Lecture Notes in

Mathematics
Edited by A. Dold and B. Eckmann

698

Emil Grosswald

Bessel Polynomials

Springer-Verlag
Berlin Heidelberg New York 1978
Author
Emil Grosswald
Department of Mathematics
Temple University
Philadelphia, PA 19122/USA

1 6. JAN. 1979

AMS Subject Classifications (1970): primary: 33 A 70

secondary: 33 A 65, 33 A 40, 33 A 75, 33 A 45, 33-01,33-02, 33-03, 35 J 05,
41A10, 44A10, 30A22, 30A80, 30A84, 10F35, 12A20, 12D10, 60E05

ISBN 3-540-09104-1 Springer-Verlag Berlin Heidelberg New York

ISBN 0-387-09104-1 Springer-Verlag New York Heidelberg Berlin

This work is subject to copyright. All rights are reserved, whether the whole or
part of the material is concerned, specifically those of translation, reprinting.
re-use of illustrations, broadcasting, reproduction by photocopying machine or
similar means, and storage in data banks. Under § 54 of the German Copyright
Law where copies are made for other than private use. a fee is payable to the
publisher, the amount of the fee to be determined by agreement with the publisher.
(c-: by Springer-Verlag Berlin Heidelberg 1978
Printed in Germany
Printing and binding: Beltz Offsetdruck, Hemsbach/Bergstr.
2141/3140-543210
To

ELIZABETH

BLANCHE

and

VI VI AN
 FOREWORD
The present book consists of an Introduction, 15 Chapters, an Appendix, two
Bibliographies and two Indexes.The chapters are numbered consecutively, from 1 to
iS and are grouped into four parts, as follows:
Part I - A short historic sketch (i Chapter) followed by the basic theory
(5 Chapters);
Part II - Analytic properties (5 Chapters);
Part I I I - Algebraic properties (4 Chapters);
Part IV - Applications and miscellanea (2 Chapters).
According to its subject matter, the chapter on asymptotic properties would fit
better into Part II; however, some of the proofs require results obtained only in
Chapter 10 [properties of zeros) and, for that reason, the chapter has been incorpo-
rated into Part Ill.
The Appendix contains a list of some 12 open problems.
In the first bibliography are listed all papers, monographs, etc., that could
be located and that discuss Bessel Polynomials. It is quite likely that, despite
all efforts made, absolute completeness has not been achieved. The present writer
takes this opportunity to apologize to all authors, whose work has been overlooked.
A second, separate bibliography lists books and papers quoted in the text, but
not directly related to Bessel Polynomials.
References to the bibliographies are enclosed in square brackets. Those refer-
ing to the second bibliography are distinguished by heavy print. So [i] refers to:
W.H. Abdi - A basic analog of the Bessel Polynomials; while ~] refers to: M. Abramo-
witz and l.E. Segun - Handbook of Mathematical Functions.
Within each chapter, the sections, theorems, lemmata, corollaries, drawings,
and formulae are numbered consecutively. If quoted, or referred to within the same
chapter, only their own number is mentioned. If, e.g., in Chapter 10 a reference
is made to formula (12), or to Section 2, this means formula (12), or Section 2 of
Chapter 10. The same formula, or section quoted in another chapter, would be refer-
red to as formula (I0.12), or Section (10.2), respectively. The same holds, mutatis
mutandis, for theorems, drawings, etc.
~%ile writing this book, the author has received invaluable help from many col-
leagues; to all of them he owes a great debt of gratitude. Of particular importance
was the great moral support received from Professors H.L. Krall and O. Frink, as
well as A.M. Krall. Professors Krall also read most of the manuscript and made valu-
able suggestions for improvements.
As already mentioned, there is no hope for an absolutely complete bibliography;
however, many more omissions would have occurred, were it not for the help received,
in addition to the mentioned colleagues, also from Professors R.P. Agarwal, W.A. AI-
Salam, H.W. Gould, M.E.H. Ismail, C. Underhill, and A. Wragg.
Last, but not least, thanks are due to Ms. Gerry Sizemore-Ballard, for her skill
 VI

and infinite patience in typing the manuscript and to my daughter Vivian for her
help with the Indexes.
Part of the work on this book was done during the summer 1976, under a Summer
Research Grant offered by Temple University and for which the author herewith ex-
presses his gratitude.

July 1978 E. Grosswald

 TABLE OF CONTENTS

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IX

PART I
CHAPTER 1 Historic Sketch ................................................

CHAPTER 2 Bessel Pol~nomials and Bessel Functions ........................

Differential equations, their d-forms and their S-forms.
Polynomial solutions. Their relations to Bessel functions.
Generalized Bessel Polynomials.

CHAPTER 3 Recurrence Relations ........................................... ,~

Recurrence relations for yn,gn,~n. Representation of BP by
determinants. Recurrence relations for the generalized
polynomials.

CHAPTER 4 Moments and Orthogonality on the Unit Circle .................... ~5

Moment problems and solutions by Stieltjes, Tchebycheff,
Hamburger; the Bessel alternative. Weight function of the
generalized BP. Moments of the simple BP. Orthogonality
on the unit circle.

PART I I
CHAPTER S Relations of the BP to the classical orthonormal pol~nomials
and to other functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . , , . . ° . . ~l~
.,

BP as generalized hypergeometric functions, as limits of

Jacobi Polynomials, as Laguerre Polynomials; their
representation by ~ i t t a k e r functions and by Lommel Polynomials.

CHAPTER 6 - Generating Functions ........................................... h"

Generating functions and pseudogenerating functions. Results
of Krall and Frink, Burchnall, Ai-Salam, Brafman, Carlitz,
and others. The theory of Lie groups and generating functions.
Results of Weisner, Chatterjea, Das, McBride, Chen and Feng,
and others. Different types of generating functions.

CHAPTER 7 - Formulas of Rodrigues Type ..................................... ~

Methods of differential operators, of moments and of generating
functions. Combinatorial Lemmas.

CHAPTER 8 - The BP and Continued Fractions ................................. ~9

The BP as partial quotients. Approximation of the exponential
function by ratios of BP.

CHAPTER 9 - Expansions of functions in series of BP ........................ 6a

Formal expansions in series of the polynomials yn[Z;a,b), or
8n(Z;a,b ). The Boas-Buck theory of generalized Appell Polynomials.
Convergence and summab~lity of expansions in BP. Applications
to expansions of powers and of exponentials.

PART llI
CHAPTER i0 - Properties of the zeros o f BP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Location of zeros. Results of Burchnall, Grosswald, Dickinson,
Agarwal, Barnes, van Rossum, Nasif, Parodi, McCarthy, DoPey,
Wragg and Underhill, Saff and Varga. Olver's theorem. Laguerre's
Theorem. Results of Ismail and Kelker. Sums of powers of
the zeros.
 VIII

CHAPTER ii - On the algebraic irreducibility of the BP ...................... ")9

Theorems of Dumas, Eisenstein, and Breusch. Newton Polygon.
Degrees of possible factors. Cases of irreducibility. Schemes
of factorization. Two conjectures.

CHAPTER 12 - The Galois Groul! of BP ......................................... 416

Theorems of Schur, Dedekind, Jordan, Cauchy, and Burnside.
Resolvent and Discriminant. The Galois Group of the
irreducible BP is the symmetric group. Details of the case n = 8.

C~PTER 13 - Asymptotic properties of the BP ................................ ~24

Case of n constant, z -~ O. Case of constant z, n + =.
Results of Grosswald, Obreshkov, Do£ev.

PART IV

CHAPTER 14 - Applications ................................ i .................. d 3~

The irrati~-ality of e r (r rational) and of ~ . Solution of
the wave equation. The infinite divisibility of the Student
t-distribution. Bernstein's theorem. Electrical networks
with maximally flat delay. The inversion of the Laplace
transform. Salzer's theorem.

CHAPTER iS - Miscellanea .................................................... q 50

Mention of the work by many authors, not discussed in the
preceding chapters.

APPENDIX - Some open problems related to BP ............................... ~62

BIBLIOCRAPI~ of books and papers related to BP .............................. J6/4

BIBLIOGRAPHY of literature not directly related to BP ....................... 17 ~

SUPdECT INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . < 75

NAME INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . "I, 79

PARTIAL LIST OF SYMBOLS ...................................................... N84

 INTRODUCTION
Let us look at a few problems that, at first view, have little in common.

PROBLEM i: To prove that if r = a/b is rational, then e r is irrational; also that

is irrational.

Following C.L. Siegel [$$] (who strea~dined an idea due to Hermite), one first deter-

mines two polynomials An(X ) and Bn(X), both of degree n, such that eX+An(X)/Bn(X)

has a zero of order (at least) 2n + 1 at x = 0. This means, in particular, that

the power series expansion of Rn(X) = Bn(X)e x + An(X) starts with the term of degree
2n+l 2n+2
2n + i, Rn(X) = ClX + c2x + ... say. By counting the number of conditions

and the number of available coefficients, it turns out that An(X ) and Bn(X) are uni-

quely defined, up to a multiplicative constant. By proper choice of this constant

one can obtain that An(X ) and Bn(X ) should have integer coefficients. By simple

manipulations one shows that An(-X ) = -Bn(X ) and that

%(x) = (n!)-ix 2n+l I 01 tn(l_t)netXdt The last assertions are proved by effective

construction of the polynomials involved (see Sections 14.2 and 14.3 for details).

It follows from the integral representation that '.Rn(X) i' ~ (nl)-lixl

' '2n+le[Xl and

that Rn(X) > 0 for x ~ 0. If now e r = e a/b were rational, also e a would be rational;

let q > 0 be its denominator. As already observed, Bn(a) and An(a ) are integers,
$o that

m = qRn(a ) = q(Bn~a)e a + An(a)) is a positive integer.

Using the bound on Rn(a), 0 < m < q.(n~)-la2n+le a and, by Stirling's formula,

0 < m < q(a2n+lea/nn+i/2e-n(2z)i/2)(l+E), where e ÷ 0 as n + =. For sufficiently

large n, 0 < m < I, which is absurd, because m is an integer. Hence, e r cannot be

rational.
Next setting x = ~i, Rn(~i ) = -An(-~i)(-i ) + An(~i )

= (-l)n+ 1 2 n + I :I tn(l-t)nsin zt dt
n! ~0

(the last equality depends on some computations and will be justified in Chapter 14).

The integrand is positive, so that Rn(ni) # 0. Let k = [~] where [x] stands for the

greatest integer function; then A ( x ) + An(-X) is a polynomial in x 2 of degree k and

2
with integer coefficients. Hence, if ~ is rational, with denominator q > O, then

qkRn(~i) = qk{AnC~i ) + An(-~i)} = m, an integer, possibly negative, but certainly

 X

# 0. Also, by using the integral representation of Rn(X), 0 < Iml =

(ql/2 2)n
qk!Rn(wi) l < ( n ) - l q k 2n+l
' - ~ ~ < nn+I/2e-n(2v)I/2 (l+e) (a ÷ 0 for n ~ =), or

0 < In! < i for sufficiently large n. This is, of course impossible for integral m.

Hence 2 , and a f o r t i o r i w are irrational. A (highly nontrivial) modification of

this proof permits one to show much more, namely that e r is actually transcendental
for real, rational r. In particular, for r = i, this implies the transcendency of
e itself.

PROBLEM 2. To prove that the Student t-distribution of 2n+l degrees of freedom is

infinitely divisible.
We do not have to enter here into the probabilistic relevance, or even into the
exact meaning of this important problem. Suffice it to say that, based on the
pioneering work of Paul L e ~ [44.] and of Gnedenko and Kolmogorov [~9], Kelker [65]
and then Ismail and Kelker [60] proved that the property holds if, and only if the

Kn_i/2 (~xx)
function ~(x) = ~x Kn+i/2(~x) is completely monotonic on [0,~), which means that

(-l)k¢(k)(x) > 0 for 0 < x < = and all integral k ~ 0. Now, it is well-knowm (see,
e.g. ~], 10.2.17) that if the index of K (z) (the so called modified Hankel func-

tion) is of the form n+i/2 (n an integer), then (2z/~)I/2eZKn+i/2(z) = Pn(I/z), with

Pn(U) a polynomial of exact degree n. Previous relation can now be written as

Pn_l(X -I/2) Pn_l(X I/2)
n
¢(x] - xl/2pn(X_i/2 ) , or, with Pn(U) = u Pn(i/u), ¢(x) - (xl/2) We now use
Pn

Bernstein's theorem (see [68]); this asserts that ¢(x) is completely monotonic if,
and only if it is the Laplace transfarm of a function G(t), non-negative on
0 < t < ~. In the present case it is possible to study the pol)~omials Pn(X) and
compute G(t). It turns out that G(t) ~ 0 for small t > 0 and also for t sufficiently
large. This alone is not quite sufficient to settle the problem, but if we also
knew that G(t) is monotonic, then the conclusion immediately follows. In fact, by
playing around with G(t), one soon suspects that it is not only monotonic, but
actually completely monotonic. In order to prove this, one appeals once more to
Bernstein's theorem and finds that G(t) is the Laplace transform of ¢(x) =

(~2 x) -1/2{i n
÷ j=l
X ~j(x+ aj)
2 - i}, where ~l,a2 ..... an are the zeros of the polynomial

Pn(U). A detailed study of these zeros permits one to reduce the large bracket to

the form xn/q(x), where q(x) is a polynomial with real coefficients and such that
q(x) > 0 at least for x > - min l~jl 2. This shows that ¢(x) > 0 for
l<jsn
 XI

0 < x < ~; hence, G(t) _> 0 on 0 < t < - and ¢(x) is indeed completely monotonic, as
we wanted to show.

PROBLEM 3. To solve the equation with partial derivatives

1 32V
(1) AV = - ~ - - where A is the Laplacian,
c ~t 2

32 2 3 i 32 3 1 32
A = ~ 3 r + r Dr + -2-r (~-~ + cot 0 ~ ) + r2sin2 e 302

with the boundary conditions (i) and (ii):

(i) V = V(r,e,¢,t) is symmetric with respect to a "polar axis" through the
origin, so that, in fact, V = V(r,O,t) only;
(ii) V is monochromatic, i.e., all "waves" have the same frequency ~;
and with the initial condition
(iii) the values of V are prescribed along the polar axis at t = 0, say
V(r,0,0) = f(r), a given function of r.

Here r, 0, and ¢ are the customary spherical coordinates, t stands for the time and
c represents dimensionwise a velocity.
Conditions (i) and (ii) are imposed only in order to simplify the problem and
can be omitted, but the added complexity can easily be handled by classical methods
and has nothing to do with the problem on hand.
Following the lead of Krall and Frink [68] we look, in particular, at solutions
of (1) of the form (obtained by separation of variables)

u = r-ly(i/ikr) L(cos 0)e ik(ct-r)

Here y and L are, so far, undetermined functions and we shall determine them precise-
ly hy the condition that u be a solution of (I), while k is a parameter related to
the frequency ~ by k = ~/c. On account of (ii), k is a well defined constant. The
(artificially looking) device of introducing complex elements into this physical
problem is useful for obtaining propagating, rather then stationary waves. The
real components v, w of u = v + iw will be real solutions of (I) and represent waves
traveling with the velocity c. For c = 0 and with x = i/ikr, one obtains, of course,
directly a real, stationary solution of (i). We now substitute u into (i), by taking
into account that 8u/3# = O, and obtain, with x = I/ikr and z = cos 8, that

L(z)(x2y"(x) + (2+2x)y'(x)) + y(x)((l-z2)L"(z) - 2z L'(z))= O,

or, equivalently,

x2~'"(x)+(2+2x))' '(x) = _ (l-zZ}n"(z)-2zn '(z)

y(x) L(z)
 XII

These two functions, each of which depends on a different independent variable, can
be identically equal only if their common value reduces to a constant, say C. It
follows that L(z) satisfies an equation of the form

(l-z2) L"(z)-2z L'(z) + CL(z) = O.

We immediately recognize here the classical equation of Legendre. If, but only if
C = n(n+l) with n an integer (it is clearly sufficient to consider only n { O,
because -n(-n+l) = n(n-l)) does this differential equation admit a polynomial solu-
tion, namely the Legendre polynomial of exact degree n; we shall denote it by Ln(Z).

Incidentally, if we would not require symmetry with respect to the polar axis, then
we would obtain here the associate Legendre nolynomials P(q)(z) instead of the sim-
n
pler Legendre polynomials, and this is the main reason for the present, more restric
tire formulation of the problem.
So far, everything has been fairly routine; now, however, it turns out rather
surprizingly that, with C = n(n+l) the equation

(2) x 2 y"(x) + (2+2x) y'(x) - n(n+l) y(x) = O,

satisfied by y(x), also admits a polynomial solution for n an integer, namely a

polynomial of exact degree n, uniquely determined up to an arbitrary multiplicative
constant. We shall denote it by Yn(X) and may normalize it, e.g., by setting
Yn (0) = I.
We have, herewith, obtained a sequence of solutions to (i), of the form

u n = Un(r,9,t ) = r-iLnfCOS,
- 3]Vn(I/irk)eik(ct-r),,
.

With each solution u n and for each constant an, also anU n is a solution of (i), and

so is the sum V = ~ anUn, if it converges. In particular, along the polar axis,

n=O
with z = cos 9 = I, Ln(l ) is equal to i, and we obtain at t = O, with previous sub-

stitution x = i/ikr,

V = V(r,O,O) = ~ a r -I e -ikryn(i/ikr ) = ik ~ a xe-i/Xy n ( x ) .

n=O n n=O n
In order to satisfy also the initial condition, we define F(x) by f(r) = f(i/ik~ =
ikx F(x), so that condition (iii) becomes

I ane-i/Xyn(X ) = F(x).
n=O

From (2) it follows that, by taking as closed path of integration the unit circle,
 XIII

Yn (Z)Ym(Z)e-2/Zdz = ~ 2(-l)n+l
mn 2n+l '

where the Kronecker delta ~ = 1 for m = n, ~ = 0 otherwise. It follows that

mn rml

an = (-i)n+l (n+i/2)#F(z)yn(Z)e-I/zdz.

With these values for an , V(r,O,t) = ~ anr-iLn(COS @)Yn(I/ikr)e ik(ct-r) is a for-

n=0
mal solution of (I), in general complex valued, that satisfies formally all boundary
and initial conditions of the problem. It is an actual solution, if the infinite
series converges. Precise conditions (that depend on the nature - especially the
singularities - of F(z)) are known (see [13]) for this convergence and will be dis-
cussed in Chapter 9. Here we add only that in the more general situation, when we
discard the restrictions (i) and (ii), the corresponding solution is of the form

V(r,e,¢,t) = ~ n~ 0 ....
an,kP~(cos e) sznlm¢+¢0Je ik (ct-r) Yn(i/ikr)/r,
k m=0
with the outer sum extended over all values of k = e/c, corresponding to all frequen-
cies ~ that occur.
What do these problems have in common? All three depend on the study of cer-
tain sequences of polynomials, An(X ) (= -Bn(-X)) in Problem I, Pn(X) in Problem 2,
Yn(X) in Problem 3. In fact, the three problems have more in common than just that,
because actually, all three sequences of polynomials are essentially the same se-
quence. There are still many other problems, in which this particular sequence of
polynomials kno~m to-day as Bessel Polynomials plays a fundamental role. It also
turns out that these polynomials exhibit certain symmetries that are esthetically
appealing and have therefore been studied for their own sake. To-day there exists
a fairly extensive literature devoted to this specific subject. Nevertheless,
recently, when the present author needed some information concerning these polyno-
mials, it turned out that it required an inordinate amount of time to search through
a large number of papers and several books, in order to locate many a particular
fact needed. It is the purpose of the present monograph, to give a coherent account
of this interesting sequence of polynomials. It may be overly optimistic to hope
that everything known about them will be found here, but at least the more important
theorems will be stated and proved. Originally an attempt was made to obtain all
important properties in a unified way, but this attempt hm~ not always been success-
ful; in fact it could hardly have been expected to be. After all, it is not sur-
prizing that the structure of the Galois group of Pn(X) requires for its determina-
tion other methods than, say, the study of the domain of convergence of an expansion
in a series of these same polynomials.
The author has made himself a modest contribution to the subject matter, but
the aim of this work is primarily expository: to systematize and to make easily
 XIV

accessible the work of all mathematicians active in this field. But mainly, unless
this book succeeds in relieving the future student of this subject o£ the need for
an exasperating, time consuming search for known items, deeply hidden in the litera
ture, it will have failed in its purpose.
 PART I
CHAPTER 1
HISTORIC SKETCH

I t may not be easy t o d e t e r m i n e t h e f i r s t occurrence in the mathematical litera-

t u r e o f t h e s e q u e n c e o f p o l y n o m i a l s t h a t we a r e i n t e r e s t e d in. They seem t o have
a p p e a r e d s p o r a d i c a l l y f o r a r a t h e r long t i m e . In 1873 Hermite [2~] proved t h e
t r a n s c e n d e n e y o f e. The p o l y n o m i a l s used by Hermite are c l o s e l y r e l a t e d t o t h e ob-
j e c t o f our s t u d y . See [83] f o r t h e c o n n e c t i o n . The p o l y n o m i a l s d e n o t e d by Olds
w i t h Tn and Zn a r e r e l a t e d t o t h e Bessel P o l y n o m i a l s yn(Z) by yn(Z) = Tn(Z) + Zn(Z};

s e e [83] and C h a p t e r 8 f o r more d e t a i l s .

These p o l y n o m i a l s a p p e a r a l s o i n 1929, in a p a p e r by Bochner [14] and in one by
Romanowsky [92]. S h o r t l y a f t e r w a r d s (1931), but q u i t e i n d e p e n d e n t l y , t h e y o c c u r i n
a long p a p e r [18] by J . L . B u r c h n a l l and T.W. Chaundy, who o b t a i n them as s o l u t i o n s
of certain differential equations. W. Hahn [58] runs a c r o s s them i n 1935 and so
d i d H.L. K r a l l [67] i n 1941.
The f i r s t s y s t e m a t i c s t u d y o f t h e s e p o l y n o m i a l s i s due t o H.L. K r a l l and O.
F r i n k , who i n 1949 c o n s i d e r them i n a fundamental p a p e r [68] p u b l i s h e d i n t h e T r a n s -
a c t i o n s o f t h e AMS. They gave t h e s e p o l y n o m i a l s t h e name o f BESSEL POLYNOMIALS,
u n d e r which t h e y have been known e v e r s i n c e . This same d e s i g n a t i o n has o f t e n been
used by v a r i o u s a u t h o r s , even when t h e y a c t u a l l y s t u d i e d r e l a t e d p o l y n o m i a l s , o r
different normalizations, etc. So, e . g . , we f i n d among t h e s e t s o f p o l y n o m i a l s
l a b e l l e d as B e s s e l P o l y n o m i a l s , b e s i d e s K r a l l and F r i n k ' s Yn(X), a l s o x n y n ( 1 / x ) ,

n-1 n
(-1) x Yn(2/x), ( x / 2 ) n y n ( 2 / x ) , and o t h e r s .

In t h e p r e s e n t monograph, we s h a l l adopt i n g e n e r a l t h e o r i g i n a l n o r m a l i z a t i o n
o f K r a l l and P r i n k and we s h a l l a b b r e v i a t e t h e d e s i g n a t i o n B e s s e l P o l y n o m i a l s , by BP
( r e g a r d l e s s o f t h e i r use i n t h e s i n g u l a r , o r p l u r a l ) .
K r a l l and F r i n k had been l e d t o t h e c o n s i d e r a t i o n o f t h e BP by a s t u d y o f t h e
wave e q u a t i o n ( e s s e n t i a l l y our Problem 3). They i n d i c a t e t h e d i f f e r e n t i a l equation,
recurrence relations, a pseudo g e n e r a t i n g f u n c t i o n and an o r t h o g o n a l i t y p r o p e r t y ;
t h e y a l s o g e n e r a l i z e t h e s e t Yn(X), by i n t r o d u c i n g two p a r a m e t e r s ( o n l y one r e a l l y
significant).
Shortly afterwards, and stimulated by Krall and Frink's work, two other papers
appeared almost simultaneously. J.L. Burchnall [17] pointed out that the BP Yn(X)

is related to the polynomials 8n(X ) studied by Burchnall and Chaundy in [18] by

8n(X ) = xn Yn (i/x) By using the machinery developed there, Burchnall obtains several

of the results o f K r a l l and F r i n k and, i n a d d i t i o n some o f t h e b e a u t i f u l symmetry

p r o p e r t i e s o f t h e z e r o s , as w e l l as a g e n e r a t i n g f u n c t i o n .
 The present author [53] studied asymptotic properties of the BP, also proper-
ties of their zeros, the irreducibility of the BP over the rationals and the Galois
group of the BP.
At about the same time, when Krall and Frink, Burchnall and the present author
started the systematic study of BP, W.E. Thomson studied certain networks used in
multistage amplifiers, and that led to a particularly desirable characteristic,
called maximally - flat delay. The investigation of the complex impedance and of
the transfer functions of these networks led to the consideration of certain polyno-
mials, proportional to the transfer function and defined by initial values and a
recurrence relation. These polynomials turn out to be exactly the BP in the normali-
zation of Burchnall and Chaundy. It is quite unfortunate that Thomson's work [107],
[108] remained essentially unknown to the mathematicians who worked on BP. Indeed,
Thomson obtained many important properties of the zeros of BP, but he was not parti-
cularly interested in the theory. "Those interested in the theory will find an out-
line in the Appendix", he writes. It is indeed just an outline, set in particularly
small print. In addition to many interesting details, some of which were rediscover-
ed only recently [see Chapter 10),the paper [108] contains also a tabulation of all
the zeros (real and complex) of the BP up to the ninth degree (inclusive). For
other similar tabulations see [99], [I00], [70], [I02], [69], and [116].
A few years afterwards, a real flood of papers appeared, with improvements of
the theorems concerned with the location of the zeros, with generating functions and
with the relations of the BP to other special functions, especially to Bessel func-
tions and the hypergeometric series; this is not too surprizing, since Yn(X) =
2F0(-n, l+n; -; -x/2). It is impossible to mention at this place all contributions

made during these years, and quoted in the following pages but a few names and
papers come to mind: Carlitz [19]; AI-Salam and Carlitz [7], [8], Ai-Salam [3],
Rainville [89], Agarwal [2]; Toscano [109]; McCarthy [74], [75]; Nasif [81], Dick-
inson [47], Brafman [15], Ragab [88], Wimp [112] and others.
Some of this material was included in abbreviated form in the books by R.P. Boas
and R.C. Buck [13] and by E.D. Rainville [90].
In 1962 Do~ev [48] obtained what until recently was the best upper bound for
the absolute value of the zeros of BP. See Chapter 10 for more recent results by
01vet [84], [~6] and by Saff and Varga [98].
While the interest in BP never completely disappeared, there were a few years,
during which the efforts of mathematicians were apparently directed into other
channels. But recently, there seems to have arisen a renewed interest in BP. In
1969 the author [54] could settle (computer assisted - although in the proof itself
no computer work is invoked) a remaining problem concerning the Galois groups of
the BP of degrees 9, ii, and 12.
 Parodi [85] r e p r e s e n t s t h e BP as d e t e r m i n a n t s and o b t a i n s new bounds f o r t h e
r e g i o n i n which one may f i n d z e r o s o f BP. I f one combines t h e s e w i t h some p r e v i o u s l y
known o n e s , one g e t s r a t h e r strong results.
Barnes [12] s t u d i e s a g a i n t h e z e r o s o f BP, as w e l l as t h e i r c o n n e c t i o n with
continued fractions and w i t h t h e e x p o n e n t i a l f u n c t i o n .
Wragg and U n d e r h i l l [113] r e l a t e t h e BP t o t h e d e n o m i n a t o r s o f t h e Pad~ a p p r o x i -
mants t o t h e e x p o n e n t i a l f u n c t i o n ; they also represent, f o l l o w i n g P a r o d i , t h e BP as
d e t e r m i n a n t s and, by u s e o f G e r s h g o r i n ' s and t h e B e n d i x s o n - H i r s c h t h e o r e m s , o b t a i n
r a t h e r s i m p l e p r o o f s f o r good bounds on t h e z e r o s o f t h e BP.
K e l k e r [65] and I s m a i l and K e l k e r [62] r e d u c e an i m p o r t a n t p r o b l e m o f p r o b a b i -
lity (the infinite divisibility of certain distributions - essentially our Problem
2) to the complete monotonicity of the function ¢(x) = Yn_l(X-I/2)/xl/2yn(X-i/2).

The proof that this property, in fact, holds, has been obtained by the present
wTiter [55], [56].
In the meantime, Thomson's contribution, ignored by mathematicians, became
rapidly common knowledge among electrical engineers. They discovered soon the iden-
tity of Thomson's polynomials with the BP and used systematically the properties
of these polynomials and of their zeros, as presented in [68], [17], and [53], in
order to perfect amplifiers, as well as filters. In 1954, L. Storch [106a,b] giving
full credit to Thomson, and quoting [108], [68], [17] and [53] treats the topic with
all the details needed for the understanding of the theory and also for the effec-
tive computation of the numerical values of the elements of the network.
Soon the use of BP in the synthesis of certain networks was treated in textbooks
like those of E. Guillemin [57] and D. Hazony [60]. More recently R.R. Shepard
[I01] indicated an almost mechanical method for the design of networks with certain
preassigned characteristics, by use (among others) of BP. For related work see
also [64], [115], and [116].
At present, the BP are accepted along with the classical orthogonal polynomials
among the "special functions" and are often mentioned in connection with either the
Bessel functions, or some other previously studied functions, to which they are re-
ducible (see Chapter 5). They are quoted, besides in [90], also in such well-known
collections as [71], [72], [20], [50] or [ll4](but not in [~]), and there does not
appear to exist in the mathematical literature any systematic discussion of their
properties. This fact, the author hopes, will be accepted as a sufficient justifica-
tion for the present monograph.
 CHAPTER 2
BESSEL POLYNOMIALS AND BESSEL FUNCTIONS:
DIFFERENTIAL EQUATIONS AND THEIR SOLUTIONS.

I. The so called "modified" Bessel functions Iv(z) and Kv(z ) - sometimes improperly

designated as being of imaginary argument -, K (z) also known as the MacDonald func-

tion, satisfy the well-known (see e.g. [61] or [~]) differential equation

(I) Lw - Z2W '' + zw' - (z 2 + v2)w = O.

For fixed v with Re v > 0 the following asymptotic relations hold (see [I]):
For z -~ O,

K (z) _= ~1 r(~)(z/2) -v, K~(z) ~-r(v+l)2~-iz v-I

(2a)
Iv(z) ~ {F(v+l)}-l(z/2) v, l$(z) a (2F(v))-l(z/2)v-l;

for z ÷ ~ and larg z I < ~,

(2b) Kv(z ) ~ (w/2z)i/2e-Z, Iv(z) ~ (2wz)-i/2eZ.

These relations suggest that functions like ¢(z), or 8(z), defined by w = z-re, or

by w = z-Ve-Zs, with e = eZ¢, may exhibit a simpler behavior than the Bessel func-
tions themselves.

2. Let us consider first 8 = e(z). By logarithmic differentiation

v 8'
w, = (- ~ - i + ~-)w

O" 2)W v ~--)w'

w"= (~ +~-- ( + (- z -I +
z

O" .0'.2 8'.2.Jw

v -i + ~--)
z

and, substituting these in Cl) we obtain

Lw = Cz8" - (2z+2v-l)%' + (2v-l)@)zw@ -I = 0.

As zw(z) ~ 0, it follows that e(z) satisfies the differential equation

(3) z@" - (2z+2v-l)e' + (2v-l)@ = 0.

By the general theory of linear differential equations (e.g., [~]) the origin
is a regular singular point and there exist (in general) two independent particular

solutions of the form @ = z ~ ~ cmzm. Here ~ is any one of the two (in general
m=0
distinct) solutions of the indicial equation. Either by Frobenius' method, or by
general considerations one finds that the indicial equation is

(4) a(a - 2u) = O.

The solutions of (4) are indeed distinct, except for v = O. In the latter case

the two solutions are, of course, proportional to e-ZIo(Z) and e-ZKo(Z), respectively

and are of no particular further interest here.

F o r u # O, s e t 8 = O(z,u) = Cmzm a n d 8 = O ( z , ~ ) = z 2~ ~ c'zm'm By d i f f e r -

m=O m=O
entiation and s u b s t i t u t i o n into (3), we o b t a i n

2u+l-2m (2u-1)(2u-3)...(2~-2m÷l)
cm - (2u-m)m Cm-1 ~ " ' " = (2~-l)(2~-2)...(2u-m)ml Co "

s o that

8(z,v) = coCl+z+ (29-i)(2v-3) z 2 (2v-l)(2v-3)...(2v-2m+l) zm

(29-I)(29-2) 2! + "'" + (29-1) (29-2) ... (2v-m) m--~ + "'" )"

Similarly,

C' 2~+2m-I c' = (2v+l)(2v+3)...(2v+2m-l)

m - (2v+m)m m-i "'" = (2~+l)(2v+2)...(2,~+m)m! c0

and
2 m
%(z,u) = c6z2~(l+z+ 2~+3 Z (2~+I)(2'~+3)...(2~+2m-I) z
29+2 2: + "'" + (2~+l)(2v+2)...(29+m) m--~ + "'') "

The f u n c t i o n s O(z,~) a n d z~eZK ( z ) , both are solutions o f (3) a n d s o i s any

linear combination with constant coefficients of these functions. Also, relations

(2a) show t h a t i f we s e t u v ( z ) = cQz~eZK ( z ) - 2 v - l r ( v ) e ( z , ~ ) , t h e n u (z) is a solution

o f [3) w i t h u (0) = u ~ ( 0 ) = 0. Consequently, u (z) v a n i s h e s identically and

(s) 8(z,v) = ~vzVeZKv(z) with ~v = c021-~(r(~))-i.

In a similar way one verifies that 8(z,~) = ~ zVeZl (z), with ~ = c62~r(~+1).

It is obvious that O(z,~) reduces to a polynomial of exact degree n if 2~ = 2n+l

and t h a t z - 2 ~ 0 [ z , ~ ) reduces to a polynomial of exact degree n if 2~ = - 2 n - 1 . In

the first case, for ~ = n+l/2, t h e two i n d e p e n d e n t solutions of 65) a r e
2 m
0 ( z , n + 1) = c 0 ( l ÷ z + 2 n - 2 z (2n-2) C2n-4)...(2n-2m+2) z
2n-1 2' ÷ "'" + (2n-1) C2n-2)...C2n-m+l) m-T + " ' "
n
(2n-2) (2n-4) ...4.2 z
+ (2n-I) (2n-2)... (n+2) (n+l) ~T. ')
and
 2 m
, 2n+l.. 2n+4 z (2n+4) (2n+6) ... (2n+2m) z
e(z,n+ i) = C o Z if+z+ 2n+3 2-! + • • • ÷

(2n+3) (2n+4) ... (2n+m+l) m-T + "" ") -

1
In the second case, with ~ = -n - ~ ,

m
(2n+4)(2n+6)...(2n+2m) z___+ ...)
e(z,-n-i/2) = e0(l+z+...+ (2n+3)(2n+4)...(2n+m+l) m~
and
m
e(z,-n-I/2) = c~z-2n-l(l+z+...+ (2n-2)(2n-4)...(2n-2m+2) z
_ _ +

(2n-l) (2n-2)... (2n-m+l) m'

n
(2n-2)...2 z
(2n-l)...(n+l) n--~ )"

If we take c o = c6, then

(6) e(z,n+i/2) = z2n+le(z,-n-i/2) and e(z,n+i/2) = z2n+le(z,-n-i/2).

It follows that, in general, it will be sufficient to consider only non-negative

values for the half integral second parameter• Sometimes, however, e.g., for rea-
sons of symmetry, it may be convcnicnt to be allowed to use also negative values
for the integer n and we shall soon find a convenient way to do it.

In @(z,n+I/2) the coefficient of zn equals CO-2nn!/(2n)!. In following Burch-

nall [17] and Burchnall and Chaundy [18], we select c 0 = (2n)!/2nn2, so that the

leading coefficient becomes one. With this normalization we shall denote e(z,n+i/2)
simply by en[Z). One now easily verifies (see [17]) that
n n
(7) @n(Z) = [ an_mZm = [ amzn'm ,
m=O m=O
where all coefficients are integers (see Theorem 3•1) given by

2m-n(2n-m)! (n+m)!
(8) a =
n-m (n-m) .m.'' , am 2m(n-m) .m.'
'
So, e.g., we find that
2
e0(z) = i, el(Z) = z+l, e2(z) = z +3z+3, etc.

If we have to deal simultaneously with several polynomials, it is useful to

identify the polynomials to which a coefficient belongs by a superscript. So, e.g.,

a~ 5) is the coefficient of z 5-3 = z 2 in es(z); by (8) its value is (5+3)~ _

23(5-3)~3!
8! 7:
. . . . 420.
8.2~3! 12

The coefficients that occur in (7) and (8) would then be denoted by a (n) and
n-m
a(n! respectively.
m
 While the polynomials en(Z ) have been defined so far only for n > O, we observe

that, if we replace formally n by -n in (8), we obtain

a(-n) = (-n-m+l)(-n-m+2)...(m-n) = (_l)2m (n+m-l)(n+m-2)...(n-m) = a(n-l).

m 2mm2 2mm! m

Hence, if we extend the definition (7) of en(Z) formally to negative subscripts,

we obtain
n
(9) e n(Z) = [ a(-n)z -n-m = z-2n+l a(n-l)z (n-l)-m
- m N
m=0 m= 0

= z-2n+l nil a~n-l)z(n-l)-m = z-2n+iOn_ l(z);

m=0

indeed, on account of the factor n-m in the numerator of a (n) the coefficient
m
a (n-l) vanishes.
n

The identity (9) may be taken as definition of @_n(Z). If we replace in (8) n

by n+l, we obtain @n(Z ) = z2n+le_(n+l)(Z ). Comparison with (6) shows that

@_(n+l)(Z) is not a new function, but is precisely the function e(z,-n-i/2).

In some contexts it is more convenient to work with the reverse polynomial

n
(i0) yn(Z) = znenCl/z) = [ am zm,
m=0
a normalization due to Krall and Frink (see [68 ]). If we want to have the first
coefficient reduced to unity, we may factor out a(n)n = ((2n)!)/2nn! = Co and obtain

yn(Z) = c0 n~ b(n)zn-m, b(n) = 2! (2n-m)(2n-m-l)...(n-m+l)

m=0 m m m! 2n(2n-l)...(n+l)

By differentiation of @n(Z ) = znyn(i/z) we obtain successively

e~(z) = nzn-lyn(1/z)-zn-2y~(1/z),
and

8n(Z ) = n(n-1)zn-2yn(1/z)-nzn-3yn(1/z)_(n_2)zn-3Yn(1/z) + zn-4yn(1/z).

If we substitute these into (3) with ~ = n+I/2, we obtain after a few simplifi-
cations (see [68]) that YnCX) satisfies the differential equation

(Ii) z2y~(z) + 2(z+l)y~Cz)-nCn+l)YnCZ) = 0.

From (99 and (i0) it follows that

 Yn [z) = znSn (z-l) = zn'z-2n-l@-(n+l) (z-l) = z-n-l@-(n+l)(Z-1) = Y-n-i (z)'

so that the polynomials yn(Z) are defined for negative subscripts by the particular-
ly simple relation (see [68] and [47])

(12) y_n(Z) = Yn_l(Z).

We shall speak occasionally of both, 8n(Z) and yn(Z) as Bessel Polynomials (BP)
and it will be clear from the context, which one is meant. When confusion could
arise, then yn(Z) will be called the n-th BP, while 8n(Z) will be referred to

(following Boas and Buck [13]) as the reverse BP.

We summarize (and slightly complete) the results obtained in this section in

the following theorem:

THEOREM I: For integral n > O, the differential equation

zw" - 2(z+n)w' + 2nw = 0

has polynomial solutions w = @n(Z ) . These are polynomials of exact degree n and

n
specifically On(Z) = ~ a(n)z n-m, with a (n) (n+m)!
m=O m m 2re(n-m)!m!
2z
If n = O, the general solution of the differential equation is cle + c2

and also in this case ~mong the solutions one finds the polynomial of degree zero
eo(Z ) = i, the same as one obtains formally by 8ettin~j n = 0 in above formulae.

The equation with n replaced by the negative integer -n, n > 0 has a rational

solution 8_n(Z) that satisfies 8_n(Z ) = z -2n+18 n_l["z"), or, equivalently,

@n(Z) = z2n+ls_(n+l)(Z).

The equation

z2w '' + 2(z+l)w' - n(n+l)w = 0

has polynomial solutions w = yn(Z) of exact degree n. Specifically, with the seQne

n
aCn)m as above, yn(Z) = a(n)z m is a solution, so t ~ t
m
yn(Z) = zn8 (l/z) and yn(Z)
n
m=O
and en(Z ) are polynomials reverse to each other.

For n = 0 the general solution of the equation i8 cle2/Z + c 2 and o~nong these

solutions one has, in particular, the polynomial of degree zero yO(z) = I.

 If n is replaced by -n-l, the product n(n+l) remains unchanged, so that

y_n_l(Z) = yn(Z).

3. We turn now to the function ~(z) = e-Z@(z). From @(z) = eZ~(z), by differentia-

tion, e' = eZ~ ' + eZ~, e" = eZ~ '' + 2eZ~ ' + eZ~ and if we substitute these in (3),
it follows that ~ satisfies the particularly simple differential equation

(13) z2~ '' - 2nz~' = z2~ .

2 d2 d
Denote the linear differential operator z dz 2 - 2nz ~ by L; then the equation

L# = z2~

stays invariant under the transformation z -~ -z. Hence, with ~(z) = e-Ze(z), also

~(-z) = eZe(-z) will be a solution of (13).

The differential operator L, of second order, can be factored with
d
the help of the first order operator ~ = z ~-~ . For future use, and also because

they are of independent interest, some of the (well known - see [17]) properties of
this operator will be developed here.
Clearly ~zn = nz n
and, more generally,

(14) ~kzn = nkz n

holds, and also

(15) (~-a)z n = (n-a)z n.

More generally, if f(z) is differentiable, ~f = zf' and, if f(z) is n times

differentiable, then
n
~nf = ~ s(m)zmf(m)
n
m=l

where S (m) are the Stifling numbers of the second kind. The general case follows by
n
induction on n, starting from n = i (which holds by the definition of 5), with the

help of the recurrence relation (see [~] p. 825) S (m) = mS (m) + S (m-l)
' n+l n n "

Next, by using (15) one verifies that the operators ~-a and ~-b commute. Con-
sequently, if P[n) is a polynomial with zeros Ul,U2, .... Um, one has

(16) P(~)z k = (k-Ul)(k-u2)...(k-Um)Z k = P(k)-z k.

 10

We observe that for every differentiable function f(z),

(~-n) zf(z) = z ~d (zf(z))-nzf(z) = z(zf'-(n-l)f) = z(~-(n-l))f(z)

It follows that a factor z may be "moved across" the operator g-n, from right to
left, provided that n is replaced by n-l.

If we iterate the procedure we obtain the result that

(17) (~-nl)(6-n2)...(6-nr)Zf(z) = z(6-nl+l)...(6-nr+l)f(z).

-bz
In particular, for f(z) = e , we h a v e

-bz
(18) ( 6 - n 1) ( 6 - n 2) . . . ( ~ - n r ) ze = z (~-nl+l) (6-n2+1)... ( ~ - n r + l ) e -bz .

It is possible to compute explicitly the right hand side of (18).

This has a particularly simple expression, if the constants nl,n 2 .... ,nr are

consecutive integers.
Indeed,

(6_n)e-bZ _(bz+n)e-bZ zn+l d (z-ne-bZ),

and

( z - n e - b Z ) + z n+2 d 2 (z-ne-bZ),
~(~_n)e-bZ = (n+l)zn+l d _ _

dz 2

so t h a t (6-n-1)(~-n)e -bz = z n+2 _ _d2 ( z - n e - b z )

dz 2

An induction on m will now complete the proof of

-bz n+m dm
(19) (~-n-m+l)(6-n-m+2)...(6-n)e = z (z-ne-bZ).
dz m
We observe in particular that for a twice differentiable function f(z),

~(6-a)f = z 2 f '' + ( 1 - a ) z f ' ,

and i f , in particular, a = 2n+l, then ~(~-2n-1)f = Lf. This yields the announced
factorization o f L. From (13) i t now f o l l o w s , that ¢(z) is a solution of

(20) 6(6-2n-1)# = z2¢.

THEOREM 2. (Burchnall [17], Chaundy and B u r c h n a l l [18]). Define the differential

operator = (~-1)(6-3)...(~-(2n-1)); then the function ~n(Z) = Q~(~)e
IA
-z is a

solution of (20).

Proof. 6(~-2n-l)~n = ~(6-2n-i)(6-1)(~-3)... (6-2n+l)e -z

 = (8-1) (8-3)... (8-2n-l)6e -z = -(8-i) (8-s) . . . (8-2n-l)ze -z

= -zS(6-2)...(8-2n)e -z = -z(6-2)(8-4)...(8-2n)Te -z

= z(6-2)(8-4)...(8-2n)ze -z = z2(8-1)(8-3)...(8-2n+l)e -z = Z2~n .

Here the first and last equality are justified by the definition of ~n(Z), the

second and fifth one by the eommutativity of the operators 8-a, the third and sixth
one by the definition of 8, and the fourth and seventh by (18). Theorem 2 may also
be proved directly, as follows:
By using (16),
,o k ~ (_l)k
(21) CnCZ) : Qn(8)e -z : % ( 6 ) [ (-1)k k: - ~ k: Qn (8)zk
k=O k=O

(_l)k ~(k)z k
= X k,---~--
k=O
and

(22) z 2,n(Z ) = ~~ (-I)

k! k Qn (k)zk+2 -- k~ 2 (k-2)'
(-i) k Qn (k-2)z k "
k=0 = "

Next,

8¢n ( z ) = (-1)k Qn(k)szk

~ = °~ (-1)k ~k (k)z k = Jl (k-l)'. Qn (k) zk
k=O k=O =

o~ (_l)k k
= ~ (k-3)(k-5)...(k-2n+l)z ,
(k-2) '
k=2

and similarly,

62¢n(Z) = k~2= (k-2)

(-l)k! k(k-3)...(k-2n+l)z k.

Hence,

(23) L¢n = (82-(2n+l)6)~n = k!2 (k-2)!

(_1)k (k-3) (k-5)... (k- (2n+l)) zk

(_l)k
(k-2)' Qn ( k - 2 ) z k = Z2¢n'
k=2 °

by (22).
For l a t e r use we s t a t e the f o l l o w i n g .
k
THEOREM 3. In the series expansion of ~ n ( Z ) , the coefficients of z vanish for
k = 1,3,5 ..... (2n-l).
 12

Proof. Qn(k) = 0 for k = 1,3 ..... (2n-l), so that Theorem 3 follows from (21).

As already observed, (15) and, hence, (20), are satisfied not only by Cn(Z),

but also by ~n(-Z). This is evident also from (22) and (23), in both of which one

only has to suppress the factor (-i) k under the summation sign.
From ¢n(Z) = Qn(g)e "z and the relation between On(Z ) and Cn(Z) it follows that

-Z
On(Z) = eZQn(~)e ,

an identity found already in [17].

For future use it is of interest to record here also the g-forms of the differ-
ential equations for 8n(Z ) and yn(Z):

(24) [g(g-2n-l)-2z(g-n)]gn(Z ) = 0 ,

(25) [26 + z(g-n)(6+n+l)]Yn(Z ) = O.

d
If one replaces g by z dK ' one immediately verifies that these are, indeed,

the same as (3) and (ii), respectively.

4. Equation (Ii) has been generalized in several ways. Krall and Frink [68] intro-
duce two new parameters and write the equation as

(26) z2y"+(az+b)y'-n(n+a-l)y = 0.

For a = b = 2, (26) reduces, of course, to (ii). Following [68], we shall denote

the polynomial solution of (26) (if such exists) by yn(Z;a,b), the generalized BP

of degree n. The reverse BP of degree n, znyn(z-l;a,b) will be denoted by en(Z;a,b).

If we set Y(z) = y(2z/b) and Z = 2z/b, then Y'(z) = (2/b)y'(2z/b), Y"(z) =

(2/b)2y"(2z/b) and we verify that

2z
z2Y"+(az+b)Y'-n(n+a-l)Y = (2z/b)2y"(2z/b)+(a--~ + 2)y'(2z/b)-n(n+a-l)y(2z/b)

= Z2y"(Z)+(aZ+2)y ' (Z)-n(n+a-l)y(Z).

This shows that if y(Z) is a solution of (26) with b = 2, then y(2z/b) is a solution
of the general equation (26). In other words, b is only a scale factor for the
independent variable and not an essential parameter. If we want to keep b = 2, but
let yn(Z) or 8n(Z ) depend also on the parameter a, we write yn(Z;a) and en(Z,a),

respectively.
By Frobenius' method or by direct verification through substitution, it may be
shown that
 13

(27) yn(Z;a) : 7. d (n)zk, where d~ n) = n' (n+k+a-2)(k)

k=O k! (n-k) !2 k

In (27) and hereafter we use the notation u (n) to mean u(u-l)...(u-n÷l). Similarly,
u will stand for u(u+1)...(u+n-l). No confusion should arise between these stand-
n
ard notations and sub, or s u p e r s c r i p t s , l i k e in d~ n) .
n
From these the coefficients f~n) of yn(Z;a,b) = ~ f~n)zk are obtained, by
k:O

replacing z by 2z/b; hence,

(28) f~n) = n2(n+k+a-2) (k)

kl(n-k)!b k

Ck)
One verifies that, for a : b = 2, fEn)r : n.'( n + k ) ( n + k ) : -- =
(a~n) ,

k~ (n-k) '2k 2kk! (n-k) !

in agreement with (8).

Some authors like Obrechkoff [82] and Docker [48] have adopted the normalizations
b = i, or b = -I rather than b = 2. They also set m = a-2 and write m as a super-
script. In their notation, therefore, yn(Z;a) = (-l)nPn (a-2)(-x/2) and, in particu-

lar, yn(Z) = (-l)nPn(0)(-x/2).

S. The corresponding generalization of 0n(Z) is obtained most conveniently by

setting

yn(Z;a,b) = z n O n ( Z - 1 ; a , b ) .

If we substitute this in (26), we obtain by routine computations the differen-

tial equation satisfied by 8n = 8n(Z;a,b ), namely

(29) z0"-(2n-2+a+bz)Snn + hnOn = 0.

For a = b = 2, (29) reduces to

Z0n-2(z+n) 0n + 2n0 n = 0,

which is, of course, (3) with 2~-I = 2n.

The obvious generalization of On(Z) is clearly Cn(Z;a,b) = e-bZ/20n(Z;a'b)" It

is convenient, however, to increase the flexibility of the presentation, by intro-

ducing a new parameter; hence, we set

(30) Cn(Z;a,b,c) = e-CZ0n(Z;a,b).

 14

From this we obtain as particular cases @n(Z;a,b) itself, for c = 0, and

~n(Z;a,b) for c = b/2; in addition also the case c = b will turn out to be of

interest.
If we differentiate (30) twice and use (29) we obtain the differential equation
satisfied by w = #n(Z;a,b,c), namely

(31) zw"-(2n+a-2÷(b-2c)z)w'+(c(c-b)z+(b-2c)n+c(2-a))w = O.

For a = b = 2, c = O, (31) reduces again to (3) with 2~-i = 2n, while for
b = 2c = 2, (31) becomes

(32) zw"-(2n+a-2)w'+(2-a-z)w = 0,

satisfied by ~n(Z;a,2) = e-Zen(Z;a,2).

Finally, for b = c, (31) simplifies to

(33) zw"-(2n+a-2-bz)w'+b(2-a-n)w = 0,

with the solution w = e-bZe (z;a,b).

n

As in the particular case a = b = 2c = 2, it is convenient to "factor" (31)

d
with the help of the first order differential operator 6 = z ~ . From the defini-

tion of 6 follows

~(~+l-a-2n)w d {zw' +(l-a-2n)w}

= z ~-~ = z(zw"+(2-a-2n)w').

By (31) this equals

(34) z2(b-2c)w'+z{c(b-c)z+(2c-b)n+(a-2)c}w.

If b ~ 2c, (34) may be written as z(b-2c)(~-n+az+B)w, with ~ = c(b-c)/(b-2c),

= c(a-2)/(b-2c), so that w = ~n(Z;a,b,c) is the solution of the differential

equation

(3S) 6(~+l-a-2n)w = -z(2c-b)(6-n+az+8)w.

For b = 2c, (34) equals {z2(b/2)2+(a-2)(b/2)z}w and w = #n(Z;a,b) satisfies

8(~+l-a-2n)w = {z2(b/2)2+(a-2)(b/2)z}w.
In particular, if b = 2, this reduces to

~(6+l-a-2n)w = (z2÷(a-2)z)w,
the f-form of (32). If we set here also a = 2, we recover, of course, once more
(20).
 15

6. For b = c, (35) becomes (see (20) i n [17])

(36) 8(~+l-a-2n) w = -bz(~-n+2-a)w,

verified by O n ( Z ; a , b , b ) = e=bZo ( z ; a , b ) .
n

In particular, for a = 2, (361 reduces to

(37) 6(6-2n-l)w = -bz(~-n)w,

with the solution w = e-bZSn(Z;2,b ).

Without restricting ourselves to a = 2, let us assume nevertheless (following

Burchnall [17]) that a is at least a non-negative integer and consider the func-
tions

(381 W(z) = CP(~)e -bz,

with P(6) = (~-n-a+l)(~-n-a)...(6-2n-a+2) and constant C.

By (191

(39) W = Cz 2n+a-I dn (z-n-a+le-bZ).

dz n

We claim that for any constant C, W is a solution of the differential equation (36).
This can be proved directly, as in the second proof of Theorem i, but an easier
proof is by use of (181. On account of the homogeneity of (36), it is sufficient
to verify the claim for C = i. We obtain successively,

6 (6+l-a-2n)W = 6(6+l-a-2n)[(6-n-a+l)...(6-2n-a+2)e -bz]

= (~-n-a+l)... (6-2n-a+2) (~-2n-a+l)~e -bz =

(6-n-a÷l)...(6-2n-a÷l)(-zb)e -bz = -bz(6-n-a+2)...(6-2n-a+2)e -bz

= -bz(6-n-a+2) [(6-n-a+l)...(6-2n-a+2)e -bz] = -bz(8-n-a+2)W,

as claimed. Here the first equality follows from (38), the second from the commu-
tativity of the operators 6-c, the third from the definition of 6, the fourth from
(18), and the last two are obvious.
One can easily verify that not all solutions W of (36), or, equivalently, of
bz
(33), are such that e w = p(z), a polynomial.
It follows that at most one of any two linearly independent solutions of (33)
(or (361) can be of this form. However, we do know one such solution, namely (38).

