, page 60, to (I) yields
(2) p.""(x) ~ (l ~ ! a).(x t l)",F. [ -n, -: : :: ~ ~ : 1
In the next section we shall obtain still another 2Fl form for P n (a.Il)(x),
namely,
(-I)n(1 + (3)n [-n,1 + a + (3 + n;
(3) p .. (a{J)(x) = , 2Fl
n. 1 + {3;
254
132] JACOBI POLYNOMIALS
255
Each of (1)-(3) yields a finite series form for P n(a,fJl(x):
Equations (4), (5), (0) are expanded forms of (1), (2), (3), re-
spectively. We must not use (3) or (6) until we derive them in the
next section.
By reversing the order of summation in (4), (5), (6), respectively,
we obtain
(
7) P (a.
fJ
)( ) = (I+a+i1)2n (X-::-l)n F[-n, -a-n;
n X n!(1+a+/1)n 2 2 I
-a- i1-2n;
x + ~ ]
x-I'
of which (9) is not yet available for use.
A generating function for the Jacobi polynomials follows readily
from (4). Consider the series
= ~ (1 + a + i1)n+2k(X - 1) ktnH
L..J J. I I (1 + ) ') k '
n , k = 0 l n . a k .....
256 JACOBI POLYNOMIALS [Ch.16
if; = i: i: (1 + a + f3 + 2k)n
tn
(1 + a + f3)2k(x - l)ktk
k=O n=O n! k!(l + O')k2k
ro (1 + a + f3b(x - l)ktk
= 6 2kk! (1 + a) k(l - t)1+aH+2k'
Since (ab = + Dk, it follows that
[
HI + a + f3), H2 + a + f3);
(10) (1 - t)-l-a-
fJ
zFl
1 + a;
- i: __
- n=U (1 + O')n '
a generating relation for the Jacobi polynomials.
133. Bateman's generating function. Equation (5) of the pre-
ceding section, which came from (2) of the same section, leads us
to what seems the most beautiful of the known generating functions
for the Jacobi polynomials. From (5) we obtain
We have thus derived Bateman's generating function:
(1) F[ -; t(X-I)] F[ -; t(X_-t12]_ f Pn(a,fJ)(x)t
n
.
o 1 1+0'; 2 0 1 1+f3; 2 - n=O (l+O')n(l+{3)n
Equation (1) was first published by Bateman [1]. Bateman's
methods and the form in which he put his result bear no resemblance
to what was done here.
If in (1) we replace x by (-x) and t by (-t), the two of!'s change
roles. Hence from (1) \ve obtain
(2)
Pn(a,fJ)( -x) = (-1) np nCfJ,a)(x).
Applying (2) to equation (1) of Section 132 yields equation (3)
of Section 132. Hence equations (3), (6), and (9) of that section
are now available for use. Equation (2) also leads to
134] THE RODRIGUES FORMULA
P
n
(a,tl)(-l) = (-l)n(1, + fJ)n.
n.
257
Two other generating functions, one of them containing an ar-
bitrary parameter, will be derived in Sections 140 and 141.
134. The Rodrigues formula. Equation (5), page 255, can be
written
(1)
If D = djdx, then for non-negative integral.') and rn,
D'x
m
+
a
= (rn + a)(rn + a-I) .. (rn + a - .') + l)x
m
-.+
a
or
(2)
From (2) we obtain
(3) Dk(X + l)nH = (1 + (3)n(X + l)n-k+fJ
(1 + (3)n-k
and
(4)
Therefore (1) can be put in the form
(5)
P (a,tl)(x) = (x - l)-a(x + 1)-tl
n 2nn!
In view of Leibnitz' rule for the derivative of a product, equation
(5) yields the Rodrigues formula
(x l)-a(x + l)-fJ
(6) Pn(a,tl)(x) = - 2nn! Dn[(x - l)n+a(x + l)n+tll,
or
(7) Pn(a,{J)(X) = - x)n+a(l+x)n+tl].
Equation (7) is more desirable than (6) \"hen we work in the in-
terval - 1 < x < 1.
258 JACOBI POLYNOMIALS [Ch.16
135. Orthogonality. By using (1), Section 132, and the dif-
ferential equation for 2Fl (a, b; c; y), ,ve obtain at once a differential
equation satisfied by the Jacobi polynomial:
(1) (I-x2)D2 P
n
(a,M(x)+[,B-a-(2+a+iJ)x]D
+ n(I + a + iJ + = 0.
Since ,B-a-(2+a+iJ)x=(1+,B)(I-x)-(I+a)(1+x), we may put
(1) in the form
(I-x)l+ <>(1 +X)l+ D2p n (a,
+ [(1+ ,B)(I-x) - (1 +a)(1 +x)](I-x)a(l +x) (a,/J)(x)
+n(I +a+ ,B+n)(l-x)a(I +x) n = 0,
which yields
(2) D[(I - X)I+a(I +
+ n(l + a + ,B + n)(1 - x)a(1 + x) n(a,/3)(x) = 0.
From (2) and the same equation with n replaced by m, it follows
(see page 173 for the corresponding work on Legendre polynomials)
that
[n(1 +a+ ,B+n) -m(I +a+ iJ+m)J(I-x) a(I +x) n m
=D [(I-x)I+a(1 +X)lHI Pn(a.fJ)(x)DP - P m(a,/J) (x)DPn(a.fJ) (x) I],
Therefore we may conclude that
(3) (n - m)(1 + a + (3 + n + m)
fb(1 - x)a(I + x) fJPn(a,fJ)(x) Pm(Ct,{J) (x) dx
a
= [(1 - X)I+a(1 + x)lHI Pn(a.fJ)(x) D P md) (x)
- Pm(a.fJ)(x)D Pn(a.tJ)(x) II
In particular, if Re(a) > -1 and Re(iJ) > -1, then (3) leads us
to the orthogonality property
(4) i> -x)a(I + dx = 0, m ;6. n.
That is, if Re(a) > -1 and Re({3) > -1, the Jacobi polynomials
form an orthogonal set over (-1, 1) ,vith respect to the weight
function (1 - x) a (1 + x) 13. For real a and ,B, the Jacobi poly-
135] ORTHOGONALITY 259
nomials are real, so we can then apply the conclusions of Chapter 9
on real orthogonal sets of polynomials.
In order to evaluate
(5)
we shall employ the Rodrigues formula and integration by parts.
This method incidentally furnishes a second derivation of the
orthogonality property (4).
From (7), page 257, we obtain
(6) (1 - x)"'(l + = - x)n+a(l + x)nHJ.
Therefore, if Re(ll') > -1 and Re(/3) > -1,
(7) f:
l
(l - x)"(l + dx
= IDn[(l - x)n+a(l + x)n+f3JlPm(a,{3)(x) dx.
On the right in (7), integrate by parts n times, each time differen-
tiating P
m
(a,{3)(x) and integrating the quantity in curly brackets.
At the kth stage the integrated part,
{Dn-k[(l - x)n+a(1 + x)nHJIDk-lP
m
ia,{3)(x)
is zero at both limits because of factors (1 - x)k+a(1 + with
Re(ll') > -1 and Re(/3) > -1. After n such integrations by
parts, we have
If m n we may choose n to be the larger (or interchange m and n
in the preceding discussion) and therefore conclude that
(9)
II (1 - x)a(l + x) (3P
n
(a,{3)(X)Pm(a,{3)(x) dx = 0,
-1
m r! n,
because P on the right in (8) is a polynomial of degree m.
260 JACOBI POLYNOMIALS [Ch.16
In (8) we have a tool for the evaluation of the gn of (5), but we
need Dnp n(a,p>(x). From
(1 +) [ - n, 1 + a + /3 + n;
Pn(a,p>(x) = ,a n 2FI
n. 1 + a;
~ J
2
and repeated application of the formula for the derivative of a
2FI (Ex. 1, page 69), we obtain
[
0, 1 + a + /3 + 2n;
~ l
1 + a + n;
~ J
2 '
from which
(10)
Now (8) with m = n yields
gn = 1
1
1(1 - x)a(I + x)P[P
n
(a,P)(x)J2 dx
= 2 (1, + a + /3)2n II (1 - x)n+a(l + x)n+8 dx.
2 nn.(l + a + (3)n -1
But, by Ex. 7, page 31,
I: (1 - x)n+a(I + x)nH dx = 2
2n
+
a
+8+
I
B(I + a + n, 1 + /3 + n)
2
2n
+
a
+8+! r(I + a + n) r(I + (3 + n)
= - - - - - - - = ~ - - - - ~ ~ ~ ~ - - ~
r(2 + a + (3 + 2n)
Hence
(1 + a + {3)
2n
2I+a+8r(1 + a + n) r(I + /3 + n)
Un = n!(l + a + (3)nr(2 + a + {3 + 2n) ,
or
(11)
2I+a+8r(I + a + n) r(I + /3 + n)
Un = n!(1 + a + {3 + 2n) r(I + a + (3 + n)
From (10) we conclude that
136] DIFFERENTIAL RECURRENCE RELATIONS 261
(
12) P (a,fil( ) _ (1 + (3)2n
xn
+ ( )
.. x - 2nn!(1 + a + (3)n 1l"n_1 X ,
in which 1l"n_I(X) is a polynomial of degree (n - 1). In the notation
of Theorem 58, page 154, we have
(1 + a + (3)2n
h,. = 2nnfCi + a + (3) .. '
Then
(13) = (n + I)! __ a + (3 + n)
Ynhn+l 2
a
+fJ(2 + a + (3 + 2n) r(I + a + n) r(I + (3 + n)
With (11) and (13) at hand, the Christoffel-Darboux formula of
Theorem 58, page 154, may now bc written explicitly, if desired.
By combining Theorem 54, page 148, with equations (11) and
(12), we conclude that
(14) i
l
l(l-x)a(I+X)I>X
k
P
n
(a,l>l(X) dx=O, k=O,I,2, ,(n-I),
and
(15) (1 - x)a(1 + x)I>XnPn(a,l>l(x) dx
2
1
+
a
+fJ+
n
1'(1 + a + n) r(1 + (3 + n)
r(2 + a + (3 + 2n)
For a > -1, (3 > -1, the zeros of Pn(a,f)l(x) are distinct and lie
on the open interval -1 < x < 1. We may also apply Theorem
59, page 156, if we wish.
136. Differential recurrence relations. In the generating re-
lation (10), page 256, put x = 1 - 2v to get
(1)
{
HI + a + (3), !(2 + a + (3);
(1 - t)-l-a-I> 2F
1 + a;
= f: (1 + a + (3)"p,,(a,f)(l - 2v)t".
(1 + a)n
N ow we see that Theorem 48, page 137, with
c = 1 + a + {3,
262 JACOBI POLYNOMIALS [Ch.16
applies to the polynomial
f (v) - (1 + a + (3)n
P

,,- (1 + a)"
The five results in Theorem 48, when put in terms of x rather than
v, yield our definition, page 254, of Pn(a,/!)(x) and the four other
properties
(2) (1 - x)n =
2n(1 + a) + 21;)P_+_<: + (3hPk (a,/!)(x)
" (1 + a + (3)n+l+k(l + a) k '
(3) (x - 1)[(a + (3 + n) D p,,(a,{J)(x) + (a + n) D P
n
_
1
(a,!1J(x)]
d
= (a + {3 + n)[nPn(a,/!J(x) - (0: + n)Pn_l(a,{J)(x)], D = dx'
(4) (x - 1) D P
n
(a,8J(x) - nPn(a,/!)(x)
= _ 'i:(l + a + (3)k.
(1 + a + (3),. (1 + a)k
[(1 + a + (3)P
k
(a,/!J(x) + 2(x - 1) D Pk(a,/!)(x)],
(
5) (x - 1) D P (a,/!J(x) - nP (a,/!J(x) = (1 + a)n
" n (1 + a + (3)n
:i (-l)n-k(l + a + f3 -I- 21,)(1 + a + (3)kP
k
(a,/!)(x).
bO (1 + a)k
The relation (3) is an ordinary differential recurrence relation,
with derivatives and shift of index but no shift in the
parameters a and (3. \Ye seek two such relations to furnish the
groundwork for others and for the pure recurrence relation. We
know
Pn(/!,a)( -x) = (-1) np,,(a,/!)(x).
Therefore in (3) we interchange a and {3 and replace x by (-x) to
obtain
(6) (x + l)[(a + {3 + n) D Pn(a,/!J(x) - ({3 + n) D p"_l(a,/!)(x)]
= (a '1- {3 + n)[nPn(a,/!)(x) + ({3 + n)Pn_1(a,/!J(x)].
Let us eliminate D Pn_1(a,/!J(x) from (3) and (6). The result is
138] MIXED RELATIONS 263
(7) (a+I1+2n)(x
2
-1) D Pn(a,fJ)(x)
=n[iJ-a+(a+ iJ+2n)x]P n(a,fJ)(x) -2(a+n)(iJ+n) P n_l(a,fJ)(x).
137. The pure recurrence relation. If we use (7) above twice in
(6), we arrive at a relation involving only shifts in indices, the pure
recurrence relation
(1) 2n(a+ iJ+n)(a+ iJ+2n-2)P n(a,fJ)(x)
= (a+ iJ+2n-l)[a
2
- iJ2+x(a+ f3+2n)(a+ f3+2n-2)]P n_l(a,fJ)(x)
-2(a+n-l) (iJ+n-l)(a+ iJ+2n)P n_2(a,fJ)(x).
138. Mixed relations. From the definition
(1 + a)" ,[-n,l + a + iJ + n;
(1) p,,(a,fJ)(x) = --T-- 21/1
n. 1 + a;
it follows, with D = d/dx as usual, that
D P (a,fJ)( ) = n(1 + a),,(l + a + iJ + n)
n X 2(1 + a)n!
[
-n + 1,2 + a + iJ + n;
zFl
2 + a;
(2 + a)n_l(l + a + iJ + n)
2(n - I)!
[
- (n - 1), 1 + (a + 1) + (iJ + 1) + n - 1;
zFl
1+(a+1);
so that
(2) D Pn(a,fJ)(x) = HI + a + iJ + n)Pn_l(a+I.fJ+1)(x).
Iteration of (2) yields, for 0 < k ~ n,
~ J
2 '
(3) Dk Pn(a,fJ)(x) = 2-
k
(1 + a + iJ + n)kPn_k(aH,fJH)(x).
In Section 132 we found that
264 JACOBI POLYNOMIALS [Ch.16
Differentiation of both members of (4) yields
D Pn(a,P)(x) = n(x + l)-lPn(a.P)(x)
2Fl[-n+I, -/3-n+1;
(n-J)!(x+I) 2 l+(a+I);
from which it follows that
(5) (x + 1) D Pn(a,P)(x) = nPn(a,(J)(x) + (/3 + n)Pn_1(a+l,p)(x).
In the same manner the known relation [(8) of Section 132]
(6)
(1 + (3),,(x _ 1)" ,[-n, -a - n;
p,,(a,P)(x) = --,---- -q- 2/11
n. . 1 + /3;
leads us to
(7) (x - 1) D Pn(a,P)(x) = nPn(a,p)(x) - (a + n)Pn_1(a,Hl)(x).
From (5) and (7) it follows that
(8) 2D Pnz,fJ) (x) = (/3 + n)P
n
_
1
(a+!,f3)(x) + (a+ n)Pn_1(a,HO(x).
If we use (2) in (8) and then shift n to (n + 1), a to (a - 1) and /3
to (/3 - 1), we obtain
(9) (a+ /3+n)P n (a, P)(x) = (/3+n)P n (a, p-o (x) + (a+n)P n (a-I. (J)(x).
We next seek to express each contiguous (a or (3 increased or
decreased by unity) Jacobi polynomial in terms of the original
polynomial.
In Ex. 22, page 72, we found that
(10) F(a,b;c;z)
= F(a - 1, b + 1; c; z) + c-1(b + 1 - a)zF(a, b + 1; c + 1; z).
In (10) put a = -n, b = 1 + a + (3 + n, c = 1 + a, Z = HI - x),
and simplify the result to obtain
(11) !{2 + a + (3 + 2n)(x - I)Pn(a+l,fJ)(x)
= (n + 1)P
n
+
1
(a,fJ)(x) - (1 + a + n)Pn(a,fJ)(x).
In Ex. 23, page 72, we found that
(12) (c - 1 - b)F(a, b; c; z)
= (c-a)F(a-1, b+ 1; c; z) +(a-l-b)(1-z)F(a, b+ 1; c; z).
139j APPELL'S FUNCTIONS 265
In (12) put a = -n, b = 1 + 0: + f3 + n, C = 1 + 0:, z = HI - x)
to obtain
(13) H2+ 0: + f3 + 2n)(x + I)Pn(a,HI)(x)
= (n + l)P
n
+
l
(a,/l)(x) + (1 + (3 + n)Pn(a,fJ)(x).
The contiguous function relation
(a - b)F(a, b; c; z) = aF(a + 1, b; c; z) - bF(a, b + 1; c; z)
leads in the same way, after a shift from (3 to (/3- 1) to
(14) (a + (3 +2n)P n c d ~ ! ) ( X )
= (a + f3 + n)l'n(a,/lJ(x) + (0: + n)Pn_1(a,/Jl(x).
From (9) and (14) it follows that
(15) (0: + f3 + 2n)P
n
(a-l,IJ)(x)
= (0: + f3 + n)Pn(a,fJ)(x) - (f3 + n)P"_I(a,!JJ(x).
Finally, (11) and (13) combine to yield
(16) (1 + x)Pn(a,HI)(x) + (1 - x)F\(a+l.fJ)(x) = U\(a,/l)(x)
and (14) and (15) combine to yield
(17) Pn(a,/l-l)(x) - Pn(a-l,/l)(x) = Pn_1(a,/ll(x).
139. Appell's functions of two variables. One of various ways in
which the 2Fl function has been generalized is to functions of more
than one variable. Of these a fruitful set is that of Appell's four
functions*
*See Appell and Kampe de Fcriet [1],
266 JACOBI POLYNOMIALS [Ch.16
The four functions above have many interesting properties, for
which see the reference in the footnote or Bailey [lJ or Erdelyi [1].
Here we need only two theorems on F
4
, and our discussion is limited
to what we need.
Consider
(1) if; = (l-x)-a(l-y)-b
F
4
( a, b; c, c'; o=-xf(1-=:y-)' [1 _y)).
From the definition of F4 we obtain
00 (a)n+k(b),,+k( -1)" t-kXI'Y"
if; = k!nT(C)k(c'),,(l - x) tI+k+a(1 _ y);;+k+b'
Now
and
(1 _ y)-b-n-k = f: (b + n k).ys.
__ 0 s.
Hence
00
if;= L
n,k,s,i=.O
(
-l)n+k(a) k (b) ,. xk+iyn+.
n+ +1., n+,,+s
I 'fI1 I() ( ')
S.L/v.n. eke n
so that
00 k (-l)n+k-i(a) (b) .xkyn+8
if; = L L 'I(k l(n)+k+8(-:,) .
L - J, .s.n. C k-, ,c n
Let us reverse the order of the inner summation and write
ro [-k, b + n + s;
L 2Fl
n,k,s_O c'
,
Then
and we reverse order in the inner summation to obtain
139] APPELL'S FUNCTIONS 267
co n (-1)8(a).+k [-k,b+n;
W = L L I ( ) I ( ') 2FI
n,k=O .=0 s. n - S C
c;
or
Now recall that in Ex. 6, page 128, we found, with different nota-
tion, that if m and 'Yare non-negative integers,
(3)
[
-m, d + 'Y;
~ I
d' ,
for m > 'Y
for 0 ~ m ~ 'Y.
With (3) in mind we turn to (2) and put c' = a, C = b.
Since
and
F[-n, a + k;
a' ,
F[-k, b + n;
b' ,
for n > k.
for k > n,
the only terms remaining in the double summation (2) will be those
for which k = n. Recall the definition (1) of 1/1. We thus arrive at
(1 - x)-a(l - y)-b
F
{ a, b; b, a; (1 _ x ) ( ~ _ y)' (1 _ x)(; _ y)
= (I - xy)-I.
268 JACOBI POLYNOMIALS [Ch.16
THEOREM 83. If neither a nor b is zero or a negative integer,
(
-x -y)
(4) F4 a, b; b, a; (1 _ x)(l _ y)' (1 _ x)(l _ y)
= (1 - xy)-l(1 - x)a(1 - y)b.
Next let us return to (2). Since the two 2Fr's with unit argument
terminate, we may write
_ f ree') r(e'-a+n:-I.;) r(e)
if/ - ",k r(e' +n) r(e' -a-I.;) r(e+k) r(e-b-n)k!n!
We know that
so we may write
(5) if/ = i: r(e' - a + n - k) r(e - b + k - n)
r(e' - a) r(e - b)
(a)k(b),,(l - e' + a)k(1 - e + b)n(-I)n+kxky"
k!n!(e')n(c)k
If we so choose e' that the arguments in the Gamma products add
to unity, we may use the formula
r(z) r(1 - z) = _. _7r_.
SIn 7rZ
Hence we choose
e' = 1 - c + a + b.
Then
r(e' -a+n- k) r(c-b+k-n)
r(e' -a) r(e-b)
r(l-e+b+n- k) r(e-b-n+k)
r(l-e+b) r(e-b)
sin 7r(c - b)
sin 7r(e - b - n + k)
sin 7r(e - b) = (_l)"H
(-I)n-k sin 7r(e - b) .
Therefore (5) yields, for c' = 1 - e + a + b,
140] ELEMENTARY GENERATING FUNCTION 269
if; = [i: (a)k(c - b)kXk][i: (b)n(I - C + b)nL]
hO k!(C)k n!(1 - C + a + b)n
[
a, C - b;

