AMERICAN MATHEMATICAL SOCIETY

COLLOQUIUM PUBLICATIONS
VOLUME XXIII

ORTHOGONAL POLYNOMIALS

BY
GABOR SZEGO
PROFESSOR OF MATHEMATICS
STANFORD UNIVERSITY

010214

PubLished by Ihe
American Mathematical Society
Providence, Rhode Island
1939
 First edition, 1939
Revised edition, 1958
Third edition, 1967
Third edition, second printing, 1974
Fourth edition, 1975

Library of Congress Catalog Number 39-33497

International Standard Book Number 0-8218-1023-5

Printed in the United States of America

TO MY WIFE
 PREFACE
Recent years have seen a great deal of progress in the field of orthogonal
polynomials, a subject closely related to many important branches of analysis.
Orthogonal polynomials are connected with trigonometric, hypergeometric,
Bessel, and elliptic functions, are related to the theory of continued fractions
and to important problems of interpolation and mechanical quadrature, and
are of occasional occurrence in the theories of differential and integral equations.
In addition, they furnish comparatively general and instructive illustrations of
certain situations in the theory of orthogonal systems. Recently, some of these
polynomials have been shown to be of significance in quantum mechanics and
in mathematical statistics.
The origins of the subject are to be found in the investigation of a certain
type of continued fractions, bearing the name of Stieltjes. Special cases of these
fractions were studied by Gauss, Jacobi, Christoffel, and Mehler, among others,
while more general aspects of their theory were given by Tchebichef, Heine,
Stieltjes, and A. Markoff.
Despite the close relationship between continued fractions and the problem
of moments, and notwithstanding recent important advances in this latter
subject, continued fractions have been gradually abandoned as a starting point
for the theory of orthogonal polynomials. In their place, the orthogonal
property itself has been taken as basic, and it is this point of view which has been
adopted in the following exposition of the subject. Choosing this same basic
property, we discuss certain special orthogonal polynomials, which have been
treated in great detail independently of the general theory, and indeed, even
before this theory existed at all. In this connection we add the names of La-
place, Legendre, Fourier, Abel, Laguerre, and Hermite to those previously
mentioned.
As regards treatises on the subject, we note that the only systematic treat-
ment thus far given is found in J. Shohat's monograph, Theorie Generale des
Polynomes Orthogonaux de Tchebichef, Memorial des Sciences Mathematiques,
Paris, 1934. Limitations of space have compelled that work to be brief, and
consequently, it does not enter into a detailed treatment of many problems
which have been especially advanced in recent years. It has therefore seemed
desirable to attempt a new and detailed development of the main ideas of this
field, devoting, in particular, some space to recent investigations of the distribu-
tion of the zeros, of asymptotic representations, of expansion problems, and of
certain questions of interpolation and mechanical quadrature.
In what follows, \Ve are concerned partly with the general theory of orthogonal
polynomials, and partly with the study of special classes of these polynomials.
As might be expected, we have more exhaustive results for these special classes,
and we cite as an instance the classical polynomials satisfying linear differential
v
Vl PREFACE

equations of the second order. Also, when the primary importance of these
special classes in applications is taken into account, it should not be at all sur-
prising that the present book is mainly devoted to their study. The general
theory, however, as developed in Chapters XII and XIII, doubtless represents
the most important progress made in recent years.
In the present work, no claim is made for completeness of treatment. On the
contrary, the aim has purposely been to make the material suggestive rather
than exhaustive. An attempt has been made to indicate the main and charac-
teristic methods and to point out the relation of these to some general ideas in
modern analysis. As a rule, preference has been given to those topics to which
we were able to make some new, though modest, contributions, or which we
could present in a new setting. Thus the book contains a number of results not
previously published, some of which originated several years ago. For instance,
we have included a discussion of the Cesaro summability of the Jacobi series at
the end-points of the orthogonality interval (the method used here is of interest
even in the classical case of Legendre series). Further, a new and simpler ap-
proach has been given to S. Bernstein's asymptotic formula for orthogonal
polynomials. We also refer to certain details of minor importance, such as:
simplifications and additions in the asymptotic investigation of Jacobi and
Laguerre polynomials and in the discussion of the expansions in terms of these
polynomials; the discussion of the cases in which the Jacobi differential equation
has only polynomial solutions; the evaluation of the number of zeros of general
Jacobi polynomials in the intervals [- oo, - 1], [- 1, + 1], [ + 1, + oo J; a new
proof of the Heine-Stieltjes theorem on linear differential equations of the second
order with polynomial coefficients and polynomial solutions, and so on.
In general, we have preferred to discuss problems which may be stated and
treated simply, and which could be presented in a more or less complete form.
This was the main reason for devoting no space to the extremely interesting
urithmetic and algebraic properties of orthogonal polynomials, such as, for
instance, the recent important investigations of I. Schur concerning the irre-
ducibility and related properties of Laguerre and Hermite polynomials. Fur-
thermore, we hav~ attached great importance to the idea of replacing incomplete
and overlapping theorems, scattered in the literature, by complete results
involving only intrinsic or necessary restrictions. We have also tried to exploit,
as far as seemed to be at all possible, definite methods, such as, for instance,
Sturm's methods in differential equations (see §§6.3, 6.31, 6.32, 6.83).
A complete treatment of Legendre polynomials was not feasible, and probably
not desirable, in the framework of the general theory. Besides, there are al-
ready complete treatises on spherical and other harmonics. 1 We have selected
and considered only those properties of Legendre polynomials which are the
starting points of generalizations to ultraspherical, Jacobi, or to more general
polynomials. Another subject which could not be included was Stieltjes'

1 For instance, E. W. Hobson 1 (see bibliography).

 PREFACE Vll

problem of moments, which has been omitted in spite of its great interest;
for this subject would have necessitated the development of a complicated
apparatus of results and methods. Orthogonal polynomials of more than one
2
variable also have not been treated.
The book is based on a course given at Washington University during the
academic year 1935-1936. Acquaintance with the general ideas __and methods
of the theory of functions of real and complex variables is naturally required.
Occasionally, Stieltjes-Lebesgue and Lebesgue integrals are considered. In the
greater part of the book, however, these integrals have been avoided, and, except
in a very few places, no detailed properties of them were used.
The problems at the end of this book are, with few exceptions, not new, and
they are not interconnected as are, for instance, those in P6lya-Szego's A ufgaben
und Lehrsatze. They are more or less supplementary in character and serve as
illustrations and exercises; they sometimes differ widely from one another both
as to subject and method.
The list of references is not complete; it contains only original memoirs, a few
text books of primary importance, and monographs to which references are made
in the text.

For the suggestion of preparing a book on orthogonal polynomials for the

Colloquium Publications, I am indebted to Professor J. D. Tamarkin, who has
also participated in the present work by offering a great number of valuable
suggestions. It is with the greatest gratitude that I mention his friendly
interest.
I have also received valuable advice from my friends and teachers L. Fejer
(Budapest), and G. P6lya (Zurich). My colleagues P. Erdos (Manchester),
G. Grunwald (Budapest), vV. H. Roever (St. Louis), A. Ross (St. Louis), J.
Shohat (Philadelphia), and P. Turan (Budapest) gave generously and unstint-
ingly of their time. F. A. Butter, Jr. (at present in Los Angeles) collaborated
with me in the preparation of the manuscript. This last aid was made possible
through a grant from the Rockefeller Research Fund of Washington University
(1936-1937). My student L. H. Kanter also rendered valuable assistance in
the preparation of the manuseript.
My gratitude for the encouragement and help of these friends, colleagues, and
institutions can hardly be measured by any formal acknovvledgment. Lastly,
I wish to express to the American Mathematical Society my great appreciation
for the inclusion of the present book in its Colloquium Series.
G. SzEGO
vVASHINGTON UNIVERSITY, 1938.

2 Cf. the bibliography in Jackson 8, p. 423.

 PREFACE TO THE REVISED EDITION
The first printing of this book published in 1939 was about exhausted in 1948.
Reprinting was arranged then but for various reasons no change in the text was
made. During the past twenty years since the preparation of the original
edition was completed, considerable progess was made in this field. A glance
at the pertinent section of the Mathematical Reviews suggests that the interest
in this topic is still very much alive. Systematic treatment of orthogonal
polynomials has been incorporated in various modern texts published in the
meantime. We refer only to the Higher Transcendental Functions published by
the Bateman Manuscript Project Staff ( cf. in particular, vol. 2, Chapter X,
edited by Professor A. Erdelyi), and to the book of F. Tricomi, Vorlesungen
iiber Orlhogonalreihen (Chapters IV-VI).
Recently the council of the American Mathematical Society has authorized
the author to prepare a revised edition of the book, adding a moderate amount
of material in order to bring it up to date. Naturally, limitations of space and
time did not allow including all new results (or, for that matter, the old ones
which were missing from the original edition). Only a few particularly interest-
ing new items have been added as well as some details which deserve attention
because of elegance of the method or originality of ideas. We mention here in
particular the important Pollaczek polynomials; they are treated in an Appendix.
Further new material was incorporated in the form of Problems and Exercises.
New bibliographic items have been included, again in a rather selective way.
Finally, misprints have been corrected and numerous minor improvements and
additions made.
The author recollects again, as was stated in the Preface of 1938, that the prep-
aration of this book was suggested to him by the late Professor J. D. Tamarkin.
Since his untimely death in 1945 his name is not too frequently mentioned. It is
justified and probably necessary to remind the younger mathematical gener-
ation, in the rush of modern developments, how much American mathematics
owes to his great energy and far-sighted intelligence.

STANFORD UNIVERSITY, 1958 G. SZEGO

ix
 PREFACE TO THE THIRD EDITION
The interest of the mathematical community for orthogonal polynomials,
classical and non-classical, is still not entirely exhausted. During the past
years I lectured about this subject several times at Stanford. The attendants
of the course were upper division and graduate students, specializing in
mathematics, mathematical statistics, calculus of probability, etc.
Only minor changes have been made in the text. I owe numerous im-
provements and corrections to various friends and colleagues. I mention
particularly Professor Paul Tur{m (Budapest, Hungary) and Professor
Lee Lorch (Edmonton, Canada). New references, published in the time
interval 1958-1966, have been included.

STANFORD UNIVERSITY, 1966 G. SzEGO

PREFACE TO THE FOURTH EDITION

Again the American Mathematical Society has taken the initiative to

reprint the present book, allowing some minor changes and new material.
Among the persons interested in the field of orthogonal polynomials who
have contributed to these changes and additions, I mention with particular
indebtedness my friend and colleague Professor Richard Askey (Madison,
Wisconsin) and the very active and original group of mathematicians. around
him. A very important set of lectures by Askey entitled, "Orthogonal
Polynomials and Special Functions," reached me too late to be incorporated
in the present edition.
Further material has been furnished by Professor Paul Tur{m (Budapest,
Hungary) and Professor Lee Lorch (Toronto, Canada). New problems and
exercises have also been included. Peter Szego (Redwood City, California)
gave me valuable assistance in preparing the present manuscript.
My gratitude goes to aJJ these friends and colleagues.
STANFORD UNIVERSITY, 1975 G. SZEGO

xi
 TABLE OF CONTENTS
P.\GI!;

PREFACE ........... . v
PREFACE TO THE REVISED EDITION ix
PREFACE TO THE THIRD EDITION. . xi
PREFACE TO THE FOURTH EDITION xi
CHAPTER I. PRELIMINARIES 1
CHAPTER II. DEFINITION OF ORTHOGONAL POLYNOMIALS; PRINCIPAL EXA:\IPLES 23
CHAPTER III. GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS 38
CHAPTER IV. JACOBI POLYNOMIALS . . . . . . . . 58
CHAPTER v. LAGUERRE AND HERMITE POLYNOMIALS 100
CHAPTER VI. ZEROS OF ORTHOGONAL POLYNOMIALS 111
CHAPTER VII. INEQUALITIES . . . . . . . . . . 159
CHAPTER VIII. ASYMPTOTIC PROPERTIES OF THE CLASSICAL POLYNOMIALS 191
CHAPTER IX. EXPANSION PROBLEMS ASSOCIATED WITH THE CLASSICAL POLYNOMIALS 244
CHAPTER X. REPRESENTATION OF POSITIVE FUNCTIONS . . . 274
CHAPTER XI. POLYNOMIALS ORTHOGONAL ON THE UNIT CIRCLE 287
CHAPTER XII. ASYMPTOTIC PROPERTIES OF GENERAL ORTHOGONAL POLYNOMIALS 296
CHAPTER XIII. EXPANSION PROBLEMS ASSOCIATED WITH GENERAL ORTHOGONAL
POLYNOMIALS . . . . 313
CHAPTER XIV. INTERPOLATION 329
CHAPTER XV. MECHANICAL QUADRATURE 349
CHAPTER XVI. PoLYNOMIALs oRTHOGONAL oN AN ARBITRARY cuRvE 364
PROBLEMS AND EXERCisgs 377
FURTHER PROBLEMS AND EXERCISES 387
APPENDIX 393
LIST OF REFERENCES 401
FURTHER REFERENCES 416
INDEX . . . . . . . 425

xiii
 CHAPTER I

PRELIMINARIES

1.1. Notation
Numbers in bold face type, like 1, refer to the bibliography at the end of the
book. The system of section numbering used is Peano's decimal system, and
the numeration of formulas starts anew in each section. Thus, reference to §9.5
and (9.5.2) means section 9.5 in Chapter IX and formula (9.5.2) in the same
section, respectively. A similar numeration has been used for the theorems.
We use the symbol: Onm = 0 or 1, according as n ~ m, or n = m.
The closed real interval a ~ x ;£ b (a and b finite) will be denoted by [a, b].
The same symbol is used if either a orb is infinite or if both are; in this case the
equality sign is excluded.
We often write for a real x
(1.1.1) sgn x = - 1, 0, + 1,
according as x is negative, zero, or positive; more generally, for arbitrary com-
plex x, x ~ 0, we write
(1.1.2) sgn x = I x l-1 x.
The symbol i denotes the conjugate complex value, ~(x) the real part, and
3(x) the imaginary part of the complex number x.
If two sequences Zn and Wn of complex numbers have the property that Wn ,= 0
and Zn/Wn ----+ 1 as n ----+ oo, we write Zn ""' Wn. If Zn and Wn are complex, Wn ~ 0,
and the sequence I Zn l/1 Wn I has finite positive limits of indetermination, we
Write Zn r-...~ Wn .
Occasionally we make use of the notation
(1.1.3)
if an > 0, to state that Zn/an is bounded, or tend~ to 0, respectively, as n----+ oo.
A similar notation is used for a passage of limit other than n ----+ oc.
A function f(x) is called increasing (strictly increasing) if x 1 < X2 implies
f(xt) < j(x2); it is called non-decreasing if Xt < X2 implies J(xt) ~ j(1:2). An
analogous terminology '"ill be used for decreasing functions.
Let p ~ 1, and let a(x) be a non-decreasing function in [a, b] which is not
constant. The class of functions f(x) whieh are measurable with respect to
a(x) and for which the Stieltjes-Lebesgue integral f~ J f(x) JP da(x) exists (see
§1.4) is called L~(a, b). In case a(x) = x we use the notationLv(a, b); in case
p = 1, a(x) arbitrary, the notation La(a, b) is used. If f(x) and g(x) belong to
the class L~(a, b), the same is true for f(x) + g(x). (Cf. Kaczmarz-Steinhaus 1.
pp. 10-11.)
1
 2 PRELIMINARIES [I l

1.11. Inequalities
(1) Cauchy's inequality. Let (a.l, {b.l, v = I, 2, . · . , n, be two systems of
complex numbers. Then

(1.11.1)

The equality sign holds if and only if two numbers >-., IJ., not both zero, exist
such that >-.a.+ ~J.b. = 0, v = 1, 2, ... , n.
(2) Schwarz's inequality. Letf(x) and g(x) be two functions of class L'!(a, b).
Thenf(x)g(x) is of class La(a, b), and

I ib f(x)g(x) da(x) ~ [b [b
2

(1.11.2) 1 I f(x) f da(x) I g(x) 1

2
da(x).

(3) Inequality for the arithmetic and geometric mean. If f(x) > 0, we have

lb f(x) da(x) > {ib log f(x) da(x)}

(1.11.3)
J.' da(x) ~ exp f.' da(x) '

provided all integrals exist, and f~ da(x) > 0. (Cf. Hardy-Littlewood-P6lya

1, pp. 137-138.)
(4) Abel's transformation and Abel's inequality. From

(1.11.4)
fogo + f1g1 + · · · + fngn
= (Jo - f1)Go + (!1 - f2)G1 + · · · + CJn-1 - fn)Gn-1 + fnGn 1

where
(1.11.5) G. = go + g1 + · · · + g. , v = 0, 1, 2, · .. , n,
we obtain, assuming fo ~ f1 ~ · · · ~ fn ~ 0, and I G. I~ G, v = 0, 1, · · · , n,
the inequality
(1.11.6)

(5) Second mean-value theorem of the integral calculus. Let f(x) ~ 0 be a

non-increasing function, and let g(x) be continuous, a ~ x ~ b, a and b finite.
Then

(1.11.7) [b f(x)g(x) dx = f(a + 0) 1~ g(x) dx,

1.12. Polynomials and trigonometric polynomials
We shall consider polynomials in x of the form
(1.12.1)
 [1.2] TRIGOXOMETRIC POLY~O.:\IIALS 3

with arbitrary complex coefficients Co, c1, c2 , ... , cm. Here m is called the
degree; and if Cm =;t. 0, the precise degree of p(x). In what follows an arbitrary
polynomial of degree m will be denoted by 11"m. If p0 (x) 1 p1 (x), ... , Pn(x) are
arbitrary polynomials such that Pm(x) has the precise degree m, every 7rn can be
represented as a linear combination of these polynomials with coefficients which
are uniquely determined.
A trigonometric polynomial in 0 of degree m has the form
(I.I2.2) g(O) = ao + a1 cos 0 + b1 sin 0 + · · · + am cosmO+ bm sin mO,
with arbitrary complex coefficients. Here m is again called the degree of g(O);
m is the precise degree if I am I + I bm I > 0. According as all the b~< or all the a~<
vanish, g(O) is referred to as a cosine or a sine polynomial.
The functions cosmO and sin (m + 1)0/sin 0 are polynomials in cos 0 = x of
the precise degree m and are called Tchebichef polynomials of the first and second
kind, respectively. These polynomials play a fundamental role in subsequent
considerations. Setting

(l.I2.3) cosmo= T m (cos 0) = T m(x), sin (m• + I) 0 -_ U m (COS O) -_ U m (X)

1
sm 0
we see that any cosine polynomial of degree m is a polynomial of the same degree
in cos 0 = x, and conversely. Any sine polynomial of degree m, divided by
sin 0, furnishes a cosine polynomial of degree m - I. Thus, a sine polynomial
can be represented as the product of sin 0 = (1 - x 2) 112 by a polynomial in
COS 0 = X.
The polynomials (I.I2.3) are special cases of the so-called Jacobi polynomials
(cf. Chapter IV). They contain only even or only odd powers of x according
as m is even or odd. Thus cos (m + !)O/cos (0/2) and sin (m + !)O/sin(0/2)
are cosine polynomials in 0 of degree m; they are also connected with the
Jacobi polynomials (see (4.1.8)).
We define the "reciprocal" polynomial of (l.I2.I) by
(l.I2.4)
If the zeros of p(x) are x1 , X2, · · · , Xm, those of p*(x) are x't, x:, . · · , x!, where
x: == x;;- 1 is the point which is obtair.ed from x~< by inversion with respect to the
unit circle I x I = I in the complex x-plane. The zeros must be counted accord-
ing to their multiplicity, and O* = oo, oo * = 0; oo as a zero of order k means
that the coefficients of the k highest powers vanish.
1.2. Representation of non-negative trigonometric polynomials
THEOREM 1.2.1. Let g(O) be a trigonometric polynomial with real coefficients
which is non-negative for all real values of 0. Then there exists a polynomial p(z)
of the same degree as g(O) such that g(O) = I p(z) 12 , where z = e;8 • Conversely, if
z = e;8 , the expression I p(z) 1 always represents a non-negative trigonometric poly-
2

nomial-in 0 of the same degree as the polynomial p(z).

4 PRELIMIKARIES [I ]

See Fejer 6. The second part of the statement is obvious. The first part is
easily derived from (I.I2.2) by introducing l + z-k for 2 cos kO and l - z-k
for2isink0. Wethenfindg(O) = z-"'G(z),whereG(z)isa1r2 ,.forwhichG*(z) =
G(z). Now those zeros of G(z) which are different from 0 and oo, and which
do not have the absolute value I, can be combined in pairs of the form z,., z"t,
0 < I z~-' I < I, where z"t has a meaning similar to that in §I.I2. Furthermore,
every real zero Oo of g(O) is of even multiplicity, and e.-80 is a zero of G(z) of the
same multiplicity. Thus
U T

(l.2.I) G(z) = cz· II (z

J.<=l
- zJ.I)(z - zt)
•-1
II (z - rY,
0 < I ZJ.I I < I, 1r.l = I; K + + (1 T = m.
- z~-' I = I z~-' II z - z! j, z = e.- ,
= I g(O) I = I G(z) I, z = / 8 , and I z 8
Since g(O)
the theorem is established.
The representation in question is, however, not unique. Indeed, if a denotes
an arbitrary zero of p(z), the polynomial p(z) (I - az)/(z - a) furnishes another
representation. Hence assuming g(O) :/= 0, we can gradually remove all the
zeros from I z I < I and obtain the following theorem:
THEOREM 1.2.2. Let g(O) satisfy the condition of Theorem l.2.I and g(O) :/= 0.
Then a representation g(O) = I h(e.-8 ) 1 exists such that h(z) is a polynomial of
2

the same degree as g(O), with h(z) :;e 0 in I z I < I, and h(O) > 0. This polynomial
is uniquely determined. If g(O) is a cosine polynomial, h(z) is a polynomial with
real coefficients.
A generalization of this normalized representation (its extension to a certain
class of non-negative functions g(O)) is of great importance in the discussion of
the asymptotic behavior of orthogonal polynomials. (See Chapters X-XIII.)

1.21. Theorem of Lukacs concerning non-negative polynomials

(I) THEOREM l.2I.I (Theorem of Lukacs). Let p(x) be a 11"m non-negative in
[-I, +I]. Then p(x) can be represented in the form
2 2 2
IA(x)) +(I- x )!B(x)J if m is even,
(I.2l.I) p(x) = { 2 2
(I + x)IC(x)) + (I - x)ID(x)) if m is odd.
Here A (x), B (x), C (x), and D (x) are real polynomials such that the degrees of the
single terms on the right-hand side do not exceed m.
The proof can be based on Theorem 1.2.2. We have
p(cos 0) = I h(e;o) 12 = I e-im0/2 h(/s) 12,
where h(z) is a 7rm with real coefficients. Now the expressions
sin (m + I)O cos (m + !)O sin (m + !)O
(1.21.2)
cosmO, sin 0 ' cos (0/2) ' sin (0/2)
 [ 1.3 1 APPROXIMATION BY POLYNOMIALS 5
are all1rm in cos(} (see §I.I2), so that
._. .
e-•uw 12h(e' 8) = {A(cos 0) +i sin(} B(cos o) if m is even,
2t cos (0/2) C(cos 0) + i2 1
sin (0/2)D(cos 0) if m is odd:
where the degrees of A(x), B(x), C(x), D(x) are, respectively, m/2, m/2 _ I,
(m - 1)/2, (m - 1)/2.
(2) The following theorem has a simpler character:
THEOREM 1.21.2. Every polynomial in x, which is non-negative for all real
values of x, can be represented in the form !A(x)) 2 +
!B(x)) 2 • Every poly-
nomial which is non-negative for x ~ 0, can be represented in the form
2
!A(x) ) +
!B(x) )
2
+
x[!C(x) )
2
+ 2
!D(x) ) ]. Here A(x), B(x), C(x), D(x) are
all real polynomials, and the degree of each term does not exceed the degree of the
given polynomial.

These representations can also be written in the form I P(x) 12 and

I P(x) 1
2
+ x I Q(x) 12, respectively, where P(x) and Q(x) are polynomials with
complex coefficients; for the degrees the same remark holds as before. In the
case when x ~ 0, B(x) and D(x) can be chosen to vanish identically. See
Achieser 4, [2.54], and Karlin-Studden 1, Chapter V, Corollary 8.1.
In connection with this section see P6lya-Szego 1, vol. 2, pp. 82, 275, 276,
problems 44, 45, 47.

1.22. Theorems of S. Bernstein

THEOREM 1.22.1. If g(O) is a trigonometric polynomial of degree m satisfying
the condition I g ( 0) I ~ 1, (} arbitrary and real, then I g' ( 0) I ~ m.
This theorem is due to S. Bernstein. (Cf. M. Riesz 1.) The upper bound m
cannot be replaced by a smaller one as is readily seen by taking g(O) = cos mO.
The following special case is worthy of notice:
THEOREM 1.22.2. Let p(z) be an arbitrary 11"m satisfying the condition
I p(z) I ~
I, where z is complex, and I z I ~ I; then I p'(z) I ~ m, I z I ~ I.
With regard to this theorem see also Szasz 1, pp. 5I6-5I7. Finally we
mention the following ·consequence of Theorem 1.22.1:
THEOREM 1.22.3. Let p(x) be a 11"m satisfying the condition I p(x) I ~ I in
- 1 ~ x ~ + 1. Then
I p'(x) I ~ (1 - x 2 )-im.
This follows by applying Theorem 1.22.1 to g(O) = p(cos 0).

1.3. Approximation by polynomials

(I) THEOREM 1.3.1 (Theorem of Weierstrass). A function, continuous in a
finite closed interval, can be approximated with a preassigned accuracy by poly-
nomials. A function of a real variable which is continuous and has the period 21r,
can be approximated by trigonometric polynomials.
6 PRELIM IN ARIES r1 1

For information concerning this theorem we refer to Jackson 4. In the second

partt · of the· theorem let the function in question be even (odd)·, then the approxt-
·
ma mg tngonometric polynomials can be chosen as cosine (sine) polynomials.
THEOR~M 1.3.~. Let w(o) be the modulus of continuity of a given function
f(x), contmuous m the finite interval [a, b],

(1.3.1) w(o) = max I f(x') - f(x") I if I x' - x"l ~ o.

Then for each m we can find a polynomial p(x) of degree m, such that in the given
interval of length l we have

(1.3.2) I f(x) - p(x) I< Aw(l/m).

In the case of a periodic function f(O) with period 21r, a trigonometric polynomial
g(O) of degree m can be found such that

(1.3.3) If(O) - g(O) I< Bw(21rjm).

Here A and B are absolute constants.

In this connection see Jackson 4, pp. 7, 15.
THEOREM 1.3 .3. Let f(x) have a continuous derivative of order 11. in the finite
irdcrvalla, b), 11. ~ 1, and let wio) be the modulus of continuity of f~>>(x). Then
a polynomial p(x) of degree m + 11. exists such that
I f(x) - p(x) I< C(l/mY'wp.(l/m),
(1.3.4) I f'(x) - p'(x) I <
1
C(l/m)r wp.(l/m),
l = b - a.

Here C is a constant depending only on 11.·

Analogous inequalities can be obtained for all the derivatives f(x), f'(x),
· · · ,lP.>(x).
For the first inequality see Jackson 4 (p. 18, Theorem VIII). To prove the
second inequality we first establish the following lemma:
LEMMA. Let f(O) be a function of period 21r satisfying the Lipschitz condition

(1.3.5)
where A is a positive constant. Then there exist for each m trigonometric poly-
nomials g( 0) of degree m such that
D'A I g'(O) I < D"A'
(1.3.6) 1 f(o) - g(o) 1 < -m,
where D' and D" are absolute constants.
For the first inequality (1.3.6) see Jackson 4, pp. 2-6. When we use his
notation and argument, it suffice:s to show that I A- l 1~(0) I is less than an abso-
lute constant. But
[ 1.3 1 APPROXIMATION BY POLYNO:\IIALS 1
h i+r/2
(1.3.7) 1~(0) = - "' IJ(O + 2u) - f(O) }F;n(u) du
2 -r/2

and

rr/2
}o
u I F~(u) Idu = 4
}o
rr/2 I
u sin.mu 131 !!:._ sin.mu I du
msmu dumsmu

(1.3.8) = O(I) rr/2 I

u sin mu 131 !I_ sin mu du I
}o mu du mu

I•
+ O(l)
1 r/2
mu
• mu j du
u sm mu 131 sm
mu '
since ujsin u is analytic in the closed interval [0, 1rj2]. On writing mu x,

O(m-1) 1"" X l.si~ X rId~ si~ X I dx + O(m-2) 1"' X I si~ X 14 dx = O(m-1).

Now we use (cf. loc. cit.) hm = O(m).

The analogue of the lemma for polynomials can be derived in the usual way.
Then in the upper bound of the first inequality of (1.3.6) the factor b - a = l
appears. It is convenient to transform the interval a ~ x ~ b into - t ~ y ~ !
(instead of -I ;£ y ~ I, cf. Jackson, loc. cit., p. I4), defining the function in
[ -1, - !] and [!, 1] by a constant.
In order to prove Theorem 1.3.3, we apply Theorem VIII of Jackson (loc.
cit., p. I8) to f'(x). (For this argwnent cf. loc. cit., p. I6.) Thus

I f'(x) - q(x) I< K(ljm)i'- 1w14 (l/m),

where q(x) is a proper 7rm+~<-l. Applying the lemma to f(x) - f~ q(t)dt,
which satisfies a Lipschitz condition with
1
>-. = K(l/m)"- w14 (l/m),
we obtain a 7rm, say u(x), such that

k(x) - i"' q(t) dt- u(x) I< K'(l/m)"w14 (l/m),

If we write J: q(t).dt +
u(x) = p(x), the statement is established.
The constants K, K', K" in the last three inequalities depend only on p..

(2) THEOREM 1.3.4 (Theorem of Runge-Walsh). Let f(x) be an analytic

function regular in the interior of a Jordan curve C and continuous in the closed
domain bounded by C. Thenf(x) can be approximated with an arbitrary accuracy
by polynomials.

See Walsh 1, p. 36. This theorem has been proved by Runge in case f(x)
is analytic on C; the general case is due to ·walsh.
We need also a supplement to the former theorem, due to Walsh (1,
 PRELL\1IN"ARIES [I J

pp. 75-76). Let C be again a Jordan curve in the complex x-plane. Let
x = ct>(z) be the map function carrying over the exterior of C into I z I > 1
and preserving x = z = CXJ. Then the circles I z I = R, R > 1, correspond to
certain curves CR, called level curves. 'Ve have

THEOREM 1.3.5. Let f(x) be analytic within and on C, and let CR be the largest
level curve in the interior of which f(x) is regular. Then to an arbitrary r,
0 < r < R, there corresponds a constant M > 0 such that, for each m, a polynomial
Pm(x) of degree m ex£sts satisfying the inequality
(1.3.9) lf(x) - Pm(X) I< llfr-m, x on C.
This holds also if Cis a Jordan arc, for example, the interval - 1 ;;£ x ;;£ + 1.
In the latter case CR is an ellipse with foci at ± 1, and R is the sum of the semi-
axes (§1.9).