Indeed, ebzw(z) = cebZp(~)e -bz is obviously a polynomial and, specifically, a con-

stant multiple of On(Z;a,b I as already mentioned immediately after (33) and (36).
 16

It follows, using (39) t h a t , f o r an a p p r o p r i a t e constant C, @ n ( Z ; a , b ) =

ebZw(z) = cebZz 2n+a-1 dn ( z - n - a + l e - b Z ) . In o r d e r t o d e t e r m i n e the constant C, we

dz n
observe that the highest power of z furnished by the Leibniz form of the derivative

dn (z-n-a+le -bz) comes from the term z -n-a+l dn (e -bz) and il equal to
dz n dz n

(-b)nz-n-a+le-bZ; hence, the right hand side equals

Ce bz z2n+a-l[(-b)nz-n-a+l+ (lower powers of z)]e -bz =

czn(-b)n+ (lower powers of z). The coefficient of zn in en(z;a,b) is i, so that

C = (-l)nb -n and this finishes the proof of the Rodrigues-type formula

(40) en(Z;a,b ) = (-l)nb-nebzz 2n+a-I dn (z-n-a+le-bZ).

dz n

F o r b = 2, i n p a r t i c u l a r ,

en(z;a) = ( _ l ) n 2 - n e 2 Z z 2 n + a - 1 _ _d n ( z - n - a + l e - 2 Z )
n
dz

and (see [ 17 ]), if also a = 2, then

( 1)n2-ne2Zz 2n+l d n (z-n-le-2Z) .

en(Z ) = _
dz n

7. It is clear that all values of b # 0 are admissible and lead to BP. For b = O,
however, (26) reduces to z2y '' + azy' - n(n+a-l)y = 0. The solutions are of the form
r
y = z , where the r are solutions of the equation r(r-l)+ar-n(n+a-1) = 0. These
solutions are n and l-a-n so that the general solution of (26) for b = 0 becomes
n 1-a-n
y = ClZ +c2z For n > a+l, this is a polynomial only if c 2 = O, when it

reduces, essentially to zn. This is a trivial case of no further interest here.

The parameter a may take arbitrary values. This is clear from the form of

the coefficients dk(n) (see (27)) and fk(n) , which remain well defined for all complex

values of a. For some of the theory to be developed, however, the values a = 0,

-1,-2,... lead to special difficulties and may have to be discussed separately. So,
e.g. if a = l-n, equation (26) becomes

y" _ n-I b
y' z Z
2
n
z
and has the general solution y = --~ + ClZ+C2+bz log z.
 17

n
This is a polynomial only if b = 0, when it reduces to y = ClZ +c2, also of no

further interest.
Whenever such cases arise, we shall tacitly or explicitly assume that a is
not zero, or a negative integer.
 CHAPTER 3
RECURRENCE RELATIONS

1. It is well known (see [I]) that the solutions K (z) of (2.1) satisfy recurrence
relations, such as

(I) K _I-K + 1 = -(29/z)K ,

(2) K'v = -K~-I - (~/z) K ,

(3) K _I+K + 1 = -2K$,

(4) K' = -K~+I + (~/z)K v.

From (2.5) and c o = (2n)'/2nn!

•
it follows that Kn+i/2(z) = (~/2)i/2e-Zz-n-i/2@ (z)
n "

If we substitute this in (i) for Kv(z), we obtain after routine simplifications

(see [17])

(5) @n+l(Z) = (2n+l)@n(Z) + Z2gn_l(Z).

On account of @n(Z) = eZ~n(Z), the same recurrence relation holds also for the

functions Cn(Z):

(6) ~n+l(Z) = (2n+l)~n(Z) + Z2~n_l(Z).

-i n+l
If we replace in (5) z by z , multiply the result by z and recall that

zn@n(Z -I) = yn(Z), then we obtain (see [68]) the recurrence relation for the BP yn(Z):

(7) Yn+l(Z) = (2n+l)zYn(Z) + Yn_l(Z).

The recurrence relations (5) and (7) allow us to compute successively the polynomials
8n(Z ) and yn(Z) with very little effort. We obtain as first few polynomials,

starting from @0(z) = 1 and @l(Z) = l+z, the following:

@2(z) = 301(z) + Z2@o(Z) = z 2+3z+3; 03(z) = 5@2(z) + Z2Ol(Z) =

5(z2+3z+3) + z2(z+l) = z 3+6z 2 +15z+15, etc.

Similarly, starting from y0(z) = I, yl(z) = l+z, we find y2(z) =

3zYl(Z ) + y0(z) = 3z2+3z+l, y3{z) = 5zY2(Z ) + yl(z) = 5z(3z~+3z+l) + z+l =

l$z3+lSz2+6z+l, etc. These results illustrate the relations znyn(Z -I) = @n(Z) and

znOn(Z-I ) = yn(Z).
 19

From yO(z) = Co(Z) = i, yl(z) = Ol(Z ) = z+l, (S) and (7) also immediately

follows

THEOREM i. The coefficients of all BP are positive, rational integers.

I f we p r o c e e d i n t h e same way, by u s i n g a l s o (see ( 2 . 5 ) s o l v e d f o r ~vK ( z ) ) ,

~ K v! =
-vz-V-le-Ze(z,v)-z-Ve-Zo(z,.) + z-Ve-Zo'(z,.) we o b t a i n from (2) t h a t On,

Cn' and Yn s a t i s f y

(8) 8n(Z ) = On(Z ) - ZOn_l(Z),

(9) @n(Z) = -ZCn_l(Z),

and

(io) z2y~(z) = [nz-l)Yn(Z)+Yn_l[Z),

respectively. For (8), see [17], for (i0) see [68]. One may remark the particular-
ly simple form of (9), which may be new.
Similarly, one obtains from (3) the following recurrence relations:

(ii) 2zOn(z ) = (2z+2n+l) On(Z) - (Z2en_l(Z)+On+l(Z)),

(12) 2z~n(z ) = -Z2~n_l(Z)+(2n+l)~n(Z)-¢n+l(Z),

(13) 2z2yn[z) = (Yn_l(z)-yn(Z))+(Yn+l[z)-yn(Z))-Zyn(Z)

= Yn_l (z) - (2+z) yn (z) +yn+ 1 (z) .

Finally, (4) is a linear combination of (2) and (3). Either directly, as

before, or by combining [8) and (ii), or [9) and (12), or (i0) and (13) we obtain
(see [17] for (14)) the corresponding relations

(14) zen(z ) : (z+2n+l)On(Z)-@n+l(Z),

C15) z~(z) = (2n+l)¢n[Z)-@n+l(Z),

(16) z2yn_l(Z) = yn(Z)-(l+nz)Yn_l(Z).

By combining above formulae, or by starting from any other among the numerous
recurrence relations known in the theory of Bessel functions, many other recurrence
relations for the polynomials 8n(Z ) and yn(Z), or for the functions Cn(Z) can be
obtained.
For future use we observe that (7) [and similarly (5) and (6)), can be &eneral-
ized to read
 2O

(7') yn(Z) = Pm(Z)Yn_m(Z) + Qm_l(Z)Yn_m_l(Z),

whore Pm(Z) and Qm_l(Z) a r e p o l y n o m i a l s i n z o f d e g r e e s m and m - l , r e s p e c t i v e l y .

Indeed, (7) i s t h e i n s t a n c e m = 1 and t h e g e n e r a l s t a t e m e n t f o l l o w s by i n d u c t i o n

on m: Assuming ( 7 ' ) a l r e a d y v e r i f i e d , we r e p l a c e Yn_m(Z) by u s e o f (7) and o b t a i n

Yn = Pro(z) ((2n-2m+l)ZYn-m-1 + Yn-m-2 ) + Qm-l(Z)Yn-m-1

= {(2n-2m+l)ZPm(Z ) + Qm_l(Z)}Yn_m_l + Pm(Z)Yn_m_2

= Pm+l(Z)Yn_m_l + Qm(Z)Yn_m_2.

This shows t h a t ( 7 ' ) h o l d s a l s o f o r m + l , h e n c e f o r a l l m f n and, i n c i d e n t a l l y shows

that Pm+l(Z) = (2n-2m+l)ZPm(Z) + Qm_l(Z), Qm(Z) = Pro(z).

Formulae ( 5 ) , (6) and (7) p e r m i t us t o e x t e n d t h e d e f i n i t i o n s o f On(Z), ~ n ( Z ) ,

and yn(Z) t o n e g a t i v e v a l u e s o f n. For @n(Z) and y n ( Z ) , t h i s has a l r e a d y been done

i n ( 2 . 9 ) and (2.12) and one has t o v e r i f y t h a t t h e two e x t e n s i o n s a r e c o n s i s t e n t ,

for instance that (7) l e a d s t o Y-n = Yn-l"

From (7) i t f o l l o w s t h a t Yn-1 = Y n + l - ( 2 n + l ) z Y n ; h e n c e , Y-1 = Yl-ZY0 = l + z - z =

1 = Y0' i n agreement w i t h ( 2 . 1 2 ) . Assuming t h a t (7) r e m a i n s c o n s i s t e n t w i t h (2.12)

for all subscripts n o t e x c e e d i n g n, by ( 7 ) ,

Y-n-1 = Y - n + l - ( - 2 n + l ) z Y - n = Yn-2+(2n-1)ZYn-1 = Yn

as we wanted t o show. The c o n s i s t e n c y o f ( 2 . 9 ) w i t h (5) f o l l o w s i m m e d i a t e l y from

what p r e c e d e s , and can a l s o be checked d i r e c t l y by t h e same method.

2. The r e c u r r e n c e r e l a t i o n (7) p e r m i t s us t o g i v e a new d e f i n i t i o n t o t h e BP Yn(X),

as a d e t e r m i n a n t . Indeed, f o l l o w i n g [85] l e t us c o n s i d e r ( 7 ) , w i t h n r e p l a c e d by
n-l, as a homogeneous l i n e a r equation in the three "unknowns" Yn' Y n - l ' Y n - 2 '

Yn Yn-2
namely ~ U ] - - ZYn_1 - 2n_-----[ = 0. Next, we r e p l a c e n s u c c e s s i v e l y by n - l , n-2,...,
Y3 Yl
3,2, and I. For n = 3 the equation reads ~-- -zY2 - 7 = O. For n = 2 we write it

Y2 1 Yo
(inhomogeneously) in the form ~-- zy I = ~ (= fL). Finally, for n = 1 (the last

one), it is Yl = z+l (= zy 0 + y_l ). We solve this system of n linear equations in

Yn' Yn-i .... 'Yl by Cramer's rule. The determinant of the system is
 21

1 1
-z 0 0
2n-I 2n-I

1 1
0 -z 0
2n-3 2n-3

D =

1
0 0 0 0 0 -Z

0 0 0 0 0 • .• 0

and YnD = Dn, where D n is obtained by replacing the first column of D by the column

vector of second members of the equations. The entries of this vector are, as seen,
all zeros, except for the last two, which are 1/3 (from n = 2) and z+l (from n = i),

respectively. It is clear that D = {(2n-l)(2n-3)...3.1} -I, because D is triangular;

hence it is equal to the product of its diagonal elements• As for

1
-z 0 ... 0 0
2n-i
1 1
0 0
2n-3 -z 2n-3 " ""

1
0 -z 0 0
2n-S " ""
D =
n
• . . . •

1 1
0 0 0 ... ~ -z
Y
z+l 0 0 0 ... 0 1

we shift the first column to last place, taking into account the sign change

(_l)n-l•

Next, we change the signs of the entries in the last column, so that

D = (-l)nMn , where
n

1
-Z
0 0 0 0
2n-i
1 1
-Z 0 0 0
2n-3 2n-3
1
0 0 0 0
2n-5 -z
M
n =
• ° . . :

0 0 0 1 -z 1__
~ -3

0 0 0 0 1 -z-1
 22

Finally, yn(Z) = Dn/D = (-i)n.i.3.5 ... (2n-l)M n.

The corresponding results for the generalized BP yn(Z;a,b) do not appear equally

useful, because they are rather complicated and lack symmetry (see, however [74])•
For On{Z ) on the other hand, we may start from (5) {with n replaced by n-l)
instead of (7) and obtain in exactly the same way that
DO n = Dn, where

2
I - (2n-l) -z 0 ... 0 0 0

2
0 1 - (2n-3) -z ... 0 0 0

D=
"2
0 0 0 0 ... 1 -5 -z -- i

0 0 0 0 ... 0 1 -3

0 0 0 0 ... 0 0 1

and
2
- (2n-l) -z 0 . .. 0 0 0
2
i - (2n-3) -z ... 0 0 0

0 1 -(2n-5) ... 0 0 0

D =(-i) n
n
2
0 0 0 ... -5 -z 0
2
0 0 0 ... 1 -3 -z

0 0 0 ... 0 i -(l+z)

with On[Z ) = D n.

3. From the recurrence relations for the BP, we obtain recurrence relations for
their coefficients We denote as before {see (2.8)) by a (n) the coefficient of z m
• m

in yn(Z), normalized by Yn(O) = i.

n

By substituting (2.10), i.e. yn(Z) = a (n) z TM into (7) we obtain

m
m=O

I I for m = 0 ,

(17) a (n) = ( 2 n = l ) a ( n ? l ) + a (n-2) for 1 < m < n-2 ,

m m-i m

(2n- i) am(nil) for m = n-i and m = n.

 23

By substituting

a(n) =
I_In-l)
(2.10) into (i0) we obtain

~0 : 1

(n_m+l)a(n)+a (n-l)
for m = O,

(18) for 1 _< m <_ n-l,

m m-i m
a (n)
for m = n.
n

Finally, if we substitute (2.10) into (16) we obtain

1 ~ I I) ~ for m = O ,

(19) a m(nl = (m+n-l)a - +a n-l) for 1 < m _< n-i ,

(2n-l)a~nil) for m = n.

In each s e t o f f o r m u l a e (17), (18), and ( 1 9 ) , the central formula holds for

all cases, if we remember that a~ n) = 1 and a (n) = 0 f o r m > n , o r m < 0.

m

4. Recurrence relations similar to (5) and (7) h o l d f o r t h e g e n e r a l i z e d BP and f o r

@n(Z'a)" However, t h e i r usefulness seems t o b e l i m i t e d and t h e p r o o f s are routine.

For t h e s e reasons it should suffice to only list a few o f them h e r e (some o f

these appear in [68]3:

(20) (n+a-1)(2n+a-2)Yn+l(z,a) = [(2n+a)(n-l+a/2)z+a-2](2n+a-1)Yn(Z,a )

+ n (2n+a)Yn_l (z, a ) ,

(21) z2(2n+a-2)Yn(Z,a) = [n(2n+a-2)z-2n]Yn(Z,a)+2nYn_l(z,a ) ,

(223 z 2(2n+a-2)yn_ l(z,a) = 2 ( n + a - 2 ) Y n ( Z ,a) - {(n+a-2)(2n+a-2)z+2n}Yn_ l(z,a) ,

(231 (n+a-1)(2n+a-2)en+l(z,a) = [(2n+a)(n-l+a/2)+(a-2)z](2n+a-1)6 n(z,a)

+ n ( 2 n + a ) Z2en_l ( z , a ) ,

(24) (l+(a-2)/2n)e~(z,a) = en(z,a ) - Zen_l(z,a ).

Recursion (233 holds unchanged if we replace @n(Z,a), by Cn(Z,a) = en(Z,a)e-Z;

however, (24) becomes

(251 (2n+a-2)¢~(z, a) = -(a-2)~n(Z,a)-2nZ~n-l(z'a1"

One sees from (25), that the surprizing simplicity of (91 was, in some sense,
rather accidental.
 24

5. The reader may have been surprized that in Chapter 2 nothing was said concerning
generalized BP for negative subscripts. This was no simple omission. The recur-
rence relation (20) permits us to define y_n_l(z;a,b) recurs ively, if we know Y-n

and Yl-n" The first step, however, the determination of y l(z;a,b) fails because

in (20) the factor n(2n+a) of Yn-i vanishes for n = 0. This is not the case when

a = 2; in that case n (in fact 2n(n+l)) is a factor common to all terms and is
cancelled. Also the difficulty with y_n(Z;a,b) for arbitrary real a is not an
accident. Indeed, let us consider the differential equation (2.26) with n = -I,

z2y '' + (az+b)y' + (a-2)y = 0.

We easily find that it has a single solution regular at z = 0 and normalized by

y(0) = i, given by

(re+a-3) (m)
(26) Y(Z)(=Y-l(z;a'b)) = m=0~ Cm zm, with cm = (-l) m bm

This solution reduces to a polynomial if and only if a equals an integer not in

excess of 2. For a = 2, y(z) (i.e., y_l(Z;2,b)) reduces to the constant
1 (= y0(z;2,b)), as we know from (2.12). The other case of interest is a = I, when

y_l(Z;l,b) = l+z/b = yl(Z;l,b) and, by (20) with a = i, y_n(Z;l,b) is a polynomial

of exact degree n. However, the theory for a = 1 has never been developed much
further, the suggestion of Krall and Frink ([68], bottom of page 109) notwithstand-
ing. See, however, [99] and [i00].
For the other cases when y_l(z;a,b) is a polynomial, that is, for a = 0, or
a equal to a negative integer, y_l(z;a,b) is given by (26) and once we know

y_l(z;a,b) all other y_n(Z;a,b) can then be obtained recursively from (20), and are

polynomials of degree n+l-a.

Unfortunately, some parts of the theory break down, or become complicated pre-
cisely in these cases, when a = 0, -I, -2, ... and so we shall not pursue this
matter further.
 CHAPTER 4
MOMENTS AND ORTHOGONALITY ON THE UNIT CIRCLE

I. The theory of moments has a long history. Its modern phase may be said to have
started with Stieltjes [$4] and it was considerably advanced by the work of Tcheby-
cheff [ST] and Hamburger [26], [27]. It is intimately connected with the idea of
orthogonality.
In the classical setting, one starts with a function w(x), real and non-nega-
tive over an interval a ~ x ~ b of the real axis, and considers the "moments"

mn = f ab w (x) x n d x . Set

m0 mI ... mn_ 1

An = A n(w) = mI m2 ... mn , A0= 1

mn_ 1 mn ... m2n_2

and let us assume that An ~ 0 for all n = 1,2,....

Next, we recall that the polynomials with real coefficients form a vector
space S over the reals. We define an inner product on S by setting

(f,g) = I ab w(x)f(x)g(x)dx.

Clearly, (f,g) = (g,f). If (f,g) = O, we s a y t h a t f a nd g a r e o r t h o g o n a l to each

other.
0 2
One may t a k e as a basis for S the powers of x, x = l, x, x .... which are
linearly independent.
By u s e o f t h e Schmidt method (see [35], p. 152]) - sometimes called t h e Gram-
Schmidt procedure - we c a n f i n d a finite sequence of linear combinations of those
powers, i.e., of polynomials Pn(X) of successive degrees 0,1,2 .... , such that each
n
n-m
one is orthogonal to all preceding ones. If Pn(X) : ~ CmX then it is, of
m:0
course, sufficient to insure that (xk,pn(X)) = 0 for k = 0,I ..... n-l. If we
n
substitute here ~ cmxn-m for Pn(X) and use the definition and linearity of the
m=0
inner product, we obtain n homogeneous equations in the n+l coefficients
Cm(m = 0,1 ..... n) of Pn(X). These have a unique solution, up to an arbitrary

multiplicative factor, if and only if their determinant is different from zero.

One immediately verifies that this determinant is precisely An and, by assumption,

An ~ 0. We may take advantage of the mentioned arbitrary constant factor of the

Cm'S , in order to obtain also (pn,Pn) = 1. A set of polynomials that satisfies

 26

these conditions, i.e. (pn,Pm) = 6nm, is called orthonormal. One may show (see [if6],

2.2.6) that

m0 ml "'" mn l

ml ?2 .mn+l

i
-1/2
Pn(X) = (An_ 1. An)
mn - 1 nln m2n- 1
1 x xn

For more d e t a i l s see [56].

This situation is usually generalized as follows (see, e.g. [56], [69], or

[63]): One replaces the interval [a,b] of the real axis by an arc (infinite, or
finite, open or closed) of a curve C in the complex plane and defines the inner
product by (f,g) = f w(z)f(z)g(z)Idzl, where w(z) is a (generally non-holomorphic)
C
function of the complex variable z, real and non-negative on C. One observes that
instead of the commutativity (f,g) = (g,f), the inner product now satisfies the
equation (f,g) = (g,f). Much of the classical theory goes through in this broader
setting. For the case in point, however, a different kind of orthogonality is of
interest• It occurs already in [68] and has been studied in detail by John W. Jayne
[63] who calls it the Bessel alternative. Still more general points of view are
possible• R.D. Morton and A.M. Krall [80] start from a weight "function" that is,
in fact a distribution. In the case of the classical orthonormal polynomials
(Jacobi, Laguerre, Hermite), use of the Fourier transform permits to recover the
usual weight functions from the distributional ones• This approach is not entirely
successful in the case of the BP. A recent paper of A.M. Krall [66] however, com-
pletes this work.
In spite of the great interest of this problem, it will not be discussed here
further. Indeed, precisely because of its breath, it encompasses much that is not
directly related to BP and the interested reader is directed to the original papers.
We shall restrict ourselves to problems of moments directly pertinent to BP
and, indeed, many of the results of this chapter can be found in [68] and in [17],
although not always in the same generality.
For arbitrary real or complex a # 0, -i, -2, ... and z # 0, set p = p(z;a,b) =

rfa) b n n
n 0 r(a+n-1) (- ~) and l e t pn(Z) = ~ cmzm be an a r b i t r a r y polynomial of degree n.
= m=O

The k - t h moment of pn(Z) on t h e u n i t c i r c l e , with weight f u n c t i o n p is d e f i n e d by

Mk(Pn;P) = 2 -1~ I ] z[=l z k Pn (z) (r [g r ( aF+(a)

r-1) (-b/z)r)dz"
0

Here we may s u b s t i t u t e the e x p l i c i t form of pn(Z) and i n t e r c h a n g e summation and

 27

integration b e c a u s e t h e sum c o n v e r g e s e s s e n t i a l l y like ~

Ibln
n! , pn(Z) c o n t a i n s
n=O
only a finite number o f t e r m s , and t h e p a t h o f i n t e g r a t i o n is finite. Most terms
v a n i s h and we o b t a i n
n+k÷l
Mk(Pn,p) = ~ Cm(r(a)/r(a+r-1))(-1)rb r = ~ (-1)rbrcr_k_lr(a)/r(a+r-1).
k+m-r=-i r=k+l
n
We a r e i n t e r e s t e d , in p a r t i c u l a r , in t h e c a s e when Pn (x) = y n ( Z ; a , b ) = ~ f(n)zm
m
m=0
where ( s e e ( 2 . 2 8 ) ) f(n)=m ( ~ ) b - m ( n + m + a - 2 ) ( n + m + a - 3 ) ' ' ' ( n + a - 1 ) " The sum may t h e n be

~Titten successively, as f o l l o w s :
n+k+l
[ (_l)rbr n! ( n + r - k + a - 3 ) . . . ( n + a - 1 ) r ( a )
r=k+l (r-k-1):(n-r+k+D~r-k-lr(a+r-1)

(_b)k+l n n! (n+v+a-2) . . . (n+a-1) • ( n - v + l ) . . . (n+k+l) =

(k+n+l)~ ~ (-l)V v! a ( a + l ) . . . ( a + v + k - 1 )
v=r-k-l=O

(_b)k+l n nl (n+k+l) ! (n+v+a-2) . . . (n+a-1) =

(k+n+l) ! ~ ( ' l ) V v! (n-v) ' (k+v+a-1) (a+l)a
v=0 . . . .

(-b) k+l ~ (-1) (n+v+a-2):..(n+a-1)

v($) (k+v+a-1) (a+l)a
v=0

so that

(1) Mk(Yn,p ) =

(_b)k+l n 1" v ' n " ...

(k+n+a-1) (a+l)a ~ (-) [v ) ( n + v + a - 2 ) (n÷a)(n+a-1}(k+v+a)(k+v+a+l)...(k+n+a-1).
"'" v=0

We claim that the sum is independent of a and is equal to the constant

k(k-l)...(k-n+l). If we substitute this value for the sum we obtain

THEOREM i. Let Yn = Yn (z;a'b) and p = p(z;a,b); then

r(a) r (k+l)
Mk(Yn;P) = (-b)k+l r(k+a+n)r(k+l-n)

Our claim concerning the sum may be well-known (see e.g. [68], p. 113); however,
as no reference appears to be readily available, a proof of the claim follows. It
is based on two lemmas, the first of which is indeed well-known.
t
k-n (k).
LEMMA i. (see (3.4) in [ZO]). ~. (n)(t_v) =
v=0
 28

k n k-n
Proof. Z (k) xt = ( x + l ) k = ( x + l ) n ( x . l ) k - n = Z (n) xv Z ck-n)xWw
t=O v=O w=O

k k n
n k-n
Z xt Z (~) (k-n)w : Z xt Z (v) (t_v)
t=0 v+w=t t=0 v=0
O~v~n
O~wgk-n

and the lemma follows from the comparison of the coefficients of x t in the first and
last member.

LEMMA 2. The function

n
(23 f(x) = Z (_l)v (v) r(k+~+x+l)
r (n+x) (k+v+x) ( k - l + v + x ) . . . (n+v÷x)
v=O

i 8 i n d e p e n d e n t o f x, and has t h e v a l u e k ( k - 1 ) . . . (k-n+l); in pc~ticulav, for k < n,

one has f ( x ) = O.

Proof. For k > n (2) is equivalent to

n
7 C - l ) v ( ~n (k+n+x) ( k + n + x - 1 ) . . . (k+v+x) . . . (n+v+x) . . . (n+x)
(3) f(x)
"V" (k+v+x) • (n+v+x)
v=O ""

f(x) is a rational function and c o u l d h a v e p o l e s a t mos t f o r x = -m, n < m 5 k+n.

In fact, as follows from ( 3 ) , all these poles cancel and f ( x ) is a polynomial of
degree at most n. We s h a l l show t h a t lim f(x) exists for all these k+l values of
X -~ - m

m in the g i v e n r a n g e and t h a t all these limits equal k!/(k-n)!, independently o f m.

It then follows that the polynomial f(x) of degree n < k+l equals k'/(k-n)! identi-
cally, as claimed.
To find the limit, set m = n+t, 0 < t < n, t E Z, x = -n-s. Then, by (2),

r(k-s+l) n 1 v n (t-s)...C-s)
lim
s-~t
f(-n-s) = lim
S-~t
(t-s) (t-l-s).
""
(-s) r(-s) v=O
[ (-) (v) (k*v--~n-sS_::~-s)

t
7 V n t (t-l)... (t-v+l)
= r(k-t+l) lira ~ (-1) (v) (k+v-n-t)' '
S+ t v=O

because for t < v _< n , -s < t - s < v - s and t h e v a n i s h i n g factor of the numerator is
not cancelled by a c o r r e s p o n d i n g factor of the denominator. It follows that

t 1 v n t. w (k-t),t, t 1 r n k-n
lira f(-n-s) = r(k-t+l) Z (-) (v) (t-v)!(k+v-n-t), = (k-n) l Z (-) (v)(t-v)"
s ÷ t v=O v=O

By Lemma 1, the l a s t sum equals (tk) , so t h a t indeed

lim f(x) = (k-t)'t~ kl = k!

(k-n)! t'(k-t)' (k-n), ' as claimed.
X "+ - m
 29

Proof of Theorem i. In the sun in (I) set a-I = x. The sum becomes
n
(-l)V(~)(n+v+x-l)...(n+x).(k+v+x-l)...(k+n÷x)
v=0

n 1 v n (k+n+x)(k+n+x=l)...(k+v+x)...(n+v+x)...(n+x) _ k2
= [ (") (V) (k+v+x) (n+v+x) (k-n)!
v=0 "'"

by Lemma 2. I f we s u b s t i t u t e this value in (1), the theorem follows.

Concerning the range of validity o f Theorem i , one o b s e r v e s t h a t t h e r i g h t hand

side has a finite value for all a # -m, m an integer of the interval 0 ~ m < k+n-l.
If a approaches one of the excluded values, then Mk(Yn(Z;a,b),0) ÷ ~.

2. From Theorem i follow several corollaries.

COROLLARY i. (see (57) in [68]; also Lemma 16.1 in [80]). For 0 f k < n,

Mk(Yn(Z;a,b);p) = 0.
Formally, the corollary follows immediately from Theorem I. It can also be
verified directly. Indeed, now we obtain instead of (I):
n
~ (yn,p) (_b)k+l [ v n (n+v+a-2) . . . (n+a-l)
= (-i) (v) (k+v+a-l) (a+l)a
v=0 "'"

and the sun, again with a-I = x, equals

n
( - )1 v n
(v) ( n + v + x - l ) ( n + v + x - 2 ) . . . (k+v+x+l).
(x÷l)... (x÷n-l) v=0

n . 1"v'n" n+v+x-i
This sun vanishes for k < n. Indeed, let F(y) = [- ) [vjy ; then
v=0

~n-k=l F n
8yn-k--------~= v=0[ (-l)V(~)(n+v+x-l)'''(k+v+x+l)yk+V+X"

The sun to be evaluated is the last one for y = i. On the other hand,

F(y) = yn+x-i n~ (-I)n (v)y

n v = yn+x-l(l-y)n.
v=0
Hence, by Leibniz' rule, the (n-k-l)-th derivative is of the form

n+x-l-s.. .n-t
asy If-y) and vanishes for y = i, because n-t > n-(n-k-l) =
s+t=n-k-i

k+l > 1.
By taking for a a natural integer and, in particular a = 2, we obtain from
Theorem I the following corollaries:

COROLLARY 2. (see [ 6 8 ] ) . For i n t e g r a l a > O,

 30

(a-l) !k!
Mk(Yn(Z;a,b);p) = (_b)k+l
(k+n+a-l)' (k-n) ! "

k!
COROLLARY 3. (see [68]). For a = 2, Mk[Y n (z;2,b);p) = (-b)k+l (k+n+l) ~ [k-n) :

(_l)n(b/z) n e-b/z
REMARK. For a = 2, p[z;2,b) = n! = so, that Corollary 3 may be
n=0
rephrased as

COROLLARY 3' [see [68]). For k > n

1
zkyn(Z;2,b)e-b/Zdz = (_b)k+l k'
; the integral vanishes for
2~i $1z]=l (k+n+l) .'(k-n) '

k<n.

For a = b = 2, in particular, we obtain

COROLLARY 4. (see [68]).

(_2)k+ik!
Nk(Yn(Z),p) = Mk(Yn,e-2/z ) = ~ z l = l zkyn{Z)e-2/Zdz = (k+n+l) I (k-n) ' fork >n

and vanishes for k < n .

COROLLARY S. (see [17]). if k ~ n, then

i
2~i I[zl=l Yk(Z;a'b)Yn (z;a'b)0(z;a'b)dz = 0.
k
Proof. Let k < n; then yk{z;a,b)__ = [ flkjzm,:
~ with 0 < m < n; hence, the result
m
m=0
follows from Corollary I.

COROLLARY 6. (see [68], [17]).

1 ~ bn! r (a)
2~i IIzI=l y (z;a,b)p(z;a,b)dz = (-i)n+l 2n+a-i I" ( n + a - l ) "

Proof. By Corollary S, ~ i I]zi= I y~(z;a,b)o(z;a,b)dz

1 f[ f~n)znyn(z;a,b)p(z;a,b)dz = f~n)Mn(Yn(Z;a,b) ,p)

2~i zI=l

and the result follows from Theorem I and (2.28).

By setting a = b = 2, one obtains from Corollary 6 also

1 2 2
COROLLARY 7. [see [68]) 2~1 IIzI=l Yn (z)e-2/zdz = (-l)n+l - -
2n+l "

3. If we replace yn(Z;a,b) by zn@n(z-l;a,b) in Theorem i and its cdrollaries, we

obtain the following results.
 3~

THEOREM 2. Let Pl ( z ; a ' b ) = z-2 ~ r(a)

r(a+r-1) (-bz) r a n d s e t
r;0

%(en;01) = Mk(On(Z;a'b);Pl ( z ; a ' b ) ) = 2 -1~ f l z [ = l z -k (z-nen ( z ; a , b ) ) Ol(z;a,b)dz;

then

~lk(On;O1) = (_b)k+l r(a)r(k+l)

r(k+a+n) r(k+l-n) "

COROLLARY 8. For k < n, ~(en,Ol) = 0.

bk+l(a-l)!kl
COROLLARY 9. For integral a > 0, {~(@n;Pl) = (-i) k+l (k+n+a-l)!(k-n): "

(-bz) r -2 -bz
For a = 2, pl(Z;2,b) = z -2 [ r! = z e and we obtain
r=0

COROLLARY 10. F o r a = 2, Mk(On'Z-2e-bZ) = (-b)k+l k~

(k+n+l)!(k-n)!

In particular, for a = b = 2, one obtains

(_2) k+l k!
COROLLARY ii. For a = b = 2, Mk(@n,Z-2e -2z) (k+n+l) ~ (k-n) ' "

COROLLARY 12. If k ~ n, then

i ii z (z-kOk(Z;a,b))(z-nOn(Z;a,b))Pl(z'a,b)dz = 0
2~i I=i ' "

COROLLARY 13. If k = n, then

1 i Izl=l (z-nOn(Z;a,b)) 201(z;a,b)d z = (-1)n+lb 2n+a-1

n! r(a)
r(n+a-1)
2~i

In particular, for a = b = 2, the following Corollary holds:

COROLLARY 14.
1
2--~/izl=l (z -n@n(Z)) 2z-2e-2Zdz = (-l)n+l 2
2n+]

4. In the case of positive integral a , Corollaries 12 and 13 may be written in

a somewhat different way.
Let us observe that in this case

01(z;a,b ) = z -2 r~0 a(a+l-~.]~-(a+r-2)

(-bz)r = (a-1)!z -2 ~[ (_bz)r
(a+r-2)'
= r=0 "

= b2 (a_l) : (_bz)-a ~ (-bz)

,~=a-2

= b2(a_l):(_bz)-a(e -bz_ ai3 -(-bz)

- ) .r
r!
r=0
 32

Hence, / [ z [ = l (z-kOk(z;a'b))(z-nen(z;a'b))Pl (z;a'b)dz =

ai3 (-bz) r
(-l)ab2-a(a-l) ! llzl= 1 z-a-k-nek(z;a,b) en(Z;a,b)(e-bZ- - - ) dz
r=O r~

= (_b)2-a(a_l)~ I I z l = 1 z-a-k-nOk(z;a,b)On(Z;a,b)e-bZdz.

Indeed, for 0 < r ¢ a-3, z - a - k - n + r has an exponent t = -a-k-n+r ~ - k - n - 3 ;

hence, a l l neglected i n t e g r a l s are of the form I l z l = l Z - t O n ( Z ; a , b ) O k ( z ; a , b ) d z and

k+n
vanish, because the highest power of z that One k can contribute is z

It follows that for a e Z + the integrals in Corollaries 12 and 13 may be

written as

(_b)2-a(a_l)I ~1 I [ z l = 1 z -a-k-no k t"z"' a ' b ) O n ( z ; a ' b ) e - b z d z = 6nk(-1)n+lb (2n+a-1)(n+a-2)!

n!(a-1)'

As a consequence the following Corollary holds:

COROLLARY 1S. (see [17]). For k = n,

1 Yl z i= 1 z
-a-k-n.. (_l)n+a+lba-1 n!
2~i Vk [z ;a,b)en(Z ;a,b)e-bZdz = (2n+a-l).(n+a-2)!'

for k # n the i n t e g r a l vanishes.

I f we r e p l a c e here z by z -1 and write yn(Z;a,b) for z n e n ( Z - 1 ; a , b ) , we obtain

COROLLARY 16.

2~ii ilz[=l za-2yn(Z;a'b)Yk(Z;a'b)e-b/Zdz = 6kn(-l)n+a+Iba-i (2n+a-l)n"(n+a-2)' "

For a = b = 2, Corollaries 15 and 16 reduce to Corollaries 14 and 7, respectively°

Finally, if we recall that,by (2.291,¢n(Z;a,b) = e-bZ/2en(Z;a,b), we obtain

from Corollaries 12, 14 and 15 that, for a c E +

1 ii -a -k -n = ( l)n+a+iba-i n!
2~i zi=l Z (z ¢k(z;a,b))(z %(z;a,b))dz - (2n+a-l]-(n+a-2)

if k = n, = 0 if k # n.
i 2
In particular, ~ /izl= I z-2(z-nCn(Z))2dz = (-I)n+l 2n+l "
 33

Note added August 30, 1978. In the present chapter we discussed only the orthogona-
lity on the unit circle and the corresponding moments. In fact, it was known from
general considerations, going back to the work of Stieltjes, that a function ¢(x),
of bounded variation, ought to exist, such that f~
_~ Yn(X)Ym(X)d$(x) = 0 should hold

for all integers m ~ n, with the integral taken along the real axis. Unfortunately,
no such function ~(x) was known.
At the moment of shipping the present typescript to the publisher, the author
receives a (handwritten) manuscript by A.M. Krall, with the following important
results.
Define the function ~ (x) by

~(8)-~(~) : lira -1 I B exp( 22~X2)sin( 22~C2)dx;

C -~ 0 X +£ X +¢

then the following holds:

1. t~(x) is of bounded variation.

2. The support of d~(x) is the origin.
5. The moments ~n xnd~(x) (-2)n+i/(n+l) for n 0,1,2 ....

4. f-Z Yn(X)Ym (x)d~(x) = (-l)n+126mn/(2n+l), where 6ran stands for the Kronecker

delta.
5. For (a,b) # (2,2), these results generalize as follows:

Let z = x+ie and set

0(8)-~(~) = lira 1 l~a Im{(-b/z)iFl [i;

a; -b/z] }dx;

then ~(x) is of bounded variation, ~n = (-b)n+i/a(a+l)"'" (a+n-l), and

P Yn(X;a,b)Ym(X;a,b)d~(x) = (-l)n+In.'b ~mn/(2n+a-l)a(a÷l)...(a+n).

 PART II
CHAPTER 5
RELATIONS OF THE BP TO CLASSICAL ORTIIONOR~L POLYNOMIALS
AND TO OTHER FUNCTIONS.

i. In Chapter 2, the BP have been defined as the polynomial factors of the Bessel
functions K (z) of half-integral order. In fact, it follows from (2.5) with

= n + 1/2 and c o = (2n)!/2nn! that

8n(Z) = 8(z,n+I/2) = c021/2-n(F(n+i/2))-Izn+I/2eZKn+i/2(z),

so that

(i) @n (z) = / 2z~ zneZKn+l/2(z).

Many of the properties of BP can be obtained in the easiest way from (i), by using
known properties of the much studied functions Kn+I/2(z). This, however, is not

the only possible point of view. The BP may also be considered as particular cases
of hypergeometric functions, (see, e.g., [90], [2], [3]) of Laguerre polynomials [3],
of Whittaker functions [2], or as limiting cases of Jacobi polynomials [2]. They
also may be defined as partial quotients of certain continued fractions (see [83]),
or, as seen in Chapter 3, as determinants of some simple matrices (see [85]).
While the identity of all these definitions is easy to establish, some of them
are more convenient than others for the proof of specific properties of the BP.
In the following sections of the present chapter we shall present the connection
of the BP with hypergeometric functions, with Jacobi and Laguerre polynomials,with
Lommel Polynomials and with the ~ittaker functions. The connections with continued
fractions will be discussed in Chapter 8.
These connections with well studied functions have been used by many authors,
in order to prove specific properties of the BP in the most convenient way.

2. Let
(al)m(a2)m...(ap) m
® zn
p F q ( a l , a 2 . . . . . ap; b l , b 2 . . . . . bq;Z) = 1+ m~1 (bl)m(b2)m" (bq) m m--i- '

where ( a j ) m = a j ( a j + l ) ( a j + 2 ) . . . ( a j ÷ m - 1 ) , and b l b 2 . . . b q ~ 0.

The c o n d i t i o n s of convergence of t h i s sum a r e w e l l known ( s e e , e . g . [90], o r

[ 6 ] ) , but we do not need them h e r e , because we a r e i n t e r e s t e d only in t h e polynomial
case o f t e r m i n a t i n g s e r i e s . The f u n c t i o n s F r e p r e s e n t e d by convergent (in p a r t i -
Pq
c u l a r , by t e r m i n a t i n g ) s e r i e s are t h e g e n e r a l i z e d h y p e r g e o m e t r i c f u n c t i o n s .
I t may be v e r i f i e d (see [90]) t h a t F is a s o l u t i o n o f t h e d i f f e r e n t i a l equa-
Pq
tion
 35

(2) [6 ~ (6+bj-1)-z ~Ti:l(6+ai)]w = O,

d
where 6 = z ~ a s i n C h a p t e r 2.

In general, if p > q+l, the corresponding series diverges for all z # O; i f ,

however, one o f t h e a i ' s is a negative integer, then the series reduces to a polyno-
m i a l and (2) i s m e a n i n g f u l without any restriction on t h e n o n - n e g a t i v e integers p
and q.
In p a r t i c u l a r , for q = 0 (i.e., if (bl)n(b2)n...(bq) n = 1 for all n),a 1 = -n,

and a 2 = l+n, the general term of F becomes

Pq
m m
Z (n+m) ' z
(-n) (l-n)... (m-l-n). (n+l) (n+2)... (n+m) -- (-l)m (n-m)~ m~ '

(n+m) ! m
so that the general term of 2Fo(-n,l+n;-;-z/2) reads Z .
2m(n-m) .m.'
'

Comparison with (2.8) shows that

(3) yn(Z) = 2Fo(-n,l+n;-;-z/2).

We also verify that in this case (2) reduces to [6-z(6-n)(6+n+l)]w = O, or upon

replacing z by -z/2, to [6+(z/2)(6-n)(6+n*l)]w = O, which is (2.25). In the same
way one verifies (see [89], [90] and [50]) that

(4) yn(Z;a,b) = 2Fo(-n,n+a-l;-;-z/b).

Just as the well-known theory of Bessel functions can be used to obtain proper-
ties of BP from (i) so the equally well-known theory of hypergeometric function can
be used for the same purpose, on account of (3).
This has indeed been done successfully (see, e.g. [3], [2], [112], [90], [15],
[47], [88], [89], [95]). In this context, it is convenient to recall also that

(s) lim 2Fl(al,a2;X;Xz) = 2Fo(al,a2;-;z).

This follows immediately from the fact that, for fixed n,

(al)n(a2) n xnx n (al)n(a2)n xn

lim -- -
X + ~ (X)n nl n~

3. We recall that the Jacobi Polynomial Pn(a'B)(z) may be defined (see [90], p. 254)
by
(i+~) n
p ~ a , 6 ) (z) = n-----~ 2Fl(-n,l+a+B+n;l+e; (l-z)/2).

If we set here a+l = X, z = l+2X(y/b), X+8 = a-l, then a = X-l, 8 = a-l-X,

(l-z)/2 = -X(y/b) and
 36

2Fl(-n,l+a+8+n;l+a;(1-z)/2) = 2Fl(-n,a+n-1;X;-X(y/b))

n! (X-l,a-X-1)(l+2X(y/b)) r(a+l)n! p(X-l,a-X-1)(l+2X(y/b))

(l+a) n Pn = r(n+a+l) n

r(x)r(n) p(X-l,a-X-1) ( l + 2 X ( y / b ) ) .
= n r (n+X----------~ n

I f we take the l i m i t for X + ~, then the l e f t hand s i d e becomes (see (5))

2Fo(-n,a+n-1;-;-(y/b)), or, by (4), y n ( Y ; a , b ). We have o b t a i n e d t h e r e s u l t (see

[2]) (with z instead of y):

(6) yn(Z;a,b) = lim nB(n,X)P n( X - l , a - X - l ) (l+2X(z/b)) ,

X-~ ¢0

where as usually B(a,b) - r(a)r(b)

r(a+b)

By using the classical differential equation, Rodrigues formula, recurrence

formulae, generating function, etc. of the Jacobi polynomials, one easily obtains
the corresponding results for yn(Z;a,b),by passage to the limit in (6). See [2]
for details.

4. Laguerre Polynomials may be d e f i n e d (see [90], p. 201) by

n (- 1)m(l+a)n xm
Ln(a) (x) = Z m~ (n-m)
m=O "! (l÷a)m °

n (-2n-a+2)n 2m -m
I t follows t h a t h(-2n-a+l)(2/Z)n = X ( - i ) m m!(n_m),,(_2n+2_a)m z
m=O

and

n.'(-z/2)nL (-2n-a+l) (2/z) =

n

n
(-l)m+n2 m-n ( - 2 n - a + 2 ) ( - 2 n - a + 3 ) . . . ( - n - a + l ) n , , z n-m
m! (n-m) .'[-2n+2-a) (-2n+5-a) . . . (-2n+m-a+l)
m=0

n
(2n+a-2)(2n+a-3)...(n+a-l)n! (z/2)n-m
m,,(n-m)!(2n+a-2)...(2n+a-m-l)
m=0

n
X (2n+a-m-2)...(n+a-l)n! (z/2)n-m
m=0 m,,(n-m)!

(n+k+a-2)... (n+a-l)n! zk
k=0 k ! (n-k) ~2k

Comparison with (2.27) now shows (see [3]) t h a t

 37

yn(Z'a) = n:(-z/2)nL (-2n-a÷l)(2/z)

' n "

Finally, if we replace z/2 by z/b, we obtain

yn(Z;a,b) = n!(-z/b)nL (-2n-a+l)(b/z)

n
and

en(Z;a'b) = znyn (z-l;a'b) = (-l)nn'b-nL(-2n-a+l)n (bz).

In p a r t i c u l a r (see [19], [3])

@n(Z ) = (_l)nn~2-nm(-2n-l)(2z).
n

5. The Whittaker functions Wk,m(Z ) are solutions of the differential equation

w"+{-i/4+kz-l+(i/4-m2)z-2}w = 0 and may be represented for z # 0 (an empty product

being set equal to one) by (see [67]):

e-Z/2z k ~ (m2-(k-i/2)2)(m2-(k-3/2)2)'''(m2-(k+i/2"r)2) + R;
Wk,m(Z) = 2 r -
r=O r z

here R = R(z;m,k,q) + 0 for fixed q and [z] + ~.

The series obtained by letting q ÷ ~ is asymptotic. However, if m+k = n+i/2

with integral n, then the sum is terminating and R = 0. Indeed m2-(k+i/2-r) 2 =
(m+k+i/2-r)(m-k-I/2+r) vanishes for r = n+l; hence,

Wk -k+i/2+n(Z)

= e-Z/2zk ~ n(n+l-2k)(n-1)(n+2-2k)...(n-r+l)(n+r-2k)
r
r=0 r!z

One recognizes in the sum precisely 2F0(-n,n+l-2k;-;(-z-l)), or, by (4),

yn(z-l'2(1-k),l), . It follows that yn(Z-1;2(l_k) ,1) = e z / 2 z - k ~~ k , _ k + l / 2 + n l Z. ) ,. or

writing z -1 f o r z and a for 2(1-k), yn(z;a,1) = el/2Zzl-a/2W1_a/2,(a_l)/2+n(1/z).

~inally, i f we r e p l a c e on t h e r i g h t z by z / b we o b t a i n (see [2] and [ 5 ] ;
observe, however, misprint i n (13) o f [ 2 ] ) :

yn(Z;a,b) =
e b / 2 z (z/b)l-a/2Wl_a/2,(a_l)/2+n(b/z).

For a = b = 2, one obtains the ordinary BP:

yn(Z) = el/ZW0,n+i/2(2/z ).
 38

6. The Lommel Polynomials Rn,v(z ) may be defined (see [61], §9.6-9.73) by

1
Rn,v(z ) = ~ ~z cosec n~ {Jn+v(z)J_~+l(Z)+(-l)nj_v_n(Z)J _l(Z)} .

If one replaces here the functions Jm(Z), for m = k + ½,half an odd integer, by

Jk÷i/2(z ) = (2~z)-i/2{i-n-leiz yn(I/z)

. + in+l e -iz yn(-i/z)} (see [68], (ii)) one
obtains (see Dickinson [a7]): For . = s + ~z , Rn," (z-l) =

1 {i-nyn+s(iZ)Ys_ l ( - i z ) + z n y n + s ' ( - i Z ) Y s _ l ( i z ) } . The r e l a t i o n between t h e Bessel

function Jk+l/2(z) and t h e BP yn(~ i / z ) follows (see Chapters 2 and 3), e . g . , by

Kn+l/2(z-1. ) = (nz/2) 1/2e-1/Zyn(Z ) and t h e r e l a t i o n s between K (z) and J ( z ) , or, as

in [68], d i r e c t l y from the d i f f e r e n t i a l equations.

7. We formalize some of the results obtained so far in this chapter in the following
theorem.

THEOREM i. With previous notations for hypergeometric functions, Jacobi Polynomials,

Beta Function, Laguerre Polynomials and ~nittaker functions, the following relations
(conong others) connect the BP with these special functions:
(i) Yn(Z;a,b) = 2Fo(-n,n+a-l; -; -z/b) ;

[ii) yn(Z) = 2Fo(-n,n+l;.;_z/2) ;

(iii) yn(Z;a,b) = lim nB(n,x)p(X-l'a-x-l)(1+2x(y/b)) ;
X -~- ~

(iv) yn(Z) = lira nB(n,x)P (x-l'l-x)(l+xy)

n
x -> oo

(v) yn(Z;a,b) = n!(-z/b)nh (-2n-a+l)(b/z)

n
(vi) yn(Z) = n~(-z/2)nL (-2n-l)(2/z).
n

(vii) Yn(Z;a,b) = eb/2Z(z/b)l-a/2w (b/z);

l-a/2, ~1 (a-l)+n
(viii) yn(Z) = e I/z W 1 (2/z);
O,n + ~-

(ix) 8n(Z;a,b) = (-1)nn'b-nL (-2n-a+l)(bz)"

n

(x) Sn(Z ) = ( - 1 ) n n ! 2 - n a ( - 2 n - 1 ) ( 2 z )
n

REMARK. In t h e s t a t e m e n t of t h e theorem a l l r e l a t i o n s referring to t h e case

a ~ b = 2 have been o m i t t e d , because t h e s e are immediately o b t a i n a b l e by s e t t i n g
b = 2 in t h e formulae f o r y n ( z ; a , b ) and 0 n ( Z ; a , b ). The formulae f o r yn(Z) and On(Z )
a r e i n d i c a t e d , because o f t h e i r r a t h e r f r e q u e n t use.
 39

8. In the preceding sections we have discussed the relations of BP to other special

functions essentially from the point of view of identifying BP with these other
functions, for particular values of their parameters.
A different kind of connection is pointed out by the following remark that,
apparently, was never made before.
Let Pn(X) be the n - t h Legendre Pol>momial. This i s , of course, a p a r t i c u l a r

case of Jacobi Polynomials, but can also be defined by the generating function

n
(7) (I-2zx+z2) -I/2 = ~ Pn(X)Z .
n=0

If P(r)(x) stands for the r-th derivative of PnfX~,.. then the following theorem holds
n
n
THEOREM 2. yn(Z) = [ p~r)(1)zr;
r=0
n
@n(Z) = I P~r)(1)zn-r
r=0
The two statements are, of course, equivalent, on account of (2.10). Both are
immediate consequences of

LE~ZIA i. a(n) = p(r)(1).

r n

On account of (2.8), Lemma 1 is equivalent to

LEMI~ 2. p(r)(1) = (n+r)!

n 2rr][n_r)!

This lemma appears to have been known to I. Schur, but this writer is unable to
locate a reference in Schur's work. However, many essentially distinct proofs are
known, by N. du Plessis, [48], E. Rainville [S0], I.N. Herstein [51], and this
writer [ZS], among others. Here is the particular simple one due to du Plessis:
We differentiate (7) r times with respect to x and obtain

(8) zr.l.3 .... (2r-l) (l-2zx+z2) -r-I/2= [ p~r)(x)z n.

n=0
By setting x = i and expanding the binomial, we obtain on the left of (8)

zr.l.3...(2r-1)(1-z) -2r-I = z r.1.3...(2r-1) [ (-2~-l)(-z)m

m=O

= 1.3...(2r-i) [ (-l)m(-2~ -I) zm+r =

m=0

(2r÷l) (2r+2) (2r+m) m÷r

1.3... (2r-l) ~ m! """ z =
m=0
 40

co

1.3... [2r-l) [ (n+r) (n+r-l)...<jrn_r~:

(2r+l) zn.
n=r

co

For x = i, the right hand side of (8) becomes ~ P(r)(1)zn and Lemma 2, hence
n
n=O
Lemma i and Theorem 2 follow b y c o m p a r i s o n o f t h e c o e f f i c i e n t s o f zn i n (8).
 CHAPTER 6
GENERATING FUNCTIONS

1. A v e r y l a r g e number o f g e n e r a t i n g functions f o r BP a r e known. Some a r e f o r m a l

series, others represent functions within certain domains o f c o n v e r g e n c e ; some a r e
o b t a i n e d by more o r l e s s classical methods, others are obtained by a d - h o c c o n s i d e r a -
tions, w h i l e some o t h e r s still a r e o b t a i n e d by t h e t h e o r y of Lie groups. An a t t e m p t
i s made t o l i s t here at least t h e more i m p o r t a n t , o r more u s e f u l o f t h e known g e n e r a
ting function, but even for many of the listed ones it will be necessary to refer
the reader to the original literature for complete proofs.

2. Let us consider the function f(t,z) = exp{[l-(l-2zt)i/2]/z}. By logarithmic

differentiation with respect to t we obtain

Ff ' (t,z) = ( l _ 2 z t ) - l / 2 , so that f'(t,z) = ( l _ 2 z t ) - l / 2 f ( t ,z), and by a further

differentiation f"(t,z) = {(l-2zt)-l+z(l-2zt)-3/2}f(t,z)

= {a~l)(l_2zt)-l+a~l)z(l_2zt)-3/2}f(t,z),

with a~ I) = a~ I) = i, the two coefficients of yl(z).

These formulae are the instances k = i and k = 2 of the

PROPOSITION. Fom k > i,

k-i
il) f(k)(t,z) = f(z,t) 7. a(k-l) zmil-2zt) -(k+m)/2
m
m=0

where f(k) it,z) = __~kf and where the a (j) are the coefficients of y~ iz), as given by
~t k m J

i2.8).

Proof. We a l r e a d y know t h a t (1) h o l d s f o r k = 1 and k = 2, and p r o c e e d t o c o m p l e t e

t h e p r o o f by i n d u c t i o n on k.

We differentiate (I) and obtain:

k-i
fik+l)(t,z ) =- ~ a(k-1)zm(1-2zt)-(k+m)/2-(1-2zt)-l/2f(t,z)
m= 0 m

k-i
+ fit,z) ~. a ( k ' l ) z m + l ( k + m ) ( 1 - 2 z t ) -(k+m+2)/2
m=O m

k
= fit,z) ~ (a ( k - l ) + (k+m- 1) a ( k - l ) z m i l - 2 z t ) - (k+m+l)/2
m m-I
m=O

By using (3.19), this can be written as

 42

k
f(k+l) (t, z) = f(t,z) a(k) zm(l-2zt) - (k+m+ I)/2
m
m=O

and this is precisely (I), with k+l instead of k. The Proposition is proved.

As immediate consequences we note the following corollaries.

COROLLARY i. f(k+l)(0,z) = yk(z).

COROLLARY 2. (see [68]7. f(t,z) = Z tkyk_l (z)/k!,

k=0
where (see (2.12)), for k = 0, y_l(Z) = y0(z) = i.