C" ,
] [
1 - C + b, b;
X 2/
1
\
1 - C + a + b;
THEOREM 84. If neither C nor (1 - C + a + b) is zero or a
negai'ive 'integer,
] [
a, b;
-x
--- 2FJ
1 - x
1 - c + a + b;
-y ]
1 - Y
In (6) replace x by -x/(l - x) and y by -y/(l - y) to obtain
the equivalent result
(7) F4(a, b; c, 1 + a + b - c; x(I - y), y(l - x)
[
(1, b;
= 2FJ
C" ,
] [
0, b;
X 2FI
1 + a + b - c;
Theorems 83 and 84 furnish the tools we need in the next two
sections.
140. An elementary generating function. On page 2':');) we
obtained the result
(1) Pn(a,{31(x) = ' (l+a)n(l+J3)n
k!(n-k)!(I+a)k(I+J3)n_k 2 2
With the aid of (1) we find that
i: P n (a,{31(x)t
n
= i: (1 + a)n+k(I +/),n+k[Hx-1)]k[}(x+ l)]n
t
nH,
k.n.(I+a)k(1+J3)n
270 JACOBI POLYNOMIALS [Ch.16
of which the right member is an Appell F4 as defined on page 265.
In fact,
ro
(2) L Pn(a,{l)(x)t
n
= F
4
(1+iJ, l+a; l+a, l+iJ;
n=O
By Theorem 83, page 268,
(3) F{ a, h; h, a; (1 _ U)(l
U
_ v)' (1---= u)(l - 1)))
= (1 - uv)-l(l - u)a(l - V)b,
and we put a = 1 + f3, h = 1 + a,
-U l(x - 1)
(4) ------- = -----
(1 - u)(l - v) 2'
- v lex + 1)
(1 -u)(l - v) =
As usual, let p = (1 - 2xl + t2) '. Consider
(5)
2
u = 1 - 1 + t + p'
From (5)
-11 = _1_(1 __ 1_) = L=_L+ P(l __ !_J + p)
(1 - u)(l - v) 1 - v 1 - u 2 2
(1 - t + p)(l - t - p)
4
(1 - l)2 - pZ l(x - 1)
------ ----
4 2
and in the same way,
-v (1+t)2-p2 l(x+1)
------
(1 - u)(1 - v) 4 2
as desired. Also
1 1
T - u = 2(1 + l + p),
1 1
--- = -(1 - l + p)
1 - v 2 '
from which
1 1 1 - uv
p=---+---1=
1 - u 1 - v (1 - u) (1 - v)
Hence
(1 - uV)-l(1 - u)a(l - V)b = p-l(l - u)a-l(l - v)b-1,
so that (2) and (3) yield
141] BRAFMAN'S GENERATING FUNCTIONS 271
co (2 )IJ( 2 )a
P n(a,/J)(x)t
n
= p-I 1 + t + p 1 - t + p ,
or
co
(6) L Pn(a,/J)(x)t
n
= 2a+/Jp-
I
(1 + t + p)-/J(1 - t + p)-a,
n_O
in which p = (1 - 2xt + t2) i.
141. Brafman's generating functions. Consider the sum
(1)
f h')n(o)nPn(a,IJ)(x)t
n
(1 + a):Cl + (3);
= f t. h)n(o)n[}(X - l)Jk[Hx_ -t l)Jn-kt
n
n_O k!(n - k)!(1 + a)k(l + (3)n-k
_ - I)JkUl(x + l)Jn
-.L.J II )(
k.n.(1 + a k 1 + (3)"
= F4( 'Y, 0; 1 + a, 1 + (3; - 1), + 1).
Theorem 84, page 269, yields
(
-u -v)
F4 a, b; c, 1 - c + a + b; (1 _ u)(1 _ v)' (1 - u)(1 - v)
-u
[
a, b;
=
] [
a, b;
--
1 - u
1 - c + a + b;
-v ]
1 - v .
c ,
The F4 in (1) will fit into Theorem 84 if \ve choose
1 + {3 = 1 - (1 + a) + 'Y + (I,
or 0 = 1 + a + (3 - 'Y, and
-u t(x - 1) -v l(x + 1)
(1 - u)(1 - v) = --2 -
(1 - u)(1 - v) = ---2 -.
These are the u and v of the preceding section and
-u 1
1 _ u = 1 - 1 _ u = 1 - HI + t + p) = HI - t - p),
-v
-1-- = 1 - HI - t + p) = HI + l - p).
- v
272 JACOBI POLYNOMIALS [Ch.16
Hence (1) becomes
(2)
f (')') .. (1 + 0'+ /3 -
n=O (1 + O')n(l + /3)n
,[')" 1 + a + /3 - ')';
= 2FJ
1 + a;
[
')" 1 + a + /3 - ')';
2F\
1 + /3;
in which p = (1 - 2xt + t2) I and')' is arbitrary.
The set (arbitrary,),) of generating functions in (2) was discovered
by Brafman [1].
142. Expansion in series of polynomials. It is at times desir-
able to expand the general Jacobi polynomial in a series of simpler
polynomials. We shall now exhibit a technique often useful for
this purpose and always ayailable when the simpler set is one of
special Jacobi polynomials.
In the next chapter we discuss briefly the ultraspherical poly-
nomial which is merely the special case /3 = a of the Jacobi poly-
nomial. As an illustration of the method under discussion we
propose to obtain in explicit form the expansion
n
(1) = L A1(k, n)Pk(a.al(x).
k_O
Expansions of the Jacobi polynomial in series of other special Jacobi
polynomials will be found in the exercises at the end of this chapter.
An expansion of Pn(a!J)(x) in a bizarre combination of elements of
different sets (varying a) of ultraspherical polynomials will be found
in Brafman [3J.
Naturally we do not use the orthogonality property for the
explicit determination of A1(k, n), since we have generating functions
available.
From equation (4), page 255, we get
(2)
142] SERIES OF POLYNOMIALS
273
The use of /3 = a in equation (2), page 262, yields
(3) (I-X)8 = (1 +a).s! (-I)k(1 +2a) k(I +2a+2h)Pk(a,al(x) .
2 k.D (s-k)!(I+2a)k+8+1(1+a)k
Consider the series
(4)
Using (2) and (3) in the right member of (4), we obtain
f(x, t) (1 _ X)'
w n (-1)'(1+a+/3)nH -2- tn
= LL-----
s_u s!(n - s)!(l + a),
(
1 - X)8
co (_1)8(1 + a + -2- t
n
+.
= L , '(1 + )
71,8=0 s.n. a 8
= i: (-1)8H(1 + a+ /3)n+2s(1 +2ah(1 +2a+2k)P
k
(a,al(x)t
n
+.
n,8=0 k=O nICs - k)!(l + 2a).+k+l(1 + a)k
= i: (_1)8(1 +a+ /3)n+2s+2k(1 +2ah(1 +2a+2k)PkCa,al(x)tnH+8.
n,k,8=O n!s!(l + 2a).+2k+l(1 + a)k
We now rearrange terms in the above to obtain
f(x, t)
= f t (-1)8(1 +a+ /3)n+8+2k(1 +2a) k(1 +2a+2k)P kCa,al(x)t
nH
n,k=O 8=0 s!(n - s)!(1 + 2a)8+2k+l(1 + a)k
co [-n,l + a + {3 + n + 2k;
= L zFl
n,k=O 2 + 2a + 2k;
(1 + a + /3)n+2k(1 + 2ah(1 + 2a + 2k)P
k
(a,al(x)t
nH
, n!(1 + 2a)2k+l(1 + a)k
The above 2Fl with unit argument has the value
(-l)n(/3 - a)n
(2 + 2a + 2k)n'
for which see Ex. 5, page 69. Therefore,
274
1f(x, t)
JACOBI POLYNOMIALS [Ch.16
= f (-l)n({3- a)n(l +a+ (3)n+2k(1 +2a) k(l +2a+2k)P
k
(a,a)(x)t
nH
n,k=O n!(l + 2a)n+2k+l(1 + a)k
'" n
= L L A (k,n)Pk(a,a)(x)t"
11_0 k_O
in which
A(k n) = (-1)n-:-k({3-a)n_kQ..a+{3),,+k(Lt?a)k(I+2a+2k).
, (n - k)!(l + 2a)n+k+l(1 + ah
Using (4), we may now conelude that
with
A(k,n) = (-l)n-k({3- a)n_k(l (3)n+k(1 + 2a) k(1 + 2a + 2k),
(n - k)!(l + 2a)n+k+l(1 + a)k
wltich is the result desired.
EXERCISES
1. Let
Use Bateman's generating function, page 256, to see that
'"
L gn(x)ln = oFI( -; 1 + a; -xl) oFI( -; 1 + (3; t)

and thus show that, in the sense of Section 126, gn(X) is of u-type zero with u =
D(} + a). Show also that gn(X) is of Sheffer A-type unity.
2. Show that

2x(a + (3 + n) D Pn (x) + [x(a - (3) - (a + (3 + 2n)] D P
n
-
I
(x)

= (a + {3 + n)[2nPn (x) - (a - (3)Pn-l (x)],
which reduces to equation (2), page 159, for a = {3 = O.
3. Show that

2(a + (3 + n) D Pn (x) + [a - {3 - x(a + (3 + 2n)] D Pn-
I
(x)
(a.il)
= (a + {3 + n)(a + {3 + 2n)Pn _
1
(x),
which reduces to equation (6), page 159, for a = {3 = O.
142] SERIES OF POLYNOMIALS 275
4. Show that
2n(a + {3 + n + 1) D
+ [(a + {3)(n + 2)x + n(a - (3)] D Pn(a,{J)(x)
+ (n + 1)[(0' - (3)x - (a + {3 + 2n)] D
= n[2(n + 1)(0' + {3 + n) + (a + {3 + n + 1)(0' + {3 + 2n + 2)]P
n
(a,{J)(x)
- (a - (3)(n + 1)(0' + {3 +
which reduces to equation (5), page L"i!), for a = {3 = O.
5. Ut;e the method of Section 142 to !;h()w that
P"(,,,13)(x) = n)n _ -+ 2k)P
(I + 0' + (3)" (n - k).(l + a)"+k+1
6. U!;e the result obtained in Section 1,12 to evaluate
ill (1 - X")"l'n(<x,6)(x)P
k
<q,(t)(x) !lx.
Ans. For k > n the integral is zero; for 0 k n, it has the value
(_1)n-k({3 - a)n_
k
2I+2"f(1 + a + n)f(l + a + k)r(1 + 0'+ fJ + n + k)
k!(n - k) !f(2 + 20' + n + k)f(l + a + {3 + n)
7. Use the result in Ex. 5 ahove to evaluate
i: (1 - x)"P,,("fJJ(x)Pk(a,u)(x) dx.
8. Use Theorem 84, page 269, with y = x = -v/(1 - v) to conclude that
] [
a, b;
v ZFl
1 + a + b - c;
[
a, b, Ha + b), Ha + b + 1);
= 4
F
S
a + b, c, 1 + a + b - c;
4v(1 - v) J
9. Use the result in Ex. 8, above, and Theorem 25, page r.7, to show that

[ 2a,2b,a+b;
y = sFz
2a + 2b, a + b + L
CHAPTER 17
Ultraspherical
and Gegenbauer
Polynomials
143. Definitions. The special case f3 = ex of the Jacobi poly-
nomial of Chapter 16 is called the ultraspherical polynomial and is
denoted P n(a,a)(x). The Gegenbauer polynomial Cn'(x) is a gen-
eralization of the Legendre polynomial and is defined by the
generating relation
en
(1) (1 - 2.rt + t2)-. = L Cn'(X)t
n
.
fI.=(J
If we put {3 = ex in equation (10), page 25G, we obtain
[
. ~ + ex, 1 + ex;
(2) (1 - t)-l-2a 2Fl
1 + ex;
= i: (1 + 2ex)nP
n
(a,a)(x)t
n
n ~ O (1 + ex)n '
in which the 2Fl degencra tcs into a binomial. Indeed, the left
member of (2) is
__ 0 [ 2t(x - 1)]-!-" __
(1 - t) 1 _a 1 - ( C ~ t)i: = (1 - 2xt + (2) ! a.
Hence
(3)
276
144] GEGENBAUER POLYNOMIALS 277
From (1) and (3) it follows that the Gegenbauer and ultraspherical
polynomials are essentially equivalent:
(4)
(5)
For some purposes the Cn'(x) is a more convenient form; for others,
the Pn(a.a)(x) leads to neater looking results.
144. The Gegenbauer polynomials. The polynomial C,:(x) de-
fined by
00
(1) (1 - 2xt + t
2
)-, = L C,:(x)t
n
n=O
retains most of the properties of the Legendre polynomial of Chapter
10. Note that
Pn(x) = en !(x).
We here derive a few of the results for the Gegenbauer poly-
nomial en,(x). Many other properties of Cn'(x) are then listed,
and it is left to the reader to obtain them either by using the methods
of Chapter 10 or otherwise.
From (1) it follows that
:t Cn'(x)t
n
= :t p ) ~ ~ i ~ ~ - 0:,
n=O ,,=0 n.
Hence
(2)
If p is neither zero nor a negati"e integer, the en,(x) form a simple
set of polynomials and
2n(P)nxn
Cn'(x) = ---,- + 7T'n_2.
n.
278 GEGENBAUER POLYNOMIALS [Ch. 17
It follows readily from (1) and (2) that
(3)
(4)
(5)
C
n
'( -x) = (-1) nCn'(x),
C
n
'(1) = (211)n
n! '
C;nj-l (0) = O.
Since, by (1),
m
L C'n'(x)t
n
= [(1 - xt)2 - t2(X2 - 1)]--
n=O
[
t2(X2 - l)J--
= (1 - xt)-2, I - ------
(1 - xt)2
it follows that
(6)
so that
(7)
For arbitrary 1', (6) yields
:t ('Y)nCn'(X)t
n
= :t (-y)n+2kXn(x2 - 1)ktn+2k
n=O (211)n n,k=O 2
2k
k!(11 + Dkn!
ro ('Y)2k(x2 _ 1)kt2k
= 62
2k
k!(11 + !)k(l - xt)'Y+
2
k'
144l GEGENBAUER POLYNOMIALS
Hence
[
h' h +!;
(8) (1 - xt)--r 2FJ
v + 1;
The methods of Chapter 10 also yield the following results:
(9) D=!:
dx'
(11) x DC,:(x) = - (21' + n)C,:(x),
(12) (X2 - 1) DC/(x) = nxC/(x) - (21' - 1 +
(15)
(21')n [-n, 21' + n;
Cn'(x) = -,- 2FJ
n. + 1.
I' 2,
Reversal of order of summation in (15) yields
279

1 - x
The use of Theorem 20, page GO, on (15) gives us
(17)
(21')n(X + l)n [-n, ! - I' - n;
Cn'(x) = -,- -2- 2Fl
n. + 1.
I' 2,
280 GEGENBAUER POLYNOMIALS [Ch.17
The property (3) of this section can be used to get other forms of
Cn'(x) from (15), (16), and (17).
Equation (17) yields the Bateman generating relation
(18)
From (2) and (6) it follows that
(19)
(20)
( )
(
? .) n [- n, - n + i;
'1 11 n _1,
C:(x) = ---,--- 2Fl
u. l-I'-n.
,
(2) n
C,:(x) = 2Fl
H. + I.
11 '2,
Brafman's generating function, page 272, becomes
With the aid of the methods of Section 95, we conclude from the
preceding (7) that
(
sin ex)n .;... (21')n
(22) Cn'(cos ex) = -'-{3 L.J ( k)' (2)
SIn n - . I' k
[
sin ({3 - ex)]n-k
C k'( cos (3)
SIn ex '
and from (1), page 276, that
(23) p-n-hC
n
' (x - t) = i: (n + k) '-f:,+k(X)tk
p k=O k!n! '
with p = (1 - 2xt + t2)i.
1.4.4] GEGENBAUER POLYNOMIALS 281
Equations (23) and (7) of this section lead us, by the method of
Section 96, to the bilateral generating relation (See Weisner [1])
(24) p-z, ex
p
[ -yt(x
2
- t) ] oFI[ - ;
P + 1
v 2,
= f
,,_0 (21')" '
in which L,,(2,-1)(y) is the Laguerre polynomial of Chapter 12. The
same procedure can be used in applying (23) to the relations (8),
(18), and (21) of this section.
From (6) we obtain
(,,/21 _ I)k
C:(x) = (2v)" {;(2k)!(n - 21.:)1(1' +
and thus deriye
(25) Cn'(X) = (21')"r(v + i'-[x + VX2 - 1 cos ip]" sin
2
,-lip dip.
n: r(!) rev) 0
For v > 0 and -1 x 1, (25) easily yields the following bound
on the Gegenbauer polynomial:
Either the differential equation (14) or the Rodrigues formula
(26) Cn'(x) =
leads to the orthogonality property
(27) i>l -x
2
)-!C
n
v(X)C
m
v(x) dx = 0,
From (26) it follows that
m r" n.
d
D = dx'
(28) gn = II (1 _ x2)'-![Cnv(X)]2dx = (2v)n .
-1 n!(1' + n) I (I')
In the notation of Chapter 9, the leading coefficient in Cn'(x) is
h" = 2n(v),,/n!, and we may now apply the theorems of Chapter 9
to the Gegenbauer polynomials, if v >
282 GEGENBAUER POLYNOMIALS [Ch.17
The differential equation (14) leads also to the evaluation
(29) (n - m)(n + m + 211)i
b
(1 - x2)v-ICnv(x)C
m
v(x) dx
= [(1 - X
2
)'H{C
n
'(X) D Cmv(X) - Cm'(X) D CnV(X) II.
Following the technique suggested in Exs. 1 and 2, pages 182-183,
we find that
(30) Cnv(x)
= (II) k(lI)n_k(X + yx2 - l)n-k(x - yx2-=- l)k
hO k!(n - "')!
From the generating relation (6), page 271, we obtain
(31)
with p = (1 - 2xt + t2)1.
Since
{
II'
(1 - 2xt + t2)-. = (1 - t)-2v IF _ :
Theorem 48, page 137, is applicable to Cnv(x) with a slight change
from x to HI - x) in variable. We thus arrive at the following
results:
(32)
(33) (x - 1) D Cnv(x) - nCnv(x)
= - (211 + n - - (x - 1) D
(34) (x - I)D Cnv(x) - nCnv(x)
n-I n-I
-211 L C k'(x) - 2(x - 1) L D C kV(X),

145] ULTRASPHERICAL POLYNOMIALS 283
n-l
(35) (x - l)D Cnv(x) - nCnv(x) = 2 L (_I)n-k(1' + k)Ckv(x).

The method used to deri\"e the formula of Theorem 65, page 181,
leads us also to
(36)
Equations (32) and (36) simplify the task of expressing various
other polynomials in series of Gegenbauer polynomials. See the
exercises at the end of this chapter.
From (30) in this section it follows that
(37)
Cn'(cos 0) = i= (I') 2k)O.
h 0 k. (n - Iv).
Various mixed recurrence relations are easily derived. For
instance,
(38) 21'(1 - = (n + 21')xC
n
V
(x) - (n +
(39) (n + = (I' - I)[C:+1(x) - C:_1(x)].
For expansions of analytic functions in series of Gegenbauel'
polynomials, see Boas and Buck [2; 58J for existence, uniqueness,
and region of convergence; use equations (32) or (36) to obtain the
explicit coefficients.
145. The ultraspherical polynomials. By means of the equation
(1)
or its equivalent,
(2)
(21' )nP n (v-;,v-;) (x)
Cn'(x) = (I' + '
(I + a)nCna+!(x)
(I + 2a)n
results obtained for the Gegenbauer polynomial Cnv(x) may be
transformed into results on the ultraspherical polynomial Pn(a,a)(x),
or vice versa. We leave it to the reader to obtain from equations
(1)-(39) of the preceding section the equivalent statements regard-
ing P n(a,a)(x).
Certain properties of the ultraspherical polynomial are due solely
to its being a Jacobi polynomial; others are peculiar to Pn(a,a)(x)
because of the equality of (3 and a.
284
EXERCISES
GEGENBAUER POLYNOMIALS [Ch.17
1. Show that the Gegenbauer polynomial Cn'(x) and the Hermite polynomial
lIn (x) are related by
C '(x) = v + n - k'
n k_O ' "k!(n-2k)!
2. Show that
II ()
In12] ( +' ')J')(" ()
= L (-1)k,F,(-k; 1 + v + n - 2k; l)v __ X .
n. k_u /.,.' ".1-2k
3. Show, using the modified Bessel function of Section (i5, that
m
eX' = L (v + n)l.tn(t)c,:(X).
n-O
4. Show that
CHAPTER 18
Other
Polynom ia I Sets
146. Bateman's Zn(x). One reason for interest m the poly-
nomials fn(x) generated by
(1)
(
-4xt) co
(1 - t)-1ft (1 _ t)2 =
\vhich were touched upon in Section 75, is that so many special
cases of the fn(x) have arisen in recent studies. In 1936 Bateman
[2J was interested in constructing inverse Laplace transforms. For
this purpose he introduced the polynomial
(2)
for which we obtained a pure recurrence relation in Ex. 1, page 243.
By Theorem 48, page 137, with c = 1 and 'Yn = (Dn/(n!)2, we
obtain
(3)
(4)
n = ( 1)2 ( -n) k(2k + l)Z k(X)
x n6 (n+k+1)! '
285
286 OTHER POLYNOMIAL SETS [Ch.18
n-l n-l
(6) xZn'(x) - nZn(X) = - L Zk(X) - 2x L Z/(x),

n-l
Equation (3) was given by Bateman [2J, (5) by Sister Celine [1],
(7) by Dickinson [IJ, and (4) and (G) are immediate consequences
of (3), Note that, because of Kummer's second formula, page 12G,
(3) may be written
[
-2xt ] [-;
(8) (1 - t)-1 exp (1 _ t)2 oFl
1 ;
= (1 - t)-1 = Zn(X)t
n
,
which is the form Bateman used.
Bateman moved from the Z n to the more general
(9)
which is an instance of one of the generalizations (Rainville [6J) of
the Bessel polynomials (Krall and Frink [1]) of Section 150. We
shall now exhibit the relation between the polynomial (9), including
Zn as the special case II = !, b = 0, and the problem of inverse
Laplace transforms.
With the usual notation
L{F(t) I = i"'e-BlF(t) dt = f(s); F(t) = L-l{f(S)},
we seek
(
10) FCt) = L-l{ reI + b) [1 _ 2 s - 1 + 2J-V}.
(l + S)b+l zs + 1 z
In terms of the Gegenbauer polynomial of Chapter 17,
FCt) = r(I + b)L-! f Dzn
(1 + S)H!
co Cn{1 -
= e-
t
r(1 + b)L-l L ----.
Sb+l
147] RICE'S Hn(r, p, v) 287
Now
so that
F(t) = e-
t
r(l + b)L-l :t 2Fl(-n, 21' + n; v +
n.s S
Let us therefore first evaluate
_ {1'(1 + b) ( . 1 _ I)}
A (t) = L 1 --- ---- 2FJ - n 21' + n v +. - -
Sb+l ' , "' s
By Ex. 9, page 106, with p = 2, q = 2, c = b, al = -n, a2 = 21' + n,
bI = v + !, b2 = 1 + b, Z = 1, we obtain
A(t) = t
b
2F
2
(-n, 21' + n; v+!, b + 1; t).
Thus, for the F(t) of (10) it follO\vs that
F(t) = tbe-
t
:t (2v),n
zn
zF'2( -n, 21' + n; v + !, b + 1; t).
n=O n ..
See also Ex. 13 at the end of this chapter.
In the same paper, Bateman [2J studied another set of functions
J nU'V which are, except for a simple factor and ehanges of notation,
the polynomials IF
2
( - n; (31, (32; x). The J n u. u occur again in Rain-
ville [3J and Langer [1]. In the latter, Langer is studying solutions
of the differential equation for
(
1 2 - A
2
X
3
)
IF2 fJ.; 3' 3; -9-
and thus encounters Bateman's J nU'V.
147. Rice's Hn(r, p, v). S. O. Rice [lJ made a considerable
study of the polynomials defined by
(1)
We obtained the pure recurrence relation for Hn in Ex. 3, page 243.
A few of the simpler of Rice's results are:
For Re(p) > Re(f) > 0,
(
2) H (r p v) = rep) Lltt-I(I - t)p-t-1P (I - 21'l) di-
n r(r) rep - r) 0 n ,
288 OTHER POLYNOMIAL SETS [Ch. 18
For Re(.\) > Re(p) > 0,
(,3) P (1 - 2v) = - t)r-p-1H (r p vt) dt
n r(p)r(.\ - p) 0 ";,,,
(4)
[
L L
(1 - 0-
1
F
p;
-4vt] '"
- ----- = '" II (r 7) v)t
n