1.4. Orthogonality; weight function; vectors in function spaces

(I) Let a(x) be a non-decreasing function in [a, b] which is not constant. If
a = - CXJ (or b = + CXJ ), we require that a(- CXJ) = limx--oc a(x) (a(+ CXJ)
limx-+oc a(x)) should be finite. The scalar product of two real functions f(x)
and g(x), where x ranges over the real interval [a, b], is defined by the Stieltjes-
Lebesgue integral

(1.4.1) (f, g) = lb f(x)g(x) da(x),

where we assume that f(x)g(x) is of the class La(a, b). This is certainly the
case if f(x) and g(x) are both continuous, or both of bounded variation, and
[a, b] is a finite interval. For a fixed function a(x) the orthogonality with
respect to the "distribution" da(x) may be defined by the relation
(1.4.2) (f, g) = 0.

We shall also use the expression "f(x) is orthogonal to g(x)."

If we permit f(x) and g(x) to be complex functions in general, definition
(1.4.1) must be modified to read

(1.4.3) (f, g) = lb f(x)g(x) da(x).

With this change in the definition of (f, g), we retain (1.4.2) as the definition of
orthogonality.
[For the definition of Stieltjes-Lebesgue integrals see, for instance, Hildebrandt
1, pp. 185-194. This definition, given originally for a monotonic a(x), can
easily be extended to the case where a(x) is of bounded variation. Hildebrandt
1, pp. 177-178, may also be consulted for the definition of Riemann-Stieltjes
integrals.
 [ 1.4 1 ORTHOGONA.LITY; WEIGHT FUKCTIOK; VECTORS 9

In what follows we sometimes need the formula for integration by parts:

(1.4.4) ib f(x) da(x) + ib a(x) df(x) = f(b)a(b) - f(a)a(a),

where a and b are finite, a(x) 'is of bounded variation, and f(x) is continuous.
The integrals are taken as Riemann-Stieltjes integrals.
The expression "distribution" used above arises from the classical inter-
pretation of da(x) as a continuous or discontinuous mass distribution in the
interval [a, b], the mass contributed by the interval [x 1 , x 2 ] of [a, b] being
a(x2) - a(xi).]
(2) If a(x) is absolutely continuous, the scalar product (I.4.I) reduces to

(1.4.5) (f, g) = ib f(x)g(x)w(x) dx,

where the integral is assumed to exist in Lebesgue's sense. Here w(x) is a non-
negative function measurable in Lebesgue's sense for which f~ w(x) dx > 0.
We shall call w(x) the weight function, referring to a weight function of, or on,
the given interval. Instead of "weight function" the term "norm function"
3
is sometimes used in the 1iterature. In the case of a distribution w(x) dx the
total mass corresponding to the interval [x 1 , x2] is obviously f~~ w(x) dx.
In what follows we refer to distributions of the form da(x) as distributions of
Stieltjes type.
\Ve use the same concept of distribution and weight function on a curve or on
an arc in the complex plane, for example, on the unit circle. Then we replace
the variable x by the real parameter which is used for the definition of the curve
or arc in question. (See Chapters XI and XVI.)
(3) Let da(x), or w(x) dx, a ~ x ~ b, be a fixed distribution, and consider a
space of "vectors" defined by the set of real functions f(x) which belong to the
class L!(a, b). The scalar product of two vectors (functions) f(x) and g(x) is
defined by (1.4.I) and the length (magnitude, norm) of a vector f(x) by II! II =
(f, J)*. Vectors (functions) with II f II = 0 are called zero-vectors (zero-func-
tions); vectors (functions) with II f II = I are said to be normalized. When
f(x) is not a zero-function, Af(x) will be normalized provided >-. :;e 0 is a proper
constant, uniquely determined save possibly for sign. If the functions a(x) and
w(x) satisfy the conditions mentioned in (I) and (2), there exist functions of
positive length for both cases. In the second case f(x) is a zero-function if and
only if lf(x) ) 2w(x), or what amounts to the same thing, f(x)w(x), vanishes
everywhere in [a, b] except on a set of measure zero. If w(x) and f(x) are
integrable in Riemann's sense, f(x) is a zero-function providedf(x)w(x) vanishes
at every point of continuity.
We note the inequality of Schwarz (cf. (l.I1.2))
(1.4.6) II fg II ~ II f II II g II ,
a Some corresponding German and French terms are: Belegungsfunktion, Gewichts-
funktion, fonction caracteristique (Stekloff), poids (S. Bernstein).

-------- ·------·---------------------------
10 PRELIMINARIES [I ]

the equality sign holding if and only if Af(x) + IJ.g(x) is a zero-function with
A and IJ. proper constants not both zero.
A finite set of functions fo(x), f1(x), , f 1(x) is said to be linearly inde-
pendent if the equation
II Aofo(x) + Ad1(x) + · · · + Az/z(x) II = 0
can be true only for
Ao = A1 = · · · = Az = 0.
Evidently no zero-function can be contained in such a system. An enumerable
set of functions (l = CXJ) is called linearly independent..if the preceding condi-
tion is satisfied for every finite subset of the given set.
The extension of these considerations to complex vector spaces is not difficult.
The scalar product is then defined as in (1.4.3).
Concerning the axiomatic foundation of these concepts see Stone 1, Chapter I.
1.6. Closure; integral approximations
(I) DEFINITION. Let p ~ I, and let a(x) be a non-decreasing function in [a, b]
which is not constant. 4 Let the functions
(1.5.1) fo(x), f1(x), /2(x), · · · , fn(x), · · ·
be of the class L~(a, b). The system (1.5.1) is called closed in L~(a, b) if for every
f(x) of L~(a, b) and for every E > 0 a function of the form

(1.5.2) k(x) = Cofo(x) + cd1(x) + · · · + Cnfn(x)

exists such that

(1.5.3) [b lf(x) - k(x) IP da(x) < E.

With regard to this definition see Kaczmarz-Steinhaus 1, p. 49. These authors
use the term "Abgeschlossenheit" for "closure."
(2) THEOREM 1.5.1. Let p and a(x) have the same meaning as in the previous
definition, and let the function f(x) be of the class L~ (a, b), a and b finite. Then
for every E > 0 a continuous function F(x) can be determined such that

(1.5.4) [b lf(x) - F(x) IP da(x) < E.

For a Riemann-integrable function with a(x) = x, this follows by a well-

known argument from the definition of the integral. In the general case, it is
convenient to use the method of W. H. Young of approximating Stieltjes-
Lebesgue integrals. (See Hildebrandt 1, p. 190.)
Applying Weierstrass' theorem, we obtain the following:
4 See the remark at the beginning of §1.4 (1).
[ 1.6 l LINEAR FPNCTIONAL OPERATIOl\'S 11

THEOREM 1.5.2. Let p, a, b, a(x), f(x) satisfy the conditionsofTheorem 1.5.1.

For every E > 0 there exists a polynomial p(x) such that

(1.5.5) lb Jf(x) - p(x) IP da(x) < E.

This means the closure of the system

(1.5.6) n = 0, 1, 2, ·. · ,
in the elas~ L~(a, b). In what follows, we shall use in particular the rases p = 1
and p = 2.
An analogous statement holds for the "mean approximation" of f(x) hy
trigonometric polynomials, which is equivalent to the property of closure of
the system
(1.5.7) 1 , cos x, sin x, cos 2x, sin 2x, · . · , cos nx, sin nx,

in L~( -1r, +1r).

(3) A more precise form of Theorem 1.5.2 is often useful.
THEOREM 1.5.3. Let p, a, b, a(x),J(x) satisfy the conditions of Theorem 1.5.1
and let f(x) be real-valued. Then we can find a polynomial p(x) which satisfies
(1.5.5) and is such that p(x) remains between the upper and lower bounds of f(x).
We refer also to the following property of Riemann-Stieltjes integrals which
plays a role in Chapter X.
THEOREM 1.5.4. Let the real-valued function f(x) be bounded in [a, b], a and b
finite, a(x) non-decreasing, and let the Riemann-Stieltjes integral f~ f(x) da(x)
exist. For every E > 0 there exist polynomials p(x) and P(x) sttch that

(1.5.8) inf f(x) -E ~ p(x) ~ f(x) ~ P(x) ~ sup f(x) +E,

and

(1.5.9) lb I P(x) - p(x) l da(x) < E.

See (for a(x) = x) P6lya-Szegi) 1, vol. 1, pp. 65,228, problem 137.

Similar statements hold for approximations by trigonometrie polynomials.
If f(x) is an even function, - 1r ~ x ~ + 1r, the approximating trigonometric
polynomials can be chosen as cosine polynomials.

1.6. Linear functional operations

(I) Let U(f) be an operation which makes a number U(f) correspond to
every function f(x), continuous in the finite interval [a, b]. This operation is
called additive if
(1.6.1)
12 PRELIM IN ARIES [ I]

whenever c1 and c2 are constants and J1(x) and f 2(x) are arbitrary continuous
functions in [a, b]. It is called continuous if U(fn) ~ U(f) whenever fn(x) ~ f(x)
uniformly in [a, b]. Additive and continuous operations are called linear.
An operation U(f) is called limited if there exists a constant M such that
I U(f) I ~ M max I f I . The greatest lower bound of these constants M is the
norm of U(f). The class of additive and limited operations U(f) coincides with
that of the linear operations.
According to a theorem of F. Riesz (1), any linear operation can be written
in the form

(1.6.2) U(f) = lb f(x) da(x),

where a(x) is of bounded variation, defined in [a, b] and independent of j(x).

It is obvious that (1.6.2) always represents a linear operation. In (1.6.2) the
function a(x) can always be so normalized that a(x - O) ~ a(x) ~ a(x + O)
or a(x + 0) ~ a(x) ~ a(x - 0) for a < x < b. Then the norm of U(f) is
given by J~ I da(x) I, which is the total variation of a(x).
(2) Let K(x) be a given function continuous in [a, b]. Then

(1.6.3) lb f(x)K(x) dx

defines a linear operation. Dirichlet's integral

(1.6.4) ]:__ l+r

27r -r
J( ) sin {(2n + 1) (x- Xo)/21 d
X sin {(x - Xo)/21 x,
where n is a non-negative integer and Xo arbitrary, is a special case of (1.6.3).
It represents the nth partial sum of the Fourier expansion of f(x) at x = xo .
Another important example is Fejer's integral

(1.6.5) 1
27r(n+1)
l+r
-r
f() (sin {(n + 1) (x- x0 )/2J) d
x sin {(x-x0 )/2l x,
2

which represents the nth Cesaro mean of the Fourier expansion of f(x). A
further illustration of linear· operations is furnished by Lagrange's interpolation
polynomial
(1.6.6) L(x) = L(f; x) = j(xo)lo(x) + f(xi)li(x) + · · · + f(xn)ln(x),
where l0 (x), l1 (x), · · · , ln(x) are the fundamental polynomials associated with the
interpolation points Xo, X1, • • • , Xn (see Chapter XIV). For a fixed value
x = ~~ the expression L(f; ~) represents a linear operation on f(x). Finally, the
general mechanical quadrature formula,
(1.6.7)
is also an example; here Ao , A1 , ••• , An are the so-called Cotes numbers (see
Chapter XV).
[ 1.6 1 LINEAR FVNCTIONAL OPERATIOXS 13

(3) We consider the sequence of linear uperations

(1.6.8) n = 0, 1, 2, · · · ,
and the operation

(1.6.9) U(f) = 16

f(x)da(x),

where an(x) are normalized as in (1.6.2). Then we have the following theorem:
THEOREM 1.6. A necessary and sufficient condition that limn_,ooUn(f) = U(f),
where f(x) is an arbitrary continuous function, is that the following two relations
be satisfied simultaneously:

n->oo
k = 0, 1, 2, ... '
(1.6.10)
n = 0, 1, 2, · · · .

Moreover, if the second condition (1.6.10) is not satisfied, a continuous func-

tion f(x) exists such that the sequence 1Un(f) l is unbounded.
This important theorem is due to E. Reily (1, pp. 268-271). See also
Banacb 1, p. 123. The first condition (1.6.10) expresses the validity of the
limiting relation for an arbitrary polynomial. The second condition (1.6.10)
states that the total variations of the functions an(x) are bounded.
(4) Let b - a = 21r, and suppose thatf(x), an(x), and a(x) are functions with
period 21r. Then the first condition (1.6.10) must be replaced by the following:

lim Un(cos kx) = U(cos kx),

n-ooo
(1.6.11) k = 0, 1, 2, ...
lim Un(sin kx) = U(sin kx),
n->oo

One of the most important applications of the preceding considerations is

to the theory of "singular integrals" of Lebesgue:

(1.6.12)

where !Kn(x) l is a given sequence of continuous functions. In this case we

are mainly interested in finding a necessary and sufficient condition that
UnCf). ~ f(xo), where :ro is a fixed point in [a, b] t>1d f(x) an arbitrary continuous
function. According to Reily's theorem, this must hold if f(x) is an arbitrary
polynomial (or trigonometric polynomial in the periodic case) and the so-called
Lebesgue constants (which arc the norms of Un(f)) are bounded:

(1.6.13) lb I Kn(x) I dx < A.

14 PRELIM IN ARIES [I ]

See Lebesgue 1, 2; in particular articles 45, 46, pp. 86-88. See also Haar 1.
For Dirichlet's integral (1.6.4) this condition (1.6.13) is not satisfied; hence
there exist continuous functions whose Fourier expansions are di,·ergent at a
preassigned point. (Du Bois-Reymond 1; Lebesgue 2, chapter IY, pp. 84-89.)
This condition is, ho\\·enr, satisfied in the case of Fejrr's integral (1.6.5) v•hich
implies the Cesaro ~ummability of the Fourier expansion of a continuous func-
tion (Fejer 2, in fJarticular p. 60). The same holds for the Cesaro means of
second order of the Legendre series (Fejer 4).
Regarding applications of Reily's theorem to the theory of interpolation and
mechanical quaJrature, see Chapters XI\" anJ XV.

1.7. The Gamma function

The Euler integral of the sf'<~ond kind

(1.7.1) r(z) = 1"" e

0
-tr·-l dt

defines the Gamma function r(z) for m(z) > 0. By analytic continuation \Ve
obtain a meromorphic function without zeros and with 1-'imple pole:;; at z = 0,
-1, -2, . . . . The functional equations

(I. 7 .2) r(z + 1) = zr(z), r(z)r(1 - z) = 1r/sin 1rZ

hold. Another important formula is
r(z)r(z + 1/n) ... r(z + (n - 1)/n)
(I. 7.3)
=ni-nz(2 1r)(n-Il/2r( nz,
) n a positive integer.
In what follows we use mainly the cases n = 2 and n = 3.
The Euler integral of the first kind

(1.7.4) B(p, q) = 1 1
xP- 1(1 - x)q-I dx, p > 0, q > 0,

can be expressed in terms of the Gamma function thus:

B( ) = !'(p)r(q)
(1.7.5) p, q r(p + q).
The integral (1.7.4) exists also for complex p and q with positive real parts for
which (1.7.5) remains valid. By means of (1.7.5) the definition of B(p, q) can be
extended to arbitrary complex p and q. (See \Vhittaker-Watson 1, Chapter 12,
pp. 237, 239, 240, 254.)
The special case n = 2 of ( 1.7 .3) is as follows:
(1.7.6) r(z) r(z + l/2) = 21 - 2 '7r 112 r(2z).
We mention also the formula
(O+l
(1.7.7)
l
-- =-
r(z)
l
21ri f-00
e1l-'dl
•

1. 71. Bessel functions

(1) The Bessel function of the first kind of order a can he defined by
 [ 1.71 l BESSEL FPXCTI0::\8 I5

"" ( -I)"(z/2),+ 2 •
(1.7l.I) J a(z) = L 1
•-0 II. r
(
II + a + I )"
Obviously, z -a J a(z) is an even integral function. Here a is arbitrary real.
If a is a negative integer, l r(11 + a + 1) l- 1 must be replaced by 0 whenever
11+ a+ I~ 0. \Vethenobtain therelationJa(z) = (-l)aJ_a(z). Ifais
not an integer, J a(z) and J -a(z) are linearly independent. We notice the
special casei:l

(1.71.2)

The function (1.71.I) satisfies Bessel's differential equation

(1.71.3) y, +z -1
y, + (I - a 2z-•)
- y = 0, Y = .J a(z).
For non-negative integral values of a we introduce the Bessel functions of the
second kind

Y a (z) _- -2 ( 'Y + I og -z) J a (z) - -I ~

2
L.....
(a - 11 - I)! (z/2)- •-a
-rr 2 1r •-o_ 11!

(1.71.4) - ! f (-l)'(z/2)2•+a II/I + I/2 + ...

1r .-o 11! (11 +a)!

+ I/11 + I/I + I/2 + · · · + I/(11 +a) J, a = 0, I, 2, · · · .

Here 'Y is Euler's constant. The fhst sum is to be suppressed for a = 0, and the
curly brackets in the second sum are to be replaced by I for 11 = 0, a = 0, and by
I/I + I/2 + · · · + I/a for 11 = 0, a > 0. This function furnishes a second
solution of (1.71.3) independent of (1.71.1). (See Whittaker-Watson 1, Chap-
ter I7, pp. 370, 372.)
The formulas

follow directly from (1.71.1) on comparing the corresponding coefficients on

both sides. The integral representation

(1.71.6)

holds for a > !. This can be verified by introducing the development of

ei" and integrating by means of (1.7.5).
(2) The following important asymptotic formula is used in various appli-
cations:

(1.71.7) z ~ + ao.

This is only a special case (p = I) of the asymptotic expansion (see Whittaker-

Watson 1, p. 368):

~~---- ------------------------------
16 PRELIMINARIES [ I]

)! cos 1 2
Ja(z) = ( -2 {P-
(z- a7r/2- 7r/4) L a.z- " + O(z- 2P)}
7rZ v=O
(1.71.8) l
+ (~)
7rZ
sin (z - a7r/2- 7r/4) {~
v=O
b.z- 2 1
"- + O(z-2 1
P- ) } ;

here p is an arbitrary positive number, a. and b. certain constants depending

only on v, and z ~ + ao. Also, ao = 1.
This expansion holds also if z is complex, I arg z I ~ 1r - o, o > 0, if we agree
that z1 = exp (! log z) with I S(log z) I ~ 1r - o. \Ve notice the following
important consequence of this formula, valid for -7r/2 + o ~ arg z ~ 37r/2- o,
0 > 0:
eari/2J a(e-ir/2z) = (27rz)-lez II + O(l z 1-1) l
(1.71.9)
+ (27rz)-l exp [- z + (a + !)1ri] II + OCI z ~- )
1
).

An asymptotic formula similar to (1.71.7) holds for Bessel functions Y a(z)

of the second kind, with the only change that cosine is to be replaced by sine.
(See Whittaker-\Vatson 1, p. 371.)
(3) It may be useful to notice th.:: order of magnitude. of J a(z) and Y a(z) for
z ~ + 0 and z ~ + ao. From the preceding formulas we see that when z ~ + 0,
J a(z) ~ za, a real, a ~ -1, -2, -3, · · · ,
(1.71.10) Ya(z) ~ z-a, a = 1, 2, 3, · · · ,
Yo(z) ~ log (1/z),
while when z ~ + ao,

(1.71.11)

1.8. Differential equations

We shall make frequent use of certain elementary transformations of homo-
geneous linear differential equations of the second order.
(1) Let K(x), M(x), N(x) be functions defined in the interval a < x < b
in which K(x) and M(x) have continuous derivatives and K(x) ~ 0; let N(x)
be continuous. If in
(1.8.1) K(x)y" + M(x)y' + N(x)y = 0
we introduce y = s(x)u(x), u(x) being the new unknown function, s(x) can be
determined so that u(x) satisfies an equation of the form
(1.8.2) u" + X(x)u = 0.

Direct calculation gives

(1.8.3) 2Ks' + Ms = 0; s(x) = exp - {f Mdx}

2
K ,
 [ 1.8 l DIFFERENTIAL EQUATIONS 17

where the integration is extended from an arbitrary point x 0 to x. Then

(1.8.4) X(x) = - d 2K
dx (M) - (M)
2K +
2
N
K"

(2) If we introduce into (1.8.1) the new independent variable 0 defined by

x = u( 0), we obtain

(1.8.5) K(x)u'(O) ~:~ + /M(x)[u'(0)] 2 - K(x)u"(O) l ~~ + N(x)[u'(0)] 3

y = 0.

If we apply the process in (I) to (1.8.5), the first derivative can be removed.
We write y = s*u; here, in view of (1.8.3),

(1.8.6) s* = exp {! - M'

2
u ;_u' u dO
2
K"} = (u') 1 s,

where s has the same meaning as in (1.8.3). Hence, y = (u') 1su, and u satisfies

( 1.8 .7) d2u * - 0

d02 +X u - '

with
2 2 2
(1.8.8) * = _!!_ (Mu' - Ku") _ (Mu' - Ku") N ,2
X dO 2Ku' 2Ku' +K u .

As an application of the above, we note the following transformations of

Bessel's differential equation (1.71.3), k ~ 0,

(1.8.9) 2
~ + (k2 + 41-x2 a 2) u = o·'
dx2

+ (kx+ I-
2 2
(1.8.10) d u a )
dx2 4x2 u = 0;

Another elementary formula, important for further exposition is the repre-

sentation of a solution y = y.(x) of the non-homogeneous equation
(1.8.11) K(x)y" + M(x)y' + N(x)y = f(x)
in terms of a fundamental system lYI(x), Y2(x) l of the corresponding homo-
geneous equation (1.8.1). We have

(1.8.12) y(x) = CIYI(x) + D.!Y2(x) +

where x 0 is a fixed value and c1 , c2 proper constants.
l z
o
Y1(x)y2(t) - Y2(x)y1(t) f(t)
yf(t)y2(t) - y~(t)yi(t) K(t) dt,

Now

(1.8.13) y~(x)y2(x) - y~(x)yi(x) = const. exp {- J~ dx }.

When we take M = 0, this expression becomes a constant.

-------- ---------------------------------------------------------------
 I8 PRELIMINARIES [I]

Applying this last remark to (1.8.9), we obtain the important formulas

Ja, ( X) J_a (X) - J_a

, (x)Ja(X ) = 2 sin ct1r
1 a non-integral,
7rX
(1.8.I4)
J:(x)Ya(x)- v:(x)Ja(x) = -~, a= 0, 1, 2, · · ·.
7rX

As regards the evaluation of the constants on the right side, see \Vatson 3,
p. 43, (2), p. 76, (I).
1.81. Airy's function
An interesting transformation of Bessel's differential equation (1.71.3) can
be obtained in the special cases a = ±1/3. If we use (1.8.7) and (1.8.8), there
is no difficulty in showing that both integral functionS'

j" ~ v! ~C~~x~3 ~ ;~),

3

k(x) = j (x/3) 1J -d2(x/3)!} =

(1.8l.I)
l(x) = j" (x/3) 1Ji l2(x/3)! l = i §~ 11
! ~C~~x~
3
~ ;3 ),
3

satisfy the equation

2
(1.81.2) dy+.l -0
dx2 aXY - .

For negative x we have k(x) > 0, l(x) < 0. Using (1.71.9), 5 we obtain for
X< 0, X~- ao

(1.81.3)
Thus, but for a constant factor, the function
(1.81.4) A(x) = k(x) + l(x)
is the only possible particular solution of (1.81.2) which remains bounded if
x ~ - ao . Indeed,
(I.8I.5) X~- ao.

(See \Vatson 3, pp. 188-I90, 202.) This function A(x) is called Airy's function;
it can be considered as the standard solution of (1.81.2) and plays an important
part in numerous questions in mathematical physics. The function l(x)/k(x)
is increasing (see Fig. I) so that an arbitrary real solution of (1.81.2) has at
most one negative zero and infinitely many positive zeros. In particular, A (x)
6
If x is negative, we have
 [ 1.82 J THEOREMS OF STURl\I'S TYPE 19

has no negative zero and infinitely many positive zeros. Since A (x) > 0 for
x < 0, we see from (1.81.2) that A"(x) > 0 for x < 0; therefore A'(x) ~ 0
asx~ -ao.

FIG. 1

1.82. Theorems of Sturm's type

The following "comparison theorems" of Sturm's type can be proved in the
usual way (see Szego 20, pp. 3-4) :

THEOREM 1.82.1. Let f(x) and F(x) be functions continuous in Xo < x < Xo
with f(x) ~ F(x). Let the functions y(x) and Y(x), both not identically zero,
satisfy the differential equations

(1.82.1) y" + f(x)y = 0, Y" + F(x)Y = 0,

respectively. Let x' and x", x' < x", be two consecutive zeros of y(x). Then the
function Y(x) has at least one variation of sign in the interval x' < x < x" provided
f(x) ¢ F(x) in [x', x"].
The statement holds also for x' = Xo [y(xo +
O) = O] if the additional condition

(1.82.2) lim 1y'(x) Y(x) - y(x) Y'(x) l= 0

:-:ro+O

is satisfied (similarly for x" = Xo).

From Theorem 1.82.1 we readily derive (loc. cit., p. 4) the following theorem:
 20 PRELIM IN ARIES lI I

'THEOREM 1.82.2. Let cp(x) be continuous and dec1·easing in x 0 < x < X0 ,

and let y be a solution of

(1.82.3) y" + cp(x)y = 0

which £s not identically zero. Then x' < x" < x'" being three consecutive zeros
of y(x), we have x" - x' < x'" - x"; that is, the sequence of the zeros of y(x)
zs convex.
The last inequality holds also under the following more general cmulition:
(1.82.4) cp(x) > cp(x") > cp(y), forx < x" < y < x"'.
In addition, it holds also tf x' = Xo [y(xo + O) = 0] provided that

(1.82.5) lim (x - xo)y'(x) = 0.

z-zo+O

In order to prove x" - x' = h < x"' - x", we compare (1.82.3) with

Y" + cp(x - h)Y = 0,

which has the solution Y(x) = y(x - h) in the interval x" 5 x 5 x"'.
Another very elementary remark of a related nature is the following:

THEOREM 1.82.3. Let f(x) be continuous and negative in x 0 < x < Xo . Then
an arbitrary solution y of y" +
f(x)y = 0, for which y ~ 0 if x ~ Xo, cannot
vanish in Xo ~ x < X o •

Suppose the contrary. Now between two consecutive zeros sgn y" = sgn y
is constant, say positive; then y is positive and convex, which is a contradiction.

A further remarkable result of Sturm's type is the following (Watson 3, p. 518,

Makai 2):
THEOREM 1.82.4. Let f(x), F(x), y(x), Y(x), Xo, X 0 , x', x" have the same
meaning and satisfy the same conditions as in Theorem 1.82.1. We denote by ~
the first zero of Y (x) to the right of x', x' < ~ < x".
Assuming that y(x) > 0; Y(x) > 0 in x' < x < ~~and

(1.82.6 ) .
I1m y (x) > 1
Y( ) = ,
x->x' +0 X

we have y(x) > Y(x) in x' < x < ~-

We conclude as usual that the function y' (x) Y (x) - y (x) Y' (x) is increasing
in [x', ~];it is zero at x = x'(Y(x') = 0), thus positive in x' < x < ~- Hence
y(x) I Y (x) is increasing. In view of (1.82.6) the assertion follows.
The statement holds also for x' = x 0 provided the condition (1.82.2) is satisfied.
We prove now the following important consequence of the last theorem
(Hartman-Wintner 1; Makai 2):

---· ---------------------------------------··---
 [ Ull] PRIXCIPLE OF ARGC".\IEXT; ROUCHl~'S THEOREl\I 21

THEORE:\-1 1.82.5. Let <P(x), x', X 11 , X 111 have the same meaning as in Theorem
(1.82.2): X - x' < X ' - X • Let y(x) describe a negative "wave" in [x', X 11 ] and
11 11 11

11
a positive one in [x", x' ]. The .first "wave" is then entirely contained in the
second one.

The meaning of the last assertion is the inequality:

0 < -y(2x" - x) < y(x), X
11
<x < 2x 11 - x'.
The proof proceeds as usual, taking into account that, Y(x) = -y(2x 11 - x),
r y(x) - I. y'(x)
1m
x ->x" - Y (2X 11 - X
) - 1m Y1 ( 2X 11
x -+x" - X
) = l.

1.9. An elementary conformal mapping

Let the complex variables x and z be connected by the relations
(1.9.1) z = x + (x 2
- 1) 1.
The exterior of the unit circle, I z I > 1, as well as its interior, is mapped onto
the whole x-plane except the closed interval [ -1, +I] (the so-called cut plane),
with z = oo and z = 0, respectively, corresponding to x = oo. If we take that
branch of x + 2 1
(x - 1) which becomes infinite at x = oo, we obtain I z I > 1;
if we take the other branch, which vanishes at x = oo, we obtain I z I < 1.
The unit circle z = ei is carried over into the closed segment - 1 ~ x ~
8
1 +
described twice since x = cos 0.
The circle I z I = r, or I z I = r- , 0 < r ~ 1, corresponds to the ellipse with
1

foci at -1, +I and with semi-axes

2
1 I r- r -1 I.
Upon replacing z by i or bye -r, we obtain the representation

(I. 9.2) X = COSh S.

It maps the half-strip
(1.9.3) ~(S) > 0, - 7r < s (S) ~ 7r

onto the same x-plane cut along [ -1, + 1] as before; now, however, the point
x = oo has to be removed.

1.91. The principle of argument; Rouche's theorem; sequences of

analytic functions
THEOREM 1.91.1 (Principle of argument). Let f(x) be analytic both inside and
on a Jordan curve C, and letf(x) ~ 0 on C. Then the variation of Sllogf(x) l =
argf(x), as x describes C in the positive sense is 21rim, where m is the number of
the zeros of f(x) in the interior of C, counted with the proper multiplicity.
22 PRELDHXARIES lI I

THEOREM 1.91.2 (Rouche's theorem). Let f(x) and g(x) be analytic inside
and on C, and let I g(x) I < I f(x) I on C. Thrn f(x) + g(:r) and f(x) have no
zeros on C and the same number of zeros in the interior of C.
THEOREM 1.91.3 (Theorem of Hurwitz). Let Un(x) l be a sequrnce of analytic
functions regular in a region G, and let this sequence be umformly convergent in
every closed subset of G. Suppose the analytic function limn ..... oofn(x) = f(x) does
not vanish identically. Then if x = a is a zero of f(x) of order k, a ne1"ghborhood
I x - a I < o of x = a and a number N exist such that if n > N, fn(x) has exactly
k zeros in I x - a I < o.
The last theorem followR immediately from Theorem 1.91.2. Concerning
the preceding theorems Ree P6lya-Szego 1, vol. 1, pp. 120-124.
In Theorem 1.91.3, let G be symmetric relative to the real axiR, and let fn(x)
be real if xis real. If .-r = a is a Rimplc real zero of f(x), then for n > N each
fn(x) has exactly one real zero in I x - a I < o. For if fn(:ro) = 0, then fn(io)
is also 0.
 CHAPTER II

DEFINITION OF ORTHOGONAL POLYNOMIALS;

PRINCIPAL EXAMPLES
2.1. Orthogonality
(1) In what follows a(x) is a fixed non-decreasing function which is not con-
stant in the interval a ~ x ~ b. (See the remark at the beginning of §1.4 (1).)
DEFINITION. An orthonormal set of functions cl>o(x), c1>1(x), · · · , cl>t(x), l finite
or infinite, is defined by the relations

(2.1.1) (c/>n, c/>m) = 1b cl>n(X)c/>m(x) da(x) = Dnm, n, m = 0, 1, 2, · · ·, l.