It follows that the function f(t,z) is a generating function for the polynomials
Yk_l(Z). By differentiation with respect to t we obtain (see [17]):

(2) ~f
~t - (l-2tz)-i/2f(t'z) = ~ tkyk (z)/k'
k=0

The ~nction ~f/3t is, therefore, a generating function for the yk(z).

3. If we replace in (27, yk(z) by zkek(z-l), we obtain

co

(l-2tz)-I/2exp{[l-(l-2tz)i/2]/z} = kZ0= tkzk@k(z-l)/k! ,

-i.
or, replacing z by z
co
(l_2tz-l)-i/2exp{z(l_(l_2tz-l)I/2 } = ~ (tz-l)k@k(Z)/k! .
k=0
We now denote tz -I by v and this leads to

(3) (l-2v)-i/2exp{z(l-(l-2v)I/2)} = ~ vk@k(Z)/k! .

k=O
We may a l s o , following Burchnall [17], set l - ( l - 2 v ) I / 2 = 2u. I f we s u b s t i t u t e this
in (3) we o b t a i n (see [17])
co

(4) (l-2u)-le2ZU =
[ {2u(l-u)}kOk(Z)/k! .
k=O
Burchnall calls this a pseudo-generating function for @k(Z ).

The corresponding pseudo-generating function for the generalized BP ek(z;a,b)

is (see [17] and also [2]).
co

(S) (l-2u)-l(l-u)2-ae bzu = ~ {bu(l-u)}kOk(z;a,b)/k! .

k=O

By replacing here @k(z;a,b) by zkyk(z-l;a,b) and then z by z -I, we obtain

co

(l-2u)-l(l-u)2-ae bu/z = ~ {buz-l(l-u)}kyk(z;a,b)/k!, or, with uz -I = v,

k=0
 43

co

(l-2vz)-iCl-vz) 2-aebv = Z {bv(l-vz)}kyk{z;a,b)/k! .

k=O

Finally, still following Burchnall [17], let 2v(l-vz) = t, so that l-2tz =

{l-2vz) 2 and we obtain the following generating function for the polynomials
yk(z;a,b) :

(6) (i-2tz) -1/2{ [i+ (l-2zt) 1/2]/2 }2-aexp{ (b/2z) (i- (l-2t z) i/2)}
oo

= ~ (b/2) ktkyk (z; a,b)/k : .

k=O
A simple proof of (6) is also given in [43] and [2] and, for b = 2 in [16], see
also [50].

4. Burchna11's generating function has been generalized by Rainville (see [89] and
[90]), as follows:

Let

~n(X) = q+2Fp(-n,c÷n,l-81-n . . . . . 1-gq-n; 1-al-n ,1-~2-n . . . . . 1-C~p-n;(-1)P÷q÷lx);

then On(X ) admits a generating function of the form

(l-4xt) -1/2{1 [l÷(l-4xt) 1/2] }l-CpFq (C~l,C~2 . . . . . C~p;81 , 82 . . . . . gq; [2t (1+ (1-4xt) 1/2) -1] )

(7)
[ ~n(z)tn (~l)n...(~p) n
n=O n! (61)n...(Bq)n "
Burchnall's generating function (6) is essentially (7) with p = q = 0, c = a-l,

x = zb -I, and t replaced by bt/2.

A different type of generating functions for BP is obtained by Rainville [89]
as follows: Set ~nCX) = p+2Fq(-n,c+n,~l,...,~p; 81 .... ,Sq;-X); then

(8) (l-t) -c p+2Fq(C/2,(c+l)/2,~l ..... ~p;B 1 ..... 8q; 4xt(l-t) -2)

= ~ On(X)(C)ntn/n~ •
n=0

For p = q = 0, On(X) = 2F0(-n,c+n;-;-x) = Yn(bX,c+l,b), so that (8) is the generating

function for the yn[Z;a,b), with z = bx, a = c+l, and the same b.

5. W.A. AI-Salam [3] uses two generating functions for Jacobi Polynomials due to
Feldheim [16] and the relation between BP and Jacobi Polynomials (see Chapter S),
 44

in order to obtain (in a slightly different notation and somewhat lesser generality)
the generating function

(9) et(l-tz/b) I-~ = [ tnyn(Z;~-n,b)/n~ ,

n=0
and the formal (divergent) generating function

(i0) (l-t)-12F0(a-2,1;-;tz/b(l-t)) ~ Z tnyn(Z;a-n-l,b).

n=0
On account of (5.4) for integral a (i0) may be written also as

(i0') (l-t)-ly2_a(Z/(l-t-l);a,b) -~ [ tnyn(Z;a-n-l,b ).

n=O

By using the connection with Laguerre Polynomials (see Chapter S) and known
generating functions of these polynomials, AI-Salam [3] also obtains

(ii) (l+tz/b)e-2exp{t/(l+tz/b)}
=Z tn Yn (z;~-2n,b)/n! "
n=0

The characteristic feature of these generating functions is that they lead t o

sums o f BP y n ( Z ; a , b ) , where the e n t r y a = a ( n ) , and i s not a constant.
oo

Next the relation L(a)(x+t) = et Z (-l)ktkL(~+k)(x)/k: leads (see (6.10) in

• n n
k=O
[~]) to

(12) etCl-tz/b)nyn(Z;a, c) = Z tkyn(Z;a-k,b)/k' ,

k=O
where Z = cz/(b-tz).

One may observe that the parameter c in the first member is idle and can be replaced
by 2• provided that one then sets Z = 2z/(b-tz). It also may be worthwhile to
observe, that in the second member the subscript of all BP is n, while the "a"-para-
meter varies, depending on k and the sum is over k.
Ai-Salam gives also several• rather complicated bilinear generating functions•
as well as the formal (no convergence) generating function

(13) L~ (n+a-2
n ) t n yn(Z;a,b] ~ (l_t)l-a2F0(a-12 , ~a ,. -,. - bzt
- )
n=O (l-t) 2 "

In case (l-a)/2 = m is an integer, this is equivalent (by (5.4)) to

.n-2m-l. n
(13') [ n ) t yn(z;l-2m,b) ~ (l-t)2mym(-4zt(l-t)-2 ; ~3 -2m,b).
n=O
For a = 2, on the other hand, (13) reduces (see [3]) to

I bzt
(13") ~ yn(Z;2,b)t n -~ (l-t)-12Fo(~ ;1;-; 2) ,
n=O (I-t)

essentially obtained already by Krall and Frink [68].

 45

From these formulae corresponding ones for en(Z;a,b) can be obtained by simple

manipulations.
A somewhat different generating function, involving Hermite polynomials is
found in [3], (6.13). It can be obtained by setting a = l-k in (13) and inter-
changing n and k. With some simplifications (also some misprints in [3] need correc
tion), this reads

(_l)k(~)tkyk(Z;l_k,b) = e - ~ i n / 2 ( t z / b ) n / 2 H n ( ~ (~)i/2).
k=0

6. Brafman [15] considers the Cauchy product of the series representations for
1 1 (t-(t2-4xt)i/2), obtains formally
2Fo(~,c-~;-; ~ (t+(t2-4xt) I/2) and 2Fo(~,c-~;-;

a sum over 3F2's, transforms the latter by use of a formula of Whipple [65] and

obtains in this way the formal result

2Fo(e,c-e;-; ~i (t_(t2_4xt]i/2))2F 0 (~,c-~;-; ~1 (t+(t2-4xt)i/2)).-"

( (~)n (c-~)n/n!)tn2F0 (-n,c+n; - ;x) .

n=0

By observing that 2F0(-n,c+n;-;x) = yn(-bx;c+l,b) we obtain the formal generating

functions for BP:

(14) 2F0(a,c-a;-; ~i (t_(t2_4xt)i/2)

)2F0 (~,c-~;-; ~1 (t+(t2-4xt)i/2)) -~

Z cntnyn(-bx,c+l,b),
n=0
where en = Cn(=,c ) = (~)n(C-~)n/n:. If ~ = -m, a negative integer, then also the

first member of (14) can be written as a product of BP, namely

Ym(Zl,C+l,2)Ym(Z2,c+l,2),

where z I = -t+(t2-4xt)I/2, z2 = -t-(t2-4xt)I/2.

We may observe that in this case the second member reduces to a finite sum,
because now (S)n = (-m)(l-m)...(n-m-l) and vanishes for n > m, and the formal

equality becomes an actual one (see [3] in somewhat different notations) and reads:
m
(15) ym(Zl, a, 2)Ym(Z2, a, 2) = [ [.m-n+l.
n )(a+m-n)ntnyn(-bx'a'b)'
n=O
with previous values for z I and z 2.

7. Carlitz [19] starts from Corollary 2, sets %(z) = znyn_l(z-l) = Zen_l(Z) and
observes that Corollary 2 can be written as
 46

(16) exp{z(l-(l-2t)i/2) } = Z fn (z)tn/n! ,

n=O
or, if we set l-2t = (l-2u) 2,

(17) e 2uz = Z 2nfn(Z)(U-u2)n/n! .

n=O

If we replace in (17), z by z -I, so that fn(Z -I) = znyn_l(Z ) and then set 2uz -I = x,
(17) becomes (see [68])

(18) eZ = Z (z-xz2/2)nyn_l(x)/n! •
n=O
In this surprizing formula, the right hand side is in fact independent of x and, for
x = O, it reduces trivially to the left hand side.
By using Brafman's formal equality (14) in the case c = i, Dickinson [47] obtains
a generating function for the modified Lommel polynomials.
Here we have reached a point, where Jt is debatable, whether we are still
dealing with generating functions in their original sense. Formulae like (17), or
(18), may be considered more properly as belonging to the theory of representation
of functions by series of BP, a topic that will be discussed in Chapter 9.

8. This chapter would be incomplete without the mention of a most important general
method that permits us to obtain a large number of generating functions, among them
most of the previous ones. It is based on the theory of Lie groups. It is, unfor-
tunately, not possible to develop here the complete theory and the interested reader
is advised to consult, e.g., the excellent presentation of the method in [73]. The
method has been introduced by L. Weisner [62], [6~], [64] and probably the clearest
presentation of the basic idea is still the one of [62]. On the other hand, the
only complete presentation knoml to the present author of all details needed for
the non-specialist is to be found in E.B. McBride [73]. A somewhat different approach
can be found in the work of S.K. Chatterjea [35], and M.K. Das [43].
Following Weisner [62], we consider a linear, ordinary differential equation,
that depends also linearly on a (not necessarily integral) parameter ~, say

d
(19) L(x, ~-~ ,~)v = 0,

with a ¢ A, say, where A is some denumerable set.

If y is another independent variable, then for every B, with v = v (x), also

u(x,y) = yBv (x) is a solution of (19). Let now, in general, u = u(x,y) be a func-

tion of the two independent variables x,y that satisfies ~ u = au; here we have set
Y
~y = y ~ , in analogy with the notation of Chapter 2. By the linearity of
L ( x , ~-~
d , e) i n ~, a n d b e c a u s e By.ye = ~ y ~ , i t follows that a function u(x,y), that
 47

satisfies 6 u = eu, and also

Y

(20) Lu = n(x, ~- , ~y)u = 0,

is of the form u(x,y) = y~va(x), with va(x ) a solution of (193. The c o n v e r s e is

fairly obvious: i f v (x) is a solution of (19), then u(x,y) = yava(x) is a solution

o f t h e s y s t e m ~yU = a u , Lu = 0. ttence, if {v ( x ) } i s a f a m i l y of solutions of (19),

then the se~ {yav (x)} is a family of solutions of (20).

L e t u s now a s s u m e t h a t independently o f what p r e c e d e s we c a n

(i) find a solution u = g(x,y) of (20); and
(ii) expand g(x,y) in a series of the form

(213 g(x,y) = ~ ga(x)y a.

sEA

If, furthermore, the series converges sufficiently well so that a termwise applica-

tion of the operator L = L(x, ~ ,6y) c a n be j u s t i f i e d , it follows that

~ d d
L ( x , ~-~, 6 y ) ( g a ( x ) y ) = 0, w h e n c e L ( x , ~-~ , a ) ( g a ( x ) y a) = y ~ L ( x , ~-~ , a ) g a ( x ) = 0,

so that {ga(x)} is a family of solutions of (193. In t h i s way we h a v e o b t a i n e d ,

according to (213, a generating function g(x,y) for a family {ga(x)} ¢ A of

solutions of (19).
The d i f f i c u l t y of this approach consists in the fact that most usual methods for
the solution o f (20) lead to solutions of the form u = g(x,y) = g a ( x ) y a and sums o f

such solutions. This, of course, reduces (21) t o a t r i v i a l identity. If, however,

solutions u = g(x,y) of (20) c a n be f o u n d i n c l o s e d form, not of the type ~ ga(x)y a,
a
then (21) y i e l d s genuine generating functions for the class {ga(x)} E A of solutions

of (193. Moreover, distinct (and l~nearly independent) solutions g(x,y) of (20)

furnish distinct generating functions.
A particular case in which non-trivial solutions of (20) c a n be f o u n d i s t h a t
when (20) a d m i t s a n o n - t r i v i a l group of transformations (it always admits the groups
x + ¢, y + i n ) . Sometimes it is possible to guess a set of differential operators
of the first order and v e r i f y on h a n d o f t h e i r commutation re!ations that they
generate a L i e g r o u p and t h a t these generators commute w i t h t h e o p e r a t o r L, o r a t
least with ~(x)L, for an a p p r o p r i a t e function ~(x). In the particular case of the
classical orthogonal polynomials, or of Bessel functions, one may t a k e as parameter
a the degree, or order n and a t t e m p t to use as generators of the Lie group the
operators that raise, or lower n. A variety of special methods, shortcuts, etc. have
been used by various authors, in order to obtain their results (see, e.g. [35], [43]).
 48

In general, however, a rather careful study of the Lie algebra, its commutation rela-
tions, conjugacy classes of operators, etc., is needed (see, e.g. [62], [63], [g4],
[73]).
As an e x a m p l e ( s e e [73] and [ 4 3 ] ) , l e t Un(X ) = x l / 2 e - 1 / X y n ( X ) , w h e r e Yn(X) i s

the n-th BP, and s e t F n ( X , t ) = e ( n + l / 2 ) t U n ( X ) . (We w r i t e t r a t h e r than y, to avoid

2 8
confusion with Yn(X)). We o b s e r v e t h a t R = et(x ~ + x ~x) is a right shift operator

for F . Indeed,
n

R-Fn = e t { x ( n + l / 2 ) e ( n + l / 2 ) t U n + X 2 e ( n + l / 2 ) t U n }

(n+3/2)t ...... 1/2 - 1 / x 2.1 -1/2 -1/x -3/2 -1/x. 5/2 - 1 / x , .

= e tLn+l/zjx.x e Yn+x t ~ x e +x e jyn+X e yn ~

= e(n+3/2)te-1/Xxl/2{((n+l)x+l)Yn+X2y~} = e(n+l+l/2)txl/2e-1/Xyn+ 1 = Fn+ 1,

by (3.16). It follows that

n ~ n ~ n
(22) e~-RFm(x,t ) = ~ ~-F.
RnF~ = ~ ~ F = xl/2e -I/x ~ ~y.
~ e(m+n+i/2)tYn+m(X)
n=0 - m n=0 m+n n=0

On the other hand, e~BFm(X,t ) = Fm(e~Bx, eWBt).

Here _Rx = etCx ~- + x 2 ~x)X

~ = etx 2, 8 + "x 2 8__}
R2x = et{x ~- 8x (etx 2) = 3e2tx 3,

and, by an easy induction on n,

Rnx = 1 . 3 . 5 . . . ( 2 n _ l ) e n t x n+l

Consequently, emBx = n=0

[ 1 . 3 . . n!
.(2n-1) mn e n t x n + l = x ( 1 - 2 x m e t ) - 1 / 2 = Xl, s a y .

Similarly, e~gt = t + Y m p'-'nt = t + 7~ 2 . 4 . . n!

.(2n-2) ~n x n e n t
n=l n=l

1 ~ (2~xet)n t - 1
= t + [ n = ~ log (1-2~xe t) = tl, say.
n=l

Consequently, Fm(e~Bx , emgt) = F m ( X l , t l ) =

exp{(m+i/2)(t-i/2 log(l-2mxet))}{x(l-2mxet)-i/2}I/2.
exp{-(1-2x~et)i/2x-l}ym(x(l-2x~et)-i/2 )

= exp{(m+i/2)t}(l_2~xet)-(m+I/2)/2xl/2(l_2x~et)-1/4
• exp{-(l_2x~et)i/2x-l}ym(x(l_2xmet)-I/2)
 49

and (22) becomes, after some obvious simplifications,

(23) (l-2x~et)-(m+l)/2exp{[l_(l_2x~et)i/2]x-l}ym(x(l_2xmet)-i/2)

n nt
L
n=O
n: Yn÷m(x).

In particular, if we set t = O, we obtain

n
(l_2x~)-(m+l)/2exp{[l_(l_2x~)i/2]x-l}ym(x(l_2x~)-i/2) = [ m Yn+m(X)
HT.'
n=O
Finally, if we take m = 0 and write t for m, this becomes

tn
(l-2xt)-I/2exp{[l-(l-2xt)I/2]x
-I} = [ ~ Yn (x),
n=O
which is, of course, precisely the generating function (2) of Burchnall.

Instead of the right shift operator, we could have used, of course, the left
shift operator

L = e-t(-x ~-~ + x 2 ~-~..)

8x ' but we obtain exactly the same result.

On the other hand, we may verify that JR, L] = RL - LR = 0, so that these operators

commute and e K = e uB+vL = eUBe vL, with u,v any two independent variables. By pro-
ceeding with K as we did before with R, we now obtain (see [43])

(l_2xu)-(m+l)/2(l_xv)m/2exp{ [l_(l_xu)i/2(l_xv)1/2] x-l}ym(x(l_2xu)-I/2(l_2xv)-l/2)

unv k
= [ ~ Ym+n-k (x)"
n,k=o " "

In the second member, for k > m+n, Ym+n_k(X) has to be interpreted (see (2.12)) as

Yk_m_n_l(X). The formula simplifies somewhat for m = O.

The generating function (2) had been obtained also by S.K. Chatterjea [35] by a
very similar approach but after he first reduces equation (2.11) to its Sturm-Liou-
rifle form

d
d-~ {p(x)u'} + l¢ (x)u = O.

In a somewhat different notation and by following very closely the exposition in [73],
Ming-Po Chen and Chia-Chin Feng [39] obtain the following generating functions:

(bt) k
(l-xt)nebtyn(Z;a'b) = ~ k! Yn (x;a-k'b)'
k=l
x
with z = ~ , which is essentially the same as (12); and
 50

yn(Z;a,b) = (l_t) n+a-I .n+k+a-2. k . x

l k )t Yn(X;a+k,b), with z = l-t "
k=O

For a = b = 2, in particular, the latter becomes:

Yn(X/(l_t)) = (l_t)n-1 ~ ~.n+k.

k ) tkyn(X;2+k;2)"
k=O

Finally, with z - xt(t-~x)-l(l-t) -I,

(t_~x)neba/t yn(Z;a,b) (l_t)n+a-I ~ ~ m+k-m+a-2. (~b) m tn+k-myn(X,a_m+k,b)

= ~ k ) m.--~-- "
m=O k=O

9. In addition to the generating functions presented here, one finds either different
ones, or different proofs for those discussed here, in the following papers: [3],
[109], [46], [36], [76], [77], [78], [79], [16], [30], [31], [44]. See also
Chapter 15.
 CHAPTER 7
FORMULAS OF RODRIGUES' TYPE

d
i. In Chapter 2, by use of the differential operator 8 = z ~ , that is by following

essentially Burchnall's [17] method, we obtained the Rodrigues' formula (2.40), i.e.:

(I) On(Z;a,b ) =(-1)'nb-nebzz 2n+a-1 dn (z-n-a+le-bZ).

dz n
We mentioned also the important particular cases of (I), corresponding to b = 2,
and to a = b = 2, which are

(i') @n(Z;a) =(-l)n2-ne2Zz 2n+a-I _ d_ n (z-n-a+le-2Z)

dz n
and

(I") On(Z ) =(-l)n2-ne2Zz 2n+l dn (z-n-le -2z) ,

dz n
respectively.
It would appear reasonable to obtain corresponding formulae for yn(z;a,b) and

its particular cases from (i), (i'), (I"), simply by writing 1/z for z and then

multiplying by zn. This is, in fact, possible, but the computation of

dn
{ d ~ (u-n-a+le-bU)}u=i/z to which one is led is not entirely trivial (see Section

S). For that reason and also in order to present a variety of approaches we shall
start out differently and follow essentially [68].

2. We shall need some known facts, which we state as Propositions and, in order to
make the presentation selfcontained, we indicate short proofs of them.
Let C be the unit circle [z I = I and let w(z) be a weight function. With
respect to w{z) we define the k-th moment of the power zj by

mjk = ~ 1 /C zk.zj .w(z)dz. With these moments we form the determinants

mOO mOl """ mo'n-I I

!
An(W) = [[mJkl]O-I = mlO mll "'" ml'n-i I

mn-l,O mn-l,l ..- mn_l,n_l]

I
PROPOSITION i. If w(z) is such that An(W) ~ 0 for all n, then the conditions

i
2~---{ IC W(Z) zkun(Z) dz = 0 for k = 0,I ..... n-i define a polynomial Un(Z ) uniquely,

up to an arbitrary multiplicative constant.

 52

n
Proof. L e t Un(Z ) = [ a.z j with a = 1; t h e n the i n t e g r a l equals
j=O ] n

n n
[ aj l
~-~ f C zk+Jw(z)dz = Z ajmjk and the conditions become
j =o j =o

n-I
Z ajmjk = -anmnk = -mnk (k = 0,I .... n-l). The determinant of the coefficients
j=0
aj (j = 0,I ..... n-l) is An(W) and, by assumption, An(W ) # O. It follows that the

a.'s are uniquely determined. If the coefficient a is allowed to be arbitrary,

3 n
then all the aj's have their previous values multiplied by an and Proposition 1 is
proved.
, " 2J+k+l
PROPOSITION 2. If w(z) = e -2/z then mjk = (-i) 3+k+l (j+k+l)!

1 " 1 zk+j 1 ~)
Proof. mjk = 2-~ fC zk+Je-2/Zdz = ~ fC (r=O
~ ~' ( r)dz

(-2) r 1
r~ 2~i IC zk+j-rdz"
r=0

The inversion of the summation and integration is easily justified. All inte-
grals vanish, except one, where k+j-r = -I, that is, r = k+j+l, and this equals
2~i; Proposition 2 is proved.

PROPOSITION 3. If w(z) = e -2/z, then An(W) # 0 for all n.

Proof. It is easy, although somewhat computational to determine the actual value

2 n-i 2n-i
An(W ) = (_l)(3n-l)n/22 n [ ] (9~)/ ~ - - (~!),
~=i 9=n

from which An(W ) # 0 follows trivially. For a proof of this formula, see [25].

COROLLARY l.The conditions f C e-2/Zzkun(z)dz = 0 for k = 0,i ..... n-I define a poly-

nomial Un(Z ) of degree n uniquely, up to a multiplicative constant.

Proof is immediate on account of Propositions 1 and 3.

PROPOSITION 4. Un(Z ) = e 2/z - -dn (z2ne -2/z) is a polynomial of exact degree n, with
dz n

constant term 2n.

Proof. d-~d (z2ne -2/z) = 2n z2n-le -2/z + 2z2n-2e -2/z = e-2/Z(2n z 2n'l + 2z2n-2) ",
 53

this is the instance m = 1 of the general statement d m (z2ne -2/z) = e-2/ZP2 n m(Z)
dz m - ,

where P2n_m(Z) is a polynomial of degree 2n-m. One completes the proof by induction

on m. Next, by considering Leibniz's Rule for differentiation, we observe that the

lowest power of z is obtained from the term z 2n dm e -2/z = z2n{(2/z2)m+...}e -2/z =

dz TM
(2mz2(n-m)+...)e -2/z. For m = n, in particular, the lowest term is the constant 2n.

LE~MA l.The first n moments of Un(Z), with respect to the weight function

w(z) = e -z/2, vanish.

Proof. - -dn (z2ne-2/z) = - -

dn ~~ (-i) v - -2~ z 2n = ~~ (-i)
~! ~ 2~ - -dn z -~+2n
dz n dz n u=O u!z u u=O dz n

= [ (-1) ~ ~2 ~ (-~+2n)(-~+2n-1)...(-~+n+l) z-~+n.

One observes, in particular, that the powers -u+n corresponding to ~ = 2n,

= 2n-l,...,w = n+l do not occur, so that (see remark to that effect in [17])

dn n 2u n-u
(z2ne -2/z) = ~ (-i) u ~., (2n-~)(2n-l-u)...(n+l-9)z
dz n ~=0

+ ~ 2~
9=2n+i 9--~ (~-2n)(~-2n+l)'''(~-n-l)z-~+n"

This is the sum of a polynomial of exact degree n and of an infinite series of

decreasing powers, that starts with z -n-I It also follows that fk,n(Z) =

k dn
z (z2ne -2/z) is the sum of a polynomial of exact degree k+n and of a series of
dz n
decreasing powers starting with zk-n-l. Consequently, for k = 0,1,...,n-l, the
residues of fk,n(Z) vanish, so that

I c e -2/z zkun(z)dz = IC e-2/Zzk{e2/Z d n (z2ne-2/Z)}dz =

dz n

I C zk - -d n (z2ne-2/Z)dz = I C fk,n(Z)dZ = 0, as claimed.

dz n

On the other hand, we know from Corollary 4.4 that f C e-2/Zzkyn(z)dz = 0 for

k = 0,1,...,n-l. From Corollary 1 it now follows that Un(Z) can differ from yn(Z)

only by an arbitrary multiplicative factor.

 54

THEOREM i. (see [68]). The BP yn(Z) admits the Rodrig~aee' formula

(2") yn(Z) = 2-ne2/z dn (z2ne-2/z).

dz n

Proof. On account of what precedes, it only remains to determine the factor of

proportionality. We know from Proposition 4, that the constant term of Un(Z ) is 2n.

Also (see Chapter 2) the constant term of yn(Z) is 1 and Theorem 1 is proved.
A somewhat different proof of the Lemma may be given, following [68]. It
depends on the remark that all functions involved are uniform. We proceed by
integrating around C, say, from +i to +i and use integration by parts. All integrated
terms vanish, because they are singlevalued. Specifically,

IC zk _ _dn (z2ne-2/Z)dz = z k dn-I (z2ne-2/z.,+l k-I dn-I

dz n dzn-i )J+l -k f C z dzn- 1 (z2ne-2/Z)dz = ... =

(-l)rk(k-l)...(k-r~l) f C zk-r dzn_

dn-rr (z2ne-2/z) dz-

In particular, for 0 < k = r < n, this becomes

dn-r dn'r-I +i
(-l)rr! fc dzn-r (z2ne-2/Z)dz = (-l)rr' dzn-r-----~(z2ne-2/z)]+l = 0.

The rest of the proof of Theorem i may be kept unchanged.

3. THEOREM 2. The general BP YnCZ;a,b) P~s the Rodrigue8' formula

b-nz2-aeb/Z dn (z2n+a-2e-b/z).
(2) yn(Z;a,b) = dz n

For a = b = 2, (2) reduces to (2"). It also is clear that not much generality is
gained by allowing arbitrary values of b # 2. One could well consider instead of
(2) simply

(2') yn(Z;a) = 2"nz2-ae 2/z dn (z2n+a-2e-2/z),

dz n
from which (2) ilr~ediately follows by a change of variable. However, the proof of
(2) is not more difficult than that of (2'), so that we proceed to prove Theorem 2
as stated.
In the proof we shall need
k
LE~ 2.For k ~ n, Z (-l)S(~)(2n+a-2-s)(n)=n(k) (2n+a-k-2)(n-k)', for k > n, the
s=0
sum vanishes.

Proof. For integers n,c,k,

 55

dn dn k k k
{xC(x-1) k} = X (-1)S(s)xC+k-s = (-1) s (ks) (c+k-s) ( n ) x c * k - s - n .
dx n dx n s=O s=O

k
s k (n)
For x = 1, in p a r t i c u l a r , t h e second member becomes [ (-1) (s) (c+k-s) On t h e
s=O
n dn-r dr
o t h e r hand, by L e i b n i z ' s r u l e , the f i r s t member equals r=O[ (;) ( d x - - ~r xC) ( J (x-1)k)"

The terms with r > k v a n i s h , because ( x - l ) k i s of lower d e g r e e than r ; f o r r < k

the l a s t f a c t o r c o n t a i n s a p o s i t i v e power o f x-1 and v a n i s h e s at x = 1; c o n s e q u e n t l y ,
only the term w i t h r = k can be d i f f e r e n t from z e r o . It exists, however, only i f

k _< n and then i t s v a l u e i s ( k ) c ( n - k ) k ' = n ( k ) c ( n - k ) . We have proved t h a t f o r an),

k
integers n,c,k, k < n, [ (-1)S(s)1" ( c + k - s ) ( n ) = nkkJc
k n - k "J ~ " " and t h a t t h e sum
s=O
vanishes for k > n. For c = 2n+a-k-2 this is precisely the Lemma.

Proof of Theorem 2. We expand the right hand side:

b-nz 2-a m~0~ bm dn ~[ (-I)s ~bS z2n+a-2-s

= m!z TM dz n s=0

b-nz 2-a [ ~b m Z -m ~[ (-1)S ~-.,

b s ( 2 n + a - 2 - s ) ( n ) z n+a-2-s
m=O s=O

Set m÷s = k; then the double sum can be written as

bkzn-k k
( - 1 ) s (~) (2n+a-2-s) (n).
k=O s=O

By Lemma 2, the inner sum equals n (k)(2n+a-k-2)(n-k) for k f n, zero otherwise;

hence, the sum becomes
n bk-nzn-k n(k) n b-k k.n. (k)
k! (2n+a-k-2) (n-k) = [ z [k) (n+k+a-2)
k=O k=O

and this is precisely yn{Z;a,b), as follows from (2.28).

4. From the Rodrigues' formula (i) for @n(Z;a,b) immediately follows the correspond
ing formula for Cn' namely:

~n(Z;a,b,c) = (_l)nb-ne(b-C)Zz 2n+a-I _ d_n (z-n-a+le-bZ),

dz n
whence, in particular

#n(Z;a,b,b) = (_l)nb-nz 2n+a-I d n (z-n-a+le-bZ);

dz n
 56

@n(Z;a,b) = C n ( Z , a , b , b / 2 )
• = ( _ l ) n b - n e z b / 2 z 2n+a-1 dn ( z - n - a + l e - b Z ) ;
dz n

Cn(Z;a) = Cn(Z;a,2) = (-l)n2-neZz 2n+a-I dn (z-n-a+le-2Z) • and

dz n

Cn(Z) = On(Z;2) = (-l)n2-neZz 2n+l dn (z-n-le-2Z ) .

dz n

S. In Section 1 it was suggested that a natural way to obtain Theorem 1 and 2 would
be to use (i), already known from Chapter 2, and the relation yn(Z) = zn@n(z-l).

In the present section we shall sketch such a proof and similar ones. We shall show
that they require certain combinatorial identities and they may appear less trivial
t h a n one may have e x p e c t e d .
We a l r e a d y know (see ( 6 . 2 ) ) that

(1-2zt)-l/2exp{[1-(1-2zt) i/2]/z) = Z Yn(Z) t n / n I ,

n=O
dn
so t h a t yn(Z) = - - ( ( l - 2 z t ) - l / 2 e x p { [ 1 - ( 1 - 2 z t ) l / 2 ] / z } ) t = O .
dt n

As z is not involved in the differentiation,

dn =
(3) e-2/Zyn(Z ) = _dt_ n ((l-2zt)-i/2exp{_[l+(l-2zt)i/2]/z)t= 0 n!c n,

where cn is the coefficient of tn in the expansion of the large bracket. Let

v = [l-2zt) I/2. Then

1 e-(l+v)/z i Z (_l)k (l+v) k i [ zk ( )vr =

= ~ k=0 k!z k = ~ k=0 k: r=0

(-i) k (-i)
- - (r
k)(l-2tz)(r-l)/2 = Z Zkk (k) 7. (_l)m 2mzmt m
k=O k!z k r=O k=O k! r=O m=O
r-I
= ~ (_l)mtm.2mz m k~O (-l)k ~ (r
k) (7) = ~ Cm tm
m=0 = k'z k r=0 m=0 "

It follows that the coefficient c of tn is

n

C = [-l)n2nz n ~ (-l)k ~ (kr)

n k=0 k!z k r=0
(?) , or, by (3),

Yn(Z) = (-l)ne2/Zn:2nzn (-l)kkZ0=

~ k!z n r=0~ (kr) (r-
0 i

By (2") this implies the validity of the following identity:

 57

(4) (-l)n2nzn k~0= k!z k r=0 (?) = n.' dz n (z2ne-2/z)"

dn ~ 2k dn 2n-k
However, - - (z2ne-2/z) = ~ ( - 1 ) k k! z
dz n k=O dz n

2k n 2k (n) zn-k ~ 2k -k+n

[ (-I)k ~ (2n-k)(n)zn-k = ~ (-i) k ~ (2n-k) + ~ ~ (k-Zn)(n)Z ,
k=0 k=0 k=2n+l

because the products (2n-k)(2n-k-l)...(n-k+l) with n < k ~ 2n all vanish.

By equating the coefficients of equal powers of z in the two members of (4),

it follows that (4) holds if, and only if

(a) f o r k ~ n, (-1)n22nn~ k~ (~) = 2k ( 2 n - k ) ( 2 n - k - 1 ) . . . ( n - k + l ) ;

(b) for n < k < 2n, ~ (~) = 0;

r=0

>_ ~ ( ) = 2k(k-2n)(k-2n+l)...(k-n-1).
r=0

These three conditions are jointly equivalent to

(S) ~ (k) _k-2n .k-n-l.

= Z ( n ) "
r=O

This formula appears to be new. The closest result available in the literature
seems to be Carlitz' formula (see [ZO])

i k .k-n-l.2k-2n (n ~ O)
(5') k k r/2 = ~ [ n-I )
(r) ( n )
r=0 = 2k (n = 0)

In fact (S) and (5') can be obtained from each other, but not trivially [22].
Both are particular cases of sums of the form
k
(+ l)r(~)((k+~ )/2) (a = i or 2, s = rational integer)
r=0
studied by H.W. Gould [21].
We have proved the purely combinatorial identity (5), valid for all integers k,
by using Rodrigues' formula (2"). Conversely, however, if one knows that (5) holds
(and (S) can be proved also directly, although not trivially - see [22]), then one
can trace all steps backwards and one obtains the Rodrigues formula (2") as a
consequence of the generating function (6.2) and (5). This is perhaps a rather
 58

artificial procedure and for this reason we abstain from proving (2), by starting
similarly from the generating function (6.6).
Finally, let us see how one can obtain (2") from (i"). By (2"),

@n(Z) = znyn(i/z) = zn.2-ne2Z{dnud--n(u2ne-2/U)}u=I/z. If we also replace On(Z)

with the help of (I") and suppress common factors, we obtain the identity

dn dn (z-n-le-2Z)
(6) { d ~ (u2ne-2/U)}u=I/z = (-l)nzn+l dz n "

Clearly, by going backwards, (2") is a consequence of (I") and (6), if the

latter can be established directly. This we now proceed to do. The left hand
member can be computed by (4) to be equal to

{ ~( kk!u
(-1)n22nn:un k=O (-1)k
r k) r=O
~~ }u=l/z (-1)n22nn' k=O
~ (-1~ zk-n r=O
~ (k)C~2
~n //
Assuming once more that (5) holds (this was already needed in order to estab-
lish (4) directly), (6) becomes

dn co (-1)k k-2n-12k-2n(k-n-l)
(7) dz n (z-n-le -2z) = 22nn! ~ ~ z
k=O

The left hand side equals

dn {z -n-I [(-~I k 2kz k}
dz n k=0

= ~ (kl~)k 2k dn zk-n-i = 2k
~ (-l)k ~. (k-n-l)(n)z k-2n-I
k=0 dz n k=0
,2k (k-n-l) (n)
The coefficient of zk-2n-l on the right side of (7) equals (-l)k nk,
n:

so that (6) is indeed an identity. As already pointed out, (6) and (i") imply (2").
This method, however, requires the use of the combinatorial identity (5) and is
hardly easier than the direct proof.
It is clear how one can prove the more general result (2), by starting from (i)
instead of (i") and making use of (5), but we abstain from reproducing it here.
 CHAPTER 8
THE BP ~ND CONTINUED FRACTIONS

I. Let [a0,a I .... ,an .... ] stands for the continued fraction

1
a 0 + al + ...
+
1
m
a
n+...

of partial quotients an and denote the n-th convergent [a0,a I .... ,an] by pn/qn , where

(pn,qn) = I. Consider, in particular, the expansion into a continued fraction of the

function

e 2/z + 1
(i) -~7~/z - 1
e = [z, 3z, Sz ..... (2n+l)z .... ],

known from the work of Lambert [4Z], (see also [~$]). See [83] for a recent proof.
If we denote the numerators and denominators of its successive convergents by
pn(Z) and qn(Z), respectively, we find, e.g.,

P0 (z) z Pl (z) 1 3z2+i

q0(z-----y= z = y ; ql(z-----y= z + 3--{ = 3z , etc.

It is well known (see, e.g. [47]) that pn(Z) and qn(Z) can both be obtained, for

n ~ 2, from the same recurrence relation

(2) Xn = anXn_ 1 + Xn_ 2,

where one has to start with the initial values p0(z) = z and pl(z) = 3z2+I for

pn(Z), and with q0(z) = i, ql(z) = 3z for qn(Z). In either case an = (2n+l)z, as

follows from [I). One obtains, in particular, for n = 2:

p2(z) = pl(z)a2+P0(z ) = (3z2+l)(bz)+z = iSz3+6z,

q2(z) = ql(z)a2+q0(z ) = (3z)(bz)+l = iSz2+l.

One may observe that (2) actually holds also for n = i, provided that one defines
p_l(Z) = I, q_l[Z) = 0. One is led to make the following remarks:

[i) All powers of z that occur in qn(Z) are of the same parity, which is that

of n, and all powers that occur in pn(Z) are of the parity opposite to that of n.

This is verified directly, on above examples, for n = 0,1,2, and holds in general
by induction on n, on account of (2).
 60

(ii) p0(z) + q0Cz) = z+l = YlCZ); PlCZ) + qlCz) = 3z2+l+3z = Y2Cz), and also

p2(z)+q2(z ) = Y3Cz). The identity pn(Z) + qnCz) = Yn+l(Z) holds, therefore, for

n = 0,1,2. It appears reasonable to make a change of subscripts and set Pn+l(Z) =

pn(Z), Qn+l(Z) = qnCz), and to conjecture

THEOREM I. PnCZ) + Qn(Z) = yn(Z).

This statement holds, as seen, for n = 1,2,3, and, with previous convention,
p_l(Z) = P0(z) = I, q_iCz) = Q0(z) = 0, it is verified to hold also for

n = 0. The proof of Theorem I for all n is completed by induction on n as follows.

Identity (2), written out explicitly for PnCZ) and Qn(Z) reads:

Pn+ICZ) = (2n+l)z PnCZ) + Pn_l(Z)

(3)
%+iCz) = (2n+l)z QnCZ) + Qn_ICZ).

If we assume that Theorem 1 holds for all subscripts up to n, then (3) yields

Pn÷l(Z)+Qn+l(Z ) = (2n+l)Z(Pn(Z)+Qn(Z)) + (Pn_l(Z)+Qn_l(Z))

= (2n+l)z yn(Z) + Yn_l(Z),

by the induction assumption. But, by (3.7), the right hand side equals Yn+l(Z), so
that Theorem 1 holds also for n+l, as claimed.
In this decomposition of yn(Z), Pn(Z) is the polynomial containing all the terms

of yn(Z) with powers of z of the same parity as n, while Qn(Z) contains the terms

with powers of z of parity opposite to that of n. In other words,

1 1
Pn(Z) = ~- {yn(Z)+(-l)nyn(-Z)}, Qn(Z) = ~- {yn(Z)-(-l)nyn(-Z)}. With this we have

proved

THEOREM 2. The n-th convergent of the continued fraction [z,3z ..... (2n+l)z .... ]

e 2/z + 1 Yn(Z)+(-l)nyn (-z)

i8
of e 2/z - I yn(Z)-(-l)nyn(-Z)

2. In what follows we shall need the following

LEMMA i. Pn+l(Z)%(z)-Pn(Z)Qn+l(Z) = (-i)n+l.

The lemma would follow immediately, if we knew that Pn(Z) and Qn(Z) are coprime.

This is indeed the case, both, in the algebraic sense (Pn(Z) and Qn(Z) do not possess

any non-trivial polynomial divisor) and in the arithmetic sense (for integer m, the
 61

integers Pn(m) and qn(m) are coprime), as stated, e.g. in [II], p. 83, but it is

more convenient to prove the Lemma directly, without the need to invoke much classi
cal theory of continued fractions; the coprimality will then follow as an easy
corollary.
We verify that

Pl(Z)Qo(Z)-Po(Z)Ql(Z ) = poq_l-P_lq 0 = 0-i 2 = -I,

p2(Z)ql(Z)-pl(Z)Q2(z) = (3z2+1) -1-z(3z) = I,

P3(z)Q2(z)-P2(z)Q3(z) = (iSz3+6z)(3z)-(3z2+l)(ISz2+l) = -i,

so that the Lemma holds for n = 0,1,2. Let us assume that it has already been
verified for all subscripts up to n. One then observes that, by using (3) and the
induction hypothesis one obtains:

Pn+l(Z)%(z)-Pn(Z)%+l(Z) : [(2n+l)ZPn(Z ) + Pn_l(Z)]%(z)

- Pn(Z)[(2n+l)z%(z)+Qn_l(Z)] = Pn_l(Z)%(z)-Pn(Z)%_1(z)

=_ [Pn(Z)Qn_l(Z)-Pn_l(Z)Qn(z)] = (_i)n+l

and the Lemma is proved.

COROLLARY i. For all integers n and k, (Pn(k), Qn(k)) = I.

Proof. If d = (Pn(k)), Qn(k)), then d divides the left hand side of the identity

of Lemma I; hence d[(-l) n+l, so that d = i, as claimed.

COROLLARY 2.

Pn (z) Pn+l (z) (_I) n

%cz)
holds f o r a l l z E £, such that O,n(Z)Qn÷l(Z) # 0.

Proof follows trivially from the Lemma.

We recall (see [47]) that any real number falls between any two consecutive of
its convergents. More precisely, we have

COROLLARY 3. I f n is even and z = x is real and positive,then

Pn+l (z) e2/X+l Pn (x)

%÷iCz)
if n is odd, the inequalities are reversed.
 62

Proof follows from the preceding remark and Corollary 2, by observing that
e2/X+l
Pn(X)/Qn(X) is the n-th convergent o f - and that Qn(X)Qn+l(X) > 0 for x > 0
e2/X_l

COROLLARY 4. Regardless of the parity of n ar~ for all real x such that
Qn(X)Qn+l(X) ~ O,

Pn(X) e2/X+l I < i

0 < I On(X ) e2/X_l Qn(X)Qn+l(X) •

Proof. The condition Qn(X)Qn+l(X) ~ 0 implies x ~ 0. The first inequality then

follows from the fact that Pn(X)/Qn(X) is a convergent with following partial

quotient different from zero (in fact, by (i) a non-vanishing multiple of x); while
the second inequality follows from Corollaries 2 and 3.

e 2/x +i 2
We now observe that - - -i = - - and
e 2/x -i e 2/x -i

Pn(X) yn (x) + (-l)nyn(-X) (-l)nyn (-x)

%(x) -I = -i = , so that by Corollary 4,
Yn (x) - (- I )ny n (-x) Qn (x)

n
2 (-1) yn (-x)
I J- z ; - 1 %cx) I< l
Qn(X)Qn+l(X) , or, equivalently,

12 Q n ( X ) - ( - 1 ) n y n ( - X ) ( e 2 / X - 1 ) [ < (e2/X-1)/qn+l(X).

I f we r e p l a c e here x by 1/x and t h e n m u l t i p l y by x n we o b t a i n

(4) ]2xnQn(1/x)-(-l)nxnyn(-1/x)(e2X-1)l < (e2X-1)x2n/xnQn+ l ( I / x ) .

In t h e f i r s t member, xnQn(1/x) = ~ ( x ) , say, c o l l e c t s t h e terms with odd powers o f

@n(X), w h i l e ( - 1 ) n x n y n ( - 1 / x ) = (-x)nyn(1/(-x)) = @n(-X) and xnQn+l(1/x) =

x +l(X), a p o l y n o m i a l with o n l y even powers of x. The f i r s t member o f (4) may

now be written as 12 ~(x)-@n(-X )(e2x-l)l = 12Qn(x)+@n(-X)-@n(-X)e2Xl. But

2~(x) = @n (X) -@n (-X ) so that 2Q(x)+@n(-X ) = @n(X). Next,by a previous remark,

[n~2] . x2 n
xn%+l (l/x) = j=0 c.x
) 2], so that (e2X-l)/xnQn+l(i/x) =

2n " (2x) TM [n~2] cjx2 j _ (2n+2) '

x [ m' / . Here c o " as f o l l o w s from Chapter 2.
m=l j=0 2n+l ( n + l ) '
 63

This value is here without further relevance, except for c O ~ 0. If Ixl is less

than the smallest zero of the denominator, it follows that the right hand side of (4)

can be written as a series x 2n+l [ km xm and has a zero of order 2n+l.

m=0

Consequently, ~
also the entire function Rn(X) = @n (x) -@n (-x) e 2x, which, except

perhaps for sign, is the left hand side of (4), has a zero of order at least 2n+l at
x = 0. Let us replace here x by x/2 and set Rn(X/2) = Rn(X). The result obtained

so far may be written as

(6) On(X/2)-@n(-X/2)eX = R (x) = x 2n+l ~ c!x J.

n j=O J

Finally, let us c h a n g e n o t a t i o n s and s e t An(x) = @n(X/2), Bn(X) = - @ n ( - X / 2 ) ,

Then (6) r e a d s :

x 2n+l
Rn(x) = A n(x) + Bn(x) e = x ~ c!x ~.
j=0 j

This is precisely the expression encountered in Problem 1 of the Introduction and

shows that indeed, except for normalization, the polynomials A n(x) and Bn(X) there

defined, are BP. In fact, a count of the number of constants involved shows that,
except for an arbitrary multiplicative constant, An(X) = @n(X/2) and Bn(X) =

-@n(-X/2) are the only polynomials of a degree not in excess of n, and for which

the linear form Rn(X ) has a zero of order ~ 2n+l at the origin.
 CHAPTER 9
EXPANSIONS OF FUNCTIONS IN SERIES OF BP

i. The topic of expansions of arbitrary functions either of a complex, or of a real

variable in series of BP has not been explored fully and many open problems remain
to be answered. To date the most comprehensive treatment is that of R.P. Boas, Jr.
and R.C. Buck [13], but the literature available on this subject cannot be compared
to that on expansion in, say, Fourier series, or in series of classical orthonormal
polynomials, etc. In view of the orthogonality of the BP on the unit circle, as
presented in Chapter 4, it is reasonable to expect that the natural problem to
investigate is that of the expansion of functions f(z), defined for a complex
variable z, and continuous, or at least integrable on a set containing the unit circle,
or some compact neighborhood of the origin.
On the other hand, one may ask just for some formal expansions, say

f[z) ~ ~ CnYn[Z;a,b), or f(z) N ~ Cn@n(Z;a,b), and then inquire in what sense

n=0 n=0

the series "represent" f(z), under what conditions they converge and, if they do,
when do they converge to f(z). Even if the series do not converge, it may still be
possible to find summation methods, under which the series are summable to f(z).
Finally, one may raise questions concerning the uniqueness of such expansions.
The question of formal expansions into series of the BP yn(Z;a,b) has been

asked and answered already by Krall and Frink [68]. By using the orthogonality

relations of Chapter 4, it follows that,if f(z) = I CmYn(Z;a,b), then

m=0

I f(Z)yn(Z;a,b)p(z)dz = f ( [ CmYm(Z;a,b))Yn(Z;a,b)~(z)dz,
r r m=O

where P is any simply closed curve that encircles the origin, and p(z) = (2~i)-1e -2/z
if a = b = 2; otherwise O(z) = O(z;a,b), as defined in Chapter 4.
If r belongs to a connected domain on which the convergence of the series

CmYm(Z;a,b) i s s u f f i c i e n t l y strong to justify the inversion of the order of

m=O
summation and i n t e g r a t i o n , the last integral is equal to
~ (-1)n÷lb.n! r(a)
~ Cm g Ym(Z;a'b)Yn ( z ; a ' b ) ° ( z ) d z = Cn ( 2 n + a - 1 ) r ( n + a - 1 ) '
m=0 r

by Corollary 4.6. In the particular case a = b = 2, this simplifies to (see [68])

i
2~i f f(Z)Yn(Z) e-2/Zdz = (-l)n+l 2 Cn'
v 2n+l
 65

or

(i) Cn = (_l)n+l n 2~i

+ 1/2 f f(Z)Yn (z)e-2/zdz"
F

If f(z) = ~ bmzm, b m = f(m)/0)/m! , then (i) becomes (see [81])

m=0

co

Cn : (_l)n+l(n+ i) ~ f(m)(0) {l__!__ f F zmyn(Z)e-2/Zdz}.

m! 2~i
m=0

If we replace here the curly bracket by its value from Corollary 4.4, we obtain

" f(m)(o) m!
Cn = (-l)n+l(n+ 9 ) m~0= m~ (-2)m+l (m+n+l)!(m-n)!

fCm) (0) (-2) TM

(-l)n(2n+l) [ (m+n+l)!(m-n)'
m=0

In the sum all terms with m < n vanish; in the others we set m = n + v, ~ = 0,I,...
and obtain Nasif's [81] formula:

~ f (n+v) (0) (-2) 9

cn = (2n+l).2 n ~ v! (2n+v+l) '
v=0

This, of course, is all formal work. In general, there is no guarantee that

the series ~ CnYn[Z;a,b) converges, or even if it converges, that its sum is f(z).
n=0

The determination of necessary and sufficient conditions on f(z), for the convergence,
or the summability to f(z) of the series appears to be an open problem.

2. Similar considerations hold if we consider expansions of functions f(z) into

series ~ cn en[Z;a,b). It is clear, however, from the work of Burchnall (see

n=0
Corollary 4.15) that the en(Z;a,b) themselves do not form a system of polynomials

orthogonal on the unit circle, because the factor z-(m+n+a)e -bz is not independent of
the polynomials em and en involved. A somewhat different approach, however, permits
one to obtain expansions of entire functions in series of the polynomials en(Z;a,b).

It turns out that we obtain particularly simple results, if we set, at least provi-

2n
sionally, pn(Z) (=pn(Z;a)) = ~., en(z;a,2 ). In this case, the theory of generali-

zed Appell Polynomials, due to Boas and Buck [13], [5], permits one to obtain not
only the coefficients c n of the formal expansions, but also to determine the number
 66

of distinct expansions (the expansions, unfortunately,are in general not unique),

their respective regions of convergence and the regions of sun~nability for these
series of generalized Appell Polynomials. The main results and sketches of the
proofs will be given here, but for detailed proofs it is suggested that the reader
consult the original presentations [5] and [13].

3. The fundamental ideas may be traced back, on the one hand to Whittaker [66], on
the other hand to the classical work of E. Borel, G. P61va,
• Phragmen
z and Lindelof
(see [32], vol.2 and [4]).

Let A(w) = [ anwn a 0 ~ O; and g(w) = [ gn wn gl ~ 0 be given functions

n=O n=l

holomorphic in some neighborhood of the origin. Let ~ be the largest simply connec-
ted region containing the origin, in which A(w) is holomorphic, and let ~w ~ ~ be

the largest simply connected subregion of ~, in which ~ = g(w) is univalent. Define

also Aw as the largest disk centered at the origin and such that h w ~ ~w~. Due

to the univalence of g(w) we have a i:i correspondence between the points w e ~w and

the images ~ = g(w). In particular, let ~ be the image of fi and A be the image
w
of Aw, so that A ~ c _ ~ in the ~-plane.

To each entire function f(z) = [ f zn we associate its Borel (or Laplace)

n=O n

transform F(w) = ~ n!fn

n+l (see [iO], p. i13), holomorphic outside some circle
n=O w

Iz] = r. We recall the inversion formula:

i eZW F I eZW ~ n!fn

(2) 2~---~ IC (w)dw = ~ IC [ --~ dw =
n=O W

z m w TM ~ n !f ~ m
1 IC (mE __~.t )(n ! w~l )dw = [ z I n !fn • 2-~
1 IC n-m+l
dw
2~i =0 0 m=O ~., n=O w

oo

= [ f zn = f ( z ) ,
n=O n

where C is a circle of radius r + e.

 67

~-plane

w-plane W ~

(
Figure 1

We shall be concerned mainly with entire functions of exponential type, i.e.,

functions f(z), such that If(z) I ~ Me alzl for finite, positive constants M, ~ and
all complex z. If T = inf a for which this inequality holds, f(z) is said to be of
exponential type T. By D(f) we denote the complement of the set of regularity of
the Borel transform F(w) of f(z).
We now define generalized Appell Polynomials pn(Z) by the generating function

(S) A(w)eZg(w) =
[ pn(Z)W n.
n=O
Two remarks are in order. The first is that the theory has been developed by Boas
and Buck [13] in a more general framework, in which the exponential function et is

r e p l a c e d by any " c o m p a r i s o n f u n c t i o n " ~ ( t ) = Ont n , s u b j e c t o n l y t o t h e r e s t r i c -

n=O
t i o n s t h a t none o f t h e c o e f f i c i e n t s ~n v a n i s h and t h a t t h e r a t i o ~n+l/~n_ d e c r e a s e s
 68

monotonically to zero. Clearly, the exponential function qualifies as "comparison

function" with ~n = i/n~ # 0 and ~n+I/~n = i/(n+l) ÷ 0. The condition on ~n+i/~n

implies, of course, that ~(t) (here et) is an entire function.

The second remark is that the case g(w) = w corresponds to the classical Appell
polynomials. Indeed, (3) then becomes

(3') A(w)eZW = [ pn(Z)W n

n=0
and we obtain by differentiation with respect to z: A(w) wezw = [ p~(z)w n, so that
n=0

A(w)e T M = ~ p~(z)w n-l, or, equivalently,

n=l

(4) A(w)e T M = [ p~+l(Z)W n

n=0

and a comparison with (3') shows that the polynomials pn(Z) satisfy the differential

equation Pn+l(Z) = pn(Z) that characterizes Appell polynomials.