( 1
- t) 2 L..-" ", ,
/,-=1)
co -1 [.\,1;
(6) (2n + I)Qn(s)H,.(.\, p, v) = 1 _ sF p;

1 - s
In (2), (3), and (5), Pn(X) is the Legendre polynomial. In (ti)
the Qn(S) is the Legendre function of the second kind defined on
page 11-)2. Sister Celine (Fasenmyer[l]) obtained the differential
recurrence relation
in which primes denote differentiation with respect to v.
Because of (4) in this section, Theorem 48, page 137, applies to
Rice's H n with the choices c = 1, x = v,
From Theorem 48, preceding equations (1) and (7) follow, as do the
results
(9) vH n'(.\, p, v) - nH n('\, p, v)
n-l
= - L [H k(L p, v) + 2vH /(.\, p, 1')],
k_O
n-l
(10) vHn'(.\,p,v) -nHn(.\,p,v) = L (_1)n-k(1 +2k)H
k
(.\,p,v).
k-O
See Rice [1] for many other properties of H n'
148] BATEMAN'S Fn(z) 289
148. Bateman's F n(z), Bateman [3] studied the polynomial
(1) Fn(z) = SF2( -n, n + 1, HI + z); 1, 1; 1)
quite extensively, and he and others kept returning to it in later
papers. Note that the variable Z is contained in a parameter of
the 3F2, not in the argument. That the F n(Z) form a simple set of
polynomials should be apparent upon consideration of the nature of
the terms in a aF2.
Bateman obtained the generating relation
(2)
[
1 '+ 'z.
(1 _ t)-1 zF, j,:i 2,
1 .
,
-4t ] en
(1 _ t)2 = ~ F,,(z)t
n
and the pure recurrence relation
together with numerous mixed relations involving a shift in ar-
gument as well as in index. Two examples are quoted here:
(4) (Z + 1)2[F n(Z + 2) - F n(Z) J
+ (z - 1)2[Fn(z - 2) - Fn(z)J = 4n(n + I)F
n
(z),
(5)
CD 2t [ - ~ ' 1 + 1z; -4t ]
~ [Fn(z - 2) - Fn(z)Jt
n
= (1 _ t)3 2F, 2; (T-=- tr .
Since Fo(z) = 1, F,(z) = -z and F
2
(z) = HI + 3z
2
), it follO\ys
from (3) that for I z I < 1, 'F n(Z) , < 1, n ~ 1. Bateman also
made much use of F n(Z) in the study of definite integrals and certain
. .
senes expansIOns.
In 1956 Touchard [IJ introduced polynomials for which he did
not give either an explicit formula or a generating relation. Later
that year Wyman and Moser [IJ obtained for Touchard's poly-
nomials a finite sum formula and a generating function. Their
generating function was equi\'alent to Bateman's (2) above, as can
be seen by applying Theorems 21 and 23 of Chapter 4. In 1957
Carlitz [IJ pointed out that Touchard's polynomials and Bateman's
F n(Z) are essentially the same, the former being
( - 1) n ( n ! ) 2 F n (1 + 2x)
2
n
(1)"
290 OTHER POLYNOMIAL SETS [Gh.18
Also in 1957 Brafman [6J obtained two generating functions for
Touchard polynomials, and one of these [his (12)J is equivalent to
that of Wyman and Moser and therefore to Bateman's (2) in this
section, by Theorem 20 of Chapter 4. Brafman's other generating
relation is a useful contribution to the study of Bateman's F n(z);
it is
(6) F (
1 1. l' t) F (1 + 1 l' ) _ ~ Fn(z)t
n
1 1 2 - 2
Z
, , 1 1 2 2
Z
" -t - L..- "
n=O n.
See also Ex. 3 at the end of this chapter.
Another polynomial in ,vhich interest is concentrated on a param-
eter is the Mittag-Leffler polynomial
gn(Z) = 2z 2F
1
(1 - n, 1 - z; 2; 2),
which was investigated to some extent in Bateman [4]. Two
generating functions
~
(7) (1 + t)z(l - t)-z = 1 + L {J,,(z)tn,
11=1
(8)
and several mixed recurrence relations for gn(Z) were given by
Bateman. He also included some discussion of a generalization of
the Mittag-Leffler polynomials to
( -r)n
(9) gn(z, r) = --,- 2F
1
( -n, z; -r; 2).
n.
149. Sister Celine's polynomials. Sister Celine (Fasenmyer
[1]) concentrated on polynomials generated by
(1)
+
which yields
Her polynomials include as special cases Legendre polynomials
149] SISTER GELINE'S POLYNOMIALS 291
Pn(1 - 2x), some special Jacobi polynomials, Rice's Hn(r, p, v),
Bateman's Zn(X) and F n(Z), and Pasternak's
(3)
[
-n, n + 1, HI + Z + m);
Fnm(z) = F
1, m + 1;
which is a generalization of Bateman's F n(Z). The simple Bessel
polynomial of Section 150 is also included.
The two major parts of Sister Celine's work are the technique for
obtaining pure recurrence relations (illustrated in Chapter 14) and
her extension of Rainville's work on contiguous function relations
to certain terminating pFq's for v;hich p > q + 1. She also ob-
tained a few results of interest for some of the simpler of her poly-
nomials.
We quote the easily derived result
which includes
1 feo
(5) P n(1 - 2x) = . r In y-!e-
lI
fn( -; - ; XV) dy
v
7r

and
(6)
Using p = 1, q = 1, al = !, b
1
= 1, we find that Sister Celine's
(4) in this section becomes
(7)
1 feo
fnC!; 1; x) = --= In y-
i
e-
lI
fn( - ; 1; xV) dy.
V7r 0
As she points out, for Bateman's Zn,
(8) fn(!; 1; x) = Zn(X),
and in terms of the simple Laguerre polynomial,
(9)
292 OTHER POLYNOMIAL SETS [Ch. 18
Equation (9) is the special case a = -n, f3 = 1, of Ramanujan's
theorem obtained in Ex. 5, page 106. By combining (7), (8), and
(9), Sister Celine obtains
(10)
The general polynomial of Sister Celine, (2) in this section, falls
under the classification of Theorem 48, page 137, with c = 1. For
the moment, denote the polynomial of (2) by Cn(x):
(11)
Then Theorem 48 yields
(12)
(13)
n-l
(14) XC,,/(X) - nC,,(x) = - L [Ck(x) + 2xC/(x)],
k_O
n-l
(15) xCn/(x) - nC,,(x) = L (- I) ,,-k(2k + I)Ck(x).

Next let us turn to the polynomial (2) with no a's and no b's.
Put
(16)
a polynomial whose pure recurrence relation we derived in Chapter
14. As Sister Celine points out,
(17)
involving the Laguerre and Hermite polynomials with the symbolic
notation of Chapter 15.
For thej,,(x) of (16) the generating function (1) becomes
(18)
[
-4xt ] ro
(1 - t)-1 exp (1 _ t)2 =
150] BESSEL POLYNOMIALS 293
Therefore this fn(x) is a polynomial of Sheffer A-type zero. In the
notation of Chapter 13,
(19) A(t) = (1 - t)-I,
-4t
H(t) = (1 - t)2
Then the inverse of H Ct) is
(20)
2 <n - ( ~ ) k+1tHl
J(t) = 1 - -_ .. = L -- -- -I'
1+V
1
-
l
k ~ O (1.:+2).
Computation of the constants appearing in Theorems 73 7G, pages
224-223, shows that
(21) Oik = 1, hk = -4(k + 1), Ek = -4(k + 1)2,
(22)
In particular, Theorems 75 and 7G yield the relations
(23) nfn(x) = L [1 - 4(k + 1)2x]fn_1_k(X),
k ~ O
n-1
(24) fn'(x) = -4 L U'- + 1)fn-1-k(x).
k ~ O
It is interesting to note that (23) and (24) eombine to yield
n-1
(25) xfn'(x) - nfn(x) = L [4k(k + l)x - 1]jn_1_k(X)
k ~ O
n-1
= L [4(n - k)(n - 1 - k)x - l]/k(x),
k_O
a different result from that given by (15),
n-1
(26) xfn'(x) - nfn(x) = L (-1)n-k(2k + l)fk(x),
k_O
which came from Theorem 48.
150. Bessel polynomials. In 1949 Krall and Frink [lJ initi-
ated serious study of what they called Bessel polynomials. In their
terminology the simple Bessel polynomial is
(1) Yn(x) = 2Fo( -n, 1 + n; -; ~ x )
294 OTHER POLYNOMIAL SETS [Ch. 18
and the generalized one is
(2) Yn(a, b, x) = 2Fo( -n, a-I + n; -; - L)
The simple Bessel polynomial Yn(X) is related to the Bessel functions
in the following manner:
Yn(:J = + inJ -n-l(r)],
In+l(r) = (27rr)-{i-
n
-
1
e;ry,,( + in+1e-;rYn(fr)],
J -n-l(r) = (27rr)-{ ine;rYn( + i-ne-;ry,,(i) J.
Let us take as a standard for polynomials of this character
(3) <Pn(C, x) = (C),n 2Fo( -n, C + n; -; x).
n.
The simple Bessel polynomial is the special case with C = 1 and x
replaced by (- To get the Krall-Frink generalized Bessel
polynomial, introduce the redundant parameter b by replacing x
by (-x/b), put C = a-I and multiply <Pn(a - 1, -x/b) by
n!/(a - l)n.
From (3) it follows that
(4)
or
(5)
Equation (4) shows that <Pn(C, x) fits into the scheme of Theorem
48, page 137, with 'Yk = + This yields the generat-
ing relation
(6)
(
-4xt ) m
(1 - t)-c 2Fo + - ; (1 _ t)2 '" <Pn(c, x)t
n
and the further properties
(7)
n _ ,.;.. (-l)k(c + 2k)<Pk(C, x)
x -n.L.. (k)'() ,
k=O n - . C n+k+l
150] BESSEL POLYNOMIALS
295
n-l
- L: [eCPk(c, x) + 2Xcpk'(C, x)]

n-J
= L: (-l)n-k(c + 2k)CPk(C, x).

For C = 1 the generating relation (6) appears in Nister Celine's
work, Fasenmyer [1]. Essentially the generating relation of The-
orem 48 for if; a 1,FQ appears in Rainville [6J with minor variations in
notation. Sister Celine in her Michigan thesis also obtained con-
tiguous function relations applicable to 2Fo's of the type of (2) or
(3) in this section. Equation (7) appears in Dickinson [IJ.
It is a simple application of Sister Celine's technique (sec Chapter
14, or Fasenmyer [2J) to obtain such relations as
(9) (e + 2n - I)x
2
cpn'(c, x)
= n[I + (e + 2n - I)x]cp,,(c, x) - (c + n - l)CPn-I(C, x),
(10) ecpn(C + 1, x) = xcpn'(c, x) + (c + n)CPn(c, x),
(11) (c+2n-I)CPn(c-I, x) = (c-I)[cpn(c, x) +CPn_l(r, x)],
(12) n(e + 2n - 3)CPn(e, x)
= (c + 2n - 2)[e - 1 - (c + 2n - 1)(c + 2n - 3)X]CPn_l(C, x)
+ (e + n - 2)(e + 2n - I)CPn_2(c, x).
For the moment let
(13) fn(x) = :t (-
k!(n - Ie')! '
of which CPn(C, x) is the special case On = 1. From (13) it follows
that
[
1 + 1 1 + 1 + 1,
00 zC zn, zC zn 2,
= L: 2FI
c+n'
,
]
0 tn
-4.rt -T
n.
= (1 + 4xt)-i f ( 2 )c+n-l On;"
1 + VI + 4xt n.
296 OTHER POLYNOMIAL SETS [eh.18
by application of the result in Ex. 10, page 70. K ow let
ro On
U
"
F(u) = .
n-O n.
It is then possible to \vrite
For F a pFq, what is essentially the preceding (14) appears in
Rainville [6] with minor changes in notation. By using On = 1,
we obtain Burchnall's generating relation
(15)
(1 + 4Xt)-!C + vj--t4XtY-I expC--+- ;:+4;.D
= x)t
n
(c)n '
",hich appeared in Burchnall [1] with different notations. For
c = 1, equation (15) appears in Krall and Frink [1]. It is evident
from (15) that the reversed Bessel polynomials X"l{Jn(C, l/x)/(c)n are
of Sheffer A-type zero, from which several properties of Bessel poly-
nomials arise. If the F of equation (14) is chosen to be a oF
q
, the
reversed polynomials are seen to be of CT-type zero, as discussed
in Chapter 13.
Brafman [4J obtained a whole class of generating relations which
in our notation become
(16) [a, C - a;
,po
,
f (a)n(c - a)nl{Jn(c, x)t
n
- (C)n '
for arbitrary a. Brafman, of course, discovered (lG) by using a
limiting process on his class of generating relations for Jacobi
polynomials, for which see Section 141 or Brafman [1]. The proof
in Brafman [4] is based on Whipple's result, Theorem 31, page 88.
151] BEDIENT'S POLYNOMIALS 297
AI-Salam [lJ repeats much of the known material, with proper
credits, and obtains many new results. He makes free use of
pseudo-Jacobi, pseudo-Laguerre, and pseudo-Bessel polynomials.
His concept of generating functions is not that defined in Chapter 8.
In addition to the various references given in this section, see
Aganval [lJ, Carlitz [2J, and Grosswald [1].
151. Bedient's polynomials. Bedient [lJ, in his study of some
polynomials associated with Appell's F2 and F
s
, introduced
(1)
(iJ)n(2x)n [-tn, -tn + t, , - (3;
R,,({3, ,; x) = I 3F2
n. 1
" - f3 - n;
;,]
and
(2) Gn(a, (3; x) =
(ab({3h(2xE 3
F
Z[-t
n
, -tn + t, 1 - a - (3 - n;
n!(a+{3)n 1-a-n l-{3-n' X2
, ,
One of several reasons for interest in Rn and G
n
is their connection
with the Gegenbauer polynomials of Chapter 17. Indeed,
(3) Lim R
n
(f3, ,; x) = C
n
l3(x),
')'+00
Lim Gn(a, f3; x) = C
n
l3(x),
,,+00
Lim Gn(a, (3; x) = Cn"(x).
13+
00
For the R
n
, Bedient obtains the generating relations
[
{3, , - {3;
(4) (1 - 2xt)-13 ZFl
,;
and
(5) IF
1
(f3;,; l(x - vxz -1)) IF
1
(f3;,; l(x + vx-z - 1))
_ f Rn ({3, ,; x)t
n
-
For the G
n
he finds that
298 OTHER POLYNOMIAL SETS [Ch. 18
=
(6) 'IJi't(a, {3; a + (3; 2xt - t2) = L Gn(a, {3; x)t
n
n=O
and
(7) (3; -; t(x - yx2 - 1) 2Fo(a, (3; -; l(x + yx2 - 1)
<v
'" L (a + (3)"Gn (a, (3; x)tn.
nt-<O
Bedient [lJ contains numerous other properties of the Rn and Gn
152. Shively's pseudo-Laguerre and other polynomials. Shively
[lJ studied the pseudo-Laguerre set
(1) Rn(a, x) = Jp()2)n IF
1
( -n; a + n; x),
n. a n
which are related to the proper simple Laguerre polynomial
L,,(x) = IF
1
( -n; 1; x)
by
(2)
Rn(a, x) = (a - l)n+t
Ln
-
k
(x).
(a - 1)" bO k.
Toscano [1] had already shown that
(3)
(
2 I ( - 4xl )
(1 - 1 + yI --4:[ exp (1+-yi- - 4tY
00
= L Rn(a, x)tn.
n=O
Shively obtained Toscano's other generating relation
(4) F (
1 1 ) L
oo
x;ln
e
2t
0 - - + t2 - xl = -----
1 ,2 2, (1 + 1 )
11.=0 2 za n
and extended Toscano's (3) in this section to
(5) (1 _ 2 _ pF q[a
1
' .. " a
p
; -4xt]
1 + yI 4t Q Q. (l + yl'=-4t)
2
tJl, , fJq,
00
= L Sn(X)t
n
,
n=O
in which
153] BERNOULLI POLYNOMIALS 299
( ) [-n, al,"', a
p
;
S .. (x) = p+IF q+l
n. an +
a n, {31,, {3q;
(6)
For the particular choice p = 0, q = 1, (3l = 1, a = 1, the Sn(X)
becomes
(7)
for which Shively has the additional generating relation
(8) oFl[-;
1;
t + + "]
= i:
n_U (211)!
The Rn(a, x) of (1) is of Sheffer A-type zero, as pointed out by
Shively. He obtains many other properties of Rn(a, x).
153. Bernoulli polynomials. Much good has come from the
study of Bn(x) defined by
(1)
text 00 Bn(x)tn
-t-- = L ---, -,
e - 1 n.
particularly in the Theory of Kumbers. The Bn(x) of 0) are the
Bernoulli polynomials, which have been generalized in numerous
directions. See Erdelyi [IJ and [3]. The BJr)hl! are of Sheffer
A-type zero, from which fact various interesting properties may be
obtained. It is also a simple matter to show that
(2)
(3)
(4)
Bn(X + 1) - Bn(x) = nx
n
-
1
,
Bn(x + 1) {I + B(x)} n,
BnCl - x) = (-1) nBn(x).
One definition of Bernoulli numbers Bn is
(5) Bn = Bn(O).
It follows that
(6)
and
300
(7)
Since
OTHER POLYNOMIAL SETS [Ch.18
n ~ 2,
t -t
-- = -t+---
e
t
- 1 e-
t
- 1 '
Bl is the only nonzero Bernoulli number with odd index. Thus the
generating relation
(8)
may also be written
t 00 B2nt2n
(9) -e=.-1 = Bo + Bit + L -(2 )1 '
e n_1 n.
in which Bo = 1, BI = -!, etc.
For us one natural extension of Bn(x) is to replace the factor ext
in the generating relation (1) by any oFq with argument xt. The
resultant polynomials are of u-type zero in the sense of Chapter 13.
154. Euler polynomials. The polynomials En(x) defined by
(1)
are called Euler polynomials, and the numbers
(2)
are called Euler numbers. The polynomials Enex)/n! are of Sheffer
A-type zero.
It is not difficult to obtain such results as
(3)
(4)
(5)
(6)
Since the
relation
(7)
En(x + 1) + En(x) = 2x",
En(1 - x) = (-1) nEn(x),
En(x + 1) ~ {I + E(x) 1 n,
nEn_l(x) = 2Bn(x) - 2
n
+
I
B
n
(!x).
Euler numbers defined III (2) have the generating
co E ntn 2e
t
2e-
t
~ 7 = e
2t
+ 1 = 1 + e-
Zt
'
it follows that EZn+1 = O.
The Euler polynomials have been generalized in vanous ways.
See Erdelyi [lJ and [3].
155] TCHEBICHEFF POLYNOMIALS 301
155. Tchebicheff polynomials. The Tchebicheff polynomials
Tn(x) and Un(x) of the first and second kinds, respectively, are
special ultraspherical polynomials. See Szego [lJ and Chapter 17
above. In detail,
(1)
n'
T (x) = --"-P (-i.-i)(x)
n n ,
(2)
These are often introduced by the relations
sin(n + 1)0
Tn(cos 0) = cos nO, U,,(cos 0) =
SIll IJ
From (1) and (2) many of the formulas on Jacobi polynomial..;,
Chapter 16, can be converted into results on Tchebicheff
nomials by choosing (l' = {3 = ! or (l' = {3 = -!.
Some of the generating functions to be found are
ro
(3) (1 - 2xt + (2)-1 = L U,,(x)t",
Tt"""O
co
(4) (1 - xt)(l - 2xt + t2)-1 = L Tn(x)t
n
,
u=O
(5)
(6)
f: U,,(x)t
n
+
1
= ext
n 0 (n + I)! V X2 - 1
Relations (3) and (4) at once yield
(7) n 1,
(8)
Of the several explicit formulas for Tn and Un (sec Chapter 17
with appropriate (l' and (3), we note
(9)
[n/2J n!xn-Zk(x2 _ l)k
Tn(x) = 6
(10)
302 OTHER POLYNOMIAL SETS [Ch.18
(11)
[n/2J (n + 1)!xn-2k(x2 _ l)k
Un(x) = L .
bO (2k + l)!(n - 21d!
Equation (10) relates 'L(x) to the Neumann polynomials of Chapter
G. See equations (4) and (G), pages 116-117.
EXERCISES
1. For the Bernoulli polynomial of Section 1.53 show that
n n!Rk(x)
xn =- t:Jk!(n --k-+ J)!
2. Let B,,(x) and Bn = Rn(O) denote the Bernoulli polynomials and numbers as
treated in Section 153. Define the differential operator A(x, D) by
00 R2kD2k
A(x, D) = (x - - .t:l (2k)! .
Prove that A(x, D)Bn(x) = nBn(x).
3. Consider the polynomials