Here cl>n(x) is real-valued and belongs to the class L~(a, b).

Functions of this kind are necessarily linearly independent. If a(x) has
only a finite number N of points of increase (that is, points in the neighborhood
of which a(x) is not constant), l is necessarily finite and l < N.
THEOREM 2.1.1. Let the real-valued functions
(2.1.2) l fim'te or infinite,
be of the class L! (a, b) and linearly independent. Then an orthonormal set
(2.1.3)
exists such that, for n = 0, 1, 2, · · · , l,
Ann> 0.
The set (2.1.3) is uniquely determined.
The procedure of deriving (2.1.3) from (2.1.2) is called orthogonalization.
(Cf. Stone 1, pp. 12-13.)
(2) For the orthonormal functions (2.1.4) the,following explicit representation
holds:
(2.1.5) n = 0, 1, 2, · · · ,
where, for n ~ 1,
Uo, fo)
(fl, fo)
(2.1.6) D n (X)
'
-- ................•..............

fo(x) ft(x)
23
24 DEFIXITIOX OF ORTHOGOXAL POLY::\0:.\IIALS [ II ]

and, for n ~ 0,
(2.1.7) Dn = [(fv, fi')Jv. 11=0,1,2, .. •,n > 0.
\Ve write D-1 = 1 and Do(x) = fo(x). The determinant (2.1.7) corresponds
to the positive definite quadratic form
II 'U{)fo + ud1 + ... + Unfn 11
2

(2.1.8)

so that Dn > 0 for each n.

Furthermore, the following integral representations can be established:

D.(x) ~ ~! t t . . f.' :(0~~1?.) ...~1~~1?........ /n.(~1:.

JO Xn-1 f1(Xn-1) fn(Xn-1)
n
(2.1.9) fo(x) f1(x) fn(x)
fn-1(xo)

fn-1(x1) da(x0) da(x 1) · · · da(Xn-1), n ~ 1,

fo(Xn-1) f1(Xn-1)
1
Dn = (n + 1)!
2
(2.1.10) fo(xo) f1 (xo) ··· fn (xo)

·1b [b .. ·1b ~o~~1? .. !~~~1~ .. .' .· .' .. ~:~~1: da(x 0)da(x1) · · · da(xn).
n+1___., lfo(Xn) f1(Xn) • •' fn(Xn)
(Cf. Kowalewski 1, p. 326; P6lya-Szeg6 1, pp. 48-49, 208, problem 68.)
(3) DEFINITION. Let {<Pn(x) l be a given orthonormal set, finite or infinite.
To an arbitrary real-valued function f(x) let there correspond the formal Fourier
expansion
(2.1.11) f(x) ""fo<Po(x) + f1<P1(x) + · · · + fn<Pn(x) + · · · .
The coefficients f n , called the Fourier coefficients of f(x) with respect to the given
system, are defined by

(2.1.12) n = 0, 1, 2, · · ·.

Every finite section of the series (2.1.11) has the following important minimum
property:
[ 2.2 I ORTHOGONAL POLYNOMIALS 25

THEOHEM 2.1.2. Let cl>n(x), f(x), fn have the same meaning as in the previous
definition. Let l ;;=; 0 be a fixed integer, and a0 , a1 , • • • , a 1 arbitrary real con-
stants. If we write
(2.1.13) g(x) = aocl>o(x) + atcl>t(x) + ··· + atcl>t(x),
and the coefficients a, are variable, the integral

(2.1.14) lb {f(x) - g(x) l 2 da(x)

becomes a minimum zf and only if av = fv , v = 0, 1, 2, · · . , l.

The minimum itself is

(2.1.15)

so that
(2.1.16) f~ + Ji + · · · + f~ ~ II! W,
and (Bessel's inequality)

(2.1.17) f~ +f~ +J~ + · ·· ~ IIJW = lb 2

lf(x)l da(x).

If the left-hand side of (2.1.17) is an infinite series, it is convergent. The dis-

cusRion of the equality sign in (2.1.17) leads to the concept of closure (§1.5).
A classical example of Fourier expansions of this kind is the ordinary Fourier
series in terms of the trigonometric functions 1, cos nx, sin nx, n = 1, 2, 3, .. · ;
-7r ~ X ~ +7r.
(4) Another important characterization of the orthonormal set (2.1.4) can
be based on the preceding minimum property of the partial sumR. Indeed, for
variable real values of Xo , At , · · · , An-I the expression
(2.1.18) II Xofo(x) + Xtft(x) + ··· + An-dn-l(x) + fn(x) II
becomes a minimum if and only if

(2.1.19) Xofo(x) + Xtft(x) + · · · + An-tfn-t(x) + fn(x) = X~~c/>n(x).

The extension of these considerations to complex function spaces is not

difficult. The scalar product of the functions f(x) and g(x) is then defined as
in (1.4.3).

2.2. Orthogonal polynomials

(1) Let a(x) be a fixed non-decreasing function with infinitely
DEFINITION.
many points of increase in the finite or infinite interval [a, b], and let the "moments"

(2.2.1) Cn =·lb x" da(.r), n = 0, 1, 2, · · · ,

26 DEFINITION OF ORTHOGONAL POLYNOMIALS [II ]

exist. If we orthogonalize the set of non-negative powers of x:

2
(2.2.2) 1, x, x , ••• 'xn, ... '
in the sensr; explained £n §2.1 (the linear independence IS shown below), we
obtain a set of polynomials
(2.2.3) Po(x), Pt(x), P2(x), · · · , Pn(x), · · ·
uniquely detrrmined by the following conditions:
(a) Pn(x) is a polynomial of precise degree n in which the coefficient of xn is
positive;
(b) the system lPn(x) l is orthonormal, that is,

(2.2.4) n, m, = 0, 1, 2, · · ·.
The existence of the moments (2.2.1) is equivalent to the fact that the func-
tions xn are of the class La(a, b).
A similar definition holds in the special case of a distribution of the type
w(x) dx. Here we assume that w(x) is non-negative and measurable in
Lebesgue's sense and that J! w(x) dx > 0. Moreover, the moments must
exist again.
We call Pn(x) the orthogonal polynomials 6 associated with the distributions
da(x) and w(x) dx, respectively; in the latter case we also speak of the or-
thogonal polynomials associated with the weight function w(x). The following
chapters are devoted to the ~tudy of these polynomials. Evidently if the
distribution is of the type w(x) dx, the system
(2.2.5) n = 0, 1, 2, · ·. ,
is orthonormal in the usual sense.
The linear independence of the functions (2.2.2) can readily be shown. In
fact if p(x) is an arbitrary real polynomial, the relation

liP W= lb {p(x)) 2da(x) = 0

is possible only if p(x) vanishes at all points of increase of a(x). Since there
are infinitely many such points, p(x) must vanish identically.
If cv.(x) has only a finite number, say N, of points of increase, the functions
1, x, i, ... , xN-J are still linearly independent. Through orthogonaJization
we obtain a finite system of polynomials· lPn(x) l, n = 0, 1, 2, · · · , N :_ 1,
~atisfying similar conditions as required in the previous Definition. See §2.8
and §2.82.
(2) Using the general formulas (2.1.5) to (2.1.8), we obtain, for n ;;=;; 1,

Sometimes these are called Tchebichef polynomials.

6 We shall reserve this terminology
for the important special cases (1.12.3).
 I 2.2 I ORTHOGONAL POLYNOMIALS 27

(2.2.6) ......................
2 n
1 X X X

where for n ~ 0
(2.2.7)
In addition to (2.2.6) we have Po(x) = Do'
= c()t. The determinant (2.2.7) is
associated with the positive definite quadratic form

(2.2.8) t t
v-0 J.&=O
Cv+14 UvU14 = 1b
a
(uo + U1X + U2X + · ·· + Unxn?da(x),
2

which is called a form of Hankel or of recurrent type. (See Szego 1.)

The determinant in (2.2.6) can be transformed by multiplying the next to the
last column by x, subtracting it from the last column, and repeating this opera-
tion for each of the preceding columns. In this way we obtain, n ~ 1,

CoX - C1 Cn-lX - Cn

.......................................
Cn-1X - Cn CnX - Cn+l

Furthermore, according to (2.1.9) and (2.1.10), we have the following integral

representations:

Pn(x) =
(D n-1D n)_,
n!
lb lb lb
a a
...
a
(x - Xo)(x - xl) ... (x - Xn-t)

(2.2.10) n

II
"~J.&-0,1,• · ·,n-1
2
(xv - x 14 ) da(xo) da(xl) · · · da(Xn-1),
v<l'

and

(2.2.11)

For (2.2.10) and (2.2.11) see, for example, Heine 3, vol. 1, p. 288. Formulas
(2.2.6), (2.2.9), (2.2.10) are not suitable in general for derivation of properties
of the polynomials in question. To this end we shall generally prefer the
orthogonality property itself, or other representations derived by means of the
orthogonality property.

-------,~----·-------------------
28 DEFINITION OF ORTHOGONAL POLYNOMIALS [ Ir I

(3) The Fourier expansion of an arbitrary function .f(x) m terms of the

polynomials I Pn(x) l has -the form
(2.2.12) f(x) "'-' foPo(x) + f1P1(x) + · · · + fnPn(x) +
with

(2.2.13) fn = lb f(x)pn(x) da(x), n = 0, 1, 2, ....

The partial sumt1 have the minimum property formulated in Theorem 2.1.2.
We notice as an important special case the following direct characterization of
the orthogonal polynomials (see §2.1 (4)). Considering the set of all poly-
nomials p(x) of degree n with the coefficient of x'" unity, we find that the integral

(2.2.14) lb (p(x)) 2 da(x)

becomes a minimum if and only if p(x) = const. Pn(x). Here the constant
factor is determined by normalizing p(x). If kn denotes the highest coefficient
2
of Pn(x), the minimum is obviously k-;. • From (2.2.8) we find for this minimum
the value Dn/ Dn-1 , so that
(2.2.15)
which also follows directly from (2.2.6).

2.3. Further remarks

(1) The restriction (a) in the definition in §2.2 (1) concerning the highest
coefficient, and the restriction (b) concerning the integral of the square, is
only one of various possible wayR of normalizing the polynomials in question.
Sometimes other kinds of normalization are appropriate, such as fixing the
value of Pn(x) at x = a or at x = b/ or fixing the highest coefficient of Pn(x),
and so on. Since Pn(x) has the precise degree n, every 'Trn can be represented
as a linear combination of po(x), p1(x), · · · , Pn(x) (see §1.12). Therefore
Pn(x), n ;:_: : 1, is orthogonal to any 'Trn- 1 . In particular,

(2.3.1) lb Pn(x)xv da(x) = 0, v = 0, 1, 2, · · ·, n - 1.

This condition determines Pn(x) save for a constant factor. Frequently, this
wider formulation of the orthogonality property is used. Observe also that
if p(x) is a 7rn and

(2.3.2) lb Pn(x)p(x) da(x) = c,

then the coefficient of x" in p(x) is ckn .

7 We have Pn(a) ,e. 0, Pn(b) ,e. 0 (see §3.3).
 [ 2.5 J A FORi\IULA OF CHRISTOFFEL 29
(2) Let [a, b] be an interval sym!lletric with respect to the origin, that is,
a = -b, and let us consider a distribution of the type w(x) dx with an even
weight function, that is, w( -x) = w(x). Then Pn(x) is an even or an odd
polynomial according as n is even or odd:
(2.3.3)

It can contain only those powers of x which are congruent to n (mod 2). In-
deed, we have for v = 0, 1, 2, · · · , n - 1

~~ Pn( -x)xv w(x) dx = ( -1Y ~~ Pn(x)xv w(x) dx = 0.

Consequently, Pn(- x) possesses the same orthogonality property as Pn(x) (in

the wider sense). Therefore, comparing the coefficients of x", we obtain
Pn( -x) = const. Pn(x) = ( -1) 11 Pn(x).
The linear transformation x = kx' + l, k ~ 0, carries over the interval
[a, b] into an interval [a', b'] (or [b', a']), and the weight function w(x) into
w(kx' + l). Then the polynomials
(2.3.4) (sgn k)" I k I! Pn(kx' + l)
are orthonormal on [a', b'] (or [b', a']) with the weight function w(kx' + l).

2.4. The classical orthogonal polynomials

1. Let a= -1, b = +1, w(:r) = (1 - x)a(l + x)~, a> -1, {3 > -1.
Then, except for a constant factor, the orthogonal polynomial Pn(x) is the
Jacobi polynomial P~a,f3J(x) (see §4.1).
2. Let a = 0, b = + oo, w(x) = e-xxa, a > -1. In this case Pn(x) is,
except for a constant factor, the Laguerre polynomial L~al(x) (see §5.1).
2
3. Let a= -oo, b = +oo, w(x) = e-x • In this case Pn(x) is, save for a
constant factor, the Hermite polynomial Hn(x) (see §5.5).
Some special cases of 1, except for constant factors, are:
The ultraspherical polynomials,. for a = {3.
The Tchebichef polynomials of the first kind, T n(x) = cos ne, X = cos e,
for a = {3 = -! (see (1.12.3)).
The Tchebichef polynomials of the second kind, Un(x) = sin.(n + 1)8/(sin e),
X = COS 8, for a = {3 = +! (see (1.12.3)).
The polynomials U2n (cos (8/2)) = sin (n + !)8/sin (8/2) of cos e = x, for
a = -{3 = ! (see §1.12).
The Legendre polynomials Pn(x), for a = {3 = 0.
A detailed investigation of these polynomials will be given in later chapters.

2.5. A formula of Christoffel

(1) THEOREM 2.5. Let I Pn(x) l be the orthonormal polynomials associated
with the distribution da(x) on the interval [a, b}. Also let

~---·-----------·----------·------·-····
30 DEFINITION OF ORTHOGONAL POLYNOMIALS [ II ]

(2.5.1) c =;e 0,
be a 1r t which is non-negative in this interval. Then the orthogonal polynomials
\qn(x)}, associated with the distribution p(x) da(x), can be reprrsr:nted in terms
of the polynomials Pn(x) as follows:
Pn (x) Pn+l (x) Pn+l(x)
(2.5.2) p(x)qn(x) = Pn(xl) Pn+l(xl) Pn+t(XJ)

In case of a zero xk , of multiplicity m, m > 1, we replace the corresponding

rows of (2.5.2) by the derivatives of order 0, 1, 2, · · · , m - 1 of the polynomials
Pn(x), Pn+l(x), ... 'Pn+t(X) at X = Xk.
This important result is due to.Christoffel (see 1, actually only in the special
case a(x) = x). The polynomials qn(x) are in general not normalized.
The proof is almost obvious. The ri'ght-hand member of (2.5.2) is a 7rn+t
which is evidently· divisible by p(x). Hence it has the form p(x)qn(x), where
qn(x) is a 7rn. Moreover, it is a linear combination of the polynomials Pn(x),
Pn+l(x), · · · , Pn+t(x), so that if q(x) is an arbitrary 7rn-l, then

(2.5.3) ib p(x)qn(x)q(x) da(x) = [b qn(x)q(x)p(x) da(x) = 0.

Finally, the right side of (2.5.2) is not identically zero. To show this, it suffices
to prove that the coefficient of Pn+t(x), that is, the determinant [Pn+v(x~'~ 1)],
v, tJ. = 0, 1, 2, .. · , l - 1, does not vanish. Suppose it to vanish; then certain
real constants Xo, X1, X2, · · · , >--1-1 exist, not all zero, such that

(2.5.4)

vanishes for x = x1, x2, · · · , x1. Hence (2.5.4) is of the form p(x)G(x),
where G(x) is a 7rn~l. Since (2.5.4) is orthogonal to an arbitrary 11"n-l, we have
the relation

[b p(x)G(x) · G(x) da(x) = 0;

whence G(x) = 0, a contradiction.
(2) The representation (2.5.2) enables us, for instance, to reduce ultra-
spherical polynomials with a = {3 = an integer, or with a +
! = {3 +! =
an integer, to Legendre and Tchebichef polynomials, respectively [cf. §4.21 (3)].
Another illustration may be obtained in connection with the polynomials con-
sidered in §2.6.
By using some special properties of da(x) or of p(x), formula (2.5.2) can be
simplified. For example, let da(x) = w(x) dx, w(x) and p(x) be even func-
tions, and a = - !J. Then, instead of (2.5.2), we have the representation (l cvC'n)
[ 2.6 l A CLASS OF POLYKOi\IIALS 31

Pn(x) Pn+4(x)

where ( ±x1, ±x2 , · · · , ±xt12l is the total set of zeros of p(x). For instance,
the orthogonal polynomials qn(x) associated with the weight function 1 - x 2
in [ -1, + 1] can be determined from

2 P n(x) P n+2(x)
(1 - x )qn(x) = = Pn(x) - Pn+2(x).
P n(l) P n+2(1)
(Cf. (4.7.27), 'A = !.)
2.6. A class of polynomials considered by S. Bernstein and G. Szego
Let p(x) be a polynomial of precise degree land positive in [ -1, +1]. Then
the orthonormal polynomials Pn(x), which are associated with the weight func-
tions
(1 - x2)-! (p(x) l-\
(1- x2)!(p(x)l-\
(2.6.1) w(x) -

( 1~ +- x)' {p(x) l-1

X

can be calculated explicitly provided l < 2n in the first case, l < 2(n + 1) in
the second, and l < 2n + 1 in the third. The polynomials of the first case
play an important role in the proof of Szegi/s equiconvergence theorem (9; cf.
Theorem 13.1.2). All three cases were later investigated by S. Bernstein (3)
in connection with his asymptotic formula (2; cf. Theorem 12.1.4).

THEOREM 2.6. Let p(x) be a 1r 1 of precise degree l and positive in [ -1, +1].
2
Let p(cos e) = l h(ci ) \ be the normalized representation of p (cos e) in the sense
8
8
of Theorem 1.2.2. Writing h(ei ) = c(e) + is(e), c(e) and s(e) real, we have the
following formulas:
Pn(cos e)- (2/7r)!ffi(eineh(ei8)l
(2.6.2) - (2/7r)! (c(e) cos ne + s(e) sin nel,
w(x) - (1 - x2)-!(p(x) l-\ l < 2n;

Pn(cos e) = (2/7r)!(sin e)-13(ei<n+ll8h(eiB)l

(2.6.3) = (2 /71' )! {c(e) sin (~ + 1)e _ s(e) cos(~+ 1)e},

sm e sm 0
w(x) = (1 - x2)! (p(x) l-\ l < 2(n + 1);
32 DEFINITION OF ORTHOGONAL POLYKO:'IIIALS (II]

Pn(cos 0) = 71'-!(sin (0/2))- 1 3(ei(n+!)Bh(eiB))

(2.6.4)
= 71'-! {c(O) sin (n +! )0 _ s(O) cos (n ! )0} +
· sin (0/2) sin (8/2) '

w(x) = (~ + ~)! (p(x) l- 1

, l < 2n + 1.
These formulas must be modified fo?· l = 2n, l = 2(n + 1), and l = 2n + 1,
respectively, by rn·ultiplying t.he right-hand member of (2.6.2) by (1 h 1/ho)-!, +
and those of (2.6.3) and (2.6.4) by (1 - ht/ho)-!, where h0 = h(O) and ht is the
coefficient of z1 in h(z).
First we observe that the right-hand members of (2.~.2), (2.6.3), (2.6.4) are
cosine polynomials with the highest terms

(2.6.5) (2/71' )! ho sin (~ 1)_~ + -! h sin (n

0
+
!)O
sm 0 ' 71' sin (0/2) '
respectively. In the first of these expressions, if l = 2n > 0, ho must be re-
placed by ho +
h 1 ; in the second and last, if l = 2(n 1) and l = 2n 1, + +
respectively, v.•e have ho - hz in place of ho.
We give the proof of (2.6.2). First we show that
+1

1 _
1
Pn(x)x"(1 - x2 )-! (p(x) )- 1 dx = 0, v = 0, 1, ... , n - 1,

or, what amounts to the same thing,

!a . Pn (cos 0) cos vO(p (cos 8))- d0 1

= 0, v = 0, 1, 2, ... , n - 1.

Now,

(2~71')! ffi {1 . eine h(eiBj (e;.e + e-;.e) I h(e'e) 1-2 do}

= (2/71')!
- 4 ffi -... e
+
{1+ ... ei(n+•)6 ei(n-•)6 } = ~/71't {~
h( iB) dO 4 ffi .
1, JzJ-1
Zn+• Zn-• } =
h( )
z z dz 0,
1 +
since the function (zn+• + zn-•) (zh(z) l- 1
is regular for Iz I ~ 1. Furthermore,

1~
1

(pn(x)) (1-
2
X
2 1
)-!(p(x)l- dx = 1. (pn(cos0)) 2 (p(cos0))- 1 d8

= Ia . Pn(cos0)(2/71')!hocosnO(p(cos0))- dO 1

= ~ (2/71')! ho(2/71')! ffi {~ r inh7) 1 dz} = -41 (2/71')ho(271'/ho) = 1.

4 }JzJ=1 1, Z Z

The proofs of (2.6.3) and (2.6.4) are similar. In place of cos vO we use
sin (v +
1)0/sin 0 and sin (v +
!)O/sin (0/2), respectively. The modifications
[2.8] STIELTJES TYPE; ANALOGUE OF LEGENDRE POLYNOMIALS 33

necessary for l = 2n, l = 2(n +

1), and l = 2n +
1, in (2.6.2), (2.6.3), and
(2.6.4), respectively, are also obvious. Finally, we notice that (2.6.2) arises
from (2.6.4), (2.6.4) from (2.6.3), and (2.6.2) from (2.6.3) by replacing p(x)
by (1 - x)p(x), (1 +
x)p(x), and (1 - x 2 )p(x), respectively.

2.7. Stieltjes-Wigert polynomials

Wigert (2, p. 7; also Stieltjes 11, pp. 507-508) found a very elegant explicit
representation for the orthonormal polynomials p,.(x) associated with the
weight function

(2.7.1) w(x) = 1r-!k exp (-e log 2 x) = 7r-!kx-k 2

Iogx, 0 < x < +oo; k > 0.
Using the notation (cf. Gauss 1, p. 16)

(2.7.2) [n]
Jl =
(1 - q")(1 - q"- 1) • • • (1- q"-•+1)
(1- q)(1- q2) ... (1 - q•) ' 0 < v < n, [~] = [~] = 1,
where
(2.7.3)
we have

(2.7.4) p,.(x) = (- 1)"q"12H I (1 - q)(1 - q2 ) • • • (1 - q")} -! f

•=0
[n] q"\-
Jl
q!xr.

For n = 0 the product in the braces must be replaced by 1.

The proof can be based on the identity of Gauss;

(2.7.5) f
•=0
[n]
Jl
2
q•<•+1)/ u" = (1 + qu)(1 + q u) 2
· · · (1 + q"u).
See Szego 12, where· other similar polynomials (related to the theory of theta
functions) are also considered. Also see Hahn 5.

2.8. Distributions of Stieltjes type; an analogue of Legendre polynomials

Tchebichef (4) investigated a remarkable finite set of orthogonal polynomials
associated with the distribution da(x) of Stieltjes type, where a(x) is a step
function with jumps of one unit at the points x = 0, 1, 2, ... , N - 1 (N is a
fixed positive integer). This is a distribution of the type mentioned at the
end of §2.2 (1). The as:-;ociated polynomials are, except ·for constant factors
(see (2.8.3)),

(2.8.1) t,.(x) = n! !l" n(x)(x-n N) , n=012 ... N-1

' ' ' ' .
Indeed, Tchebichef shows (4, pp. 547, 552; see also A. Markoff 1, pp. 21-22) that

(2.8.2) 1-: 00

t,.(x)lm(x) da(x) = x=O,l. ~

.. ,N-
1
t,.(x)tm(x) = 0, if n ¢. m,
34 DEFINITION OF ORTHOGONAL POLYNOMIALS l II]

and

1-: 00

Iln(X)}
2
da(x) = x=0. ,
1 2~ ·, N-
1
Iln(X)}
2

(2.8.3) 12)(N2 - 22) · · · (N 2

2
N(N - - 1/)
=
2n 1 +
n, m = 0, 1, 2, · · ·, N - 1.
These formulas hold for all non-negative values of n and m, but they are trivial
for n ~ N or m ~ N, since tn(x) = 0 for x = 0, 1, 2~ · · · , N - 1, if n ~ N.
In (2.8.1) we used the symbols
!:J.f(x) = f(x + 1) - f(x),

(2.8.4) t:J."f(x) = t:J.It:J."- 1/(x)}

=f(x + n)- (~)f(x + n- 1) + ... + (-1)"f(x).

By the mean-value theorem
(2.8.5) 0 < () < 1,
(see, for example, P6lya-Szego 1, vol. 2, pp. 55, 241, problem 98), we obtain
for a fixed value of n the remarkable formula
(2.8.6) lim N-"t,.(Nx) = Pn(2x- 1),
N-+oo

where Pn(x) is the Legendre polynomial of degree n (see §4.1 (3)). The repre-
sentation (2.8.1) is the "difference" analogue of (4.3.1), a = {3 = 0. The
proofs of (2.8.2) and (2.8.3) are analogous to those in §4.3.
Tchebichef also considers (1, 2) the more general case in which the points
0, 1, 2, ... , N - 1 are replaced by an arbitrary set of N distinct points. In
this connection he obtains an interpolation formula having a certain significance
in mathematical statistics. (See Jordan 1.)

2.81. Poisson-Charlier polynomials

These polynomials have become important in some recent investigation~ con-
nected with the calculus of probability and statistics (see DoetBrh 2 and the
literature quoted in E. Schmidt 1, also Meixner 1, 2). They bdong to the
distribution da(x) where a(x) is a step function \\'ith the jump
(2.81.1) j(x) = e-aax(x!)-1 at the point x, x = 0, 1, 2, · · · ; a > 0.
Obviously, the total variation of a(x) is
00

a(+ oo) - a(- oo) = L

x=O
j(:r) = 1.

The corresponding orthonormal polynomials are:

[ 2.82] KRAWTCHOUK'S POLYKOMIALS 35

(2.81.2)
Pn(x) = an 12 (n!)-! f (-1r-·(n) v! a-·(x)
•=0 Jl Jl

= an 12 (n!)-!(-1rU(x)}-l~nj(x- n).

A simple proof of (2.81.2) can be given by means of the method of generating·

functions (see §4.4; cf. Doetsch 2, p. 260, and Meixner 1, 2). Let, for a suf-
ficiently small I w I ,

G(x, w) = f a-n (n!)-!pn(x)wn = f :t (-n.1t-• (n)v!a-•(x)wn

n=O
12
n=O •=0 Jl Jl

(2.81.3) = f f (-n!1r-· (n)v!a-·(x)wn

•=0 n=• Jl Jl

= ~ -• (X) we
L.Ja • =e-w( 1 + a w.)x -w -1

·-0 Jl

Then
L
x=0,1.2,· · ·
j(x)G(x, u)G(x,· v)

(2.81.4)
x=0,1,2,• • •

so that
(2.81.5)

n, m = 0, 1, 2, · · · .
The polynomials (2.81.2) are connected with Laguerre polynomials (§5.1)
by the relation
(2.81.6)
Concerning the expansion problem af'ROciat.ed with Charlier-Poisson poly-
nomials, we refer to E. Schmidt 1.

2.82. Krawtchouk's polynomials

Considerations in the calculus of probability lead also to the following dis-
tribution da(x).
Let a(x) be a step function with the jump, at the point x, of

(2.82.1) '( ) (N)

JX= X
x N-x
p q' X = 0, 1, 2, ... 'N.

Here N is a positive integer, p > 0, q > 0, and p q = 1. +

See Krawtchouk 1. The total variation of a(x) is 1. The associated set
of orthogonal polynomials is again finite as in §2.8.
 36 DEFINITIOl\ OF ORTHOGONAL POLYNOMIALS [ II]

(1) The method of generating functions yields the formula

n = 0, 1, 2, · · · , N.

In case xis an integer, 0 ~ x ~ N, the last summation can be extended over

all values of v, n = 0, 1, 2, · · · ; v ~ n, since the general term vanishes if
N- x < n- v or if x < v. (From N- x ~ n- v, x ~ v, we haveN ~ n.)
Thus

K(x, w) =
•-0 Jl
£ (-1)"-• (Nn -- x) p"-•wn-•
'£ (x) q•w• n-· Jl
(2.82.4)
= £(X)
•-0 Jl
q"w"(1 - pw)N-z = (1 + qwY(l - pw)N-z,

from which
L
x=O,l. 2. · · ·.N
j(x)K(x, u)K(x, v)

(2.82.5)
= L ( ~) pzqN-z( 1 + qu)z(I _ pu)N-z(l + qvY(l _ pv)N-z

= I p(1 + qu)(1 + qv) + q(1 - pu)(1 - pv) l N = (1 + pquv)N,

so that in fact

(2.82.6) L
x=o.u ... ·.N
j(x){(N)}i
n
12
(pq)" pn(x){(N)}i
m (pq)m 12 pm(x) = (N)
n (pq)nOnm,

n, m = 0, 1, 2, · · · , N.
If n > N, obviously Pn(x) = 0 for x = 0, 1, 2, · · · , N.
(2) Two other classes of polynomials can be derived from the polynomials
(2.82.2) by two different limiting processes.
(a) Let z be real and let x denote the greatest integer less than or equal
to pN + z(2pqN)~ where p, q, z are fixed, and N ~ ((). Then for a fixed n
(2.82.7) lim Pn(x) = (2nn!)- 1Hn(z)
N-+oo

if H n(z) denotes the nth Hermite polynomial (§5.5). This follows readily from
(2.82.3) and (2.82.4), since for x an integer, 0 < x < N,
[2.91 FUHTII!m SI'I·;CL\L CASES 37

li~ to {(~) }\pqr pn(:r) 12

I (2/N)twl n
=lim (1 + (2/N)tqw}x(l- (2/N)tpw}N--x
N-+CO

(Use Titchmarsh 1, p. 95; sec (5.5.7).) It is instrueth·c to obsf'rvc that the

same limiting process applied to the given distribution rla(x) kads to the di;.;tri-
2
bution e-z dz of the Hermite polynomials; more precisely,
(2.82.8)
(b) Let pN = a, where a is a fixed positive number, N - oo, p - 0, q -----) 1.
Then for a fixed n and a fixed integer x ~ 0, we find that. llms~oo Pn(.r) exists
and is identical with the Poisson-Charlicr polynomial (2.R 1.2). In fad ( (2.81.:)))
lim (1 + qwY(1 - pw).v--r = lim (1 + qwY(l - pwr-r(l - pw)'' t,.
N-~ p~o
(2.82.9)

2.9. Further special cases

Concerning other dif-ltrihutions da(x) of Stieltj0s type, sec A. lVIarkoff 1, pp.
7-18; Sticltjes 11, pp. 546 5.')5; and Gottlieb 1.
Markoff considers the case for whieh a(:r) is a st<'P funrt ion with the jump,
atthepointqX,ofj(x) = qX, x = 0, 1,2, ···, N -1,and q > O,q ~ 1. This
distribution is very much similar to that in §2.8. Analogues of (2.8. I) and
(2.8.6) hold.
Stieltjes and Gottlieb investigate the eas.<~ for which a(x) is a st<·p function
with the jump q:r at tlw point..x, x = 0, 1, 2, · · · , 0 < q < 1.
In addition to these "discrete" distributions as well as to those studied in
§2.8 and §2.82, see Karlin-McGregor 2, Eagleson 1, and Gasper 6, 7.
A remarkable distribution can be defined by the weight function
(2.9.1) w(x) = (x(a - x)({3 - .r) }-l, 0 < x ~ a, cv < {3.
Heine (3, vol. 1, pp. 294-296) derives a second ord<~r linear cliffcrelltial cquat iou
for the associated orthogonal polynomials which arc related to the .Jarobian
elliptic functions. Recently Achieser (1) investigated the orthogonal poly-
nomials associated with the wright function
1(1 - x ) (a - i) (b - x)} -ll c - x
2
I, - 1 ~ x ~ a, b ;;£ x ;;£ + 1,
(2.9.2) w(x) =
. 10, a < x < b,
where -1 < a < b < + 1 and c depends in a proper way on a and b. These
polynomials are also related to the cllipti(' funrtions.
In some cases the condi~ion of the positiveness of the wright function <~an be
removed to a certain extent. ( Cf. Szegi> 19.)
Concemiug the pol~·uominls of Poll:l<'zl'k (1-4 ), s<'<' .\p]wudi.-.;.
 CHAPTER III

GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS

In this chapter we shall deal with properties of orthogonal polynomials

which hold for distributions restricted only by certain conditions of integrability.
Usually, we shall consider distributions of the Stieltjes type da(x), but at times
we shall be concerned with distributions of the special type w(x) dx. However,
a(x) and w(x) will alway:-; Le taken subject to the conditions formulated in
§2.2 (1).