We now return to the general case. By use of the univalent map ~ = g(w), we
can reduce the general case to a more manageable one, close to the classical (4).
Indeed, let w = W(~) be the map inverse to ~ = g(w) and set A(w) = A(W(~)) = B(~).
To any compact set C ~ A ~ corresponds a class of entire functions f(z), such that

D(f) C C; let us denote that class by K[C]. Let F be a simply closed curve in A ,

enclosing C and passing through no zeros of B(~). Now (3) becomes:

eo

(5) B(~)e~Z = ~ pn(z)W(~) n,

n=0
co

and it follows that on r, ~ pn(Z)W(~)n/B(~) converges uniformly to e ~z.

n=O

We now set

(6) Cn = in(f ) = i W(~) n F(~)d~

and obtain, using (2), that

1 F(~) PnW(~)n)d~ = ~1 IF F(~)eZ~d~ = f(z).

n--0 CnPn(Z) = ~ IF ~ (n 0

The interchange of summation and integration is justified for f c K[C], by the

oo

uniform convergence of [ pn(z)W(~) n.

n=0

In order to have f e K[C] we need to know that D(f) c C~A~. In particular, if

r I is the radius of DI, the largest disk, centered at the origin and inside A~, a
 69

sufficient condition for D ( f ) ~ D1 ~A is that T < r I. We consider also D2, the

largest disk centered at the origin, such that D 2 ~ ; let r 2 be the radius of D 2.

The following theorem holds and most of its assertions follow readily from the
preceding discussion.

n
THEOREM i. Let A(w) = [ an wn, a 0 # 0 and g(w) = [ gn w ' gl ~ 0 be holomorphic
n=0 n=l

in some neighborhood of the origin, with g(w) univalent on nw; let W(~) be the

function inverse to g(w), defined on ~ , the image of ~w' set A(W(~)) = B(~) and

define the polynomials pn(Z) by (5).

Let £(z) = ~ f zn be an entire function of exponential type with Borel-

n=O n

Laplace transform F(~), and define the constants cn by (6). Then f(z) i8 repre-

sented by the series f(z) N ~ CnPn(Z). Furthermore, if D I, r I and D 2, r 2 are

n=0

defined as above and if the type • of f(z) satisfies r < r I, then f E K[C] for some

C~D[f) and the series converges to f(z). If T < r2, then the series is Mittag-

Leffler svz~nable to f(z).

RE~RK i. These conditions insure convergence, or summability for all entire func-
tions of exponential type T, bounded by rl, or by r 2, respectively. However, what

one needs is only the existence of simply closed curves F, that encircle D(f).
Hence, for specific functions, for which D(f) is explicitly known, one can often
establish convergence, or summability, when the sufficient conditions T < r I, or
T < r 2 do not hold.

RE~RK 2. We abstain here from a proof of the Mittag-Leffler summability in the case
r I < r < r2, for which the reader may want to consult [13].

We observe that F(C) has an essential singularity at the origin (unless f(z) is
a polynomial, in which case T = 0 and F(C) has a pole at the origin), so that the
cn do not all vanish. B(~) may, or may not have zeros. If it has no zeros, then

a~y F selected as before, will lead to the same values for cn = in (f) and the series

obtained coincides with Whittaker's basic series [66]. If, however, B(~) has zeros
in A , outside D(f), then the values of the c n will, in general, depend on the set

of zeros of B[~) enclosed by F, so that the expansion of f(z) need not be unique.
Further different expansions may be obtained, if one selects F outside ~ (so that
 70

g(w) = ~ is no longer univalent), but, fortunately, these complexities need not

worry us in the present case.

4. We start from Burchnall's generating function (6.6). As usually we shall set

b = 2 (otherwise, we make the change of variables z'/2 = z/b) and, writing also 2z
for z, obtain

{~1 [1+(1_4zt)1/2]}2-a(l-4zt)-l/2exp{[1-(l-4zt)
i/2]/2z} = Z Yn(2Z;a)-- tn
n!
n=0

The polynomials Yn(2Z;a)(=Yn(2Z;a,2)) as here defined, do not appear to be (generali

zed) Appell polynomials, according to the definition (3). Let us replace, however,
t by w/z; we obtain

= Z
wl [l+(l_4w)i/2}Z-a(l_4w)-i/4exp{[l_(l_4w)i/2]/2z} i Yn (2z;a) wn
{~ n n!
n=O z

2n 1 ~ 2n @n((2z)-I a)w n
Z n! Yn (2z'a)wn = Z ~., ,
n=O (2z) n ' n=O

or, replacing i/2z by z,

2n n
1 [l+(l_4w)i/2}2-a(l_4w)-i/4exp{z[l_(l_4w)i/2]} = Z ~., On(Z;a)w •
n=O

This is precisely (3), with A(w) = 1 [1+(l_4w)i/212-a(1_4w)-1/4, g(w) = l-(l-4w) I/2,

n
2
and pn(Z) = ~., On(Z;a).

In order to insure that A(w) and g(w) are holomorphic, it is sufficient to insure

the singlevaluedness of (l-4w) 1/2. A convenient way to obtain this is to cut the
complex w-plane along the positive real axis from + 1/4 to ~. One easily verifies
that g(w) is also univalent in the cut plane; hence, the cut plane is flw and A w is

the d i s k lwl < 1/4.

The image ~ of ~w under ¢ = g(w) = l-(l-4w) I/2 is found, by observing that the

image of the boundary [1 < w < ~ of flw becomes 1 % i(4w-l) I/2 = 1 ; iv (0 < v < =)

in the c-plane. It follows that fl~ is the half-plane Re ~ < i. The function

w = W(~), inverse to g(w) is obtained from (l-4w) I/2 = i-~ l-4w = I-2~+~ 2
1 1
w = $ ~(2-¢). The image of Aw, i.e., of the disk lw] < T is, therefore, that part

of the inside of the lemniscate 1c(2-c) I < 1, that is contained in ~ , i.e., the

left loop of it. The lemniscate is symmetric with respect to the x-axis, which it
meets in the points ~2-2~± 1 = 0. For the plus sign we obtain the double point
 71

t = i of the lemniscate, on the boundary of at; for the minus sign we obtain the

vertices of the lemniscate, at t = 12 ¢~. The first one,t = 1 + /2 , is not in

~t' while ~ = I-/2 = -(fr2-1), is in ~ . The largest disk D 1 inside A has, there-

for, a radius r I = ¢~ -i. This requires some verification, in particular, that no

point of the lemniscate is closer to the origin than this vertex at _(¢r~ -I).

E-plane

w-plane D2 Re t = 1

4" -i 1 +/2"
= ~w

fit

Figure 2

The disk D 2 has the radius r 2 : I. A simple computation shows that B(t) = A(W(~)) =

{~1 [1+1-~]}2-a(i-~)-1 = (i- t/2)2-a/(l-~)is singlevalued and does not vanish in ~ ,

2n
where Re g < i. It follows that the s e r i e s Cn@n(Z;a) has uniquely defined
n=O

coefficients c
n
. For f(z) = Z fn zn, Ln(f ) becomes successively
n=O
1 i ~n(2-g)n(1- t) 1 1
F ( ~ ) d ~ = - - - - I F ~n(1-g)(1-~/2)n+a-2F(g)d~
2~---i-IF ~ (1_~/2)2-a 2n 2~i

m!f
2nl 2~ii I F ~n(l-~)(l-~/2)n*a-2( --~)d~m =
m=O

1 ~ 1 (1-~)(1-~/2) n+a-2
2--
ff [ m! fm 2-~ IF m-n+l dE.
m=O
 72

n+a-2 (-i) r (n+a-2.

In particular, for integral a ~ I, the numerator equals r=O 2r - r )( r_ r+l)

n+a-2
and the integrand becomes [ (-i) r -n+a-2) 1 1 }. The integrals
r=0 2r [ r { m-r-n+l m-r-n

vanish, except for m = r+n and m = r+n+l. The first case contributes
n+a-2
r=0
[ (-l)r2
r (n+~-2) fr+n(r+n)! ' while the second contributes

_n+i-2 (_i) r
r=0 2r (n+$ 2)fr+n+l(r+n+l)!, so that

n+a-2 (-i) r .n+a-2.

i
(7) Cn = Ln(f) = 7 r=O 2r [ r )(r+n)!{fr+n-(r+n+l)fr+n+l}.

In the important particular case a = 2, this formula simplifies to

°i i (1)r
c 2n 2-7-- !{fr+n-(r+n+l)fr+n+l}
n r=0

and the coefficients of en(Z) in the expansion of f(z) are

2n n
[ (-I)r (r+n)! (r+n+l)
n-T Cn = r=O 2 r r!(n-r)! {fr+n- fr+n+l }

C8) n
= [ (-l)ra~n)(fr+n-(r+n+l)fr+n+l),
r=O
n
with the a (n) given by (2.8), as coefficients of yn(Z) = ~ a (n) zr.
r
r=0 r

On account of the fact that r I = / 2 - i and r 2 = i, it follows that, as long as

2n
a is a natural integer, the expansion f(z) = ~ ~., Cn@n(Z;a ) with the cn given by
n=0
[7) are convergent and converge to f(z), at least for entire functions f(z) of ex-
ponential type T < a - i . If Y < i, then the series is at least Mittag-Leffler sum-
mable to f(z).

5. As an application, let us find the expansions of powers of z in series of BP. If

fCz] = z k, then fk = i, fr = 0 for r ~ k, so that, by (8),

k (-I) r k! k'
z = [ 2r r:(n-r)' 8n(Z) - [ (-l)r
n+r=k " n+r+l=k 2r r:(n-r)! en(Z)"
0<_r<n 0<r<n

In the first sum r < n < k-r, 0 < n-r <_ k-2r and the sum extends over 0 < r <_ k/2;
 73

in the second sum, similarly, 0 < n-r < k-2r-i and the sum extends over 0 E r < (k-l)/2
Consequently,

k k! k 2 (_l)r 1 (k-l)/2 (-1) r 1

z = r=0 2r r!(k-2r)' ek-r(Z)-k" ' ~ 2r r! (k-2r-l)' . ek-r-l(Z)"
• r=0

In the second sum replace r+l by r, which will then run from 1 to (k+l)/2 and reads,
(~+1)/2 r
including sign, (-i) 1
r=l 2r-1 (r-1)!(k-2r+l)! @k-r (z)

= (~+1)/2 (_l)r.2r
ek_r(Z ). For 1 < r < k/2 we can combine the two sums.
r=l 2rr~(k-2r+l) l

The general term is

k! (-l)r (k-2r+l)+2r (k+l)'

2r r!(k-2r+l)' ek-r(Z) = (-l)r " @k-r (z)"
• 2rr!(k_2r+l)!

Finally, we verify that for r = 0 and r = (k+l)/2 this general term represents pre-
cisely the omitted terms, the one for r = 0 of the first sum, and the one for
r = (k+l)/2 of the last sum, respectively, so that we obtain the final result:

k (k+l)/2 r e k - r (z)
(9) z = (k+l): [ (-1)
r=0 ,2rr!(k-2r+l)!

found by C a r l i t z [19]. The c o r r e s p o n d i n g r e p r e s e n t a t i o n of z k by sums o f t h e p o l y -

nomials yn(Z) i s (in a somewhat d i f f e r e n t notation) in Dickinson [47] and A1-Salam

[3] and reads

k
k 2r+l
z = 2kk! [ (-1)r (k-r)'(k+r+l)! Yr (z)"
r=0
This formula may be proved by u s i n g (1) t o g e t h e r with C o r o l l a r y 4.4.
Let us v e r i f y (9) f o r i n s t a n c e in t h e p a r t i c u l a r case k = 3:

3 , 1 1 + __/__1 el(z)}
z = 4.{~-., e3(z ) 2-1!2! e2(z) 222!0,

= e3(z) - 6e2(z ) + 3 e l ( z ) .
If we substitute for ej (z) the corresponding polynomials (see Chapter 3) we verify

that indeed [z3+6z2+iSz+lS) - 6(z2+3z+3)+3(z+l) = z 3.

6. As another application, let us find the expansion of the exponential function

k
f ( z ) = e az = ~ fkzk. Clearly, fk = ~a and, by (8),
k=0
2n Cn =
n-T n~ ( - 1 ) r (r+n)! {a r+n ~r+n+l(r+n+l)
r=0 2r r! (n-r) ! (r+n) ! (r+n+l) ! }
 74

n r ar+n n n n
[ (-I) (l-a) _ (l-~)a (2)r
r=0 2r rI(n-r)! n! r=0~ (-l)r()

(l-a) (l-a/2)n~n/n~ .

It follows that the expansion of e ~z in a series of BP @n(Z) is:

(i0) e az - (l-a) [ (a-a2/2)n@n(Z)/n: .

n=0
One easily verifies that (i0) is precisely the result that one obtains, if one diffe
rentiates Carlitz' formula(6.17)with respect to u and then sets 2u = a.
As for convergence, it is clear that the right hand side of (i0) converges for
all complex values of a and z. Also, we recall that the Mittag-Leffler summability
method is regular, so that, for convergent series, it leads to the same value as the
direct summation of the series. Hence, for lal = T < i, the series (i0) converges to

the function eaz ~at happens for I~I ~ i? One observes that for s = I, the right
hand side of (i0) vanishes, while the function e z does not. Hence, although the
~Z
series converges for all a, for lal > 1 it does not converge to the function e
One may observe, for instance, that for a = 2, the sum of the series equals -i,
2z
which is different, in general, from f(z) = e
If we set z = 0 in (i0) and replace @n(0) by its value (2n) I/2nn2, we obtain
after a short computation that, for lal < i/2,

(ii) ~ (2n
n ) (a-a2) n = l__!___= [ 2mare.
l-2a
n=O m=O

From (ii) one can obtain many combinatorial identities. In particular, by

comparing the coefficients of a k, it follows that

k~2 ..r.2k-2r. k-r 2k.

(-lJ [ k-r ] ( r ) =
r=0
 PART III
CHAPTER i0
PROPERTIES OF THE ZEROS OF BP

i. In this chapter we shall be concerned mainly with properties of the zeros of the

simple BP, either yn(Z) J

or On(Z), which we denote by a~ n) and Bk(n) (k = 1,2 ....
"
n) '

respectively. W~enever possible (and convenient) we shall extend the results to the

zeros of yn(Z;a) , or en(Z;a), d e n o t e d by a ~ n ) ( a ) , o r 8 k( n ) ( a ) , respectively, for ar-

bitrary a (often, h o w e v e r , we s h a l l have to restrict ourselves to values a ~ 2).

From t h e s e , of course, analogous results for the zeros of yn(z;a,b), or en(Z;a,b)

(which we may d e n o t e by o ~ n ) ( a , b ) - and g ~ n ) ( a , b ) , respectively) can be o b t a i n e d imme-

diately by a change of variable and we shall usually refrain from writing out expli-
cit results for b # 2. Also, whenever possible the subscripts and/or superscripts
will be suppressed.
From

n
(1) z Yn(1/z) = 0n(Z) = eZCn(Z)

it is clear that t h e z e r o s o f On(Z) and of Cn(Z) a r e t h e same and a r e t h e r e c i p r o c a l s

of the zeros of yn(Z), so t h a t Ok8 k = 1 (k = 1,2 . . . . . n). These considerations remain

valid e v e n f o r a ~ 2 / b, i n f a c t , even for Cn(Z;a,b,c). Hence, any i n f o r m a t i o n about

t h e z e r o s o f any one o f t h e s e functions c a n be t r a n s l a t e d immediately into information

about the zeros of all o f them.
Some o f t h e p r o p e r t i e s of the z e r o s o f BP h a v e a t t r a c t e d the attention quite
early (see [17] and [ 5 3 ] ) . Others, s u c h as good u p p e r and l o w e r b o u n d s f o r [Ok], or

[Ski, respectively, were d i s c o v e r e d and i m p r o v e d o n l y l a t e r .

2. Some o f t h e o l d e s t known r e s u l t s are formulated in the following three theorems.

THEOREM 1. ([17], [53], see also [47].) (a) A l l z e r o s o k (k = 1 , 2 . . . . . n) o f y n ( Z )

(or 8k of On(Z)) are simple. (b). No two consecutive polynomials y n ( Z ) , Yn+l(Z)

(or On(Z), en+l(Z), respectively) have a zero in comwnon. (c) No BP of even degree

has a real zero and those of odd degree have exactly one real, negative zero.

THEOREM 2. (see [53]). (a) All zeros of the BP yn(Z) satisfy [o~n)[ < I, except

for ~i) =-i; correspondingly, 18~n) l > i, except for 8~i) = -i. (b) The real

zeros °(2m-l)m o~. the polynomials Y2m_l(Z) form a monotonically increasing sequence:
 76

-i = al(1) < a~3) < .. . < ~(2m-l)m < O.

THEOREM 3. ([17]; see also [62] and [2]). The zeros a r = a r(n) o f yn(Z) s a t 4 s f i y

the equations
n n
2m-i
(2) Z %:-i, Z % = 0 for m = 2,3 . . . . . n.
r=l r=l

Correspondingly, the zeros 8 r = 8 r(n) of @n(Z) satisfy

n n
- l-2m
(2') Z ~rI =-l, Z ~r = 0 for m = 2,3 ..... n.
r=l r=l

3. P r o o f o f Theorem 1. (a) If a is a multiple zero of yn(Z), t h e n Yn(a) =

!
y~(a) = 0. By ( 3 . 1 0 ) , also Yn_l(a) = 0 and by ( 3 . 1 6 ) i t now f o l l o w s that Yn_l(a) = 0

Hence, if a is a multiple zero of yn(Z), it is also a multiple z e r o o f Yn_l(Z) and

s o , by i n d u c t i o n on n, a m u l t i p l e zero of yl(z) = z+l, which is absurd. (b) By

(3.7) it follows that i f a i s a common z e r o o f Yn+l(Z) and y n ( Z ) , then it is also a

zero of Yn_l(Z). By i n d u c t i o n on n i t is also a zero of y2(z) and o f y l ( z ) , i.e.

a = -1. However, y 2 ( - 1 ) # 0 and t h e a s s e r t i o n follows. (c) This assertion follows

a l m o s t i m m e d i a t e l y from t h e t h e o r y o f t h e B e s s e l f u n c t i o n s K (z) f o r ~ = n + 1/2

(see, e.g. ~], p. 377 and F i g . 9 6 ) , b u t can a l s o be p r o v e d d i r e c t l y , as follows.

By (1) i t is sufficient to prove the statement f o r @n(Z). We r e c a l l that f(z) =

@n(Z) and g ( z ) = @n(-Z) a r e b o t h s o l u t i o n s (easily s e e n t o be i n d e p e n d e n t ) of (2.13).

It follows that

zf" - 2nf' = zf,

zg" - 2ng' = zg,

d
whence z (f'g - fg') = z(gf" - fg") = 2n(f'g - fg'), or, by integration,

(3) f,g - fg, = Cz 2n.

While i r r e l e v a n t for our present purpose,we observe that (2.2b) and ( 2 . 5 ) lead to

lim z-2n(f'g-fg ' ) = 2 ( - i ) n+l and (3) shows t h a t this value remains unchanged for
Z ~

all z, so that C = 2(-1) n+l.

We now c o m p l e t e t h e p r o o f o f ( c ) , by f o l l o w i n g Burchnall [17]. We r e t u r n to
On(Z), r e p l a c e f and g i n (3) by t h e i r respective values and o b t a i n

d - z -z d
d--{ (e Z e n ( Z ) ) - ( e e n ( - Z ) ) - ( e On(Z))- ~-~ (eZen(-Z)) = (-l)n+l.2z 2n. This simplifies
 77

t o (see [17])

(4) -2On(Z)On(-Z)+O~(Z)On(-Z)+On(Z)O'(-z) = 2 ( - 1 ) n + l z 2n.

The coefficients of On(Z) are all positive. Hence, any real zero of 0n(Z) is negative

Let -a, -8(a > 0, ~ > 0) be two consecutive real zeros (assuming that such exist ).
Then (4) shows that

8~[-a)en(a ) = 2(_l)n+l~ 2n,

and similarly

e n C - S ) e n ( 8 ) = 2 ( _ l ) n + l B 2n.

As 0n(a ) and 0n(B ) are both positive, O~(-a) and @~(-B) have the same sign, which

is impossible for consecutive zeros. Hence, the assumption that there exist two
real zeros is false. Finally, the reality of the coefficients of en(Z) shows that

complex zeros occur as complex, conjugate pairs and Theorem 1 is completely proved.

4. Proof of Theorem 2. The proof of Theorem 2 is based on a Theorem of Enestr~m

[14] and Kakeya [40]. We shall state it in the stronger version due to Hurwitz [34],
in the form of a sequence of three Theorems. For completeness, we shall sketch their
proofs, but the reader, willing to accept their statement, may skip these.
n
THEOREbl A. Let b 0 > b I > ... > b n > O; then the polynomial f(z) = ~ b.z J has
_ _ 9_,=~
v J
no zero with Izi < i.

Proof. Let Izl = r < i; then, with b_l = bn÷ I = 0,[f(z)(1-z)i=

i n " I n n+l
b° - j!l (bj l-bj)zJ-bnzn+l > b 0 - j=l~ (bj-l-bj)rJ - bnrn+l = - j=0~ (bj-l-bj)rJ"

n-i
consequently, {f(z)(1-z)] >- - j=O
[ (bj-l-bj) r j = f(r).(1-r) > 0 "

THEOREM B. (Kakeya). Let the coefficients bj of f(z) be positive and set

Maxj (bj/bj+ I) = O, minj (bj/bj+l) = o; then all zeros zk (k = 1,2 ..... n) of f(z)

satisfy the inequalities e < ]Zk] < p.

n b.p j b.
Proof. Consider f(zp) = ~ bjpJz j = g(z). Then J 1 hence,
j=0 bj+ipj+l ~ ~ Maxj bj+l = i;

the coefficients of g(z) are increasing, those of h(z) = zng(I/z) are decreasing and
Theorem A applies to h(z). If u is a zero of h(z), i/u is a zero of g(z) and p/u
is a zero of f(z). By Theorem A, lul ~ i, so that any zero zk of f(z) satisfies
 78

[Zk[ = [ p / u [ ! o, as claimed. The o t h e r inequality follows in the s a m e way by

consideration of znf(~/z).

THEOREM C. (Hurwitz). The e~ation f(z) = 0 with positive, monotonically decreasing

coefficients has a zero of absolute value e ~ a l to one, if and only if the coeffi-
cients can be partitioned into sets of m consecutive e ~ a Z coefficients, where m is
a divisor of n+l.

n+l
i
Proof. For a zero z, from f(z)(1-z) = - ~ (bj_1-bi)z~ = 0 follows

n+l
(bj-l-bj )z J = b o" For Iz[ = 1 also follows
j=l
n+l n+l n+l
l(bj_l-bj)zJl = ~ (bj_l-b j) = b o = i I (bj_l-bj)zJl.
j--i j=l j=l

Consequently, all summands have the same argument, i.e. (bj_l_bj)z j = pje ia ,

n+l n+l
real, pj > O. In fact, a = O, b e c a u s e ~ (bj 1-bi)zJ = e ia [ pj = b .
- j=l - " j =1 o

From bnzn+l = Pn > 0 now follows zn+l > 0 and Using also Izl = i, zn+l = i. Let

m be the smallest integer such that zm = I; then m divides n÷l and m >_ 2 (otherwise

n . n z-3
z = 1 and 0 = ~ b.z J = ~ bj > O, a c o n t r a d i c t i o n ) . From ( b j _ l - b j ) = pj > 0
j=0 J j=o

follows, for m ~ j , that b j _ l - b j = O. In particular, b o = b I = ... = bm_l; bm =

bm+ I = ... = b2m_l, etc. which proves the necessity of the condition. On the other

t km m-I "
hand, if the condition on the b.'s holds, then f(z) = ( ~ bkm z )( [ zJ), with
J k=O j =0
t = (n+l-m)/m, so that f(e 2~ir/m) = 0 for r = 1,2,...,m-l, and this proves the

sufficiency. The proof of Theorem 2(a) is now i m m e d i a t e . From ( 2 . 8 ) it follows

n

that the coefficients o f @n(Z) = ~ an_mZm a r e decreasing from an = (2n)!/2nn : to

m=0
a ° = 1. Hence, Theorem A is applicable with b m = an_ m a n d a l l zeros 8 k o f @n(Z )

a b
n-nl m _
satisfy [Bk[ >_ 1. Going beyond this, we v e r i f y that, r m = an-m-1 = bm+l

1 (m+l)[2n-m) (m = 0 , 1 .,n-l), with p = Max r = n(n+l)/2 and ~ = min r = 1.

2 n-m ''" m m
m m

Theorem 2(a) is now a simple corollary of the stronger

 79

THEOREM 4. For n > 1, a l l zeros 8 k = 5~ n) of 8n(Z) s a t i s f y 1 < IBkl E n ( n + l ) / 2 .

The z e r o s a k o f yn(Z) (n > 1) s a t i s f y 2/n(n+l) ~ Jak[ < 1. For n = 1, t h e u n i q u e

zero of e1(z) and of YlCZ) is a~1) = B~I) : -1.

Proof. On account of p = n(n+l)/2, o = 1 and of Theorem B, the only statements of

Theorem 4 still to be justified are the omissions of equal signs for n > I. For
that we use Theorem C. The equal sign requires that all coefficients fall into sets
a
n-m
of at least two consecutive equal coefficients. However, the ratio r
m an-m- 1

equals (m+l)C2n-m)/(2n-2m) > m+l >_ I, with equality only for m = 0, when r 0 = i.

Hence, equality can exist only between the first two coefficients. The sets of
consecutive equal coefficients required hy Theorem C reduce, consequently, to a
single set of two coefficients, so that n = i, when yn(Z) = 81(z ) = z+l, and when
we have indeed a I = 81 = -I.

Theorem 4 and with it Theorem 2(a) is proved.

The proof of Theorem 2(h) is based on (3.7). Let a (n) be the unique real zero

of yn(Z) (n odd). Then (3.7) with n-I instead of n and z = a (n-2) becomes

YnCa in-2)) = (2n-l)a(n-2)yn_l(a(n-2)). By Theorem i, ~(n-2) < 0 and Yn_ICa (n-2)) > 0

because n-i is even, Yn_l(Z) does not change its sign and Yn_l(0) = 1 > 0; consequent-

ly, }'n(a i n - 2 ) ) < 0, YnCa in)) = O and, by Theorem 1, yn(Z) > 0 f o r z > a (n), so t h a t

a (n-2) < a (n) < 0, as c l a i m e d .

S. Proof of Theorem 3. (see [17]). By logarithmic differentiation of (i) we obtain,

successively:

@n'CZ) 8'CZ)n n 1 n 1 1 n ~ zk-I

%c=---Y + l = - = =- Z - - - - I l
@nCz) r=l Z-Br r=l fir l-Z/Br r=l k=l Bk
r
CS)
n m
= - Z zk-i where ~m = Z ~r"
k=l U-k' r=l
By Theorem (2.3), ~n(Z) contains no odd powers of z with exponent less than 2n*l;

it follows that neither ~ ( z ) , nor ~ ( z ) / @ n ( Z ) contains even powers of exponent less

than 2n. Hence, by comparing coefficients in (S) for k odd, 0 < k-I < 2n, one has
n
O_k = 0. For k = I, by (S), 1 = -a_l. Consequently, ~ Bi-2mr = 0 for m = 2,3 ..... n,
r= 1
n
while 8 -I = -i. We recall also that the a ' s are the reciprocals of the _
8r'S
r r
r=l

and with this Theorem 3 is proved. For a proof of the converse, namely that every
 8O

solution of the system (2) (or (2') consists of the zeros B$n) of @n(Z) (or a(n)r of

yn(Z), respectively) taken in some order, see [17].

6. It is easy to improve some of previous results. First we observe that the proofs
of Theorem l(a) and (b) go through for yn(z;a), provided only that we use (3.21) in-

stead of (3.10); (3.22) instead of (3.16).

Next, the statement of Theorem l(b) can be strengthened to read as follows
(see [47]).

THEOREM l(b'). If n # m, then no zero of yn(Z) can be a zero of ym(Z); in fact, no

zero of yn(Z;a) can be a zero of ym(Z;a).

Proof. For the proof of the first statement (see [47]), it is sufficient to use
(3.7') instead of (3.7) and for the second one to use the obvious generalization of
(3.20) instead of (3.7').
We have proved

THEOREM i'. The statements of Theorem l(a) and l(b') hold for the generalized BP.

REN~RK. The proof of Theorem l(c) does not translate immediately into one for
yn(Z;a). The problem of determining the values of a for which Theorem l(c) holds
appears to be open.

Also Theorem 2(a) can be extended to the case a # 2. Indeed, by (2.27), with
d b
the coefficients d k = d~ n) of @n(Z;a ), rm = dn-m-ln-----~m
= = bm + l (m+l)(2n-m+a-2)2(n_m) . If we

consider for a moment m as a continuous variable and take the derivative, we verify
a-2
that in general r m increases monotonically with m, from o(a) = 1 + ~ for m = 0

to o(a) = n(n+a-l)/2 for m = n-l. An exception can occur only for n < -a+2, i.e.,
only for a < 2, and even in those cases only for finitely many values of n. As
these can be studied individually, we shall ignore them here and assume that
n ~ -a+2. By using Theorems A and B as before we obtain

THEOREM 2'. For n B Max(-a+2,2) (i.e., if a > O, for all n > 2)

1 + (a-2)/2n ~ IBk(a)[ < n(n+a-l)/2,

2/n(n+a-l) 2n/(2n+a-2).

For n = i, the unique zero of yl(z;a) is ~) = -2/a and the zero of @l(z;a) is

S~ I) = - a / 2 .
 81

7. In previously mentioned paper [47], Dickinson, by using the connection of BP with

Lommel polynomials (see Chapter 5), showed that no a kon) is purely imaginary. We

shall use a different method (see [12]) to prove the stronger result (see [iii] and
[9s]) of

TIIEOREM 5. For a l l n, the z e r o s a~n) o f yn(Z) s a t i s f y Re a~ n) < O.

In the proof we use the following known theorem.

n . (n+~)/2 n-2k+l
TIIEOREM D. Let P(z) = ~ a.z n-3 and set Q(z) = L a2k_iZ
j =0 3 k=l
All the zeros of P(z) have negative real parts, if and only if, in the notation of
Chapter 8, Q(z)/P(z) = [0,ClZ÷l, c2z ..... CnZ ] with the coefficients cj (j = 1,2 ..... n)

all positive.
For a proof of Theorem D, see, e . g . [111].

Proof of Theorem 5. In the notations of Chapter 8,

1
yn(Z) = Pn(Z) + Qn(Z), with Qn = 2 {Yn (z) - ( - 1 ) n Yn(-z)}"

Yn (z) Pn (z)
By Theorem 8.1, Qn(Z) - 1 + Qn(Z-----
~ By Theorem 8.2 this is the n-th convergent of

e2/Z+l
= 1 + [z, 3z . . . . . (2n-1)z . . . . ] = [l+z, 3z . . . . . (2n-1)z . . . . ].
e
2/z 1

Hence,
%(z3
yn(Z ) - [0, l+z, 3z . . . . . (2n-1) z].

As a l r e a d y observed i n Chapter 8, Qn(Z) c o n t a i n s p r e c i s e l y a l l powers of yn(Z) o f

p a r i t y o p p o s i t e to n, so t h a t we may i d e n t i f y Qn(Z) and y n ( Z ) , with Q(z) and P(z)

r e s p e c t i v e l y of Theorem D and the r e s u l t f o l l o w s .

Wimp [112] proved t h a t Theorem 5 holds a l s o for the zeros akn)(a)" of yn(Z;a) at

l e a s t for a > 2.

8. Another result of Dickinson [47], namely that the origin is a limit point of

zeros of BP yn(Z), as well as Nasif's [81] bound [e~n)] ~ {(n_l)/(2n_l)}i/2 and

McCarthy's [74] generalizations of some of these results are all consequences of a

result of DoPey [48], that we shall prove directly in the slightly more general
form of

TIIEOREM 6. For any complex a are positive b the zeros a ~n) (a,b) of yn(Z;a,b)

satisfy the inequality

 82

b
la n ) ( a ' b ) [ ~ n - 1 + Re a

COROLM~RY 1. F o r a = b = 2, [~n)[ ~ 2/(n+1).

We o b s e r v e that this result is sharp, because for n = 1 it holds with the equal sign.
The u p p e r b o u n d o f T h e o r e m 6 may be c o m b i n e d , w i t h t h e lower bound of Theorem 2'
and t h e r e s t r i c t i o n of Theorem 5 in order to l e a d u s c l o s e to what are essentially
the best presently known s t a t e m e n t s concerning the locations of the zeros of yn(Z;a)

and @n(Z;a), respectively, that are valid for all n. See, however, besides [85] a l s o

[98], [46], and [84], the latter with better results, but valid, only asymptotically,
f o r n ~ ~.

THEOREM 7. For a > 2, all zeros ak(a ) of yn(Z;a) belon~ vo the semi-~nulus d e ~ n e d

by the inequalities:

2 i 2
n(n+a-1) <- '~k ( a ) [ <- n + a - 1 ' Re ~k < 0.

The zeros Bk(a) of @n(Z;a) similarly satisfy

(n+a-1)/2 ~ [Sk(a)[ ~ n(n+a-1)/2, Re Bk < 0.

The inequalities for ]ak(a)[ and [~k(a)[ arc sharp and become equalities for n = i.

COROLLARY 2. The zeros ~k of yn(Z) and the zeros ~k of 0n(Z), belong to vhe s~mi-
annuli

2
n ( n2+ l ) <- [akl < ~ , Re ak < 0

and

(n+l)/2 5 Igkl 5 n(n+l)/2, Re Sk < 0,

respectively.

Theorem 7 and its corollary are immediate consequences of previously quoted

results, of which only Theorem 6 remains to be proved. The proof relies heavily on
Theorem E, due to Laguerre. Up to now most theorems not directly connected with BP
and whose proof requires more than just a few lines have only been stated for ease
of reference and with an indication where a proof may be found. It is, however, not
entirely trivial to formulate Laguerre's theorem (first published in 1882; see [4~])
in the form here needed. In fact, the clearest presentation that the author could
find is in a sequence of problems in P~lya and Szeg~'s "Aufgaben und Lehrsatze aus
der ~lalyse" ([49], vol. 2, Problems 102-118). For these reasons the proof of
Laguerre's theorem will be sketched here.
 83

9. I]IEOREM E. (Laguerre [41]). Let f(x) be a polynomial of degree n and set

X(x) = x-2(n-l)f'(x)/f~'(x). Let x 0 be a simple zero of f(x) arZ consider a circle

C (possible a straighv line) ~hro~h the point represenved by the complex number
Xo, such that no zero of f(x) belongs to one of ~he two open, cira~iar regions de-

term.ined by C. Then X (Xo) belo~sz to ~he other region.

Proof of Theorem E. Let Zl,Z 2 .... ,z n be complex numbers with the center of mass
n
= (j =i
~ z j)/n. We generalize this concept and shall say that ~ = ~ as just defined

is the center of mass of the points z. (j = l,...,n) with respect to the "pole"
J
z 0 = . In order to define the center of mass ~z0 of the zj 's with respect to an

arbitrary pole z 0, different from all the points zj, we proceed as follows:

First, we map z 0 into the point at infinity, by a linear fractional transformation,

say z' = a + b (a,b arbitrary complex numbers). Under this map zj is sent into
z-z 0
n
z!] = -zJz 0 a + b. We now find ~' = g~' = (j=l
~ zj)/n as before and then map ~' back.

The inverse of ~' under our mapping will be denoted by ~ = ~_ and is the desired
~0
generalized center of mass of the zj.'s with respect to z 0. We now show that ~z0

depends only on z 0 and the zj's and is independent of the specific auxiliary map

selected. Indeed, ~ z~ = a ~ (zj-z0) -i + nb, so that ~ = ~a ~ 1 0

z -z + b. How-
j3 j j j

1 1 _ 1 , and
ever, also ~'~ uz 0~ + b holds, so that ~ ~. -zJz O ~Uo~z-Z
J

-I -l
~z0 = z 0 + n/~j (zj-z0) = z0-n/~j (z0-z j)

is indeed independent of a and b.

n F' 1
Next, let F(z) = j=iI
] (z-zj) be a poly~lomial. Then ~ (z) = ~ -z-zj
- , so that

the center ~z of the zeros with respect to an arbitrary pole z is given by

(6) ~ = z-nF(z)/F'(z).
z

From the Gauss-Lucas theorem we know that the (ordinary) center of mass ~= of the

zeros of F(z) is inside the convex closure of those zeros and that this is the
smallest convex set containing all z~'s. Each side of the polygon is a straight line
 84

through two zeros. It may be considered also as a circle through any two zeros and
the pole z = ~. It follows that ~ is inside the corresponding polygon of the

(zi)'s. When we map back, the sides of that polygon become circles through any two

of the z.'s and the pole z. Of the two circular domains determined by such a circle,
J
it may happen that one contains no other zeros of F(z). Let us "delete" it then
mentally and keep only the one that contains all other zeros. The intersection of
all these remaining circular regions (the image of the intersections of the corres-
ponding half planes through the (zi)'s, taken two by two) yields a curvilinear

polygon, say Cz, that contains 5z' the image of ~ .

Claim I. If the set of zeros z. (j = 1,2,...,n) belongs to any circular domain D

J
and if z is outside D, then C is in D.
z

Proof of Claim I. The smallest convex polygon containing all the (z~)'s is inside

any circle D', containing all the (zj')'s (because that polygon is the smallest con-
vex set containing all (z~)'s and a circle is a convex set), while z' = ~ is outside

D'. When we map back, set inclusions are preserved and Claim 1 is proved.

Claim 2. Let x be a simple zero of F(z). The center of mass of the remaining zeros
with respect to x is

(7) x = x-E(n-1)F' (x)/F"Cx).

Proof of Claim 2. Let FCz) = (z-x)f(z); then F'(z) = f(z) + (z-x)f'(z), F"(z) =
2f'(z) + (z-x) f"(z), so that F'(z) = f(x) and F"(x) = 2f(x).
We now apply (6) to f(z) (whose degree is n-i with respect to x) and obtain
X = x-(n-l)f(x)/f'(x). If we replace f(x) and f'(x) by F'(x) and F"(x)/2, respective
ly, we obtain (7) and Claim 2 is proved.

Let D 1 be a circle through x that contains all other zeros of F(z). As x is a

simple zero, one can deform D 1 into a circle D that leaves x outside, but still con-

tains all other zeros. Let C be the curvilinear polygon of these other zeros of
x
F(z), with respect to x. We know from Claim 1 that X = X(X) belongs to C and
x
Cx~D because x ~ D. This finishes the proof of Laguerre's Theorem E.

i0. From Theorem E i~nediately follows the

COROLLARY 3. If x 0 is one of the zeros of largest modulus of a polynomial f(x), then

 85

Proof. No z e r o belongs to the open circular domain ]z I > [x0l; hence,

IX(x0) ] > ]x01 is not possible and the conclusion follows.

LEb~ i. If the polynomial f(z) satisfies a linear differentia~ equation of second

order of the form,

P(z)y" + Q(z)y' + R ( z ) y = O,

then, ybr each zero z0 of f(z),

f ' ( z O) P ( z O)
(8)
E ' ( z O) q ( z O) "

Proof. This follows trivially from f(z0) = y(z 0) = 0.

LEMb~ 2. Let a be a zero of yn(Z;a,b); then

2(n-l)~ 2
X(~) = c~ + as + b

Proof. By (2.26), yn(Z;a,b) satisfies a linear differential equation as described

2
in Lemma i, with P(z) = z and Q(z) = az+b, so that the result follows from (7) and
(8).

LEbtMA 3. If a is one of the zeros of iar;]est modulus of yn(Z;a,b) then

2(n-l)
ll ÷ D ~ g - 51.

Proof. By C o r o l l a r y 3, i f a is a z e r o o f maximum m o d u l u s , then Ix(~)l ~ I~1 and

the result follows by Lemma 2.

-1
P r o o f o f T h e o r e m 6. Let v = ~ , where a is one o f t h e z e r o s o f l a r g e s t modulus of
yn(Z;a,b). T h e n , by Lemma 3, ll+2(n-1)/(a+bv)l 5 1. We s t a r t by d e t e r m i n i n g the

locus of the complex variable v, f o r w h i c h we o b t a i n equality. It is given by

2(n-l) - e i~ 0 < * < 27

(9) 1 + a+bv ' - '

or, solving for v, by

-i
v = -b (a+n-l+i(n-l)cotg ~/2)

= -b-l{Re a+n-l+i ((n-l)cotg ¢/2 + Im a)}.

As # varies from 0 to 27, v describes the straight line L parallel to the imaginary
axis, of abscissa -(n+Re a-l)/b. Consequently, the locus of v -I for which one obtains
the equality (9), is the inversion of L in the unit circle, q'his is the circle
through the origin, with center on the real axis and passing through -b/(n+Re a-l).
Its equation is
 86

b b
Iiz + 2(n+P,e a - l ) ] = 2(n+Re a - l )

The required inequality for a corresponds to points inside the circle. In particular,

[a[ < b / ( n + R e a - i )

and Theorem 6 ( h e n c e , a l s o Theorem 7) i s p r o v e d .

REbLCRK. Do~ev considers only the case b = -i and denotes a-2 by m.

ii. In 1976 appeared a paper by Saff and Varga [98] with an important improvement of
DeWey's result. It became available too late for a complete treatment here; there-
fore, only a brief outline will be given, although it leads to the strongest pre-
sently known results concerning the location of the zeros of yn(Z;a) and @n(Z;a).
The method is based on the following

THEOREM F. Let {Pn (z)}kn=O be a sequence of. ~olynomials of re~rective~ degrees n,

which satisfy the three-tez~ recurrence relation

Z Z
(10) p n ( Z ) = (~--- + 1 ) P n _ l ( Z ) - ~ - P n _ 2 ( z ) (n = 1,2 . . . . . k),
n n

where the b n's and cn's are .~;ositive real nwnbers anJ uhere p_l(Z) = 0, pO(z) =

PO ~ O. Set a = min bn(l-bn_iCn I) (b 0 = 0). If ~ > O, then vhe parabolic

i <n<k
region

P = (z = x+iy E Cly 2 < 4~(x+a) x > -~}

contains no zeros of any of the polynomials pn(Z), n = 1,2 ..... k.

Sketch of Proof. Let z 0 c P , such that none of the pn(Z) vanishes at z 0 and set

~n(Z) = ZPn_l(Z)/bnPn(Z ) (n = 1,2 ..... k). First we verify that ~n = ~n(Z0 ) has

Re ~n E 1 (n = 1,2,...,k). This is clear for ~I = z0/(zo+bl)' because in P ,

Re z ~ -bl; and for 1 < n < k that property follows by induction on k, by use of (I0)

and the definition of P~. Next, we observe that Pn(0) ~ 0 for all n = 1,2,...,k.

Indeed, by (i0), Pn(0) = Pn_l(0); hence, by induction, for every n, Pn(O) = P0(0) =

PO ~ 0. It follows that if pn(Z0) = 0, then z 0 ~ 0. Assume now that pn(Z0) =

Pn_l(Z0) = O; then, by (I0) and z 0 ~ O, also Pn_2(z0) = 0 and, by induction

Pn_j(z0) = 0 for all j < n, and this if false for j = n. Consequently, pn(Z) and

Pn_l(Z) have no common zeros. Let now, contrary to the statement of Theorem F,

pn(Zo) = 0 for some n in 1 <_ n < k and z0 ~ Pa. By (10),Pl(Z) = (z+bl)P0/bl, so

 87

that z = -b I is the (only) zero of pl(z). By the definition of ~, -b I ~ - ~, so that

-b ~ P ; hence, 2 5 n < k. Again by (i0),

_ z0+b n z 0 Pn_2(z0 )
(Zobnl+l)Pn_l(Z0) = Z0cnlpn_2(z0) , or bn - Cn Pn_l(zO) ;

this is meaningful, because pn(Z0) = 0 implies, as seen, that Pn_l(Z0) # 0. This

cn Z0Pn_2(z 0)
shows that - - (z0+bn) - = Un_1(z0) and, by z 0 ~ P , Re ~n l(Z0 ) < I.
bn_ibn bn_iPn_l(Z 0) - -

C c C
Hence, Re n n n , whence
bn_ibn (z0+bn) <_ i, or --Rebn_ibn z 0 <_ i- --bn_l

Re z 0 j -bn(l-bn_ic~l) ~ -~ . This, however, shows that z 0 could not have been in

P , where for all z, Re z > - ~. Theorem F is proved.

n
Consider now the polynomials @n(Z;a) = ~ d(n) zn-m, with d (n) = n:(n+m÷a-2)(m)
m=0 m m m:(n_m)!2 m '

as in (2.27). By comparing the coefficients of zn-m(m = 0,1,...,n) on both sides,

one verifies that

(ii) @n(Z;a) = (z+n-l+a/2)@n_l(z;a+l) - ~1 (n-l)z On_2 (z; a+2).

For each polynomial in (Ii), the sum of the degree and the a-entry is the same,
say n+a = s. With a = s-n, (ii) becomes

@n(Z;s-n) = (z+(s+n°2)/2)0n_l(z;s-(n-l)) - ~1 (n-l)z @n_2 (z;s-(n-2))"

If we set b n = (s+n-2)/2 and divide by bn, we obtain

(12) 2(s+n-2)-iOn(Z;s-n) = (zbnl+l)0n_l(Z;s-(n-l))-(n-l)(s+n-2)-iZ@n_2(z;s-[n-2)).

Equation (12) is almost of the required form (10),except for the factor
2/(s+n-2) of @n(Z;s-n). To eliminate this discrepancy, we multiply (12) by

2n-I/(s+n_3)(n-2) and obtain

2n
(s+n-2)(s+n-3)...s @n (z's-n) =

2n-I ,)n-2
Z Z -
(~--+I) (s+n-3)...s @n-I (z;s-(n-l)) - c (s-~n-4)...s @n-2 (z;s-(n-2))'
n n

where we have set c = (s+n-2)(s+n-3)/2(n-l). This is precisely of the form (i0),

n

2n
0n(Z;s-n ) and bn, c n as defined. Also,
=
with pn(Z) (s+n-2)(n°l)
 88

bn(1-bn_lCn 1) = ( s - l ) / 2 i n d e p e n d e n t l y o f n, so t h a t a = ( s - l ) / 2 . By Theorem F, none

o f the p o l y n o m i a l s pn(Z), hence none o f the polynomials @n(Z;s-n), can have zeros

i n s i d e the p a r a b o l a y2 5 2 ( s - 1 ) ( x + ( s - 1 ) / 2 ) = (s-1)(2x+s-1). This may be w r i t t e n as

Izl 2 = x2+y 2 ~ x2+2(s-l)x÷(s-l) 2 = (x+s-l) 2 ) or Iz[ < Re z ÷ s-l. In polar coordinates

s-i @ _ 1-cos
r 5 r cos @+s-l, or r < l-cos n+a-i O " Consequently, for all zeros g~n)(a)

n+a-i @ ' and therefore the zeros

(k = 1,2 ..... n-l) of @n(Z;a) we have ]B n)(a)l > 1-cos

e~n)(a) of yn(Z;a) are inside the cardioid I ~ n)(a)] < (1-cos @)/(n+a-l). The bound

(1-cos @)/(n+a-l) is clearly better than 2/(n+a-l) obtained in Theorem 7. By re-

calling another result of this chapter, we may state here an improved version of
Theorem 7, as

COROLLARY 4. For r e a l a > -n+l and b > O, t h e z e r o s ~ n ) ( a , b ) of yn(Z;a,b) satisfy

the inequalities

b ~n) 1-cos @ b;
n(n+a-l) f [a (a,b)[ S n+a-i

~(n)(a,b) of @n(Z;a,b) satisy~4

and the zeros ~k

n+a-1 ~n) n(n+a-1)

(1-cos @)b 5 18 (a'b) l ~ b

Under certain conditions we know that Re ~n)(a,b) < O. This happens when a = 2

(see Theorem 5), also for real a ~ 2, sufficiently large (see [74]) and, presumably,

for all real a. Whenever that is the case, also Re B~n)(a,b) < 0 and @ in above

inequalities may be restricted to I~-01 < ~/2.

12. During the years 1956-59, Parodi perfected an approach for the determination of
bounds for the characteristic values (eigenvalues) of matrices. It is based on the
following classical theorem of Gershgorin, a proof of which may be found in [45].

THEOREM G. Let A be an n × n matrix with complex coej~'~icients ajk. All character-

n
istic values of A lie in the union D 1 of the disks Ix-ajjl < ~ lajk I. They also
k=l
n k~j
lie in the union D 2 of the disks Ix-ajjl S ~ [ajkl; hence they lie in the intersea-
j=l
jCk
tion Dlf~ D 2.

In connection with the representation of BP by determinants (see Chapter 3),

Parodi's method leads to bounds for the zeros of yn(Z). These are weaker than those
 89

given by Theorem 7, but the method is simple, elegant, and may well be improved
further. For these reasons we state here

THEOREM 8. The zeros ~k of yn(Z) belong to the intersection of the regions D 1 and

D2, defined by D 1 = union of the circles Izl 5 2/3 and ll+zl 5 i; D 2 = union of the

circles I~l ~ 6/s ~nd ll+zl ~ 1/3.

Proof. Let M be the determinant defined in Chapter 3. Then the zeros of yn(Z) are
n
the same as those of M . On the other hand M = IA-Izl, where I is the n x n unit
n n
matrix and
1
0 0 ... 0 0 0
2n-i
1 1
0 ... 0 0 0
2n-3 2n-3
1
0 0 ... 0 0 0
2n-5
1
A= 0 0 0 0 0
2n-7 """

i 1
0 0 o ... y 0 -y

0 0 0 ... 0 1 - 1

It follows that the zeros of yn(Z) are precisely the characteristic values of A.

These are located, by Theorem G and looking at rows, in the union of the circles

[z[ < ~ 1 , Iz[ 2

f 2n-2k+l (k = 2,3, . ,n-l)
. . and
. Iz+l] < i. Most circles are con-

centric and their union is D I. In the same way, proceeding by columns, we obtain

D 2 and the Theorem is proved. It is obvious that it can be improved, by adding the

condition of Theorem 5 that Re ~ < 0. Parodi shows in the same paper [85], how pre-
vious result can be sharpened. If we multiply the next to last column of M n by l+z

and add to the last and expand the resulting determinant by the last line, we see
that, with an irrelevant (but easily found) non-vanishing constant C n, yn(Z) =
CnlB-Iz], where B is the (n-l) x (n-l) matrix
 90

1
0 0 ... 0 0 0
2n- 1

1 1
0 0 0 0
2n-3 2n-3 """

1
0 0 ... 0 0 0
2n-S
1
0 0 ... 0 0 0
2n- 7
B =

1
0 0 0 ... 0 --- 0
7
1 1
0 0 0 ... ~- 0 - g (l+z)

0 0 0 ... 0 31 31 -z (l+z)

By applying Gershgorin's Theorem G we now find that the zeros of [B-Izl, hence, those
of yn(Z), belong to the intersection of the regions Di and D½, where Di = union of

[z[ f 71 ' [z[ 5 g1 + g1 i i + z I , and 1 + z(l+z)[

[X < g1 ; D2, = u n i o n of lz[ f g1 + g1 and

@+ z(l+z) l s g 1 ,l + z [ • A sketch shows inmlediately, that D I / ' I Di i s much s m a l l e r

than and contained in DI/'~D 2. The result can be improved further, by removing the

portions of D I f ~ D 2, or of D~/~ D½ in the right half plane.

13. A more powerful method to improve Parodi's approach is due to Wragg and Under-
hill [113]. They use, in addition to Gershgorin's theorem also the following
Theorem H, due to Bendixson [3] and to Hirsch [33]. For a more recent presentation
and proof, see [45].

THEOREM H. Let A be an n × n matrix with complex entries, denote by A its complex

conjugate and by hj (j = 1,2 ..... v) is characteristic values. Set A = B + iC,

where B = ~1 (A+A*) and C = ~i (A-A*) are .qez~itian matrices and denote their respec-

tive characteristic values by ~j and vj, respectively (j = 1,2 ..... n). Then

rain ~j ~ Re h k < Max Vj and min vj < I m ~k < Max ~. for k = 1,2 ...,n.
j - j j - - j 3 '

TIIEOREM 9. If ~ = ~+in is a zero of yn(Z), t h e n - ~2 5 ~ < ( 2 n - 3 )2 ( 2 n - 1 ) < 0 and

In[ < 8/15.

Proof. Similar matrices have the same characteristic values. Let W = p-IAp, with A
as previously defined and P = llPij[I, an n × n matrix with Pij = 1 if i+j ~ n+l,

Pij = 0 otherwise. Then p-i = IIP~jll with p~j = 1 if i+j = n+l, p~j = -I if

i+j = n, p~j = 0 otherwise. One verifies that

 91

2 1
y y o 0 ... 0 0 0

1 2 1
0 ... 0 0 0

1 2 1
o ~ 3~ 7 "'" o o o
W=
oo.

-i -2 1
0 0 0
0 """ 2n-3 (2n-3) (2n-l) 2n-i
-i -i
0 0 0 0 ... 0
2n-I 2n-I

We now apply Theorem H, by writing W = B+iC, with B the diagonal matrix of the dia-
gonal elements of W, and

o -y o ... o o o

i i
0 --- •.. 0 0 0
Y 5

C = "

i -i
0 0 0 0
2n-5 2n-I
i
0 0 0 0 0
2n-i

The characteristic values of B are the diagonal entries themselves, so that, by

2 2
Theorem H, - ~ < ~ < - (2n-3)(2n-l) By Theorem G, the characteristic values of

1 1 1
C lie in the union of the circles Izl < ~ • Izl <- ~-~-3 + ~ (k = 3,4, "'" ,n) and

Izl _< ~ -1~ ; they are concentric and the largest corresponds to k = 3, whence the

theorem.

14. In 1952 and 1954 appeared two important papers of F.W.J. Olver (see [84] and
[46]). In them the author discusses expansions of functions that satisfy a certain
class of differential equations and studies their zeros. In particular, he puts
special emphasis on Bessel functions. Specifically, he determined, ([84], p. 353)
among others the distribution of zeros of the Hankel functions H(1)(z) and H(2)(z),

for integral and for half-integral ~. From these the zeros of K (z) follow immedi
ately because (see [i], 9.6.4).
 92

I
~i ~- i
= i ~ie H (1)(ze 2 )(=~ < arg z < ~/2)

K(z) = 1 ~i
i . - ~- ~ i --
= ~ ~le 1t(2) (ze 2 ) ( _ ~ < arg z < ~).

For half integral values of ~ = n ÷ 1/2, the zeros of K (z) are (see Chapter 2) of

course the as the that is the reciprocals of the The theory o f

Olver and even his precise conclusions cannot be presented here and the interested
reader is advised to consult Olver's original papers [84], [46]. A simplified
version of his results, however, is easy to state, lie shows that the zeros of

H(1)
n+ i/2 (ze~1 ) lie on a convex arc (see Fig. 15 on p. 352 of [84]), symmetric with

respect to the imaginary axis. For n ÷ =, the points of intersection with the real

axis approach -n and +n, while that with the imaginary axis approaches inz0, where

z 0 = t~02 -i ~ .66274 .... Here t O Z 1.19967... , is the (unique) positive root of

the equation coth t = t.

The zeros of Kn+ i/2(z), i.e. of @n(Z), are then located (see e.g. [I], p. 377

and Fig. 9.6) on an arc obtained by rotating the previous arc by n/2. Finally, the
zeros of yn(Z) are on an arc obtained from the preceding one by inversion in the
-i
unit circle. (see Fig. 1 ). Let v = z 0 ~ 1.50888...

i
K then the results of Olver have as a
corollary the following statements.