1/;n(C, x, y) - (c)" SF2 e _, _ IX _ n'
, 2 2 ,
Show that
F (
' ,..) F (' ,.. ) _ 1/;,,(c, x, y)t
n
1 l'i - 'ix, C, yl 1 1'2 + 'iX, e, -I -- L.. -,--,
7l=() n.
-yt
Z
]
(l - yt)(l + t)
and that 1/;n(l, X, 1) = F,,(x) , where Fn is Bateman's polynomial of Section 148.
4. Sylvester (1879) studied polynomials (sec page 255 of ErdeJyi [3J),
xn ( 1)
"'n(x) = ,zFo -n, x; -; - - .
n. x
Show that
co
(1 - t)-xe
xt
= L "'n(x)t",
5. For Sylvester's polynomials of Ex. 4 find what properties you can from the
fact that "'n(X) is of Sheffer A-type zero.
155] TCHEBICHEFF POLYNOMIALS 303
6. Show that Bateman's Zn(X) , the Legendre polynomial Pn(X) , and the
Laguerre polynomial Ln(x) are related symbolically by
Zn(X) 'i= P
n
(2L(x) - 1).
7. Show that Sister Celine's polynomial
jn(X) = 2F2( -n, n + 1; 1,!; x)
of equation (16), page 292, is such that
lWe-Xjn(X) dx = (- 1 )n(2n + 1).
8. For Sister Ccline's jn(X) of Ex. 7, show that
dx
[
-n,n+ 1, -m,m+ 1;
= (-l)n+m(2n + 1)(2m + 1)4F3 1,
o k n,
= 0; k > n,
10. Gottlieb [1] introduced the polynomials
Show that
A) = e-nA ,F
I
( -n, -x; 1; 1 - eA).
- 1)" n n!(1 - A)
A + JJ.) = _ e}.)" t:o k!(n -=IF'
L
o:> -.1 . _ _ res -
o e 'Pn(X, t) dt - res + l)r(s + x - n + 1)'
(x - A) = - 1; A) - A),
A) - A)] = (x + 1) [<Pn(X + 1; A) - 'Pn(X; A)],
(x + n + A) = xeA'Pn(X - 1; \) + (n + l)e
A
<Pn+lex; A),
and see Gottlieb [1] for many other results on A).
304 OTHER POLYNOMIAL SETS [Ch.18
11. With fn denoting Sister Celine's polynomials of Section 149, show that
and
i'x!(t - x)!fn( -; -; x) dx = 7rZ
n
(t) ,
i'x!(t - x)!fn( -; -; x(t - x dx = 7rZ
n
(it2) ,
12. Define polynomials <Pn(X) by
<Pn(X) = -n, C + n; 1, 1; x).
n.
Show that
<Pn(X) = t (-I)n-k(c -
(n-k)!(n+k+I)!
13. Show that the F(t) of equation (10), page 2<..;(), can be put in the fonn
-4zt ]
(I - Z)2
CHAPTER 19
Elliptic
Functions
156. Doubly periodic functions. Let the functionfCz) have two
distinct periods 2Wl and 2W2:
(1) f(z+2wl) =f(z),
(2) fCz+2w2) = fCz).
Further, let the periods
have a nom'eal ratio,
wr/ W2 not real, so that
there exists a network of
period parallelograms,or
meshes, as shown in 0
Figure 8. The period
parallelograms are called
z-p!ane
fundamental if they are Figure 8
the smallest such meshes for the function fCz).
The values of fez) in any parallelogram P in Figure 8 are repeated
in each of the other parallelograms, as indicated in (1) and (2) above.
Doubly periodic functions are a natural extension of such simply
periodic functions as the familiar trigonometric functions of ele-
mentary mathematics.
305
306 ELLIPTIC FUNCTIONS [Ch. 19
157. Elliptic functions. If a doubly periodic analytic function
fez) has in the finite plane no singular points other than poles, then
fez) is called an elliptic function.
Consider a fundamental period parallelogram P of an elliptic
function fez). There are only a finite number of poles and zeros
of fez) in the closed region P because any point of condensation of
poles or of zeros of fez) would be an essential singularity of the
function, which is contrary to the definition. Therefore there
exists a point t in P such that the parallelogram with vertices at t,
t + 2W1, t + 2Wl + 2W2, l + 2W2 has on its boundary no poles and
no zeros of fez). That parallelogram is ealled a cell, shown in
Figure 9.
z-plane
o
Figure 9
We replace the net-
work of fundamental
period parallelograms of
Figure 8 with a network
of cells, each of which
contains no pole and no
zero on its boundary.
One reason for this is
that we wish to perform
integrations along the
boundary of a cell; we
could not integrate fez)
around the boundary of
P for fear of encountering a singular point of the integrand.
158. Elementary properties. Consider an elliptic function fez)
with periodicity properties
(1)
(2)
fez + 2Wl) = fez),
fez + 2W2) = fez),
and a network of cells as defined in the preceding section.
Certain simple properties of fez) are at once available.
THEOREM 85. The number of poles of fez) in any cell1"s finite, and
the number of zeros of fCz) in any cell is finite.
THEOREM 86. An elliptic function with no poles in a cell is a
constant.
158] ELEMENTARY PROPERTIES 307
We have already seen that Theorem 85 is a consequence of the
fact thatf(z) is permitted no essential singularity in the finite plane.
For Theorem 86, note that if fez) has no poles in a cell, it is
analytic and bounded in the closed cell and also throughout the
finite plane because of its periodicity. Then by Liouville's the-
orem, * it follows that fez) is a constant.
THEOREM 87. The sum of the residues at the poles 'in a cell of an
elliptic function 'is zero.
Proof: Let SR be the sum of the residues at the poles of the
elliptic function fez) in the cell shown in Figure 9. Then 27riS
R
equals the integral of fez) around the cell:
(3) 27riS
R
= J
1
-/
2
w'f(z) dz + f
l
+
2w
,+2w'f(Z) dz
t t+2Wl
in which each integration takes place along a fitraight line path.
Let us change variable of integration in the second integral on
the right in (3) by replacing z by z + 2WI, and in the third integral
on the right in (3) by replacing z by z + 2W2. We thus obtain
f
l
+
2W
, fl+2W,
(4) 27riSR = t fez) dz + t fez + 2WI) dz
+ ~ 2 W / ( Z + 2
W
2) dz + f:2w!(Z) dz.
But fez + 2W2) = fez), so that the first and third integrals differ
only in sign (direction of integration), and becausef(z + 2WI) = f(z) ,
the second and fourth integrals also have the sum zero. Then
27riS
R
= 0; so, SR = 0, and the proof is complete.
THEOREM 88. For an elliptic function f(z), the number of zeros in
a cell equals the number of poles in a cell.
Proof: Let N. be the number of zeros and N
p
the number of
poles of fez) in a cell. Here the multiplicity is to be counted; a
pole of order two counts as two poles; a zero of order three counts as
See any book on complex variables or function theory; for instanc<', Churchill
[3; 96].
308 ELLIPTIC FUNCTIONS [Ch.19
three zeros, etc. Then from the study of functions of a complex
variable we know that
(5)
. if'(z) dz
27r1,(N. - N
p
) = B fez) ,
in which the integration is to be carried out around the boundary
B of the cell.
From
(6) and fez + 2wz) = fez),
it fo11o\Y8 that
(7) and f'(z + 2 w ~ ) = f'(z).
Thcrefore f'(z) and abo j'(z)/f(z) ha\'e the samc periods as doeH
fez). Hence the deyice used in showing that the integral Lf(Z) dz
is zero may be used to show that
i
j'(z) dz = o.
B fez)
Thcrefore N. - N
p
= 0, which we wished to prove.
The actual situation is that Theorem 88 is merely a particular
instance of the fact that the number of times an elliptic function
fez) assumes any particular value c in a cell is independent of c. For
zeros of fez), c = 0; for poles of fez), c = ro.
THEOREM 89. For an ell?phc function fez) and for any jimte c,
the number of zeros of fez) - c is independent of c.
Proof: The function fez) - c has the same periods as fez) and
so doesf'(z). The poles of fez) - c are the poles of fez). Then the
difference between the number of poles of fCz) and the number of
zeros of fez) - c is
1 i f'(z) dz
27ri B fez) - c'
which can be shown to be zero by the same device as was used in
the proofs of Theorems 87 and 88.
159. Order of an elliptic function. For an elliptic function
fez) we define as its order the number (counting multiplicity) of
160] WEIERSTRASS FUNCTION P(z) 309
poles of fez) in a cell. We already know that the order cannot be
zero unless fez) is constant. If fez) were of order one, then in a
cell there would be a single, and simple, pole. By Theorem 87,
page 307, the residue at that pole must be zero, so the point is not
a pole but a point of analyticity of f(z). lIence f(z) has no poles in
the cell, and fez) must be constant. We have proved the following
result \vhieh will be used over and over in Chapter 20.
THEOREM 90. A n elliptic function of order less than two is a
constant.
From the standpoint of its order, the simplest elliptic function
will be one of order two. There are essentially t\VO types of elliptic
functions of order two. If there is but one pole in each cell, that
pole must be of order two and the residue at the pole must be zero.
If there are two simple poles in each cell, the residues at those poles
must be negatives of each other. We shall encounter specific
instances of each of these types of elliptic functions of order two.
160. The Weierstrass function P(z). Let m and n be integers
and define
(1)
where 2"'1 and 2"'2 are to be the fundamental periods of our elliptic
function and "'1/"'2 is not real. We shall show that the function in
(2), below, devised by Weierstrass, is an elliptic function of order
two, with each cell containing a second order pole.
Let a prime attached to the summation symbol indicate that the
indices of summation are not both to be zero at once. Then define
the Weierstrass function P(z) by
(2) P(z) = ~ + f' 1 1 __ _ ~ .
Z2 m,n=-ro (z - Om,n)2 (""12
.l'm,TI
The deletion of the m = n = 0 term from the summation is forced
upon us by the last term in (2). Since the general term in the
series in (2) is 0(0;;;\), the series is absolutely and uniformly con-
vergent in any closed region which excludes the origin and all the
other points z = am, n. Then P(z) is analytic throughout the finite
plane except for a double pole at each of the points z = am, n for
integral m and n.
310 ELLIPTIC FUNCTIONS [Ch. 19
Now
(3)
P'(z) = f -2 3
m,n=-oo (Z - Qm,n)
in which the m = n = 0 term need not be deleted and the differen-
tiation is justified by the uniform convergence of the series in (3).
Now P'(z) is an odd function of z, since
(4)
m 2 ~ 2
P'(-z) = L ----3 = L ---(--3'
rn,TL>=.-= (z + Qm,n) m,1t=--cn (z - .. lm,n)
where the second scrips is obtained by replacing m by (-m) and
n by (-n) in the first series. By (4),
(5) P'( -z) = -P'(z).
The same device can be used on the series in (2) to show that
P(z) is an even function of z,
(6) P( -z) = P(z).
We next show that P'(z) has the periods 2w! and 2W2' Consider
Since Qm,n = 2mwl + 2nw2, we have
m -2
P'(z + 2Wl) = m . ~ - = (z - Qm_l,n)3
and a shift of index from m to (m + 1) shows that
(7) P'(z + 2Wl) = P'(z).
It follows in the same way that
(8) P'(z + 2W2) = P'(z),
and we now see that P'(z) is an elliptic function.
Integration of each member of (7) yields the following equation:
(9) P(z + 2Wl) = P(z) + c,
with c constant. In (9) use z = -WI to obtain
P(WI) = P( -WI) + c.
But by (6), P(WI) = P( -WI), so c = 0 and (9) yields
(10) P(z + 2WI) = P(z).
162] DIFFERENTIAL EQUATION FOR P(z) 311
The same device shows that
(11) P(z + 2W2) = P(z).
Hence P(z) is an elliptic function with poles of order two at each
of the points z = Qm,n, m and n integral.
161. Other elliptic functions. From the Weierstrass function
P(z) we easily obtain an elliptic function with any specified set of
zeros and poles in a cell.
First consider an even elliptic function which, in a cell, is to ha ve
zeros at aI, a2' . " an and (automatically) at the points congruent
to -aI, -a2,"', -an. Let multiplicity be accounted for by
repetition, that is, by equality of a's. Let the desired poles in the
cell be at b
l
, b
2
,' " b
n
and at the points congruent to -bj , -b
2
,' "
- b
n
, with multiplicity treated as with the zeros. Then
(1)
has the desired zeros and poles in a cell and is an even elliptic
function. Odd elliptic functions may be constructed in the same
way, using P'(z), or may be treated by using the product of P'(z)
and a rational function of P(z). Since any funetion may be split
into its even and odd parts, the procedure sketched here permits
construction of elliptic functions with any desired permissible dis-
tribution of zeros and poles in a cell. See Whittaker and Watson
[1; 448 ff].
162. A differential equation for P(z). We know that P(z) is
analytic in a deleted neighborhood of its second-order pole at z = o.
Indeed, from the definition
(1)
P(z) = 12 + I:' { 1 2 __ 1_},
z (z - Qm,,,) ,,2
,n
we find that P(z) -Z-2 is analytic at the origin and is zero there.
Also P(z) - Z-2 is an even function of z, so there exists a Taylor
expansion of the form
CD
(2) P(z) - Z-2 = L: a
n
z
2n
n=!
convergent in a region around z = O.
From (2) we need only a few terms. We write
312 ELLIPTIC FUNCTIONS [Ch.19
(3) as z ----? O.
We keep in mind that the terms lumped together in O(Z5) are actually
as exhibited in (2), so that differentiations with respect to z are
legitimate.
From (3) we obtain
(4)
Thc series (3) for P(z) contains onc negatiye power, Z-2; the series
(4) for P'(z) eontains one negative power, _2Z-3. In order to
eliminate negatiye pO\vers of z, we are thus led to consider [P(Z)]3
and [P'(Z)]2. In detail,
(5)
and
(0)
Then
But
[P'(Z)]2 = 4z-
6
- 8a1Z-2 - 16az + o(z)
[P'(Z)J2 - 4[p(z)Ja = -20alz-
2
- 28a
2
+ o(z).
and we arriYe at
(7) [P'(Z)J2 - 4[p(z)Ja + 20a
1
P(z) + 28a2 = o(z).
The left member of (7) is an elliptic function with periods 2wI
and 2W2. That function is analytic at z = 0, and therefore also at
z = Qm.n. The function is therefore constant, and that constant
is zero, as can be seen by letting z ----> 0 and noting the right member
of (7).
We have shown that P(z) satisfies the first-order nonlinear dif-
ferential equation
(8)
in which al and a2 are the specific coefficients stipulated in (2). It
is customary to put 20al = gz and 28a2 = ga. We then express our
result by stating that y = P(z) is a solution of the differential
equation
(9)
(
dy)Z 4 3
dz = Y - gzY - ga.
Since equation (9) is invariant under a translation in the z-plane,
the general solution of (9) is Y = P(z + a), with a an arbitrary
163] ELLIPTIC INTEGRALS 313
constant. It can be shown * that every solution of (9) is of the
form y = P(z + a).
163. Connection with elliptic integrals. By elementary means
it follows from the differential equation
(1)
that one solution is
(2)
But Y = P(z + a), and if 've let y - 00, z - 0, so that P(z) has a
pole at z = a. Then a = Qm,n for some integers 112 and n. Then
y = P(z + Qm,n) = P(z).
That is, the inverse of the integral in (2) is the Weierstrass function,
(3) y = P(z).
Any nondegenerate integral whose integrand contains the square
root of a cubic or quartic polynomial in the variable of integration
together with rational functions is called an elliptic integral. The
name arose from the occurrence of one such integral in the problem
of finding the length of arc on an ellipse. As the foregoing discus-
sion shows in relating (2) and (3), an elliptic function is the inverse
of an elliptic integral. A corresponding fact in elementary singly
periodic functions is that the integral
(4)
f
v dt
z -
o VI - t2
has as its inverse the simply periodic trigonometric function
(5) y = sm z.
See pap;es 437 and 484-485 of Whittaker and Watson [1]. The major problem is to
show that Wj and w. exist for given g. and g .
CHAPTER 20
Theta
Functions
164. Definitions. In this chapter we study in some detail four
functions first treated extensively by Jacobi. Thcse are the four
theta functions:
OJ
(I)
8
1
(z, q) = 2 L.: (_l)nq<n+!)' sin(2n + I)z,
n=O
OJ
(2) 82(Z, q) = 2 L.: q(n+!)' cos(2n + I)z,
n=O
00
(3) 8
3
(z, q) 1 + 2 L.: qn' cos 2nz,
71=1
en
(4) 8
4
(z, q) = 1 + 2 L.: (-Irqn' cos 2nz.
It seems safe to say that no topic in mathematics is more replete
with beautiful formulas than that on which we now embark.
In the functions defined by equations (1)-(4), we ordinarily think
of each 8 as a function of z, with q playing the role of a parameter.
When only the z is to be emphasized, the q will be suppressed, and
we shall write
(5) 8
i
(z) = e,(z, q), i = 1,2,3,4.
In the definitions (1)-(4) we need only to require that I q I < 1
to get absolute conycrgencc for all finite z. At times the role of the
314
165] ELEMENTARY PROPERTIES 315
parameter q becomes important, and we use the additional notation
(6)
The requirment I q I < 1 implies that the coefficient in the imag-
inary part of T be positive, Im( T) > 0. When the dependence of
Oi(Z,q) upon T is to be emphasized, we write Oi(Z) = OJZIT) for
i = 1, 2, 3, 4.
One last set of notations is needed. For i = 2, 3, 4, the value of
Oi(Z, q) at Z = plays an important role, as does the value of
d
-d-- 01(Z, q) nt z = o. For these funetions of q alone we use the
_z
simplified notations
(7) Oi = 0;(0, q), i = 2,3,4,
(8) 01' = [dd OI(Z, q)] .
z Z ~ O
165. Elementary properties. Some properties of the theta func-
tions follow at once from the definitions (1)-(4) of the preceding
section. The function OI(Z) is an odd function of z; the functions
02(Z), 03(Z), and 04(Z) are all even functions of z.
Since
sin[(2n + l)(z + h)] = cos(2n + 1)z sin[H2n + 1)7I"J
= (-1)ncos(2n + l)z,
cos[(2n+ l)(z + h)] = -sin(2n + l)z sin[H2n + 1)7I"J
= (-1)n+lsin(2n + l)z,
and
cos[2n(z + h)] = cos 2nz cos n7l" = (-l)ncos 2nz,
it follows readily that
(1) 01(Z + !71") = 02(Z),
(2) 02(Z + h) - 01(Z),
(3) 03(Z + h) 04(Z),
(4) 04(Z + !71") 03(Z).
From (1) and (2) above we obtain
01(Z + 71") = 02(Z + h) = -OI(Z),
and we perform similar operations on the other theta functions to
316 THETA FUNCTIONS [Ch.20
arrive at the results
(5)
(6)
(7)
(8)
8
1
(z + 11") = - 8
1
(z),
8
2
(z + 11") = - 8
2
(z),
8
a
(z + 11") = 8a(Z),
8
4
(Z + 11") = 8
4
(Z).
We now know that 8
a
(Z) and 8
4
(Z) have the common period 11" and
that 8
t
(z) and 8
2
(z) have the common period 211". This is, of
a reflection of the periodicity of the sine and cosine functiolls. But
the exponential function is also periodic,
(9) exp(u + 27ri) = exp(u),
and the theta functions involve an exponential in that q = exp(7riT).
Hence we are led to investigate the possibility of another period for
the theta functions. It turns out that these functions are not
doubly periodic but that they miss that property by so small a
margin (a multiplicative factor) that it is a simple matter to use
the theta functions to construct functions which are doubly periodic.
166. The basic property table. The discussion in the last para-
graph of the preceding section leads us to investigate for each 8
i
(Z)
the value of IJi(z + !n).
We defined the theta functions by the equations
en
(1) 8
1
(z) = 2 L ( -1) nq(n+!l' sin(2n + l)z,
n_O
en
(2) 8
2
(Z) = 2 L q("+!l' cos(2n + l)z,
n_O
co
(3) 1 + 2 L q"' cos 2nz,
n-I
co
(4) 8
4
(z) = 1 + 2 L ( -Irq"' cos 2nz.
71=1
Let us now put the trigonometric functions involved into exponen-
tial form. We may write, for instance,
en
8
1
(Z) = i-I L (_lfq(n+!1'[e(2n+!)<z _ e-(2
n
+1
1
iz]
"" 0
en en
-i L (_lrq(n+!1'e(2n+
1l
iz + i L (_I)n
q
(n+!1'e-(
2n
+lliz.
n=O n=O
166] BASIC PROPERTY TABLE 317
In the last summation, replace n by (-n - 1) to obtain
m
8
1
(z) = -i L (_1)"q(n+!)'e(2n+I)>z + i L (_It+
1
q(-n-
l
)'e(2n+l)i,,
n=O n= --1
or
'"
(5) ( )
" (-1 )nq(n+
ll
' ". (2
n
+l)iz.
81 Z = -2 L.. '
The same procedure yields also, and even more easily,
(6)
(7)
(8)
00
8 ( )
- " (n+!l' (2n+l)i.
zZ-L..q e ,
rt=-{X)
'"
8 ( )
- '"' n' 2niz
sZ-L..qe,
n=-co
en
8
4
(z) = L (_lrqn'e
2ni
'.
n=-a)
We are now in a position to obtain 8
i
(Z + !n) for l' = 1, 2, 3, 4.
Since e"ir = q, it fo11O\vs that
exp[(2n + l)i(z + !n)] = exp[(2n + l)iz] cxp[(n + Din]
or
(9) exp[(2n + l)i(z + !7rT)] = e(Zn+!)izqn+!.
We may now write from (5) and (9),
co
8
1
(z + !n) = -i L (_lfq"'+n+l
e
(Zn+l)iz
q
n+
i
n=-'X)
co
. -I '"' ( l)n (n+I)' (2,,+l)i,
= -2q L.. - q e .
Tt=-=
In the last summation replace n by (n - 1) to obtain
m
8
1
(z + !7rT) = _iq-l L (_1)"-l
q
n'e(2lL-l)iZ
n=-ro
=
. -I -IZ " ( l)n n' 2niz
= 2q C L.. - q e
n=-co
We have thus deriyed the result
(10)
318 THETA FUNCTIONS [Ck.20
Equations (6) and (9) yield
en
02(Z + !1rT) = L qn'+n+i
e
(2n+l)iZ
q
n+!
n=-c:o
co
-I '" (n+1)' (2n+l)i&
= q ~ q e
n--oo
,<,
- i ""' nl (2n -l)iz
= q ~ q e ,
na::::-t'l)
from which we get
(11)
Since exp[2ni(z + ~ 7 r T ) J = e
2niz
qn, we may obtain from (7)
m
(
+
') '" n' 2niz n l '" Q(n+l)'e2niZ,
Oa Z 21fT = ~ q e q = q ~
n=-co n=-oo
so that we find
(12)
In the same way we derive
(13)
We now use equations (10)-(13) to obtain the functions of
argument (z + 7rT). Using first (10) and then (13), we obtain
01(Z + 7rT) = iq-
1
e-
i
(z+!u)8
4
(z + ~ 7 r T )
= iq- ie-izq-!iq-le-iz81(Z),
which yields
(14)
From (11) and (12) it follows that
or
(15)
02(Z + 7rT) = q-
1
e-
i
(z+!U)8
a
(z + hT)
= q- ie-izq-! q-
1
e-
iz
02(z),
In the same manner we obtain
(16)
(17)
Oa(Z + 7rT) = q-le-2iz8a(z),
8
4
(Z + 7rT) = -q-
1
e-
2iz
04(Z),
167] LOCATION OF ZEROS 319
Examination of equations (14)-(17) show that the theta functions
barely missed having another period 1f'T. Indeed certain ratios of
theta functions do have that period. We shall eventually build
doubly periodic functions from the four theta functions by means
of well chosen ratios.
The information contained in equations (1)-(8) of Section 165
and in equations (10)-(17) of this section is tabulated below. This
table is one to which we refer almmit constantly in this and the
next chapter.
Y=
z z + 4,,- z+tr z + Z + trT
OI(Y) =
OI(Z) 02(Z) - O,(z) iq-le-;'O.(z) -q-
l
e-
2
"8
1
(z)
82 (y) = 8
2
(z) -OI(Z) -02(Z) q-l
e-;'03(z) q-
l
r';z8
2
(z)
03(y) = 03(Z) O,(z) 03(Z) q-le-
iz
02 (z) q-
1
e-
2iz
03 (z)
-
O,(y) = O,(z) 03(Z) O,(z)
iq-1e- iZO, (z) - q-
1
e-
2
;zO,(z)
The basic table may be used also to depress, as well as to augment,
the argument of a theta function. For instance, since 8
4
(z) is an
even function of z,
8
4
(z - 1f'T) = 8
4
( -z + 1f'T) = -q-le2iz84( -z) = -q-
1
e'iz(}4(Z),
a result also easily obtained by reading the table in reyersc order.
We can in these ways obtain a set of results useful to us later:
8
1
(z - 1f') - I/l(Z), 8,(z - 1f') = -I/,(z),
1/
3
(z - 1f') 1/
3
(z) , 8
4
(Z - 1f') = 1/4(Z),
8
1
(z - 1f'T) -q-
1
e
2iz
l/l(Z), 8
2
(z - 1f'T)
=
8
3
(z - 1f'T) = q-
1
e
2iz
l/iz), 8
4
(z - 1f'T)
=
-q-
1
e
2
<Z1/4(z).
167. Location of zeros. The basic table makes it a simple mat-
ter to obtain the zeros of the theta functions. Since
OJ
(1) 8
1
(z) = 2 L (-l)nq(n
H
)'sin(2n + I)z,
n=O
we see that 8
1
(0) = 0; Z = 0 is a zero of 8rCz). Again from the
table it follows that
8
2
(!1f') = 8,(0 + !1f') = -8
1
(0) = 0,
so that z = !1f' is a zero of 8
2
(Z). Since
320
THETA FUNCTIONS [Ch.20
we find that
8
3
071' + ~ 7 r T ) = q-
l
e-!
iK
8
2
(h) = 0;
Z = !1r + 7 r T is a zero of 8
3
(z). Finally, from
it follows that
z-plane
8
1
(z + ~ 7 r T ) = iq--1e ;'8
1
(z)
Z = ~ 7 r T is a zero of 01(Z).
The zeros obtained so far are
exhibited in Figure 10. In the
figure an encircled number in-
dicates \vhich theta function pos-
se:s:ses that zero. For instance,
@ ncar ~ 7 r + !7rT is a reminder
that z = ~ 7 r + ~ 7 r T is a zero of
03(Z). Note the counter e!ock-
wise order of the zeros in the
figure.
The fourth and sixth columns
Figure 10
of the basic table, page 319, show
that, if we add 71' or 7rT to (or sub-
tract 71' or 7rT from) a zero of any theta function, we arri,-e again
at a zero of that function. So far ,ye have shown that if m and n
are any integers,
) 01(Z) = 0
at z = m7r + n7rT,
8
2
(z) = 0 at Z = ~ 7 r + m7r + n7rT,
A
(83 (Z) = 0
at
!7r + ~ 7 r T + m7r + n7rT, Z =
8
4
(Z) = 0 at Z = 17rT + m7r + n7rT.
Our next task is to show that these are the only zeros of the theta
functions.
Directly from the basic table we obtain
(2) k = 1, 2, 3, 4,
in which the minus sign is to be used for k = 1,2, and the plus sign
for k = 3, 4. Logarithmic differentiation of the members of (2)
yields
167] LOCATION OF ZEROS
321
(3)
Ok'(Z + 71") o/(z)
Ok(Z + 71") = Ok(Z);
k = 1,2,3,4.
Note that the relation (3) is the same for all pertinent k values.
The basic table also yields
(4) k = 1, 2, 3, 4,
in which the plus sign applies for k = 2, 3, and the minus sign for
k = 1, 4. Logarithmic differentiation of the members of (4) gives
(5) O/? : 7rT} -2i + 0/(0)2; k = 1, 2, 3, 4.
Ok Z 71"T Ok z
Let t be any point in-
side the parallelogram of
Figure 10. Consider the
parallelogram P shown
in Figure 11, that with
vertices at t, t + 71", t + 71"
+ 7rT, t + 71"T. Let Nk
be the number of zeros of
Ok(Z) inside P. Then,
since Ok(Z) has no poles,
(6) 2
'AT _ r o/(z) dz
71"Z k-J
p
Ok(Z) ,
or
(7)
z-plane
o
Figure 11
+ fl+.-< o/(z) dz + fl o/(z) dz
I+r+rr (h(z) I+n Ok(Z)
In the second integral in (7) replace Z by (z + 71"); in the third
integral in (7) replace Z by (z + 7rT). The result is
(8)
271"iNk = ft+ro/(z) dz + jt+rTO/(z + 71") dz
I Ok(Z) 1 Ok(Z + 71")
+ It O/(z + 7rT) dz + It O/(z) dz
t+r Ok(Z + 7rT) t+rT olif-
Next use the relation (3) on the second integrand in (8) and the
relation (5) on the third integrand in (8) to obtain
322
(9)
THETA FUNCTIONS [Ch.20
27riN k = It+ .. O/(Z) dz + It+r<O/(Z) dz
t Ok(Z) I Ok(Z)
+ It [-2i + O/(Z)J dz + I' dz.
'+r Ok(Z) t+r< Ok(Z)
From (9) it follows that
27riN k = It -2idz = 27ri,
t+r
so that Nk = 1 for each k = 1, 2, 3, 4.
We have thus proved that the zeros tabulated in A are the only
zeros of the theta functions and that earh zero is a simple zero.
168. Relations among squares of theta functions. We know
that the theta functions have no singularities in the finite plane,
and we know the location of all zeros of these functions. Let us
employ this knowledge together with the properties exhibited in the
basic table to obtain relations among the theta functions.
The technique is simple. Vsing the information described in the
preceding paragraph we construct an elliptic function of order less
than two. The function is therefore (Theorem 90, page 309) a
constant, and we thus obtain a relation among the component parts
used to form the degenerate elliptic function.
Suppose, for instance, that we wish to obtain a relation il1\-olying
012(Z). We set up the function
(2)
(3)
z-plane
(1)
t+1CT
f+:r+n which \ve shall force to be
r-------------------,
t+JC
Figure 12
Oj2(Z+7r) = 0\2(Z),
an elliptic function of
order less than two with
period parallelogram C,
as shown in Figure 12.
The denominator on the
right in (1) has a double
zero at z = 0, and that is
its only zero inside C.
Furthermore, the basic
table, page 319, yields
012(Z + 7rT) =
168] SQUARES OF THETA FUNCTIONS 323
We therefore choose the numerator N(z) in (1) so that N(z) will
have at least a single zero at z = 0 and also possess the properties
(4)
(5)
N(z + 7r) = N(z),
N(z + 7rT) = q-2
e
-4izN(z).
Then cp(z) will have the two periods 7r and 7rT and will have at most
a simple pole inside the cell C. Thus cp(z) will be an elliptic funC"tion
of order less than two and will be constant.
The basic table shows us that the square of any theta function
will havc the properties (4) and (5). We form for N(z) a lincar
combination such as
A022(z) + B03
2
(Z).
Recall the notation Ok = Ok(O) for k = 2, 3, 4. If we choose
(6)
surely N(z) will yanish at z = O. Furthermore, N(z) will ha\'c the
properties (4) and (5), as we see from the basic table. Hence
(7)
a constant, since cp(z) is an elliptic function of order less than two.
To evaluate the constant h \ve use any convenient z. Choose
z = !7r. Then from the basic table, page 310,
Hence
1:/r
2
(!7r) = 0
2
2
(0) = 0
2
2
,
02207r) = 0,
8
a
2
( t7r) = 0
4
2
(0) = 0
4
2