3.1. Extremum properties; closure

(1) Let f(x) be a given function of the class L!(a, b), and let xn belong to
La(a, b) for n = 0, 1, 2, · · · . Then it is evident that the integrals

(3.1.1) 1b lf(x) 1da(x), 1b f(x)xn da(x),

2
n = 0, 1, 2, · · · ,

exist in the Stieltjes-Lebesgue sense. Next, denoting by IPn(x)} the ortho-

normal set of polynomials associated with the distribution da(x) in [a, b], we
state the following theorem:
THEOREM 3.1.1. The weighted quadrat£c deviation

(3.1.2) 1° lf(x) - p(x) 12 da(x),

where p(x) ranges over the set of all 11'n, becomes a minimum if and only if p(x)
is the nth partial sum of the Fourier expansion

J(x) "-' foPo(x) + f1P1(x) + f~p2(x) + · · · + fnp,.(x) + · · · ,

(3.1.3)
fn = 1° f(x)pn(x) da(x), n = 0, 1, 2, · · ·

See Theorem 2.1.2 and §2.2 (3). The minimum itself is

(:-U .4)

This implies Bessel's inequality, that is,

(3.1.5)

(2) On replacing n by n - 1 and taking f(x) = xn, we obtain the following

direct characterization of Pn(x):
38
[ 3.1 l EXTREMUM PHOPERTIES; CLOSURE 39
THEOREM 3.1.2. The integral

(3.1.6) 1b I p(x) 2
1 da(x),

where p(x) ranges over the set of all1rn with the highest term xn, becomes a minimum
if and only if p(x) = const. Pn(x).
See §2.2 (3). If kn is the coefficient of xn in Pn(x), the minimum of (3.1.6)
is attained for p(x) = k";/pn(x).
(3) THEOREM 3.1.3. Let Xo be an arbitrary complex constant, p(x) an arbitrary
11"n with complex coefficients, normalized by the condition

(3.1.7) 1b I p(x) 2
1 da(x) = 1.
2
The maximum of I p(xo) 1 is given by the polynomials
(3.1.8) p(x) = e(Kn(Xo, Xo) }-! Kn(Xo, x), IE I = 1,
where
Kn(Xo, x) = Po(Xo)po(x) + Pt(Xo)Pt(X) + · · · + Pn(Xo)Pn(x).
(3.1.9)
= Po(io)po(x) + Pt(io)Pt(X) + · · · +Pn(io)Pn(x).
The maximum itself is Kn(Xo, xo).

If we write p(x) = Aopo(x) + AtPt(x) + · · · + AnPn(x), condition (3.1.7)

becomes I Ao 1 + I At 1 + · · · + I An 1 = 1, and by Cauchy's inequality it
2 2 2

follows that
n n
(3.1.10) I p(xo)
2
1 ~ L
v=O
I A, 1
2
L
v=O
I p,(xo) 1
2
= Kn(Xo, Xo).
The latter bound is attained for A, = Ap,(x0 ), where A is to be determined
according to the condition
n
. I
2
A1 L
v=O
I p,(xo)
2
1 = 1.

Thus the statement is established.

The "kernel polynomials'; Kn(Xo, x) = Kn(x, xo) = Kn(i, io) can be used
for the representation of the nth partial sum sn(x) of the Fourier expansion
(3.1.3) in the form of an integral. In fact we have

Sn(X) = fopo(x) + ftPt(X) + · · · + fnPn(X)

(3.1.11) n
= ~ p,(x)
1b a j(t)p,(t) da(t) =
1ba f(t)Kn(t, x) da(t).

As a consequence of (3.1.11) we obtain

40 GEXERAL PHOPERTIES OF ORTHOGONAL :POLYXOl\IIALS [ Ill J

(3.1.12) 1b Kn(t, x)p(t) da(t) = p(x),

where p(x) is an arbitrary 7rn. We may easily show that this is a characteristic
property of Kn(t, x) as a 7rn in l. As a further consequence we notice the
following theorem:
THEOREM 3.1.4. Let a and Xo be fint'te, xo ~ a. Then the polynomt'als
{Kn(xo, x) l are orthogonal with respect to the dt'stn'bution (x - x0 ) da(x).
This follows immediately from (3.1.12) by writing x = x0 , p(t) = (t - x0 )r(t),
where r(t) is an arbitrary 7rn-l. A similar result holds if b is finite.
(4) According to the previous results the expression (3.1.4) decreases as n
increases, and consequently it tends to a non-negative limit as n ----t oo. We
have Parseval's formula

(3.1.13) I fo
2
1 + I fi
2
1 + I h
2
1 + ·· · + I fn
2
1 + ··· = 1b I f(x)
2
1 da(x)

when and only when this limit is zero.

The validity of (3.1.13) is evidently equivalent to the fact that the integral
(3.1.2) can be made arbitrarily small by a proper choice of the polynomial p(x).
This is, however, the same as the closure in L!(a, b) of the system (pn(x) l or
of the system (xn} (see the definition in §1.5 (1)). Thus, according to Theorem
1.5.2 we have the following:
THEOREM3.1.5. Thesetoftheorthogonalpolynomt'als {pn(x)},n = 0, 1,2, · · ·,
assoct'ated wt'th the distribution da(x) on afint'te t'nterval [a, b], is closed in L!(a, b).
More generally it is closed in L~(a, b), p E;; 1.
For a function f(x) of the class L!(a, b) Parseval's formula (3.1.13) holds.
A function f(x) of the class L!(a, b), for which fn = 0, n = 0, 1, 2, · · · , is
necessarily a zero-function.
The finiteness of the interval considered is an essential restriction. Some
cases of infinite intervals will be studied later. (See §5.7.)
The assumption fn - 0 in the last part of Theorem 3.1.5 is equivalent to
the fact that

(3.1.14) n = 0, 1, 2, .. · .

The discussion of this condition is closely connected with the uniqueness of

Stieltjes' problem of moments. An example showing that Theorem 3.1.5 does
not hold generally in case of an i'nfinite interval is the following:
(3.1.15) da(x) = exp (- x'" cos p.1r) dx, f(x) = sin (I'" sin p.1r), 0 < p. < 1/2.
Here (3.1.14) is satisfied (cf. P6lya-Szego 1, vol. 1, pp. 114, 285, 286, problem
153), and yet f(x) is not a zero-function. In the same case, if p(x) is an arbitrary
polynomial,
[3.11] GENERALIZATIONS 41

2
A = 1oo lf(x)} da(x) = 1oo f(x) lf(x) - p(x)} da(x)

:ii f 1/(x) - p(x) I da(x) :ii {f 1/(x) - p(x) I' da(x)}'{{ da(x)Y,
which shows that the integrals

1r.o lf(x) - p(x) I da(x) and 1oolf(x)- p(x)l 2 da(x)

cannot be made arbitrarily small.

3.11. Generalizations
Numerous analogous problems arise if the weighted quadratic deviation
(3.1.2) is replaced by other types of deviations. The most interesting cases are

(3.11.1)

p being a fixed positive number, and the limiting case p ~ oo 8 (called also the
"Tchebichef deviation"):
(3.11.2) max llf(x) - p(x) I w(x)}.
a~x~b

In the last case we assume that f(x) and w(x) are continuous. Similarly, the
integral (3.1.6) might be replaced by the expression

(3.11.3) 1b 1. p(x) IP da(x),

or by
(3.11.4) max {I p(x) I w(x)}.
a~x~b

The polynomials of fixed degree which minimize (3.11.1) and (3.11.2) represent
a generalization of the nth partial sum of the expansion of f(x) in terms of the
orthogonal polynomials associated with da(x) or w(x) dx. The polynomials
of fixed degree, and with highest coefficient unity, which minimize (3.11.3)
and (3.11.4) represent a generalization of the orthogonal polynomials them-
selves.
Since the number of investigations which can be classified under this general
point of view is very considerable, only the most important aspects can be
indicated here.
(1) For p = 2 the polynomials minimizing (3.11.1) are the partial sums of the
8 Replacing da(x) by {w(x)lP dx, we have (a and b finite, f(x) and w(x) continuous)
]1/P = max
lim
p-oo [1a
b
If(x) - p(x) jP { w(x) l P dx
a~x~b
ll f(x) - p(x) I w(x)}.
42 GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS [III]

expansion of a given function f(x) in a series of orthogonal polynomials (Theorem

3.1.1). In case a= -1, b = +1, w(x) = (1 - x2)-t, we obtain the expansion
of f(x) = f (cos e) in a cosine series; in cat>e a = -1, b = + 1, w(x) = 1, we
obtain the Legendre series. (Cf. Chapters IX and XIII.)
(2) In the general case (3.11.1) the existence and uniqueness of the minimizing
polynomials have been investigated. See Jackson 1, 2, 3; Shohat 1, pp. 509-513,
4, pp. 160-161. Both authors consider only distributions of the type w(x) dx.
For general distributions see Tamarkin 1, p. 118.
(3) Let a, b be finite, da(x) = dx, and p = 1. For problems (3.11.1) and
(3.11.3) see S. Bernstein 2, in particular pp. 135-137, where references arc also
given to the earlier literature. (Cf. also Geronimus 5.) Recently Achycscr (2)
discussed the problem of minimizing

(3.11.5) 1q I p(x) I dx + j" I p(x) I dx,

where [p, q] and [r, s] are given disjoint finite intervals and p(x) ranges over
all 11"n with the highest term xn. The minimizing polynomials can be repre-
sented in terms of elliptic functions.
(4) In the case where a and b are finite, f(x) continuous, and w(x) = 1, the
minimum problem corresponding to (3.11.2) leads to the closest approximation
of continuous functions by polynomials. The connection between the closeness
of this approximation (as n ~ oo) and the continuity properties of f(x) has been
investigated in great detail. (See Jackson 4.)
(5) If a, bare finite, f(x) continuous, and da(x) = dx, the minimizing polJno-
mials of (3.11.1) (for a fixed n) tend to the minimizing polynomial of (3.11.2) as
p ~ oo. (We have existence and uniqueness in both cases.) See P6lya 2;
also Shohat 1, pp. 513--514, 4, p. 171. Both authors consider only distributions
of the form w(x) dx. For general distributions, see Tamarkin 1, p. 125.
(6) If a = -1, b = +1, w(x) = 1, then problem (3.11.4) has the solution
p(x) = 21--nTn(x) (see the notation in (1.12.3)). This is a classical result due to
Tchebichef and is the starting point of various investigations of the highest
interest. (Cf. S. Bernstein 1.)

3.2. Recurrence formula; Christoffel-Darboux formula

(1) THEOREM 3.2.1. The following relation holds for any three consecutive
orthogonal polynomials:
(3.2.1) Pn(x) = (Anx + Bn)Pn-1(x) - CnPn-z(x), n = 2, 3, 4, · · · .
Here An, Bn, and Cn are constants, An > 0 and Cn > 0. If the highest coeffi-
cient of Pn(x) is denoted by kn, we have
kn
(3.2.2) An= k-,
n-1

For the proof, we first determine An so that Pn(x) - AnXPn-1(x) is a 'll"n-1 •

[ 3.2 l RECURRENCJi; FORMULA 43

This can be represented as a linear combination 'X0 p0 (x) + 'A 1p 1 (x) + . · · +

An-tPn-t(x), and beca~se of the orthogonality it is readily seen that 'A. = 0
if v < n - 2. Therefore (3.2.1) follows. The first part of (3.2.2) is a conse-
quence of (3.2.1); the second par:t follows from

1b Pn(X)Pn-2(x) da(x) = 0 = An 1b XPn-t(X)Pn-2(x) da(x) - Cn,

since the integral of the right-hand member is equal to

1b Pn-t(x)(kn-2Xn-l + ·· ·) da(x) = ~:=: 1b (Pn-t(x)} 2da(x).

'The recurrence formula (3.2.1) is valid also for n = 1 if we write P-t(x) = 0,

with the understanding that C1 it> arbitrary. The first formula in (3.2.2) then
holds for n = 1.
Concerning a converse of Theorem 3.2.1, see Favard 1.
(2) THEOREM 3.2.2. We have

Po(x)po(y) + Pl(x)pt(y) + · · · + Pn(x)pn(y)

(3.2.3)
kn Pn+l (x )pn (y) - Pn (x )Pn+l (y)
- kn+l X- y

For the special case da(x) = dx, see Christoffel!; see also Darboux 1. This
important identity can be easily derived from the recurrence formula. For
we have
Pn+t(X)Pn(Y) - Pn(X)Pn+l(y)
= ( (An+tX + Bn+t)Pn(X) - Cn+lPn-t(X) }pn(Y)
- Pn(x) ( (AnHY + Bn+t)Pn(Y) - Cn+tPn-t(Y)},
= An+t(X - y)pn(x)pn(Y) + Cn+d Pn(X)Pn-t(Y) - Pn-t(X)pn(Y)} ·
By (3.2.2), this becomes

kn Pn+l(x)pn(y) - Pn(X)Pn+l(y)
X- y
_ ( ) (· ) + kn-1
- Pn X Pn Y -k
Pn(X)Pn-l(y) - Pn-t(X)pn(y)
--- ,
n X - Y
which holds also for n = 0, with the understanding that k_ 1 is arbitrary. On
replacing n by 0, 1, 2, · · · , nand adding, we obtain (3.2.3).
We notice the special case x = y:

(po(x) }
2
+ (pt(x) ) + · · · + (pn(x) )
2 2

(3.2.4)
= kkn (p~+t(x)pn(x) - P: (x)Pn+t(x)}.
n+l
44 GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS [III]

(3) The left-hand member of (3.2.3) is identical with the "kernel" Kn(i, y) =
Kn(Y, x) introduced in (3.1.9). Using (3.1.12), we may derive (3.2.3) in a
different way by showing that the right-hand member of (3.2.3) (replacing y
by t) satisfies (3.1.12). For we have

~ (b Pnn(x)pn(t) - Pn(x)Pn+l(t) p(t) da(t)

kn+l }a X- t
kn
= k-
n+l
[b
a
lPn+l
() (X Pn t) - Pn(X)Pn+l(t)}
p(t) - p(x)
X -
t da(t)

+~
kn+l
p(x)1b a
Pn(t) Pn+I(X) - Pn+l(t) da(t)
X- t

+ kkn p(x) (b Pn-tl(t) Pn(t) - Ptn(x] da(t).

n+l ~ X -

Here the first and third integrals of the right-hand member vanish (also for
n = 0). The second term is p(x) since

kn Pn+l(x) - Pn+l(t) = kntn + .....

kn+l X - t
Another proof of (3.2.3) may be obtained by combining Theorem 3.1.4 with
Theorem 2.5.

3.3. Elementary properties of the zeros

THEOREM 3.3.1. The zeros of the orthogonal polynomials Pn(x), associated
with the distribution da(x) on the interval [a, b], are real and distinct and are
located in the interior of the interval [a, b].
In special instances, particularly in the classical cases (see §2.4), we shall
obtain later more exact information concerning the position of the zeros. (See
Chapter VI.)
As a consequence of Theorem 3.3.1 we have a(a) < a(x1 - 0) and a(xn + 0) <
a(b), where x1 and Xn are the least and the greatest zeros of Pn(x), respectively.
(1) The usual proof of the preceding theorem is based on the orthogonality
property. From

1b Pn(x) da(x) = 0, n ~ 1,

we are assured of the existence of at least one point in the interior of [a, b] at
which Pn(x) changes sign. (The function a(x) has an infinite number of points
of increase.) If x1 , x2 , · · · , Xt denote the abscissas of all such points, the
product Pn(x)(x - x1)(x - x2) · · · (x - Xt) has a constant sign (that is, is non-
negative or non-positive throughout [a, b]); we have l ~ n. On the other hand,
if l < n,

(3.3.1) [b Pn(x)(x - x1)(x - x2) · · · (x - Xt) da(x) = 0.

[ 3.3 l ELK\!El\TARY PROPEHTIES OF THE ZEROS 45

Since the integrand is not a zero-function, this is impossible. Therefore we

have l = n.
(2) A slight variation of this argument may be made as follows. Let x 0 be
an arbitrary zero of Pn(x). The coefficients of Pn(x) being real, we see that
Pn(x)j(x - i'o) is a 7T"n-l. On the other hand

(3.3.2) rb Pn(x) XPn(X~

}a - Xo
da:(x) = lba (x - Xo) II'
2
Pn(x) 1 da(x) = 0,
. X - Xo
so that

(3.3.3) Xo t i Pn(:r) ( da(:r) = {b :r! Pn(:r) 12 da(x).

}a 1
X - Xo 1 }n I X - Xo
In other words, xo is the centroid of a mass distribution on the interval [a, b].
The integral in the left-hand member of (3.3.3) being positive, x0 is real. From
(3.3.2) we see that a < xo < b.
If xo were a multiple zero, we should have

l b () lb{ 2
(3.3.4) Pn X ( Pn(x) ) 2 d a (X) = Pn(x) } da ( X) = 0,
X - Xo a X- Xo
which is a contradiction.
(3) The statement concerning the location of the zerot> (not their simplicity)
follows also from the minimum property formulated in Theorem 3.1.2. ·were
a zero Xo to lie outside [a, b], the distance I x - Xo I could be diminished simulta-
neously for all x in [a, b] by a proper displacement of x 0 • Hence the corre-
sponding integral (3.1.6) could not be a minimum.
For an extension of this argument to polynomials possessing an analogous or
a more general minimum property in the real or complex region, see Fejer 7,
Szego 5.
From the orthogonality property of the kernel Kn(x 0 , x) (Theorem 3.1.4),
we can similarly derive some theorems concerning the location of its zeros in x
if x 0 is regarded as a parameter. (See Szego 5, pp. 241-244.)
(4) The reality apd simplicity of the zeros (without the more exact statement
concerning their lob:tion in [a, b]) follow from the recurrence formula by means
of Sturm's theor~m (Perron 4, vol. 2, pp. 7-9). For, the polynomials
(3.3.5) Po(x), P1(x), P2(x), · · · , Pn(x)
form a Sturmian t>equence in [a, b] since (a) if p,(xo) = 0, v ;?; 1, it follows from
(3.2.1) that Pv-l(xo)Pv+l(xo) < 0; (b) Po(x) is a constant ~ 0, and Pn(x) is of
precit>e degree n; (c) at a point Xo where Pn(xo) = 0, we have (xo)Pn-l(xo) > 0. ]J:
The latter fact follows from (3.2.4) if n is replaced by n - 1 and x by Xo (see
below). Now the number of variations of sign in (3.3.5) is n if x < 0 and I x I
is sufficiently large; it is 0 if x > 0 and sufficiently large. (Cf. §6.2 (1) and the
footnote 32.)
(5) From (3.2.4) we obtain the important in ;quality
(3.3.6) P:+l(x)pn(x) - p:(x)Pn+l(x) > 0, x real.
 46 GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS [III]

As a first consequence we point out that Pn(x) and Pn+ 1(x) cannot have common
zeros. Furthermore, we obtain the following separation theorem:
THEOREM 3.3.2. Let X1 < X2 < · · · < Xn be the zeros .of Pn(x), Xo = a, Xn+l = b.
Then each interval [x., x.+l], v = 0, 1, 2, · · · , n, contains exactly one zero of
Pn+l (x) ·

In fact if ~ and 77, ~ < 77, are two consecutive zeros of Pn(x), we have
p:(~)p:(7J) < 0. On the other hand, (3.3.6) yields -p:(0Pn+l(0 > 0,
-p:(7J)Pn+l(7J) > 0, so that Pn+l(~)Pn+l(7J) < 0. This indicates an odd number,
that is, at least one zero of Pn+I(x) in~< x < 7J. Now let~= Xn be the greatest
zero of Pn(x); then p:(~) > 0, and (3.3.6) yields Pn+l(~) < 0. Since Pn+l(b)
is positive, we obtain at least one zero of Pn+ 1 (x) on the right of~ = Xn, and
similarly at least one on the left of the least zero x1 of Pn(x). Consequently,
we can have only one zero of Pn+ 1 (x) between x. and x.+ 1 , v = 0, 1, 2, · · · , n.
By interchanging the role of Pn(x) and Pn+ 1 (x), we can prove as before the
existence of at least one zero of Pn(x) between two consecutive zeros of Pn+l(x).
This shows again that we cannot have more than one zero of Pn+l(x) between
two consecutive zeros of Pn(x).

(6) THEOREM 3.3.3. Between two zeros of Pn(x) there is at least one zero of
Pm(x), m > n.

See Stieltjes 11, pp. 414-418. For the following proof see Popoviciu 1.
. Let b , ~2 , • • • , ~m be the zeros of p,.(x) in increasing order. According to
Theorem 3.4.1 we have

(3.3.7)

where p,~<) are the Christoffel number.<.; associated with ~~~<) (see §3.4) and
p(x) is an arbitrary 1r n-1 • Now an argument :,;imilar to that used in (1) shows
that the sequence lPn(b), p,.(h), · · · , p,.(~,,)) displays at least n, and therefore
exactly n, variations of sign. Here sgn Pn(b) = ( -1)", Pn(~,.) > 0. Thus
there are n distinct intervals
v = 1, 2, ... , n; 1 ~ J.l.l < /-L2 < . . . < 1-Ln+l ~ m,
coutainiug exactly ouc zero of p,.(x), respectively. This establir;hes the state-
ment.
Other simple consequences of (:3.3.6) are:

THEOREM 3.3.4. Let c be an arbitrary real constant. Then the polynomial

(3.3.8)

has n + 1 distinct real zeros. If c > 0 (c < 0), these zeros lie in the interior of
[a, b], with the exception of the greatest (least) zero which lies in [a, b] only for
C ~ Pn+l(b)/Pn(b), [c ~ Pn+l(a)/pn(a)].
[ 3.4] GAUSS-JACOBI l\IECHANICAL QFADRATURE 47

Indeed, the function Pn+I(x)/Pn(x) increases from- ooto + oo in the intervals

x. < x < x,+1 v = 0, 1, 2, · · · , n, where Xo = - oo, Xn+l =
1 + oo.
THEOREM 3.3.5. The fallowing decomposition into partial fractions holds:

(3.3.9) Pn(x) = f _z._ , l. > 0,

Pn+l(x) v=O X - ~.

where ~~.) denote the zeros of Pn+1(x).

For we have

(3.3.10) [, = Pn(~.) = P:+1(~.)pn(~,.) - p:(~.)Pn+l(~.) > O

P~ ·tl (~.) l P:+1 (~.) ) 2
•

3.4. The Gauss-Jacobi mechanical quadrature

(1) THEOREM 3.4.1. If X1 < X2 < · · · < Xn denote the zeros of Pn(x), there
exist real numbers A1 1 A2 , · · · 1 An such that

(3.4.1) lb p(x) da(x) = A1p(x1) + A2p(x2) + · · · + Anp(xn),

whenever p(x) is an arbitrary 1l"2n-1 . The distribution da(x) and the integer n
uniquely determine these numbers >-•.
The set lx. = x.n) of zeros, as well as the set of numbers l>-. = >-.nl, depends,
of course, on n. Sometimes the numbers>-. are called Christoffel numbers. See
Gauss 2, Jacobi 1, Christoffel 1, Tchebichef 1, Mehler 1; Heine 3, vol. 2,
pp. 1-31.
It suffices to prove (3.4.1) for the special cases p(x) = xk, k = 0, 1, 2, ... ,
2n - 1. These cases represent 2n conditions which uniquely determine, as
we shall prove, the Christoffel numbers >-. and the points x. . (If the distinct
points x. are given arbitrarily, the numbers >-. can be determined so that (3.4.1)
holds for every 7T"n-1 .)
To prove (3.4.1) we construct the Lagrange interpolation polynomial L(x)
of degree n - 1 which coincides with p(x) at the points x., that is,
n Pn(X) n
(3.4.2) L(x) = L
•-1
p(x.) ' ( )(
Pn Xv X - Xv
) = L p(x.)l.(x),
v=1

where the l.(x) are the fundamental polynomials associated with the abscissas
X1, X2, · · · , Xn of the Lagrange interpolation (see §14.1). Now p(x) - L(x)
is divisible by Pn(x), so that p(x) - L(x) = Pn(x)r(x), where r(x) IS a 7T"n-1.
Therefore

lb p(x) da(x) = lbL(x) da(x) + lb Pn(x)r(x) da(x)

b n lb
= L(x) da(x) = ~ p(x.) a l.(x) da(x).
l
 48 GE::\ERAL PlWPERTIES OF ORTHOGO:\'AL POLY:\'0:\IJALS [ III l

This establishes (3.4.1) with

(3.4.3) >... = [bl.(x) da(x) = 1b Pn a

1 (
Pn(x)
Xv
)(-
X
__
X.
-) da(x), v = 1, 2, · · ·, n.
Conversely, let (3.4.1) hold for an arbitrary 1r2n-l, called p(x). Then we
choose p(x) = l(x)r(x), where l(x) = (x - x1)(x - x 2 ) ••• (x - x,) and r(x)
is an arbitrary 7l"n-1. \Ve find from (3.4.1) that

[b l(x)r(x) da(x) = 0,

so that l(x) = const. Pn(x).

The interpretation of the left-hand member of (3.4.1) as a mechanical quad-
rature is obvious. For an arbitrary function f(x) defined in [a, b] it may
be written (cf. §15.1)
(3.4.4)

Then Theorem 3.4.1 can be formulated as follows: Qn(f) = f~ f(x) da(x)

provided· f(x) is an arbitrary 7r2n-1 • Further, from (3.4.3) the Christoffel
numbers >... are the values of Qn(f) for f(x) = l.(x). Also we can discuss for
a fixed functionf(x) the convergence of the sequence l Qn(f) l as n---> oo. (Com-
pare Theorem 15.2.3 and also Problem 9 below.) Concerning mechanical quad-
rature formulas holding for an arbitrary 7r2n-k, see Shohat 7, p. 465.
(2) THEOREM 3.4.2. The Christoffel numbers >... are pos£tive, and

(3.4.5) A1 + A2 + · · · + An = 1b da(x) = a(b) - a(a).

The following representations hold:

2
(3.4.6) >..,. =
l b (
1 (-
Pn Xv
Pn(x)
)(-
X
_--_ )
X,.
)
da(x),

(3.4.7)

(3.4.8)

= Kn(x., x.).

Here the previous notations are used.

Concerning (3.4.8) see Shohat 3, p. 456. The special case a = -1, b = +1,
da(x) = dx is particularly important. Here the abscissas x. are the zeros of
the nth Legendre polynomial, and the sum of the Christoffel numbers is 2, the
length of the interval of integration. This is the case originally considered by
[ 3.4 I GAUSS-JACOBI ~IECHA:\'ICAL QPADHATlfiU<; 49

Gauss and Jacobi. Another important special case, namely a = -1, b = + 1,

2
da(x) = (1 - x )-! dx, is due to Mehler (1).
The po;.;itivencss of 'A. is clear from any of the representations (3.4.6), (3.4.7),
(3.4.8). In case (3.4.7) we take into account (3.3.6). According to (3.4.5)
the sum of the >--. is the total masR of the diRtribution da(x) spread over the
given interval.
The discussion of the representation (3.4.7) can be carriPd further in the case
of the classical orthogonal polynomials. (Cf. §15.3 (1).)
For the proof of Theorem 3.4.2 we write p(x) = ll.(x)) 2 in (3.4.1); this
furnishes (3.4.6). Furthermore, writing y = x. in (3.2.3), multiplying by
da(x), and then integrating, we obtain, according to (3.4.3),

1 = kn {b -pn(x)Pn+1(x.) da(x)
kn+l }a x-x.

This establishes (3.4.7). Combining (3.4.7) with (3.2.4) for x = x., we get
(3.4.8).
(3) As an application of (3.4.1) we obtain, for arbitrary real constants
Uo , 'U1 , • • • , Un-1 ,

= L 'A.(uo + U1Xv + · · · + Un-1X:-1)2,

•=1
(3.4.9)

= L 'A.x.(uo + U1Xv + · · · + Un-1X:- 1t

v=1

Therefore, the characteristic values of the pencil

n

(3.4.10) G(u) -~F(u) = L x.(x. -

•=1
~)(uo + U1Xv + ... + Un-1x:- 1)2
are precisely ~ = X1 X2, · · · , Xn. With the notation (2.2.1), the quadratic
1

form of the left-hand member becomes

n-1 n-1

(3.4.11) L L (c•+~<+1 -
v=O 1'-0
~c•+~<)u.u~<.