THEOREM i0. All the zeros ~ n ) of yn(Z)

are located on an arc symmetric with
respect to the real axis, that runs in
the left half-place. For n ~ ~, the inter-
section of that arc with the imaginary
axis approach ± i/n, while that with the
real axis approaches -v/n (see Fig. i).
The absolute largest zero of the polyno-
mials yn(Z) of odd degree is the real one,

say ~(n)(m = (n+l)/2) and a(n) =

m m
i
n -vn -i + 0 (n-2).

Figure 1
 93

To give an idea of the degree of approximation in Olver's results, we may consi-

der the following tabulation:
n] -v/n
~m
n [
I -1.50888.. -i
3 .5029... .4305...
5 - .3017... 2742.

While Olver's results contain an element of finality, they yield exact results
only in the limit, as n + =.
Numerical investigations have led to the following

CONJECTURE. For arbitrary a > O, b = 2 and odd n + ~ the real zero e (n) (a,2) of
' m
yn(Z;a,2) satisfies the asymptotic equality
a(nl
m ~,2)r,J-2(1.52548 n + a - (~+1)/~} -1.

This follows conjecturally from [71], vol. 2, p. 194, by taking into account
the differences of notation and normalization.

15. Burchnall's results (2), (2') turned out to be useful in some rather unexpected
applications (see, e.g. [62], [55]). In fact, it would be very desirable to have
simple explicit formulae for the sum of arbitrary powers of the zeros of 8n(Z), or
yn(Z). No such formulae, valid for all powers seem to be known.

A certain amoung of information about these sums, can, of course, be obtained

from Newton's classical formulae (see, e.g. [SO], p. 208). Let us denote the sums
of the r-th power of the zeros of a polynomial by o r and, if we want to put into

evidence that the polynomial is yn(Z), then we write a(n). For a general polynomial
n r
f[x)= ~ ckxn-k with c 0 = 1 Newton's identities are
k=0

r-i
c. o .+rc = 0 for r = 1,2,...,n;
j =0 ] r-] r

(13)
n
c. o = 0 for r > n.
j=O .1 r-j

In t h e c a s e o f y n ( Z ; a ) , we may t a k e c j = dn_(n)j/d n(n), w i t h t h e dk(n) g i v e n by ( 2 . 2 7 ) .

2n 22 2j
T h i s l e a d s t o c 1 = 2n+a-2 ' c2 = (2) (2n+a-2)(2n+a-3) ..... cj = (~.) (2n+a-2) ( j ) .....
2n = 2n (n+a-2) !
Cn = ( 2 n + a - 2 ) ( n ) (2n+a-2)! " By (13) (assuming n large) we o b t a i n o 1 = - c 1 =

-2n = = 4n(n+a-2)
2n+a-2 ' °2 - a l C l - 2 C 2 = c21 - 2c2
(2n+a- 2) 2 (2n+a-3)
 94

o3 = -8(a-2)(n+a-2)n , etc. It is immediately apparent that we obtain

(2n+a-2)3(2n+a-3) (2n+a-~)
significant simplifications, if we restrict ourselves to the case a = 2, of the

2j
simple BP. In that case cj = (~) - - .) and previous formulae reduce to o I = -i,
(2n) (3
1 -i 2
o2 = ~ , o 3 = 0; also, ~4 = ' °5 = 0, o 6 = ,
(2n-l) 2 (2n-3) (2n-l) 3 (2n- 3) (2n-5)
10n-17
o 7 = 0, a 8 = - , o 9 = 0, etc.
(2n-l)4(2n-S)2(2n-5)(2n-7)

The values obtained for odd subscripts are, of course, those predicted by
Burchnall's theorem (2).
The sums of negative powers of the zeros ~ n ) of yn(Z) are the same as the sums

of the positive powers of the zeros ~n) of Gn(Z) and conversely. In order to avoid

confusion in what follows we shall denote by ~r = °(n) the sum of the r-th powers
' r

(r positive, zero, or negative) of the zeros a~ n) of the simple BP yn(Z). The sums
o_r of negative powers can of course be computed as sums of positive powers of the

zeros 8~n) of 8n(Z), by use of Newton's formulae (13). The coefficients cj in (13)

are now precisely the a.'s of (2.8).

3
<n+l)n (n+2) (4) (n+l) n
With a 0 = i, a I 2 ' a2 = ~ ,... we obtain O_l = -a I = - - 2 '
2"2!
2 (n+l)n 1 n (n+l) {n (n+l)-6},
o2 = -a°-l'2a2 = al-2a2 2 ' o-3 =

°-4 = - ~i n(n+l){n(n+l)-3}, 1 n (n+l) {n2 (n+ i) 2-28n (n+l) +60 } etc.

o_S = - --16

It turns out that we obtain somewhat simpler results if we set n(n+l)/2 = N,

say. 1 N(N+I) , o_3 = ~1 N(N-3),

Then o_i = -N, o_2 = ~- o_4 = -N(2N-3),

o_5 = _ ~1 N(N2_I4N+IS) , etc •

In this way we are able to compute effectively the sums of an}' given power
(positive or negative) of the zeros of the BP yn(Z); however, no obvious pattern

seems to emerge and an explicit formula for o (n) and/or o (n) does not seem to be
r -r
known.

A more convenient way to approach this problem is through a recursion formula

(see [62]), valid at least for positive values of r. While of a rather complicated
appearance, it permits one to go beyond Burchnall's (2), for which incidentally it
gives a new proof.
 95

We start by observing that, on account of Theorem 1 all zeros of BP are simple.

Hence,
On_l (z) n A .
_ _ = ~ nj with Anj = lim (z-Bj) @n-l(Z) @n-l(BJ)
On(Z) j=l z-Bj' z + flj @n (z) @n(~j )

By (3.8) the last fraction equals -B -I , so that

J
(14) -en_l(Z
- = ) n~ 1
On(Z) j=l gj(Bj-z) "

The corresponding formula for yn(Z) is easily obtained from (14), by replacing On(Z)

by znyn(i/z), 8j by ~i I and then simplifying the result. We obtain

2
Yn_l(Z) n ~.
Z-~.
Yn (z) J=] 3

Returning to (14), we observe that it can be written as

en_l(Z ) n n 8i2 r l o ( z~r zr n

j=l 8~(1-z/Bj) j=l = q r=O j = l Bj r=O °r+2Z "

On the other hand, by (3.5) (with n replaced by n-l),

On(Z) _ 2n-l+z 2 On-2 (z)

On_l(z ) @n_l(Z ) or

en-l(Z) I i [ (_l)m z" en-2(z) m

(16) @n(Z ) = 2n_l+Z2On_2(z)/O n_l(z) - 2n-~ m= 0 (2nXl @n_l(Z)) "

We now equate the right hand sides of (15) and of (16) and then replace the ratios
@n_2(z)/@n_l(Z) with the help of (15). We obtain
~ 2m
r (n) = z ~l)}m
I ~ ~r+2 I (-l) m { I zk~
r=O m=O (2n-l) 2m+l k=O

2n_l + [ (I zko -
m=l (2n-l) m+l k=O

I + tr Zr (-i) m (n-l) (n-l) _(n-l)

- 2n-i
r=l 2m÷kl+...+km=r (2n-l) m+l °kl+2 °k2+2 ""Ukm+2 •
m >- 1 , k .j >- 0

1
By comparing coefficients, we obtain first o~ n) = 2n-I ' then, in general, for r > 0,
 96

(-i) m (n-l) _(n-l)

(17) o (n) =
r+2 (2n_1) m+l ° k i + 2 " " ° k m + 2 •
m
2m+ [ k.=r
j=l j
m ~ i, k.j >- 0

Formula (17) p e r m i t s one, in principle, t o compute t h e sums o (n) r e c u r s i v e l y , if one

m

a l r e a d y knows t h e sums o ( n - 1 ) . In f a c t , (17) iooks rather forbidding and one may

m
wonder w h e t h e r i t can be p u t t o any u s e . Fortunately, the answer is positive, al-
though (so far, at least) (17) c o u l d n o t be made t o y i e l d explicit formulae for all

o(n)
m

To s t a r t w i t h , we may u s e (17) in order to obtain a new p r o o f o f ( 2 ) , (2')• We

observe that f o r n = i and n = 2 we have

eo(Z) 1 ~ (_l)rz r o ( 1 ) zr
Ol(Z) l+z r=0 = r=0 r+2

and

-
@i (z)
- =
l+z =
1
- -
l+z =
l 1
82(z) 3+3z+z 2 3 l+z+z2/3 3 l+z2/3(l+z)

2 4
i z 3 z 2 3 .2
= -3- {i- ~ (l-z+z2-z +...) + ~-- (l-z+z -z +.. ) - ...)

1 z2 z3 2 4 1 5
= y- 7+ ~ - - 27 z + ~ - z + ... = ~ 0 (2) z r
r+2 "
r=0

We note that o ~2) = 0; this is the instance n = 2 of the claim that o (n) = 0 holds
2~+i
(1)
for ~ = 1,2,...,n-l. (For n = 1 this is also true - vacuously.) Next o 3 =

o i)= -I and o~ 2)- ~ , o 27 , so that

(183 (n) (-I) n o{n) (-I) n

°2n+1 = { 1 . 3 . . . . . ( 2 n _ i ) } 2 ' 2n+3 = { 1 . 3 . , . ( 2 n _ l ) } 2 ( 2 n _ 1 )

both hold for n = i and n = 2. Let us assume that o,k~f

~ = 0 has already been veri-
2~+i
fied for all k with 2 < k < n-i and all ~, I ~ v < k-l. Then (17) shows that if

r+2 = 2~+i is odd, also o (n) = o (n) = 0 for ~ = 1,2,...,n-i Indeed, let k = n-l;
r+2 2v+l
if r+2 is odd, then at least one of the k.'s is odd. Also, all odd k.'s satisfy
3 3
1 < k.] <- r-2m = 2~-i-2m <_ 2~-3 ~ 2n-5, for ~ < n-l._
In particular, we find in each

product at least one factor o (n-l) with 3 < k + 2 < 2n3= 2On l) 1
• k.+2 - 3 '
1 Km j
and this vanishes by the induction hypothesis. It follows that all summands in (17)
 97

vanish and a,n,l~ = 0 for all 9 = 1,2, ,n-i in agreement with (2) and that is what
29+1 "'"
we wanted to prove.
If r+2 = 2n+l, so that r = 2n-l, then one of the summands in (17) has m = 1
and k I = r-2 = 2n-3 (no other k.'s occur) and does not vanish. Indeed, (17) now
3
_(n-l)
U
2n-i _(n-l) _(n-i)
reduces to o -n" = - The numerator is U2n_l = O2(n_l)+l and we have
2n+l (2n_i)2"

verified that (18) holds at least for the superscripts 1 and 2; if we assume that
the first of (18) has already been recognized as valid up to the superscript n-l,

then s(n) _ i (-i) n-I = (-i) n

2n+l = (2n-l) 2 {1"3"''(2n-3)} 2 {1"3'''(2n-1)} 2 and (18) h o l d s a l s o

for superscripts n , and t h i s concludes its p r o o f by i n d u c t i o n .

I f r+2 = 2n+3, t h e n t h e r e a r e t h r e e p o t e n t i a l l y non-vanishing terms in (17),
corresponding to m = i, k I = 2n-l; m = 2, k I = 2n-3, k 2 = 0; and m = 2, k I = 0,

k 2 = 2n-3. Hence, assuming that the second of (18) has been verified for ~(k)
2k+3

and k up to n-l, (17) and the induction hypothesis yield

_(n-i) (n-l) (n-l)
~(n) = - U2n+l 2 O2 °2n-I = (-l)n
2n+3 (2n_i)2 + (2n_i)3 {l.3...(2n_3)}2(2n_3)(2n_l)2

i (-I) n-i i !:1)_~

n (i__ 2
2"
2n-3 {1.3...(2n_3)}2 (2n_i)3 {i.3...(2n_i)}2 2n-3 (2n-i)(Zn-3)q

= (-i) n

{l.3...(2n-l)}2(2n-l)

and this finishes the proof by induction of (18).

One may, of course, go on. However, the number of non-vanishing terms in (17)
increases fast and the formulae become rather complicated.
We shall finish this section with the remark that, although we do not know

explicit formulae for a (n) , valid for all integral values of r and n, we still can
r

at least prove the following

THEOREM ii. For n = i, a (I) = (-i) r. For n = 2, c (2) = 3-[m/2]h and a (2) =
r m m -m
3[(m+l)/2]hm , where h m depends only on the residue class of m .~dulo 12, as fo~low~:

m 0 1 2 3 4 5 6 7 8 9 I0 ii modulo 12

h 2 -i 1 0 -i 1 -2 1 -I 0 1 -i
m

Proof. For n = i, the result is obvious. For n = 2 we obtain by Newton's first

 98

formula (13), that G 0 = 2, o I = -i, o 2 = 1/3. Next, by the second formula (13), we

verify that also ~3 = 0, o 4 = -1/9,..., Oli = -3 -5 , conform to the statement of the

theorem. In general, by (13),

1
(19) om = -°m-I - ~ °m-2"

We assume that for all values of k _< m-i we have already verified that Theorem ii
holds. Then we check b y (19) that also o takes on the value prescribed by the
in
theorem. This requires 12 checks, according to the residue class of m (mod 12). We
may reduce these to 6 checks, by observing that hm+ 6 = -hm. As an example, wo work

out one of these checks, say for m - 5 (mod 12). Then m-I - 4 and m-2 - 3 and, by
the induction hypothesis and (13),

o m = -°m-i - ~1 °m-2 = _3[(m-l)/2]h 4 _ ~1 . 3- [ (m - 2)/2]h 3 = 3-(m-2)/2 = 3-[m/2]

and this proves that Theorem ii holds also for m, provided that m - 5 (mod 12), and
in fact, also for m ~ ii (mod 12).

The assertion concerning o (2) could be proved in the same way, by using
-m

Newton's formulae for 82(z) = z2+3z+3. It is much easier, however, to observe that

-m -m ~ m m = 3m-[m/2]h = 3[(m+l)/2]hm , as claimed.

a-m = el + ~2 = (~ + ~2)/(~i~2) = 3m°m m

No similarly simple results seem to hold for o<n~6

~ , even when n = 3.
m
 CHAPTER ii
ON THE ALGEBRAIC IRREDUCIBILITY OF THE BP

i. Much of the present chapter covers material found in [53]. However, in [53],
there are many errors of detail and the presentation leaves much to be desired. In
the present chapter many procfs have been reworked completely, in order to make the
treatment not only correct, but also readable.
The term irreducible will mean here irreducible over the field of rational num-
bers. This is equivalent to the statement that if f,g and h are polynomials with
rational, integral coefficients, then f = gh is possible only if either g, or h re-
duce to a constant polynomial, in fact to a rational integer, and the other factor
has the same degree as f. The polynomials here considered will, in fact, be monic.
It is extremely likely that all BP are irreducible, but no proof of this general
statement exists. What we can prove are the four theorems of the present chapter.
The symbol p, with or without subscripts, stands for "an odd prime, while q and m are
positive inters. Also, for given n, we set k I = min (n-Pl), k 2 = min (P2-n-1),
and k = k(n) = rain (kl,k2). Pl -< n P2 > n

THEOREM i. (a) The BP of degrees n = pm are irreducible.

(b) The BP of degrees n = qp-l, with q < p/2 are either irreducible, or else can
~ v e only irreducible factors of degrees rp or rp-i (r = 1,2 ..... q). If p-i > k(n),
then the BP is irreducible. In particular, the BP of degrees n = p-I are irreducible.
(c) The BP of degrees n'= qp+l, with q < p/2 are either irreducible, or else can
have only irreducible factors of degrees rp, or rp+l (r = 0,i ..... q). if also
p > k(n) and q is odd, then the BP is irreducible. In particular, the BP of degrees
n = p+l are irreducible.
(d) The BP of degrees n = qp with q < p/2 can have irreducible factors only of de-
grees that are multiples of p. If also either p > k(n), or q <_ 17, then the BP is
irreducible.
(e) The BP of degrees n = pk-i can only have irreducible factors with degrees that
are multiples of p-l. If also p-i > k(n), holds, then the BP is irreducible.
(f) If p is the largest prime factor of n, or of n+l, then the BP of degree n can-
not have any factors of degree less than p-l.

THEOREM 2. For every integer n, the BP of degree n contains an irreducible factor

of degree Ann, with l im A n = I. For every~ n, A n > 16/17 and no BP can have an
n->oa

irreducible factor of degree d, with n/17 < d < (16/l?)n.

THEOREM 3. All BP of degree n ~ 4QQ are irreducible.

There is no doubt that with more work the bound n ~ 400 could be increased.
In fact, there are only four values of n, 301 < n < 400 for which a separate verifi-
cation was needed. Also the fraction 1/17 that occurs in Theorem 2 could be
 100

decreased and Theorem i could be improved by the addition of other types of n, for
which irreducibility can be guaranteed. There does not seem, however, to be much
point to such a task, because in any case it would fall short of the proof of the
already mentioned

CONJECTURE i. All BP are irreducible.

2. The proofs of these theorems rely heavily on Theorem A, due to Dumas [lZ], to be
stated presently. It is based on the theory of the Newton polygon of a polynomial
f(x), with rational, integral coefficients, with respect to a fixed, rational prime
p (see [~Z] and [aS]). This frequently occurring sentence will be abbreviated to
read "N.p. of f(x), w.r. to p", or simply "N.p."

n e
m n-m
DEFINITION. Let f(x) = ~ amP x , am,e m rational integers, p~a m, and consider
m=0

the points Pm = [m'em]" For want of a better name we call these particular lattice

p o i n t s o f the plane the sgots o f f ( x ) . The N.p. of f(x) w.r. to p i s the unique
open polygonal line w~th vertices only at the spots of f(x), convex dowr~)ards and
with no spots below its sides (see Fig. i).

n e
m n-m
THEOREM A. (Dumas [~2]). Let f(x) = ~ amp x , P~am; let h r are v r be the
m=O

horizontal and vertical projections, respectively, of the r-th side of the N.p. of
f(x) w.r. to p and let their greatest common divisor be t r = (hr,Vr). Then, if M

i8 the number of sides of the N.p. and h r = trSr, all factors of f(x) have degrees
M
of the form ~ ~rSr, with integers ~r' 0 < ~r < t
r=l - - r"

RE~RK. If v r = O, then t r = h r and sr = i.

COROLLARY i. ~ e 0 = 0, em ~ men/n for m = 1,2 ..... n, then M = i and all factors of

f(x) have degrees of the form un/t, t = (n,en), 1 < ~ < t. If, furthermore,

(n,e n) = i, then ~ = t = 1 and f(x) is irreducible.

Proof of the Corollary follows immediately from Theorem A.

In what follows it will be convenient to refer (improperlyl) to the horizontal
projection h r, as the length of the r-th side of N.p. We shall not give here a

proof of Theorem A. Indeed, an excellent presentation can be found in [59] (see

also [It], or [60]), and none of the available proofs is really easy.
We use square brackets rather than parentheses, in order to avoid confusion with
the notation (m,em) used for the greatest common divisor of the integers m,e m.
 101

e (¢m)

f
, I

I
I
,i J I

1 2 3 4 5 7 8 9 i0 ii 12 13 14 15 m

Newton Polygon of the polynomial zls(x) = zlS+S(clzl4+c2zl3+c3zl2+c4zll+cszl0)

7 6..3 S .4. 4 3 2 , 5[cj
+52(c6z 9 +cvzS+csZ, +CgZ )+b Cl0Z +b [CllZ +el2 z +CI3Z +C14Z+C15)

(j = 1,2 ..... 15), modulo p = 5.

Figure i.
 102

Sometimes a particularly simple case of Theorem A is sufficient for the purpose

on hand. This is known as Eisenstein's criterion of irreducibility.
n
n n-m
THEOREM B. (Eisenstein; see [59]). Let f(x) = x + p ~ amX , with P~an;
m= i
then f(x) is irreducible.

Proof. Here e 0 : 0, e n : 1, so t h a t t h e N.p. of f(x) reduces to a single line of

l e n g t h n. As a l s o e m >
_ 1 for 1 < m < n and ( n , e n ) = (n,1) = 1, t h e C o r o l l a r y to

Theorem A a p p l i e s and f ( x ) is irreducible•

To c o n c l u d e t h i s section, we s t a t e also, for ease of reference, another theorem

that will be n e e d e d l a t e r . I t i s w e l l known ( s e e , e.g. [4~]) t h a t t o e v e r y E > 0,
there exists an N = N ( E ) , such that, for x > N(e), there is at least one p r i m e p i n
the interval x < p ~ (l+~)x. Unfortunately, the explicit function N = N(e) i s n o t
known. Bertrand's postulate, p r o v e d by T c h e b y c h e f f , states that N(1) = 1 a n d , after
further improvements of this result by I . Schur [51],R. Breusch [8] p r o v e d

THEOREM C. (R. B r e u s c h ) . N(I/8) = 48.

3. Let us denote by e(m) (= ep(m); we shall suppress the subscript) the exact power

of p that divides m.
For any integer n, it is clear (see, e.g. [24]) that

(i) e(n!) : [~] + [~'] + .... [~-] + ...

P P
where [x] stands for the greatest integer not in excess of x; there should be no
danger of confusion with Pm = [m'em]' where there are two arguments enclosed in

square brackets. It is sufficient to stop in (i) with the term r = [log

Llog n]
pj, because

all other terms vanish. For n = pm, (i) immediately yields

=n-1
(2) e((pm) !) p-i "

k
More generally, let n = ~ ajp J, 0 < a• < p-i be the p-adic representation of n;
j=0 -J-
then we obtain by direct substitution in (i),
e(n!) = + akpk-1 + a k _ l pk-2 + ...
+ a2P + a I
k-2 k-3
+ akP + ak_iP + ... + a 2
÷ ° ....

+ akP + ak_ 1

+ ak
 103

= ak (pk-1 + p k-2 +...+ p+l) + ak _ 1 (pk-2 + p k-3 + .... + 1)÷...+ a2(P+l) + a1

pk_l_i 2_ 1
= ak + ak-1 p-1 + • "" + al P p-i

k k
1
p-i (j!O ajp3 - j=O[ aj}.

This completes the proof of the following lemma, due to Legendre:

LF~IA i. e(n:) = n-o(n) where a(n) ( = Gp(n)) = ~ aj, s(0) = 0.

p-i • j =0

From Lemma i immediately follows

LEMMA 2. Set

=
(n+m) !
(3) Cm m'• (n-m) ! '

m+a (m) +o (n-m) -o (n+m)

then e(Cm) = p-i

Proof. By Lemma i, e(Cm) = ~ {n+m-~(n+m)-m+e(m)-(n-m)+o(n-m)} and Lemma 2 is proved

4. It is clear that Yn(X) and On(X) are either both reducible, or both irreducible•

In fact, there is an obvious one-to-one correspondence between their respective

irreducible factors. It is somewhat more convenient to apply previous criteria of
irreducibility to monic pol}momials and, therefore, we shall be concerned here with
@niX), rather than Yn(X). We obtain a slight additional simplification, if we ob-

n
2-n (n+m) ! n-m
serve that On(X/2 ) = ~ m:(n-m)' x , so that, by using also (3), Zn(X) =
m=0
n
2nOn(X/2) = [ c x n-m It is sufficient to study the irreducibility of Zn(X) and,
m=Q m

in this chapter only, conclusions referring to BP of degree n will apply to any of

the three polynomials YnfX), @n(X) or Zn(X ).

Proof of Theorem i. Part I. (i). Let n = pk, then 2n = 2p k and, by Lemma I, or 2

k-i
e(Cn) = ~ {2n-2-n+l} = n-lp_l= j=0[ pJ. Clearly, (e(Cn),n) = i holds, but we ob-

serve that this would no longer be the case for p = 2, n = 2m, where 2n = 2m+l,
 104

e(Cn) = n, so that (e(Cn),n) = n. This is the reason, why p is assumed to be an

odd prime only.

We claim that, for 1 < m S n-l,

m m(n-l)
(4) e(Cm) > ~ e(Cn) = n(p-l) "

Assuming for a moment (4), it follows that the N.p. of Zn(X) reduces to the straight

line from P0 = [0,0] to Pn = [n, e(Cn) ]. On account of (n,e[Cn)) = i, Corollary 1

shows that Zn(X), is irreducible. It remains to prove (4).

k-i k-i k-i

Let m = [ ajp J, 0 < a. < p-l, with o(m) = [ a.. Then n+m = pk + [ ajp3,
j=O ]- j=O J j=O

o(n+m) = o(m) + 1 and, by Lemma 2,

e(Cm) = p _ ~ (m+o(m)+o(n-m)-o(n+m)) = p _ ~ (m+o(n-m)-l).

To prove (4), we have to verify that n(m+o(n-m)-l) > m(n-l), or that n(o(n-m)-l) > -m
This is obviously true, because o(n-m) and m are both positive integers. Part (a)
is proved.
(ii) We now dispose first of a few other easy to prove statements of Theorem A.

If n = p-l, then PlCm for m = 1,2 ..... n and p2~c n ( = p(p+l)...(2p-2)), so that

Theorem B applies and Zn(X ) is irreducible. If n = p+l, then plc m for 2 < m < n,
2
p ~c n, because c n = (p+2)(p+3)...(2p)(2p+l)(2p÷2). It follows that the N.p. of

Zn(X ) consists of two sides, the first of length one, from P0 = [0,0] to P1 = [i,0],

and the second from P1 to Pn = In,l) of length (n-l) and without spots on it. By

Theorem A, Zn(X ) is either irreducible, or else splits into a linear factor and an

irreducible polynomial of degree n-i = p. The last alternative, however, is impos-

sible. Indeed, the linear factor has a real zero, (which, in fact, must be a ne-
gative integer) and we know (see Chapter i0) that BP of even degree (n = p+l is even)
has no real zeros. It follows that Zn(X ) is irreducible. Let us observe at this

point that we already have proved so far a weaker form of Theorem 3, namely

THEOREM 3' All BP of degrees n < 20 are irreducible, except, perhaps, for n = 15.

Proof. The statement is obvious for n = i and n = 2. For 3 S n ~ 20 , n # 15, all

odd integers are prime powers and all even integers are of the form p ± i. As for
the irreducibility of ZlS(X) , see Section 5.

[iii) In the more general case n = qp-l, q < p/2, one has n = (q-l)p+(p-l),
 105

o(n) = p+q-2, 2n = (2q-l)p+(p-2), o(2n) = 2q+p-3, e(Cn) = (2n-o(2n)-(n-~(n))/(p-l)=q,

and (n,e(Cn)) = (pq-l,q) = I. For 1 < m ~ n-l, m = rp+s, 0 < r < q, one has o(m) =

r+s. Also, n+m = (r+q)p+(s-l), with o(m+n) = r+q+s-l, if s # 0; n+m =

(r+q-l)p+(p-l), with o(m+n) = r+q+p-2, if s = 0. Finally, n-m = (q-r-l)p+(p-s-l)
with ~(n-m) = p+q-r-s-2. By Lemma 2, e(Cn) = r+l if s # 0, e(c n) = r if s = O.

Consequently, the line of slope i/p through the origin runs through all spots [rp,r],
corresponding to the values m = pr (with s = 0), while all other spots including
Pn = [n'en] = [(q-l)p+(p-l),q], are above that line. It follows that the N.p. of

Zn(X ) consists of two sides, one from the origin to P(q-l)p = [(q-l)p,q-l], and the

other from P(q-l)p to Pn" The first side has length h I = (q-l)p, and v I = q-l,

t I = (hl,Vl) = q-l, while the second has h 2 = p-l, v 2 = i, t 2 = (h2,v2) = 1 and

1 1
slope p_-~> ~ , as required by the convexity (downward) of the N.p. By Theorem A,

all factors of Zn(X ] have degrees of the form d = pp + ~(p-l) = (~+~)p-~,

= 0,1,...,q-l; 9 = 0,1. With this the proof of the first and of the last state-
ment of Theorem 1 (b) is complete.
[iv) The treatment of the general case n = qp+l, q < p/2, is entirely analogous.
One finds, successively, that o(n) = q÷l, 2n = 2qp+2, a[2n) = 2q+2, and e(Cn) = q.

For m = rp+s [I f m ~ n-l), one obtains that 0 < r ~ q, with r = q only if s = 0,

~[m) = r+s. Next, n+m = (q+r)p+(s+l), with a(m+n) = q+r+s+l, if 0 < s ! p-2;

n+m = (q+r+l)p, with o(m+n) = q+r+l, if s = p-l.

Furthermore, n-m = (q-r)p+l(l-s), with o(n-m) = q-r+l-s, if s = 0,i;

n-m = (q-r-l)p+(p+l-s), with o(n-m) = q-r+p-s, if 2 < s < p-l.

I r if s = 0, or i,
This leads to e(Cm) = r+l if 2 ~ s ~ p-2,
r+2 if s = p-l.

In particular, for m = 1 (i.e., r = 0, s = i), e(cl) = 0. As under (iii) we observe

that the straight line through Pl = [i,0], with slope i/p contains all spots with

m = rp÷l, e(Cm) = r, including n = qp+l, while all other spots lie above that line.

Consequently, the N.p. of Zn(X ) consists now of two sides, one horizontal, from

P0 = [0,0] to PI = [1,0], of length one, and the second from Pl to Pn = [n,q] of

length h = n-i = qp, v = q, of slope l/p, and t = (h,v) = (qp,q) = q. It follows,

by Theorem A, that any factor of Zn(X ) has a degree d = up+~, with ~ = 0,i ..... q

and ~ = 0,i. With this the first and the last statement of Theorem i (c) are proved.
 106

= n-q
(v) Let n = qp, with q < p/2; then ~(n) = q, 2n = 2qp, ~(2n) = 2q, e(Cn) p-i = q"

Also, for m = rp+s, with 0 < r ~ q-l, 0 < s ~ p-l, r+s ~ i, one has ~(m) = r+s and,
as before, computes v(n+m) = r+q+s, o(n-m) = q-r if s = 0, ~(n-m) = q-r-s-i if
1 ~ s < p-l. By Lemma 2, e(Cm) = r+l if s ~ 0, e(Cm) = r if s = 0. The line of

slope i/p through the origin contains all the spots [rp,r], including Pn = [qP'q]'

while all other spots [rp+s, r+l], s ~ i, lie above that line. The N.p. consists of
a single side, from P0 = [0,0] to Pn = In,q], with h = qp, v = q, and t = (qp,q) = q.

By Theorem A, all factors of Zn(X ) have degrees of the form d = ~p (~ = 1,2,...,q).

This finishes the proof of the first statement of Theorem 1 (d).

(vi) The next t>Te of n to be considered is n = pk-l. Now n = (p-l)(pk-l+...+p-l),

k-I
so t h a t g [ n ) = k(p-1). Also, 2n = 2pk-2 = pk + [ (p-1)pJ+(p-2) and a(2n) = ~ ( n ) .
j=l
k-I k-i k-I
By Lemma 2, e(Cn) - p-ln _ ~ pj. For m = ~ ajpj , ~ ( m ) = ~ a.. Also,
j=0 j=0 j=0 J
k-i . k k-i .
n-m = ~ [p-l-aj)p J, with o(n-m) = k[p-l)-g(m). Finally, n+m = p + ~ ajpJ-l.
j =0 j =0

k-i
If a~J = 0 for j = 0,i ..... r-l, while a r ~ O, then n+m = pk + ~ ajpj + (ar_l)pr +
j=r+l
r-i k-I
[p-l) j=O~ pj and o(n+m) = j=r
~ aj+r(p-l) = o(m)+r(p-l). By eemma 2, e(Cm) =

l-i-p-1 (m+~(m)+k(p-1)-~(m)-~(m)-r(p-1)} = ~ {m-~(m)} + k - r . We shall verify t h a t

me (Cn)
(5) m
n p-1 < e (Cm),

with equality if, and only if for all j ~ r, aj = p-l. Written explicitly, (S)
reads m-a(m) + [p-l)(k-r) ~ m, or

k-1
(6) ( p - 1 ) ( k - r ) >_ o ( m ) = ~ a..
j=r J

For given r, the right hand side is maximized by the choice of a. = p-l. In t h a t
J
case, and that case only, (6) becomes an equality. For any other choice (6) is a
strict inequality. This shows that all the spots that correspond to
k-i
m = [p-l) [ pJ (r = 0,i ..... k-2) lie on the straight line y = (e(Cn)/n)x = x/(p-l),
j=r
of slope i/(p-l) through the origin, while all other spots are above that line.
 I07

The N.p. of Zn(X ) consists, therefore of a single side with h = n = pk-l,

v = e(Cn) = pn__~= pk-i

p-I ' t = (h,v) = Pk-I
p-I ' and s = n/t = p-l. According to

Theorem A, all factors of Zn(X ) have degrees of the form ~(p-l) (~ = 1,2 ..... t).

This proves the first statement of Theorem 1 (e).

REMARK. A slightly stronger form of Theorem A can be obtained, by the remark that
any splitting of f(x) into irreducible factors can be effected by splitting f(x)
first into two (not necessarily irreducible) factors, then splitting any factor
that is not already irreducible into two factors, etc. At each stage the sum of
the degrees of the factors equals the degree of the polynomial being split. By
keeping track of this equality one can obtain the mentioned stronger version of
Theorem A. Before stating it, it is convenient to define a new term. ~enever
there are spots on a side of a N.p., these divide the side into collinear segments,
that we shall call (following Wahab [60]), the elements of the N.p. If bl,b 2 ..... bg

are the lengths (i.e., the horizontal projections) of the different elements, then
we have the following version of Theorem A, which in a somewhat different formula-
tion appears to be due to Wahab.

THEOREM A'. (Wahab [60]). Let f(x) = fl(x)f2(x) ... fz(x); then ds, the degree of

fs(X) is of the ! ~ ds = I ~sjbj, with ~sr = 0,i and 6sr6tr = 0 if s ~ t.

3
For a complete proof of Theorem A', see [60].
If we use this version of Theorem A, we have some added information concerning
the possible degrees of the factors of Zn(X), n = pk-l. Indeed we already know

that the spots on the single side of the N.p. of Zn(X) have the abscissae

k-i
(m-l) ~ pJ (r = 0,I ..... k-2). It follows that the lengths of the elements are
j=r

k k-i k-i k-2

b I = p -p , b2 = p -p ..... b k = p-l. By Theorem A', any factor of Zn(X) has a

k
degree ds = J ~ ~sj ( p k - j + l - p k - j ) (~s~~ = 0 o r 1). This result, not needed in the

p r o o f o f Theorem 1, i s m e n t i o n e d h e r e , because it seems t h a t this kind of closer

analysis may l e a d e v e n t u a l l y to a proof of Conjecture 1.
(n+m)! (n-m+l)(n+m) c and (n-m)(n+m+l)
(vii) By (3), cm : m!(n-m)! m m-i Cm+l = (m+l) Cm"

It follows that (without the earlier restriction q < p/2), e(Cm) = e(Cm_l),

unless p divides at least one of the factors n-m+l, n-m, or m. For pln this can
 108

h a p p e n o n l y f o r m ~ 0, o r 1 (mod p ) . T h i s means t h a t the spots arrange themselves in

horizontal rows o f e q u a l e ( C m ) ' S w i t h s t e p s u p , o r down, o n l y f o r m ~ 0 , 1 (mod p ) .

From t h e f a c t that t h e N.p. o f any Zn(X ) s t a r t s a t P0 = [ 0 , 0 ] and t h e downward c o n -

vexity of the N.p. it follows that if [ml, e(Cml) ] i s a s p o t on a s i d e o f t h e N . p . ,

w i t h e(Cml ) > 0, a l l other s p o t s w i t h m2 > m1 h a v e e(Cm2) > e ( C m l ) . F o r e(Cml) = 0

no such r e s t r i c t i o n exists and we h a v e s p o t s at all integers m w i t h e(Cm) = 0. It

furthermore follows that a s p o t w i t h e = e(Cm2) can be a v e r t e x , o r i n d e e d even a

spot on a side of a N.p. only if m 2 is the largest value of m for which e(Cm) =

e(em2 ) [see Fig. 2 ).

/
/

/_

ml m2 m
Fig. 2 Fig. 3

Quite generally, once e(cj) > 0,vertices or even spots on the sides of a N.p. can

occur only at abscissae m > j, such that e(Cm+l) > e(Cm). If n ~ 0 (mod p), this

requires m÷l E 0,i (mod p), or m ~ -I, 0 (mod p). The distance between two consecu-
tive abscissae of this type is either p, or p-l, or p+l, and this is the distance
between consecutive spots on a side of the N.p. This does not necessarily mean that
the length of the complete side of the N.p. has one of these values. It is quite
possible to have a side continue in the same direction and hit another spot after
another interval of length p-l, p, or p+l. The side of the N.p. cannot continue
with a smaller slope than on a preceding interval, as that would violate the convex-
ith downwards, but it may continue with the same slope.

After the side has traversed a number of spots, it either reaches Pn = [n'e(Cn)]

and the N.p. terminates, or else the polygonal line bends again upwards, thus start-
ing a new side of the N.p.
We may verify that the spots on any side lead to elements whose lengths
satisfy the requirement of Theorem A namely to be of the form ~s, ~ ~ t = (h,v). To
do this we observe that along a given side the spots that occur as interior points,
 109

say Q2' Q3 in Fig. 3 are lattice points (they have integral coordinates). No such

points can appear, if (h,v) = i. Indeed, vj = (v/h)(mj-ml) with h > mj-m I for mj

not the last abscissa so that v. cannot be an integer, unless v/h is a reducible
]
fraction, i.e. (h,v) = t > i. If v = tw, h = ts, (s,w) = i, then vj = (w/s) Cmj-ml)

with sl(mj-ml) , so that mj-m I = ~js f h = ts, ~j f t, vj = w~j, and finally mj-m I =

(vj/w)s = ~js, ~j ~ t, in agreement with Theorem A. From the point of view of

Theorem A', m 2-m I, m3-m 2 are the lengths of the elements of the N.p. It follows

that the length of an element (hence, that of the permitted degrees of a factor)
cannot be less than p-l.
In case n ~ -i (mod p), the factors n-m+l, n+m, and m are congruent mod p to
-m, m-l, and m, respectively. Hence, e(Cm+l) > e(Cm) only for m ~ 0, or I (mod p)

and the N.p. can have spots on a side of the N.p., or vertices, only for m ~ -I, or
0 (mod p). From here on the proof proceeds as before and its details may be sup-
pressed. The proof of Theorem 1 {f) is complete.

S. A given integer n may belong to several of the types considered in Theorem I.

So, e.g., for n = 9, we may ~Tite n = pm (p = 3, m = 2), or n = qp-i {q = 2, p = 5).

The prime with respect to which one considers the N.p. will of course be different
in each representation. In general, after selection of a definite prime p, Theorem
A indicates factorizations that may he possible for a given n.

So, e.g., for ZlsCX), the N.p.w.r. to p = 13 has two sides; one of the sides,

of length h I = 2, has a spot in the middle, the other one of length h 2 = 13 contains

no spots. The N.p. contains 3 elements and, by Theorem A (or A'), any polynomial
factor of Zls{X ) may only have degrees obtained as sums of the integers i,I, and 13.

We state such a result succintly in the form (15) = (1)(i)(13) and call it a scheme
of factorization. The N.p. of the same zls{x ) but with respect to p = 3 leads to
{IS) = {9)[4){2). These two results, while different, are not contradictory. They
can be reconciled in two ways. Either trivially, as (1+1+13) = {15) = (9+4+2) if
ZlsCX) is irreducible and none of the potentially allowed factorizations actually
exists, or else by observing that both schemes allow the non-trivial factorization
(i+1) C13] = {2){13) = C2)(9+4).
Two such f a c t o r i z a t i o n schemes, with a common, n o n - t r i v i a l f a c t o r i z a t i o n , w i l l
be c a l l e d oompatible. Otherwise, we c a l l them incompatible. I t is c l e a r that i f
one can e x h i b i t two incompatible f a c t o r i z a t i o n schemes, one thereby w i l l have proved
the C r r e d u c i b i l i t y o f the polynomial considered. In the case o f zisCx), the N.p.

w.r. t o p = S leads to the scheme (1S) = ( S ) { 4 ) ( 6 ) . Also t h i s i s compatible with

C15) = { 9 ) { 4 ) ( 2 ) , because (S+4)(6) = (9)(6) = (9)(4+2). I t i s , however, incompatible
 110

with the scheme (15) = (11(I)(13) obtained from the N.p.w.r. to p = 13 and we con-
clude that Zls(X ) is irreducible.

There should be no need now for a more formal proof of the

PROPOSITION i. If the N.p. of Zn(X ) with respect to ~wc different primes lead to

incompatible schemes of factorization, then Zn(X) is irreducible.

6. In this section we shall prove Theorem 4.

This will be needed in the proof of Theorem 2 and only by use of Theorem 2 are
we going to complete the proof of Theorem i.
We already know that Zn(X ) is irreducible if n < 20. This follows from Theorem

3' and the irreducibility of Zl5(X ) proved in Section 5. Also, Zn(X ) is irreducible

if n = p. If n is not a prime, let Pl < n < P2' with Pl and P2 consecutive primes

and set n = Pl÷kl = P2-k2-1. Then k I = n-Pl < p2-Pl and k 2 = P2-n-i < p2-Pl-i or,

k 2 ! p2~Pl-2.From Theorem C we know that p2-pl ~ Pl/8 for Pl ~ $3. For Ii ~ Pl < 53,

Pl
we verify that p2-Pl < -~ (the largest value is attained for Pl = 23, with 29-23 =

6 = (6/23)Pi). This inequality holds by Theorem C for all Pl ~ ii, which is more

than sufficient because we are concerned only with n ~ 21 (Pl ~ 191. We note, in

particular, that Pl÷2kl < Pl+2(P2-Pl ) < 2Pl. This insures that the N.p. of Zn(X)

w.r. to Pl has the first k I coefficients not divisible by PI" Indeed,

_ ~l(n÷ll (n-l)n(n+l)(n+2) (n-kl+l)"'(n+kl)

Cl i! ' c2 = 2! ''''' Ck I = kl! =

(Pl+l)...(Pl+2kl
and Pl < (Pl ÷I)' (Pl ÷2k) < 2PI"
kl!

We claim that all other coefficients are divisible by Pl to exactly the first power.

For k I < m < PI' the factor Pl which now occurs in the numerator is not cancelled

by any factor of the denominator. For m = PI'

(kl÷l)[kl÷21""Pl'"(2Pl)
c = and pl [ 2 still holds. None of the fac-
Pl Pl ~ cpl' Pl~Cpl

tots by which cpl has to be multiplied in order to obtain the coefficients Cm,

Pl < m ~ n, contain multiples of PI' either in the numerator, or the denominator,

and the claim is proved. We may verify it, in particular, for m = n,

 1ii

cn = (n+l)...(2n) = (Pl+kl÷l)...(2Pl)...(2Pl+2kl) with Pl+kl+l > Pl and,

2P1+2k I < 3Pl. It follows that the N.p. of Zn(X ) w.r. to Pl consists of 2 sides:

the first one o£ length kl, on the real axis, with spots at all the integers and the

second one from Pkl = [kl,O ] to Pn = [n,l], with no spots on it and of length

h2 = n-kl = PI" By Theorem A this leads to the factorization scheme

(n) = (1)(1)...(l~(Pl) and it follows that z (x) contains an irreducible factor of

n
k I times

degree at least equal to PI"

We study in exactly the same way the N.p. of Zn(X ) w.r. to P2" In particular,

(n-k2+l)...n(n+l)...(n+k2) (P2-2k2-1)...(P2-k2-1)(P2-k2)...(P2-1)
= =

Ck 2 k2~ k2~

and all coefficients cm with 0 5 m ~ k 2 are not divisible by P2' while

(P2-2k2) ... (P2-1)P2

Ck2+l = (k2+l) ! and all successive ones contain P2 to exactly the

first power. Indeed, n < P2' so that no factor o£ the denominator can cancel the

factor P2 in the numerator. In particular, cn = (n+l)...(2n) = (P2-k2)...

p2...(2P2-2k2-2) and the last factor is less than 2P2. It follows as above that

Zn(X) contains an irreducible factor of degree at least equal to n-k 2. If we

combine this result with the preceding one and recall the notation k = k(n) =
min(kl,k2), we can state

THEOREM 4. For every n, the BP of degree n contains an irreducible ~ c t o r of degree

at least equal to n-k(n) and can contain r~ factor of degree d with k(n) < d < n-k(n)

7. Proof of Theorem 2. As already seen in Section 2, for every a > 0 and x ~ N(E),
there is at least one prime p in the interval x < p ~ (l+e)x. In particular, it
follows with x = Pl' that there is at least one prime P2 such that Pl < P2 ! (l+e)P l-

Let Pl < n < P2 with PI' P2 consecutive primes; we want to estimate the largest

possible value of k(n). In k I = n-Pl and k 2 = P2-n-l, set n = rlPl, P2 = nr2' so

that k I = n(l-i/rl) , k 2 = n(r2-1- i/n). Now, k(n), the smaller one of kl,k 2, takes

its largest possible value, if n happens to fall between Pl and P2 in such a way as

to make k I = k 2. This means, l-rl I = r2-l-n-l. Also, nP2 = nrlr2P I, whence

rlr 2 = p2/Pl < l*e. The most unfavorable situation (large k) occurs when the
 112

separation between Pl and P2 is largest, i.e. for p2/Pl = i+~. By solving the system

2+e (2+ i/n)(l+e)

of 2 equations obtained for rl, r2, we find r I = 2+ i/n r2 =
' 2+ C

From this we obtain k I = n-Pl f n(l-r i I) = ne-I

2+e and verify that also

k2 P2_n_l < n(r2_l_n-l) n¢-i Consequently, k = k(n) < n~-I e The

= - = 2+¢ " , _ ~ < n 2+--T "

fraction k/n can be made arbitrarily small by taking ¢ sufficiently small. We may
indeed take ¢ > 0 arbitrarily small, provided we take n sufficiently large, n ~ N(a).
By Theorem 4, the degree of an irreducible factor of Zn(X ) exceeds

n-k[n) > n(l- e = I-

_ 2-~a) = Ann , say, with A n 2+¢c i
I+ g/2 "

For n + ~, one has a + 0, hence lim A = 1 and this proves the first state-
n

ment of Theorem 2.
We now recall that, by Theorem C, we may take a = 1/8 for n ~ Pl ~ 53. Hence,

for n >- 53, k(n)

n < 2+ 1/8
1/8 i
= i--7 and A n -
> I+ i1/16 16 '. this proves the validity
- i-Y

of Theorem 2.

8. Proof of Theorem i. Part II. It is now easy to complete the proof of Theorem
I. Let us assume first that n > 53. We shall show that under the assumptions of
the theorem the N . p . w . r . to p (of Theorem i) and the N . p . w . r . to Pl or P2 (largest

prime less than n, or least prime larger than n) lead to incompatible schemes of
factorization, so that the irreducibility follows from Proposition i.
If we assume that an(X ) is not irreducible and n = qp-l, q < p/2, as assumed

in Theorem 1 (b), then the lowest possible degree of an irreducible factor, satis-
fies, as seen, d ~ rp-i ~ p-l. From Theorem 4 it follows that d < k(n), so that
p-I < k(n). Under the assumption p-I > k(n), this is not possible, so that Zn(X) is

irreducible. Similarly, under the conditions of Theorem 1 (c), if Zn(X) is not irre-

ducible, the minimal degree d of one of its factors satisfies either d = 1 or d ~ p.

By Theorem 4, d ~ k(n), so that d • p implies p ~ k(n). If we know, however, that
p • kin), the last inequality is impossible, so that Zn(X ) must be either irreducible

or else has a linear factor with a real (in fact, integral) zero. If q is odd, then
qp+l is even and (see Chapter I0) the last alternative cannot occur.
Exactly the same reasoning, applied to the conditions of Theorem 1 (d) and (e),
with p • k(n) or p-i • kin), respectively,proves the irreducibility in those cases.
If, instead of p • k(n), or p > k(n) + 1 we know instead that q ~ 17, then, say,
in Theorem I (d) we obtain p = n/q ~ n/17. As proved in Section 7, k(n) < n/17, so
that p > k(n). If Zn(X ) is not irreducible, then all factors have degrees equal to
~p, ~p ~ p • k(n), and this contradicts Theorem 4 and proves the irreducibility of
 113

Zn(X ). It is clear that, while not stated in the Theorem, the condition q ~ 17 can

be substituted for p-i > k(n) also in Theorem i (b) and (c). With this, Theorems 1
and 2 are completely proved, at least for n 5 20 and n > 53.

9. It still remains to be shown that Zn(X ) is irreducible for 2 1 S n s 53. In the

proof we may use all previous results, except, of course, those statements of
Theorems 1 and 2, such as those referring to q f 17, that were obtained under the
assumption n > 53.
From Theorem 1 (a,b,c), we know that for all n = pm and n = p ± I, the Zn(X)

are irreducible (without the restriction n > 53). After we eliminate these n, we
still remain with eight values. For each of them it is easy to find k = k(n) by
inspection. So, e.g., for n = 33, Pl = 31, P2 = 37, k I = 2, k 2 = 3 and k(33) = 2.

For five of these 8 values of n we may use Theorem l(d) as follows: 21 = 3.7, k = I;
33 = 3.11, k = 2; 34 = 2.17, k = 2; 39 = 3.13, k = i; 51 = 3.17, k = i. In each
case any factor would have to have a degree d that is a multiple of the larger of
the two prime factors of n, and at the same time have d ~ k. For instance, if
n = 3S, d = ~.Ii ~ Ii and d < 2. This is not possible and hence, z33(x) is irreduc-

ible. The last remaining three values of n are:

35 = 2.17+I = S.7, with factors of degrees 17r, or 17r+l only, k = i, and also
with the factorization (35) = (10){25) (mod 5);

45 = 4.11+1 = 32.5, with factors of degrees llr, or llr+l only, k = i, and also
with the factorization (45) = (25)(4)(16) (mod 5);

50 = 3.7-I, with factors of degrees 7r, or 7r-i only, and k = 2.

By Theorem A and Theorem 4, all corresponding polynomials Zn(X ) are irreducible.

With this Theorem 1 and Theorem 2 are completely proved.

10. Proof of Theorem 3. We already proved a weak form of Theorem 3, namely

THEOREM 3". All BP of degree n, with n < 53 are irreducible.

In order to complete the proof of Theorem 3, we proceed as before. We first
suppress from the list of all integers n, 53 < n < 400, all primes, odd prime powers,
and integers of the form p ± i. For the remaining integers n, one determines k(n),
by looking for the primes closest to n. Next, one tries to fit each n into one of
the patterns of Theorem i, such as n = qp+~, 6 = -i, 0, +i, q < p/2 or n = pm-l.
This will impose upon the smallest degree d of a factor of Zn(X) the condition
d ~ p, d > p-i or d = i. On the other hand, d < k(n) and, if d # 1 and k(n) < p-l,
the resulting contradiction is proof of the irreducibility of the corresponding Zn(X )
If d = 1 and n is even no linear factor may split off, with the same result.
 114

After the odd p r i m e powers and n = p ± 1 h a v e been eliminated, there remain 61

integers n with 53 < n < 200. These are listed in the T a b l e i, e a c h w i t h at least

one representation n = qp+~ or n = pm-l. In all cases p ~ ii a n d k(n), which is

not listed, never exceeds 6.

Table 1

55 = 5.11 115 = 5 . 2 3 154 = 5.31-i

56 = 3.19-1 116 = 4 . 2 9 155 = 5.31

S7 = 3.19 117 = 2 . 5 9 - 1 159 = 3.53

63 = 2.31+1 = 32.7 118 = 2.59 160 = 7 . 2 3 - 1

64 = 5.13-1 119 = 7 . 1 7 161 = 7.23

65 = 5.13 120 = 112-1 165 = 2 . 8 3 - 1

69 = 3.23 122 = 2.61 170 = 9 . 1 9 - 1

75 = 4.19-1 123 = 3.41 171 = 9.19
76 = 4.19 124 = 4 . 3 1 175 = 6 . 2 9 + 1 = 7.52

77 = 6.13-1 129 = 3.43 176 = 3 . 5 9 - 1

85 = 5.17 133 = 7 . 1 9 177 = 3.59

86 = 2.43 134 = 2.67 183 = 3.61

87 = 3.29 135 = 8 . 1 7 - 1 184 = 5 . 3 7 - 1

91 = 4.23-1 141 = 3.47 185 = 5.37

92 = 4.23 142 = 2.71 186 = 6.31
93 = 3.31 143 = 2 . 7 1 + 1 = 11.13 187 = 4 . 4 7 - i = 11.17
94 = 2.47 144 = 5 . 2 9 - 1 188 = 4 . 4 7

95 = 5.19 145 = 5 . 2 9 189 = 1 0 . 1 9 - 1

99 = ? 146 = 2 . 7 3 195 = 2 . 9 7 + 1 = 15.13

105 = 2.53-1 147 = 4 . 3 7 - 1

iii = 3.37 153 = 8 . 1 9 + i = 9.17

From Table 1 it a p p e a r s that all integers u p to 200 fit r e a d i l y into one of t h e

patterns of Theorem i, e x c e p t for n = 99. The polynomial z99(x), however, is irre-

ducible by Theorem 1 (f) a n d T h e o r e m 4. Indeed, P m = Ii is the largest prime divisor

o f 99, so that z 9 9 C x ) c a n n o t have factors of degrees less than I0. On the other

hand, k<99) = k 2 = 101-99-1 = i and z99(x), if not irreducible, can split at m o s t

into a linear factor and an i r r e d u c i b l e factor of degree 98, whence the irreducibility

follows. Also for n = 63, 143, 175, and 195, all odd and of type n = q p + l <q < p/2)
one must still show that the}, do n o t c o n t a i n linear factors. F o r b o t h n = 63 and

n = 175, the largest prime factor is P m = 7 and b y T h e o r e m 1 (f) n o factors of de-

grees less than 6 exists. F o r n = 143 and n = 195, P m = 13 and n o factors of de-
grees less than 12 exist.
 115

~ile not needed for the purpose on hand, it is instructive to see to what
factorization schemes one is led in these S cases, by the N . p . w . r . to the corres-
ponding largest prime factor Pm" One obtains (99) = (!i)(i0)(78), (63) = (7)(7)(49),

(143) = (13) ( 1 2 ) ( 1 1 8 ) , (17S) = ( 7 ) ( 7 ) ( 4 9 ) ( 4 9 ) ( 4 9 ) ( 6 ) ( 8 ) and (19S) : (139(139(169).

One may remark that, while n : 63 and n = 195 lead to d ~ Pm' for n = 99, 143,

and 175, one obtains nothing better than d ~ Pm-l, so that the statement of Theorem

1 (f) cannot be improved without some added restrictions.