With this value for h, equation (7) may be rewritten as
(8)
Additional relations are
(9)
(10)
(11)
8
a
2
0
1
2
(Z) + 8
4
2
8
2
2
(Z) = 02
20
4
2
(Z),
(}22012(Z) + 04
2
(}a
2
(Z) = 8
3
2
8
4
2
(Z),
0
2
Z
8
2
2
(Z) + 0
4
2
8
4
2
(Z) = 03
20
3
2
(Z),
324 THETA FUNCTIONS [Ch.20
each of which can be discovered by the process used to get (8). \Ye
leave (9) and (10) as exercises and now prove the validity of (11).
Consider the function
(
12) ( ) _ - 84
2
('h)83
2
(Z)
<PI Z - 822(Z) .
The basic table yields at once
In the cell C of Figure 12, the function <Pl(Z) can have no singular
point except possibly at Z = the zero of 8
2
(z). At Z = h, the
denominator in (12) has a double zero and numerator at least a
single zero. Renee <Pl(Z) is an elliptic funetion of order less than
two, and it must be constant. But
<pIOn) = .
82
2
( '27l"T)
From the basie table, page 319, we get
8
3
2
( hT) 83
2
(0 + !n) = q-
t
e
o
8
2
2
(0) = q-
t
8
2
2
,
8
2
2
( hT) 82
2
(0 + !n)
= q- teo8
3
2
(0) = q-
i
8
3
\
8
4
2
(!7l" ) 8
3
2
(0) 8
3
2
,
8
3
2
0 7l") 8
4
2
(0) 8
4
2

Therefore the constant value of <Pl(Z) is
1) - 8a
2
q-
t
82
2
<PI(Z) = <PI(2n = = -82
2
and (12) now yields
8
4
2
8
4
2
(Z) - 8
a
2
8
3
2
(Z)
82
2
(Z)
which is merely a rearrangement of (11).
From (11) we obtain by putting Z = 0 the identity
(13)
Now 8
2
, 8
3
, and 0
4
are functions of the parameter q. From the
original definitions of the theta functions we obtain
169] PSEUDO ADDITION THEOREMS 325
co
(J4 = 1 + 2 L ( - 1 r qn',
n=l
IT;
(J3 = 1 + 2 L q"',
Tl...<:l
so that (13) may bc interpreted as a cOlnpad form of the statement
that, for !qi < I,
169. Pseudo addition theorems. Wc shall now deriyc formulas
inYoh'ing Ok(X + y) and (h(x - y) for k = 1,2,3,4, The technique
used will be much the same as that employed in getting relations
among the squares of the theta functions.
Suppose, for instance, that we \vish to construct a formula for
Ol(X + y) Ol(X - V). We use that product as the denominator of
a fraction which \ve force to become an elliptic function of order
less than two when considered as a function of y with x fixed. It is
desirable first to determine the periodicity properties and the zeros
of the product. Let
Dll(y) = (Jl(X + y)(Jl(X - V).
Then, from the basic table, page 319, we find that
Dll(y + 7l') Ol(X + y + 7l')(Jl(X - y - 7l') = Ol(X + y)Ol(X - v),
or
(1)
Also
Dll (y + n)
or
(2)
Dll(y + 7l') = Dll(Y)'
Ol(X + y + n)OJX - y - n)
-q-
1
e-
2
i(x+
y
lO
t
(x + y)( - V),
Therefore our denominator DttCy) has periodicity factors unity
for period 7r and q-2e-4i
y
for quasi-period n. The square of any
indi\'idual theta function of y will also haye thosc periodicity factors.
vVe may then choose as our numerator such a combination as
(3)
326 THETA FUNCTIONS [Ch.20
and thus obtain a fraction \vith periods tr and trT. The A and B in
(3) may be any functions of x.
We need next to locate the zeros of our denominator Dll(y) and
knock out one of them by an appropriate choice of A and B. Since
Dll(y) = OJ(x + y)OI(X - y),
we observe that Dll(y) has simple zeros at x+y=O and at x-y=O.
Let us force the numerator (3) to yanish at y = x by choosing
A = 02
2
(X), B = - OI2(X). Then the fraction
022(X)O}2(y) - 08X)022(y)
OI(X + y)Ol(X - y)
(4)
is doubly periodic with periods tr and trT and has in a cell no singular
points except possibly a simple pole at y = -x. (Actually the pole
at y = -x was accidentally remoyed along with that at y = x.)
Then (4) is an elliptic function of order less than two and is therefore
constant,
(5)
02
2
(X)OI
2
(y) - OI
2
(X)02
2
(y) = C
OI(X + y)OI(X - y) .
The constant C might depend upon x, of course. To evaluate C,
let us put y = 0 and thus obtain
Insertion of the yalue of C into (5) leads us to the desired relation,
We may move from (6) to other expressions for the product
OI(X + y)OI(X - y) by using on the right the various identities in-
volving squares of theta functions, equations (8), (9), (10), (11) of
Section 168. As an example, the known identity
+ =
used first with argument x, then with argument y, permits us to
rewrite (6) as
8428228I(X + y)OI(X - y)
822(y)[022032(X) - 8
3
2
8
2
2
(X)] - 822(X)[822032(y) - 8
3
2
0
Z
2
(y)]
= 822822(y)832(X) - 822022(X)032(y).
169] PSEUDO ADDITION THEOREMS
327
We thus obtain
(7) 84
2
81(x + y)8
1
(x - y) = 8
2
2
(y)8
a
2
(x) - 8
2
2
(X)8
a
2
(y),
a result also easily discovered by the method we used to get (6) in
the first place.
The reader can easily show that for each k = 1, 2, 3, 4, the product
Dkk(y) = 8
k
(x + y)8k(X - y)
has the properties
Dkk(y + 7r) = Dkk(y),
The derivation of the consequent formulas for the various products
Dkk(y) is left for exercises at the end of this chapter.
Let us next obtain one formula in which two different theta func-
tions, one of argument (x + y) and one of argument (x - y), arc
involved. Let
D
l4
(y) = 8
1
(x + y)e
4
(x - y).
Then the basic table, page 319, yields the results
D
l4
(y + 7r) = -D
l4
(y),
D
l4
(y + 7rT) = q-2
e
-4i
u
D
I4
(y).
Now the products 8
2
(y) e
3
(y) and 8
1
(y) e
4
(y) behave in that same
way. That is,
The fraction
(8)
e
2
(y + 7r)e
3
(y + 7r) = -8
2
(y)oiy),
el(y + 7r)04(y + 7r) = -e
1
(y)e
4
(y),
8
2
(y + 7rT) ea(y + 7rT) = q-2e-4 iue2( y) ea(y),
8
1
(y + 7rT)04(Y + 7rT) = q-
2
e-
4iu
OI(y)04(Y).
Ao2(y)oa(Y) + Bo
I
(y)04(Y)
BI(x + y)e
4
(x - y)
has periods 7r and 7rT in its argument y, as long as A and B are in-
dependent of y. The denominator in (8) has simple zeros at
x + y = 0 and at x - y = By choosing A = OI(X)04(X) and
B = 02(X)Oa(X), we make the numerator of (8) vanish at y = -x.
Then
(9)
.OI(X)04(X)02(y)03(Y) + 02(X)Oa(x)e
l
(y)o4(y) = C
+ - y) ,
328 THETA FUNCTIO"VS [Ch.20
where C is independent of y because the left member of (9) is an
elliptic function of order less than two. The yalue of C in (9) may
be found by using y = 0:
C = = 020
3

01(X)04(X)
We have thus discoyererl the identity
(10) 0
2
0
3
0
1
(x+y) 04(X - y) = 0
1
(x) 8
2
(y) 03(y) 01(X) + 0
1
(y) 02(X) ol x) 04(y).
Similar results for other products Ok(X + y)Oh(X - y) \\ill he found
in the exercises.
An important special case of (10) is obtained by ehoosing y = x.
The result is
(11)
which may be called a duplication formula for OI(X), Similar
results for the other theta functions appear in the exercises.
170. Relation to the heat equation. The simple (one dimen-
sional) heat equation, or equation of diffusion,
(1)
appears in many phases of applied mathematics. The four theta
functions, , ... hen considered as functions of z and T, satisfy an equa-
tion of the form (1).
At the start of this chapter we defined the theta functions by
senes. As an example, consider
m
(2) 04(Z, q) = 1 + 2 L ( _l(qn, cos 2nz.
With q = exp(1riT), we write (2) in the form
co
(3) 04(Z IT) = 1 + 2 L ( -1) n exp(n21riT) cos 2nz.
n=l
From (3) we obtain
and
329
Hence
(4)
Indeed, examination of the series definitions of the theta functions
shows that equation (4) is still valid when the subscript 4 is replaced
by 1, 2, or 3. That is,
a 1 . a
2
(;: ar8k(z/r) = k = 1,2,3,4.
171. The relation 0'1 = 0,0
3
0
4
, The values 8
3
, 8
4
of three
theta functions at z = 0 haye already entered our work in many
plaees. We define 8
1
' as the value of o/(z) at z = O. The four
functions of q alone, 0
1
', O
2
, 03, 04 are connected by the remarkably
neat relation
(1)
which we now proceed to prove.
Equation (11) of Section 169 gives us
(2)
In (2) take logarithms of each member and differentiate throughout
with respect to z to obtain
(3) = + t
01(2z) 01(Z) Ok(Z)
Differentiate each member of (3) with respect to z, thus arriving at
40/'(2z) _ = 8/'(z) _
8
1
(2z) 0
1
(2z) 8
1
(z) 01(Z)
+ t [O/'(Z) _
k=2 8k(Z) Ok(Z)'
or
(4)
= t -
We wish to let z -> 0 in (4). The functions 02(Z), 03(Z), 8
4
(z) are
even functions of z; the function 8
1
(z) is an odd function of z. Thus
330 THETA FUNCTIONS [Ch.20
the limits of the terms on the right in (4) are easily found, but the
terms on the left yield indeterminate forms. We shall attempt to
simplify the proof by employing three lemmas.
Lemma 13.
Lemma 14.
. O/'(z) 0/"(0)
Lun ---- = --.
.-+0 Ol(Z) 0/(0)
Lemma 15.
Lim _ {f!!J. Z)}2] = 20
1
"'(0) .
-+0 OI(2z) Ot(z) 0
1
'(0)
Lemma 13 follows at once from equation (2) of this section.
Lemma 14 is a consequence of a single application of L'Hopital's
rule.
To obtain the result in Lemma 15, note first that
where
4{f!!J2z)}2 _ =
01(2z) OJ(z) 012(Z)OI2(2z)
= ABC,
A = OJ(z)
01(2z)'
B = 201(z)o/(2z) + OJ(2Z)OI'(Z)
OJ(2z) ,
c = 201(z)8j '(2z) - OJ(2z)o/(z).
Oj3(Z)
By Lemma 13, Lim A =!. Using Lemma 13 again we obtain
.-+0
Lim B = 2 + OJ'(O) = 28/(0) .
-+0
Finally, with the aid of L'Hopital's rule, we get
L
C L' 401(Z)OI"(2z) - OI(2z)0/'(z)
lIn = 1m
.+0 z-+o 30j
2
(z)01'(Z)
171] THE RELATION 0'1 = OZ0304
331
Hence, employing Lemmas 13 and 14, we find that
Lim C = __ l_ . 2 - 2 . OI'''CQ2] = .
z+o 30/(0) 0/(0) 0/(0) [0
1
'(0)]2
Since the left member in Lemma 15 is the product of the limits of
A, B, and C, it equals
1 ,20/"(0) 20
1
"'(0)
"2 20
1
(0)[0
1
'(0)]2 = -0
1
'(0) ,
as stated in the lemma.
We return now to equation (4); let z ---> 0 and use Lemmas 14
and 15 to obtain
40/"(0) 0
1
'"(0) 20
1
"'(0) 4 0/'(0)
0(0) - 0
1
'(0) - 0
1
'(0) = 6 th(O) ,
or
(5)
0
1
"'(0) 4 0/'(0)