Its determinant is a 7rn in ~ which vanishes for ~ = x1 , X2) · · · Xn , and is 1

therefore a constant multiple of Pn(~). We thus arrive at a new proof of Pquation

(2.2.9).
Sec also Problem 10.
50 GENERAL PROPERTJ.ES OF ORTHOGONAL POLYNOMIALS [ III I

3.41. Separation theorem of Tchebichef-A. Markoff-Stieltjes

In 187 4 Tchebichef stated a very remarkable property of the Christoffel
numbers (see 6, 8), proofs of which were given independently by A. Markoff
and Stieltjes. Let n ~ 2. In view of the positiveness of>... and of (3.4.5), there
exist numbers Y1 < Y2 < · · · < Yn-1, a < Y1, Yn-1 < b, 9 such that
(3.41.1) A, = a(y,) - a(Yv-1), P = 1, 2, · · · , n; Yo = a, Yn = b.
THEOREM 3.41.1. The zeros x 1 , X2, ... , Xn, arranged in increasing order,
alternate with the numbers Y1 , Y2 , · · · , Yn-1 ; that is,
(3.41.2) Xv < Yv < Xv+l ;

more precisely
a(x. + 0) - a(a) < a(y.) - a(a) = A1 + A2 + . ·. + X.
(3.41.3)
< a(x.+l - 0) - a(a), v = 1, 2, · .. , n - 1.
In view of (3.41.1) the quadrature formula (3.4.1) becomes
b n
(3.41.4) [
p(x) da(x) = L p(x.) l a(y.) - a(y,_1) ).
•-1
Since Y·-1 < x. < y. , the right-hand member has the character of a "Riemann-
Stieltjes sum."
As a further consequence of the inequalities (3.41.3) we notice that
a(x. + 0) < a(x.+l - 0). Thus we have proved the following:
THEOREM 3.41.2. In the open interval (x., x,+I), between two consecutive zeros
of Pn(x), the function a(x) cannot be constant.
Or in other words: In an open interval in which a(x) is constant, Pn(x) has
at most one zero.

3.411. First proof of the separation theorem 10

Let v be an integer such that 1 ~ v ~ n - 1. Choose for p(x) in (3.4.1)
a special 7r2n-2 subject to the following 2n - 1 conditions:

~ {~ k = 1, 2, ... 'v,
if
p(x,)
(3.411.1) if k = p + 1, p + 2, ... 'n;
p'(xk) = 0 if k = 1, 2, . · · , v - 1, v + 1, ... , n.
Then this polynomial is uniquely determined.
9
To find such numbers y, it might be necessary to modify a(y) at some of its points of
discontinuity, which has, of course, no influence on (3.4.1). It should be also observed
that y, is in general not uniquely determined.
1o Cf. A. Markoff 1, 2, Stieltjes 1, A. Markoff 3.
 [ 3.412 1 SECOND PROOF OF SEPAHATION THEOREM 51

By Rolle's theorem p'(x) has at least one zero in each of the open intervals
(3.411.2) (Xv-11, Xv+2), • • · , (Xn-1, Xn).
These zeros together with Xk, 1 ~ k ~ n, k ~ v, furnish (n - 2) + (n - 1) =
2n - 3 zeros for p'(x). Since p'(x) is a 1l'2n-a, it follows that these are the only
zeros of p'(x), and also that they are all simple zeros. Hence p(x) is monotonic
between any two consecutive zeros of p'(x); in particular, it is monotonic
between the zero in (xv-1 , x.) and x.+l , and therefore also in [x., x.+Il· Fur-
thermore, p(x) is decreasing in [x. , x.+l] since p(x.) = 1, p(x.+l) = 0. From

these considerations the graph of p(x) is easily seen to have the shape given in
the figure. Therefore we have
p(x) ~ 1 In a ~ x ~ x. ,
(3.411.3)
p(x) ~ 0 In x. ~ x ~ b.
For this special case the general formula (3.4.1) gives

A1 + A2 + · · · + >-. = [b p(x) da(x)

rx.+O l:r.+O
> }a p(x) da(x) > a da(x),

which establishes part of the inequalities (3.41.3).

To prove the remaining part we consider the distribution d[ -a( -x)] in
[- b, -a]. The associated orthonormal set is l (-1) nPn(- x)} with the zeros
-xn < -Xn-1 < · · · < -xl . In place of the numbers Y1, Y2, · · · , Yn-1 we
now have - Yn-1 , - Yn-2 , · · · , - Y1 . Then, according to the preceding result,
-a(Xn-•+1 - 0) < -a(Yn-v), or a(y.) < a(xv+1 :- 0).

3.412. Second proof of the separation theorem 11

Let the non-decreasing step-function V(x) be defined by the following
conditions:

u See Stieltjes 12, especially pp. 588--592.

_____ _____________________________
,,
52 GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS [ III ]

o, a ~X < X1,
Al ' Xl ~ X < X2 '
f
(3.412.1) V(x) = A1 + A2 , X2 ~ X < X3 ,

l~: ;·~:; ·. .·.·; -~~:. -~~ ~ -~ -~· ~-

Then (3.4.1) can be written in the form

(3.412.2) 1bp(x)d!a(x)- V(x)J = [bp(x)d(a(x)- a(a)- V(x)f = 0.

An integration by parts (see (1.4.4)) yields

(3.412.3) lb Ia(x) - a(a) - V(x) l p'(x) dx = 0,

since V(a) - 0 and (see (3.4.5)) a(b) - a(a) - V(b) = 0.

The function V(x) is constant in the open intervals (a, x1), (x 1 , x2 ), ••• ,
(Xn-1, Xn), (xn, b); hence a(x) - a(a) - V(x) = {3(x) is non-decreasing there.
We have {3(x) ~ 0 (but not {3(x) =
O) in the first interval, and {3(x) ~ 0 (but
not {3(x) =0) in the last interval. In the other intervals (x., x.+ 1 ) the function
{3(x) is either of constant sign (constantly non-negative or non-positive), or
there exists a pointy, x. < y < x.+l, such that {3(y - 0) < 0 and {3(y + 0) > 0.
Thus the total interval [a, b] can be subdivided in at most 2n intervals in which
{3(x) is non-negative and non-positive alternately, without being identically zero.
The end-points of these intervals are some of the zeros x. and some of the points
y previously defined. Now from (3.412.3) we conclude, by means of an argu-
ment similar to that in §3.3 (1), that the number of these intervals is at least 2n,
and then exactly 2n. Less precisely, {3(x) has exactly 2n variations of sign in
[a, b] which are located at the zeros x. as weq as at the points y mentioned,
whose number is n - 1. · At the points y there is a transition from negative to
positive values. Hence at the points x. there is a transition from positive to
negative values.
Consequently,
(3.412.4) {3(x. - O) > 0 > {3(x. + 0), v = 1, 2, • · · , n,
which is equivalent to the statement of Theorem 3.41.1. (The first inequality
is trivial for 11 = 1; the same holds for the second one for 11 = n.)
We can also prove the above statement by a slight modification of the argu-
ment used. Let y. denote the point in (x. , x.+l) with the variation of sign of
the type y; then we have
(3.412.5) {3(x. + 0) < {3(y.) = 0 < {3(x.+l - 0).
(Cf. the footnote above.)
 .-\XOTHER SEPARATIOX THEORE:\1 53

3.413. Third proof ofthe separation theorem 12

Let the functions cf>k(x, t) be defined as follows

(x - t)k if x ~ t,
(3.413.1) cf>k(X, t) =
l 0 if X > t,
where k is a non-negative integer, x and t arbitrary and real. Lee 3

(3.413.2) Fk(t) = Ib cf>k(x, t) da(x) - ~ Avcf>k(x., t).

Then according to (3.4.1)
(3.413.3) k = 0, 1, 2, ... , 2n - 1.
Also Fk(t) is continuous if k ~ 1; furthermore, we readily see that F0 (t) = {3(t),
where {3(t) has the same meaning as in §3.412. Now
(3.413.4) F;(t) = -kFk_1(t), a <t < b, k ~ 1.
[Fork= 1, t = x., this means F~(x. ± 0) = -F0 (x. ± 0).] If we take into
account (3.413.3) and (3.413.4), Rolle's theorem furnishes for the number of
zeros of Fk(t) (including t = a and t = b) the lower bound 2n + 1 - k,
1 ~ k ~ 2n - 1; this holds also for k = 0 in the sense that F0 (t) has at least
2n - 1 variations of sign. From this point on the statement follows by an
argument similar to that in the second proof.

3.42. Another separation theorem

If X1n <
X2n < · · · < Xnn denote the zeros of Pn(x), we know (Theorem 3.3.2)
that the system (x.n) alternates with the system !xv,n+I}, that is,
(3.42.1) Xv-l,n < Xv,n+l < Xvn, II =· 1 2 · · · n
' ' '
+ 1. Xo n =
' t
a Xn+l
' J
n = b.
Let now (Avn) denote the system of Christoffel numbers associated with
Pn(x), and let IYvnl be the numbers y. defined in (3.41.1). Stieltjes showed
(12) that in addition to Theorem 3.41.1 the following separation theorem holds:
THEOREM 3.42. We may assert that

(3.42.2) Yv-l,n < Yv,n+l < Yvn ;

or

(3.42.3)
Aln + A2n + + Av-l,n < Al,n+l + A2,n+l + ••• + Av,ntl
< Aln + A2n + ... + Avn' = II 1, 2, ... 'n.
For 11 = 1 the inequalities involving Yon and Aon must be disregarded.
12
This proof is due to P6lya and Uspensky (written communication).
13
In case k = 0, t = a the integral in the right-hand member should be replaced by 0.
54 GENERAL PROPERTIES OF ORTHOGONAL POLYKOMIALS [ III]

The analogy between (3.42.1) and (3.42.2) is obvious.

For the proof we use the same function V(x) as in §3.412, denoting it now
by V ,(x), and introducing the corresponding function V n+I(x) associated with
the system (Av.n+If· Then by (3.412.3) we have

(3.42.4) lb I v ,(x) - vn+I(x) Jp'(x) do:(x) = o,

where p(x) is an arbitrary 1l"2n-1. Hence V ,(x) - V n+I(x) has at least 2n - 1

variations of sign. Such variations can occur only at the points Xvn and xv.n+I .
In the first case V n+l(x) is constant in the neighborhood of this point, and
since V ,(x) increases, the variation of sign is necessarily a transition from
negative to positive values. The opposite is true for Xv,n+I . The total number
of points (xvn) and (xv.n+Il is 2n + 1. Now V,(x) and V,+I(x) are identical
in the intervals a ~ x < x' and x" < x ~ b, where x' is the minimum and x"
the maximum of all the zeros x.~ and Xv.n+I. Hence no variation of sign is
possible at x' or x". This means that a variation of sign actually occurs at
each of the other zeros and is of the type described above.
As a first consequence of this, we- again obtain (3.42.1), that is, Theorem
3.3.2, and as a second consequence, the inequalities
(3.42.5) V n(Xvn - 0) - V n+J(Xvn - 0) < 0 < V ,.(Xvn + O) - V n+l(Xvn + 0).
These are the same as the inequalities (3.42.3).

3.5. Continued fractions

Historically, the orthogonal polynomials (p,(x) l originated in the theory
of continued fractions. This relationship is of great importance and is one of
the possible starting points of the treatment of orthogonal polynomials. See
Tchebichef 1-8, Heine 3, vol. 1, pp. 260-297, St.ieltjes 11.
(1) For an infinite continued fraction we use the notation

a1l ~I a, I
(3.5.1) bo + I~ + ib;" + ... + Ib, + ...
Here, as usual, the convergent R,j S,, n = 0, 1, 2, ... , is defined as the finite
fraction obtained from (3.5.1) by stopping at the term b,. (See, for example,
Perron 3.} We have
Ro = bo, R1 = bo b1 + a1 ,
(3.5.2)
So= 1, s~ = b1,

and the recurrence formulas

(3.5.3) R, = b, Rn-1 + a, Rn-2 , S, = b, S,_l + a, Sn-2 , n = 2, 3, 4, · · · ,
which hold also for n = 1 if we define R-1 = 1, S-1 = 0. Also, we easily obtain
(see Perron, loc. cit., p. 16)
[ 3.5 l CONTIKUED FRACTIONS 55

R .. S ..-1 - Rn-1 S .. = (-1)"- 1 a1 a2 ·. · a.. ,

or
R.. Rn-1 ( -1)"- 1 a1 ~ • • • a.,
(3.5.4) n = 1, 2, 3, · · ·.
s.. - s.. -1 = s.. s..
-1

(2) Let jp .. (x)} be the orthonormal set of polynomials associated with the
distribution da(x) on [a, b]. The recurrence formula (3.2.1) then suggests the
consideration of the continued fraction

Ca! _ ... _ c.. I

IAax + Ba IA .. x +B..
Here A .. , B .. , C.. have the same meaning as in (3.2.1). Therefore,
(3.5.6) bo=O, b.,=A ..x+B .. , n~1; a 1 =1, a.,=-C .. , n~2.

Next we prove the following theorem:

THEOREM 3.5.1. The convergents R .. / S.. of (3.5.5) are determined by the
formulas

R .. = R .. (x) = co 1(coc2 - ci)t (b p.. (x) - p.. (t) da(t),

(3.5.7) }a X- t

S.. = S.,(x) = c~p .. (x), n = 0, 1, 2, · ...

Here c., has the same meaning as in (2.2.1).
Accordingly, the orthogonal polynomials are identical with the denominators
of the convergents of the continued fraction (3.5.5).
The second part of the statement follows immediately by comparing (3.2.1)
with (3.5.3) for n ~ 2 and observing that the statement is true for n = 0 and
n = 1. As regards t.he first part, we notice that it holds also for n = 0 and
for n = 1. (Since P1(x) = k1x + const., the corresponding integral becomes
k1c0 • Then we use (2.2.15) and (2.2.7).) Finally, if n ~ 2, we have
c~(CoC2 - ci)-t(R., - b.,R.,_t - a.,Rn-2)
= lb{p.
a
(x) - p.
(t) - (A .. x +B.,) lPn-1(x) - Pn-1(t)}
X- t

+ Cn !Pn-2(x) - Pn-2(t) l}da(t)

X- t

= lb- (A .. t + B ..)p.,_1(t) + (A .. x + B ..)p.,_1(t) da(t)

a X- t

= A., lb Pn-1(t) da(t) = 0,

which establishes the statement.

5() GENERAL PROPERTIES OF ORTHOGONAL POLYNOMIALS [ III ]

Therefore, the numerators of the convergents are expressible in terms of

p,(x). ObYiously, R, is a polynomial of degree n - 1 in x.
(3) On expanding the rational function R,(x)jS,(x) in descending powers
of x, we obtain for n ~ 1

(3.5.8) Rn(x) d -1
S,(:r) = onX
+ ({JnX
-2
+ w.!"X + ...
..1 -3

According to (3.5.4) this expansion agree~ 'vith that of R,_1 (x)/ Sn_ 1(x) up to,
2 2
and including, the term x -< "- l Whence there exists a power series
(3.5.9) dox- 1 + d1x- 2 + d2x- 3 + ...
such that, for n ~ 1,

+d + + d2n-1X-2" +v=2n
00
R,(x) _ d -1
(3.5.10) S,(x) - oX 1X
-2
L d.,x-•- 1.
This is generally true for the convergents of any continued fraction of the
type (3.5.5).
THEOREM 3.5.2. The equality
(3.5.11) d• = Co-2( Co C2 -
2)! c. ,
C1 II= 0, 1, 2, • • •,
is valid.
In fact, if d. is defined by these equations and we use (3.5.7), we find
1
R,(x) - S,(x)(dox- + d1x- 2 + · · · + d2n-1X- 2")

= Co
-!( CoC2 - C1
2)!{1bp,(x)- p,(t) da ()
a X- t
t - p" ( X)( CoX
· -1 + ·· · + C2n-1X-2n)} .
Here the expression in the braces can be written in the form
bp,(x) - p,(t) da(t) - p,(x) lb !_=.___x-2nt2n da(t)
[ x-t a x-t
= rb p,(x) - p,(t) x-2nt2n da(t) - rb p,(t) 1 - x-2nt2n da(t)
}a X- t }a X - t

= X-2n lb p,(x) - Pr.(t) t2n da(t)

a X - t

- x-2n ib p,(t)(x2n-1 + x211-2t + ... + xt2n-2 + t2n-1) da(t).

Since the first integral of the right-hand member is a 1r,_1 , the expansion of
1
the first term starts with x-"- • Since the contributions of the powers
2 1
1, t, t , ••• , t"- in the second integral vanish, the expansion of the second
term starts with x-·2"x"-1 = x-"- 1• On dividing by S,(x), we obtain an expan-
sion of the form (3.5.10). This requirement uniquely determines the numbers
do , d1 , ... , d2n-1 , and therefore, the whole sequence Id.).
[ 3.5] CONTINUED FRACTIOXS 57

(4) THEOREM 3.5.3. T~e following decomposition into partial fractions holds
for the convergents of (3.5.5):

(3.5.12) Rn(x) = co2(Coc2 - ci)t

Sn(X)
f: X·~-,
- Xvn
v=l

where "A. = Am and x. = Xvn have the same meaning as in §3.4.

For, we find from Theorem 3.5.1

R;(x.n) = co (~oC~ - ci)t

1
rb
-.Pn(t) da(t).
Sn (Xvn) C(J Pn (Xvn) } a t - Xvn
Now we can apply (3.4.3).
From (3.5.12) we see that the zeros of Rn(x) are real and that they alternate
with those of Sn(x).
(5) Finally, we consider the special case in which [a, b] is a finite interval.
Then the expansion (3.5.9) represents the function

(3.5.13) F(x) = co2(c0c2 - .ci)t lba

da(t)
X - t
provided I xI is sufficiently large. In various special cases, a function F(x)
representable in the form (3.5.13) may be developed directly into a continued
fraction of the type (3.5.5), the denominators of this r"raction being the or-
thogonal polynomials associated with the distribution da(t). Such an approach
to these polynomials is essentially different from that used in Chapter II.
THEOREM 3.5.4. Let [a, b] be a finite interval. Then

(3.5.14) .
l 1m Rn(x) F( )
n-oo
-S()=
n X
x,

if x is an arbitrary point in the complex plane cut along the segment [a, b]. The
convergence is uniform on r:very closed set having no points in common with [a, b].
This theorem is due to A. Markoff (5, p. 89).
If x be real, x > b, we may combine Theorem 3.5.3 with Problem 9 to get

(3 .5 . 15) F( X) _ Rn(X) _ _:2( _ 2)!( _ t)-2n-lk-2 < t < b

Sn(x) - Co CoC2 Cl X <; n ' a= <; = .

This tends to 0 as n ---+ oo (Problem 52) provided x is sufficiently large. On

the other hand, the left-hand member of (3.5.15) is uniformly bounded in the
exterior of an arbitrary closed curve containing [a, b] in its interior, since "A.n > 0
and (3.4.5) holds. Now the statement follows by use of Vitali's theorem
(Titchman-;h 1, p. Ui8). Another proof <'nn he ba:-;ed on Theorem 1:).2.:3.
Concerning further properties of the convergents we refer to Sherman 1
and the bibliography given there. Regarding the relation of the continued
fraction (3.5.5) and of the orthogonal polynomials to the problem of moments,
see Hamburger 1, 2, M. Riesz 2, and the bibliography quoted in these papers.
 CHAPTER IV

JACOBI POLYNOMIALS
In this chapter we shall be concerned with the main properties of Jacobi
polynomials, which include as special cases the ultraspherical polynomials,
particularly, the Legendre polynomials. Among the topics which are not con-
sidered here, but which are reserved for later study, are the properties of the
zeros, asymptotic expressions, expansion problems, and properties connected
with interpolation and mechanical quadrature.,
Addition theorems for Legendre and ultraspherical polynomials have also
been omitted, as have the relations of these polynomials to spherical and surface
harmonics of various dimensions. Limitations of space and the existence of
exhaustive treatises on these subjects are the chief reasons for such an omission.
The interested reader may well consult Whittaker-Watson (1, pp. 326-328,
335) and Hobson (1).
4.1. Definition; notation; special cases
(1) The definition of the Jacobi polynomials P~a,f3l (x) has been given in
§2.4, 1; they are orthogonal on [ -1, + 1] with the weight function w(x)
= (1 - x)a(1 + xt Assurance of the integrability of w(x) is achieved by
requmng a> -1 and {3 > -1; the normalization of p~a,f3l(x) is effected by14

(4.1.1) p~a,f3)( 1 ) = (n ~a).

The orthogonal polynomials with the weight function (b - x) a(x - a/ on the
finite interval [a, b] can be expressed in the form

(4.1.2) const.P~a,f3l{2~ _:.- 1}

·(see the last remark in §2.3). The case a = 0, b = 1 is often used (Jacobi 3;
Jordan 1, vol. 3, pp. 231-234; Courant-Hilbert 1, pp. 76-77). 15
Stieltjes (6, p. 75) writes a and {3 for (/3 + 1)/2 and (a + 1)/2, respectively,
in terms of our notation. The same notation is used by Fejer (13, p. 42).
Jordan's function Zn(u) in our notation is

(-1r{(n +:- )}-~ P~a-')',')'-l)(2u-

1
1).

14 According to §3.3 the zeros of p~a,f3) (x) are in -1 < x < + 1, so that p~a.{1J (1) ~ 0.
lb The statement on p. 76 in Courant-Hilbert must be corrected so as to read
p(x) = x!l-1(1 - x)r!l, q > 0, p - q > -1.
58
[4.1] DEFIXITIOX; XOTATIOX; SPECIAL CASES 59

Courant-Hilbert'~function Gn(P, q, u) is the ~ame a~ Zn(u) with p = a, q = 'Y·

The important identity
(4.1:3)
is readily derived by means of the last remark in §2.3. Combining (4.1.3) with
(4.1.1), we have

(4.1.4)

(2) For a - {3 we have the ultraspherical polynomials. They are even or

odd polynomials according as n is even or odd ( §2.3 (2)).
THEOREM 4.1. The following formulas hold:

p<a.a)(x) = r(2v + a + 1)r(v + 1) p<a,-!)(2 1) 2 -

r(v + a + 1)r(2v + 1)
2
" v X

= ( _ 1)" r(2v + a + 1)r(v + 1) p<-Lal(l _ 2 2)

r(v + a + 1)r(2v + 1) v X '
(4.1.5)
p<~,a)(x) = r(2v + a+ 2)r(v + 1) p<a.i)(2 1) !! -
2
'+1 r(v + a + 1)r(2v + 2) X Xv

= ( -1)" r(2v + a+ 2)r(v + 1) p<i.a)(1 - 2x2).

r(v + a + 1)r(2v + 2) X v

As a consequence of these important relations, Jacobi polynomials with

a or {3 = ±t may be expressed by ultraspherical polynomials. In order to
establish the first relation, it suffices to prove that

L~ P~a.-il(2x 2
1

- 1)p(x)(1 - x 2 )a dx = 0,

where p(x) is an arbitrary 1r2.-1. This is trivial if p(x) is odd. Let p(x) be even
and equal to r(x\ where r(x) is an arbitrary 1l"v-1 • Then we have

r+ p~a,-i)(2x2
J-1
1
- 1)r(x2) ( ( - x 2t dx = 2 t p~a,-i\2x 2
}o
- 1)r(x 2)(1 - x 2t dx

= 1P~a.-i>(2x
1
- 1)r(x)(1 - xtx-i dx

~-~ p~a.-i>(x)rl t(1 + x)) (1

1

= 2-a-i - xt(1 + x)-i dx = 0.

A similar argument may be used to prove the second relation. The constant
factors are determined according to (4.1.1), (4.1.3).
The case a = -1, b = +1, w(x) = \ x \2k, k > -t, can also be reduced to
Jacobi polynomials. The corresponding orthogonal polynomials are (see
Szego 2, p. 349):
60 .JACOBI POLYNOMIALS [IV]

if n = 2v,
(4.1.6)
if n = 2v + 1,
where the constant factors are different from zero; they depend on v and k.
The proof is similar to the previous one.
(3) The simplest cases of ultraspherical polynomials are 16
<-L-i> 1· 3 · · · (2n - 1) 1· 3 · · · (2n - 1)
Pn (x) = 2 _4 ... 2n Tn(x) = _ ... 2n cos n8,
2 4

(4.1.7) p~U>(x) = 2 ~· 3 · .. ~~n + ~~ Un(x)

·4 ··· + n
= 1· 3 · · · (2n + 1) sin (n + 1)8
2
2 · 4 · · · (2n + 2) sin 8 '
where x = cos 8, and Tn(x) and Un(x) denote the Tchebichef polynomials of
the first and second kind [(1.12.3)]. This follows from

n ~ m,

because of (4.1.1).
17
In this connection, two ttmixed" cases of importance may be mentioned:
p<L-i\ ) = 1·3 · · · (2n - 1) sin {(2n + 1)8/2)
n X 2·4 • · · 2n sin (8/2) '
(4.1.8) X = COS 8.
p<-U>() = 1·3 · · · (2n- 1) cos {(2n + 1)8/2)
n X 2·4 · · · 2n COS (8/2) '
The proof is similar to that of (4.1.7) (or is obtained by setting a= ! and a = -!
in (4.1.5)). Formulas (4.1.7) and (4.1.8) also follow from §2.6 by putting
p(x) = 1 there.
Another important ultraspherical case is a = {3 = 0, that is, the Legendre
(0 0) .
polynomials Pn · (x) = Pn(x). Less elementary cases are a = {3 = -!, and
a = {3 = --§-,for which Koschmieder (1) gave representations in terms of elliptic
functions.

4.2. Differential equation

(1) THEOREM 4.2.1. The Jacobi polynomials y = P~a,fJ>(x) satisfy the following
Linear homogeneous differential equation of the second order:
(4.2.1) (1 - x 2 )y" + [{3 - a - (a + {3 + 2)x]y' + n(n + a + {3 + l)y = 0,
16 In the first equation the coefficient of T n(x) is 1 for n = 0.
17 For n = 0, the numerical factor on the right side is 1.
 [ 4.2 l DIFFEREXTIAL EQCATIOX 61

or

+ n(n +a+ {3 + 1)(1 - xt(1 + x)fiv = 0.

To prove this, we note that since y is a ?rn, the expression
dl(1 - x)a+1(1 + x)fJ+ly'J/dx
has the form (1 - x)a(1 + x)fiz, where z is also a ?rn. In order to show that
z = const. y, we prove the orthogonality relation

1 +1
-1
dd
X
I (1 - xt+1(1 + x)fJ+ly'Jp(x) dx = 0,

where p(x) is an arbitrary ?rn-1. An integration by parts reduces the left-hand

member to

- J_~ 1

(1 - xt+ (1
1
+ x)fJ+ly' p'(x) dx,
since a + 1 and {3 + 1 are positive. A second integration by parts gives

1 +1
-1
Y dd
X
I (1 - xt+l(l

In the last integrand the coefficient of y is of the form (1 - x)a(1

+ x)fJ+ p'(x) l dx.
x)fJr(x),
1

+
where r(x) is a ?rn-1. Hence this integral vanishes and the statement is estab-
lished. The constant factor -n(n + a + {3 + 1) may be determined by com-
paring the highest terms.
An alternative form of (4.2.1) is
2
(1 - x )Y" +[a- {3 + (a+ {3- 2)x]Y'
(4.2.3) + (n + 1)(n +a+ {3)Y = 0,

Y = (1 - x)a(1 + x)fJy = (1 - x)a(1 + x)fJp~a,fJ>(x).

(2) Replacing n(n + a + {3 + 1) in (4.2.1) by/', we may ask: For what

values of 'Y has this equation a polynomial solution which is not identically
zero?
THEOREM 4.2.2. Let a > -1, {3 > -1. The differential equation
(4.2.4) (1 -
2
x )y" +[/3 - a - (a + {3 + 2)x]y' + ')' y = 0,
where 'Y is a parameter, has a polynomial solution not identically zero if and only if
'Y has the form n(n + a + {3 + 1), n = 0, 1, 2, · · · . This solution is const.
·P~a,fJ>(x), andnosolutionwhichislinearlyindependentof P~a,fJ>(x) can be a poly-
nomial.
To prove this, substitute y = .L:;o..o a.(x - 1)" in (4.2.4). We find
62 JACOBI POLYKOMIALS [IV]

- (x + 1) L"" v(ll - .1)a.(x - 1)"- 1

•=2

+ 1) + (a + {3 + 2)(x
'lO

+ 'Y L
""

- [2(a - 1)] L 11a,(x - 1)"- 1 a.(x - 1)" = 0,

-1 -o
which yields the recurrence formula

(4.2.5)
['Y - 11(11 + a + {3 + 1)]a. - 2(11 + 1)(11 + a + 1)a.+ 1 = 0,

II= 0, 1, 2,
Assuming that y is a polynomial, let us suppose that an is the last nonzero
coefficient. Then we see from (4.2.5) that for 11 = n the coefficient of an must
vanish, that is, 'Y = n(n a {3 + + +
1). Conversely, if this condition is 13atis-
fied, then an+1 = an+2 = · · · = 0 since the coefficient of a.+1 never vanishes.
Now let 'Y = n(n a {3 + + +
1), and let z be a second solution of (4.2.1) or ·
(4.2.2). If we let x --7 ±1 in the relation
(4.2.6) (1 - x)a+ 1 (1 + x)fJ+ (y' z -
1
yz') = const.,
we see that y and z cannot both be polynomial13 unless the constant in the right
member is zero, that is, unless y and z are linearly dependent. This argument
shows that z cannot even be regular at x = -1 or at x = 1, unless y and z +
are linearly dependent.

4.21. Hypergeometric functions

(1) Substitution of x = 1 - 2x' in (4.2.1) yields

ddX~2 + [a + 1 -
2

(4.21.1)
x'(l - x') (a + {3 + 2)x'] ddy'X
+ n(n + a + {3 + 1)y = 0,
which is the hypergeometric equation of Gauss. On account of the second part
of Theorem 4.2.2, for n ~ 1, we obtain the important representation:

= (n+a)n -n, n +a + F( {3 + 1; a+ 1;
1-
2
x)
(4.21.2) 1 L (n) (n +a+ +
1 n
{3 1) · · · (n +a + {3 + 11)
n. ·-o
=
v

·(a+"+ 1) ... (a+ n)(x ~ 1y.1s

ts The general coefficient

(: )<n + a + {3 + 1) · · · (n + a + {3 + ,)(a + " + 1) · · · (a + n)

is to be replaced by (a+ 1)(a + 2) · · · (a+ n) for" = 0, and by
(n + a + {3 + l)(n + a + {3 + 2) · · · (2n + a + {3) for " =n.
[ 4.22 l GENERALIZATION 63

Here, and in what follows, F(a, b; c; x) is the usual notation for the hyper-
geometric series
F(a, b; c; x)
(4.21.3) ~ a(a + 1) · · · (a + v - 1) b(b + 1) · · · (b + v - 1)
•
=1+~ X
•-1 1·2···v c(c+1)···(c+v-1)'
convergent for I x I< 1 and satisfying
2
d d
(4.21.4) x(l - x) ·dx~ + [c - (a+ b + 1)x] d~ - aby = 0.

(See Whittaker-Watson 1, p. 283.) For latter reference we observe that

(4.21.3) is without meaning if c is a non-positive integer. However, it is readily
seen that if m is a positive integer,
lim
c->-(m-1)
(c +m - 1)F(a, b; c; x)

(4.21.5) = ( -l)m-1 a(a + 1) · · · (a + m - 1)b(b + 1) · · · (b + m - 1)

m!(m- 1)!
·xmF(a + m, b + m;m + 1;x),
and the function xmF(a + m, b + m; m + 1; x) satisfies the equation (4.21.4)
with c = - (m - 1).
(2) In the formula (4.21.2) the hypergeometric series stops with the term
in xn. The constant factor in the first part of (4.21.2) is determined by (4.1.1).
Using (4.21.2), note that the coefficient z~a.fJ> of the highest term xn in p~a,fJ\x) is

(4.21.6) z~a,{J) = lim x-n p~a,{J)(x)

x-~
= 2-n ( 2n +a+
n
{3).
(3) Another application of (4.21.2) is the useful formula

(4.21.7) :x (P~a,fJ>(x)) = Hn +a+ {3 + 1)P~~i 1 'fJ+O(x),

which follows immediately when we expand both sides of (4.21.7) according
to (4.21.2).
As an application of (4.21.7) we observe that the successive derivatives
r:(x), T~(x), T~'(x), ... of the Tchebichef polynomial Tn(x) are, but for
constant factors, P~1:..1{(x), P~~~i(x), P~1:!](x), . . . . The first is, except for a
eonstantfactor, Un-l(x) (see (4.1.7)). We note also that the derivatives P:(x),
P~ (x), ... of the Legendre polynomial P n(x) are constant multiples of p~l_:_ { (x),
1

P~ :.. i (x), ... , respectively.