The verification that also all Zn(X ) w~th 200 < n < 400 are irreducible is
equally easy - and dull• For 201 < n ~ 300, all integers fit one of the patterns of
Theorem i, except for 209. Here Pm = 19, k(209) = i, so that z209(x) is irreducible

by Theorem 1 (f) and Theorem 4. In 301 < n < 400 we find four values of n that have
to be considered separately, namely n = 323 (Pm = 19, k = 6), 324 (324+i = 52 13

Pm = 13, k = 6), 351 (Pm = 13, k = i) and 391 (Pm = 23, k = 2). Three of them are

irreducible by Theorem 1 (f) with n = qp and Theorem 4, while for n = 324 one uses
Theorem 1 Cf) with n = qp-i and Theorem 4.
In spite of the ease with which we are able to prove the irreducibility of any
single polynomial Zn(X), a general method still eludes us. Such examples as n = 99

and n = 209, etc., show that we cannot take it for granted that we shall be able to
fit all integers n, into any finite set of patterns, for which the irreducibility
can be proved. By working out a large number of N.p., one is led to make certain
conjectures.
There also is some hope that the method used in the proof of Theorem i (e) and
1 (f) could he perfected, in order to prove

CONJECTURE 2. If p' # p" and p'p"In, then the N.p. of Zn(X) w.r. to p' and to p"

lead to incompatible schemes of factorization.

It is clear that the proof of Conjecture 2 would go a long way towards the
proof of Conjecture i. In fact, the only integers not covered would be those of

the form n = 2mp k and which also cannot he represented as n = qp"' ~ i, with appro-
priateiy small q.
 CHAPTER 12
THE GALOIS GROUP OF B.P.

I. The Galois group of the BP of degree n will be denoted by Gn, the symmetric

group on n s}~bols by S n and the corresponding alternating group by A n .

A group is said to be of degree n, if it is isomorphic to a permutation group on

n symbols.The terms "transitive", "primitive", etc., have their usual meaning from
group theory as defined, e.g., in [9]. The term irreducible will mean irreducible
over the rationals, as in Chapter ii.
The material of this chapter is essentially contained in [53, Section 5] and in
[54], but several proofs have been reworked.
The main result is

THEOREM i. For every n, if G n is transitive, then G n = S n.

This theorem can be rephrased. Indeed, the Galois group of a polynomial is

transitive, if and only if the polynomial is irreducible. Hence, we may state

THEOREM i' If yn(Z) i8 irreducible, then G n = S n.

We saw in Chapter ii, that a large class of BP consists in fact of ireducible

polynomials and we formulated there Conjecture i, according to which all BP are ir-
reducible. This leads us to state

THEOREM i". Cor~ecture 1 of Chapter ii ~plies tb~t, for every n, G n = S n.

2. Theorem i' is equivalent to Theorem 1 and implies Theorem I". Hence, it is

sufficient to prove Theorem i'. For that we shall need several known results. As
in Chapter ii, we shall have to content ourselves with the statements only and refer
the reader to easily available proofs in the literature.

THEOREM A. (I. Schur [SZ], Theorem V). Let

n
f(x) = ~ a.x n-3 , a. rational integers,
j=o J J

have the discriminant D and be irreducible, and let G n be its GaZois group. Let us

assume that there exists a prime p, which satisfies the following conditions:

pm]D with m > n, P]an, p2~a n.

Assume also that if f(x) ~ xkg(x) (mod p) and d is the discriminant of g(x), then
p~d. Any such prime is a divisor of the order of Gn-
If, furthermore, n/2 < p < n-2, then G n = A n in case D is a perfect square,
Gn = Sn, otherwise.
 117

THEOREM B. (Dedekind; see [$2]). Let us suppose that the polynomial f(x) splits
modulo p into a product of r irreducible polynomials fi(x) (i = 1,2 ..... r), incon-

gruent in pairs modulo p. Then the Galois group of f(x) contains at least one per-
mutation of r cycles, such that each cycle corresponds to one of the factors fi(x),
and the order of each cycle is equal to the degree of the polynomial to which it
corresponds.

THEOREM C. (Burnside; see [9],XI). (i) Ef n (> 3) and 2n+l are both primes, then
any triply transitive group of degree 2n+3 contains the alternating group (§ 165,
page 214, Exercise).
(2) For n = 4, there are no primitive groups, except A 4 a n d s 4 (§ 166, page 214,
(ii)).
[3) Every triply transitive group of degree seven contains A 7 (§ 166, pages 216-218,
(v); see also [52(a)], p. 449, or [S2(b)], p. 197).

(4) If pe is the highest power of p contained in n! and if p < 2n/3, then pe-I
is the highest power of p that divides the order of any primitive group of degree n,
which does not contain A n 6§ 160, pages 207-208).

THEOREM D. (Cauchy; see [30], p. 74). If p divides the order of a group G, then G
contains an element of order p.

THEOREM E. (Jordan [39], Note C; see also [9], p. 214, Theorem I and [5Z](a), p.448;

[S2(b)], p. 196). If G is a primitive transitive ~arouv~ of dearee~ n and p > n[ ~s" a

divisor of the order of G, then the degree of transitivity of G is at least n-p+l.

3. The method of proof follows, in general outline that of Schur ([52]), and it is
largely based on Theorem A. We also shall need the following

LS~MAI.For (n+l)/2 ~ p < n, 0n(X ) ~ xPOn_p(X ) (mod p).

Proof. Trivially n z (n-p) (mod p); consequently, for m < n-p+l, one has
m < n - (n+l)/2+l = (n+l)/2 < p, so that

a(n) = (n-m+l)...n(n+l)...(n+m)
m 2mm!

((n-p)-m+l)...(n-p)(n-p+l)...((n-p)+m) = a(n-p) (rood p).

2mm! m

On the other hand, for n-p ~ m < p, a (n) contains the factor p in the numerator,
m
but not in the denominator and a (n) E 0 (mod p). For p ~ m ~ n, the denominator
m
contains the factor p to exactly the first power (because n < 2p-i < 2p), while the
numerator contains at least two multiples of p (p itself and 2p, because n ~ p,
 118

m ~ p imply n+m ~ 2p). It follows that

n n-p n-p (n-p) -m+p

OnCX) = [ a Cn) xn-m _ [ aCn)x n-m _ [ aCn-P)x
m m m
m=0 m=0 m=0

- xp nip a ( n - P ) x ( n - p ) - m = xP0n_p(X ) (mod p ) .

m:0 m

We now verify that the conditions of Theorem A hold for 0n(X ) . Specifically, we
shall do the following:

(a) compute the discriminant Dn of On(X);

(b) show that for n > 14,and also for n = i0, there exists a prime p such that
a (n) , p 2 ~ a ~ n ) ;
(2n-1)/3 < p < n-2, and a l s o pn]D n , p n

(c) show t h a t @n(X) ~ xPOn_p(X ) (mod p) and P2Dn_p;

(d) verify that Dn i s n o t a p e r f e c t square.

Theorem A can be a p p l i e d to those values of n, for which these conditions can

be v e r i f i e d (in fact, the present condition (h) i s more s t r i n g e n t than a correspond-
ing condition o f Theorem A) and we c o n c l u d e t h a t f o r t h o s e n, G n ~ A n . From (d) i t

follows, however, that Gn # An, so t h a t i n d e e d Gn = S n.

If those conditions h o l d f o r some p t h a t falls outside the range required by

(b), then it still follows from Theorem A, t h a t p divides t h e o r d e r o f Gn. By

Theorem D i t t h e n f o l l o w s t h a t Gn c o n t a i n s an e l e m e n t o f o r d e r p. I f a l s o p > n/2,

then the group G n (transitive by assumption) is obviously primitive. Theorem E then

insures a rather high degree of transitivity, which, in some cases, suffices to show
that G n ~ An, so that, by (d) above, G n = S n-

In the few remaining cases we shall succeed to reach that conclusion, by appeal-
ing to Theorem B.

4. (a) We r e c a l l (see, e.g. [Sg], § 28) t h a t R(f,g), the resultant o f two p o l y n o -

n n-i m .
mials f(x) = ~ c.x of degree n and g[x) = [ d.x m-3 of degree m, is a polyno-
i=0 I j=O 3
mial function of the coefficients ci, dj (i = 1,2 ..... n; j = 1,2 ..... m) that vanishes
n m
if and only if f and g have a common zero. In fact, R(f,g) = Cod m n0i _ ~ 1 I I (xi-Yj),
"= j=l

where the x i range over the zeros of f(x) and the yj over those of g(x). Clearly,

n m
R(f,g) = c~ ~.= g(xi) = (-l)nmd~ ~.= f(yi) = (-l)nmR(g,f). We shall be concerned

here with monic polynomials, so that c O = d o = I, and nm will always be even, so

 119

that these formulae simplify accordingly.

The discriminant D of a polynomial f(x) is, essentially, the resultant of f(x)
and f'(x). More precisely,

(i) R(f,f') = (-1)n(n-1)/2Co D.

n
It follows that D = c~ -2 ~ - [ f ' ( x i ) : (-1) n(n-1)/2^2n-2
~0 [ I (x i-xj)
i=l i#j

c 2n-2
o I 1I (xi_xj)2 ' where the x.'s ~ d x.'s run through the zeros of f(x). The
i<j z J
last expression is often taken as definition of D.
In order to compute Dn, the discriminant of @n(X), we first compute R(@n_l,@n).

By (3.5), @n+l(Xi) = X~@n_l(Xi) for each zero x i of 0n(X). By taking the product

over all these zeros and recalling also that O is monic, we obtain
n

n • 2 ~n (a (n)
R(On'0n+l) i On+l(X i) = (xlx 2. -xn) i:i en-l(Xi) : - n )2R(0n'en-i )"
= i=l]

Either n, or n-I is even; hence R(en,0n_l) = R(0n_l,0n) and we obtain by an obvious

inductive argument, {a(n)a (n-l) .a~l)}ZR(Ol or, by using (2 8)

R(On'@n+l) = n n-i "" 'O0)' "

and R(@I,eO) = I,
n-i
(2) R(0n,@n_ I) = {]--F -(2k):
- }2.
k=l 2kk!
On the other hand, by (3.8), On(Xi) = -XiOn_l(X i) and, if we take the product over

all zeros x i of en(X ), we obtain

n n

R(@n,@n) = [ I @n(Xi) : ( - l ) n X l . . . X n @n_l(Xi) = (-l)na~n)R(@n,@n_l).

i=l "=

By (i), the first member equals (-l)n(n-l)/2D and it follows that

n
D = (-l)n(n+l)/2a(n)-
n Rn(@n,@n_l), so that, by (2),
n

n(n+l) n-i (2k)! ~2

(3) D n = (-I) 2 (2n)! {T-T - - ~ •
2nn! k=l 2kk!
n-i
The denominator in (3) may be written as (2.4.6...2n) ! I (2.4...2k) 2 and a
k=l
fairly easy counting atgument shows that each even integer 2m occurs exactly to the
power 2n-2m+l. This is exactly the same power to which it occurs in the numerator,
so that ]Dn! reduces to the product of the odd integers in the numerator. The odd

factor 2m-l (1 ~ m ~ n) occurs exactly once in (2n)!, and also once in each factorial
(2k)! for m < k < n-l, so that 2(n-l-(m-l))+l= 2n-2m+l is the exact power to which
 120

2m-i occurs in [3). It follows that

n
(4) + Dn = ~ T [2m-l) 2n-2m+l = 12n-132n-3---(2n-3)S(2n-l) 1
m=l
9
(b) Set n-3 = ~ x and recall from Chapter II that, for x _> 48, there is a prime in
9
the interval x < p <_ ~ x. It follows that there is a prime p that satisfies the in-
equalities
2n-i 8n 8 9
3 < ~-- - ]- = x < p _< ~ x = n-3 < n-2,

9
provided that n > 3 + ~ • 48 = 57. It is an easy exercise to verify that for all

14 < n < 56 there also exists such a prime (e.g. for n = 20, 2.20-I
3 _ 13 < 17 < 20-2)

2n-i 19
Also for n = I0 we find ~ = - ~ < 7 < 8 = n-2. By (4), it follows that any

such prime divides D n to the exact power 2n-p _> n+2 > n. It also follows from (2.8),

that Pla(n)n , p2~an(n).- Indeed, p < n-2 < n+l < 2p < 2n < 3p+l; however, 2n is even,

so that the last inequality is equivalent to 2n < 3p.

(e) By the Lemma, 8nCX ) - xPen_p(x ) (mod p) for (n+l)/2 <_ p < n, a f o r t i o r i for

(2n-i)/3 < p < n-2. Next, by ~4), IDn_pl = 12n-2p-132n-2p-3...(2n-2p-3)3(2n-2p-l) I"

however, p > 2n-2p-l, because, by assumption, 3p > 2n-l, and P2Dn_ p.

(d) All primes that divide (2n)'. only once, divide D n to an odd power, so that,
except for n = I,D n is not a perfect square.

With this, all conditions of Theorem A are verified and it follows that for
n > 14 and also for n = I0, the Galois group G n of the irreducible BP @n(X) is the

symmetric group S n.

5. For n = 13, 12, ii and n ~ 9, no primes exist in C2n-i)/3 < p < n-2. For n = 1
G 1 = SI, both reducing to the identity. For n = 2 and n = 3, the polynomials are

irreducible and the discriminants are not squares; hence G 2 = S 2 and G 3 = S 3.

For the remaining values of n, we try to find primes p = Pn that satisfy at

least the weaker inequalities (2n-i)/3 < p < n. We find exactly one such prime for
all 4 < n < iS, (n ~ 10), except for n = 5 and n = II. Specifically, P4 = S, P6 =

P7 = 5, P8 = P9 = 7, P12 = PlS = ii. We may recall from Chapter ii, that all BP

involved are in fact irreducible, hence, the G ' s are transitive and, as also in
n
each case Pn > n/2, the respective Gn'S are primitive. By Theorem E, the Gn's have

degrees of transitivity £n ~ n-Pn+l' so that £4 ~ 2, £6 ~ 2, £7 ~ 3, £8 ~ 2, £9 ~ 3,

£12 ~ 2, and £13 ~ S.
 121

By Theorem C (2) there are no primitive groups of degree 4, except for A 4 and
$4, so that G 4 ~ A4 and (recall (d), whence G 4 # A4), G 4 = S 4.

By Theorem C (3), the only triply transitive groups of degree seven are A 7 and
$7, whence, as before, G 7 = S 7. For n = 13 we invoke Theorem C (I), observe that

Pl = 5 and P2 = 2PI+I = ii are both primes and conclude that G13, with n = 2Pi+3 = 13,

contains AI3, so that GI3 = S13.

6. The only values of n for which Theorem I' has not yet been proved are
n = 5,6,8,9,11, and 12.
In these six cases we shall use Theorem B and observe that we have

O5(x ) ~ (x3+x2+4x+5)(x-2)(x-l) (mod 17)

O6(x ) ~ (x3+4x2-3x+6)(x2+x-l)(x+3) (mod 13)

O8(x ) ~ (xS+9x4-7x3+8x2-7x+4)(x2+2x+2)(x+6) (mod 19)

(s)
@9(x) ~ (xS+x4-Sx3-x2+Tx+13)(x3+12x2-13x+3)(x+3) (mod 29)

7+ 6+ 5+ ~ 4+ 3 2
@ll(X) ~ (x 39x 35x l.lx 148x 60x +25x+S3)(x4+27x3+14x2+74x+23) (mod 149)

@12(x) ~ (x7÷2x6+27xS+28x4+30x3+38x2+lSx+3)(xS+76x4+65x3+63x2+29x+lS) (mod 89)

In each of these six factorizations, the polynomial factors on the right are mutually
incongruent and irreducible modulo the respective prime modulus. The first three
factorizations were done by hand, by the present author, in 1949; O9(x) has been

factored modulo 29 by M.Ne~Ean and K. Kloss on the IBM 704 at the National Bureau of
Standards, in October 1961, while @ll(X) and Ol2(X ) have been factored by J.D. Brill-

hart and R. Stauduhar, on the CDC 6400 of the University of Arizona, in February 1969,
by using an algorithm of E. Berlekamp.
It is, of course, quite easy to verify these factorizations by hand) once they
are found. The discovery of a prime that leads to a usable factorization and the
determination of the factors is perhaps less trivial. The verification that the
factors are indeed irreducible modulo the respective prime is an operation of inter-
mediate difficulty. For whatever interest it may present, the factorization of
@8(x) will be discussed in detail in Section 7 and is fairly typical for all similar
factorizations.
By Theorem B we know that for n = 5,6,8,9)ii, and 12 the corresponding G n con-

tain at least one permutation Pn' of the following structures, respectively:

 122

P5 = (a,b,c) P9 = (a,b,c,d,e)(f,g,h)

P6 = (a,b,c)(d,e) PII = (a,b,c,d,e,f,g)(h,i,j,k)

P8 = (a,b,c,d,e)(f,g) PI2 = (a,b,c,d,e,f,g)(h,i,j,k,£).

P5 leaves two symbols invariant; P6' P8' and P9 each leave one symbol invariant,

while PII and Pl2 move all symbols.

G 5 contains a cycle of order three, hence the order of @5 is divisible by 3; as

3 < (~) • 5, 3[5:, and 32~5: , Theorem C (4) with e = 1 shows that G S ~ AS, whence

G 5 = S~.
D

With P6' G6 contains also P63 = (e,d), a transposition. As a primitive group

that contains transpositions, G 6 = S 6.

With P8' G8 also contains P82 = (a,c,e,b,d), an element of order 5 The order

of G 8 is, therefore, divisible by 5 and, by Theorem C (4) with e = l and the remark

that 5 < (32-) .8, it follows that GS~A8, i.e. G 8 = S 8.

Similarly, G 9 contains P93 of order 5(< (~) • 9), 519!, 52~9 ' and we conclude as
4
before that G 9 = S 9. For Gll we use Pll of order 7(< ( ) • ll) and for Gl2 we use

P125 also of order v, to reach the same conclusion. With this the proof of Theorem l'

hence also that of Theorem 1 and of Theorem I", is complete.

7. In order to prove that G n ~ A n, the first step is to decide on the type of a

desired factorization. For G 6 we could show that it contains a transposition, but

in general, this cannot be achieved. In that case we try to find a Pl < 2n/3, such
2,
that Plln!, Pl~n., in order to apply Theorem C (4). For n = 8, there is little

choice; only Pl = 5 can be used, so that we aim at a factorization from which the
existence of an element of order 5 can be inferred. Next, we have to choose a prime
p as a modulus in Theorem B. No primes less than 2n can he used, because for them

a~n)~ a~ n) ~ 0 (mod p) and en(X ) ~ xZg(x) (mod p) already exhibits a repeated factor,

which makes Theorem B inapplicable. For n = 8, the smallest prime to be tried is,
therefore p = 17. We first look for linear factors modulo 17, but there are none,
because 88(k) ~ 0 (mod 17) for all integers k. Next one may want to consider a

decomposition e8(x) ~ f2(x)f6(x) (mod 17), with a quadratic factor f2(x) and a

factor f6(x) of degree 6. This, however, would not be useful. Indeed, in the
 123

absence of linear factors one cannot obtain a factor of degree 5, by further factor-
ing of f6(x). Hence, we attempt to split @8(x) as a product f3(x)fs(X) (mod 17).

If we write monic polynomials f3 and f5' of degrees 3 and 5 respectively, with un-

determined coefficients, we are led to a system of congruences modulo 17. It turns

out that this system has no solution. As a factorization 04(x ) ~ f4(x)f4(x) (mod 17)

would not be useful, we pass to the next higher prime p = 19. Here we find that
e8(-6 ) ~ 0 (mod 19) and, by factoring out x+6, we obtain

(6) @8(x) ~ (x+6)(x?-8X6-6xS-7x4-Sx3+6x2-6x+8) (mod 19).

By direct substitution of integers k (k = 0,1,...,18) into the factor of ?-th degree

we verify that there exist no other linear factors. Hence, we look for a congruence
of the form

(7) xT-8x6-6xS-Tx4-Sx3+6x2-6x+8 ~ (x2+~'x+B')(xS+~x4+Bx3+~x2+~x+E) (mod 19).

This leads to the following system of simultaneous congruences, all modulo 19:

a+a' ~ -8 y+~'B+~'~ ~ -7 E+~'~+B'y ~ +6 ~'~ ~ 8.

B+~'~+~' ~ -6 6+~'y+8'B ~ -5 ~'~+B'6 ~ -6

This system has a solution modulo 19, which, when substituted in (6), leads to
the third congruence of (S).
Before we can use Theorem B, we must still verify that all factors are irreduci-
ble [mod 19). As there are no linear factors, the only further possible decomposi-
tion would be

xS+gx4-Tx3+Sx2-Tx+4 ~ (x2+~'x+~')(x3+~x2+Bx+y) (mod 19).

We are again led to a system of congruences (mod 19) in the undetermined coefficients
and it turns out that this system has no solution. It follows that the fifth degree
polynomial is irreducible modulo 19 and Theorem B may be applied to the factorization
of 88(x) in (5).
 CHAPTER 13
ASYMPTOTIC PROPERTIES OF THE BP

i. Asymptotic properties of BP were considered already in [53]. It was sho~m there

that, for constant z # 0 and n ÷ ~,

(2n)! znel/Z.
(i) yn(Z) ~ 2nn!

Moreover, for n > i,

(2) [yn(Z) (2n)! znel/Z[ _< (2n)! [zln-Zel/[z [ = kn(Z )[(~n)! znel/Z[,
2nn~ 2n+2n!(n-1) 2 n!
1 -2 i i
where kn(Z ) - 4(n-l) [z exp(~ - ~)[ and kn(Z ) + 0 for fixed z # 0 and n + ~.

If one uses Stirling's formula for the factorials, (i) is seen to be equivalent
to

(3) .2nz. n ~
yn(Z) ~ (--~--) el/z

Formula (3) has been generalized by Obreshkov [82] to the polynomials Yn(X;a,b) with
b = -i. If in Obreshkov's formula one makes the change of variable -2x = z needed
to reduce b to its standard value b = 2, and recalls that in his notation m = a-2,
the formula of [82] reads

.2nz.n~a-3/2 1/z
(4) yn(Z) ~ (--~-) z e .

Formula (4) has been improved further by K. DoPey [48], who proved that

(S) yn(Z) = (--e--)'2nz'n~a-3/2z

el/Z''ll- l+6(a-2)(a-l+2z-l)+6z -2 + 0(I a2j }.
24n
n

It is clear from Do~ev's work that the terms of the large bracket are only the
first few terms of an asymptotic series, which, in principle, can be computed to
any desired number of terms.
Formulae (I) to (5) are meaningful only for z # 0. In [53] the remark is made

that close to z = 0, yn(Z) behaves essentially like en2z/2, and even an error term

is computed. In fact, it is clear from (2.10) and (2.8) that if one wants to approxi
mate ynCZ) in the vicinity of z = 0, then the natural candidate is en(n+l)z/2 In-

deed, ynCZ) = l+(nCn+l)/2)z+((n-l)n(n+l)(n+2)/2.4)z 2 + 0(z 3) and e n(n+l)z/2 =

l+(n[n+l)/2)z+(n2Cn+l)2/2.4)z 2 + 0(z3), so that an elementary computation leads to

THEOREM i. For fixed n a ~ [z] ÷ 0,

yn(Z) _ en(n+l)z/2 = _ 1 n(n+l)z 2 + 0(z3).

4
 125

in particular, yn(Z) r - j e n ( n + l ) z / 2 .

This result is better than that of [53], but it seems that neither of them has
any particular significance. Therefore we shall not consider them in what follows

2nn~ z-n
and turn to discuss the approximation of the exponential e I/z by (2-~., yn(Z) (for

2nn!
z ~ 0), or, equivalently, that of e z by ~ en(Z ) (for z finite), as expressed by

formulae (i) to (5). This approximation is precisely the property of BP used in the

proof of the irrationality of e r for r rational (see the Introduction, Problem I, and
Chapter 14).

2. It is, of course, sufficient to prove (5), from which the other formulae readily
follow This proof, which is the object of Section 5, uses rather deep results,
such as Docev's estimate of the zeros of yn(Z,a) (see Theorem 10.6). For this rea-

son we present in sections 2 to 4 an independent, elementary, although somewhat

lengthy , proof of the following slightly stronger version of (i) and (3).

THEOREM 2. For constant z ~ O,

yn(Z) = c 0 z n e l / Z + Rn(Z),

(2n)!
with c o = 2nn! , and where, for Iz[ e 2/n, one ~ s IRn(Z) I < c01z] n ~(n,z) with

0 < k(n,z) <

1=l-2el/Izl
2(2n-i) (l+~(n)); here 0 < ~(n) ~ 2 and, for n ~ 00,

~a(n) = n n-3/2" , with lim an = 0.

n -> ¢~

REMARKS. From Theorem 2 one obtains (i), by the observation that

lim IRn(z)/yn(Z)l = 0. Also the error term estimate in (2) is an immediate conse-
n ,~. ~

quence of t h e bound on IRn(Z) l in Theorem 2. Finally, by Stirling's formula,

(2n) ! ( 2 n ) 2 n + l / 2 e -2n 2n+l/2e-n

~ = ,
c O = 2nn, 2nnn÷l/2e-n
so t h a t (3) f o l l o w s from (1).

3. Proof of Theorem 2. In Chapter 2, by u s e of ( 2 . 1 0 ) , we o b t a i n e d t h e r e s u l t

n (2n)' b(n) 2m (2n-m)! n!

Yn (z) = c o X h(n)z n-m, with c o = " and = m' (n-m) (2n)
m= 0 m 2nn., m ' !
 126

n -m n 2m (2nzm_) !n,.~ Z -m
Hence, yn(Z) = c0zn{ ~ z
m' X (i- (n-m) : (2n) ," F-T- }
m=0 m=0

ao -m n -m
= c0zn{el/Z _[ ~ z z
(l-f(m,n)) ~ )]
m.-~- +
m=n+ 1 m=0

= c0znel/Z - c0z n ( R l ( Z ) + R 2 ( z ) ) , say,

2mn~ (2n-m) : Also,

where f(m,n) (2n) ] (n-m) ' "

IRlCz) I Iz[ -n-1 ~ Iz] -2 ]zl -n-i i < 2[zl -n-I

<- (n+l) ! (i+ n+2 + (n+2) (n+3) "'') -< (n+l)' [z[-1 (n+l)2
l-
n+2
eor I zl > 2/~.
n -m
In o r d e r to estimate la2(z) l = ! y (l-f(m,n)) m.--]--[,
z we use the following
m=O
LEb£%L~ i. For' 0 < m < n,

(6) 0 < 1 - f(m,n) < re(m-l)

2 (2n-l) '

with equality ~br m = 0 and In = 1; the inequality on the left is strict for 3 < m <_ n ,

and on ~he right for 4 < m <_ n.

If we accept for a moment (6), it follows that, for z ~ 0,

n m(m-l) Izl -2
!e2(z) l < m~2 2 ( 2 n - 1 ) ml =

IZl-2 n-2 ~ izl-2 m~0 i~l-m 1=!-2 el/,.= I

2(2n-i) m! 0 . :< 2(2n-i) m~ = 2(2n-i)

If we set R3(z ) = Rl(Z ) + R2(z), then

iR3(z) I
-<
21~l -n-1
(n+l)! +
[zl-2el/]:!
2(2n-i) =
[zl-2el/Izl
2(2n-i) {i+
4]zl-n+l(2n_l)
e-l/Izl}
(n+l)! "

For positive u = Izl -I, the function eUu l-n increases to infinity for u ÷ ~ and also
. e .n-1
for u ÷ 0 (n > i) and it attains its minimum for u = n-l, when it equals [n_--~) .

Consequently, ]R3(z)l < [z1-Zel/Izl

2(2n-1) (l+~(n)), with ~(n) =
4(2n-i)
(n+l):
.n-l.n-i
IT ) "

One e a s i l y verifies that, f o r n > 1,

(2n+l)n . n .n-1 1 2n+l
~(n+l) = ~(n) (2n-l)(n+2) (n---~) e < 2n+3-2/n ~(n), so that, for n > 2,

p(n+l) < p(n). Also, by direct computation p(1) (= lira 4(2t-i)

F(t+2) (t-l.t-l)
--~-) = 2,
t÷l
 127

~(2) = 2/e, ~(3) = I0/3e2,...," hence, ~(n) _< 2. Finally, by use of Stirling's for-

mula, ~(n)~ 8 ~ n-3/2 f o r n + ~.

2~7~- It follows that yn(Z) = c0znel/Z÷Rn(Z) ' with

Rn(Z) = c 0 z n R 3 ( z ) a n d

(7) IRn(Z) I < c0

-
Ill n Lzl2 (-Net/Ill
2n-1)
(l+~(n))
"

I f we now c o m p a r e t h e s e c o n d member o f (2) w i t h (7), it follows that (2) is a

1 l+~(n) 1
consequence of Theorem 2, provided that 4(-~_I) ~ 2(2n-i) ' or >(n) < 2(n-l)

This obviously holds for large n, because ~(n) = 0(n-3/2); in fact, this holds for
8(2n-l)(n-l) n-I n-I
any n, provided that (n+l)., (~) < i as follows by using the explicit

formula for ~(n). By direct computation one may check that this is indeed the case
for all n > i. The proof of (I), (2) and of Theorem 2 are complete, except for the
proof of Lemma i.

4. I n tile p r o o f o f Len~a 1 we s h a l l need

LE~ 2. The inequality

(8) (2n-l- ~)(2n-2)(2n-S)...(2n-m+l) f (2n-2)(2n-4)...(2n-2m+2)

holds f o r 3 < m < n a n d is strict f o r 4 < m 5 n.

Proof of Lemma 2. For m(m-l) ~ 4n-2 the statement is trivial, because then the
left hand side is non-positive, while the right hand side is positive. For m = 3,
(8) reads (2n-4)(2n-2) f (2n-2)(2n-4) and reduces to an equality.

We now assume that (8) holds for some m with 3 < m < n and show that it also
holds for m+l. If we replace m by m+l, the first factor in (8) becomes

2n-l- (m+l)m
2 = 2n-l- m(m-l)
2 -m, and we have to verify that

(2n-l- m(m-l) -m)(2n-2)(2n-3) (2n-m+l).(2n-m) < (2n-2)(2n-4) (2n-2m+2)(2n-2m)

2 "'" - . . . .

This is equivalent to

(2n-l- m(m-l) (2n-2)(2n-3) (2n-m+l).(2n-m)-m(2n-2)(2n-3) (2n-m+l)(2n-m)

2 . . . . . .

< (2n-2)(2n-4)...(2n-2m+2)(2n-2m).

By the induction assumption (8), it is sufficient to prove that

(2n-2)(2n-4)...(2n-2m+2).(2n-m)-m(2n-2)(2n-3)...(2n-m)

( 2 n - 2 ) ( 2 n - 4 ) . . . (2n-2m+2) • ( 2 n - 2 m ) ,

or
 128

(2n-2)(2n-4)...(2n-2m+2).m < (2n-2)(2n-3)...(2n-m)'m.

This clearly holds, because the number of factors is the same on both sides of the
inequality and corresponding factors are either equal (namely the first and last),
or are larger on the right, than on the left.

m(m-l)
Proof of Lemma i. For m = 0 and m = i, f(m,n) = 1 and 2(2n-i) = O, so that both

equalities hold For m = 3, (6) becomes 0 < l-f(3,n) = 3 = m(m-l)

• 2n-i 2(2n-i) "

To complete the proof by induction, we assume that 3 < m-i < n and that (6) holds
for m-l. We show now that (6) also holds if m-i is replaced by m. The inequality

on the left is immediate, because f(m,n) = 2(n-m+l) f(m-l,n) < f(m-l,n) The
2n-m+l
inequality on the right is equivalent to

2m-l(n-l)(n-2) (n-m+l) (2n-2)(2n-4)..(2n-2m+2) m(m-l)

1 - "'" = 1 - " ~ - -
(2n-l)(2n-2)...(2n-m+l) (2n-l)(2n-2)...(2n-m+l) 2(2n-i) "

By clearing of denominators we obtain

(2n-l)(2n-2)...(2n-m+l)-(2n-2)(2n-4)...(2n-2m+2) < m(~-l____~)(2n-2)(2n-3)...(2n-m+l),

or (2n-l- ~ ) ( 2 n - 2 ) ( 2 n - 3 ) . . . ( 2 n - m + l ) < (2n-2)(2n-4)...(2n-2m+2), which holds

on account of Lemma 2. This completes the proof of Lei~la I, and with it that of
(i), (2), (3) and of Theorem 2.

S. In this section we shall give the proof of

THEOREM 3. (Do~ev [48])• For oonstant z ~ 0 ~a~d n + ~,

(9) yn(Z'a) =
.2nz,n~a-3/2
L--~) z e I/z~.
tl- 1 z -2 f"a-2"'a-l+2z-l"
~f ~ + 0(-Tj
1 ~}
' 24n 4n 4n "
n
In fact, in order to illustrate the remark made after (5) we shall actually
improve (9) to (14) with the better error term 0(n-3).

Proof of Theorem 3. From (2.27) and with the notations of Chapter 2,

Yn (z;a) = d(n) ~ {d(n)Id(n)}zk d(n) i c(n) k

n k "n = n n-k z '
k=O k=O
where
(i0) d(n)( = d(n)(a)) = 2-n(2n+a-2) (n) = 2 -n F(2n+a-l)/r(n+a-l)
n n

c(n):_
n=k k- d~n)/d~n)) = 2n-k(~)F(n+k+a_l)/F(2n+a_l).

If ~j(= ~(n))
j (j = 1 '2, • ..,n) are the zeros of yn(Z;a), then

(ii) yn(Z;a) = d(n) znp(z)

n )
 129

where
n

(12) P(~) = I] (i-~j/z).

j=l

By (I0) and Stirling's formula

d(n)= 2 -n (2n+a-l) 2n+a-3/2 e-2n-a+l l+(12(2n+a-l))-l+(288(2n+a-l)2)-l+0(n -3)

n (n+a_l)n+a-3/2 e-n-a+l l+(12(n+a-1))-l+(288(n+a-1)2)-l+o(n -3)
By routine (but lengthy) computations we obtain the following development, with
error term O(n-3):

d(n) = 2n+a-3/2nne-n{l_ 6(a-l)(a-2)+l + h(a) + O(n-3)}

n 24n i152n 2
where h(a) = 12(a-l)(a-2)(3a+7)+l. Next, by (12),
n e. n ~ r ~ or
logP( ):logl I ° - --v,
j=l j=l r=l rz r=l rz
n

where o = [ ~i. The values of the first few o ' s have been computed in Chapter
r
3 3 r

i0. -2n °2 = 4n(n+a-2)

Specifically, o I = 2n+a--------~
' (2n+a_2)2(2n+a_3) '

2
-8(a-2)(n+a-2)n Also, by Theorem 6 of Chapter i0, [ej[ -< n-l+Re a
03 =(2n+a-2)3(2n+a-3) (2n+a-4)

r
so that [~jl = O(n -r) and or = O(n l-r) Consequently,

[ ~ r ~ n cn ~ 1 c 1
- ~ I _< c r=4 ~?zl r <- --4
r=4 -rz r=4 (n[z I) r = ~ n z[ (l-(n ] z [)-1 )
3] 4 '

with a constant c, that depends only on a.

For fixed z and n sufficiently large,

oo (7
1-1nz] -1 > 1/2, SO that ] ~ ~[r < ~C 1 = O(n-3)
r=4 rz - n 3 ]z[ 4
In fact, this estimate holds uniformly in z, provided that [z[ _> n, for some fixed
n > O.

We conclude that

log P(z) =

2n 1 2n(n+a-2) --
1 +
8(z-2) (n+a-2)n 1 + 0(n-3)
-y
2n+a-2 z (2n+a-2)2(2n+a-3) z2 3(2n+a-2)3(2n+a-3) (2n+a-4)
-I (.(a-2)z-I zi2)n-i (a-2)2z -I zi2)n-2 + 0(n-3)
=z - 2 +-- + ( 4 ---
 130

By exponentiation we obtain

(13) P(z) =

el/Z{l_((a-})z-I + ~)n -I + (.(a-2)2z-I + [ ( a - 2 ) 2 - 1 ] z -2 ( a - 2 ) z -3 z -4. -2 ~ . - 3 .

4 8 + 8 + 3-T--Jn +urn )}.

Finally, we substitute (13) in (Ii) and obtain

.2nz.n~a-3/2 i/z .
(14) Yn(Z; a) = L--~-J z e gLz,a),

where

g(z,a) = l-[(a-2)(a-l+2z-l)+(l + z-2)](4n) -I

+ [(a-2) ((a-l) (3a+7)+(12(a-2) (a+l)+2) z-l+6(3a-5) z-2+12z -3)

+ (1~ + llz -2 + 3z-4)] (96n2) -I + O(n-3).

If we combine in (14) the terms of order n -2 with the term 0(n-3) , then we
obtain precisely Do~ev's formula (9). In the particular case a = 2, (14) reduces to

. 2 n z . nVZ
yn(Z) = I.---~-) ~ e l / Z { 1 - ( 1 +z -2) 1 1 + llz-2 + 3z -4) - -1 + 0 (.~_)},
4~-- +(T~
96n 2 n

from which (i) and (3) immediately follow.

 PART IV
CHAPTER 14
APPLICATIONS

i. In the present chapter we shall discuss some of the problems, in which BP have
found applications and which, in fact, led to their study. Three of these have al-
ready been mentioned in the Introduction and sections 2-S of the present chapter are
mainly an expanded version of the earlier presentation. Somewhat unfortunately for
this exposition, those problems present their own difficulties; while these can be
overcome by classical, well established methods, even the short sketches of those
procedures given here, may well divert the attention of the reader from the role
played by the BP themselves in the solutions of the respective problems.

2. The irrationality of the exponential function e c for rational c. While the

irrationality" of e is an almost immediate consequence of its usual series expansion,

the irrationality of e c for rational c is less obvious. We shall prove it here, by

using results of Chapter 8.
We recall that the polynomials Pn(X) collect the terms of Yn(X) with powers of

x of the same parity as n, while Qn[X) is the sum of the other terms of Yn(X). Let

x be a fixed, positive, rational number and set x/2 = s/r, with coprlme integers s

and r, so that e 2/x = e r/s. It is clear that rnPn(2S/r) = An and rn-l~[2s/r) = Bn

are integers. Also, ~ C x ) = an-I

[n) xn-I + an-3
(n)x n-3 + "'" > an-lX
(n) n-i , so that, in par-

ticular, Qn[2S/r) > a ~ 2n-l(s/r)n-I = 2(n~)(2n)![~)'s'n-iand

= rn-l%(2s/r) > ~(2n)! sn-l. If er/S would be rational, then er/S+l a

Bn er/S_i = ~ , with
a,b coprime positive integers would follow. By Corollary 8.4
Pn [2s/r) a A 1 r 2n-I
0 < Iqn(2S/r ) - ~ I = Ir-~nn - ~1 < Qn(2S/r)Qn+l(2s/r) - BnBn+l ,

or
0 < [bAn-arBn[ < r2nb < r2n(2b) (n+l)'. < 2br2n (n+l)'.
Bn+ 1 sn (2n+2)! (2n+2)!

2
.r e.n+l..
By Stirling's formula, the right hand term is less than (--~) [l+en)(lim en=0);
er n-~=

hence, for sufficiently large n, the positive integer IbAn-arBn[ is less than one,

which is impossible and shows that er/s cannot be rational. In particular e itself
and all its integral positive and negative powers are irrational.

S. The irrationality of ~. We shall use again results of Chapter 8 and follow, in

the main, the presentation of Siegel [55].
 132

First we obtain a new representation for Rn(X), defined by (8.6). We differen-

dn+l
tiate (8.6) n+l times, observe that ~dx @n(X/2) = 0, use Leibniz's formula for

the derivative of a product and obtain

dn+ 1 dn+l n+l r+l n+l -r (r) ex
(i) - - Rn(X ) dx n+l {@n(- x/2)eX} = r=0~(-I) ( r )2 en (- x/2) = eXs(x);
dx n + 1
n
h e r e S(x) = [ a.x j is a polynomial in x of degree at most n. We now observe that,
j=O J
dn+l
on t h e one h a n d , by ( 8 . 6 ) , d x - ~ Rn(X) contains only powers of x with exponent at

least n; on t h e o t h e r h a n d ,

eXs(x) = ao+(al*aO)x+...+(aj +aj_l+...+ao/J!)xJ+...+(an+an_l+...+ao/n!)xn+ ....

It follows that a 0 = a I = . . . = an_ 1 = 0. As for an, only the term of (i) with

r = 0 contributes to it and it is immediate from (i) that a n = (-l)n+12 -n, so that

S(x) = (-l)n+12-nx n and ~ddEn+l Rn(X ) = (- l)n+12-nxneX . By an (n+l)-fold integration

eX (x-t)ntnetdt"
from 0 to x it now follows that Rn(X ) = (-l)n+l
2nn! Jo Finally, if we

replace t by tx, we obtain the desired new representation

2n+l
(2] Rn(X ) = (_l)n+l x ii (l_t)ntnetXdt.
2nn! o
For x = iT, in particular,
2n+l 2n+l
IRn [iw) i f ~ , ,i (l-t)ntndt
~o = ~ n , B(n+l, n+l),
2nn! 2 n.

r (u) r (v)
where B( u , v ) F (u+v) is Euler's Beta function. Consequently,

2n+l
IIPRn(iW)l < ~ n! , or, hy Stirling's formula,
- 2n (2n+l)!

2 /f
2n+l re__~n+l l+c n = (~)n+l (l+en) ( lim e n = 0).
(S) IRn(i~)I <_ ~ 2 n <4n, ~ ~e n ÷

Let us assume that w2/4 = a/b is rational,with a,b coprime positive, rational inte-
gers. B Z setting x = wi in (8.6), we obtain, with ~ = [n/2],

Rn(i~) = @n(Wi/2)-@n(-~i/2)e wi = @n(Wi/2)+en(-~i/2) = 2 ~ a (n) (wi/2) 2k =

n-2k
k=0

2 ~ (-l)ka n = 2 ~ (-i)
k=0 k=0
 133

with a,n,r~ given by ( 2 . 8 ) . By Theorem 3.1 i t follows t h a t Ib~Rn(i~)[ =

n-2k

12 Z (-I) ka(n)
n-k akb~-k I = N is a non-negative integer. In fact, N > 0, because (2)
k=0
shows that for x = ~i the integrand has the positive imaginary part (l-t)nt n sin ~t
(0 < t < i). On the other hand, by [3),

n/2 ~2e n+l /8 (l+en) < .w2ebl/2 . n + l ~ _ l+en < 1

0 < N = ]b~RnfiW) l < b (~) ~e - ~--~) ~e

for sufficiently large n and this is not possible for any integer N. This shows
2 . 2 .
that the assumption of the rationality of ~ is not tenable, ~ is irrational and
this holds, of course, for ~ itself.

4. Solution of the wave equation in spherical coordinates. This problem is, of

course, classical and the reader may want to consult, e.g. [55] for a traditional
solution. The equation to be solved is
2
(4) Au ~ 1___ 3__~u
c2 ~t 2
where A is the Laplacian and c is the speed of propagation of the waves. In spheri-
cal coordinates

A = ---~- + ? ~-~ + -- + cot e ~ ) + 2 8¢2

3r r r sin2O

It is sufficient to consider only monochromatic waves, i.e., waves of a single fre-

quency; indeed, the general case is reducible to the monochromatic one by ordinary
Fourier analysis, that means by superposition of a countable number of monochromatic
waves.
In order to solve (4), we use the method of separation of variables, i.e.,
we set

(s) u = R(r)O(O)¢(¢)T(t),

where each factor depends only on a single independent variable. Specifically, r

is the distance of a point P from the origin O, 8 is the angle of OP with a fixed
"polar" axis [0 < 8 ~ =), $ is the angle that a plane through P and the polar axis
makes with a fixed plane through the polar axis, and t is the time variable.
If we substitute this in (4) and divide by u, we obtain

R" 2 R' 1 @' 1 @" 1 @" 1 T"

(6) ~-- + ~ ~-- + -~-~-- cot 8 + ~ @- + 2 . 2 @ 2 T
r r r sln @ c

Here the right hand member depends only on the time t, while the left hand member
stays unchanged when t varies; hence, both sides of (6) are constant. In order to
obtain a periodic solution, this constant must be negative and we denote it by -k 2.
 134

Hence, T" + k2c2T = 0 and, with proper choice of the origin for time, and after the
suppression of a multiplicative constant (that would anyhow be absorbed into a pro-
duct of similar constants - see (S)) we obtain T = cos kct. We proceed in the same
way with the separation of the other variables. First, we obtain

2 . 2A.R" 2 R' O'-- 1 O' ¢"

(7) r sln ~t~-- + ~ ~-- + O + ~ cot O. 7 + k2} = - ¢-- = ~"
r
In order to obtain a solution $(¢) that is single valued and periodic with period
2
2~ (as it has to be), it is necessary to take ~ = q , with q an integer, so that

¢(~) = a cos q~ + b sin q¢. Next, (7) with ~ = q2 leads to

q q
2
[8) r 2 R" R' k2r 2 e" C' q } = X.
~-- + 2r ~-- + = -{~-- + cot e ~-- + sin2 @

To solve the last (ordinary) differential equation of (8), set cos e = x, @(0) =

F[cos e) = F [ x ) = (l-x2)q/2v(x). Then V(x) satisfies the differential equation

(9) (l-x2)V '' - 2(q+l)xV' + (~-q2-q)V = 0.

Equation (9) has, in general, only solutions by (often divergent) series; it admits,
however, pol)momial solutions if and only if X = n(n+l), for some integer n. In

dq
that case, its solution is V(x) = Ln(X), where Ln(X ) is the Legendre polynomial
dx q
of degree n. The proof is immediate, by q-fold differentiation of the differential
equation for Legendre polynomials:

(l-x2)L~-2xL~+n(n+l)en = 0. It follows that @(O) = sinqo{ dq Ln(X)} x cos e

dx q = ,
a function usually denoted by Pq(cos e) the associated Legendre function. It is
n
clear that for q > n, Pq (cos e) = 0 identically. We now solve the remaining ordi-
n
nary differential equation. By (8) with ~ = n(n+l), this is

(I0] r2R '' + 2rR' + (k2r2-n(n+l))R = 0.

Equation (i0) may be transformed into Bessel's equation. We prefer, however,

to proceed differently. In order to obtain polynomial solutions we set

x = (ikr) -I and R(r) = (kr)-le-ikry(i/ikr) = ixe-Xy(x), so that (i0) becomes

(ii) x2y '' + 2(x+l)y' - n(n+l)y = 0.

In (ii) we recognize equation (2.11) of the BP Yn(X). The result so far is that (4)

admits solutions of the type

 135

(12) u(r,¢,8,t) = (ikr)-le-ikryn(I/ikr) (aq cos q¢+bq sin q¢)P~(cos @)cos(kct),

a fact that can be verified by direct substitution of (12) in (4), or (5).

With (12) also
n
U(r,O,¢,t) = ~ (ikr)-le-ikryn(i/ikr) ~ (an,qCOS q¢+bn,qSin q¢)P~(cos O)cos(ckt)
n=O q=O

are solutions,for any choice of the constants an, q and bn, q. One can use this fact,
in order to choose the constants so as to satisfy initial and boundary conditions.
Let us assume that at t = O, U(r,@,¢,O) = F(r,@,¢) is given. Then, if we set
(irk) -I = x and x-iF((ikx)-l,o,¢) = Fl(X,@,¢), we obtain
n
(13) ~ e-i/Xyn(X) X (an,qCOS q¢+bn,qSin q¢)P~(cos O) = F(x,O,¢).
n=O q=O
We now multiply both sides of (13) by Ym(X)e"I/x and integrate around the unit

circle. By taking into account Corollaries 4.S and 4.7 we obtain

m
(-l)m+l 2m+12 ~ (am,qCOS q~+bm,qSin q¢)P~(cos 0) = ~m(O,¢), where
q=O

i / Fl(X,O,¢)Ym(X)e-I/Xdx. We rewrite this as

~mCO'¢) = ~ l~l--i
m
(am,qCOS q¢+bm,qSin q¢)Pq(cos 0) = Km(O,¢), where Km(O,¢) =
q=O

(-l)m~l(m + i/2)~m(@,¢ ). By using the orthogonality of the trigonometric functions

and of the associate Legendre functions, one obtains (see, e.g. [SS]) explicit values
for am,q and bm,q :

4~am, 0 = (2m+l) #2~ /=0 Km(e,~)Lm(COS O)sin Oded¢; and for q # O,

0

= (m-cO :
2~am, q (2re+l) (re+q)! /o2~ ivo Km(e'¢)Pq(c°s O)cos q¢ sin 0 dOd¢,

(m-q) :
2~bm, q = (2m+l) (m+q)' fo2~ ITO Km(O'¢)Pq(c°s O)sin q¢ sin @ dOd¢.

Here Lm(COS 8) = PmO(cos @) is the Legender polynomial of degree m. Replacing again

m by n, the function

U(r,e,¢,t)
(14)
n
-i -ikr
X (ikr) e Yn(i/ikr) X (an,qC°S q¢+bn,qSin q~)P~(cos @)cos(kct)
n=O q=O
 136

is the required solution for the wave component corresponding to the given k. If
there are several frequencies present, then the superposition of solutions of the
type [14) will give the complete solution.

5. The infinite divisibility of certain probability distributions.

Let f(x) be a probability density function. Then it is customary in probability

theory to call its Fourier transform ¢(t) = I ~ f(x)eitXdx, the "characteristic

function" of f(x). Given a probability density f(x) and its characteristic function

¢[t), let us set c m [ t ) = {¢(t)} I/m. The question arises: is it true that for every

integer m, ~m(t) is the characteristic function of some probability distribution?

In other words, given any integer m, is it true that there exists a non-negative,
Lebesgue integrable function g(x) (= gm[X)), such that ~m(t) = f ~ g(x)eltXdx?

If such a function g(x) exists for every m, we say that the probability density
function f(x) is infinitely divisible. The same terminology is used in the case of
a [not necessarily differentiahle) distribution function F(x). For a more detailed
discussion of this topic see [~7]. Much work has been done on the investigation of
conditions that insure infinite divisibility (see, in particular, [45] and [19]).
Recently, Thorin [SS] and Bondesson [7] have determined large classes of distribution
functions that are infinitely divisible. In fact, the results of [7] contain as
particular cases those presented here; however, due to the generality of the problem
considered in [7], the methods used there are rather deep. Here we shall restrict
ourselves to some simple, specific cases, namely the Student t-dlstributions of an
odd number of degrees of freedom. The corresponding density function is
2
x )-k
(15) f2k_l(X) = Ck(l+

r(k)
with the normalizing constant c k = ¢[2k-I)~ F(k - 1/2)

2
REMARK. Let us observe parenthetically that if Xm is a chi-square random variable

with m degrees of freedom, then the infinite divisibility of fm(X) implies also
2-i
that of (Xm) ; we shall prove this infinite divisibility, as stated, for all odd

m = 2k-l. A proof valid for all m (in fact, for all real m > 0) can be found in
[56]. Still more general results are obtained in [36].
The present proof is based on the following Theorem A of Kelker [65]. We re-
call that a function ¢(t) is said to be completely monotonic on an interval I, if

on I the function and all its derivatives satisfy the inequality (-l)n¢(n)(t) > 0
[n = 0,i,2 .... ).
 137

THEOREM A. The Student t-distribution of 2m degrees of j~eedom is infinitely divi-

Km_ 1 (,~-)
sible, if and only if the function Cm(t) = #t Km(/t) is completely monotonic on

[0,~).
Here Km(U ) is the Bessel (or Macdonald) function discussed in Chapter 2.

Kelker [65] and Ismail and Kelker [62] proved the complete monotonicity of
Cm(t) for some small half-odd integral values of m, from which the infinite divi-
sibility of f2k_l(X) as given by (15) followed for the first few integral values of

k. They also conjectured that ~m(t) is completely monotonic for all real m, from

which follows, in particular, the infinite divisibility of f2k_l(X) for all integers

k. For further results on this problem, see besides [55] and [56], also [35], [36].

It is clear that if ¢(f) = I~ g(u)e-tUdu, with g(u) > 0 for 0 < u < ~,
0

then ~(t) is completely monotonic. That this sufficient condition is (essentially)

also necessary is less obvious and is contained in the following theorem of S. Bern-
stein:

THEOREM B. (S. Bernstein; see [6S]). The function ¢(t) i8 completely monotonic, if
and only if it is the Laplace transform of a (not necessarily j%nite} measure, i.e.

if are onlu if it is of the Form ¢(t) = f~ e-tUdF(u)

In particular, if F(u) is differentiable, so that F'(u) = g(u) exists, then

one has ¢(t) = Ie g(u) e-tUdu, with g(u) > 0 on 0 < u < ~. We shall denote the

Laplace transform of g(u) by i(g)(t) and the inverse transform of ¢(t) by i-l(¢)(u).
We shall prove

THEOREM i. For every odd integer n ~ i, set ¢(t) = On_l(tl/2)/On(tl/2 ) and let

8j (= 8~n)) (j = 1,2 ..... n) be the zeros of the BP @n(Z); then, if g(u) = L-l(¢)(u),

we have
2
(16) g(u) = (~u)-I/2- 2~ -I/2 i e ~.u
J I~ e _v2dv,
j=l -~j~-
and g(u) is completely monotonic; in particular, g(u) ~ 0 on 0 < u < ~.

COROLLARY I. For every odd integer n, fn(X) (as defined by (iS) with n = 2k-l)

and (X~) -I are infinitely divisible.

Proof of Corollary i. By (S.l), Kn+i/2(z) = (~/2z)i/2z-n-i/2e-ZOn(Z); hence

Kn_i/2(z) @n_l (z)
=
ZKn+I/2 (z) 0n(Z)
 138

and
Kn_l/2(tl/2) On_l(t 1/2)
¢n+l/2(t) = tl/2Kn(tl/2) en(t 1/2)

Now the Corollary follows from Theorem 1 on account of Theorem A and the Remark
preceding it. It remains to prove Theorem i.

On_lCtl/2) n
Proof of Theorem i. By ( i 0 . 1 4 ) , = 1 ; hence,
en(tl/2 ) j=l 8j (Sj-t 1/2)
n n
= I i-i 1 - - Z I i-i I
g(u) = /-ic¢)(u) j!l ~j 8j-t I/2 j=l ~ j tl/2+C-Bj )
By Theorem i0.5 and Corollary 10.2, Re 8j < O; hence, Re(-8j) > 0 and the last in-

verse Laplace transform may be computed. In fact, it is knowm (see e.g., (29.3.37)
in [~]) and we find
2
n
1 1 + 8je 8.u
] erfc(-Bjul/2)}, where erfc(-Bju I/2) =
g Cu) : - l
j=l ~ { ~
2
2 - 1 / 2 /-Sj~ e-v dv. By using also (10.2') we obtain (16). It remains to show

the complete monotonicity of g(u). In fact, we only need the positlvity of g(u),
in order to apply Theorem B, but it is easier to prove the stronger statement.
Indeed, let
n -1/2 2 -i
@(x) = w-l{x-i/2 + [ 8ix (x+Bj) }.
j=l

We verify by direct computation, or on hand of tables of Laplace transforms

(see, e.g., [I], (29.3.4) and (29.3.114)) that i(@)(u) = g(u). If we set ~[x) =

~xl/2@(x), then by Theorem B,g(u) is completely monotonic, provided that

n n
,(x) o l÷ Z o. , o . . ,cx) o p(x)/qc.), with qCx) - I I (.÷5)." a

j:l j=Z
polynomial with real coefficients.
As just recalled, we know from Chapter I0 that none of the 8j's is purely

imaginary; hence, q(x) has no positive zeros and, consequently, does not change
n
sign on [0,~). As q(0) = ( ~ 8j) 2 = {(2n)'./2nn!} 2 > 0, one has q(x) > 0 for
j=l

0 < x < ~. Also

n n
(17) p(x) = q(x)+ j=IZ 8j [k~j
[ (x+6k2) = xn + j=l
~" Yjxn-"J
 139

is a polynomial of exact degree n; it also has only real coefficients (namely symme-
tric functions of the 8j's) and so is real for real x. For x + ~, ~(x) =

p(x)/q(x) ÷ i, so that, by q(x) > O, also p(x) > 0 for large x. We now claim the
validity of

n
LEM~ i. p(x) = x .
Assuming the Lemma for a moment, p(x) > 0 for 0 < x < ~; hence, ~(x) > 0 on
0 < x < ~, so that g(u) is indeed completely monotonic and the proof of Theorem 1
is complete. It only remains to justify the Lemma.