0
1
'(0) kd Ok(O)
In Section 170 we found that
(6) IT) = IT)
aT 4 az
2
,
k = 1,2,3,4.
From (6) we also get
(7) I T) = - hi}3301(Z IT).
aT oZ az
Because of (6) with k = 2, 3, 4, and of (7), equation (5) may be
put in the form
(8)
N ow integration with respect to T leads us to the result
Or'(OIT) = Eo2 (0IT)Oa(0IT)o4(0IT),
or
(9)
in which O
2
, 0
3
, 0
4
, and or' are functions of q = exp(7riT), but E is
constant. In (9) insert a factor q-! on each side to write
q-lor' = E(q-
l
o2)o304,
332 THETA FUNCTIOXS [Ch.20
and then let q - O. From the definitions of the theta functions
we see that
Lim (q-
1
8
2
) = 2
q+O
and
Lim 83 = Lim 8. = 1.
q+O q+O
Hence E = 1, which concludes the proof that
(1)
Equation (1) is interesting in itself, but it will also prove of value
to us in future developments.
172. Infinite products. Since we know all the zeros of the
theta functions, it is natural to seek infinite product representations
for those functions. From anyone such representation the others
will follow by using properties from the basic table, page 319. It
turns out that either 83(Z) or 8.(Z) furnishes a simple starting point.
Consider o.(z). Its zeros are at
for integral nand m. We wish to form an infinite product which
vanishes for
(1)
for every integer rl. ~ i n c e we wish our product index to run from
1 to co, we separate the zeros in (1) into
(2) and
for n integral, n ~ 1. With these zeros in mind we form the
product
ro
(3) fez) = II [(1 - q2n-
1
e
2iz
)(1 - q2n-
1
e-
2iz
)],
n=l
which is absolutely convergent for I q I < 1. The function fez) has
exactly the zeros possessed by 8
4
(z).
In order to employ our customary technique, we recall that o.(z)
has the periodicity properties
(4)
(5)
04(Z + 7r) = 8
4
(z),
04(Z + 7rT) = -q-
J
e-""04(Z),
172] INFINITE PRODUCTS 333
and seek corresponding information about fez).
Since e
2i
(z+;r) = e
2iz
, we get
(6) fez + 11") = fez).
Now
'"
fez + n) = II [(1 - q2n-lc2iH2i"')(1 - q2n--lc-2iz-2in)]
n ,...1
".
= II [(1 - q2n+IC
2iz
)(1 - q2n-
3
e-
2iz
)].
n,;;;l
Hence
__ . qe
2iz
- 1
= q Ie 2lZ------.-j(z)
1 - qe
2tz
,
so that
(7)
fez + 1I"T) = -q-1e-Zizf(z).
Therefore (J4(z)/f(z) is an elliptic function of order less than two
and must be constant. It is customary to call the constant G. It
is a function of q but not of z. We have shown that
'"
(8) (J4(Z) = G II [(1 - qZn-
1
e
2iz
)(1 - q2n-I
e
-2iz)],
n=l
which may also be written
co
(9) (J4(Z) = G II (1 - 2q2n-l cos 2z + q4n-2).
n=l
The value of G will be obtained in the next section.
We now wish to find infinite products for the other theta functions.
From the basic table, page 319, we have
(J3(Z) = (J4(Z + !1T),
so that (9) leads to
'"
(10) (J3(Z) = G II (1 + 2q2n-l cos 2z + q4n-2).
u=l
Again from the basic table we obtain
(Jl(Z) = -iqe !iz(J4(Z + !1rT),
which, in view of (8), leads to
334 THETA FUNCTIONS [Ch.20
00 co
8
1
(z) = -iqieizG IT (1 - q2n-le2iz+i",) IT (1 - q2n-le-2iZ-i1fT)
n=! n ~ l
co co
= -iq1eizG IT (1 - q2neZiz) IT (1 - qZn-Ze-
Ziz
).
n=! n-l
A shift of index from n to (n + 1) in the last infinite prociuct yields
m ~
(JrCz) -iq1eizG II (1 - q2ne2iz) II (1 - q2ne-Ziz).
n=l n=O
Therefore
00
Ol(Z) = -iqleiz(l - e-Ziz)G II [(1 - qZne2iz)(1 - q2ne-2iz)].
n=l
Now -1"e
iz
(1 - e-
2iz
) = -i(e'
z
- e-
iz
) = 2 sin z" Hence
ro
(11) Ol(Z) = 2qiG sin Z II [(1 - q2ne2iZ)(1 - q2ne-2iZ)],
which may also be written in the form
co
(12) Ol(Z) = 2qlG sin z II (1 - 2qZn cos 2z + q4n).
n=!
The basic table yields 02(Z) = OI(Z + !7r), with the aid of which
(12) yields
co
(13) 02(Z) = 2qlG cos z IT (1 + 2q2n cos 2z + q4n).
n=!
Equations (9), (10), (12), and (13) are the dcsired infinite product
forms for the theta functions except that we have yet to determine
how G depends upon q.
173. The value of G. We shall find G from the known (page
332) relation
(1)
and the infinite products obtained in the preceding section. By
differentiating \\"ith respect to Z each member of the identity
00
Ol(Z) = 2qlG sin Z IT (1 - 2q2n cos 2z + tn)
n=1
and then using Z = 0, we arrive at
co
(2) 0/ = 2qlG II (1 - q2n)2.
71=1
173] THE VALUE OF G 335
From the identities
co
02(Z) = 2qiG cos Z II (1 + 2q2n cos 2z + q4n),
n ~ 1
00
Os(z) = G II (1 + 2q2n-1 cos 2z + q4n-2),
'I_I
'"
04(Z) = G II (1 - 2q2n-1 cos 2z + q4n-2),
n ~ 1
it follows that
CD
(3) O
2
= 2qlG II (1 + q2n)2,
11,=1
00
(4) 0
3
=
G II (1 + q2n-I)2,
n=l
=
(5) (J4 = G II (1 - q2n-I)2.
n=!
Since 0/ = 0
2
0
3
0
4
, we are led to the relation
co co r:o co
(6) II (1 - q2n)2 = G2 II (1 + q2n)2 II (1 + q2n-!)2 II (1 - q2n-!)2,
n=l n=l n=l n=l
in which each of the infinite products is absolutely convergent
because of the restriction I q I < 1.
Since the set of even positive integers plus the set of odd positive
integers is the set of all positive integers,
00 = CD
II (1 + q2n)2 II (1 + q2n-!)2 = II (1 + qn)2.
n=l n=!
Equation (6) may now be put in the form
00 00 co
II (1 - q2n)2 = G2 II (1 + qn)2 II (1 - q2n-!)2,
n=l n:a:l n=l
from which
co co co co
II (1 - q2n)4 = G2 II (1 + qn)2 II (1 - q2n-I)2 II (1 - q2n)2.
n=l n=l n=l n=l
But
00 00 00
II (1 - q2n-I)2 II (1 - q2n)2 = II (1 - qn)2.
n=l n=l n=!
Hence we have
336 THETA FUNCTIONS [Ch.20
ro CD ro
II (1 - q2n)4 = G2 II (1 + qn)2 II (1 - qn)2,
n=l n=l n=l
or
ro m
(7)
II (1 - q2n)4 = G2 II (1 - q2n)2.
n ' ~ 1
But I q I < 1, so there are no zero faetors in (7). Therefore
'"
(8) II (1 - q2n)2 = G2.
n=l
For q = 0, G = +1, as may be seen from
w
(4) OJ = () II (1 + q2n-l)2
n ~ l
and the series
00
(9) OJ = 1 + 2 L qn.
71,=1
obtained early in this ehaptcr. We may therefore eon elude from
(8) in this seetion that
co
(10) G = II (1 - q2n).
n=l
EXERCISES
1. Show that Oi' = 2q1G3.
2. The following formulas are drawn from Section 168. Derive (9) and (10):
(8) 0,20
1
2(Z) + 032022(Z) = 02203
2
(Z) ,
(9)
(10)
(11)
0
3
2
0J
2
(Z) + 0,20l(z) = O
Z
2
0,2(Z),
022012(Z) + 0
4
2
0l(z) = 0
3
2
04
2
(Z),
022022(Z) + (j,2(j4
2
(Z) = 03
2
(j3
2
(Z).
3. Use (8) and (10) of Ex. 2 and the equation O
2
' + (j44 = (j3
4
to show that
(j14(Z) + (j3
4
(Z) = 02'(Z) + (j,4(Z).
4. The first of the following relations was derived in Section 169. Obtain the
other three by using appropriate changes of variable and the basic table, page 319.
For example, change x to (x + !7r), or x to (x + !7rT), etc.
(jl(jl(X + Y)(}l(X - y) = (jJ2(X)(J2
2
(y) - Ol(X)012(y),
(j2
2
(j2(X + y)02(X - y) = (j22(X)022(y) - (j12(X)0\2(y) ,
02
2
0a(X + y)03(X - y) = Ol(X)(j2
2
(y) + 8,2(X)8\2(y),
02
2
04(X + y)(}.(x - y) = (j,2(X)022(y) + (jg2(X)012(y).
173] THE VALUE OF G 337
5. Use the identities in Ex. 2 and the relation 8, + 8. = 8
3
4
to transform the
first relation of Ex. 4 into the first relation below. Then obtain the remaining
three relations with the aid of the basic table, page 319.
8
2
'Ol(X + y)8
1
(x - y) = fJ.2(x)8
3
2
(y) - 8
3
2
(X)8,'(y),
8
2
'8b + y)Oz(x - y) = 03
2
(X)03
2
(y) - O.Z(X)O.'(y) ,
8
2
Z
0
3
(X + y)8
3
(x - y) = Oz2(x)Ol(y) + 8
1
'(x)O,'(y),
8,28,(x + y)8
4
(x - y) = Ol'(x)Ol(y) + 8,'(x)8l(Y).
6. Use equation (10) of Ex. 2 and thc first relation of Ex. 5 to obtain the first
relation below; then use the basie table to get the remaining three results.
8
3
2
8
1
(x + y)8
1
(x - y) = 8
1
'(x)O,2(y) - 03
2
(X)012(y) ,
8,'O,(x + y)8
2
(x - y) = 022(X)Ol(y) - 8l(x)012(y) ,
83'8s(x + y)8
3
(x - y) = 8,2(X)O,'(Y) + 8
1
2
(X)OI2(y) ,
8,'O,(x + y)8
4
(x - y) = 8
4
Z
(X)03
2
(y) + O,'(x)(N(y).
7. Use the first relation of Ex. 6 and identities from Ex. 2 to obtain the first
relation below. Derive the other three relations from the first one.
O,'8
1
(x + y)8
1
(x - y) = O,'(x)8
z
'(Y) - Oz'(X)O,2(y) ,
8,'8
z
(x + y)8,(x - y) = O,2(x)8,2(y) - 8
I
'(x)8,'(Y) ,
8,28,(x + y)O,(x - y) = 8
2
'(x)O,2(y) + 8,2(X)8,'(y),
8,'O,(x + y)8
4
(x - y) = OI2(X)O,'(Y) + 8l(x)8,'(y).
8. Use equation (10) of Ex. 2 and the first relation of Ex. 6 to obtain the first
relation below. Derive the others with the aid of the basic table, page 319.
8
4
2
0
I
(X + y)8
1
(x - y) = 8
1
'(x)O,'(y) - 8,'(x)81
2
(y) ,
8l8,(x + y)8
2
(x - y) = 8,'(x)O,'(y) - 8,'(x)81'(y) ,
8,'8
3
(x + y)8
3
(x - y) = O,'(x)8,'(y) - 8
2
'(x)8
I
'(y) ,
8,28,(x + y)8,(x - y) = O,'(x)8l(y) - OI'(X)OI'(Y).
9. The first relation below was derived in Section 169. Obtain the other three.
(N8b + y)8
1
(x - y) = 8l(x)O,2(y) - 8
2
'(x) 8
3
2
(y) ,
8,'O,(x + y)8
2
(x - y) = O,'(x)8,'(y) - 81'(x)03'(Y) '
8,'8
3
(x + y)O,(x - y) = 8,2(X)8
3
2
(y) - Ol'(X)O,'(y) ,
OlO.(x + y)8
4
(x - y) = O,z(x)O,z(y) - 8,2(X)Ol(y).
10. Use the method of Section 169 together with the basic table, page 319, to
derive the following results, one of which was obtained in Section 169.
O,8/ll (x + y)8,(x - y) = Ol(X)O,(X)03(y)8,(y) + 8
1
(y)Oz(y)83(x)8,(x),
8,O,OI(X + y)8
3
(x - y) = 8tCx)8,(y)03(X)8.(y) + 8
1
(y)8z(x)83(y)8,(x),
828
3
8b + y)8,(x - y) = Ol(x)8,(y)8
3
(y)8,(x) + 8
1
(y)O,(x)83(x)8,(y),
8,8,8,(x + y)8
3
(x - y) = 8,(x)8
3
(x)8,(y)8
3
(y) - 8
1
(x)8,(x)8
1
(y)8,(y),
8,B.8,(x + y)8,(x - y) = 8,(x)8,(x)O,(y)8.(y) - 8tCx)03(X)OI(y)83(y),
O,B.8,(x + y)O,(x - y) = 03(x)8,(x)8
3
(y)8,(y) - 8
1
(x)O,(X)Ol(Y)O,(y).
338 THETA FUNCTIONS [CIl.20
11. Use the results in Exs. 4-10 above to show that
02
S
02(2x) = 02'(X) - Ol'(X) = Os'(x) - 04'(X),
OsSOs(2x) = Ol'(X) + Os'(x) = 02'(X) + 04'(X) ,
ON,(2x) = O,4(X) - Ol4(X) = Os'(x) - 02'(X),
Ol0202(2x) = Oz2(X)OS2(X) - 012(X)O,2(X) ,
O/0202(2x) = 022(X)042(X) - O,2(X)OS2(X) ,
OlOsOs(2x) = 02'(x)Ol(x) + O,2(X)O,2(X),
O,20s0s(2x) = Os"(x)Ol(x) - O,Z(x)Oi(x),
0220,O,(2x) = O,2(:r)Os"(x) + Ol(X)O.,2(X).
O}0404(2x) = O,2(x)Ol(x) + Ol(X)04
Z
(X) 0
12. Use the method of Section J GS to show that
o ( Z2) (z I-I)
8
s
(zlr) = (-zr)-lexp -c Os - .0.- 0
7rtT .T T
From the above identity obtain corresponding ones for the other theta functions
with the aid of the basic table, page 319.
CHAPTER 21
Jacobian
Elliptic
Functions
174. A differential equation involving theta functions. The
basic table, page 319, shows that the ratio
8j (z)
8
4
(z)
has the periodicity factors minus one for period 7r and plus one for
period 7rT. Then
(1)
d 8j (z)
dz 8
4
(z)
has the same periodicity factors. Again from the table it can be
seen that
(2)
82(Z) 8
3
(z)
8
4
(z) 8
4
(z)
has the periodicity factors minus one for period 7r and plus one for
period 7rT. Therefore the function
(3)
8
4
2
(Z) d 8j (z)
cp(z) = ~ ( Z ) e ; ( z ) dz 8
4
(z) ,
the ratio of the functions in (1) and (2), has 7r and 7rT as periods.
Actually, t7rT is a period of cp(z), a fact which we now proceed to
prove.
From (3) we get
339
340
(4)
JACOBIAN ELLIPTIC FUNCTIONS [Ch.21
cp(z)
04(Z)0I'(Z) - 01(Z)0/(Z)
02(Z)03(Z)
The basic table yiclds
(5)
and
(6)
From (5) and (6) it follows that
(7)
(8)
Hence
or'(z + = lq-le-i.[o/(z) - iO.(z)J,
o/(z + !7fT) = iq-le--i'[Or'(z) - io
1
(z)].
o.(z + !7fT)or'(z + hT) - 01(Z + + !7rT)
= -q-
l
e-
2i
'[01(Z)0/(z) - O.(z)or'(z)J.
The basic table gh-cs us
02(Z + !7fT)03(Z + !7fT) = q-
l
e-
2iz
()2(z)03(z),
and we may therefore concludc that
(9) cp(z + !7fT) = cp(z).
The function cp(z) of (3) or (4) is now known to have the periods
7r and !7fT. The only singular points which cp(z) can have in the
finite plane are the zeros of 02(Z) and 03(Z). As wc saw on page 322,
the zeros of ()z(z) and ()3(Z) are all simple ones and are located at
z = !7r + m7r + n7rT,
z = !7r + !7rT + m7r + n7rT,
for integral m and n. Of these zeros only one is containcd in a ccIl
of Hence cp(z) is an elliptic function of ordcr less than two
and is therefore a constant.
From
( )
_ ()4(Z)0r'(z) - ()1(Z)O/(Z) _
cp z - 02(Z)03(Z) - c,
we obtain c by using z = 0:
04()r' - 0 04020
3
0
4
2
C = --- = --- = O.
02()3 ()203 '
with the aid of the knO\yn rclation or' = 0
2
0
3
0
4
. Thus we arrive at
174] A DIFFERENTIAL EQUATION
the differential equation
d 01(Z)
(10) dz 04(Z)
0
4
2
02(Z)OS(Z)
04
2
(Z)
341
The equation (10) involves all four of the theta functions. We
have previously (Section 168) found the relations
(11)
(12)
0
3
2
012(Z) + 0
4
2
02
2
(Z) = 02
20
4
2
(Z),
(N012(Z) + 0
4
2
03
2
(Z) = Oi042(Z).
With the aid of (11) and (12) equation (10) can he written in a form
involving only the dependent variable oJ(Z)/04(Z), To that end we
square both sides of (10) and write
(13)
[
!! 01(Z)]2= 04
2
022(Z) 04
2
0S
2
(Z).
dz 04(Z) 04
2
(Z) 04
2
(Z)
Equations (11) and (12) may now be used to put (13) in the form
(14)
[
!i 01(Z)]2 = [022 _ Oi ~ ~ ( Z ) ] [OS2 _ 022 ~ 2 ( Z ) ] .
dz 04(Z) 04
2
(Z) 04
2
(Z)
With the dependent variable w = 01(Z)/04(Z) we now have
(15) ( ~ y = (02
2
- OS2W
2
) (OS2 - 022W
2
).
In an attempt to simplify the appearance of equation (15), let
us introduce a new independent variable u by Z = u/Os
2
and a new
dependent variable y by w = 02Y / Os. Then (15) becomes
or
(16)
in which
(17)
In retrospect, we have shown that the function
(18) y = Os 01(OS-2U)
0204(OS-2U)
342 JACOBIAN ELLIPTIC FUNCTIONS [Ch.21
is one solution of the differential equation (16) with the constant
k being defined by (17).
175. The function sn(u). The y of equation (18) of the preced-
ing section is called the Jacobian elliptic function sn(u), defined by
(1)
and also written
(2) sn(u, k)
sn(u)
(h (M 0
3
-Zu)
020/03-
2U
)
= (h .
O2 01(03
2U
)'
when it is desired to exhibit the parameter or "modulus" k. 'Ve
know that y = sn(u, k) is a solution of the differential equation
(16) of the preceding section. We next prove that sn(u, k) is an
elliptic function, and we shall study some of its properties.
The basic table, page 3Hl, shows that the ratio 01(Z)/04(Z) has the
period 7fT in z and the periodicity factor minus one for an increase
of 7r in the argument z. Hence, as a function of z, Ol(Z)/04(Z) has
periods 27r and 7rT. Therefore, as a function of u,
0
1
(03-
2
U)
04(03-
2U
)
has periods 27r03
2
and 7rT032. That is,
(3)
(4)
sn(u + 27r(3
2
) = sn(u),
sn(u + 7fT(3
2
) = sn(u),
so that sn(u) is a doubly periodic function.
Let us now examine the singular points of sn(u). Since o,(z) and
04(Z) are analytic for all finite z, the only singular points of sn(u)
are at the zeros of the denominator 04(03-2U). Hence sn(u) has
simple poles where
Because of the periods 27r03
2
and 7rT03
2
, \ye see that a representative
cell for sn(u) has in it two simple poles, one at
and one at
u = 7r)03
2

Thus sn(u) is an elliptic function of order two, and in contrast to
176] THE FUNCTIONS cn(u) and dn(u) 343
the Weierstrass P(z) of Chapter 19, sn(u) has two simple poles
rather than one double pole in each cell.
A common notation for the periods of sn(u, k) is 4K and 2iK',
where
(5)
(6)
K = !7r8
3
2
,
K' = - i1:7rT032.
With reference to the notation in (6), recall that we require Im( T) > 0
so that Iql = le"i'l < 1. If T is pure imaginary, then K' is
positiYe.
One more notation will simplify some of the relations to be en-
countered later in this chupter. We have already defined in asso-
ciation \"ith sn(u) the modulus
(7)
We now define a complementary modulus k' by
(8)
k' = 8
4
2

Oi
In view of the relation, page 324,
we may conclude that
(9)
176. The functions cn(u) and dn(u). We have one Jacobian
elliptic function sn(u) defined by
(1)
snCu) = 03I!.!i
0
3-
2U
) .
0
2
8
4
(03-
2U
)
Closely associated with sn(u) are two other functions defined by
(2) cn(u)
0
4
(h(0;j-2
u
)
0
2
0l"03-
2U
)
and
(3) dn(u)
0
4
03(03-
2U
)
0
3
04(03-
2U
)
\Ye may refer to the basic table, page 319, to see that cn(u) is
an elliptic function of order two with periods 27r032 and (7r + 7rT) 0
3
2
344 JACOBIAN ELLIPTIC FUNCTIONS [Ch.21
and with a representatiYe cell containing two simple poles, one at
u = !7I"T83
2
, one at u = (71" + !7I"T) 83
2