2 2

4.22. Generalization
(1) The second formula (4.21.2) furnishes the extension of the polynomial
p~a,fJ\x)to arbitrary complex values of the parameters a and {3. It is a poly-
64 .JACOBI POLYI\0::\IIALS [IV]

nomial in x, a, and {3. In the following, we again denote this ?rn by P~a,J'i!(x) .
.:\1any of the properties of P~,a,/3J(x) may be extended to this general case. The
polynomial P~a./3J(x) satisfies the differential equation (4.2.1), and the formulas
(4.1.1), (4.1.3), (4.1.4) hold. Some other results, howeYer, (for instance, the
theorem on the location of the zeros, cf. §6.72) must be essentially modified.
Using (4.1.3), the representation of P;,a./3!(x)·as a 7rn in x + 1 can easily be
derived.
(2) By comparison of the corresponding powers of x - 1, we obtain the
identity
p~a,f3>(x)

= ( n
2
+a+ {3)(~)n F(-n -n- a· -2n- a- [3· - 2 )

c; xy
(4.22.1) n 2 ' ' '1-x

= p~a',/3>(~ ~),= a' = - 2n - a - {3 - 1.

Furthermore,

(4.22.2) (7) p~-l,f3>(x) = (n t {3)(x;

1
YP~1!/(x), Zaninteger, 1 ~ l ~ n,
and

(4.22.3)
n + a + {3 + k = 0, k an integer, 1 ~ k ~ n.
In connection with (4.22.2) see (4.21.5). .
(3) Let n ~ 1. A reduction of the degree of P~a,/3>(x) occurs if and only if
n + + +
a {3 k = 0 for a certain integer k, 1 ~ k ~ n. In this case ttx = 'X)

is a zero of order n - k +
1," this being the precise order unless a = -l, l an
integer, k ~ l ~ n. If
(4.22.4) n + a + {3 + k = 0, a= -l, 1 ~ k ~ l ~ n,
the polynomial P~a,f3>(x) vanishe;; identically.
By setting n + + +
a {3 k = e, a +
l = 'TJ, it can be shown that P~a,/3\x)
= er(x) + 'TJS(x), except for term;; of higher order, if e -7 0, 'TI -7 0. Here r(x)
and s(x) arc certain 7rn independent of e and 'TI· In view of (4.22.2) and (4.22.3)
it follo\vs that, apart from constant nonzero factors,
r(x) = (1 - x) 1 P~1:_zn+l-k>(x), s(x) = Pt~·-n+l-k>(x),
or
(4.22.5)

a, {3, n integers, a ~ -n, {3 ~ -n, a+ f:J ~ -n - 1, n ~ 1. Both polynomials

r(x), s(x) are solutions of (4.2.1); they are linearly independent since r(1) = 0,
[ 4.23] SECOND SOLUTION 65

s(1) ~ 0. Also, they have the precise degrees nand k - 1 = -n - a - {3 - 1,

respectively. In this instance the general solution of (4.2.1) is a polynomial.
(4) Once more let n ~ 1. We then see that p~a,tll(1) ~ 0 unless a = -l,
1 ~ l ~ n. If a = -l, x = 1 is a zero of order l, and this is the precise order
unless n + a + {3 + k = 0, 1 ~ k ~ l, in which event we recognize the ex-
ceptional case (4.22.4).
According to (4.~.3) we have p~a,tll( -1) ~ 0, unless {3 = -l, 1 ~ l ~ n.
In this case x = - 1 is a zero of order l, and here this is the precise order unless
n + a + {3 + k = 0, 1 ~ k ~ l, which is again essentially the case (4.22.4).
(5) Let n ~ 0. From (3) a second case may be derived in which the general
solution of (4.2.1) is a polynomial. If we replace n by -n - a - {3 - 1, the
differential equation (4.2.1) remains unchanged, which leads to the linearly
independent polynomial solutions:

(4.22.6)
a, {3, n integers, a < -n, {3 < -n, n ~ 0.
4.23. Second solution
(1) According to the theory of hypergeometric functions, a second solution
of (4.2.1) is given by

(4.23.1) (1- x)-aF( -n- a, n + {3 + 1; 1- a; 1 ; x),

unless a IS an integer. (See Whittaker-Watson 1, p. 286; cf. in particular

19
y 1 and Y2 . The functions (4.21.2) and (4.23.1) are then linearly independent!
Now let a be an integer. If a = -l, 0 ~ l ~ n, the function (4.23.1) is,
but for a constant factor, identical with p~a,tll(x) (see (4.22.2)). The same is
true if a ---+ ao = a positive integer, provided we multiply (4.23.1) by a - ao
before passing to the limit a---+ ao (see (4.21.5)).
Finally, for integral values of a, a < -n, P~a,tlJ (x) and (4.23.1) are linearly
independent, since

p~a,tll( 1 ) = (n: a)~ O

a.nd (4.23.1) vanishes for x = 1. The latter function is a polynomial if and

only if n + {3 + 1 is a non-positive integer, that is, {3 is an integer less than -n.
This is the case referred to in §4.22 (5).
(2) Numerous other representations are obtained for the solutions of (4.2.1)
by using the classical transformation formulas ·of hypergeometric functions.
The only singularities of this differential equation are at x = + 1, -1, and oo.
Interchanging a and {3, and replacing x by -x, we obtain the expansions about
X= -1.

19 This can be readily shown by introducing in (4.21.1), y = x'-az. Analogous methods

can be used in the cases (4.23.2), (4.23.3).
66 JACOBI POLYKOMIALS [IV]

The expansions about x = (X) are especially important. From Whittaker-

20
Watson 1, p. 286, we obtain the solutions

(4.23.2) ( 1 - x) n F ( - n, - n - a; - 2n - a - {3;
1
~ X) ,
(4.23.3) (1 - x)-n-a-{J-1 F (n +a+ {3 + 1, n + {3 + 1; 2n +a+ {3 + 2; 1- -2 -).
X

The first function iR, except for a constant factor, P~,a,fJ>(x) [cf. (4.22.1)]. The
second function is obtained from the fir:-;t by replacing n by -n - a - {3 - 1.
Apart from constant factors, the expre:-;sions (4.23.2) and (4.23.3) arise from
(4.21.2) and (4.23.1), respectively, by replacing a by - 2n - a - {3 - 1,
(1 - x)/2 by 2/(1 - x), and then multiplying by (1 - x) n. Consequently,
(4.23.2) and (4.23.3) are linearly independent unless -2n - a - {3 - 1 is an
integer not less than -n.
.
THEOREM 4.23.1. Let a, {3 be arbitrary, n ~ 0 an integer. The general solution
of (4.2.1) can be represented in the forms

AP~a,fJ>(x) + B(1- x)-aF( -n- a,n-+ {3 + 1; 1- a; 1 ; X)

if a ~ -n, -n + 1, -n + 2, · · ·,

AP~a,f!>(x) + B(1 + x)-f!F( -n- {3,n +a+ 1; 1- {3; 1 ; x)

(4.23.4)
1j {3 ~ + 1, -n + 2, · · ·,
-n, -n

AP~a.fJ>(x) + B(l - x)-n-a-fJ-l F ( n +a+ {3 + I, n + {3 + 1;

2n +a+ {3 + 2; 1- -2 -X) if a+ {3 ~ -n- 1, -n- 2,

respectively. Here A and B arc nrb£trnry constants.
(~) The preercting mmlt.s enable us to pro\'e the following:
THEOREM 4.23.2. If a and {3 arc arbitrary and n ~ 0 is an intrgrr, thPn
(4.22.5) and (4.22.6) arc the only cases in which the general solution of (4.2.1)
is a polynomial. They can be charactcrizrd by one of tlw following sets of conditions:
(a) a, {3 negative integers, a ~ -n, {3 ~ -n, a+ {3 ;£ -n- 1, n ~ 1,
(4.23.5)
(b) a, {3 negative integers, a < -n, {3 < -n, n ~ 0.
VIe see from (4.2.6) that in the case in question a and {3 must be negative
integers. Now, let a < -n, {3 ~ -n; then (4.23.1) is a non-polynomial solu-
2o In particular see the functions denoted by Y21 and Y22 . (There is a misprint in the
corresponding formulas: the exponent of -x should be -A in Y21 and -Bin Y22.)
[ 4.3 l RODRIGUES' FORl\lULA; ORTHONORMAL SET 67

tion. The case a ~ -n, {3 < -n can be excluded by making use of (4.1.3).
Finally, let a ~ -n, {3 ~ -n, a + {3 ~ -n. Then (4.23.3) is a non-polynomial
solution.
(4) Let a be an integer.' In the exceptional cases, excluded in Theorem 4.23.1,
we can show that the second solution contains logarithmic terms in its repre-
sentation about x = +1 (similarly for x = -1, x = oo ). (See (4.61.6).)
An extension of the preceding discussion to arbitrary values of n is also
possible. However, in what follows, we shall confine ourselves to non-negative
integral values of n.
The consideration of the second solution, properly normalized, will be resumed
in §4.61, where some other representations will also be given.

4.24. Transformation of the differential equation

Applying §1.8 to (4.2.1),.we obtain the following important transformations
of the differential equation of Jacobi polynomials:
2 2 2
d u {1 1 - a 1 1 - {3
(i;}. + 4 (T-= x) 2 + 4 (1 + x) 2
(4.24.1) +~(~±a 7- {3 _+ })_±_(~_± _!2~+}_2(3} u = 0,
1 - x2

d 1~ +
diJ·
{1 - ri + 4i - ,, (} + (n +a + 2 + 1)
A • ~ (}
{3
2
f3
2
} u = O,
·-t -;l!l" <'OS"
(4.24.2) 2 2
u=:u(8)= (i->in 2e)"' t ( cos
8)t~+!
2
P~"· 13 '(cos8).

The special cases a = ± ~, {3 = ± ~ arc to he particularly noted.

4.3. Rodrigues' formula; the orthonormal set

(1) Given a and {3 arbitrary, we have

(4.3.1) (1 - xt(l + x/ p~a,{j)(x) = ~=_IX (!!_)n I (1 - xr+a(l + xr+tll.

2nn! dx
First, take both a and {3 greater than -1. A simple application of Leibniz' rule
then t->hows that the right-hand member is of the form (1 - x)"(l + x)tlp(x), p(x)
being a 1r"n . To show that p(x) = const. P~a.tl> (x), it suffices to prove that
(d)n
1 _

where r(x) is an arbitrary

+1
1
dx I (1 -
1r"n-I .
xr+a(1 + xr"+tlJr(x) dx = o,
But integration by parts n times yields the
result
68 JACOBI POLYNOMIALS [IV]

1~
1

( -1t (1 - xr+a(1 + xt+tlr(n)(x) dx,

which vanishes since r<n>(x) = 0. The constant factor can then be determined
by setting x = 1 and using (4.1.1) (see (4.3.2)). ,
Since P~a.fJ> (x) is a polynomial in a and {3 by (4.21.2), and since the same is
true for the right-hand member of (4.3.1) when divided by (1 - xt(l + x/,
it follows that (4.3.1) is valid for arbitrary a and {3.
On calculating the nth derivative in (4.3.1) by Leibniz' rule, we obtain the
important representation

p~a,fJ>(x) = ~ ( : + ~) (n ~ {3) (X 2
1} (X~ 1)n-•

(4 .3 .2) = (n + a) (:__±J)n
n 2
:t n(n - 1) · · · (n - + 1~ (n + {3) (~)·
v=o(a+1)(a+2)···(a+v)
v
x+1Jl

= ( n 2
(X
n + a) -+-1)n F ( -n - n- [3· a+ 1· x-- -1) .
' ' 'x+1
(2) The argument used in (1) readily leads to the formula

[~
1

(1 - xt(l + x/{P~a,fJ>(x) ) 2 dx
(4.3.3) 1
- 2a+fJ+ r(n + a + 1)r(n + {3 + 1) = h (a,{J)
2n + a + {3 + 1 r (n + 1) r (n + a + {3 + 1) n •

(Here we have the inequalities a > -1, {3 > -1, and for n = 0 the product
(2n + a + {3 + 1)r(n + a + {3 + 1) must be replaced by r(a + {3 + 2).)
In fact, because of (4.3.1) and (4.21.6), we have
+1 1+1
1_
1
(1 - x)a(l + x)fi{p~a,fJ>(x) ) 2 dx = z~a,fJ> _ (1 - x)a(1 + x/P~a,fi>(x)xn dx
1

= (-
1
2nn!
r z~a,{J) !+1 (!i)n
-1 dx
{(1 - xr+a(l + xr+{Jl xn dx

Now we employ (4.21.6) and (1.7.5).

Using the notation (4.3.3), we obtain as the orthonormal set associated with
the weight function (1 - xt(l + x/ in [ -1, +1]
Pn(x) = {h~a,{J))-!p~a,{J)(x)

(4.3.4) ={~~±_?a ++fJ_+{31___j-_~ !~~n(__ j- 1)r(~t

~ rn+a+
t + {3 +11))}!
rn+f3+
p~a,fJ>(x),
n = 0, 1,2, ···.
[ 4.4] GENERATING FUNCTION 69

4.4. Generating function

(1) Formula (4.3.2) can be written as follows:

(4.4.1) p~a,{J)(x) = 2~i J (1 +X~ 1 z)n+a(1 +X 2 1 z)n+fJZ-n-1 dz'

where we assume that x ~ ±1. The integration is extended in the positive

sense along a closed curve around the origin, such that the points - 2(x ± 1) - 1
lie neither on it nor in its interior. ewe define the first and second factors of
the integrand to be 1 for z = 0.) Hence for sufficiently small values of I w I ,

(4.4.2) £ p~a,fJ}(x)wn = ~ f (
1 +X+ 1 )a( 1
-2- z
+ 1 ){J
X -
-2- .z dz.
n=O 27rt z- W ( 1 + X ~! z)( 1 + X 2 1 z)
The denominator is
(4.4.3) - i(x
2
- 1)wz
2
~ z(xw - 1)-w = t(1 - x2 )w(z- zo)(z- Zo),
where
2 xw- 1 + R
(4.4.4) zo = zo(w) = - -2 ----....:~
1- x w
For Zo = Zo(w) there is an. analogoufl expression with - R instead of R.
Here zo and R are regular analytic functions of w provided I w I is sufficiently
small; we take R(O) = 1. At w = 0 the function z0 has a zero, and the function
Zo has a pole. For sufficiently small I w I , zo lies in the interior, and Zo in the
exterior, of the integration curve of (4.4.2), so that by Cauchy's theorem

to P~a,fJ>wn = [i (1- x )w
2 J- 1
(1 +X~ 1 zo)a(1 + ::_ 2 1
Zoy(zo- Zo)-
1
•

Now, we readily get

1 + X2 +- 1 Zo = 2( 1 - w + R )-1 , 1 +X
-- 2- 1
Zo = 2(1 + w + R)
-1
,

so that
00

L P~a,fJ>(x)wn = 2a+fJR- 1 (1- w + Rra(1 + w + RrfJ

(4.4.5)
= 2a+fJ(1 - 2xw + w2r! 11 - w + (1 - 2xw + w2)!)-a
·11 + w + (1 - 2xw + w2)!)-fJ.
This is the generating function (series) of the Jacobi polynomials (Jacobi 3,
pp. 193-194) which may be established directly for x = ±1. The expressions
I l-a and I l-fJ must be taken positive for w = 0.
70 JACOBI POLYKO}IIALS [ IV]

(2) A slight variation of this argument may be made by writing (4.3.1) in

the form

(4.4.6) p~a,{J)(x) = _!_,

27rt
f(! t 2

2 t -
-
X
1)n(1 - t)"(1
1- X
+ t){J ~.
1+X t - X

Here x ~ ±1, and the integration is extended in the positive sense around a
closed contour enclosing the point t = x, but not the points t = ± 1. Also the
functions ((1 - t)/(1 - x))" and ((1 + t)j(l +
x))fJ are assumed to ~educe to 1
fort = x. We next write
2
(4.4.7) -1 t-- -1 = w-1 , t = w-111 1- ( 1- 2xw + w)2 ! l = x + 1- (x 2 - 1) w + .. ·.
2 t -X 2
Here that branch of (1 - 2xw + w )! must be taken which is equ~l to +1 for
2

w = 0. Then if w describes a small closed curve around the origin, t describes

a curve of the type mentioned above. Furthermore,

(4.4.8) -1 -- t = 2t1
1
- w + (
1 - 2xw + w2)i}-1 ,
1 -x

so that

(4.4.9)
P~a,fJ>(x) = __!_.
2?rt
J W-n2"(1 - w + (1 - 2xw + w2)!}-a
·~11 +w+ (1 - 2xw + w2)!)-fi(1- 2xw + w2r!w-1 dw,
which is the desired result.
(3) A third method of deriving the generating function is based on the fol-
lowing remark. If the function F(x, w) of the right-hand member of (4.4.5) is
developed in a power series in w, it is seen that the coefficient of wn is a poly-
nomial of degree n in x. To identify this polynomial with p~a.fJ> (x) we show
that
+1
(4.4.10)
1_
1
(1 - x)"(l + xiF(x, u)F(x, v) dx,

considered as a function of u and v, is a function only of the product uv, which

is equivalent to the orthogonality property. For x = 1, the identity (4.4.5)
can be proved directly, and this procedure furnishes the normalization of the
coefficients.
In Legendre's case: a = {3 = 0, the integral can be calculated explicitly
(Legendre 1, p. 250). In the general case, Tchebichef (5) transformed this
integral into the form

(4.4.11) 2"+fJ+I 1 1
tfJ(1 - t)"(l - uvt)-"(1.- uvt 2 r 1
dt,

from which the statement follows.

Concerning a fourth method based on Lagrange series, see P6lya-Szeg6 1,
vol. 1, pp. 127, 303, problem 219.
[4.5] RECURRENCE FORMULA 71

4.5. Recurrence formula

(1) In the pret>ent case the general formula (3.2.1) becomes
2n(n + a + !3)(2n, + a +!3 - 2)P~a,fJ> (x)

J3 )P~~~>(x)
2
= (2n +a +!3- 1) I (2n +a +!3)(2n +a +!3- 2)x +a2 -
(4.5.1)
- 2(n +a- l)(n + !3- 1)(2n +a+ J3)P~~~>(x), n = 2, 3, 4, · · ·;

Here, the coefficient of xP~~~> (x) may first be verified by means of (4.21.6);
then, by alternately setting x = +1 and x = -1, the coefficients of P~~~>(x)
and P~~~> (x) may be calculated. Actually, the formula is l;>ut a special case
of the relations between contiguous Riemann ?-functions (see Whittaker-
Watson 1, pp. 294-296).
(2) Using the notation (4.3.3), we obtain the following expression for the
tlkernel" (cf. (3.2.3)):
n
K~a,{J)(x, y) = L (h~a,{J))- 1 P~a,{J)(x)P~a,{J)(y)
v=O

2-a-{J r(n + 2)r(n +a+ !3 + 2)

(4.5.2) =----
2n + a + !3 + 2 r (n + a + 1) r (n + !3 + 1)
P~+~> (;r)P~a,{J) (y) - p~a,fJ> (x)P~+~> (y)
x-y
In particular, for y = 1:
K~a,fJ>(x, 1) = K~a,{J)(x)

= :t 2v +a+
v=O
!3 +
2a+fJ+!
1. r(v +a +!3 + 1) p(a,{J)(x)
f(a + 1)f(v+ j3 + 1) v

- 2-a-{J n + a + 1 r(n + a + !3 + 2)
(4.5.3) - 2n +a+ !3 + 2 r(a + 1)I'(n + !3 + 1)

p(a,{J)(x) -~-t_!:___ p(a,{J)(x)

n n +a+ 1 n+l
1- X

= 2-a-{J-I I',(n +a + J3 + 2) p(a+I.{J)( )

r(a + l)f(n + j3 + 1) n X •

The last representation in (4.5.3) is a consequence of Theorem 3.1.4, since

(1 - xr(l + x)fi(1 - x) = (1 - x)a+ 1 (1 + x)fi. We also note that
p~a+l,fJ>(x) = 2 (n +a+ 1)P~a,fJ>(x) - (n + l)P~+~>(x)
2n+a+i3+2 1-x
(4.5.4) {J) ( {J)
p(a,fJ+l>(x) = 2 (n + !3 + 1)P~a, (x) + (n ~_l)Pn'+l (x)
n 2n + a + !3 + 2 1 X +
The second formula follows from the first one if we interchange a and !3 and use
(4.1.3). Finally, by using (4.21.7) and the last formulas, we obtain
72 JACOBI POLYNOMIALS [IV 1

where
A = ~(n + a)(n + ~)(n + a+~+ 1)
(2n+a+f3)(2n+a+~+1)'

(4.5.6) B = (a _ ) 2n(n + a + ~ + 1)
~ (2n+ a + m(2n + a + ~ + 2)'
C = _ 2n (n + 1) (n + a + ~ + 1)
(2n +a+~+ 1)(2n +a+~+ 2)"
Here P~~~) (x) [or P~~~) (x) 1 can be expressed by means of (4.5.1) in terms of
xP~a.$)(x), P~a,$)(x), P~~~)(x) (or P~~~)(x)). This yields

. (2n +a+ ~)(1 - x2) :X (P~a,$)(x)l

= -n( (2n +a+ ~)x + ~- a\P~a,$)(x) + 2(n + a)(n + ~)P~~~)(x),
(4.5.7)
(2n +a+~+ 2)(1 - x 2):X (P~a.~)(x)l
= (n +a+~+ 1) ( (2n +a+~+ 2)x +a- ~lP~a.~)(x)

- 2(n + 1)(n +a+~+ 1)P~~~)(x).

In addition, we notice the following consequence of the last formula (4.5.3):
K~a,~)(l, 1) = K~a,~)(l)

(4.5.8) = Ta-~-~ r(n + a + ~ + 2)r(n j-~_+ ~

r(a + 1)r(a + 2)r(n + 1)r(n + ~ + 1).
4.6. Integral representations in general
The representation (4.3.1) and its integral form (4.4.6) are closely related to
a classical method used for the integration of the hypergeometric equation and
others of similar type. We again start from the formula (4.4.6):

(1 - xt(1 + x)~ P~a,M

.
(x) = ( !:n J(1 -
27r't
tr+a(1 + tr+~(t - x)-n-I dt,
(4.6.1)
where x ~ ± 1. The integratiOn is extended in the positive sense over a closed
curve enclosing x, but not the points t = ±1. Using an idea of Euler (1), we
try to integrate (4.2.1) and (4.2.3) by means of

(4.6.2) Y = (1 - xt(l + x)~y = J (1 - tr+a(1 + trHCt- x)-n-1 dt,

with a proper choice of the contour of integration. Here x ~ ± 1, and the

path of integration must avoid the points -1, +1, and x. However, we allow
+
-1 and 1 as end-points of a path provided the integrals (4.6.2) and (4.6.3)
are convergent.
 [ 4.6] 11'-i'tEGRAL REPRESENTATIONS IN GENERAL 73

Substituting (4.6.2) in (4.2.3), we obtain

2
(1 - x ) Y" + [a - ~ + (a + ~ - 2)x] Y'+ (n + 1) (n + a + mY
= J (1 -_ tr+a(1 + tr+fJ(t - x)-n- (n + 1)(n + 2)(1
3
( - x 2)

(4.6.3) + (n + 1)[a - ~ + (a + ~ - 2)x](t - x)

+ (n + 1)(n +a+ ~)(t - x) dt 2
)

- - (n + 1) j:t I (1 - tr+a+I(1 + tr+fJH(t - x)-n-2 ) dt.

Therefore, we see that (4.6.2) satisfies (4.2.3), provided one of the following
two conditions is fulfilled:
(a) The path of integration is a closed contour along which the expression
(1 - t) n+a+ 1(1 +
t) n+fJ+ 1(t - x)-n- 2 , or what amounts to the same thing,
(1 - t) a(1 +
tl returns to its original value.
(b) The integration is extended along an arc, finite or infinite, such that the
first expression mentioned vanishes at the end-points.
Specialization of the contour according to these restrictions yields numerous
important integral representations for Jacobi polynomials as well as for other
solutions of (4.2.1). (Cf. §4.61, §4.82.) For a special contour integral allowed
in the sense of (a) and (b), we must first show that y is not identically zero; then y
can be identified with a constant multiple of P~a,fJ) (x), or with some other
particular solution of (4.2.1); finally, the constant factor must be determined.
The resulting integral representations hold, save for some exceptional values of
a and~-
Further integral representations are obtained by replacing n by- n- a-~- 1
in (4.6.2); this does not affect (4.2.3). Thus,
(4.6.4) y = (1 - x)a(l + x)(Jy = f (1 - t)-n-(J-1 (1 + t)-n-a-\t - xr+a+(J dt,

where the contour is chosen as in (a) or (b) above. Instead of the first expres-
sion in (a) we now have
(1 - t)-n-(3(1 + t)--n-a(t - x) n+a+(J-1.

Rodrigues' formula (4.3.1) is a special case of (4.6.2), the path of integration

being a closed curve which encloses x but not± 1. Condition (a) is then satisfied.

4.61. Application; functions of the second kind

(1) THEOREM 4.61.1. Let x be arbitrary in the complex plane cut along the
segment [- 1, +
1]. Let a > - 1, ~ > - 1, n ~ 0. Excluding the case n = 0,
a+~+ 1 = 0, a solution y = Q~a,fJ)(x) of the differential equation (4.2.1), which
is linearly independent of P~a,fJ)(x), can be obtained in the form
Q~a,(J)(x)

(4.61.1)

In the exceptional case: n = 0, a ~ + +1 = 0, we have Qcia,fJ)(x) = const.;

then a non-constant solution is given by
74 JACOBI POLYNOMIALS [IV]

Q(a)(x) = log (.r + 1) + 7r- 1 sin 7ra (x - n-a(x + 1)-li

(4.61.2)
·
1+1 (1 - t)a(1 + t){3
log(1+t)dt.
-1 X- t
The function Q~a.f3)(x) is called Jacobi's function of the second kind. \Yc use
this notation also if n = 0, a + (3 + 1 = 0, for the· function Q(a)(x). In the
special case a = (3 = 0 we write Q~ ' \x) = Qn(x) (Legendre's function of the
0 0

second kind). (Cf. Jacobi 3, pp. 195-197.)

Both (4.61.1) and (4.61.2) are multi-valued (except if a and (3 arc integers).
Both integrals are single-valued and regular in the complex plane cut along the
segment [- 1, + 1]. Obviously, Q~a.f3\x) "'-' x-n-a-{3- 1 as x--. rYJ, which shows
that Q~a,f3\x) is linearly independent of P~a,f3)(x) (except if n = 0, a+ (3 + 1 = 0,
see below). The corresponding property of Q<a\x) is clear. \Ve have, as is
easily seen,

(4.61.3)

The function Q~a,f3)(x) satisfies the differential equation (4.2.1); this follows
from (4.6.2) since the segment [-'- 1, + 1] which is the path of integration,
satisfies the condition (b) of §4.6. Differentiating (4.2.1) with respect to (3, and
substituting (3 = - a - 1, we obtain the solution Q<al (x) since Q6a,f3)(x) = const.
if (3 = - a - 1.
(2) THEOREM 4.61.2. The following representations hold:
+1 p<a,{3)(t)
(4.61.4) Q~a,f3)(x) = !(x - 1)-a(x + 1)-{3
1 -1
(1 - t)a(l + t)f3 n
X -
dt
t

= 2n+a+f3 r(n + a + 1)r(n + (3 + 1) (x - 1)-n-cr-1(x + 1)-{3

(4.61.5)
r(2n +a+ (3 + 2)
2
· F ( n + a + 1, n + 1 ; 2n + a + (3 + 2; x) .
1
Furthermore,
Q(a)(x) = log (x + 1)

(4.61.6) 1- X
1 00
+ ( 1 - - 2 )a+ · 2:
•= 1
(a + v)(1- + -1 + ... + -1)( - 2 )• + c,
v 1 2 v 1- X
r'(-a)
c = r'(1) - - - - - l o g 2.
r(-a)
By use of Rodrigues' formula, (4.61.1) may be integrated by parts n times.
This establishes (4.61.4). From (4.61.1) we readily obtain
(x - 1)a(x + 1l Q~a,f3)(x)
= (- 2)-n-I L
00

•=o
(

n -I
j
n + v) (1 - x)-n-•- 1 ·+1 (1 - tr+•+a(l + tr+f3 dt

= ( -1r+I2a+f3-1 f. (n + V) f(n + V+a+ 1)f(n + (3 + 1) (-2-)n+•+

1
•=0 n f(2n + v + a + (3 + 2) 1- X
 [ 4.61 l APPLICATION; FUNCTIONS OF SECOND KIND 75

This, in connection with the notation (4.21.3), yields (4.61.5).

Turning to the exceptional case, we first notice that for n = 0, a + (3 + 1 ~ 0,
(4.61.5) becomes

Q (a,f3)(x) = 2a+f3 r(a + 1)r((3 + 1) ( - 1)-a-1( + 1)-/3

o f(a + +
(3 2) X X
(4.61.7)
·F(a+l,1;a+f3+2; 1
2
x)·
For a+ (3 + 1 = 0 this will, in fact, be a constant since ((1.7.2), second formula)

~ r(a + 1
1)r((3 + 1)(x - 1)-a- (x + 1)-/3 F(a+ 1, 1; 1; :l
2
x)
= - ! ~--- (x - 1)-a-1(x + 1)-/3(1 - _.2_)-a-1 1 7r

2 Sin 1ra 1- X 2 Sin 1ra ·

Now from (4.61.7), taking into account.(4.61.3), we obtain (4.61.6).
(3) The case n = 0 may be treated in another way by means of the relation
(4.2.6). For z = 1 this becomes

(4.61.8)
This yields an integral representation for y.
THEOREM 4.61.3. Let a> - 1, (3 >- 1. We then have the following integral
representations:

(4.61.9)
Q~a,{J)(x) = - 2a+f3 r(a + 1)r((3 + 1)
r(a + (3 + 1)
1"'co
(t- 1)-a-\t + 1)-/3--1 dt

if a + (3 + 1 > 0,

Q~a,{J)(x) = - 2a+f3 r(a + 1)r((3 + 1)

+ +
r(a (3 1)
{1"' co
[(t- 1)-a-1(t + 1)-{J-1
(4.61.10)
- t-a-{3-21 dt - a---::~-a-;-/3~:-1-:-1} if a + (3 + 1 < 0,

(4.61.11)
Q<a)(x) = 1"'
co
1
[(t- 1)-a- (t + 1)-/3- - --
t 1
1 1
dt +log (x + 1)
+
J
if a (3 1 = 0. + +
2
In the first case the integrand is ,._,_, t-a-/3- , as t ----+ oo, so that the integral is
convergent. The constant factor in (4.61.9) can be obtained by comparing the
principal terms in (4.61.7) and (4.61.9).
In the second case the principal term of the integrand is (a - (3)Ca-f3- 3, so
that the integral is convergent. In the third case the principal term of the
integrand is 2(a + 1)(t + 1)-2, so that here too we have convergence.
(4) Another very general integral representation of a second solution of (4.2.1)
is obtained by choosing the path of integration in (4.6.2) as in Jordan-Poch-
hammer's integral for the Gamma function (see the figure in Whittaker-\Vatson
1, p. 257). This path can be defined by the scheme
76 JACOBI POLYNOMIALS [IV]

(4.61.12) (- 1 - ), (+ 1 +), (- 1 + ), (+ 1 - ).
Condition (a) is then satisfied, x ~ ± 1. The principal term is x-n-a-~-- 1 pro-
vided the integral f (1 - t) n+a(1 + t) nH dt, extended over the contour in question,
does not vanish. This is the case (see Joe. cit., p. 257) unless one of the following
conditions is satisfied:
n +a = 0, 1, 2, · . . ; n + (3 = 0, 1, 2, ...
(4.61.13)
2n + a + (3 + 2 = 0, - 1, - 2, ....
In the special cases, where a + (3 or a - (3 is an integer, the contour can be
simplified.