Proof of Lemma i. From q(O) / 0 follows that ~(x) and p(x) vanish (if at all) to
n
the same order at x = O. However, by (10.2') 4(0) = l+ [ B? 1 = 0 and
n j=l l
~(m)(0) = ~ 8j (2m+l) = 0 (m = 1,2 ..... n-l), so that ~(x) has a zero of order at
j=l

least n at x = 0. This shows that p(x), of degree n, also has a zero of order n

at x = 0, so that in (17) all yj = 0 (j = 1,2 ..... n) and pox) = x n, as claimed. With

this the Lemma is proved and all assertions of this sections are justified.

6. Electrical networks. Let us consider a [perhaps very complex) electrical net-

work. It can be decomposed into a finite collection of loops, each one of which
may contain resistors, coils [self-inductances) and condensers [capacitors). If a
closed loop contains no source of electrical potential, then the Ohm-Kirchhoff laws
state that the total difference of potential around the loop vanishes. We recall
that if the current i = i(t) is measured in amperes then the difference of potential
between the terminals of a resistor of R ohms equals Ri(t) volts, while that between

di (t)
the terminals of a coil of self inductance of L henrys equals L ~ volts, and

that between the terminals of a condenser of capacity C equals C -I I i(t)dt volts.

If the loop is closed by a source of potential of v(t) volts, then the total drop
of potential around the loop equals v(t).
To "solve" the network means to determine the currents in all branches and the
potential (voltage) at all terminals (or "nodes"), with respect to some fixed poten-
tial, usually the "ground". In order to do that it is necessary to know also the
initial conditions, i.e., the value of one such current and of such a difference of
potential at, say, t = O. We assume that in each loop, R, L, and C are "lumped",
i.e., concentrated, rather than distributed and that they are constant in time. We
consider, in particular, "passive" networks, that is networks without internal
sources of energy. They will have,in general (see Fig. i) two pairs of terminals
("ports"). To the first one we may connect an outside, known source, that applies
there a given, variable potential el(t), the input signal. The second port may be
 140

connected, e.g., to a large resistor, say p, and we are interested in the output
potential e2(t) between its terminals, or, equivalently, in the output current

i2(t ) = e2(t)/p.
lnl

il(t) i2(t)
Rn3~~l C , ~ C n 2
: I ÷

e~(t 2( p
A

Cn3 ~ / ~ %2

Fig, 1 Fig. 2

A typical loop, say, the n-th loop may look like Fig. 2.
If we write the Ohm-Kirchhoff equations for the n-th loop, this has the form
of an integro-differential equation
• d .

(18) J[ {Rnjlnj(t)+Lnj ~ Znj(t) + ~Cn. I inj(t)dt} = en(t)

(en(t) = 0 if the loop is passive).

In order to solve this system, we replace each equation by its Laplace trans-
form. This means that we multiply each term by e-st (s = ~i~, a complex variable)
and integrate with respect to t (= time) from t = 0 to t ~. In this way each =

integro-differential equation becomes an algebraic equation in the new variable s.

We s e t I n j ( S ) = I~o Z" n j ( t ) e - S t d t (whenever p o s s i b l e we s h a l l s u p p r e s s t h e s u b s c r i p t s )

for the transform of a given current i(t) = inj(t) in a given branch j of the n-th

loop and E(s) = I ° e ( t ) e - S t d t f o r t h e t r a n s f o r m of a d i f f e r e n c e of potential e(t),

between two s p e c i f i e d points (or n o d e s , o r t e r m i n a l s ) of t h e n e t w o r k . A term Ri(t)

becomes R I ( s ) , Li'(t) becomes s L I ( s ) , and C-1 / i ( t ) d t becomes ( C s ) - l I ( s ) . For t h e

n-th loop, the transform of equation (18) becomes

(19) ~ { R n j + S a n j + ( S C n j ) - l } I n j ( S ) = En(S ) (En(S) = 0 i f e n ( t ) = 0 ) .

J
The problem o f " s y n t h e s i s " o f a network t h a t we a d d r e s s now c o n s i s t s in divi-
s i n g one t o be p u t i n t o t h e " b l a c k box" o f Fig. 1, c o n n e c t e d t o t h e f o u r t e r m i n a l s ,
so t h a t , given a certain t y p e o f i n p u t , we s h o u l d o b t a i n a d e s i r e d t y p e o f o u t p u t .
So, e . g . , we may want t h a t , i f a t t h e i n p u t we a p p l y a l a r g e number of s u p e r p o s e d
potentials, each one o s c i l l a t i n g with a different frequency, all but those within
a g i v e n , n a r r o w band s h o u l d be s u p p r e s s e d , w h i l e t h o s e w i t h i n t h a t band s h o u l d be
 141

collected at the output with a minimum of distortion. For instance, at the input
we may collect through an antenna the e(t)'s due to many broadcasts, but at the out-
put we want to eliminate all, except one, and this one with as little distortion as
possible. Such a network, called a filter, would give us, ideally, an output like
that of Fig. 3, with the "cut-off frequencies" ~i and ~2" Often, with no particular

loss of generality, we may set ~i = 0 and

l 1 then speak of a low-pass filter.

the purpose is to reproduce the input as
Sometimes

~1 °~2
faithfully (distortionless) as possible at
Fig. 3
the output. Then we speak of a time-delay
network.

In view of the linearity of the Ohm-Kirchhoff equations and of the Laplace

transforms, also the relations between the transformed input and output currents and
potentials, ll(S), I2(s), El(S ) and E2(s ) are linear. In case one connects at the

output a large resistor, in order only to collect there e2(t), the network may be

considered a one-port (two-terminal) one. The relation between ll(S ) and El(S )

being linear, there exists a function Z(s), called driving point impedance function,
such that El[S) = Z(s)II(S ).

Similarly, E2(s ) and El(S ) are related hy E2(s) = T(S)El(S ) (T(s) = transfer

function). Both, Z(s) and T(s) are obtained as solutions of linear equations of
the form (19), so that both are rational functions of s. In many cases one is
interested in Z(s), or in T(s), for s = i~ (m = 2~f, f = frequency, T = f-I = period)
purely imaginary; in that case one speaks of Z(iu) simply as the (complex) impedance.
The following properties of Z(s) may be proved (see, e.g. [60]): If the net-
work is passive and consists only of resistors, coils and condensers, then Z(s) is
real for real s, Re Z(o+im) ~ 0 for o ~ 0, with Re Z(c+im) = 0 possible only for
a = 0. Any function with these properties is said to be a p.r. [positive, real)
function; it maps the closed right half-plane into itself, so that any purely ima-
ginary boundary points of the image can have as inverse images only points also on
the imaginary axis. If Z(s) is p.r., then so is Z(s) -I and also the composition of
Z(s) with any other p.r. function. Neither zeros, nor poles of Z(s) may belong to
the open right half-plane and if a pole is purely imaginary, then it must be simple,
with positive residue. The difference between the degrees of numerator and denomina-
tor of Z(s) cannot exceed one. Many of these properties are shared by T(s), hut not
the last restriction.
If a polynomial has the property that it has no zeros in the right half-plane,
it is called a Hurwitz polynomial. By what precedes, both, the numerator and the
denominator of Z(s) (and also of T(s)) must be Hurwitz polynomials. The following
Proposition holds:
 142

PROPOSITION I. The ratio of the s~n of the even powers to the s~n. of the odd power8
of a Hurwitz polynomial is a p.r. function.

The converse of this proposition is not quite true; indeed, if H(s) is a Hurwitz

polynomial, then K(s) = (s2-a2)H(s) is not Hurwitz; nevertheless, the ratio of the
sum of its even powers to that of its odd powers is the same as for H(s), and hence
is a p.r. function. We also remark that, after the elimination of factors that are
common to the sums of even and of odd powers, the even and odd "parts" of a Hurwitz
polynomial are themselves Hurwitz polynomials, as numerator and denominator of a p.r.
function.
If we now consider the ideal output like the one in Fig. 3, it is rather clear
that continuous functions will not lead to it. We may try to obtain an acceptable
approximation to it by continuous functions, by settling for a graph of one of the
two shapes of Fig. 4(a), or (b).

(a) (b)

Fig. 4

We face, however, still another difficulty. As seen, passive networks, consisting

only of resistors, coils and condensers lead only to rational functions for Z[s) and
for T(s). However, the realization of an output like those of Fig. 4., may require
for Z(s) (or T(s)) a transcendental function, say F(s). In this case we may have
to approximate F(s) by a rational function of s, which, in addition, will have to
be also p.r.; this then can be realized by an R-L-C network.
Indeed, once Z(s), or T(s) have been determined, with the restrictions mentioned,
some very simple rules tell us how to realize the corresponding network, i.e., how
to find the geometry of the network and how to compute the numerical values of the
resistors, coils, and condensers to be used.
We shall discuss here only how to determine T[s) (the procedure for Z(s) is
similar) and ignore the "hardware problem". The interested reader can find exhaus-
tive treatments of this problem, i.e., in [ST], or [60]. For the "Bessel case",
among others, an almost automatic procedure is outlined in [I01]; see also [64].
We start with the consideration of an ideal network that introduces no distor-
tions, only a fixed delay. This means that, given e I = el[t) for t > 0, e I = 0 for

t < 0, we want to obtain an output e2(t ) = el(t-t0). By taking Laplace transforms

 143

we obtain E2(s) = f~o e2(t)e-Stdt = i~o e I (t-to) e-Stdt =

e-st° fo -s(t-t°) = e-St° ~ -st o f~o

el(t-to)e d(t-to) /-t el(t)e-Stdt = e el(t)e-Stdt =
o
-st
o
e El(S) •

Let us take, for simplicity, t o as unit of time (i.e., set t o = i); we shall

indicate later the modifications needed in a different time scale.

-s -s .
We have obtained E2(s)/El(S) = T(s) = e As e is not rational, one could

attempt to approximate it by partial sums of its Maclaurin series. However, not only

the partial sums of e -s (l-s, l-s+s2/2 etc, have obviously zeros in the right hand

plane), but also the partial sums of e s with more than 4 terms have zeros with posi-
tive real parts (see [37]), hence, are not Hurwitz polynomials. Such a transfer
function cannot be realized by an R-L-C network. On the other hand, we know from
Chapter i0, that @n(Z) is a Hurwitz polynomial and from Chapter 13 that

anl@n(S ) (an = (2n)!/2nn:, as in (2.8)) approximates e s. This leads us to consider

s
a succession of transfer functions Tn(S ) , that approximate e :

(20) Tn(S ) = anenl(s) = an(@n(S)e-S)-le -s = An(S )'e -s

Formula (20) puts into evidence the "distortion factor" An(S) = anSnl(s)e s. This

approach seems to have been discovered by W.E. Thomson around 1949 (see [107] and
[108]), by the study of certain multistage amplifiers that gave desirable outputs.
The corresponding transfer functions led him to the BP (at that time not yet so
named) and, specifically, to the recursion relation (3.5). Implicit in his work
seem to be also Burchnall's relations (10.2), (10.2'). Thomson also tabulates the
zeros of @n(S) for n = 1,2,...,9 to four decimal places (Table 1 in [108]).

If we had A(s) = i, then T(s) = e -s would be the ideal transfer function, that

reproduces the input function faithfully at the output with only the fixed delay 1
(or t o in the general case). In fact, An(S ) ~ I; specifically, for s = i~, or

imt
rather s = i~t ° if t o ~ i, An(i~to) = an@ ~ ici~to) e o = An e iE , say, with A n =

IAn(imto) I. For s u f f i c i e n t l y large n, ]i~tol is negligible with respect to an, so

i~t
that indeed, An(i~to) ~ e o, IAnl ~ 1 and c = mt o. In other words, the filter re-

produces faithfully at the output the variation of the applied potential el(t), with
only a fixed time delay t o .
 144

On the contrary, if ~ is so large that On(imt o) = z.nwn tno , then

An an/~nt ~ and decreases to zero as ~ -* =. In other words, high frequencies (and

what here "high" means depends on n) are practically not transmitted, so that the
filter can be made to work (by proper choice of n and by the realization of the net-
work corresponding to that Tn(S) ) as a low-pass filter.

What happens for intermediate values of ~? At this point it is perhaps worth-

while to remember that T(s) is the result of a Laplace transform so that it should
not be considered as the real value of the ratio e2(t)/el(t). Nevertheless, as seen

in the two extrema cases, it does convey much information on the dependence of e2(t)

on el(t ) . With this caveat in mind, we now proceed to determine the exact value of

An(S)-

By (S.I), On(S)e -s = /2~ sn+i/2Kn+I/2(s). By (9.6.4) and (9.1.4) of [3] we

obtain

On(S)e-s = A ~ sn+i/2(- ~i~i.i-(n+I/2))H~/2(-si

)
= _i-(n-1/2) ~ sn+l/21I(~!.~(-si)
n l/z

= _i-(n-i/2) ~ sn+i/2 i j • .
sin (n~+~/2) _(n+i/2)(-sz)-12n+iJn+i/2(-si)).

For s = i~, in particular,

0n(i~) e-i~ = _ ~ (-])n(i~)n+I/2i-(n-3/2) ( J _ ( n + i / 2 ) ( ~ ) + ( - l ) n
( - i ) Jn÷I/2 (~))

~-~ n+i/2((_l)n J
= ~ -(n+i/2)(~)-iJn+i/2(~))

and

n
a e
-im ,
(21) Tn(im ) = mn+l /~___
2~ {(-l)nJ-(n+i/2)(~)-iJn+i/2 (~)}
a result already obtained by L. Storch (see [106]). If to ~ I, we have to replace
everywhere m by ~t .
o

From (21) immediately follows

a
n
(22) A =
n n+l{ ~ 2
~-~ (J_(n+I/2)(~) +J +i/2(~)} I/2 '

also found in [I06].

Formulae (21) and (22) permit one to compute the loss (usually in decibels),
the phase delay and other characteristics of the network. For these topics we have
to refer the reader to [106], [64], [i01], [I15] and [116].
 145

It is of some interest to observe here that the vanishing of many of the sums
of odd powers of zeros of the BP (explicitly quoted in [106] and in [108]) is the
fundamental reason for the favorable characteristic of the networks based on BP,
that earned it the name of maximally flat delay network.
For whatever interest it may have, the general t~)e of network that realizes

a transfer function Tn(S ) = (an@nl(s)e s) is:

Fig. 5

7. Inversion of Laplace transforms. As already pointed out by Krall and Frink

(see, e.g., their remark on p. 106, after (23) in [68]), there seems to exist sur-
prizing analogies between the BP and the Legendre Polynomials. One more instance
of such similarity is the following application, that seems to attract increasing
attention.
We recall that the Laplace transform F(s) = f= f(x)e-SXdx is inverted by the
o
formula

(23) f(x) = ~ 1 fc+i~

c-i~ F(x)eZXdz

valid for Re c sufficiently large.

Often the function F(s) is of a nature that precludes an integration of (23) in
closed form. In these cases one is led to m m m r i c a l integrations.
In the case of a path of integration along the real axis, one favorite method
is that of Gaussian quadratures based on n nodes. These formulae use the zeros of
the Legendre Polynomials and are exact for polynomials of degrees up to 2n-l. In
an analogous way, in the present case, a method of Gaussian quadrature has been de-
vised, with the zeros of the Legendre Polynomials replaced by the zeros of generali-
zed BP. The method is due to H.E. Salzer (see [99] and [i00]) and has later been
elaborated also by other mathematicians (see [70], [102], [103], [69], [87]). Here
we can only sketch the basic ideas and refer the interested reader to the original
papers. First, following [99], we simplify (23), by setting zx = s, and obtain

(24) xf(x) = ~ 1 icx+i~

cx-l~ eSF(s/x)ds = ~ 1 fc'#i~
c'-i~ eSG(s)ds.
 146

Here G(s) depends, of course, also on x, but, for simplicity this dependence will
not be emphasized by the notation. Also, we shall write again c, rather than c'
for the (rather arbitrary) abscissa of integration and denote an integral taken
along a parallel to the imaginary axis, at the abscissa c, simply by f(c)" Next,

we recall (see, e.g. [15], p.128) that F(z), being a Laplace transform, has to
vanish for z ÷ ~ and satisfies also some other conditions. In many cases of inte-
-i
rest it is either a polynomial in z , without constant term, or at least can be
well approximated by such a polynomial. In analogy with the real case, the quadra-
ture formula to be obtained will be exact, if G(s) is a polynomial without constant

term in s -I , say P2n(s-l), of a degree not in excess of 2n, where n is the number

of terms in the formula.

We t h e r e f o r e assume t h a t G(s) = P2n(S -1) i s such a p o l y n o m i a l and s e l e c t (to

start with, arbitrarily) n distinct points s 1, s 2 . . . . ,s n, t o which we a d j o i n a l s o

n+l i)/ n+l ~l_s 1),

Sn+ 1 = ~. We now observe that if L!n+l)(s-l)1 = I I (s-l-sk ! I (s k then
k=l k=l
k#i k~i
L~ n+l)(skl ) = 6ik (Kronecker delta), so that the Lagrange interpolation polynomial

L(n+l)(s -I) of degree n+l, that coincides with P2n(S -I) at all points sk (also at

Sn+ 1 = ~, where both vanish) is

n+l
L(n+l)(s-1) [ L!n+l) -1 -1
= 1 (s )P2n(Si ).
i=l

It follows that the polynomial P2n(s-l)-L (n+l)(s-1), of degree 2n, contains the
n
factor s-lpn(S-1), where Pn(S -1) : I I (s-l-ski) • Hence, P2n(S -1) = L ( n + l ) ( s -1) +
k=l
s -lpn(S-1)rn-l(S-1) , w i t h a p o l y n o m i a l rn_ 1 ( s - l ) o f d e g r e e at most n - 1 . Consequently,

1 f(c) e s P2n (s-l)ds = ~ 1

2~i I(c) eS{L (n+l) (s-l)+s-lpn(s-l)rn_l(s-l) }ds =

n
1
2~i $c es ~ L~ n+l)(s - l)P2n(sil)ds
- + ~ -1 f(c) e s s - ipn(s-l)rn_ I (s-l)ds. In the
i=]

first integral, the last te~ (corresponding to Sn+ 1 = =) has been left out, because
-i
it vanishes on accoant of P2n(Sn+l) = P2n(0) = 0. The second integral vanishes

identically, if the polynomials Pn(S-1) (that is, ^ ( n ) ,j have been s e l e c -

i f t h e s k = ~k

t e d i n such a way, t h a t the Pn(S-l)'s are orthogonal t o each o t h e r , with the weight
s-1
function e s Indeed, if
 147

(25) I(c)eSs-lpn(S-1)pla(s-1)ds = 0

for all 0 ~ m < n, then it i s i~mncdiate t h a t also

(26) l ( c ) e S s - 1 Pn(S -1 ) s - 3 d" s = 0

for j = O,l,2,...,n-l, whence

(27) 2~ii /(c)eSs-lpn(s-l)rn_l(s-l)ds = 0

follows.
We shall see that such a choice is possible. Assuming this for a moment , (24)
becomes
n
(28) x f(x) = [ A~n)P2n(sil) w i t h A! n ) - 1 eSs-lL(n+l)(s-1)ds.
i=l ' t - --Y(c)
2~i i

It is clear that the A~njf~, the so called Christoffel constants, are independent
i
of the functions involved, as t h e L!n+l)l d e p e n d o n l y on t h e c h o i c e o f t h e s~ n)

(1 5 k S n ) . The c o e f f i c i e n t s o f P2n(S -1) d e p e n d , of course, a l s o on x.

The Gaussian quadrature formula (28), exact for polynomials 02n(S -1) without

constant term and of degree at most 2n will have been established, as soon as we
show how one can select the constants s k = s~n)- so that (25) should hold.
n-I
In g e n e r a l , pn(Z) = z n + ~ b z u, so t h a t (26) becomes

n
(29) ~ b 1 s -i
~=1 u 2--~T l ( c ) e s • s-~-Jds = 0, b n = 1.

m
The i n t e g r a l i n (29) e q u a l s ~ 1 1 s ds. Here a l l integrals
m=0 m! 2~i l(C) s~+J+l
1 ds
vanish, except that for m = u+j, when ~ l(c) --~ = i. The sum over m becomes

1
(~+j)! and (29) y i e l d s

n b
(30) ~ (u+j), = 0 for j = 0,i, . ..,n-l.
~=0
One may verify that the determinant of the b ' s (0 ~ ~ S n-l) is different

from z e r o f o r a l l 0 ~ n e Z, so t h a t t h e b 's ( b = b ~( n ) )_ a r e u n i q u e l y determined.

Once we know t h a t tile b ' s (hence, t h e pn(Z) a n d , t h e r e f o r e , the Sk'S) are uniquely

determined, we can a v o i d t h e l a b o r o f a c t u a l l y solving (30), by u s i n g some o f t h e

results o f C h a p t e r 4.
We f i r s t o b s e r v e t h a t the infinite vertical path of integration along the
a b s c i s s a c may be r e p l a c e d by t h e open p o l y g o n a l c o n t o u r ( - ~ - i Y , c - i Y , c+iY, -~+iY)
 148

(as Y ÷ ~, the contributions to the integral in the original and in the modified
path, essentially reduce to that from c-iY to c+iY); next this open contour may
be replaced by the closed rectangle of vertices c±iY, -X±iY (observe that the inte-
grand has no singularities with Re s < -X, if X is sufficiently large); in a third
deformation, the rectangle may be replaced by a circle of diameter from -X to c;
finally, we invert this circle in the unit circle and take as new variable z = i/s.
In this way, it is seen that (25) is equivalent to

(31) I c el/Zz-lpn(Z)pm(z)dz = 0

for m = O,1,...,n-1, with C a circle of diameter (-X - 1 , c - i ) . The o n l y s i n g u l a r i t y

of the integrand is at the origin so t h a t one may a c t u a l l y take for C the unit circle.
We now r e c a l l Corollary 4.5. First, it is clear that f o r any a , one may add
any c o n s t a n t to p(z;a,b), without affecting the validity of the corollary, because
t h e p r o d u c t o f two p o l y n o m i a i s is holomorphic in the unit circle. We, t h e r e f o r e ,

may r e p l a c e in Corollary 5 the weight function p(z;a,b), by p ( z ; a , b ) =

r(a) b n
n=l F(a+n-1) ( - ~) " N e x t , we c o n s i d e r the particular c a s e a = 1, b = - 1 , so that

p(z;l,-l) = ~ (r(n))-iz -n =
z- l e l / Z . Corollary 5 now reads
n=l

2wi I,lz I=I e l / Z y n ( Z ; l'-l)Ym(Z;l'-l) d-!z

z = 0

for0<m<n.
It follows that if we take in (31) pn(Z) = CnYn(Z;l,-l), then (25), hence,

also (26) and (27) h o l d . Here c n = {(-l)nn(n+l) . . . ( 2 n - i ) } - 1 , as f o l l o w s by c o m p a r i n g

the coefficients o f z n on b o t h s i d e s of the equality. On t h e o t h e r h a n d , i t follows

from t h e u n i q u e n e s s of the b 's (i.e., of the Pn(S)'S) , that the coefficients of

CnYn(Z;1,-1) are the only solution of (30). The s~ n) (k = 1,2 . . . . . n) a r e t h e n t h e

reciprocals of the zeros of yn(Z;1,-1), i.e., they are the zeros of the polynomials
@n(Z;1,-1). This finishes the proof of the exact quadrature formula (28).

I n c a s e G(s -1) i s n o t a p o l y n o m i a l o f d e g r e e 5 2n, t h e n

-i -i -i
G[s
-lj-- = L(n+l)(s-lj- + s Pn(S )g(s ), where g(z) is not, in general, a polynomial

of degree < n-I but is holomorphic for Re s > c (see [13]). Instead of (28) we now
obtain
n
xf(x) = [ A~n) G ( s i l ) + R,
i=l
1 s-i -
where R = 2 ~ i l(c) e s CnYn(S 1 ) g ( s - l ) d s is an e r r o r term. While this, in gene-

ral, is not identically zero, it may still be possible to obtain an upper bound for
 149

IRI, in a way similar to the classical case of real paths of integration and (28),
while not exact, gives a usable approximation to the inversion.
In [99] Salzer lists the polynomials yn(z;l,-l) for n = 1(1) 12 (this notation

means that n runs from 1 to 12, with the difference between consecutive entries
indicated in the parentheses), the zeros of the corresponding @n(Z;l,-l) to 8 signi-
ficant figures and coefficients A {i) to 7 or 8 figures.
n
In [I00] Salzer completes the theory of BP, by quoting Krall and Frink, Burch-
nall, Agarwal, A1 Salam, Ragab and Brahman. He extends his list of polynomials to
n = 16, and tabulates the zeros of the On(Z;l,-l) and the Christoffel numbers for

n = 1(1)16 to 15 significant figures.

Already in 1959, Kublanowskaya and Smirnova [70] had published a tabulation of
the zeros of On(Z ) (= On(Z;2,2)) and of hen(Z) + Z2On_l(Z) for n = 1(i)30. Other

tables for the numerical inversion of the Laplace transform were published by

Skoblja [102] for s and A k in the formula ~ 1 /(c)eZz-S,(z)dz n A~n)~(Zk)

~ ~ with
k=l
n = i(i) i0 and s = .i (.1)3, to about 8 and 7 significant figures for the zk and

~n), respectively. See also [103]. Next, Krylov and Skoblja, in [69] tabulate

s and A k, as follows: s = i(i)5 for n 1(1)15 to 20 significant digits and for

s = .01 (.01)3 and n = i(i)i0 to 7 or 8 significant digits. A generalization of

these procedures can be found in R. Piessens [87]. Further comments may be found in
the books of Y.L. Luke [71], vol. II, p. 189 et seg., and [72], p. 229 and 433.
 CHAPTER 15
MISCELLANEA

Many papers related to BP have not been considered in detail in the preceding
chapters. Sometimes, when such a work was related to a topic discussed, or was a
rediscovery of an older result, it was at least quoted. This was the case, e.g.,
with many papers on generating functions of BP. In other cases, however, when a
paper treated some isolated topic, not easily fitted into some broader category, the
paper often was not mentioned at all. Among these papers are some very valuable
ones, other of a more routine character and also some (fortunately very few) with
erroneous claims. It also is not surprizing that among the so far ignored papers,
many are by authors some of whose other work on BP has already been prominently
displayed.
The purpose of the present chapter is to attempt to mention all these, so far
neglected papers and their authors and to state at least briefly and without proofs
the results obtained.
A major difficulty is to make such a presentation in a systematic way. It
could, perhaps, be considered desirable to arrange the material by subject matter.
In fact, when several papers discussed related topics, in general they could already
be fitted into one of the preceding chapters. Most of the material here considered,
consists of papers each of which discusses a more or less isolated subject.
In spite of many advantages, also a strictly chronological presentation would
not be very useful. Indeed, some authors worked on different topics related to BP,
throughout many years and their names would be intermingled in a rather confusing
way. For these reasons it has appeared that the most convenient way is to list
these papers by authors. In order to facilitate the search for a specific paper,
whose author is known, the authors are listed in alphabetic order. The papers of
any given author, however, will then be listed in chronological order, under the
heading of their author.
With very few exceptions, it has been possible to obtain the original papers.
Whenever this turned out to be impossible, or to require an excessive investment of
time, the information concerning its contents was obtained through the reviewing
journals, especially the Mathematical Reviews.
Even in the cases, when the original paper was accessible, no critical apprai-
sal of its contents appeared possible, or even desirable. An exception was made
when a result appeared to be plainly erroneous, or valid only under restrictions not
stated explicitly.

i. W.H. Abdi [i]. We recall the definition of the basic hypergeometric series (see,
a(a+l) ~:: Ca+n- l)
e.g. [6]). A new parameter q is introduced, lql < i. The ratios c(c+l) Cc+n-l)

that occur in the definition of the hypergeometric series, are replaced by

 151

(l-qa~(l-~a+l)'''~ (a-qa+n-l) One observes that the limit of the new ratio, for
(l-q c) (l-q c+l) ... (l_q c+n-l) •

q ÷ i, is the previous ratio. In analogy to the definition of the basic hypergeome-

tric function, the author defines a basic analog of BP. For this he starts from

2 ~ o ( q - n ,q a+n-1 ; - , x ) (the basic analog of 2Fo(-n,a÷n-1;-;-x/b), (see [6]), but with

x i n s t e a d o f - x / b ) and d e f i n e s t h e b a s i c BP by

(1) J(q;a-1;n;x) = { ( 1 - q a - 1 ) n / ( q ; n )} 2 ~ 0 ( q - n ; q a + n - 1 ; x ) , where

¢o

(l+x) n = I [ ( l + x q k ) / ( l + x q k + n ) , and ( l - q ) n = ( q ; n ) .
k=0

The a u t h o r s t u d i e s many p r o p e r t i e s of J(q;c,n;x), such as i n t e g r a l representa-

tions, recurrence relations, etc.
The m o t i v a t i o n f o r t h i s p a r t i c u l a r definition (1) may, p r e s m n a b l y , be found i n
(a-l) n
Rainville's [90] normalization ~n(a-l,x) - n! 2F0(-n,a÷n-l;-;x). On the one
(a-l) n
hand, this coincides with n~ yn(-bx;a,b); on t h e o t h e r h a n d , i n t h e s p i r i t of a

"basic" function, lira _ J ( q ; a - 1 ; n ; x ) = ~n(a-l,x). Otherwise, the connection of

q÷l
J(q;a-1;n,x) t o Yn(X;a,b) i s r a t h e r tenuous.

2. R.R. Asarwal. In [ 2 ] , i n a d d i t i o n t o t h e m a t e r i a l m e n t i o n e d e a r l i e r (see Chap-

t e r 5 ) , one f i n d s a l s o many o t h e r r e s u l t s , such a s :
(a) Some a d d i t i o n a l recurrence relations for yn(Z;a,b).

(b) Integral representations, like

2-a
n~ z n u ebU/Zdu '
Yn ( z ; a ' b ) = ~ (b) ¢]ul=l u(l_u)n+l

and

i f~ ta-2+n(l+tx/b)ne-tdt.
Yn (z;a'b) = r(n+a-l) o

(c) Expansion formulae, of which the following is typical:

2n (-2n)2r(a+2n-l)r .x. 2r
Y2n ( 2 x ; a ' b ) = ~ r! (~) Y2n-2r ( x ; 2 a + 2 n + 4 r - l ' b ) "
r=0

3. W.A. A1-Salam. In addition to the contributions already presented, the follow-

i n g s h o u l d be m e n t i o n e d .
In [3] we f i n d :
(a) New i n t e g r a l representations and new g e n e r a t i n g f u n c t i o n s .
(b) A connection with Jacobi Polynomials:
 152

Yn(X;a) = n!(-x/2) n lim p(-2n-a+l,8)(l_(x8)-l).

n

(c) Characterizations of BP, as follows:

(i) Let a sequence of polynomials fn(X,a), of degrees n = 1,2,..., depend

ing also on a parameter a, and normalized by fn(0,a) = i, satisfy the recurrence

relation (due to Krall and Frink; see Section 19 of the present chapter)

1
(2) f~(x,a-2) = ~ n(n+a-l)f_l(x,a);

then fn(X,a) = Yn(X;a)-

(ii) Let AaYn(X,a ) = Yn(X;a+l)-Yn(X;a); then

1
AaYn(X;a ) = ~ nXYn_l(x,a+2),

and,more generally,

A~Yn(X;a ) = n(k)(x/2)kyn_k(x;a+2k);

in particular,

A~Yn(X;a ) = n!(x/2) n.

From this and Newton's formula f(a+v) = ~ (~)Arf(a) the author infers that the
r
BP satisfy the identity

(31 Aaf(X,a ) = {x/(n+a-l)} f'(x,a).

Let now fn(X,a) be any sequence of functions, not necessarily polynomials, de-

pending, besides the variable x, also on a parameter a. Furthermore, let fn(X,a)

be normalized by fn(O,a) = fo(x,a) = 1 and assume that the fn(X,a) satisfy (3) and

the "initial condition" fn(X,2) = Yn(X); then fn(x,a) = Yn(X;a).

(d) Expansion formulae, such as

Yn(X) = l+x{(2n-l)Yn_l(X ) + (2n-3)Yn_3(x)+...};

k
Yn (x'k-n+2) = e- x n~ (-n)s(-X/2)SL~ °)(~) ( L(o)
s (~) = Laguerre Poly-
k=o s=o nomial);

(_l)kk2kyn(x,2k_n+2) 1 f~ e-t(l+xt/2)n sin 2Xt dr;

k=o =~i- o t

(z/2) ~-leXZ2/8 = ~ (2n+8-i) F(n+B-l)

n! Yn(X;S)J2n+B-i (z)
n=o

(6 ~ O, -i, -2 .... ; J (z) = Bessel function).

 153

(e) Product expansions (see also Brafman [15]):

n (-n)r(n+a-i)
Yn(U'a)Yn ( v ' a ) : ~ r' (- ~ ) 2 y r ( u +~ ,a)
r=O
and the c o r r e s p o n d i n g i n v e r s i o n s

u+v~n . uv n (-n) s ( 2 s + a - 1 ) r ( s + a - 1 )
(- ~ J Yn[u-~-~v,a) = ~ s'F(n+s+a) Ys(U'a)Ys ( v ' a ) ;
S=O

also similar formulae involving @n(X,a).

(f) Product representation formulae, like the following two:

Yn(U'a)Yn ( v ' a ) _ r(n+a-1)

n! I : ta-2p(a-2,0)(l+t(u+v+
n 2))e-tdt

and
yn(~)yn(
_ x ~)x = ~l (~)'x'2nf~o t2nL(-n-i/2)(-s2/2tx2)e-t/2dtn

2n÷2 2
s 11' f~o tnyn (-8x2t' 2-3_n)e-tS dr.

In [4] the author considers the second (non-polynomial) solution of (2.11),

which i s qn(X) = (-1)ne2/Xyn(X), and e s t a b l i s h e s r e c u r r e n c e s for qn(X), analogous

to (3.16) and (3.22) (for a = 2). Ile also gives new characterizations of the BP,
of which the f o l l o w i n g i s t y p i c a l :
Given a sequence of polynomials fn(X), of degrees n (n = 0 , 1 , 2 . . . . ) , n o r m a l i z -
ed by fo(x) = 1, then fn(X) = Yn(X) i f and only i f

(4) x2 d )2
~" (fn(X)fn_l(X)) = (fn(X)-fn_l(X) •

In [5] the a u t h o r o b t a i n s a second s o l u t i o n z ( a ) ( x ) to (2.26) and a second

n

polynomial s o l u t i o n v n( a ) ( x ) to ( 3 . 2 0 ) . z n( a ) ( x ) can be r e p r e s e n t e d by a d e f i n i t e

integral and also in terms of Whittaker functions. For integral a > -n, z(a)(x) is
- n

an entire f u n c t i o n ; otherwise i t i s holomorphic only i n the p l a n e cut along the

p o s i t i v e r e a l a x i s , where z ( a ) ( x ) has along x > 0, the d i s c o n t i n u i t y

n
-2i sin(a~)Yn(X;a). For z ( a ) ( x ) , the author proves r e c u r r e n c e r e l a t i o n s , generating
n

f u n c t i o n s , m u i t i p I i c a t i o n theorems, e t c .

The " a s s o c i a t e p o l y n o m i a l s " V(a) (x) are orthogonai on the u n i t c i r c l e with

n

r e s p e c t to the weight f u n c t i o n ~(x,a) = x { 1 F l ( 1 ; a ; - z x - 1 ) } - I = ~ B r ( a ) x l - r ; here

r=o
m
the c o e f f i c i e n t s may be computed r e c u r s i v e l y by ~ ( - 2 ) r g m _ r ( a ) / ( a ÷ l ) = Bm(a) for
r=o
 154

m ~ O, 6o(a ) = i; for a = 2, 6n(2 ) = (-l)n2nBn/n!, with B n the Bernoulli numbers.

As particular cases, the author recovers several results of Ai-Salam and Carlitz
[7]. Next, some very general identities of Christoffel-Darboux type are presented.
Several properties of the BP yn(Z;a) are obtained, some of them new, others already

known (but often with new proofs), such as orthogonality (see [68] and [93]), sim-
plicity of zeros (see [53]), and others. The author shows that, for a > 2, all
zeros of yn(Z;a) are inside the unit circle (for a = 2 the result had been proved

in [53]) and those of v(a)(x) inside Ix[ < J 3/5 Also the irreducibility of
n
yn(Z;a), for a an integer, is proved for certain values of n.

In [6] the author generalizes some results of Chatterjea [23].

4. W.A. AI-Salam and L. Carlitz. In [7] the authors study polynomials Un(Z), Vn(Z),

which are, essentially , the polynomials Pn(Z), Qn(Z) of Chapter 8. Here are some

characteristic results:

Define coefficients gn and yn by 6n = 2nBn/n! (Bn = B e r n o u l l i numbers) and

2(eX+l) -1 = Yn(X/2)n; then, for 0 < m < n, ym+lun(Y) = (-1) n ( 2 2nn!

n+l~6mn and
n=o

6m+2vn(6) ( l n 2n+l(n+l)"
= ,-I, 6 . IIere the 6's are Kronecker deltas and the
(2n+3)! mn

)'powers" are symbolic: a f t e r ym+lun(Y) and Bm+2vn(6) are computed, the exponents

are lowered to become i n d e c e s . Also,

2~il f l z [ = l e2/Z_12 zmun(z)dz = (2n+3)(2n+2). ~ 1 2

f l z [ = l e2/Z 1 zmvn(Z)

( _ l ) n + l ( n + l ) ! 2 n+l
6
(2n÷1)! mn"

In [8] the a u t h o r s study the r e l a t e d polynomials Un(X) = i - n u n ( i X ) , Vn(X) =

i-nvn(ix) • Let a(x) be a s t e p f u n c t i o n with s a l t i J(Xk) = x k2 at Xk 2 2k+I
~ 1 (k E Z ) ;
then
f~ xmUn(x)da(x) = 2nn! ,~
and j_~ gn(x)d~(x ) = mn .
-~ (2n+l)-------~-6mn' Um(X) 2n+l
3
S i m i l a r l y , i f 6(x) i s a step f u n c t i o n with s a l t i j(x~) =
2k2 at the p o i n t s
1
x~ = ~ (k = ±1, ~2 . . . . ) , then

f~-~ Vm(X)Vn(X)da(x) = 2n+33 ~mn.

S. W.A. Ai-Salam and T.S. Chihara. The classical polynomials of Jacobi, Hermite,
and Laguerre are the only orthogonal polynomial solutions of
 155

(5) ~(x)P~(x) = (~nX+Bn)Pn(X) + ¥nPn_l(X) (n > i),

with ~(x) a polynomial and ~n' 6n' Yn constants.

In [9] the authors show that, if the classical meaning of orthogonality is

relaxed to admit weight functions of bounded variations, then the BP have to be
added to the solutions of (5) (indeed, see (3.10)). This result solves affirmative
ly an older conjecture of Karlin and SzegS.

6. I. Bailey [i0] studies the convergence and, especially the C-I summability of
series of BP.

7. D.P. Banerjee [ii] discusses some non-linear recursions; his "Turin inequality"
is incorrect as printed (see [28] for correct form; an equivalent result had been
obtained earlier by L. Carlitz [19]).

8. L. Carlitz defines (see [19]) the functions fn(X) = X@n_l(X) introduced in

Section 6.7. In addition to the results discussed there, he proves recursion
identities for fn(X), the formula
n
fn(U+V) = [ (~)fr(U)fn_r (v);
r=o
also representations of Laguerre Polynomials and of BP as sums of fn(X)'S. He
shows that
n
@n(X)en(Y ) = r=o
~ (n-r):r!2
(n+r): r (xy)n-r@r(X+Y)'

and proves the inequalities of Tur&n type @n+l@n_l-e~ ~ 0 and en@~+l-@n+10~ ~ 0 for

x > 0. Finally, the author investigates some arithmetic properties of @ (z).

- n

9. S.K. Chatter~ea published at least 15 papers related to BP. In addition to the

author's work quoted in Chapter 6, tile following deserves mention. In [21] he gives
two representations of yn(Z) as a determinant (one of them essentially equivalent to
2
that discussed in Chapter 3). In [22] he proves the formula Y2nY2n+2 = Y2n +
2n
(4n+S)x 2 ~ (2k+l)y~ + (4n+3)x(x+l). From this it easily follows that Y2n(X) > 0
k=l
for x ~ -i, a result used by Cima [40] in his proof that Y2n(X) # 0 for all real x.

In a sequence of three papers [23], [24], [25], the author establishes and
d
then uses operational formulae. In [23] he proves (with D = ~x ) that
n n
~-~ {x2D+(2j+a)x+b} = ~ bn-rx2ryn_r(X;a+2r+2,b)Dr , which he uses to show that
j=l r=o
min~n,m)
Yn+m(X;a,b) = (~) (~)r! x
(m+2n+a-l)r(~) 2r Yn_r(X;a+2r,b)Ym_r(X;a+2n+2r,b).
r=o
 156

In [24] he proves the formulae

n
x 2n (D+2 (nx+l) x- 2) n = ( : ) 2 n - r x 2 r y n _ r (x; 2+2r, 2) Dr;
r=o

and
xn(D_(2x+n+l)x-l) n = ~ (-2)n-rxrOn_r(X;2+r,2)D r.
r=o

In [25] the author proves that

bx n -bx n
e [ I (xD-a-2n+j+l)e y = [ (~)(-b)n-rxr@n_r(X;a+r,b)Dry,
j=l r=o

from which follows that

n
(-b)n@n(X;a,b) = ] [ (xD-bx-a-2n+j+l)'l.
j=l
In [26] the a u t h o r o b t a i n s a r e p r e s e n t a t i o n of the BP as a double i n t e g r a l .

In [27] the author gives a new generalization of the BP.

In [28] some of the Christoffel-Darboux identities contained in (or easily derived

from) [5] are rediscovered.

In [29] the author considers the polynomials defined by

Mn(X,k) = x(2-k)nk-nek/XDn(xkne -k/x) (2 < k ~ Z).

These polynomials generalize the BP, to which they reduce for k = 2 (see (7.2"),
while Mn(X,3 ) = Yn(X;n+2,3).

2
In [32] the author proves Tur~n's inequality &n(X) = ynYn+2-Yn+l >_ 0 for x ~ 0

(which as already observed is actually an immediate consequence of Carlitz's result

in [19]), by first obtaining an explicit formula for An(X). He then proceeds to
use this formula to prove also

f]xl= 1 (x-2an(X)-X-1)e-2/Xdx = 8 ~ i ( - 1 ) n + l ( l + [ ~ ] ) .

In [33], in analogy with Burchnall's operator Q(~) (see [17] and Theorem 2.21 the
author introduces the differential operator Q(6) = 6(6-2)...(6-2n+2) and studies

the polynomials defined by cn(Z) = (-l)neZQ(6)e -z (n ~ i), $o(Z) = I.

In [34] the author gives a proof, different from Carlitz's [19], for Tur~n's
inequality

@n_l(X)@n+l(X)-@~(x ) > 0 for n > i, Ixi < i.

10. C.K. Chatter~ee [38] proves (a) t h a t f o r n > 1, y~(x) and y~_l(X) have no

common zero; (b) the r e l a t i o n (x 3 d~) k (xne 1/Xyn_l(X)) = xn+k e - 1 / x Yn+k_l[X)

- " " and

s i m i l a r ones; and (c) some i n e q u a l i t i e s for products of BP.

 157

Ii. J.A. Cima [40] gives a correct proof for the statement that Y2n(X) # 0 for
all real x, whose proof in [53] had been incorrect.

12. M.K. Dan published at least 5 papers related to BP. In addition to his work
mentioned in Chapter 6, we find in [41] that, if Y is a sufficiently differentiably
d
function and 6 = x ~-~x , as in Chapter 2, then

[x(~+a+k_l)]n(xn+ke-b/Xy) = e-b/x ~ (~)bn-Pxn+2p+kyn_p(X;a+2p+2,b)DPY,

p=o
and the author makes some nice applications of this and related formulae.
In [45] the author proves a very general operational identity. As particular
cases he obtains, among others, again the formula of [41].

13. D. Dickinson's paper [47] contains, in addition to the work discussed in

Chapters 5, 9,and 10, and material not directly related to BP, also the following
result:
Let D be an arbitrary finite set of integers and let s be a fixed integer.
Also, let fn(X) be a sequence of algebraic functions such that

fn(X)Jn+s+i/2(s ) vanishes identially (J (s) = Bessel function); then

neD

both sums, [ fn(i/ix)inyn+s[X) and [ fn(-I/ix)i-nyn+s(X) also vanish identi-

neD neD
cally (observe the misprint s for x in the theorem of [47]).

14. K. DoPey [48] completes and refines work of Obreshkov [82], both, on the zeros
of BP and on the expansions of functions in series of BP. The author's contribution
to the first topic was presented in Chapter 10. In connection with the second, the
author proves the following Abelian theorem: Let us assume that the series

anYn(X;a,-I ) converges in a subset D of Izl < IXol to the function f(x), that
n=o
x ° belongs to the closure of D,and that the series converges also at x = x ° to s;

then lim f(x) = s, provided that the path along which x approaches x ° is inside D.
x+x
o

15. M. Durra S.K. Chatter~ea, M.L. Moore study in [49] a class of orthogonal poly-

nomials, that can be represented by H~(x) = x n 2F 0(_ ~, _ ~i (n+~_l);_;_i/x 2) and

note the similarity of these polynomials to the BP.

16. M.T. Eweida [51] indicates connections of BP with Meijer's G-functions,

Laguerre Polynomials and Bessel Functions and gives integral representations, of
which the following are characteristic:
 158

-12n+In!eXe2n+l f~ cos t dt
@n (x) = o (t2+n2)n

@n(X;a) = (_l)n+a-12(a-l)/2e2Xx(2n+a-1)/2 io ~ e-tt-(a-l)/2J2n+a_l(2 2/~x)dt

@n(U)@n(V) =
(2~)-I/2eU+V I~ e -(t+(u+v) 2/t)/2~n-I/2~
L
rU+V~j.
~ n L'--~J UL.

In [52] the author rephrases some (known) ~nfin~te integrals that involve Kn+i/2(x),

by replacing the Bessel function by its value (see Section 3.1)

2-1/2~l/2e-Xx-n-i/2On(X).

17. A.M. Hamza [59] e v a l u a t e s some i n t e g r a l s that involve BP.

18. M.E.H. Ismail [61] g e n e r a l i z e s and s o l v e s a problem that had previously been
a s k e d b y R. A s k e y [ Z ] . The p r o b l e m , i n i t s generalized formulation, is the
following: Let Pn(x,a) be a s e q u e n c e o f p o l y n o m i a l s , orthogonal o v e r an i n t e r v a l I,

with respect to a weight function w(x,a), where both, the polynomials and t h e w e i g h t
function depend besides the variable x aiso on a p a r a m e t e r a. Given also a se-
q u e n c e a n (n = 0,1,2,...), integers N a n d M, and c o n s t a n t s a,b,c,d, it is required

to determine a function f(x) in such a way that

I llf(x)m(x,c)Pn(x,a)dx for n = 0,i ..... N;

an = , /if(x)~(x,C)Pn(X,d)dx for n N+I, N+2 ..... M;

Iif(x)m(x,b)Pn(X,b)dx for n > M+I.

Conditions on Pn(x,a), necessary for the solvability of this problem are

determined and among the solutions Pn(x'a) are found also the BP.

The author obtains as one of his results also the following formula (previous-
ly obtained by Ai-Salam [3]):
(a+n-l)k (4k+c-n-a)n_ k
Yn(X;a) = k=o~ (2) (c*n-l)k (2k*C)n-k Yk(X;e)"

Other contributions of the author with Kelker to the theory and applications of BP
have been mentioned in the Introduction, and in Chapters i, i0, and 14.

19. H.L. Krall and O. Frink coined the name Bessel Polynomials in [68] and much of
their work has been discussed in preceding ch~ters. Nevertheless [68], contains
also some topics, mainly related to the generalized BP yn(Z;a,b), a ~ 2, that were
not discussed so far and they deserve mention at this place.
Several recurrence relations occur in [68], not mentioned in Chapter 3. Also
the differential equation for y'(z;a,b),
 159

(6) x2 d 2 Y_,__~n + (ax+2x+b) d-~

dy~--= n (n+a+l) y~
dx 2

appears in [68]. From (6) it follows that y~(z;a,b) is proportional to a BP with

the value a+2 for its principal par&meter. As y~(z;a,b) is a polynomial of degree

n-l, it follows that y~(z;a,b) = ClYn_l(z;a+2,b), or y~(z;a-2,b) = CYn(Z;a,b ). By

comparing the constant terms on both sides we obtain that c = n(n+a-l)/2. It is

this result that was generalized by Ai-Salam [3] to (2).

20. P. J. McCarthy's contribution in [74] to the location of the zeros of BP was

mentioned in Chapter i0. In [75] he shows how to generalize AI-Salam's formula (4)
(see [4]) to other systems of orthogonal polynomials.

21. N. Obreshkov's paper [82] has already been mentioned in connection with DeWey's
work on the zeros of BF in Chapter i0. In the same paper the author also studies
expansions of holomorphic functions in series of BP.

22. A. Pham-Ngoc Dinh [86] studies the function Y~(z) = e-Zz(l-a-2n)/nen(Z'a)"

The author obtains differential-difference relations satisfied by Y~(z), o p e r a t o r s

that raise, or lower the value of the parameter a, and expansions of Ya(z-utzl/2)
n

in series of the form X Cn(P,Z,tu)Y~+P(z), with rather complicated coefficients

p=o
Cn(P,Z,tU), and convergent for lutl ~ ~ , unless a is a negative integer or

a = +i.

23. F.M. Rahab ~8] states and offers proofs for five summation formulae, of which
the following is the simplest:
n
(n) P(2-a) 2x, r ,x .
(7) £(2-a-r) (- --b-J Yr(2 ' a,b) = Y2n(X;a-2n,b).
r=o

It seems that certain restrictions (not explicitly stated in [88]) on the

values of a are needed, in order to insure the validity of the summation formulae.
So, e.g., if the arguments of the ga~na functions are zero, or negative integers,

the ratio F(2-a)

£(2-a-r) presumably has to be interpreted as its limiting value

(-l)r(a-l)a(a+l)...(a+r-2). For a = 1 this product vanishes, except when r = 0, so

that (7) becomes yo( ~x ," l,b) = Y2n(X;l-2n,b), which is obviously incorrect. Indeed
x
yo( ~ ; l,b) = i, while Y2n(X;l-2n,b) is a polynomial of degree 2n. For n = I, e.g.,
2
Y2(x;-l,b) = l+(4/b)x+(6/b2)x .
 160

24. A.K. Rajagopal [91] verifies that a certain function, closely related to the
BP satisfies a certain equation (known as Truesdall's equation) and from this fact
he draws some (essentially known) conclusions about the BP themselves.

25. H. van Rossum, in [93] and [94] studies the orthogonality of the BP on hand
of the Pad~ table. He calls a sequence of real numbers c m "strictly totally posi-
tive" if all determinants ICm+i_jl ~,j = 0,1,2 ..... n-l) are positive for all

m = 0,1,2,... and n = 1,2,3 ..... If the Pa~e rational fraction of the square

~,v) of the power series ~ Cm zm has the denominator V ,~(z), then the polynomials
m=o

B (k)(z) = ZUVk,k+u(-z-l) (k = fixed integer,~ = 0,i ...) are said to be "totally

positive". They are orthogonal on the circle Izl = p+~ (p = lira sup {m C~m}, c > 0),

with respect to the weight function ~(k)(z) = ~ (_l)m+k+icm+k+l z-m-l; they have
m=o

only positive coefficients; m d they have all their zeros inside the circle
Izl < Cl/C o. The BP are an instance of totally positive polynomials, corresponding
= =
to c o cI i. The author's further contributions to the location of zeros of BP

[95] have already been mentioned in Chapter i0.

26. P. Rusev [96] indicates sufficient conditions for thc convergence of series in
BP, similar to those corresponding to series in Jacobi polynomials inside an
ellipse with foci at +i and -I.

27. H. Rutishauser [97] uses continued fractions, in order to give elegant proofs
of results previously obtained by Ai-Salam and Carlitz [8] and by D. Dickinson [47],
concerning the orthogonality on the real axis of certain functions (denoted in [8]
by Un(X ) and Vn(X)), closely related to BP.

28. S. L. Soni [104] obtains operational formulae for the generalized BP by using
operators different from D(= d/dx) and 6[= x d/dx).

29. II. M. Srivastava [105] establishes expansions of a large class of functions in

series of von Neumann type, involving (among other systems) also BP.

30. L. Toscano, in addition to his work on gencrating functions [109] quoted in

Chapter 6, also obtain (in [Ii0])
(a) representations of BP as Laplace transforms of a Legendre Polynomial:

Yn ( - 2 / p ) o Ln(l_2t)e-Ptdt
= P f~ ' valid f o r Re p > 0; a n d , i n a somewhat c h a n g e d

notation,
(b) representations of the type
 161

Yn(x-l;a) = x-nen(X;a) = F(a)-i i=

o u a-i 2Fl(-n,n+a-l;~,-ux/2)e -u du.

31. It is not possible to conclude this chapter, as well as this monograph, with-
out the sincerest apologies of the present author to the many contributors to the
theory, or the applications of BP, whose work was either overlooked, or misinter-
preted, or underevaluated. Ideally, every one of them, who could be located, ought
to have received the intended text (at least in as far as it related to his/her
work), with a request for approval and/or comments. Due to the large number of
contributing mathematicians, their dispersal throughout the world, the many years
that have elapsed since many of the papers were written - and, last, but not least,
the fact that, unfortunately, some of them are no longer alive, this ideal proce-
dure was not practicable. Even the partial execution of the possible part of this
program would have entailed such an additional delay for the present publication,
that the value of this monograph itself would have been put in question, l:or this
reason the present author counts on the understanding of his colleagues, and on
their forgiveness of his sins of omission and commission.
 APPENDIX

SOME OPEN PROBLEMS RELATED TO BP

Quite frequently papers appear with improvements of kno~m results concerning

BP. So, e.g., new generating functions are obtained, sharper bounds for the loca-
tion of the zeros of BP are determined, new recursion formulae are established, etc.
Such improvements of known results are of course always welcome; here, however, we
want to call attention to a number of unsolved problems related to BP and of a some-
what different nature.