In the same way it can be shown that dn (u) is an elliptic func-
tion of order hvo with periods 71"832 and 271"T83
2
and with a representa-
tive cell containing two simple poles, one at H = one at
u =
In terms of the parameters J( and K' given by
(4)
the periodieity properties of the Jacobian elliptic functions sn (u),
en eu), dn (u) may be written as follows:
(5) sn(u + 4K) = sn(u + 2iK') = sn (u),
(6) cn(u + 4K) = cn(u + 2K + 2iK') = en eu),
(7) dn(u + 2K) = dn(u + 4iK') = dn (u).
From our knowledge of the properties of the theta functions, we
obtain certain elementary facts about the Jacobian elliptic functions.
Recall that 8
m
(z) is an e\'en function of Z for In = 2, 3, 4 and that
8
1
(z) is an odd function of z. It follows that
sn( -u) = -sn(u), cn( -u) = cn(u), dn( -u) = dn(u);
that is, sn (u) is an odd funetion of u and both cn (u) and dn (u)
are even functions of u. Furthermore we ha \'e
sn(O) = 0, cn(O) = 1, dn(O) = 1.
177. Relations involving squares. Information on the theta
functions acquired in Chapter 20 naturally reflects itself in infor-
mation on the Jacobian functions. We found in Section 168 that
(1)
(2)
832812(Z) + 8
4
2
8
2
2
(Z) = 8
2
2
8
4
'(z),
8
2
2
8
1
2
(Z) + 8
4
2
8
3
2
(Z) = 8
3
2
8
4
2
(Z).
The choice Z = 8
3
-
2
u permits (1) and (2) to be written as
(3)
8
3
2
888
3
-
2
u) + 842822(83-2U) _ 1
82
2
8
2
2
-
and
(4)
178] DERIVATIVE RELATIONS 345
By the definitions of sn (u), en (u), dn (u) in the preceding section,
we see that (3) and (4) yield
(5)
(6)
sn
2
(u) + cn
2
(u) = 1,
k2 sn
2
(u) + dn
2
(u) = 1,
in which k = 6
2
2
/6
3
2
, as usual.
The elimination of sn (u) from (5) and (0) yields
dn
2
(u) - k2 cn
2
(u) = 1 - k2,
or
(7)
with k = 6
2
2
/6
3
2
and k' = 6
4
2
/6
3
2
satisfying the relation
k2 + (k')2 = 1
as noted earlier.
178. Relations involving derivatives. On page 341 we obtained
the result
(1)
From the definition
(2)
it follows that
(3)
d
-- sn(u)
du
6
3
1 d 6\ (1J
3
-
2
U)
1J
2
1J
3
2 d(1J
3
-
2
U) 1J
4
(1J
3
-
2
U)
By (1), equation (3) becomes
(4)
But, by definition,
(5)
(6)
so that (4) yields
(7)
d
du sn(u) = cn(u) dn(u).
346
JACOBIAN ELLIPTIC FUNCTIONS [Ch.21
We now know how to differentiate sn (u) and we know a relation
(8) sn
2
(u) + en
2
(u) = 1
between sn (u) and en (u). Then \ve differentiate each member of
(8) to obtain
d
sn(u) en(u) dn(u) + en(u)a,u en(u) = 0,
from which it follows that
(9)
d
du en(u) = -:m(u) dn(u).
In the same manner we conclude from (7) and the known relation
(page 345)
(10) k2 sn2 (u) + dn
2
(u) = 1
that
d
k2 sn(u) en(u) dn(u) + dn(u)du dn(u) = 0,
which yields
(11)
d
du dn(u) = - k2 sn(u) en(u).
We already know from equation (lG), page 34,1, that Yl = sn (u)
is a solution of the differential equation
(12)
in which k = 0
2
2
/0
3
2
Let us obtain corresponding differential
equations satisfied by en (u) and dn (u).
Put Y2 = en (u). Then, by (9),
( ~ ~ 2 y = sn2(u) dn2(u).
But sn
2
(u) = 1 - en
2
(u) and dn
2
(u) = (k')2 + k2 en
2
(u), for
which see Section 177. It follows that Y2 = en (u) is a solution of
the differential equation
(13)
179] ADDITION THEOREMS 347
Now let Ya = dn (u). By (11) we get
( ~ ~ 3 y = k' sn2(u) cn2(u).
But k2 sn
2
(u) = 1 - dn
2
(u) and /,;2 cn
2
(u) = dn
2
(u) - (k')2.
Hence Ya = dn (u) is a solution of the differential equation
(14)
179. Addition theorems. W e have defined the Jacobian elliptic
functions by
(1)
(2)
(3)
and we have established formulas invoh'ing the theta functions with
arguments (x + y) and (x - y). We should therefore be able to
conclude something of interest about sn(u + v), cn(u + v), etc.
In Section 169 and Ex. 8, p. 3:)7, we found that
(4) (Jz{J
3
0
1
(x+y) 8,(x - y) = 8
1
(x) 8
2
(!J) 03(Y) 8
4
(x) + 8
1
(y) 8
2
(X) 8
3
(X) 8,(y),
(5) 8,20
4
(X + y)8
4
(X - y) = 8
4
2
(X)8,2(y) - 8
1
2
(X)8
1
2
(y).
Equations (4) and (5) yield
028
3
8
1
(x + y) 8
1
(x)02(y)8
3
(y)04(X) + 8
1
(y)8
2
(X)8
3
(X)0,(y)
8
4
2
O,(x + y) 8
4
2
(X)04
2
(y) - 8
1
2
(X)8
1
2
(y)
8
1(X) 82(y183 (y) + 01(Ji182(X) 03(X)
O,(x) 8
4
(Y) 8
4
(Y) 8,(y) 8
4
(x) 8
4
(X)
= ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
1 - ~ 1 ~ ( ~ 1 ~ 1 ~ 0 ! 1
0,2(X) 8
4
2
(y)
Now put x = 8
a
-
2
u and y = 8
a
-
2
v and use (1), (2), and (3) to obtain
82
2
82
2
2 -2 sn(u) cn(v) dn(v) + 2 sn(v) cn(u) dn(u)
82 (+) 8, 8
4
8
4
2
sn u v = 824
1 - - sn2(u) sn2(v)
8
3
4
348 JACOBIAN ELLIPTIC FUNCTIONS [Ch.21
from which it follows that
(6)
sn(u + v) = sn(u) cn(v) dn(v) + sn(v) cn(u) dn(u) .
1 - k2 sn2(u) sn2(v)
Equation (6) is called an addition theorem. The companion results
(7)
cn(u + v) = cn(u) cn(v) - sn(u) sn(y) dn(u) dn(v)
1 - k2 sn2(u) sn2(v)
and
(8)
dn(u + v) = - k2 sn(u) cn(v)
1 - k2 sn2(u) su
2
(v)
may be obtained in a similar manner and arc left as exercises.
EXERCISES
1. Derive the preceding addition theorem (7) by the method of Section 179.
You may use results from thc exercises at the end of Chapter 20.
2. Derive preceding (8) by the method of Section 179, with the aid of results
from the exercises at the end of Chapter 20.
3. Show that jcn
3
(x) dn(x) dx = sn(x) - i sn
3
(x) + c.
4. Show that if g(x) and hex) are any two different ones of the three functions
sn(x), cn(x), dn(x), and if m is a non-negative integer, you can perform the inte-
gration
5. Obtain the result
jsn(x) dx = Log[dn(x) - k cn(x)] + c.
6. Show that
d (2 )
_ k (2) = (1 - k)[l + k sn2(x)].
n x cn x 1 _ k sn2(x)
Bibliography
Agarwal, R. P.
1. On Bessel polynomials, Canadian Journal of Math., 6, 1954, pp. 410-415.
AI-Salam, W. A.
1. The Bessel polynomials, Duke M ath. Journal, 24, 1957, pp. 529-545.
Appell, Paul
1. Sur ies Fonctions Hypergeomf:triq1les de Plusieurs Variables. Paris: Gauthier-
Villars, 1925.
Appell, Paul, and Kampe de Feriet, J.
1. F onctions H ypergeometriques et H ypersphCriques; Polynomes d'll ermites. Paris:
Gauthier-Villars, 1926.
Bailey, W. N.
1. Generalized Hypergeometric Series. Cambridge: Cambridge Univ. Press, 1935.
2. Products of generalized hypergeometric series, Proc. London 1'>1 ath. Soc.,
series 2, 28, 1928, pp.
3. Some theorems concerning products of hypergeometric series, Proc. London
Math. Soc., series 2,38, 19:35, pp.
Barnes, E. W.
1. A transformation of generalized hypergeometric series, Quarterly Journal of
Math., 41, 1910, pp.
2. A new development of the theory of the hypergeometric function, Proc.
London Math. Soc., series 2, 6,1908, pp.
Bateman, H.
1. A generalization of the Legendre polynomial, Proc. London Math. Soc., series
2,3,1905, pp.
2. Two systems of polynomials for the solution of Laplace's integral equation,
Duke M ath. Journal, 2, 1936, pp.
3. Some properties of a certain set of polynomials, Tohoku Math. Journal, 37,
1933, pp.
4. The polynomial of Mittag-Lefler, Proc. Nat. Acad. Sci., 26, 1940, pp. 491
5. Polynomials associated with those of Lerch, M onatsheften fur Math. und
Physik, 43, 1936, pp.
6. The transformation of a Lagrangian series into a Newtonian series, Proc. Nat.
Acad. Sci., 25, 1939, pp.
7. Relations between confluent hypergeometric functions, Proc. Nat. Acad. Sci.,
17,1931, pp.
349
350 BIBLIOGRAPHY
8. An orthogonal property of the hypergeometric polynomial, Proc. Nat. Acad.
Sci., 28, 1942, pp. 374-377.
9. Note on the function FCa, b; c-n; z), Proc. Nat. Acad. Sci., 30, 1944, pp.28-30.
10. The polynomial Fn(x), Annals of Math., 3.5,193+, pp. 767-77.5.
11. The polynomial FnCx) and its relation to other functions, Annals of Math.,
38, 1937, pp. 303-310.
12. The k-function, a particular case of the confluent hypergeometric function,
Trans. Amer. Math. Soc., 33, 1931, pp. 817-831.
Bedient, P. E.
1. Polynomials related to Appell functions of two variables, Michigan thesis,
1958.
Bhonsle, B. R.
1. On a series of Rainville involving Legendre polynomials, Proc. Amer. Math.
Soc., 8,1957, pp. 1{}--14.
Boas, R. P., Jr., and Buck, R. C.
1. Polynomials defined by generating relations, Amer. Math. Monthly, 63, 1956,
pp. 626-632.
2. Polynomial Expansions of Analytic Functions. Berlin: ::;pringer, 1958.
Brafman, Fred
1. Generating functions of Jacobi and related polynomials, Proc. Amer. Math.
Soc., 2, 19.51, pp. 942-949.
2. Some generating functions for Laguerre and Hermite polynomials, Canadian
Journal of Math., 9,1957, pp. 18{}--IS7.
3. A relation between ultraspherical and Jacobi polynomial sets, Canadian
Journal of Math., 5, 19.53, pp. 301-305.
4. A set of generating functions for Bessel polynomials, Proc. Amer. Math. Soc.,
4, 1953, pp. 275--277.
5. Unusual generating functions for ultraspherical polynomials, Michigan Math.
Journal, 1, 1952, pp. 131-138.
6. On Touchard polynomials, Canadian Journal of Math., 9, 1957, pp. 191-193.
7. Series of products of Gegenbauer polynomials, Math. Zeitschr., 62, 1955, pp.
438-442.
8. A generating function for associated Legendre polynomials, Quarterly Journal
of Math., Oxford 2nd series, 8,1957, pp. SI-83.
9. An ultraspherical generating function, Pacific Journal of Math., 7, 1957, pp.
1319--1323.
Brenke, W. C.
1. On generating functions of polynomial systems, A mer. Math. Monthly, 52,
.1945, pp. 297-301.
BIBLIOGRAPHY 351
Burchnall, J. L.
1. The Bessel polynomials, Canadian Journal of Math., 3, 1951, pp. 62-68.
2. A relation between hypergeometric series, Quarterly Journal of Math., Oxford
series, 3, 1932, pp. 318-320.
Carlitz, Leonard
1. Some polynomials of Touchard connected with the Bernoulli numbers,
Canadian Journal of Math., 9, 1957, pp. l S 8 ~ ~ 1 9 1 .
2. On the Bessel polynomials, Duke Math. Journal, 24, 19.57, pp. 151-162.
3. On Jacobi polynomials, Boll. Unione Mat. Ital., series 3, 9,1956, pp.371-381.
Chaundy, T. W.
1. An extension of the hypergeometric function, Quarterly Journal of Math.,
Oxford series, 14, 1943, pp. 55-78.
2. Some hypergeometric identities, Journal London Math. Soc., 26, 1951, pp.
42-44.
Churchill, R. V.
1. Operational Mathematics, 2d ed. New York: McGraw-Hili Book Co., 1958.
2. Fourier Series and Boundary Value Problems. New York: McGraw-Hili Book
Co., 1941.
3. Introduction to Complex Variables and Applications. New York: McGraw-Hili
Book Co., 1948.
Copson, E. T.
1. Theory of Functions of a Complex Variable. London: Oxford Univ. Press, 1935.
Curzon, H. E. J.
1. On a connexion betwecn the functions of Hermitc and the functions of Legen-
dre, Proc. London Math. Soc., series 2, 12, 1913, pp. 236-259.
Dickinson, D. J., Pollak, H. 0., and Wannier, G. H.
1. On a class of polynomials orthogonal over a denumerable set, Pacific Journal
of Math., 6,1956, pp. 239-247.
Dickinson, D. J.
1. On Bessel and Lommel polynomials, Michigan thesis, 1953. (See also 2 below).
2. On Lommel and Bessel polynomials, Proc. Amer. Math. Soc., 5, 1954, pp.
946-956. This is partly a digest of Dickinson [11.
3. On certain polynomials associated with orthogonal polynomials, Boll. Unione
Mat. Ital., (3), 13, 1958, pp. 116-124.
352 BIBLIOGRAPHY
Dixon, A. C.
1. Summation of a certain series, Froc. Lond. Math. Soc., series 1,35, 1903, pp.
285-289.
Erdelyi, A.
I. With Magnus, W., Oberhettinger, F., Trieomi, F. G., et al. Higher Transcen-
dental Funclions. Kew York: MeGraw-lIilI Book Co., 1)53, volume l.
2. ---, vol. 2.
3. ---, vol. 3.
4. Expansions of Lame functions into series of Legendre fllnetions, Proc. Royal
Soc. of Edinburgh, 62, 19-1X, pp. 2!72u7.
5. llypergeometric functions of two variables, Acta Math., 83, 1950, pp. 131-164.
Fasenmyer, Sister M. Celine
1. Some generalized hypergeometric polynomials, Bull. Amer. Math. Soc., 53,
1947, pp. 806-812.
2. A note on pure recurrence relations, Amer. ill ath. M onlhiy, 5u, 19-W, pp. 14-17.
Favard, J.
1. Sur les polynomes de Tchebicheff, Comptes Rendus Acad. Sci. Paris, 200,
1935, pp. 2052-2053.
Gottlieb, M. J.
1. Concerning some polynomials orthogonal on a finite or enumerable set of
points, A mer. Journal of Malh., 60, 1938, pp. 453-451\.
Grosswald, Emil
1. On some algebraic properties of Bessel polynomials, Trans. Amer. M alh. Soc.,
71,1951, pp. 197-210.
Hall, N. A.
1. A formal expansion theory for functions of one or more variables, Bull.
Amer. Math. Soc., 46,1940, pp. 824-832.
Hobson, E. W.
1. The Theory of Spherical and Ellipsoidal Harmonics. Cambridge: Cambridge
Univ. Press, 1931.
Huff, W. N.
1. The type of the polynomials generated by fCxt) 'PCt), Duke Math. Journal,
14, 1947, pp. 1091-1104.
BIBLIOGRAPHY 353
Huff, W. N. and Rainville, E. D.
1. On the Sheffer A-type of polynomials generated by 'P (t)f(xt) , Proc. Amer. Math.
Soc., 3, 1952, pp. 296-299.
Jackson, D.
1. Fourier Series and Orthogonal Polynomials, Carus Monograph Ko. 6., Math.
Assoc. of Amer., Menasha, 1941.
Kampe de Feriet, J.
1. La Fonction IlypcrgeomCtrique. Paris: Gauthier-Villars, 1937.
Kazarinoff, N. D.
1. Asymptotic expansions for the Whittaker functions of large complex order m,
Trans. Amer. Math. Soc., 7S, 19.').5, pp. 305-328.
2. Asymptotic forms for the Whittaker functions with both parameters large,
Journal of Math. and Mechanics, 6,1957, pp. 341-360.
Krall, H. L., and Frink, Orrin
1. A new class of orthogonal polynomials: the Bessel polynomials, Trans. Amer.
Math. Soc., 65, 1949, pp. 100-115.
Lagrange, Rene
1. Polynomes et Fonctions de Legendre. Paris: Gauthier-Villars, 1939.
Langer, R. E.
1. The solutions of the differential equation v'" + A
2
ZV' + 3}LA
2
V = 0, Duke
Math. Journal, 22, 1955, pp. 525-541.
MacRobert, T. M.
1. Spherical Harmonics. Kew York: Dover, 1948.
Magnus, Wilhelm, and Oberhettinger, Fritz
1. Formulas and Theorems for the Special Functions of Jt[ athematical Physics.
New York: Chelsea, 1949.
2. Anwendung der Elliptischen Funktionen in Physik und Technik. Berlin:
Springer, 1949.
Mehlenbacher, Lyle E.
1. The interrelations of the fundamental solutions of the hypergeometric equa-
tion; logarithmic case, Amer. Journal of Math., 60, 1935, pp. 120-128.
354
BIBLIOGRAPHY
Orr, W. M.
1. Theorems relating to the product of two hypergeometric series, Trans. Camb.
Phil. Soc., 17, 1899, pp. 1-15.
Pa1as, F. J.
1. The polynomials generated by j(t) exp [p(x)u(t)], Oklahoma thesis, 1955.
Pasternack, Simon
1. A generalization of the polynomial F,,(x) , Phil. Mag., series 7, 28, 1939,
pp. 209--226.
Pennell, W. O.
1. Recurrence formulae involving Bessel functions of the first kind, A mer.
Math. Monthly, 36, 1929, pp. 385-386.
Preece, C. T.
1. The product of two generalized hypergeometric functions, Proc. London
Math. Soc., series 2, 22, 1924, pp. 370---380.
Rainville, E. D.
1. Elementary Differential Equations, 2d ed. New York: The Macmillan Co.,
1958.
2. Intermediate Differential Equations. New York: John Wiley and Sons, 1943.
3. The contiguous function relations for pF q with applications to Bateman's
Jnu. v and Rice's H" (r, p, v), Bull. Amer. Math. Soc., 51,1945, pp. 714-723.
4. Certain generating functions and associated polynomials, Amer. Math.
Monthly, 52, 1945, pp. 239--250.
5. Notes on Legendre polynomials, Bull. Amer. Math. Soc., 51, 1945, pp. 268-271.
6. Generating functions for Bessel and related polynomials, Canadian Journal
oj Math., 5,1953, pp. 104-106.
7. Symbolic relations among classical polynomials, Amer. Math. Monthly, 53,
1946, pp. 299--305.
8. A relation between Jacobi and Laguerre polynomials, Bull. Amer. Math. Soc.,
51, 1945, pp. 266-267.
Rice, S. O.
1. Some properties of aF2 (-n, n + 1, r; 1, p; v), Duke Math. Journal, 6,1940,
pp.108-119.
ro
2. Sums of series of the form L anJn+a(z)J,,+p(z), Phil. Mag., series 7,35, 1944,
pp. 686-693. 0
Saalschiitz, L.
1. Eine Summationsformel, Zeitsehrijt Jilr Math. und Phys., 35, 1890, pp. 186-
188.
BIBLIOGRAPHY 355
2. Uber einen Spezialfall der hypergeometrischen Reihe dritter Ordung, Zeit8-
chriftfur Math. und Phys., 36,1891, pp. 278-295, 321-327.
Sheffer, 1. M.
1. Some properties of polynomial scts of typc zero, Duke Math. Journal, 5, 1939,
pp. 59(}-622.
2. Some applications of certain polynomial classcs, Bull. Amer. Math. Soc., 47,
1911, pp. 885--898.
3. Notc on Appell polynomials, Bull. Amer. Math. Soc., 51, 1945, pp. 739---7H.
4. A diffcrcntial equation for Appcll polynomials, Bull. Amer. Math. Soc., 41,
1935, pp. 914-!J23.
5. Concerning Appell scts and associated lincar functional equations, Duke
Math. Journal, 3, 1937, pp. 593-609.
6. Expansions in generalized Appell polynomials, and a class of related linear
functional equations, Trans. Amer. Math. Soc., 31, 1!J29, pp. 261-280.
7. On the properties of polynomials satisfying a linear diffcrential equation:
Part I., Trans. Amer. Math. Soc., 35, 1933, pp. 184-214.
Shively, R. L.
1. On pseudo Laguerre polynomials, Michigan thesis, 1953.
Shohat, J.
1. ThCorie Generale des Polynomes Orthogonaux de Tchebichef. Paris: Gauthier-
Villars, 1935.
2. Thc relation of the classical orthogonal polynomials to the polynomials of
Appell, Amer. Journal of Math., 58,1936, pp. 453-464.
Smith, F. C.
1. On the logarithmic solutions of the generalized hypergeomctric equation
when p = q + 1, Bull. Amer. Math. Soc., 45, 1939, pp. 629-636.
2. Relations among the fundamental solutions of the gcneralized hypergeometric
equation when p = q + 1. Konlogarithmic cases, Bull. Amer. Math. Soc., 44,1938,
pp. 429---433.
3. See [2] above. Part II, Logarithmic cases, Bull. Amer. Math. Soc., 45, 1939,
pp. 927-935.
Smith, R. T. C.
1. Generating functions of Appell form for the classical orthogonal polynomials,
Proc. Amer. Math. Soc., 7, 1956, pp. 636--641.
Szego, G.
1. Orthogonal Polynomials. New York: Amer. Math. Soe. Colloquium Pub!.,
1939, vol. 23.
2. On an inequality of P. Turan concerning Legcndrc polynomials, Bull. Amer.
Math. Soc., 54, 1948, pp. 401-405.
356 BIBLIOGRAPHY
Toscano, Letterio
1. Funzioni generatrici di particolari polinomi di Laguerre e de altri da ess!
dipendenti, Boll. Unione Mat. Ital., series 3,7,1952, pp. 160-167.
Touchard, Jacques
1. Nombres exponentiels et nombres de Bernoulli, Canadian Journal of Math.,
8, 1956, pp. 305-320.
Truesdell, C.
I. On the functional equation uF (z,a)/uz = F(z, a+I), Proc. Nat. Acad. Sci.,
33, 1917, p. 8R.
2. Special Functions. Princeton: Princeton Uni\'. l ' r e ~ s , Annals of Math. Studies
No. is. 1948.
Watson, G. N.
1. A Treatise on the Theory of Bessel Functions, 2d ed. Cambridge: Cambridge
Univ. Press, 19-14.
2. Dixon's theorem on generalizer! hypergeornetric functions, Proc. Lond. Math.
Soc., series 2, 20, 1922, pp. xxxii-xxxiii.
3. Notes on generating functions of polynomials: (4) Jacobi polynomials,
Journal London Math. Soc., 9, 19:34, pp. 22-28.
4. The product of two hypergeometric functions, Proc. London Math. Soc.,
series 2, 20, 1922, pp. 189-195.
Weber, Maria, and Erdelyi, Arthur
1. On the finite difference analogue of Rodrigues' formula, Amer. M alh. Monthly,
59, 1952, pp. 163-168.
Webster, M. S.
1. On the zeros of Jacobi polynomials, with applications, Duke Math. Journal,
3, 1937, pp. 426-442.
Weisner, Louis
1. Group-theoretic ongms of certain generating functions, Pacific J oumal of
Math., 5,1955, pp. 1033-1039.
Whipple, F. J. W.
1. Some transformations of generalized hypergeometric series, Proc. Lond.
Math. Soc., series 2,26, 1927, pp. 257-272.
Whittaker, E. T.
1. An expression of certain known functions as generalized hypergeometric
functions, Bull. Amer. Math. Soc., 10, 1904, pp. 125-134.
BIBLIOGRAPHY 357
Whittaker, E. T., and Watson, G. N.
1. Modern Analysis, 4th ed. Cambridge: Cambridge Press, 1927.
Wilkins, J. E.
J. Keumann "eries of functions, Trans. A mer. Math. Soc., 64, HlIS, PI'.