4.62. Further properties of the functions of the second kind

In the following considerations we again assume that a > - 1 and (3 > - 1.
(1) The possible singular points of Q~a,m(x) are + 1, - 1, oo. In order to
discuss this function near x = + 1 (and also for later purposes), we write (4.61.4)
in the form

(4.62.1) +! (x- 1)-a(x + 1)-~p~a.m(x) 1+1 (1 -

-1
t)a(l +
X- t
t)~ dt

+1 ) p<a.m(t)
+ tl-p<a.~)(
= - !(x - 1)-a(x + 1)-~
+ p~a,m(x)Qcia.~)(x).
1 -1
(1 - l)a(1 n X -
X- t
n - dt

The last integral is a ?rn-I in x (a constant multiple of the numerator Rn(x) of the
nth convergent of the continued fraction defined in §3.5; see the first p~rt of
(3.5.7)). Therefore, if x approaches+ 1, the behavior of Q~a·~\x) is to a certain
extent determined by that of Q~a·~\x).
The discussion of Q6 a.~) (x) near x = + 1 is not difficult. Expanding the
factor (t + 1)-~-I in the integrand of (4.61.9) into a power series in t - 1, we
obtain for a + (3 + 1 > 0, a not an integer,

Qcia.~)(x) = const. + (x - 1)-aM C 2

__:;);

here M(u) is a power series of u, convergent for I u I < 1, and M(O) ~ 0. A

similar representation holds if a+ (3 +
1 < 0 [(4.61.10)]. (In the exceptional
case a+ (3 + 1 = 0, this is not true for Q6a,m(x); however, it is true for Q<a\x),
cf. (4.61.11).) Now, let a be an integer; then we use (4.61.9) again. In view of

the integration furnishes a logarithmic term. Thus,

[ 4.62 J FURTHER PROPERTIES 77

rCOnst. + (.y- o-aJ\11 (

1
2
.T)
(4.62.2) Q6a.~) (x) =
I if a > -1, /3 > - 1; a~ 0, 1, 2, .. · ; a+ /3 + 1 ~ 0,
(--1t
- - Jog -1- + ( :~ , - 1)-aM2 (1-- - X)
2 X- 1 2
if a = 0, 1, 2, · · · ; /3 > - 1.
Here M1(u) and M2(u) are power series convergent for I u I < 1 with M 1(0) ~ 0,
M2(0) ~ 0 (see below). A representation similar to the first one holds for Q<a\x).
We have, for instance,

(4.62.3) Qo(O,O) ( x ) = Qo ( x ) = 1"'oo1-t 2

-+
-dt- = 21-log x-1
-1 . X

[The statement M2(0) ~ 0 requires further comment for a = 0. Taking

x > 1, and then integrating by parts, we have

i"' (t - 1)-1 (t + 1)-~- 1 dt

= (x + 1)-~- 1 log (x - 1) + (/3 + 1) i"' (t + 1)-~- 2 log (t - 1) dt,
so that
M2(0) = lim I Qcia.~)(x) +~log (x- 1) l
x-1+0

= - (/3 + 1)2~ 1 1
(t + 1)-~-2 log (t - 1) dt ~ 0.]

(2) Now we prove the following theorem:

THEOREM 4.62.1. Let x be real, x > 1, and take (x - 1t, (x + 1l real and
positive. We then have, for x ----+ 1 0, +
(x - 1)-a, a > 0,
(4.62.4) Q~a.~)(x)""' log(x- 1), a= 0,
{
1, a< 0.
More precisely,

2a-1 r(a)r(n + 13 + 1~ (x - 1)-a

a> 0,
(4.62.5)
r(r. +a+ 13 1) +
'

a= 0.

The behavior near x = - 1 of Q~a.~) (x) is similar.

The case a > 0 follows from (4.61.1):
+1
Q~a·~\x) "'-' Tn-~- 1 (x - 1)-a
1 -1
(1 - tt-1(1 + t) n+~ dt.

Inthecasea = Oweuse(4.62.1), (4.62.2),andP~a.~)(1) = 1. Inthecasea < 0

the first term of the right-hand member of (4.62.1) vanishes as x ----+ 1 + 0.
78 JACOBI POLYNOMIALS [IV]

Thus the statement is equivalent to Qcia.~)(x) "" 1. This is immediately clear

from (4.61.9) if a+ fl + 1 > 0. If a+~+ 1 < 0, we use (4.61.10) and show
that

+ 1)-~-1 l-a-~-2] dt +
(4.62.6)

On writing
! co [(t ·- 1)-a-l(t -
a+~+1
1 < 0•

(t - 1)-a-l(t + 1)-~-1 = (t - 1)-l(t + 1)-a-~-IG ~ ~y,

we see that (4.62.6), as a function of a and~~ increases with a if a+~ is constant.
But when~ approaches - 1, we find for the left-hand member of (4.62.6) the
result

!
co 1
[(t - 1)-a-l - ca-l] dt + - = 0.

We have, as an instance, [(4.62.1), (4.62.3)]

a

(4.62.7)

where R(x) is a 11"n-I. The logarithmic factor is chosen so that it tends to 0

as x ---+ oo.
(3) THEOREM 4.62.2~. Let a be an integer, a .~ 0. We consider Q~a,~)(x) (real
and positive for x > 1) in the complex plane cut along the line [- oo, + 1]. The?}
(4.62.8) Q~a·~\x + iO) - Q~a,~)(x - iO) = (- 1t- 1 7riP~a·~\x),

-1<x<+l.
This follows from (4.62.1) and (4.62.2).
On the other hand, the function

(4.62.9)

is analytic on - 1 < x < + 1 and satisfies the differential equation (4.2.1).

As x---+ 1 - 0, it displays a behavior similar to that of Q~a.~) (x). In particular,
we find for Qi0 ' 0) (x) = Qn(x)

(4.62.10) Qn(x) = R(x) + !Pn(x) log 11 +

_
X
X,

(4.62.11) Qn(- x) = (- 1)"+1Qn(x), - 1 <X< + 1,

(4.62.12) lim Qn(x)
z-1-0
= + oo.

Here R(x) has the same meaning as in (4.62.7).

In general, if a and ~ are both integers, the function Q~a .~) (x) is regular and
single-valued in the whole plane cut along [- 1, + 1].
(4) The functions of the second kind satisfy the same recurrence formula as
p~a,~)(x) [(4.5.1)], that is,
[ 4.62) FURTHER PRORERTIES 79

2n(n +a+ mc2n +a+~- 2)Q~a.~)(x)

(4.62.13) = (2n + a+ ~ - 1)! (2n + a+ ~) (2n + a + ~ - 2)x + a 2 -

2
~ l

· Q~~~) (x) - 2(n +a - 1)(n + ~- 1)(2n +a+ ~)Q~~~> (x),

n = 2, 3, 4, ....
'
This follows from (4.62.1) on account of (3.5.3) and Theorem 3.5.1. There is,
however, an essential difl:erence if n = 1. We then have, according to (4.62.1),
Qia·~\x) = ~[(,a+~ + 2)x +a- ~] Q6a,i3J(x)

(4.62.14) - 2•+>--'(a + p + 2) r(~(! ~~(~ t) 1) (x- 1)-"(x + o-'.

Therefore, both systems of functions [(4.3.3)]
Pn(X) - lh~a.~) ~-ip~a.~)(x),
(4.62.15)
qn(X) - lh~a.~) ~-!Q~a.~)(x), n = 0, 1, 2, .. · ,
satisfy the same recurrence formula of the type (3.2.1) for n ~ 1, provided we
define
(4.62.16) P-1Cx) = 0,
Thus a p.rocedure similar to that used in §3.2 (2) furnishes, for n ~ 1,
kn Pn+t(x)q,.(y) - Pn(x)qn+l(y)
kn+l X- y
(4.62.17)
_
- Pn
(X )qn (Y) + kn-1 Pn(x)qn-l(y) - Pn-l(x)qn(y)
-· ..
kn X - Y

Here kn denotes the coefficient of xn in the "normalized" polynomial Pn(x).

This formula also holds for n = 0 if we modify it as follows:

(4.62.18)
ko PI(x)qo(y) - po(x)qi(y) _
-
( ) ( )
- Po x qo y
+ cons t . -
q_I(y)
-.
k1 x-y x-y
Adding, we obtain the important result:

f2v+a+~+J: r(v+1)r(v+a+~+1) p<a.~)( )Q(a.~)c)

v=O 2a+Hl f(v+a+1)f(v+~+1) v X ' y

= ~ (y _ 1)-a(y + 1)-~ + 2-a-~

4 62 19
( · · ) 2 y- x 2n+a+~+2
r(n + 2)r(n +a+~+ 2) p~~~) (x)Q~a.~) (y) - p~a.~) (x)Q~~~) (y)
r(n+a-l-1)r(n+~+1) x-y
The constant ~ in the right-hand member can be determined by substituting
n = 0, multiplying by ya+H\ then permitting y ----+ oo, and finally using (4.61.5).
We shall return to thils formula in §9.2, where it will be used in a classical
manner for the expansion of an analytic function in terms of Jacobi polynomials
or of Jacobi functions of the second kind.
80 JACOBI POLY?\Ol\IIALS [IV]

4. 7. Ultraspherical polynomials
(1) If a = (3, Jacobi's polynomial P~a.PJ (x) is called an ultraspherical poly-
nomial.21 The following is the customary notation and normalization:
p<X)( ) = r(a + 1) r(n + 2a -t- }J p(a,a)( )
n X r(2a + 1) r(n + a+ 1) n . X
(4.7.1)
_ r(x + !) r(n + 2X) p<x-l.X-1)( )
a= X- !.
r(2X) r(n + X+ t) n .x '
Here we assume first that a > - 1, or X > - !. Some important special cases
are (cf. (1.12.3))
(4.7.2) p~i)(x) = Pn(x), P~l)(x) = Un(x).
If a or X = 0, the polynomial P~") (x) vanishes identically for n ~ 1.
= - !,
This case will be treated later. (Cf. (4.7.8).)
We next observe a number of formulas and theorems which can be obtained
immediately from the theory of general Jacobi polynomials by setting a = (3 =
X- t, X> - !:

(4.7.3) p~X)(l) = (n + ~- 1);

(4.7.4) p~X)(- x) = (- 1)np~X)(x);
2
(1 - x )y" - (2X + 1)xy' + n(n + 2X)y = 0, y = p~X) (x),
(4.7.5)
(1 -
2
x )Y" + (2X - 3)xY' + (n + 1)(n + 2X 1)Y = 0,
y = ( 1 _ x2)X-ip~x\x);

p~X)(x) = (n + ~- 1
)F(- n, n + 2X; X+ t; 1
; x)

(4.7.6) ( ) ( 2 )
= 2n n + ~- 1 (x _ 1t · F - n, - n- X+ t; - 2n- 2X + 1; _ x .
1
The last formulas define p~XJ (x) for all values of X. If necessary, for som~
special values of X, say X = Xo, the formulas may be interpreted as limits for
X -+ Xo .22 For X = - m, m = 0, 1, 2, · ·. , we obviously have (cf. the first
formula (4.7.6)) P~x) (x) = 0 if n > 2m. In this case

• p~X) (x) ·- { d p(X) ( )}

I Im ·- - n X
x--m X+ m dX X--m
(4.7.7)
== 2 (2m) !(n- 2m- 1)! F ( - n n- 2m.- m
n! ' '
+ !.2. 1 -
' 2
x)
exists. These polynomials are, except for constant factors, again the Jacobi
polynomials P~a,a)(x), c~ = - m - t. For instance (cf. (1.12.3))

(4.7.8) lim X-Ip~X)(x) = (2/n)Tn(x), n~l.

x-o
21They are sometimes called Gegenbauer's polynomials. See the papers of G;;genbauer
(1-7). See also Heine 3, vol. 1, JlV· ~~7-301, 449-464. Occasionally, the notation c;(x)
is used instead of P ~x) (x).
22 This should also be done in all the subsequent formulas.
r 4.7 1 UI./fHASPHERICAL POLY:t\01\IIALS 81

In case a = - l, (n + 1)/2 ~ l ~ n, the polynomial P~a,a\x) vanishes identi-

cally ( §4.22 (3)). However, the corresponding expression P~)o.) (x), as a limit, has
a meaning and does not vanish identically.
(2) Further formulas involving P~'lo.) (x) are

(4.7.9) lim x-"p~'lo.)(x) = 2"(-n +X - 1);

z-~ n

(4.7.11) u = (sin el p~)o.) (cos 8);

(4.7.12) (1 _
- X
2)'/o.-tp<x) ( ) = (~2t r(n + x)r(n + 2X)
" X n! r(X)r(2n + 2X) dx
(:!__)" (1- _ X
2t+'lo.-t.
'

(4.7.13) p<'lo.) ( ) = (n + 2X - 1)(x + 1)"F(- _ _ '" + 1.. '" + .! . x - 1) .

n X n 2 n. n 1\ 2 '1\ 2 'X + 1 '

(4.7.14) .!!:_ p<'lo.)

dx n
(x) = 2XP('Io.+I)
n-1
(x) ,·

(4.7.15)
f +I
-1
c1 _
- X
2)'/o.-t 1p<'lo.) ( ) l2 d = 21-2'/o. r('") l-2
n X X 7r
1 "'
r(n + 2X)
( n + X) r (n + 1)'

X> - .! X ;:e 0 •23

2' '

(4.7.16)
t
n=O
r(2X)
r (X + t)
r(n +X+ t)
r(n + 2X)
p~'lo.)(x)w"

2
= 2'/o.-i(l- 2xw + w )-il1- xw + (1- 2xw + w2)ili-'lo.;

nP~'Io.) (x) = 2(n +X- 1)xP~)o.21(x)- (n + 2X- 2)P~)o.2 2 (x) n= 2 3 4 ····

(4.7.17) ' ' ' ' '

(4.7.18)

(3) We obtain as a second solution of (4.7.5), which is linearly independent

of P~x\x), (cf. (4.23.1), (4.23.3), (4.61.1), (4.61.4), (4.61.5))

(4.7.19)
y ~ (1 - x)HF(- n - h + ~. n + h + t; f - h; 1
2
"),

X ;:C - n + 1/2, - n + 3/2, - n + 5/2, ... ;24

In the limiting case A~ 0, n ~ 1, we multiply (4.7.15) by A- 2,

2
3 Use (1.7.3), n = 2.
},. ~ 0 [(4.7.8)].

1/2, -n - 3/2, · · · this is a polynomial linearly independent of p~'lo.) (x)

2
• For A = -n -

(cf. §4.23 (1)).

82 JACOBI POLYNOMIALS [IV]

2
y = (1 - x)'-n- "'F(n + 2X, n + X + t; 2n + 2X + 1; 2 )
(4.7.20) 1 -X '

X r= - n/2, - n/2 - 1/2, - n/2 - 1, ... ;25

(4.7.21)
X> - 1/2, X r= 0;

2
y = (1 - x)-n->.-i(l + x)'-"'F(n +X+ t, n + 1 ;2n + 2X + 1; - -),
(4.7.22) 1 -X
X r= - n/2, - n/2 - 1/2, - n/2 - 1, .... 26

According to Theorem 4.23.2 the general solution of (4.7.5) is a polynomial if

and only if X - t is an integer and X ~ - n/2.
(4) Another generating function, essentially different from (4.7.16) (cf.
Problem 16), and much simpler, is often used as a definition of the ultraspherical
polynomials, namely:
P~"'l (x) + Pi"'l (x)w + P~"'l (x)w 2 + ... + P~"'l (x)wn + ...
(4.7.23)
= (1 - 2xw + w2)-"'.
For the proof, we consider the recurrence formula (4.7.17), from which
00 oo co
L nP~"'l (x)wn-1 = 2x L (n +X- l)P~"'.2 1 (x)wn- 1
- L: (n + 2X - 2)P~~ 2 (x)wn-I
n-1 n•~1 n-1
and iu which we define P~Y.(x) = 0. If the left-hand member of (4.7.20) Le
denoted by h(w), the last equation may be \vritten in the form

h'(w) = 2.xw 1-\w"'h(w))' - w2 - 2\w2"'h(w))'

= 2.x1Xh(w) + wh'(w) l - f2Xwh(w) + w2h'(w) l;
or h'(w)/h(w) = 2X(x - w)(1 - 2xw + w2)"- 1• Since h(O) = P6"'l(x) = 1,
(4.7.23) follows by integration.
On differentiating (4.7.23) with respect to X, we obtain
(4.7.24) - (ll - 2xw + w2 )-"' log (1 - 2xw + w2)
which, for X = - m, m = 0, 1, 2, ... , gives the generating function of the
polynomials (4.7.7), provided only those terms are considered for which n > 2rn.
The::.e polynomials are, as mentioned in (1), constant multiples of the Jacobi
polynomials P~a,al(x) with a= - .m- t. The polynomials of degree n ~ 2m,
defined by this new generating function (4.7.24), are essentially different from
Cf. §4.23 (2).
2&

Formula (4.61.5) has been considered only under the restriction a > -1, {3 > -1.
26

We see immediately, however, that except for the indicated values of>., the function (4.7.22)
is a solution which is ""X_"_ 2A as x ~ oo, so that it cannot be a polynomial.
[ 4.7) ULTRASPHERICAL POLYNOMIALS 83

the P~"'\x) which are given in this case by the corresponding terms of (4.7.23).
We have, for instance, [(4.1.7)],

rPi"(x) = I, P~"(x) : 0, n ~ I,

(4.7.2q ) {\'" =,O 1- log (1 - 2xu: + w ) = n~

2

0
- 1 (2/n)Tn(x)wn

= L ~ . 2·4 · · · 2n pC-!.-ll(x)wn.27
n-1 n 1· 3 ... (2n - 1) n '

Pi- 1 l(x) = - 2x, p~- 1 l(x) = 1,

p~- l (x) = 0,
1
n ~ 3,
(4.. 7.26) X = - 1
- (1 -- 2xw + w ) log (1- 2xw + w2) = 2xw- (2x 2 + 1)w2
2

- 32.£ 2·4 ... (2n- 6) p~-Jd\x)wn.

n-3 1 · 3 · · · (2n - 3)

(5) From (4.5.5), (4.5.6), (4.5.7), (4.7.14), we obtain the relations

2
(1- x ) :x {P~"'l(x)}
1
= [2(n + x)r {(n + 2X- 1)(n + 2X)P~"'21(x) - n(n + 1)P~"'~1(x) l
4 7 27
( · · ) = - nxP~"'l (x) + (n + 2X - 1)P~"'21(x)
= (n + 2X)xP~"'l (x) - (n + 1)P~"'~1(x)

n ~ 0, P~{(x) = 0.
We can then derive the following identities:

nP~"'l(x) = x:x IP~"'l(x)}- d~ {P~"'21(x));

(4.7.28)
(n + 2X)P~"'l(x) = :x {P~"'~1(l:)J- x :x {P~"'l(x)}.
Adding these formulas we find, by use of (4.7.14),
_!:__
dx
{P~"'~~(x) - P~"'l.1(x) l = 2(n + X)P~"'l(x)
(4.7.29)

n ~ 1, P~{(x) = 0.
(6) Finally, we give some special formulas involving hypergeometric functions.
Combining (4.1.5) and (4.21.2), we obtain

p~:)(x) = (
211
+ ;~- 1)Fc- II, II+ X; X+~; 1- X
2)

(4.7.30) = ( -- 1)" (
11
+ ~- 1) F(- 11, 11 +X;~; x2),
27
The first of these identities can be derived directly by writing 1 - 2w cos 0 + w2 =
(1 - wei 6 ) (1 - we-i 6 ).
84 JACOBI POLYNOMIALS [IV I

P 2v+1 __ (211 ++ 2X) X F( - 11 II+ X+ 1,. X +

<>-l .( X ) -- 1 •
2, 1 2)
-X
211 1 1
(4.7.30)

= (- 1)"2X (
11
~X) xF(- 11, 11 +X+ 1; ~; x2).
The constant factors ean be calculated by substituting x = 1 and comparing
the highest powers [or by computing P~;l(O) and P~;!;1 (0) from (4.7.23)].
Another way of writing the second part of (4.7.30) is the following:

(4.7.31 ) Pn<>-l( X ) = ~~~

1
( )m r(n - m + X) ( )n-2m
~ -1 r(x)r(m.+ 1)r(n- 2m+ 1) 2x .
This last expression is an explicit representation of the ultraspherical poly-
nomials. (Cf. Problem 15.)
The formulas (4.1.5) may be proved by a more general consideration. For
instance, in the first case, we may start from the corresponding differential
equations
2
(1 - x )y" - 2(a + 1)xy' + 211(211 + 2a + 1)y = 0,
(4.7.32)
(1 - x~)z" -- (a+!+ (a+ Vxlz' + 11(11 + a+ -!)z = 0
and show the relation y(x) = z(2x2 - 1) between their general solutions. This
argument furnishes at the same time relations between the non-polynomial
solutions of (4.7.32) similar to (4.1.5). To be specific, let us replace the quanti-
ties n, a, {3, x in the expression (4.23.3) first by 11, - t, a, 1 - 2x2 , respectively,
and then by 11, + t, a, 1 - 2x , respectively; we then find (in the second case
2

after multiplying by x)

y = x-n- 2 a- 1Ji'([(n + 1)/2] +a+ 1/2, [n/2] +a+ 1;n +a+ ~;x- 2 )
(4.7.33) 2>. 2
= x-n- F((n + 1)/2 + X, n/2 + X; n + X+ 1; x- ),
as a second solution of (4.7.5), which is linearly independent of p~>-l(x), provided
X r= -[(n + J.)/2]- k;k = 0, 1, 2, ....
Starting from the fir~;t formula in (4.7.21), we find, save for a constant factor:

Y = (1- i)i->-x-n-1 f (n +211 211)x-2• 1+1 (1- t2r+>--it2"dt

v-o -1

= (1 _ i)t->-x-n-1 i:: (n + 211) r(n + X+ t)r(~~ + t) x-2v

(4.7.34) v-o 211 r(n + X+ II+ 1)

= r(n + X + t)r(!) ( _ 2)t->- -n-1
1
f(n +A.+ 1) X X

2
·F((n + 1)/2, 1 + n/2; n +X+ 1; x- ).

For further properties of ultraspherical polynomials the reader may consult

Whittaker-Watson 1, pp. 329, 330, 335, and the literature quoted there. See
also Wangerin 1, pp. 7a0-731. The function C~(x) of these authors is identical
with p~l (x) in our notation. The Legendre associated functions P'::(x), m an
integer (cf. Hobson 1, p. 90), can be represented in the form (cf. (4.21.7))
(4.7.35)
[ 4.8) REPRESENTATIONS FOR LEGENDRE POLYNOMIALS 85

4.8. Integrall representations for Legendre polynomials

In the important case a = {3 = 0, that is, for Legendre polynomials, the

method of §4.6 leads to various integral representations. According to (4.6.1),

(4.8.1) Pn(x) = -1;

211't
f 2
(1-t -- -1)n -dt
2t - X
-.
t - X
Here the path of integration encloses the point x. At present, the position of
this contour with referenee to the points t = ± 1 is immaterial.
(1) Integrals of Dirichlet-Mehler. (Cf. Dirichlet 1, Mehler 5.) Take x in
the interior of the interval [- 1, + 1], so that x = cos 8, 0 < () < 11'. For the
contour in question we choose the circle
(4.8.2) It - 1 I = 1eie - 1 1 = 2 sin~, t = 1
. ()
+ 2 sin 2- .e
iY,

'
so that

1 t2 _ 1 1 + sin ~ · ei"'
(4.8.3) - - - -()- -
2 t - cos() .
1 + sm .e-il/1
2
Next let 1/; vary from - 1r to + 1r; then 1 + sin !O ·ei"' describes the small circle
in the figure. The expression in the right-hand member of (4.8.3) has the

a= sine/2
F10. 3

absolute value 1 and an argument twice that of the numerator. Thus, if we

write
2
(4.8.4) 1 t - 1 =cit/>
2 t - cos() '
the quantity cp varies from () to + () and again from + () to - 8. Solving
(4.8.4), we obtain
(4.8.5) t == ei<l> + ei<l>/ 2 (2 cos cp - 2 cos e)t.
Here the po:;;itive value of ( )t corresponds to I 1/; I < (1r + 8)/2, that is, to
the "exterior" arc, while the negative value corresponds to I 1/t I > (1r + 8)/2,
that is, to the "interior" arc. Furthermore, from (4.8.4) it follows that
t dl~ = eit/> dt + (t - cos e)ieit/> dcp;
whence
 86 JACOBI POLYNOMIALS [IV]

dt iei<l> dcp iei<1>! 2 dcp

t----co_s_e = t - ei<l> - (2 cos cp - 2 cos e)i.
Finally,

p (cos e) = -
1 1+8 e'nt/>. - - - - -dcp- - - - + -1 1-8 e'ntj>.
ei<i>/2
e
itl>/2 dcp
--- .
n . 271' -8 (2 cos cp - 2 cos e)i 271' +8 - (2 cos cp - 2 cos e)l'
or,

(4.8.6) , e) -_
P n ( COS 21
cos (n + ! )cp · dcp
-
o (2 cos cp - 2 cos e)i
1r '
8

where the square root of 2 cos cp - :2 cos e must be taken with positive sign.
This is the first formula of Dirichlet-l\1ehler. Substituting 1r - e for e we ob-
tain, because of (4.1.3), the second formula

(4.8.7) p n (cos e) = ~
1r }8
r7r ( sin (n + !)cp )i dcp.
2 cos e - 2 cos cp
(2) First integral of Laplace. (Cf. Whittaker-Watson 1, pp. 312-313.) Let
x be different from ± 1, and choose the circle
(4.8.8) It - x I = I x2 - 1 Ii
as the contour of integration. Writing .t = x + (x 2 - 1)iei<l> (with an arbitrary
but fixed determination of (x 2 - 1)'), we find that
2
1t - 1 ( )i
(4.8.9) - - - ==
2 t- X
X + X
2
- 1 COS cp -dt- = '/,'d cp.
' t- X

Consequently,

Pn(X) =
2
1
71' -..
1+ . lx + (i - 1)i COS cp} n dcp
(4.8.10)
= 71' -l 1. {X + (x 2 - 1) 1 COS cp j n dcp,

an expression known as Laplace's first integral. It holds for arbitrary values

of x.
(3) Second integral of Laplace. (Cf. Whittaker-Watson 1, p. 314; Jacobi 2,
p. 153.) This integral is given by

Pn(:t) = 171' -..

2
1+.. lx + (i - 1)i COS
--
cp} n 1 dcp
(4.8.11)
= 71'-
1
1. (x + (x 2 - 1)i coscp}-n- 1 dcp.

It may be derived from the first integral of Laplace in the following manner.
Let 0 <r< 1 and
1+ r
2
2 )i _ 2r
X=-- ( X- 1 ---
(4.8.12) 1 - r2 ' 1 - r2 '
Z - ei<l>
- '
whence,
[ 4.8 J HEPRESENTATIONS FOR LEGENDRE POLYNOMIALS 87

(4.8.13) - --- -lrn

P n (X ) = (1 -- [ I 1 + rz 12" I dZ I.
271' /z/=1

Now, substituting
2 2
w-r 1- r dz 1 - r
(4.8.14) z = ---
1 -no'
1 + rz =
1 - rw
,
dw
-
(1 - rw) 2 '
28
we obtain

(4.8.15)
Pn(x) = (1 -- r2)-n r (1 - r2)2n 1 - r2 I dw 1-
211' }lw/=1 )1 - rw) 2n )1 - rw) 2
Substituting w = -ei<l>', we obtain (4.8.11 ). Using analytic continuation, this
formula may be immediately extended to arbitrary complex values of x.
(4) As a special case of (4.4.9) we note the representation

(4.8.16)

+0

Fw•. ·I

The integration is extended in the positive Hen.se along a contour enclosing the
origin but neither of the points e±iB. By a proper choice of the contour, formu-
las (4.8.6) and (4.8.10) can be derived again from (4.8.16) (see P6lya-Szego 1,
vol. 1, pp. 115, 287-288, problem 157).
(5) Integral representat£on of Stieltjes. (Stieltjes 8.) Suppose 0 < () < 1r.
This important representation can be obtained f.rom (4.8.16) by using the con-
tour in the figure. (The derivation from {4.8.1) seems too complicated.) Since
the integrand is O(w-n-2 ) as w ~ oo, the contribution of the large arcs tends to
zero as the radius becomes infinite. The same is true of the contributions of
the small arcs around e±i8 (the integrand is 0 II w - e±i8 1-i l there). Thus,
we have

Pn(cos e) = 2'/R 2~i I 1

w-n- (1 - 2w cos() + 2
w )-i dw,

extended twice over the straight line w = t-V 8 , where t increases from 0 to 1
and then decreases from 1 to 0. We have·, then,
(1 - 2w cos () + w2)-i = ± te-i\1 - t)-i(l - te_2,.8 )-i,

2s Concerning this argument see Szego 21.

-------·~·······
88 JACOBI POLYNOMIALS [IV j

where the signs + and - correspond to the first and second cases, respectively
218
[(1 - t)-i = (1 - te- )-i = 1 fort = 0]. Therefore,

Pn(cos e) = 49? _1_.

271't }o
etn+1e-i(n+1)8te-i8(1 - t)-i(l - te-2i8)-i( -t-2ei8) dt
(4.8.17)
= !S {ei(n+1l811 t(1 - t)-i(1 - te2i8)-i dt}.