I. It seems that the polynomial solutions of equation (2.26), with a = I,

b = 2 have interesting properties. This was observed already by Krall and Frink,
who suggested in [68] that such a study be undertaken. It seems nevertheless that
these polynomials were never investigated thoroughly, although it may be worthwhile
to do so. As we have seen in Chapter 14, these polynomials occur in the inversion
of the Laplace transform. Do they occur also in some different context? Do they
have other practical, or theoretical applications?
2. The BP satisfy a very specific condition, recently called Bessel orthogona-
lity (see [63]]. Are there other interesting sets of functions (perhaps even of
polynomials) that have this property?
3. In Chapter 9 we studied expansion in series of BP. While the work of Boas
and Buck (see [13]) is remarkable, our knowledge of these expansions in series of
BP cannot be compared, as completeness, with the theory of expansions in Fourier
Series or in series of classical, orthonormal polynomials. Many open problems
remain, concerning, e.g., the speed of convergence, behavior of the series on the
boundary of its domain of convergence, summability methods (see, however [i0] and
[96]), Lebesgue constants and other similar questions. The generalizations to the
case a # 2 have, apparently not even been touched.
4. Are there reasonable generalizations of the BP to polynomials in several
variables?
5. In Chapter i0, certain regions were determined, where all the zeros of
yn(Z;a) (or of @n(Z;a) are located (see, e.g. the work of Parodi, DoPey, Saff and
Varga, etc). In Section 12, on the other hand, following Olver, a curve is indica-
ted, close to which all these zeros lie, the approximation becoming better with
increasing n. There is a gap to be filled between these two types of results. So,
e.g., one may determine a strip, enclosing Olver's curve, such that all the zeros
of yn(Z;a) belong to that strip.
6. Wimp [112] proved that the zeros ~!n)(a) (j = 1 2 ..... n) of yn(Z;a) have
j
strictly negative real parts. One may conjecture that the property holds for a
larger set A 1 of values of a. (Conjecture: A 1 = ~). Also the set A 2 of values

of a, such that yn(Z;a) = 0 has a real zero, if n is odd, may deserve study.
 163

7. Determine good upper and lower bounds for the real zero of yn(Z;a), n odd,
where a ~ A 2 (see Problem 6).

8. Find explicit formulae for ~(n) and ~(n), the sums of the r-th powers of
r -r
the zeros of yn(Z), and of 0n(Z), respectively; if possible, generalize to the case

of arbitrary a ~ 2.
9. Prove that yn(Z) (hence, also 0n(Z)) is irreducible over the rational field

for all n; failing that, prove the irreducibility for a large class of values of n.
i0. If possible extend the results of 9 to yn(Z;a), for a ~ 2.

ii. In Chapter 12 the proof that the Galois group of yn(Z) is the symmetric

group on n symbols required almost a case by case consideration for each n < 13.
The problem is to find a unified approach that works for all these small values of n.
12. Define the function fn(Z;a) by (see (13.9)) fn(Z;a) =

(2nz/e) -n 23/2-ae-i/Zyn(Z;a); devclop fn(Z;a) into a complete asymptotic series.

 BIBLIOGRAPHY OF BOOKS AND PAPERS RELATED TO BESSEL POLYNOMIALS

I. W.H. Abdi - A basic analog of the Bessel Polynomials - Math. Nachr.,vol. 30

(1965), pp. 209-219; MR 32 #7795.

2. R.P. Agarwal - On Bessel Polynomials - Canad. J. Math., vol. 6 (1954), pp. 410-
415; MR 15-955.

3. W.A. Ai-Salam - The Bessel Polynomials - Duke Math. J., vol. 24 (1957), pp. 529-
545; MR 19-849.

4. W.A. Ai-Salam - On the Bessel Polynomials - Boll. Un. Mat. Ital. (3), vol. 2
(1957), pp. 227-229, MR 19-542.

5. W.A. Ai-Salam - Some functions relayed to the Bessel Polynomials - Duke Math.J.,
vol. 26 (1959), pp. 519-539; MR 22 #120.

6. W.A. Ai-Salam - Remarks on some operational formulas - Rend. Sem. Mat. Univ.
Padova, vol. 35 (1965), pp. 128-131; MR 31 #4935.

7. W.A. A 1 - S a l a m a n d L. C a r l i t z - Bernoulli N~mbers and Bessel Polynomials - Duke

Math. J . , v o l . 26 ( 1 9 5 9 ) , p p . 4 3 7 - 4 4 5 ; MR 21 # 4 2 5 6 .

8. W.A. Ai-Salam and L. Carlitz - Bessel Polynomials and Bern~oulli Numbers - Arch.
Math., vol. 9 (1959), pp. 412-415; MR 21 #3597.

9. W.A. Ai-Salam and T.S. Chihara - Another characterization of classical, orthogo-

hal polynomials - SIAM J. Math. Anal., vol. 3 (1972), pp. 65-70; MR 47
#5320.

I0. I. Bai6ev - Convergence and summability of series of generalized Bessel Polyno-

mials (Bulgarian, Russian and English Summaries) - B"igar. Akad. Nauk.
Otdel. Mat. Fiz. Nauk. Izv. Mat. Inst., vol. I0 (1969), pp. 17-26; MR 44
#5703.

II. D.P. Banerjee - On Bessel Polynomials - Proc. Nat. Acad. Sci. India, Sect. A,
vol. 29 (1960), pp. 83-86; MR 26 #1505.

12. C.W. Barnes - Remarks on the Bessel PoLynomiaLs - Amer. Math. Monthly, vol. 80
(1973), pp. 1034-1041; MR 49 #660.

13. R.P. Boas, Jr. and R.C. Buck - Polynomial expansions of Analytic Functions -
Springer-Verlag, Berlin, 1958; MR 20 #984.

14. S. Bochner - ~ e r Sturm - Liouvillsche Polynomsysteme - Math. Z., vol. 29

(1929), pp. 730-736.

15. F. Brafman - A set of generating functions for Bessel Polynomials - Proc. Amer.
Math. Soc., vol. 4 (1953), pp. 275-277; ~ 14-872.

16. J.W. Brown - On Burchnall's generating relation for Bessel Polynomials - Amer.
Math. Monthly, vol. 74 (1967), pp. 182-183; MR 36 #4034.

17. J. Burchnall - The Bessel Polynomials - Canad. J. Math., vol. 3 (1951), pp. 62-
68; MR 12-499.

18. J. Burchnall and T.W. Chaundy - Cor~nutative ordinary differential operators II -

The identity pn = Qm _ proc. Royal Soc. Ser A, vol. 134 (1931), pp. 471-
485.
 165

19. L. Carlitz - A note on the Bessel Polynomials - Duke Math. J., vol. 24 (1957),
pp. 151-162; MR 19-27.

20. B.C. Carlson - Special functions of applied mathematics - Academic Press, New
York - London, 1977.

21. S.K. Chatterjea - On the Bessel Polynomials - Rend. Sem. Mat. Univ. Padova,
vol. 32 (1962), pp. 295-303; MR 26 #373.

22. S.K. Chatterjea - A Note on Bessel PoLynomials - Boll. Un. Mat. Ital. (3),
vol. 17 (1962), pp. 270-272; ~ 26 #1506.

23. S.K. Chatterjea - Operational formulae for certain classical polynomials I -

Quart. J. Math. Oxford Ser. (2), vol. 14 (1963), pp. 241-246; ~ 27 #2662.

24. S.K. C h a t t e r j e a - Operational formulae for certain classical polynomials II -

Rend. Sem. Mat. Univ. Padova, vol. 33 (1963), p p . 1 6 3 - 1 6 9 ; MR 2 7 # 3 8 4 5 .

25. S.K. Chatterjea - Operational formulae for certain classical polynomials III -
Rend. Sem. Mat. Univ. Padova, vol. 33 (1963), pp. 271-277; MR 27 #3846.

26. S.K. Chatterjea - An integral representation for the product of two generalized
Bessel Polynomials - Bull. Un. Mat. Ital., vol. 18 (1963), pp. 377-381;
MR 28 #5200.

27. S.K. Chatterjea - A generalization of the Bessel Polynomials - Mathematica

(Cluj), vol. 6 (29) i (1964), pp. 19-29.

28. S.K. Chatterjea - On a paper by Banerjee - Bull. Un. Mat. Ital., vol. 19 (1964),
pp. 140-145; ~ 29 #3695.

29. S.K. Chatterjea - A new class of polynomials - Ann. Mat. Pura Appl., vol. (4)
65 (1964), pp. 35-48; ~ 29 #6073.

30. S.K. Chatterjea - Some generating functions - Duke Math. J., vol. 32 (1965),
pp. 563-564; ~ 31 #5989.

31. S.K. Chatterjea - Some generating ]~nctions of Bessel Polynomials - Math.Japon.,

vol. i0 (1965), pp. 27-29; MR 32 #7797.

32. S.K. Chatterjea - An integral involving Turan

" i ' s expression for Bessel Polyno-
mials - Amer. Math. Monthly, vol. 72 (1965), pp. 743-745; MR 32 #2630.

33. S.K. Chatterjea - Operational derivation of some results for Bessel Polynomials
Mat. Vesnik, vol. 3 (18), 1966, pp. 176-186; MR 34 #7833.
t
34. S.K. Chatterjea - On Turan's expression J~r Bessel Polynomials - Bangabasi
Morning College Ma~., vol. 2 (1966), pp. 18-19; ~ 35 #6876.

35. S.K. Chatterjea - Sur les polynomcs~ de Bessel, du point de vue de l'algebre de
Lie - C. R. Acad. Sci. Paris, Serie A, vol. 271 (1970), pp. 357-360;
MR 42 #3329.

36. S.K. Chatterjea - Operational derivation of some generatir~ functions for the
Bessel Polynomials - Math. Balkanica, vol. 1 (1971), pp. 292-297;
MR 44 #7008.

37. S.K. Chatterjea - Some properties of simple Bessel Polynomials from viewpoint
of Lie algebra - C. R. Acad. Bulgare Sci., vol. 28 (1975), pp. 1455-
1458; (not reviewed in MR up to 1977).
 166

38. C.K. Chatterjee - ~ Bessei Polynomials i. - Bull. Calcutta Math. Soc., vol. 49
(1957), pp. 67-70; MR 20 #3308.

39. M.P. Chen and C.C. Feng - Group theoretic origins of certain generating func-
tions for generalized Bessei Polynomials - Tamkan$ J. Math., vol. 6
(1975), pp. 87-93; bIR 51 #8495.

40. J.A. Cima - Note on a theorem of OrosswaLd - Trans. Amer. Math. Soc., vol. 99
(1961), pp. 60-61; MR 22 #11156.

41. bl.K. Das - Operational representations for the Bessel Polynomials Bull.
- Soc.
Math. Phys. Mac~doine, vol. 17 (1966), pp. 27-32 (1968); MR 39 #501.

42. M.K. Das - A generating j~nction ~br the general Bessel Polynomial - Amer. Math
Monthly, vol. 74 (1967), pp. 182-183; MR 34 #7843.

43. M.K. Das Sur les po~ynomes

- 7 ~ de Bessel, du point de vue de l'algeore
~ de Lie -
C. R. Acad. Sci. Paris, Ser. A., vol. 271 (1970), pp. 361-364; MR 42
#3330.

44. M.K. Das - Sur les potynomes

~ 4 de Bessel - C. R. Acad. Sci. Paris, Ser. A, vol
271 (1970), pp. 408-411; MR 42 #6300.

45. M.K. Das - Operational ~ b ~ u l a s connected w~"~k some classical orthogonal ~oiy-
n~nials - Bull. Math. Soc. Sci. Math. R.S. Roumaine (N.S.)vol.14 (62)
(1970), pp. 283-291; MR 49 #10936.

46. G.K. Dhawan and D.D. Paliwal - G e n e r a t i r ~ I~nctions of Gegenbx~er, Bessel and
Laguerre Polynomials - Math. Education, vol. I0 (1976), pp. A9-AI5;
MR 53 #8963.

47. D. Dickinson - On Lom~el and Bessel Polynortials - Proc. Amer. Math. Soc., vol.
5 (1954), pp. 946-956; MR 19-263.

48. K. Do~ev - On, the generalized Bessel Polyno~qials - Bulgar. Akad. Nauk. Izv.
Mat. Inst., vol. 6 (1962),pp. 89-94; MR 26 #2645.

49. M. Durra, S.K. Chatterjea, M.L. Moore - On, a class of generalized Hermite poly-
nomials - Bull. Inst. Math. Acad. Sinica, vol. 3 (1975), No. 2, pp. 377-
381; b~ 52 #11149.

50. A. Erdoly~, W. Magnus, F. Oberhettinger, F.G. Tricomi - Higher transcendental

functions (3 volumes) (Based in part on notes left by H. Bateman) -
McGraw-Hill Book Co., New York - Toronto - London, 1953; ~ 15-419 and
MR 16-586.

51. M.T. Eweida - On Bessel Polynomials - Math. Z., vol. 74 (1960), pp. 319-324;
MR 22 #8153.

52. M.T. Eweida - infinite integrals involvin~ ~ Bessei Polynomials - Univ. Nac.
Tucuman Rev. Ser. A, vol. 13 (1960), pp. 132-135; MR 24(A) #262.

53. E. Grosswald - On some algebraic properties of the Bessel Polynomials - Trans.

Amer. Math. Soc., vol. 71 (1951), pp. 197-210; b~ 14-747.

54. E. Grosswald - Addendy2n to "~n some algebraic properties of the Bessel Polyno-
mials" - Trans. Amer. blath. Soc., vol. 144 (1969), pp. 569-570; MR 40
#4246.

55. E. Grosswald - The student t-distribution for odd degrees o f f reedom is infini-
tely divisible - Ann. Probability, vol. 4 (1976), pp. 680-683; MR 53
#14591.
 167

56. E. Grosswald - The student t-distribution of any degree of ~Weedem is inj~nite-

ly divisible - Z. W a h r e s c h e i n l i c h k e i t s t h e o r i e und verw. Gebiete, vol. 36
(1976), pp. 103-109; MR 54 #14037.

57. E.A. Guillemin - Synthesis of passive Networks - John W i l e y & Sons, Inc., New
York; Chapman & Hill, Ltd., London, 1958; MR 25 #932.

58. W. Hahn - Ueber die Jacobischen Polynome und zwei verwxndte Polynomklassen -
Math. Z., vol. 39 (1935), pp. 634-638.

59. A.M. Hamza - Integrals involving Bessel Polynomials - P r o c . Ma t h. Phys. Soc.

A . R . E . , v o l . 35 ( 1 9 7 1 ) , p p . 9 - 1 5 ; MR 48 # 8 8 8 8 .

60. D. Hazony - E l e m e n t s o ¢J n e t w o r k synthesis - Reinhold Publ. Corp., New York;

Chapman & Hall, Ltd. London, 1963.

61. M.E.H. Ismail - Dual and triple sequence equations involving orthogonal poly-
nomials - Nederl. Akad. Wetensch. Proc. Ser. A, vol. 78 - Indag. Math.,
vol. 37 (1975), pp. 164-169; MR 52 #3901.

62. M.E.H. Ismail and D.N. Kelker - The Bessel Polynomials and the Student t-distri
bution - SIAM J. Math. ~lal., vol. 7 (1976), pp. 82-91; MR 52 #12164.

63. J.W. Jayne - Polynomials orthogonal on a contG~r - the Bessel alternative -

(to appear).

64. J o h n s o n , J o h n s o n , B o u d r a , S t o k e s - Filters using Bessel type polynomials -

I . E . E . E . T r a n s . C i r c u i t s a n d S y s t e m s , v o l . CAS-23, No. 2, F e b r u a r y 1 9 7 6 ,
pp. ~6-99.

65. - Infinite divisibility ~nd various mixtures of the normal distri-

D.N. K e l k e r
bution - Ann. Math. Statist., vol. 42 (1971), pp. 802-808; MR 44 #3415.

66. A.M. K r a l l - Orthogonal polynomials through moments generatin~ functionals -

SI~ J . Math. A n a l . , v o l . 9 ( 1 9 7 8 ) , p p . 6 0 0 - 6 0 3 .

67. H.L. Krall - On derivatives of orthogonal polynomials II - Bull. Amer. Math.

Soc., vol. 47 (1941), pp. 261-264; MR 2-282.

68. H.L. Krall a n d O. F r i n k - A new class of orthogonal polynomials: the Bessel

Polynomials - Trans. Amer. Math. Soc., vol. 65 (1949), pp. i00-i15;
MR 10-453.

69. V.I. Krylov and N.S. Skoblia - Handbook on the numerical inversion of the
Laplace transform (Russian) - Izdat. "Nauka i Tehnika", Minsk, 1968, 295
pages; MR 38 #1814.

70. V.N. Kublanowskaya and T.N. Smirnova - Zeros of Hankel i~unctions and other
functions associated with them (Russian) - T r u d y Mat. I n s t . S t e k l o v ,
vol. 53 (1959), pp. 186-191; MR 22 #Ii05.

71. Y.L. Luke - Special P~nctions anff their Approximations (3 volumes),Academic

Press, Inc., New York - London, 1969; MR 39 #3039 and MR 40 #2909.

72. Y.L. Luke - f.~thematical Functions and their Approx~nations - Academic Press,
Inc., New York - San Francisco - London, 1975.

73. E.B. McBride - Obtaining generating functions - Springer Tracts in Natural

Philosophy, vol. 21, Sprin~er-Verla$, New York - Heidelberg, 1971;
M R 43 #5077.
 168

74. P.J. McCarthy - Approximate location of the zeros of generalized Bessel Poly-
nomials - quart, d. Math., Oxford, Ser. (2), vol. 12 (1961), pp. 265-267;
M R 25 #243.

75. P.J. McCarthy - Characterizations of classical polynomials - Portugal. Math.,

vol. 20 [1961), pp. 47-52; MR 23 #A3866.

76. H.B. Mittal - Some generatir~~ ]hnctions - Rev. da Faculdade de Ciencias de

Lisboa, Set. 2, vol. 13 (1970), pp. 43-45; MR 46 #7600.

77. H.B. Mittal - Polynomials deigned by generaving ~hnctions - Trans. Amer. Math.
Soc., vol. 168 [1972), pp. 75-84; MR 45 #3811.

78. H.B. Mittal - Same generating functions for polynomials - Czechoslovak Math.J.,
vol. 24 [99) [1974), pp. 341-348; M R 50 #660.

79. H.B. Mittal - Unusual generating relations for polynomial sets - J. Reine
Angew. Math., vol. 271 (1974), pp. 122-137; MR 51 #949.

80. R.D. Morton and A.M. Krall - Distributional weighv functions for orthogonal
polynomials - SI~M J. Math. Anal., vol. 9 (1978), pp. 604-626.

81. M. N a s i f - Note on the Bessel Polyna~ials - Trans. Amer. Math. Soc., vol. 77
(1954), pp. 408-412; MR 16 #818.

82. N. Obreshkov - About certain orthogonal polynomials in the complex plane -

B"l~ar. Akad. Nauk Izv. Math. Inst., vol. 2 (2) [1956), pp. 45-68.

83. C.D. Olds - The simple continued ]>action expansion of e - Amer. Math. Monthly,
vol. 77 (1970), pp. 968-975.

84. F.W.J. Olver - The asymptotic expansions of Bessel Functions of large order -
Philos. Trans. Roy. Soc. London, Sec. A, vol. 247 (1954/5), pp. 328-368;
MR 16-696.
4
85. M. Parodi - Sur les Polyn~,es de Bessel - C. R. Acad. Sci. Paris , vol. 274A
(1972), pp. AII53-AII55; MR 46 #416.
2
86. A. Pham - Ngoc Dinh - Un nouveau type de devetopment des polyn~mes generalzses~
p " ~
de Besael - C. R. Acad. Sci Paris, Ser. A, vol. 272 (1971), pp. AI393-
A1396; MR 44 #2959.

87. R. Piessens - Gaussian quadrature formulas for the numerical integration of

Bromwich's integral and the inversion of the Laplace transfor~ - J.Engrg.
Math., vol. 5 (1971), pp. 1-9; MR 42 #8664.

88. F.M. Ragab - Series of products of Bessel Polynomials - Canad. J. Math., vol. ii
ii [1959), pp. 156-160; MR 21 #733.

89. E.D. Rainville - Generating functions for Bessel and related polynomials -
Canad. J. Math., vol. 5 [1953), pp. 104-106; MR 14-872.

90. E.D. Rainville - Special Functions - The Macmillan Co., New York, 1960; M R 21
#6447.

91. A.K. Rajagopal - On Bessei Polynomials - Bull. Un. Mat. Ital., vol. (3) 13
(1958), pp. 418-422; MR 21 #1408.

92. V. Romanovsky - Sur quelques classes nouuelles des polynomes orthogonaux - C.R.
Acad. Sci. Paris, vol. 188 (1929), pp. 1023-1025.
 169

93. H. van Rossum - (a) A theo~j of orthogonal polynomials, based on the Pad$
Table (Thesis), University of Utrecht, 76 pages - van Gorcum, Assen,
1953;
(b) Systems of orthogonal and quasi orthogonal polynomials connected
with the Pad~ Table, I, 1~, I1~ - Nederl. Akad. Wetensch. Proc. Ser. A,
vol. 58 - Inda$. Math., vol. 17 (1955), pp. 517-525; 526-534; 675-682;
MR 19-412.

94. H. van Rossum - Totally positive polynomials - Nederl. Akad. Wetensch., Proc.
S e r . A, v o l . 6 8 - I n d a g . Math., voi. 27 (1965), pp. 305-315; MR 3 5 # 6 8 8 0 .

95. H. van R o s s u m - A note on the location of the zeros of the generalized Bessel
Polynomials and totally positive polynomials - Niew Arch. Wisk., vol.
(3) 17 (1969), pp. 142-149; MR 41 #534.

96. P. Rusev - Convergence of series of Jacobi and Bessel Polynomials on the

boundaries of their regions of convergence (Bulgarian; Russian and
English summaries) - B"Igar. Akad. Nauk. Otdel. Mat. Fiz. Nauk. Izv.
Mat. Inst., vol. i0 (1969), pp. 17-26; MR 34 #8074.

97. H. Rutishauser - Bemerkungen zu einer Arbeit yon A1-Salom unM Carlitz - Arch.
Math., vol. 10 (1959), pp. 292-293; MR 21 # 4 2 6 2 .

98. E.B. Saff and R.S. Varga - Zerofree parabolic regions for sequences of poly-
nomials - SIAM J. Math. Anal., vol. 7 (1976), pp. 344-357; ~ 54 #3060.

99. H.E. Salzer - Orthogonal polynomials arising in the n~erical evaluation of

inverse Laplace transforms - ~ t h e m a t i c a l Tables and other Aids to
Computation (MTAC), vol. 9 (1955), pp. 164-177; MR 17-1203.

i00. H.E. Salzer - Additional formulas and tables for orthogonal polynomials
originating from inversion integrals - J . Math. and Phys., vol. 40 (1961),
pp. 72-86; MR 23B #B2612.

i01. R.R. Shepard - Active filters: Part 12 - Shortcut vo network design - Electro-
nics, August 1969, pp. 82-91.

102. N. S k o b l j a - Tables for the numerical inversion of the Laplace transforms

f(x) = ~ 1 fc+i~
c-l~ eXP F(p)dp - Izdat. "Nauka i Tehnika", Minsk, 1964,

44 pages; MR 29 # 7 2 5 .

103. N. S k o b l j a- The distribution o f zeros of polynomials connected with the nu-

merical integration of the Laplace transform (Russian) - Dokl. Akad.
Nauk BSSR 9 (1965), pp. 288-291; MR 32 #4841.

104. S.L. Soni - A note on Bessel Polynomials - Proc. Indian Acad. Sci. Sect. A,
vol. 71 (1970), pp. 93-99; MR 41 #7166.

105. H.M. Srivastava - On Bessel, Jacobi and Laffuerre Polynomials - Rend.Sem. Mat.
Univ. Padova, vol. 35 (1965), pp. 424-432; MR 33 #4346.

106. L. Storch - (a) Synthesis of constant-time-delay ladder nevwork using Bessel

Polynomials - Proc. IRE, vol. 42, pp. 1666-1676, Nov. 1954;
(b) An application of modern network synthesis to the design of con-
stant-time-delay networks with low-q elements 1954 IRE Convention
Record, Part 2, Circuit Theory, pp. 105-117; ~ 16-1182.

107. W.E. Thomson - Delay network having maximally flat frequency characteristics -
Proc. Instituto Electr. Engineers, vol. 96 (1949), Part Ill, p. 487.
 170

108. W.E. Thomson - Networks with maximally flat delay - Wireless Engineer, vol. 29
(1952), pp. 256-263.

109. L. Toscano -Funzioni generatrici di partico!ari polinomi di Laguerre e di altri

da essi dipendenti - Bull. Un. Mat. Ital., vol.(3) 7 (1952), pp. 160-
167; MR 14, p. 269.

ii0. L. Toscano - Osservazioni, confronti e c o ~ l e m e n t i su particolari polinomi

ipergeometrici - Le Matematiche, vol. i0 (1955), pp. 121-133; MR 17,
p. 733.

iii. H.S. Wall - Polynomials whose zeros have negative real parts - Amer. Math.
Monthly, vol. 52 (1945), pp. 308-325; MR 7-62.

112. J. Wimp - On the zeros of a confluent hypergeometric function - Proc. Amer.

Math. Soc., vol. 16 (1965), pp. 281-283; MR 30 #4001.

113. A. Wragg and C. Underhill - Remarks on the zeros of Bessel Polynomials - Amer.
Math. Monthly, vol. 83 (1976), pp. 122-126; ~ 52 #1146.

SUPPLEt6ENTA2Y BIBLIOGRAPHY OF TITLL~ OBTAINED AFTER COAIPLETION OF THE PRESENT MONO-

GRAPH.

114. W. Miller, Jr. - Encyclopedia of mathematics and its applications, vol. 4,

Sy~netry and separation of uariables - Addison-Wesley Publ. Co., Reading,
Mass., 1977.

115. L. Weinberg - Network analysis and synthesis - McGraw-Hill Book Co., Inc., New
York, 1955; Krieger Publishing Co., Huntington, N.Y., 1975.

116. A.H. Marshak, D.E. Johnson and J.R. Johnson - A Bessel ratior~! filter -
IEEB Trans. Circ. Supt. vol. CAS - 21 (1974), 797-799.
 GENERAL LITERATURE, NOT DIRECTLY RELATED TO BESSEL POLYNOMIALS

i. M. Abramovitz and I.E. Segun - Handbook of Mathematical Functions - Dover Publi-

cations, Inc. New York, 1968.

2. R. Askey - Dual equations and classical orthogonal polynomials - J. Math. Anal.

Appl., vol. 24 (1968), pp. 677-685.

3. I. Bendixson - Sur les racines d'une ~quation fondamentale - Acta Math., vol. 25
(1902), pp. 359-365.

4. R.P. Boas - Entire functions - Academic Press, New York, 1954.

5. R.P. Boas and R.C. Buck - Polynomials d e ~ n e d by generating relations - Amer.

Math Monthly, vol. 63 (1956), pp. 626-632.

6. W.N. Bailey - Generalized hypergeometric series - Combridge Tracts in ~ t h e m a -

tics and Mathematical Physics, No. 32 - Cambridge University Press, Cam-
bridge (England), 1935.

7. L. Bondesson - A general result on in]~nite divisibility (to appear).

8. R. Breusch - Zur Verallgemeinerung des Bertrandschen Postulates - Math. Z.,

vol. 34 (1932), pp. 505-524.

9. W. Burnside - The theory of groups o f ~ n i t e orders, 2nd edition - Cambridge

University Press, Cambridge (England), 1911.

I0. M.L. Cartwright - Integral Functions - Combridge tracts in Mathematics and

Mathematical Physics, No. 44 - Cambridge University Press, Cambridge
(England), 1962.

II. H. Davenport - The higher arithmetic - Harper Torchbook - Harper & Brothers,
New York, 1960.

t" I , • • • J ~
12. M.G. Dumas - Sur quelques cas d zrreducvzbzl~te des polynomes a coefficients
rationels - J. Math. Pures Appl. (6), vol. 2 (1906), pp. 191-258.

13. G. Doetsch - Theorie und Anwendung der Laplace Transfor~ation - Springer Verlag,
Berlin (1937).

14. G. Enestr~m - Remarques sur un tneoreme

~ ~ relatif xax racines de 1 '~
equatzon... -
Tohoku Math. J., vol. 18 (1920), pp. 34-36.

15. L. Euler - Introductio analysin in~nitori°~n, L~asanne 1 (1748), Chapter 18 -

Commentarii Acad. Scientiarum Imperialis Petropolitanae, vol. 9 (1737) -
Operae S e r . I, vol. 18.

16. E. Feldheim - Relations entre les polynSmes de Jacobi, Laguerre et Hermite -

Acta Math., vol. 74 (1941), pp. 117-138.

17. W. Feller - An introduction to probability theor~ and its applications (2 vol.),

3rd edition - John W i l e y & Sons, New York, 1967.

18. L.R. Ford - Differential equations, 2nd edition - McGraw Hill Book Co., Inc.,
New York, 19S5.

19. B.V. Gnedenko and A.N. Kolmogorov - Limit distributions for s ~ s of independent
random variables - Addison-Wesley Publ. Co., Inc., Cambridge (Mass.), 1954.

20. H.W. Gould - Combinatorial identies - Morgantown, W. Va., 1972.

 172

21. H.W. Gould - Generalization of binomial identies of Carlitz, Grosswald and

Riordan ~to appear).

22. H.W. Gould - Private communication (letter of 9 August 1977).

23. E. Grosswald - On a simple property of the derivatives of Legendre Polynomials -

Proc. Amer. Math. Soc., vol. 1 (1950), pp. 553-554.

24. E. Grosswald - Topics from the theory of numbers - The Macmillan Co., New York,
1966.

25. E. Grosswald and H.L. Krall - Evaluation of ~wo determinants - Amer. Math.
Monthly (Advanced Problems; to appear).

26. H. Hamburger - B e i t r ~ e zur Konvergenz~heorie der Stieltjesschen Kettenbr~che -

Math. Z., vol. 4 (1919),pp. 186-222.

27. H. Hamburger - Ueber eine E r ~ e i t e ~ n g des Stieltjesschen Momentproblems - Math.

Ann., vol. 81 (1920), pp. 235-319; vol. 82 (1920), pp. 120-164 and
pp. 168-187.

28. K. Hensel and O. Landsberg - Theorie der algebraischen Funktionen - Teubner,

Leipzig, 1902.

29. C. Hemnite - Sur la fonction exponentielle - C. R. Acad. Sci. Paris, vol. 77

(1873), pp. 18-24; 74-79; 226-233; 285-293; also Oeuvres, vol. 3,
pp. IS0-181.

30. I.N. Herstein - Topics in algebra - Blaisdell Publ. Co., Waltham (Mass.), 1964.

31. I.N. Herstein - Private co~unication (letter of 21 November 1951).

32. E. Hille - Ar~lytic function theor~q - Ginn & Co., Boston, 1962.
f
33. M.A. Hirsch - Sur les racines d'une equation fonde~nentale - Acta Math., vol. 25
(1902), pp. 367-370.

34. A. Hurwitz - Ueber einen Satz des Herrn Kakeya - Tohoku Math• J., vol. 4
(1913), pp. 89-93.

35. M.E.H. Ismail - Bessel functions and the inIKnite divisibility of the Student
t-distribution - Ann. Probability, vol. S (1977), pp. 582-585.

36. M.E.H. Ismail - Integral representations and complete monotonicity of various

quotients of Bessel functions - Canad. J. Math., vol. 29 (1977), pp. 1198-
1207.

37. K.E. Iverson - The zeros of the partial s~ms of e z - Math Tables and other
Aids to Computation (~AC), vol. 7 (1953), pp. 165-168.

38. D. Jackson - Fourier Series and orthogonal polynomials - Carus Monograph No. 6,
The Math Assoc. of America , Oberlin (Ohio), 1941.
• • .

39. C. J o r d a n - Sur la limite de transztzvzte des groupes non altern~s - Bull. Soc.
Math. France, vol. 1 (1872/3), pp. 40-71.

40. S. K a k e y a - On the limits of the roots of an algebraic equation - Tohoku Math.

~., vol. 2 (1912), pp. 140-142.

41. E. Laguerre - Sur la distribution dans le plan des racines d'une equation alg~-
brique, dont le premier membre satisfait ~ une Equation lineaire du second
ordre - C.R. Acad. Sci. Paris, vol. 94 (1882), pp. 412-416; 508-510;
 173

Oeuvres, vol. i, pp. 161-166.

42. J.H. Lambert - Memoire sur ~aelques proprietes remarquables des q u a n t ~ e s

transaendentes, circulaires et l o g a r i t ~ i q u e s - Histoire de l'Acad, ro£ale
des sciences et belles-lettres - Berlin, 1761 and 1768.

43. E. Landau - Hand~uch der Lehre yon der Verteilung der Primzahlen - Teubner,
Leipzig, 1909.

44. P. L~vy - T~eor~e

" ~ " de l'addition des variables aleato~res
~ " - Gauthier-Villars,
Paris, 1937.

45. M. Marcus and H. M i n c - A survey of matrix theory and matri~r inequalities -

Prindle, Weber and Schmidt, Boston, 1964.

46. F.W.J. Olver - The asymptotic solution of linear diJ~'erential equations of the
second order Ibr large values of a par~neter - Philos. Trans. Roy. Soc.
London, Set. A, vol. 247 (1954/5),pp. 307-327.
. , H
47. O. Perron - Die Lehre yon den Ketvenbruchen - Teubner, Leipzig, 1929.

48. N. du Plessis - A note abou~ the dewLuatives of Legendre's Polynomials - Proc.

Amer. Math. Soc., vol. 2 (1951), p. 950.

49. G. P~lya aad G. Szeg~ - Aufgaben und Lehrs~tze ~as der Analyse - Julius Sprin-
ger, Berlin, 1925.

80. E. Rainville - On a simple property of the derivatives of Legendre's Polyno-

mials - Unpublished manuscript (letter of 21 October 19SO).

51. I. Schur - Einige Sa~ze u~er Pz~mzahlen - Sitzun$sberichte der Preussischen

Akad. der Wissenschaften (1929), pp. 125-136; Gesammelte Abhandlungen,
vol. III, No. 64, pp.140-151.

52. I. Schur - Gleichung~en ohne A~fekt - Sitzungsberichte der Preussischen Akad.

der Wissenschaften (1930), pp. 443-449; Gesammelte Abhandlungen, vol. III,
No. 67, pp. 191-197.

53. C.L. Siegel - Transcendental Numbers - Annals of Mat~em.

' ~ StuZies No. 16
Princeton University Press, Princeton, 1949.

54. T.J. Stieltjes - Recherches sur les ~ a c t i o n s convinues - Annales de la Fac.

des Sci. Toulouse, vol. 8 (1894), 122 pages and vol. 9 (1895), 47 pages;
Oeuvres Completes, vol. 2, pp. 402-566.

55. J.A. Stratton - Electromagnetic Theory - McGraw Hill, New York, 1947.

56. G. Szeg~ - Orthogonal polynomials - Colloqui~,~ Publication, vol. 23 - Amer.

Math. Soc. , New York, 1939.

57. P.L. Tchebycheff - Sur l'Interpolation - Zapiski Akad. Nauk., vol. 4, Supple-
ment 8 (1864); Oeuvres, vol. ii, pp. 539-590.

58. O. Thorin - On the infinite divisibility of the log normal distribution -

Scand. Actuar. J. (1977), pp. 121-]48.

59. B.L. van der Waerden - Modez~ Algebra (Revised English edition) - Frederick
Unsar Publ. Co., New York, 1949.

60. J.H. Wahab - New cases of irreducibility for Legendre Polynomials-Duke Math.J.,
vol. 19 (1952),pp. 165-176.
 174

61. G.N. Watson - A treatise on the theory of Bessel Functions, 2nd edition -
Cambridge University Press, Cambridge (England), 1946.

62. L. Weisner - Group theoretic origin of certain generating functions - Pacific

J. Math., vol. 5 (1955), pp. 1033-1039.

63. L. Welsher - Generating functions for Hermite functions - Canad. J. Math., vol.
ii (1959), pp. 141-147.

64. L. W e i s n e r - Generatin~ ~ n c t i o n s for Bessel functions - Canad. J. Math., vol.

II (1959), pp. 148-155.

65. F.J.W. Whipple - Some transfo~nations of generalized hypergeometric series -

Proc. London Math. Soc. (2), vol. 26 (1927), pp. 257-272.

66. E.J. Whittaker - Sur les sertes de base de polynomes quelcon~ues - Gauthier-
Villars, Paris, 1949.

67. E.J. Whittaker and G.N. Watson - Modern Analysis, 4th edition - Cambridge
University Press, Cambridge (England), 1927.

68. D.V. Widder - The Laplace Transform. - Princeton University Press, Princeton,
1941.

69. H. Wilf - Mathematics for Physical Sciences - J. Wiley & Sons, Inc., New York
and London, 1962.
 SUBJECT INDEX

The numbezs following each entry refer to page numbers

I snands for any page in the Introduction.

Arithmetic properties of BP - 155.

Asymptotic properties, or relations - 2,4,82,93,124-130,]63.
Asymptotic series - 124,163.

Bendixson-Hirsch Theorem - 3,90,91.

Bernoulli numbers - 154.
Bernstein's theorem - I, ! 3 7 , 138.
Bertrand's postulate - 102.
Bessel alternative (orthogonality) 26,162.
"Black-Box" - 140.
Borel transform - see Laplace transform.

Cardioid - 88.
Cauchy product - 45.
Center of mass - 83,84.
Characteristic functions - !36.
Characteristic (eigen-)values - 88-91.
Christoffel constants - 147,149.
Christoffel-Darboux (type) identities - 154, Ig6.
Commutation relations - 47,48.
Continued fractions - 3,34, 59-63,81, 160.
Convergence - 64,65,66,68,69,72,74,155,157,160,162.
Cramer's rule - 20.
Cycle - !17.

Delay,
flat - !45
maximally flat - 2, 1 4 5 .
phase - 144.
time - 141,142,143.
Differential equations of BP - I, 1 , 4 - 1 7 , 3 6 , 38,85.
D1 f f e r e n t i a l equations of ~ -form - 9-12,14-15,35.
Differential equations of Sturn-Liouville form - 49.
Differential equations of y~(x) ]58.

Differential - difference relations - 159.

Discriminant - I16, IJ8, I19.
Distortion factor - 143.

Eigenvalues - see Characteristic values

Eisenstein's criterion (irreducibility) - 102.
Elements (of Newton polygons) - 10"7,109.
Expansions in series of BP - 1,46,64-74, 157, 159, 160, 162.
Expansions in series of other functions - 11,47,91.
Expansion formulae - 151-153, 160.

Filter - 141,143,144.
Fourier analysis - 133.
Fourier transforms - 26,136.
Frequency - I, 1 3 3 , 1 3 6 , 1 4 0 , 1 4 1 , 144.
Frobenius' method - 5,12.
Functions
Associate Legendre - 134,135.
Bessel - 2,3,4,!9,35,38,47,91,152,157,158.
Bessel,modified - 4,5,18,19,34,76,91,92,137,144,158.
 176

Beta - 36,38,!32.
Exponential - 1,3,67,68,7~,74,125, J 3J.
Exponential type - 6"7,69,72.
Greatest integer - 1,97,()8, I02, 132.
Hankel (see a!sc Bessel) - 91,92.
Hanke], modified (see a] s c Besse!, modified) - I.
Hypergeometric - 2,34,38,[50.
Hypergeometric, generalized - 34.
Lebesgue integrable - 136.
MacDonald - see Besse] , or Hanke] , modified.
Meijer's G - 157.
Positive real (p.r.) 141,147.
Step - 154.
Whittaker - 34, 37, 38, ]53.

Gauss-Lucas theorem - 84.

Generating functions - 1,2, 36,4]-50,z7,58,67,70, 150, 131, 153, 160, 162.
Generating functions, bilJnear - 44.
Gershgorin's theorem - 9,88,90,01.
Gram-Schmidt method - 25.
Group, Alternating - 116-118,120,122.
Galois - 1,2, i16-123, 163.
Primitive - I[A-I13, ~ 20-!22.
Transitive - 116-!18,120,121.
Symmet£ic - i16, i~$, 120-122, l(i3.

Hermitian matrices - 90.

Impedance (driving point, complex) - 2,141,142.

Indiciai equation - 5.
Infinite divisibility - I, 3, ! ] 6 - 1 3 9 .
Inner product - 25,26.
Inversion formula - 66.
Inverse transform - see !.aplace transform.

Irrationality of e r (r razional~ i, 1 2 3 - ] 31 .
Irrationality of ~ - I, 1 3 1 - 1 3 3 .
Irreducibility - 2,9g-i 17, 120, 12!, 123, 154, !63.

Kronecker delta - 1,26,32,146,154.

Lagrange interpolation - [46.

Laguerre's theorem - 82,83,84.
Laplace (or Borel) transofrm arid inverse transform - 1,66,67,69, 137, 138,
!40-142,144-149.
Lap!acian - see Operators, Laplacian.
Lebesgue constants - 162.
Legendre's equation - I, 1 3 4 .
Leibniz's rule - 16,29,53,55,132.
Lemniscate - 70,71.
Lie group - 41,46,47.
Lie algebra - 48.

Measure - 137.
Moments 25-32,51,53.
Monotonicity, Monotonic - 1,68,75,78,80.
Monotonity, complete - 1,3, 136-139.
Multiplication theorems - !53.

Networks, electrical -
2,3, 139-14b.
Networks, "Bessel"case - 142.
Networks, passive - 141, 142.
 177

Networks, R-L-C - !43.

Newton ~dentities (coefficients, zeros) - 93,94,97,98.
Newton polygon - 100-115.

Ohm-Kirchhoff laws, equations - 13g-1.q7 .

Operators
Difference - 152,156.
Differen%iai - 1,9-12, 14, 15, 35,46,47,51, 133, 155-157, 160.
l aplacian - I, 1 3 3 .
Shift - 48,49.
Other - 159,!60.
Operational formulae - 155,157,160.
Orthogonality (see also Polynomials, orzhogonal and the Bessel a] t e r -
native) 1,25-32,64, 135, 154, 155, !58, !60.
Orthonormality - 26.

Pad6 table and approximanzs - 3, 1 6 0 .

Permutation - 117,121.
Po]ar axis - I, ] 3 3 .
Pole (of center of mass) - 83,84.
Polynomials
Appe!i (classical, or generalized) 65-68,70.
Associate BP - 153.
Basic BP -151.
Genera]ized BP - 12-~7, 22-24,27,29-~2, 35-38,42-45,49-51,54,5[~,64,
65,70,75,80-82,85-88,93,94, 125,128-130, 148, ~49,
15~-!63.
Hermite - 26,45, 154.
Hurwitz - 141-~43.
Jacobi - 26,34, 35,38,43, 15] , 154, ~60.
Laguerre - 26,34,36,38,44,152,154,155,!60.
Legendre - 1,3g, 134, 135, ~45, 160.
Lommel - 34,38,46,81.
Monic - 99,103,1!8,119,123.
Orthogonal, c]assica] - 3,26,34,47,64,154,162.
Orthogonal, other - 153, 157-160.
Reverse BP - 7,8,12.
Totally positive - 160.
Probability distribution, densizy function . 136.
Pseudogenerating functions of BP - 1,42.

Quadrature, formulae - 146,148.

Quadrature, Gaussian - 145,147.

Random variable - 136.

Recurrence relations - 1,2,9,]8-24,36,5(],60,86,87,94,96,143,i5!-153,
!55, [58, 162.
Re] ations of BP to other functions - 2,54-40,154,i55,[57,158.
Representation formulae
of BP by determinants - 3,20-22,88, 155.
of BP by other functions - 35-38,155.
of BP by integrals - ] 5 ! , 1 5 6 , 15"I, i 6 0 , 1 6 1 .
of other functicns by BP - 46.
others - 45,46,151,153,155.
Resultant - 118,119.
Rodrigues (type) formulae - 16,36,51-58.

Scheme of factorization - ± O 9 - ] I I, 1 1 3 , lJ 5.
Scheme of factorization, compatibility or incompatibility - ~09,110,
112,115.
Schmidt method - see Gram-Schmidt method.
 178

Separation of variables - I, 1 3 3 , ~ 34.

Singularities, singular points - 4,69,!48.
Special functions - 3.
Spots - IOO, 104-iO9, I~ I.
Stifling formula - I, 1 2 4 , 1 2 5 , 1 2 7 , 1 2 9 , 1 3 1 , 1 3 2 .
Stirling numbers - 9.
Student t-distribution - I, i 3 6 - I 3 9 .
Summability - 64-66,69,72,74,155,[62.
Synthesis (of a network) - 3,140.

Transfer function - 2, 1 4 1 - 3 4 3 , ]45.

Transposition - 122.
Truesdell equation - 160.
Turin inequality - 155,[56.

Vector space - 25.

Wave(s) - I, 1 3 3 , 1 3 6 .
Wave equation - I, 1 3 3 .
Wave, monochromatic - I, [ 3 3 .
Wave, propagating - I.
Wave, stationary - I.
Weight functions, w(x) , p(x) , etc. - 25-27, 30-'~2,51-53,64, 146, 148, 153,
155,15~, 160.
Whittaker basis series - 69.

Zeros of BP - 1,2,3, 137, 145, 148, ] 56, 157.

Zeros of BP, location - 2, 3 , 7 5 - 9 3 , 125, 138, 154, 155, 157, 159, 160, 162, 163.
Zeros of BP, properties - 1,2, 3,75-93, 129, 145, ~54, 163.
Zeros of BP, tabulations - 2,143,179.
 NAME INDEX

The numbers following each name stand for the page numbers; the letters
F and I stand for any page of the Foreword, or of the Introduction, re-
spectively.

W.H. Abdi - F, 150, 164. F.M.G. Eisenstein - 102.

M. Abramovitz - F,171. G. Enestr6m - 77,171.
R.P. Agarwal - F,2, 149, 151, 164. A. Erd@lyi - 166.
W.A. Ai-Salam - F,2,43,44,73, 149, L. Euler - 171.
151,154,158-160,164. M.T. Eweida - 157,166.
R. Askey - 158,171.

V
Baieev - 155,164. E. Feldheim - 43,171.
N. Bailey - 171. W. Feller - 171.
P. Banerjee - 155,164. C.C. Feng - 49,166.
W. Barnes - 3,164. L.R. Ford - 171.
Bendixson - 3,90,171. O. Frink - F,I, 1,2,7, 12,24,44,64,
R. Berlekamp - 121. 145,149,152,158,162,167.
Bernstein - 1,137.
R.P. Boas, Jr. 2,8,64,65,67,162,
164,171. 8. Gershgorin - 3,88,90.
S. Bochner - 1,164. B.V. Gnedenko - I, 171.
L. Bondesson - 136,171. H.W. Gould - F,57, 171,172.
E. Borel - 66. E. Grosswald - 1,2,3,39, 121, 166,
P.W. Boudra, Jr. - 167. 167,172.
F. Brafman - 2,45,46,149,153,164. V. Grosswald - F.
R. Breusch - 102,171. E. Guillemin - 3,167.
J.D. Brillhart - 121.
J.W. Brown - 164.
R.C. Buck - 2,8,64,65,67,!62,164, W.Hahn - 1,167.
171. H. Hamburger - 25,172.
J.L. Burchnall - 1,2,6,10,15,42,43, A.M. Hamza - 158,167.
49,51,65,70,76,93,94,143,149, D. Hazony - 3,167.
156,164. K. Hensel - 172.
W. Burnside - 117.171. C. Hcrmite - 1,1,26,172.
I.N. Herstein - 39,172.
E. Hille - 172.
L. Carlitz - 2,45,57,73,74,154-156, M.A. Hirsch - 3,90,172.
160, 164,165. A. Hurwitz - 77,78,141,142,143,172.
B.C. Carlson - 165.
M.L. Cartwright - 171.
A.L. Cauchy - 117. M.E.H. Ismail - F,I, 3, 1 3 7 , 158, 167,
S.K. Chatterjea - 46,49,154,155,]57, 172.
165,166. K.E. Iversen - 172.
C.K. Chatterjee - 156,166.
T.W. Chaundy - 1,2,6, !O, J64.
M.P. Chen - 49,!66. D.Jackson - 172.
T.S. Chihara - 154,!64. J.W. Jayne - 26,167.
J.A. C±ma - 155,157,166. D.E. Johnson - 167.
J.R. Johnson - 167.
C. Jordan - 1i7,172.
M.K. Das - 46,157,166.
H. Davenport - 171.
R. Dedeking - 117. S. Kakeya - 77,172.
G.K. Dhawan - 166. S. Karlin - 155.
D. Dickinson - 2,38,46,73,81, J 57, D.N. Kelker - 1,3, 136, 137, 158, 167.
160,166. K. Kloss - 121.

K. Do~ev - 2,!3,81,86,124,125,128, A.N. Kolmogorov - 1,171.

130,157,159,!62,166. A.M. Krall - F,26, 167, 168.

H.L. Krall - F,I, 1,2,7, 12,24,44,64,
G. Doetsch - 171.
145,149,152,158,162,167,172.
M.G. Dumas - 1OO,171.
M. Dutta - 157,166. V.I. Krylov - 149,167.
V.N. Kublanowskaya - 149,167.
 180

Laguerre - 26,82,83,84,172. T J. Stieltjes - 25,173.

H. Lambert - 59,173. V P. Stokes - 167.
Landau - 173. L Storch - 3,144,169.
Landsberg - 172. J A. Stratton - 173.
S. Laplace - I. G Szeg5 - 82,173.
M. Legendre - I, iO3. P Szeg5 - J55.
L@vy - 1,173.
Lindel~f - 66.
L. Luke - 149,167. P.L. Tchebycheff - 25,102,173.
W.E. Thomson - 2,3,143,169,170.
O. Thorin - 136,173.
Magnus - 166. L. Toscano - 2,3,160,170.
Marcus - 173. E.G. Tricomi - 166.
B. McBride - 46,167.
McCarthy - 2,81,159,168.
Miller, Jr. - 170. C. Underhill - F,3,90, 170.
Minc - 173.
B. Mittal - 168.
L. Moore - 157,166. R.S. Varga - 2,86,162,169.
D. Morton - 26,168.

B.L. van der Waerden - 173.

M. Nasif - 2,65,81,168. J H. Wahab - i07, !73.
M. Newman - 121. A S. Wall - 170.
G N. Watson - 174.
L Weisner - 46,174.
F. Oberhettinger - 166. F J.W. Whipple - 45,174.
N. Obreshkov - !3,124,157,159,168. E J. Whittaker - 66,69,174.
C.D. Olds - 1,168. D.V. Widder - 174.
F.W.J. Olver - 2,91,92,93,162,168, H. Will - 174.
173. J. Wimp - 2,81, ~62, !70.
A. Wragg - F, 3,90, 170.

D.D. Paliwal - 166.

M. Parodi - 3,88,89,90,162,168.
O. Perron - 173.
A. Pham-Ngoc Dinh - 159,168.
E. Phragm@n - 66.
R. Piessens - 149,168.
N. du Plessis - 39, 173.
G.P61ya - 66,82,173.

F.M. Ragab - 2,149,159,168.

E.D. Rainville - 2,39,43,151,168,
173.
A.K. Rajagopal - 160,168.
V. Romanovsky - 1,168.
H. van Rossum - 160,169.
P. Rusev - 160,169.
H. Rutishauser - 160,169.

E.B. Saff - 2,86,162,169.

H.E. Salzer - 145,149,169.
I. Schur - 39,102,116,117,173.
I.E. Segun - F, 171.
R.R. Shepard - 3,169.
C.L. Siegel I, 131,173.
G. Sizemore-Ballard - F.
N. Skoblja - 149,167,169.
T.N. Smirnova - 149,167.
S.L. Soni - 160,169.
H.M. Srivastava - 160,169
R. Stauduhar - 121.
 PARTIAL LIST OF SYMBOLS

£ field of complex numbers

Z - the rational integers
Z+ - the positive integers
Jv(z),I (z) ,K (z) - standard notations for Bessel and MacDonald
functions
H (I) (z) H (2) (z) standard notations for Hankel functions

Yn (z) , Yn (z ; a, b) Resse] Polynomial, simple and generalized,

standardization of Krall and Frink.
On(Z),en(Z;a,b) Bessel Polynomial, simple and generalized,
standardization of Burchnall.
Pn(a'b) (z) - Jacobi Polynomial

Ln (x) Legendre Polynomial

Ln(a) (z) Laguerre Polynomial

Hn(Z) - Hermite Polynomial

Rn,~(z) Lommel Polynomial

pq(z) associate Legendre function

Wk,m(Z) - Whittaker function

pFq(a 1 . . . . . ap;b 1 . . . . . bq;Z) generalized hypergeometric function

r(z) gamma function

B(u,v) - beta function
Ix] greatest integer function
erfe(z) complementary error function
w(x) ,p (x;a,b) - notations for weight functions
T(s) - transfer function
Z(s) - driving point impedance
An(S) distortion factor

An = [An(i~to) [ - size of the distortion factor for s = i~t

o
A Laplacian operator d
differential operator x ~-~

R - right shift operator

L - left shift operator

L(x, d - linear, ordinary differential operator that

El' ~) depends also linearly on a parameter

a - difference operator with respect to the

parameter a.

L(f) (t) - the Laplace transform of f(x) at t

L- 1 (g) (x) - the inverse Laplace transform of g(t) at x

Mk(f,P) - k-th moment of the function f with weight

function p
 - binomial coefficient

[ a o , a I, •.., a n , •.. ] - continued fraction

B - n-th Bernoulli number

n

A!n) - Christoffel constants

1

s(m) - Stirling numbers of the second kind

n

(~..
±3 - takes only the values O, or i; w h e n used
specifically as a Kronecker delta
(6ij = i if and only if i = j), this is

stated explicitly.

~k(n) a(n)(a,b)
' k - k-th zero of yn(Z), or of yn(z;a,b), respectively.

[n) B~n) (a,b)

~k ' - k-th zero o f @n(Z), or of @n(Z;a,b), respectively.

r - sum o f r-th powers of the zeros o f yn(Z)

(a,b) - greatest common divisor of a and b

[a,b] - spot of coordinates a,b
alb - a divides b
e m (P) - the exact power of p that divides m

a-b - a is congruent b with respect to some modulus

f~ g - asymptotic, or formal e q u a l i t y
f = O(g)
- means If(z)/g(z)' < C, some constant
P - p e r m u t a t i o n on n symbols
n

(a,b,c) (d,e) cycles of a permutation; also used for a

scheme of factorization.
N.p. - Newton polygon

A* - complex conjugate of the m a t r i x A

- notations for c h a r a c t e r i s t i c (= eigen) values
of a m a t r i x
R(f,g) - resultant o f the polynomials f and g
D, Dn - denotes u s u a l l y a discriminant, D also a
d i f f e r e n t i a t i o n operator
(n), ap (n) - sum of the p-adic digits o f n

~z - center of mass with respect to the pole z

a a (n)
m' 111 - m-th coefficient of the n - t h pol}momial, often
o f yn(Z), or @n(Z)

a m,(a)m,or a(m ) - abbreviation for a(a+l)...(a+m-l)

a (m)
- abbreviation for a(a-l)...(a-m+l)