Wyman, Max, and Moser, Leo
1. On some polynomials of Touehard, CIlTIlldirln Journal of Ala/h., FI, 19.')fi, JlJl.
321-322.
Index
Absolute convergence, infinite product,
3,4
Addition theorems, Jacobian elliptic
function, 3-18
Agarwal, R.P., 2G7, 34G
AI-Salam, W.A., :349
Appell, Paul, 14.5, 265, 2G7, :349
Appell functions, 265, 268-271, 2!l7
relation to Jacobi polynomials, 270 -271
Appell polynomials, 145
Asymptotic series, 3:3-44
about infinity, 36
algebraic properties, 38-39
defined, 33-36
divergence, 34
error function, 38
integration of, 39-40
obtained by integration by parts, 37-38
product, 38
sum, 38
uniqueness, 40-41
\Vatson's lemma, 41-44
A-type, Sheffer's, 221 (see also
Sheffer classification)
Bailey, W.N., 74, 87, 266, 349
Barnes, E.W., 94, 97-99,101-102,123,
349
Barnes integral, 94-102
Barnes path of integration, 95
Bateman, Harry, 63, 127, 130, 162-163,
183, 233, 243, 256, 274, 285-287,
289-291, 302-30:3, 349-:350
Bateman function J nU'v, 287
Bateman polynomial F
n
, 289-291, 302
bound on, 289
defined, 289
generalized, 291, 302
generating functions, 28n-290
hypergeometric form, 289
mixed recurrence relation, 28n
relation to Touchard's polynomial,
289-290
Bateman polynomial, Zn, 233, 243, 285-
286, 291-292, 303-304
defined, 285
differential recurrence relations,
285-286
expansion of xn, 285
finite series of, 285, 304
generalized, 286, 291
generating functions, 28.5-286
hypergeomet ric form, 285
pure recurrence relat.ion, 243
symbolic relation, 30:3
Bedient, 1'.E., 1GB-170, 240, 245, 297-
2G8, :150
Bedient. polynomials, 2G7--2!J8
generat.ing functions, 297-298
hypergeometric form, 2n7
relation to Gegenbauer polynomials,
2G7
Bernoulli numbers, 299-300, 302
generating funetion, 300
relation to lkrnoulli polynomials, 299
Bernoulli polynomials, 299-300, 302
expansion of :rn, :302
generat.ing function, 29n
mixed recurrence relations, 299
relation to Bernoulli numbers, 299
-Euler polynomials, 300
Sheffer A-type zero, 2n!l
Bessel functions, 74, 105, 108-122, 127-
128, 16.'5, 168, 171, 183, 201, 284,
294
bounds, 120-121
defined, 108-lOn
differential equation, 109-110
index half an odd integpr, 114-116, 294
-minus one half, 115
-one half, 115
integral form, 114, 120, 121
integrals involving, 122, 128
116, 121-122, 127,284
Keumann series of, 11!l--120, 122
produf't. of two, 121, 18:3
pure T!:lation, 111
relation to Bessel polynomials, 294
spherical, 116
zeros, 121
Bessel inequality, 155-156
Bessel integral, 114
Bessel polynomials, 245, 286, 291,
293-297
contiguous polynomial relations, 295
differential recurrence relation, 295
expansion of xn, 294
finite series form, 294
generalized, 294-296
generating functions, 294, 296
hypergeometric form, 293-294
, pure recurrence relation, 295
relation to Bessel functions, 294
reversed, 296 '
359
360
Bessel polynomials, (Continued)
Sheffer A-type zero, 296
simple, 293
Beta function, 18-20, 31
relation to factorial function, 31
-Gamma funetion, 19
Bhonsle, B.R, 181, 350
Bilateral generating functions, 170-171,
184, 213, 215, 281
Bilinear generating functions, 197-H18,
211-212,215
Boas, R.P., Jr., 1:30, 140, 143, 181, 224,
230, 245, 283, 350
Boas and Buck generating functions,
140-145,230
Brafman, Fred, I, 1Il4, 167, 190,
198,212,271-272, 290,
Brenke, W.C., 350
Buck, RC., 130, 140, 143, 181,224,
230, 245, 28:J, 350
Burchnall, J.L., 296, 351
Carlitz, Leonard, 289, 297, 351
Celine (see Sister Celine)
Cell, for elliptic function, 306
Chaundy, T.W., 202, 351
Christoffel-Darboux formula, 153-154
polynomials of Hermite, 193
-Jacobi, 177-178
-Legendre, 261
Churchill, R.V., 41,108,117,307,351
Confluent hypergeometric function, 106,
123-128
Barnes integral, 124
contiguous function relations, 124
defined, 123
differential equation, 124
integral form, 124, 128
product of two, 106
transformation of, 124, 126
Contiguous function relations, 50-53,
71-72, 80-85, 291
generalized hypergeometric function,
80-85, 291
hypergeometric function, 50-53, 71-72
Convergence, infinite product, 2
Copson, E.T., 351
Curzon, H.E.J., 191, 351
Darboux (see Christoffel-Darboux formula)
Dickinson, D.J., 140, 153, 286, 295, 351
Differential operator associated wit h
polynomial set, 218-220, 226-228,
302
Differential recurrence relations, 132-133,
135-137,141,144,146
(see also listing under each specific
function)
Divergence, infinite product, 2
Dixon, A.C., 92, 94, 105, 352
Dixon's theorem, 92, 105
Doetsch, G., 190
Doubly periodic function, 305
(see also elliptic funetion)
INDEX
Duplication formula, :'1, 328, 338
Gamma function, 23-21, 31
theta function, 328, 3:38
Elliptic function, 305-348
cpll,300
defined, 306
Jacobian,
onkr ddined, :308-309
-gren,ter than two, 311
-less than two, 309
-two, 311, 31:3-:)44
poles, numTwr in c('lI, 306-308
sum of, ,')07
Weierst mss, 309-:31 :3, :34:3
zeros, number in cell, :J06-308
Elliptie integral, 71,107, :H:3
Erd6lyi, Art hur, 6:3, 74, 127, 156, 206,
2!)!)-300, 302, 352, 356
Error function, 36-38, 41, 127
complementn,ry, 44
Euler, Leonlmrd, 1,8-9, 11-12, 15,
26-27, 31, 47, 60, 300
Euler constant, 8-9
Euler integral, Gamma function, 15, 17
Euler numbers, 300
Euler polynomials, 300
Euler product for Gn,mma function, 15, 17
Euler summn,tion formula, 26-28
Expn,nsion of polynomi:tls in of
polynomials, 147, 151
(see also listing under en,ch specin,l
polynomin,l)
Factorial function, 22-23, 31-:32
generalimtion of bctorial, 22
rehtion to Beta function, :31
-Gn,mma function, 23
Fasenmyer, Sister :\1. Celine (see Sister
Celine)
Favn,rd, J., 153,352
Frink, Orrin, 286, 293-294, 296, 353
Frobenius, F.G., 78
Fuchsin,n equn,tion, relation to confluent
hypergcometric function, 123
Fundamental pn,rallclogram, elliptic
function, 305
Gn,mma function, 8-32, 127
argument large, 29-31
-one half, 21
-positive, 11
INDEX
Gamma function, (Continued)
-positive integral, 12
-unity, 10
defined,9
derivative, 10--11,31
argument one half, 31
-unity, 11
difference equation, 12, 19-21
duplication formula, 24
Euler integral, 9, 15-18
Euler product, 11
evaluation of infinite products, 13-15
gcneralization of factorial, 12
incomplete, 127
limit form, 11-12
multiplication theorem, 26
product form, 9, 11
relation to Beta function, 18-Hl, 31
-factorial function, 23
singularities, 9
Weierstrass product, 9
zeros, lack of, 9
Gauss, K.F., 1, 24, 26, 47, 50, 65, 80
Gauss multiplication theorem, 2-1-26
Gegenbaucr, L., 68
Gegenbauer polynomials, 68, 276-284,
286, 297, 301
argument unity, 278
-zero, 278
bilateral generating function, 281
bound on, 281
contiguous polynomial relationo, 283
differential equation, 279
-recurrence relations, 279, 282-283
expansion of Hermite polynomials, 284
_xn, 283
-(1 - x)n, 282
finite series forms, 277-279, 281-283
-Qf, 280, 282-284
generating functions, 276, 278-282
hypergeometric forms, 279-280, 282
integral form, 281
integrals involving, 281-282
mixed recurrence relations, 283
orthogonality, 281
pure recurrence relation, 279
relation to polynomials of Bateman, 286
-Bedient, 297
-Hermite, 284
-Jacobi, 276
-Legendre, 277, 284
-ultraspherical polynomials, 277
Rodrigues formula, 281
in series of other Gegenbauer
polynomials, 280
special properties, 280
Generalized hypergeometric function,
73-107, 291
361
argument unity, 85-88, 92-94, 10.5-106
contiguous function relations, 80--85,291
convergence of series, 73-74
defined, 73
derivative, 106
differential equation, 75, 74-80
integral form, 85, 100-102
integrals involving, 102 1 04
Laplace transform, 106
product of, 105-106
transformation of, 88, 90, 106
Generating functions, 11:3, 12Y-I4ti, 157,
16:3-16.5, 168, 170-171, 181,
186-187,190,197-198,211 212,
222, 228, 230, 232, 2:39, 247-250,
252-253,256,271-272,274,2.6,
278-282,285-290,292, 2Y4, 2U630:3
Bernoulli numbers, 300
Bessel functions, 11:3
171,184,198,21 :3,215, 2S1
bilinear, 197-198,211-212,215
defined, 129
Euler numbers, 300
polynomials of Appell, 145
-Bateman, 285-287, 289-290
-Bedient, 297-298
-Bernoulli, 299
-Bessel 294, 296
-Boas and Buck type, 140, 143
-Euler, 300
-Gegenbauer, 276, 278-282
-Gottlieb, 303
-Hermite, 44, 130, 187, 189-190,
197-198
-Humbert, 146
-Jacobi, 256, 271-272
(Bessel), 294, 296
-Laguerre, 130, 1:34-135,201-202,
212-213, 215
-Legendre, 130,157,163-165,167-168,
170--171,184-185
2\10
-pseudo Laguerre type
(Toscano-Shively), 298
-Rice, 288
-Sheffer q, 230
-Sheffer zero, 222
-Shively, 298-299
zero, 228
-Sister Celine, 290, 292
-Sylvester, 302
-Tchebicheff, 301
-Toscano (Shively), 298
ultraspherical polynomials, 276 (sfe also
Gegenbauer polynomials)
unnamed polynomials, 131,132, 1 :)4-135,
137, HO, I-1G, lRtl, :nn, 21.250,
252-25::5, 2(16, ;W2
362
Gottlieb, M.J., 303, 352
Gottlieb polynomials, 303
Grosswald, Emil, 297, 352
Hall, N .A., 352
Heat equation, relation to theta functions,
328--329
Hermite polynomials, 44,130-1:32, 187 IU9,
207,214-216,220-221,2:31, Zl9-2-l0,
244, 246--248, 250, 252-253, 284,
292
alternative notation, 189
argument zero, 188
bilateral generating function, 198
bilinear generating function, 197 -198
Christoffel-Darboux formula, H ) : ~
derivatives, 188
differential equation, 188
-recurrence relations, 132, 188
expansion of polynomials, 19:3-196
_xn, 194
finite series forms, 187, 199
-of, 194-195,213-216, 240, 281
generating functions, 44, 130, 187, 190,
197-198
hypergeometric form, 191
integral form, 190-191
integrals involving, 191-193, 199
notation in statistics, 189
orthogonality, 192-193
pure recurrence relation, 188
relation to polynomials of Lagw'rre, 207,
215-216
-Legendre, 191, 194-196, 199
Rodrigues formula, 189, 199
in series of polynomials of Gegenbauer,
284
-Laguerre, 207, 214
-Legendre, 196
Sheffer A-type, 220-222, 231
special property, 197
statistics, notation in, 189
symbolic relations, 246-250, 252-253
table of, 189
zeros, 193
Hobson, E.W., 352
Huff, W.N., 222, 352-353
Humbert polynomials, 146
Hypergeometric differential equation,
53-55
logarithmic solution, 54-55
Hypergeometric function, 45-72, 128
argument minus one, 68
-Dne half, 69
-unity, 48-50, 69, 128
contiguous function relations, 50-53,
71-72
convergence of series, 45-46
INDEX
derivative, 69
differential equation, 53-55
function of its parameters, 48, 55-56
integral form, 47
product of two, 183, 275
transformation of, 59-70
(see also generalized hypergeometric
function)
Incomplete Gamma function, 127
Infinite produet, 1-7,9,11,13-15,332-336
absolute convergence, 3-4
associated series of logarithms, 2-:3
convergence, 2
defined, 1-2
divergence, 2
evaluation, 13-1&
Gamma function, 9, 11, 13-15
theta functions, 3:32-336
uniform convergence, 5-6
Weierstrass M-test, 6
zero factors, 2
Jackson, Dunham, 156, 189,353
Jacobi, C.G.J., 68,167, :314
Jacobi polynomials, 167, 251-275, 276,
283, 296, :301
argument minus one, 257
-unity, 254
Christoffel-Darboux formula, constants
for, 260-261
contiguous polynomial relat ions,
264-2G5
defined, 254
derivatives, 263
differential equation, 2.58
-recurrence relations, 261-264, 274-275
expansion of polynomials, 272-274
-(1 - x)n, 262
finite series forms, 254-255
-of, 2G2, 274-275
generating functions, 256, 271-272
hypergeometric forms, 254-255
integrals involving, 258-2Gl, 275
mixed recurrence relations, 263-265
orthogonality, 258, 259
pure recurrence relation, 263
relation to Legendre polynomials, 254
Rodrigues formula, 257
in series of ultraspherical polynomials
27-l
zeros, 2Gl
Jacobian elliptic functions, 342-348
addition theorems, :348
complementary modulus, 343
defined, 342-343
derivatives, 345-346
differential equations, 346-347
INDEX
Jacobian elliptic functions, (Continued)
integrals, 348
modulus, 342
complemcntary,343
order, 342-344
periodicity, 342-344
poles, 342, 344
relations among squares, 345
Kampe de Feriet, .T., 2G5, 349, 353
Kazarinoff, N.D., 353
Krall, ILL., 286, 2!);3294, 296, :353
Kummer, E.E., 54, 60, fi3, fi8, 124-126,
128, 239, 28G
Kummer's first formula, 124-125, 128
-second formula, 125-12G
Lagrange, Rene, 3.")3
Laguerre polynomials, 130, 134-136, 146,
170-171, 200-217, 231, 235, 2:n,
243-244, 246-253, 281, 291-2!)2,
303-304
bilateral generating function, 170- 1 7 I,
213, 215
bilinear generating function, 211--212,
215
Christoffel-Darboux formula, 206, 214
contiguous polynomial rclations, 20:3
defined, 200
differential equation, 146, 204
-recurrence relations, 134, 136, 146,
202-203,213-214
expansion of polynomials, 206-208
-Tn 207 214
f i n i t ~ seri;)s forms, 201, 213
-of, 207, 209, 214-2Hi
generating functions, 130, 134-135,201,
202,211-213
hypergeometric form, 200, 213
intcgrals involving, 205-206, 214, 216,
303
Laplace transform, 216-217
mixed recurrence relations, 203
orthogonality, 204-205, 214
pure recurrence relation, 202, 214
relation to polynomials of Bateman, 292
-Hermite, 207, 214-216
-Legendre, 208, 216
-Shively, 298
-Sister Celine, 291-292
Rodrigues formula, 204
in series of polynomials of Hermite,
215-216
-Legendre, 215-216
Sheffer A-type, 231
simple Laguerre polynomials, properties
of, 213-215
defined, 200, 213
special properties, 136, 209-215
symbolic relations, 246-253, 303
table of, 201
zeros, 206, 214
Langer, R.E., 287, 353
363
Laplace, P.S., 41-42,117,171,253
Laplace integral for Legendre polynomials,
171
Laplace transform, 41-42, 71, 106, 117,
2IG-217, 2.5:3, 285-287
Lcgendn', A.:\1., I, 24, G8
(see also Lcgendrp polynomials)
Lcg()ndre difTcrenlial equation, 161
second solut ion of, 182-18:3
Legendre duplication formula, 2:1-24, :)1
Legendre polynomirrls, 130-132, 157-186,
191, 194-196, 199, 208, 215 -216,
2:32,244,2+7-218,251-2.'):1,276-277,
284, 287-288, 2fJO-2fJl, 303
argument minus one, 158
-unity, 1.')8
-zero, 158
bilat ('r:d g()!lcrat ing functions, 170-171,
181, 21.5
bounds on, 172-173
Christoffel-Darboux formula, 177-178
defined, 157
differential equation, 161, 182-18:3
-recurrence relations, 1:32, 159, 183, 185
expansion of analytic functions, 181-182
_xn, 181
-(1 - x)n, 185
expansion thcorpm, 177
finite series forms, 157, 162,164,183
-of, 168, 184, 284
generating functions, 157, 163-165, 168,
170-171, HH
hypprgeometric forms, 165-167, 183, 185
integrals involving, 174-176, 184-185,
287
Laplace's integral, 171
orthogonality, 173-174
pure recurrence relat ion, 160
relation to polynomials of Gegenbauer,
277,284
-Hermite, 191, 194-196, 199
-Jacobi, 257
-Laguerre, 208, 215-216
-Rice, 287-288
-Sister Celine, 2fJO-291
Rodrigues formula, 162
in series of polynomials of Gegenbauer,
284
-Hermite, 195
-Laguerre, 208, 215
Sheffer A-type (for transformed
polynomials), 2:32
special properties, 168-169, 184,232
364
Legendre polynomials, (Continued)
symbolic relations, 247-248, 251-253,303
table of, 160
zeros, 174
Legendre function of the second kind,
182-183, 185, 288
Lommel polynomials, 112
!\IacRobert, T.M., 353
!\Tagnus, Wilhelm, 352-353
:".Tascheroni constant, 8-9
;\Iehlenbacher, L.E., 353
!\littag-Lemcr polynomials, 290
!\fodified Bessel function, 1 Hi, 121-122,
127, 2114
:".loser, Leo, 2119-290, 357
!\I ultiplication theorem of Gauss, 24 -26
Xeumann polynomials, 116-119,302
Xeumann series, 119-120, 122
Xewtonian polynomials, 231
X ormalization of orthogonal polynomial
set, 1.55
Oberhettinger, Fritz, 352-353
Order of elliptic function, 308-309
Order symbols, 12-13
Orr, W.M., 354
Orthogonal polynomials, 147-156, 240
Bessel's inequality, 155-156
defined, 147-148
expansion of polynomials, 151
normalization, 155
three term recurrence relation, 151-153
zeros, 149
Orthonormal polynomials, 155-156
Palas, F.J., 354
Pasternack, Simon, 291, 354
Pasternack polynomials, 291
Pennell, W.O., 3,54
Period parallelogram, 305-306
fundamental,305-306
Pochhammer-Barnes confluent
hypergeometric function, 123
Pollak, H.O., 1.53, 351
Polynomials (see under name of specific
polynomial)
Preece, C.T., 354
Pseudo addition theorems, theta functions,
325-328, 336-337
Pseudo Laguerre polynomials, 245, 298
generalized, 298-299
Pure recurrence relation, 233-245
polynomials of Bateman, 243, 289
-Gegenbauer, 279
-Hermite, 188
-Jacobi, 263
INDEX
-Krall-Frink (Bessel), 295
-Laguerre, 202, 214
-Legendre, lGO
-Rice, 213
-Sister Celine, 235, 243
pseudo Laguerre polynomials, 245
unnamed polynomials, 237, 2:39, 243-244
Rainville, E.D., 54, 80, 110, 123, 127, 132,
168, HJl, 222, 28G-287, 2!1l,
295-29G, 3fi3-354
Recurrence relations, 233-215 (see also
pure recurrence relations)
Rice, B.O., 2 4 : ~ , 287288, :3.54
Ric(' polynomials, 2117-21111, 291
ddined, 287
differential recurrence relations, 288
expansion of v
n
, 288
finite s(,ries of, 288
g(merating function, 2118
hypergeometric form, 287
relation to polynomials of Legendre,
21172118
-Sister Celine, 291
Rodrigues formula, 162
polynomials of Gegenbauer, 281
-Hermite, 189
-Jacobi, 257
-Laguerre, 203-204, 214
-Legendre, 161-lG3
Saalschiitz, L., 86-87, 354-355
theorem of, 86-88
Schlafli's integral, 114
Sheffcr, 1.:".1., 130, 221, 226, 231-232, 355
(see also Sheffer classification)
Sheffer classificat,ion of polynomial sets,
218-226
A-type greater than one, 222, 230-231
A-type unity, 222, 231, 274
A-type zero, 222-226, 231, 249-250, 293,
296, 299, 300, 302
generating funct.ion, 222
extension of, 226-231, 250
genemting function, 228
Shively, RL., 200, 240, 245, 298-299, 355
Shively polynomials, 298-299
generalized pseudo Lagucrre, 298-299
generating functions, 298-299
hypergeometric form, 298-299
relation to Laguerre polynomials, 298
Sheffer A-type zero, 299
Shohat, J., 15G, 355
,,-type of polynomial set, 227-228
,,-type zero, polynomials of, 228-232, 250,
25:3, 274, 2!JG, 300
Simplc Laguerre polynomials, 200,213-216
( ~ e e also Laguerre polynomials)
INDEX
Simple set of polynomials, 147, 218-219,
227
defined,147
Sister Celine (Fasenmyer), 1, 137,233-235,
240, 243, 286, 288, 290-292, 295,
303-304, 352
Sister Celine polynomials, 233, 2:35, 2-1:3,
290-293, 303-304
defined, 2HO
difTerentjal recurrence relatjon, 21)2 Z93
expansion of xn, 292-ZH:3
generating functions, 290, 2HZ
hypergeomctric form, Z\)O, 292
integral form, 2Hl
integrals involving, 2H1, 30:3-:304
pure recurrence relations, 235, 2-13
relation to polynomials of Bateman, 2Hl
-Laguerre, 291
-Legendre, 291
-Pasternack,291
-Rice, 291
Sheffer A-type zero, 293
Sister Celine's technique, 233-240,243-245
Smith, F.C., 80, 355
Smith, R.T.C., 355
Sonine polynomials, 200
(see also Laguerre polynomials)
Spherical Bessel funetions, 114-116
Summation formula of Euler, 26-28
Sylvester's polynomials, 302
Symbolic relations, 246-253, 2\)2, 2\)9, 300,
303
Szcgii, G., 156, 181,247,301,3.55
Tchebicheff polynomials, 301-302
defined, 301
finite series forms, 301-302
generating functions, 301
relation to each other, 301
-Neumann polynomials, 302
-ultraspherical polynomials, 301
Theta functions, 314-348
argument zero, 315
basic property table, 319
defined, 314
differential equations, 339-341, 346-347
duplication formulas, 32S, 33S
infinite products, 332-336
-series, 314
large tau, 338
periodicity properties, 31.5-319
pseudo addition theorems, 32.5-328,
336-337
relation among fourth powers, 336
-squares, 322-32.5, 336
-theta constants, 324, 329-332, 336
solutions of heat equation, 328, 329
transformation of, 338
value of G, 336
zeros of, 31\)-322
tabulated, 320
365
Three term recurrence relation, orthogonal
polynomials, 151
Toscano, Lettcrio, 29S, :3.56
Touchard, Jacques, 2S9-2!JO, 356
Touchard polynomiab, 289-200
Tricomi, F.C;., 352
Truesdell, C., 1 no, 356
Type of polynomial sd (sec Sheffer
classification and u-type of
polynomial set)
U1traspherical polynomials, 272, 276-277,
283, ;)01
defined, 276
generating function, 276
relation to polynomials of Geg(mbauer,
277
-Jacobi, 276
series expansion for Jacobi polynomial,
274
Uniform convergence, infinite product,
5-6
Wallis' formula, 19
Wannier, C.H., 153,351
Watson, G.;\'., 41-41, 4n, lOS, 112, 114,
127, lSI, 311, 313, 356-357
Watson's Lemma, 41 44
Weber, :Vlaria, 356
Webster, :\1.8. 356
Weierstrass, Karl, 6, H, 10-11, 13-15, 4S,
309,343
Weierstrass function P(z), 309-313, :H:l
derivative, 310
differential equation, 312
periodicity, 310
poles, 311
relation to elliptic integral, 313
series form, 309
\Veierstrass JI-(est, infinite product, 6
Weierstrass produet for Camma funct ion, \J
Weisner, Louis, 130, J 70, IS4, :3.')6
Whipple, F.J.W., 8S, 90, n2, lOti, 29(;, :35t>
Whipple's theorem, SS, !JO, 92, 2!J6
Whittaker, E.T., 4\), lOS, 114, 127, 181,
311, 313, 356-357
Wilkins, J.E., 357
Wyman, Max, 289-290, 357
Zero factors, infinite product, 2
Zeros of elliptic functions, 306-30S
-orthogonal polynomials, 149
-theta functions, :320