(6) Further integral representations are obtained by replacing n in (4.8.1) by

- n - 1 (cf. (4.6.4)) and observing the conditions (a), (b) formulated in §4.6
2
(here the expression (1 - t )-n(t - 1
xr-
must be considered). We then have

4 8 18
( · · ) Y
= _1.f (~2t-x
271't
t2 - 1)-n-1 __!!!:_ •
t-x
Integrals of this type may represent Legendre polynomials as well as Legendre
functions of the second kind, or a proper linear combination of these, according
to the special choice of the path of integration.
By using the contour in (1), we again obtain Dirichlet-Mehler's formula
(4.8.6), since the expression in the right-hand member does not change if we
replace n by - n - 1. The same procedure transforms Laplace's first integral
(4.8.10) into Laplace's second integral (4.8.11), and conversely. Thus, choos-
ing the contour in (4.8.18) as in (2), we obtain the second integral. The ex-
pression in the right-hand member of (4.8.11) cannot represent a solution other
than Pn(x), Rince it is finite and equal to + 1 at x = + 1.

4.81. Legendre functions of the second kind

(1) Let x be in the complex plane cut along the segment [- 1, + 1], and x =
Hz + z- 1), I z I < 1. We deform the path of integration in (4.61.1) into the
circular arc through - 1, z, + 1 (cf. Whittaker-Watson 1, p. 320, example 1).
This deformation is permitted since sgn ,Sx = -sgn ,Sz. Then
t _ (z + 1)eT + (z - 1)
- (z + 1) eT - (z - 1) '
(4.81.1) 2
1t - 1 dt 1
- - - = - - - = (x·+ (x 2 - 1)i cosh rj- 1•
2 t -X dr X - t
The new variable r is real and ranges from - co to + oo; furthermore (i - 1)i
I'V x as x ~ oo. This furnishes the following integral representation for
0 0
Q~ · l(x) =. Qn(x), which is very similar to Laplace's second integral:

(4.81.2)

(2) Let X be real, X > 1; we write X = cosh r, r > 0. Introducing in (4.81.2)

COSh r + Sinh r COSh T = /,
(4.81.3) e en
dr e e
de = sinh r sinh T = -(2_C_O_S_h_e---2-CO_S_h_r........,)P
we obtain
 [4.82] GENERALIZATIONS 89
oo

j
-(n+i)9

rr· de.
(4.81.4) Qn(cosh r) = (2 he 2 h
cos e- cos
In this formula (Watson 2, p. 154) we first assume that r >
0. By analytic
continuation it can he extended to the half-strip (1.9.3). The path of integra-
tion is the horizontalline ~R(e) ~ ~(r), .g(e) = 3(r).' This formula corresponds
to Dirichlet-Mehler's formula.

Flo. 5

4.82. Generalizations
(1) Generalization of th~1 Dirichlet-Mehler integral. (See Fejer 12.) Suppose
0 < e < 1r and 0 < X < 1. According to the generating function (4.7.23) we
have

(4.82.1)

Here we take the unit circle I w I = 1 as the path of integration, avoiding the
9
singular points w = e±i , in the neighborhood of which the integrand is
0 (I w - e±iB 1-x l· (See Fig. 5.) For w = ei<~>, 0 ~ cp < e, we have
(1 - 2w COS e + W )-X = t:-iXtj>(W- 1 2 COS e + wr-X = e-i>.tj>(2 COS cp- 2 COS e)->-,
2
-

and for w = ei<~>, e < cp ~ 1r,

(1 - 2w cos e+ w2)->. = e-i>.(t/>-rl(2 cos e- 2 cos cp)->-,

so that

p~X)(COS e) = 2~{~
27rt
re e-i(n+ll<i>e-iXt/>(2
}o
COS cp - 2 COS e)-\eitj> dcp

whence

P~Xl (cos e) == 1r -
1
l r F(cp) I 2 cos cp - 2 cos e 1-x dcp,
(4.82.2)
+ X)cp o ~ cp < e,
l
cos (n if
F(cp) ==
cos [(n + X)cp- X1r] if e < cp ~ 7r.
The expression resulting from (4.82.2) for X = 1/2, is the sum of the expres-
sions (4.8.6) and (4.8.7).
(2) Generalization of the integral of Stieltjes. (Stieltjes; Hermite-Stieltjes 1,
vol. 2, p. 122, no. 284.) If we choose in (4.82.1) the same contour as in §4.8 (5),
an argument similar to that used there leads to the representation
 90 JACOBI POLYNOMIALS [IV]

(4.82.3) p;" (cos 8) = (2/T) sin XT ;s {•"'" f.' t"""''-'(1 - t)-\; - te';')->at},
l(8) = (n + 2X)8 +(! - X)1r; o< 8 < 1r, o <X < 1.
The formulas in §4.8 (1) and (2) can be extended to Jacobi polynomials
P~a.f3\x) for integral values of a. The resulting representations, however, are
rather complicated. Finally, we notice the follo~ing generalization of (4.81.2)
arising from (4.61.1) in a manner similar to that used in §4.81 (1) (notation the
same as there):

Q~a.f3)cx) =! (r ~ z)a(r ~ zY [:"" lTI (1 + z)eT + 1- zra-{3

(4.82.4)
· {x + (x 2 - l)i cosh r l-n-1 dr,
1
x = t(z + z- ), I z I < 1; a >- 1, {3 >- 1,
with a proper determination of the functions involved.

4 . ~. Trigonometric representations
(1) Finite cosine expansion of Legendre polynomials. From the generating
function (4.7.23) we obtain, for X = !,
(1 - 2w eos 8 + w2)-i = (1 - we,. 8)-!(1 - we-i8 )-i
(4.9.1)

where

(4.9.2) go= 1, 1 · 3 · · · (2m - 1)

gm == m = 1, 2, 3, · · · ,
2 · 4 ···2m
so that
n
P,.(cos 8) = L: gmeimBgn-me-i(n-m)B
m=-0

(4.9.3)

2g(n-1)/2g(n+1)/2 cos 8, n odd,

+ { 2
g n/2 ' n even.
Consequently, P ,.(cos 8) is a trigonometric cosine polynomial with non-negative
coefficients.
Relation ( 4.9.3), as an identity in 8, can be written as follows in terms of
Tchebichef polynomials:

P,.(x) = 2gog,.T,.(x) + 2g1gn-1Tn-2(x) + · · ·

(4.9.4) 2g(n-1) /2g(n+1)/2T1 (x), n odd,
+ { 2
gn/2, n even.
(2) Infinite sine expansion of Legendre polynomials. (Heine 3, vol. 1, pp.
19, 89.) We have
[ 4.9 J TRIGONOMETRIC REPRESENTATIONS 91

4 ~~ · 4 · · · 2n . .
Pn(cosfJ) =; 3 . f) ... ( n + 1) lfo sm (n + 1)8 +
2 f 1 sm (n + 3)8 + · · ·
(4.9.5) + f, sin (n + 211 + 1)8 + · · ·},
fo = 1;
(n + 1)(n + 2) .. · (n.+ 11)
f, = f.n = g, (n + V(n + V ... (n + 11 + !) , 11 = 1, 2, 3, · ...

For n = 0 the factor (2 · 4 · · · 2n) / (3 · 5 · · · (2n + 1)) must be replaced by 1.

Abel's transformation ( §1.11 ( 4)) shows this expansion to converge for the values
0 < 8 < 1r, and even uniformly for E ~ 8 ~ 1r - E, 0 < E < 1rj2. The conver-
gence may be deduced also from the elements of the theory of Fourier series. In-
deed, (4.9,.5) is the formal expansion of the function defined by Pn(cos 8) for 0 < 8
< 7r' and: by - p n (cos e) for - 71' < (J < 0 (see below). It is a generalization
of the classical expansion (n = 0)
(4.9.6) 1r/4 = (sin 8)/1 + (sin 38)/3 + (sin 58)/5 +
First pr:oof (Heine, loc. cit.; cf. also Fejer 20, pp. 24-26). Using the notation
(1.12.3), We have

(4.9.7) 1r Pn(cos e) sin (m + 1)8 de = 1-: 1

Pn(t)Um(t) dt.
This integral vanishes .if :m < n, or if m ~ n, and m - n odd. Now write
m = n + 211; using (4.9.3), we find

{" Pn(cos 8) sin (m

}o
+ 1)8 de :~:: t
k-o
gkgn-k {"cos (n - 2k)8 sin (m + 1)8 de
}o
(4.9.8)
= .t;
n
(J kgn-k
( 1 1
m + 1 + n - 2k + m + 1 - n + 2k
)

gkgn-k
= 2 :L:
=
n
gkgn-k L: n

k=O m + 1 - n + 2k k=O II + k + ! .

Considering 11 as a continuous variable momentarily, we easily verify (by calcu-

lating the residues) that

(4 .9 .9 ) t gkgn-k = (11 + 1)(11 + 2) · · · (11 + n) = 2 2 · 4 · · · 2n j.,

k-o 11 + k +! (11 + !)(11 + D · · · (11 + n + !) 3 · 5 · · · (2n + 1)
which establishes the statement.
Second proof (see Fejer 19, pp. 202-203). Expansion of the last factor in
the integrand of (4.8.17) gives

(4.9.10)

where gm has the same meaning as in (4.9.2). The last integral can be calcu-
lated by means of (1.7 .5), and the 1'Jtatement is thus established.
Third proof. We use mathemat[cal induction with respect ton. According
to the recurrence formula of Pn(x) ((4.7.17), X=!), it suffices to show that
n2

(4.9.11)
n2 - .l !vn = !v,tl-1 + fv+1, 71-1 - fv+1, n-2,
4

II= 0, 1, 2, '· ·, n = 1, 2, 3, '' · jfv+1,-1 = 0.

92 JACOBI POLYNOMIALS [IV]

For n - 0 we have (4.9.6). The identity (4.9.11) can be verified by direct

calculation.
Remark. Formula (4.9.5) is closely related to certain expansions of the
functions of second kind. We find from (4.61.4), a = (3 = 0,

(4.9.12) Q~O,O)(x) = Qn(x) = -~

2
!+
-1
1
Pn(t) dt.
X- t
If we substitute x = !(w + w- ) (§1.9), the function Qn I !(w + w- 1)) will be
1

regular for I w I ~ 1, w ~ ± 1. Using (4.7.23), ;\ = 1, we obtain for I w I < 1,

+1 1---~__r:___:_+ 1+1 Pn(t)\f L
Qn I !(w +
·
1
W- ) l = 'W
1
so that, because of (4.9.7), (4.9.8), (1.9.9),
-1
p (t)
~ -.LtW
__ dt =
W
2 'W
-1
00

\_m=O
}
Um(t)wm dt,

(4.9.13) I w I< 1.

Now let I w I< 0

1, w --t c' , 0 < lJ < 1r. Then !(w + w- 1 ) --t cos 0 - iO, so that

'O) = 2 2 · 4 • · · 2n ~ j i(n+2v+l)6
(4.9.14) Qn ( cos ll
u
_
z . 3 . 5 ... (2n + 1) ~ .e ' 0 < 8 < 1r.
(Here use was made of the convergence of the last series and of Abel's con-
tinuity .theorem; see Titchmarsh 1, pp. 9-10.) Replacing i by - i, we obtain

(4.9.15) Qn ( cos un + '/,'0) = 2 3 . 25 ·...

4 ·" 2n ~~ --i(n+2v+l)6 0 < 8 < 7r.
(2n + 1) ~ v e '
From here (4.9.5) follows again. because of (4.62.8), a = {3 = 0. By use of
(4.62.9), a = {3 = 0, we find (Heine 3, vol. 1, p. 130)

2 · 4 · · · 2n ~
(4.9.16) Qn(cos 8) = 2:
3 5
... ( n + ) f;:of• cos (n + 2v + 1)8,
.
2 1
0 < 8 < 7r.
Another variation of these considerations (see Hobson 1, pp. 57-58) is to
introduce x = !(w + w-1) andy = wn+ 1z into (4.2.1). Then z, as a function of
w2 , satisfies a hypergeometric differential equation from which (4.9.13) (therefore
also (4.9.5)) again follows.
By means of (4.9.9) we see that the sequence lf.) is completely monotonic
(cf. §6.5 (4)).
(3) Another trigonometric representation of Legendre polynomials. (Stieltjes
7, 8.) Starting again from (4.8;17), we write
i(6-r/2)}-l
(1 - te 2i6)-l = ei<" 14- 612 >(2 sin 8)-l{ 1 - (1 - t) e . .
2sm8
Cvnsequently, for 2 sin 8 >l
Pn(cosO) = ~3{ei<n+l)6ei<" 14 -fl 12 >(2sin8)-l
7r
f. g. ei•<~-" > fl t(1- t)•-l dt},or
•-0 (2 sm e)• }o
12

(4.9.17) Pn(cos 8) ~)8) 5; +

=!7r 3 ' ~.' '4' ·(~·. n 2~ 1) '£h. cos{ (n + v ~ Slll !)7r/2) '
v=O (} v

where
[ 4.!) J TRIGOKOMETRIC REPRESE.NTATIOKS 93

(4.9.I8) ~=I; hv = hvn =g. t ! ···

(v - 12 )
v=I,2,3,···.
(n+V(n+V · · · (n+v+i)'
Expansion (4.9.I7) is convergent if 1rj6 < 8 < 57r/6. Its importance will be
more fully realized in discussing the asymptotic behavior of Pn(cos 8) for large
values of n ( §8.5). Concerning the connection of this representation with that
of Heine [(4.9.5)], see Stieltjes 8, p. 244.
(4) Ultraspherical polynomials. From (4.7.2I) we obtain as in (I):
p~>-->(cos 8) = 2aoan cos nO+ 2a1an-1 cos (n- 2)8 + · · ·
(4.9.I9) 2a(n-l)/2a(n+ll/2 COS 8, nodd,
+
1an/2,
2
n even,
where
(4.9.20)
The:t:efore,
(4.9.2I) an- _(n +A-
n
I)' n = 0, I, 2, · ...
Here the cases A - 0, - I, - 2, · .. are to be excluded. In particular, if
A > 0, the coefficients an are positive, so that p~>--> (cos 8) as a trigonometric
cosine polynomial again has non-negative coefficients.
Expansion (4.9.5) can likewise be extended. We have (Szego 19, pp. 508-
509) for A > 0, A r= I, 2, 3, · .. , 0 < 8 < 1r,

. )2>---1 <>-->(
(sm ) - 22-n r(n + 2A) ~~<>--> . ( + 2 + I)O
8 P n cos 8 - r (A) r (n + A + I) ~ • sm n v ,
(4.9.22) JO
;<>--> -_ I·, !<>--> _ !<>-->
v - vn

(n + I)(n + 2) · · · (n + v)
I . 2 ... v (n +A+ I)(n +A+ 2) · · · (n +A+ v)'
v = I, 2, 3, ....
The generalization of the third proof given in (2) is particularly simple. The
special case n = 0 is

(4.9.23) (sin 8) 2>-.-1 = 21r -ir(A + t) ta (I - A~~A ~A}~· ~t- A) sin (2v + I)O.

It can be verified in various ways. (Cf. Whittaker-Watson 1, p. 263, problem

40; multiply this formula by cos x, and substitute x = 1r/2 - 8, s = 2A - 2.)
Another extension of (4.9.5) is the following:

>--> 2 ~ r(n + v + 2A) )

P~ (cosO)= f(A) ~a"r(n + v +A+ I) cos((n + 2v + 2A 8- A7rl,
(4.9.24)
0 < A< I, 0 < 8 < 1r.
It follows readily from (4.82.3) by means of the argument used in the second
proof given in (2).
The extension of (4.9.I7) (Szeg617, pp. 57-60) is
 94 JACOBI POLYNOMIALS [IV]

p~>--)(cosO) = (2/7r) sin A7r r(n + 2 A) f r(v + A)f(v.....: A+ I)

(4. 9 .25 ) f(A) •=o v! r(n + v +A+ I)
. cos( (n+v+A)O- (v+Ah/2l
(2 sin 8)•+>-- ' O < A < I; 7r /6 < 8 < S1r /6.
It follows from (4.82.3) by the same argument as used in (3).
4.10. Further properties of Jacobi polynomials
(I) Rodrigues' formula. A generalization of Rodrigues' formula (4.3.I) is the
following:
(I - x)a(l x)flp~a.f3)(x) +
(4.IO.I)
=
(-I)m
_
(d)m
( (I-x)m+a(l+x)m+flp~~ta,m+fl)(x)l.
2mn(n-I) · · · (n-m+ I) dx
Here n ~ m. The proof is similar to that given in 4.3 (I).
(2) Int~gral representations for ultraspherical polynomials. (a) Combining
(4.1.5) and (4.3.2) we find
P~~,a)(x) = P~~.a)(I) · x 2•F(-v, -v +!;a+ I; I - x- 2),
(4.10.2)
P~~+i(x) = P~~:fi(I) · x 2•+ 1F(-v, -v- !; a+ I; I - x- 2)
so that, in view of (4.7.I) and (4.7.3),
p~>--)(x) = p~>--)(1) · xnF(-n/2, -n/2 +!;A+!; I - x- 2)

= 21-2>-- r(n + 2A) l~l (-I)k(n) r(k + !) xn-2k(I- x2)k

(4.10.3) r(A) n! k=o 2k r(A + k + t)
1
_
- -2>-. _f(n + 2A)1 ... (
2
[r (A) j2 n! X
+ ( 2 _ I)!
X COS cp
In sm
· 2>-.-1 cp d cp.
0

We used (1.7.3), n = 2, and (1.7.5). This generalization of (4.8.IO) holds pro-

vided that A > 0. See Gegenbauer 1, Seidel-Szasz 1.
(b) Another remarkable integral representation valid for non-negative integral
values of A - ~ is the following: ·
(x2 _ I) >--/2-1/4p~>--) (x)
(4.I0.4) =2>-.-!r(A+!) r(n+2A) +( 2-I)l .l ( . . (
In+>--!
r(2A) r(n +A+!) x coscp
71" Jo x
· cos (A - t )cp dq;.
For the sake of simplicity we assume here that x > I, (x - I )l > 0.
2

For the proof we use (4.4.6) with the circle (4.8.8) as contour of integration.
Since

we have, a = !,
f
A -
2
(x2 _ I)ap~a,a)(x) = -.
I (I_ t_- _I)n (t2 _ I)a _dt_
27rt 2 t - X t - X

which yields easily (4.I0.4).

[4.10] FURTHER PROPERTIES OF JACOBI POLYNOMIALS 95

(3) Generating functions. (a) A slight modification of the argument in

4.4 (3) (proving the formula ( 4.4.5) for the generating function F (x, w)) is the
following. Let -1 < w < 1; we write (4.4.7) as follows:
(4.10.5) x = t + !w(1 - t 2)
so that x and t describe the interval l-1, + 1] at the same time. Henee, using
the formula (4.4.7) and the two first formulas of (·1.-1.8), we find
(1 - x)a(l + x)fJF(x, w) dx = (1 - t)a(l + t)fJ dt
so that for any non-negative integer n

f _
+l
1
(1 - x)a(l + x)fJF(x, w)xn dx = _
!+1 1
(1 - t)a(l + t)fJ[t + !w(1 - t 2)]n dt,

i.e., a polynomial of degree n in w. This shO\vs that the coefficients of F(x, w)

must be, apart from constant factors, the Jacobi polynomials. These factors
can be determined by writing x = 1.
(h) The following generating function holds for ultraspherical polynomials:
p~~) (x) wn - r (2;\) (~)
00 00
"
n~ p~~l(l) n~O
•

(4.10.6) n! - r(n 2;\) Pn (x)w +

= 2~-tr(;\ + !) exw[(1 - x 2 )iwj!-~J~-![(1 - x 2)lw].
It is different from (4.7.16) and (4.7.23). Setting ;\ = ! we obtain for the
Legendre polynomials:
p (x)
(4.10.7)
oo
2:
_n_,_ wn = exwJo[(1 - x2)iwJ.
n=O n.
For the proof of (4.1 0.6) we use ( 4.10.3); we obtain for the left-hand side
of (4.10.6):
2 1 - 2 ~ r (2;\) [""
exp[wlx + (x 2 - 1)tcoscpj]· sin 2 ~- 1 cpdcp
[f(;\)]2 0

- 21-2~f(2;\) xw oo [ (x2 - 1 )!wpm1"" -2m . n-1

- [f(;\))2 e m~o (2m)! cos cp sm cp dcp
0

which is eu.sy to identify with the right-hand expression in (4.10.6).

Concerning further formal properties of the Legendre and ultraspherical
polynomials, see Bateman Manuscript Project, vol. 1, Chapter 3 n.nd vol. 2.
Chapter 10. Cf. also Problems and Exercises, 61-66, 69-71, 84.
( 4) The following remarkable identity, involving ultraspherical polynomials,
is due to Feldheim 5, p. 278:
r(2;\) p(>-)( ) 2r(;\+!)
= -----=----
f(n +
2;\)
n COS(}
r(JL + t) r(;\ - JL)
ll"/2
r (2)
JL . . . . . . ..
(4.10.8) · sm 2"cp cos 2 >--~- 1 !1'( 1 - sm 20 cos 2~;?)" 12
r(n + 2JL) !uo
·P,.(!') (
( 1 - sm
·2n
. l < <
cos 8 2)1;2 ) d 1P.A>JL>-2,;\~0.JL~0.0=0=7r·
u cos II'
 96 JACOBI POLYNOMIALS [IV]

It is an easy consequence of the generating function (4.10.6). Indeed, multi-

plyingbothsidesof(4.10.8) byw"andextendingthesummationovern = 0, 1,2, · ·.
we obtain
2>.-l2f(A + ~) ew"osO(W sinO) I Z->. J>.-I;2(W sinO)
1

__ _...:__ __:_~- . l"l2 sin'"~~'

2f(;\ + ~.) 2 cos2>--2"-IIP. 2"-1· 2r(J.L + & )e""·oso
r(J.L + ~ ) r(;\ - J.L)

or

(4.10.9)

This identity is due to Sonine (Watson 3, p. 373, (1)).

(5) Positivity of certain sums. The first result of this kind is due to L. Fejer
4, p. 83:

(4.10.10) P0(x) + P 1(x) + P2(x) + · · · + P,(x) ~ 0 for - 1 ~ x ~ 1.

This fact is important in the study of summability of the Laplace series. It
follows from (4.8.7) by observing the identity

£ sin(2~+l)O = (sin(~+l)0\ 2

v=O Sm0 SinO /

Generalizations of (4.10.10) are due to Fejer 12, Feldheim 5, Szego 25.

( 6) The formulas ( 4.8.6) and ( 4.8. 7) of Dirichlet-Mehler can be extended to

(1 -X)
a+"
p<a+~t.P-"\x)
p~a+~t.P-")(1)
n =
f(a + J.l + 1)
f(a + 1) f(J.L)
i x
1

(
1-
y)
p<a,P)(y)
p~a,p)(l)
a n
(y
-X
)
"-1d
y,
(4.10.11)
a> - 1, J.l > 0, - 1 <X< 1,
and
p~a-~t,P+~<>(x) f({J + J.l + 1) fx p<a,p)(y)
(1 + x)P+" p~P+~t,a ")(1) ='f({J + 1) f(J.L) -1 (1 + y) p p~P,a)(l) (X- y) "-1dy,
(4.10.12)
{J >- 1, J.l > 0, - 1 <X< 1.
Feldheim's integral (4.10.8) follows by using the quadratic transformations
( 4.1.5) and the integral
(1 _X) a+" p~a+~<.P>(x)
( 1 +X) n+a+1 p~a+~t.Pl(l)

( 4.10.13) = 2"f(a + J.l + 1) { 1 ( 1 _ y)a(y _ x) "-1 p~a,P>(y) d

r(a + 1) f(J,L) J X (1 + y)n+a+~t+l p~a,{J)(l) y,

a> - 1, J.L > 0, - 1 < x < 1. These three integrals are all special cases of
known integrals connecting hypergeometric functions which were found by
Bateman 2. See Askey-Fitch 2 for other useful integrals. Among the conse-
quences of these integrals is the following theorem on positive sums.
 [4.10] FURTHER PROPERTIES OF JACOBI POLYNOMIALS 97

THEOREM 4.10.1. If 0 ~ J.L ~ v and {3 >- 1 and if

( 4.10.14) to
n p~a,p)(X)

ak p~P,a)( 1) ~ 0, -1 ~X~ 1,

then
p~a-~t,P+v)(y)
( 4.10.15)
ton
ak P1P+v,a-"\l) ~ 0, -1~y~l.

Feldheim's integral ( 4.10.8) can be used to prove a related theorem.

THEOREM 4.10.2. If v > 0, a >- 1 and if
n p~a,a)(X)
( 4.10.16) "ak
~ p<a,a)( 1) >
= 0' - 1 ~ x ~ 1,
k=O k

then
n p~a+v,a+v)(y)
(4.10.17) t; ak p~a+v,a+v)( 1) ~ 0, -1~y~l.

As Feldheim 5 pointed out, (4.10.10) implies

( 4.10.18) - 1 <X~ 1, a> 0.

Askey-Gasper 4 and Askey-Steinig 2 have obtained generalizations of ( 4.10.18).

They proved

( 4.10.19) -1 <X< 1,
for
(i) a+ {3 ~ -2, {3 ~ 0 (when a=.- 2, {3 =,0, assume n ~ 1),
+
(ii) - {3 ~a~ {3 1 (a= t, {3 =.- t omitted),.
+
( iii) - {3 1 ~ a ~ 3/2,
(iv) -{3+2~a~{3+3.

The case a= t. {3 = - t of the sum (4.10.19) is Fejer's sum (6.4.3), which

is nonnegative, while the case a = 3/2, {3 = - t is equivalent to

:!:_ £
sin(k + 1)8 <
0 0 < 8 < 7r.
de k=o(k+l)sin(0/2) '
It is also equivalent to
n 2 p J~\ X) 1 . n p J~> (X)
L: (k + o p<2>o)
k=O 2k
> 4- L: p<2>o).
2k k=O

and, since the right-hand side of this inequality is the even part of ( 4.10.18)
when a = 3/2, it is nonnegative. The positivity of the left-hand side is equivalent
to the positivity of K~ 2>(x) in (15.5.1).
(7) Jacobi polynomials satisfy the addition forrnnla
98 .JACOBI POLYI\'0;\1IALS [IV]

p~a,P)( 21 cos 81 cos 82 +rei"" sin 81sin 82l 2 - 1)

n k
=.L: L c~~(~(sin81sin8 2 )k+m(cos8 1 cos82 )k-m
k=O m=O
( 4.10.20)
· P~a_tk+m,f3+k-m)( COS 281) P ~~Jk+m,/3+k-m)( COS 28z)

. p~a-P-1.P+k-m\2r2- l)rk-m {3 + k- m Cf-m(COS!p),

{3
where
c<a p) - (k +m +a) r(n +k +a +f3+1) r(k +a) r({J+l) r(n +f3+1) r(n -k +0
n,k,m -, r(n +a +f3+1) r(n +m +a+ 1) r(k +f3+ 1) r(n- m +f3 + 1)

and the limit relation

. f3+n CP( ) {2cosn!p, n =,1,2, ... ,

11m-- n COS!p =
p-o {3 · 1, n =,0,

is used when {3 =.0.

When r =.1 and a =,{3 this formula is the addition th~orem of Gegenbauer
I for ultraspherical polynomials. It was discovered by Sapiro I in the case
{3 =,0 and independently by Koornwinder I in the general case. For other
proofs see Koornwinder 2, 3, 4.
Among the special results contained in ( 4.10.20) are
p~a,P)( COS 281) p~a,p)( COS 282)
p~a,/3>( 1) p~a,/3>( 1)

( 4.10.21)

a> {3 > - t. where

· !p )2Pd !p dr,
dma,P ( !p, r ) = A a,P ( 1 - r 2)a-P-1r 2P+l( sm
and

and also

( 4.10.22)

When a__, {3, ( 4.10.22) has ( 4.10.3) as a limit. In addition to the papers listed
above, see Gasper 3, 4 and Askey IO.
Another useful formula was found by Bateman I:

p<a.P) ( 1 + xy)
n X +Y (X +Y) n n p<a.P) X p~a,P>(y)
( 4.10.23) p~a,P)( 1) -2-
-
- 6 k,n k ( ) p~a,p)( 1)
C

where the ck,n's are defined by

[4.10] FURTHER PROPERTIES OF JACOBI POLYNOMIALS 99

+
1 x)n n
= LCk,np~a,P>(x).
( 4.10.24) ( --
2 k=O
The ck,n can be explicitly computed by use of Rodrigues' formula and
orthogonality and they were given in Bateman I. However, in many applica-
tions only the positivity of the ck,n is needed (see e.g. Horton I). It can easily
be proved by induction. Bateman 3 also discovered the inverse to ( 4.10.23):

( 4.10.25)

where the bk n' s are defined by

(4.10.26) p<a.P>(x)
n = ~ bk,n ( 1 + x)k
.~
k=O 2
This has been used by Koornwinder 3. Again the specific form of the bk,n's
was not needed.
(8) Gegenbauer 6 generalized (4.9.19) to obtain
(n/2)
( 4.10.27) P~">(x) =."'L ak,nP~A!.2k(x)
k=O
where
r( A) ( n - 2k +A) r( k + JL - A) r( n - k + JL)
( 4.10.28) a -----~~~~--~~--~----~~--
k,n - r( JL) k ! r( JL - A) r( n - k +A + 1)

Proofs are given in H ua I and Askey 2. This can be inverted to give

( 4.10.29) (1- x )"-

2 112
P~">(x) = "'Ldk,nP~~2k(x)(l -x 2
k=O
r'- 112
, JL >(A -1)/2,

where
( 4.10.30)
d _ r(A)2v.- 2"(n +2k +A) r(n +2k + 1) r(n +2JL) r(n +k +A) r(k +A -JL)
k,n- r(A-JL)r(JL)r(n+l)r(k+l)r(n+k+JL+l)r(n+2k+2A) ·
See Askey I. When A = ), (4.10.29) reduces to ( 4.9.22). Generalizations of
these formulas to Jacobi polynomials as well as generalizations of Problem
84 are given in Askey 4, Askey-Gasper I, 2, Gasper I, 2. In the general case,
the coefficients are much more complicated and one needs to obtain asymptotic
formulas and positivity when it holds. For an application of ( 4.10.27) and
Problem 84 in which only the positivity is used see Askey-Wainger 3. The
linearization result in Problem 84 can be used to define Toeplitz operators
and much of the classical theory of Toeplitz operators and the newer work
on finite sections of such operators can be extended to these more general
operators. See Hirschman 2 and Davis-Hirschman I and further references
given in these papers.
 CHAPTER V

LAGUERRE AND HERMITE POLYNOMIALS

Many of the properties of the polynomials with which we shall deal in this
chapter are very similar, and more or less analogous, to the properties of Jacobi
polynomials. For this reason we shall be brief and omit details, unless essen-
tial differences in statement or proof make the contrary necessary. Here, as
in the case of Jacobi polynomials, the treatment of some special problems
(zeros, extrema, and so on) is reserved for later chapters.

5.1. Elementary properties of Laguerre polynomials

(I) We define the Laguerre polynomials 1L~">(x) l, for a > -1, by the
following conditions of orthogonality and normalization:

(5.1.1)
10() e-xx"L~">(x)L~">(x)dx = r(a + 1) (n ~ a)onm,
n, m = 0, 1, 2, · · · .
In addition, we require that the coefficient of xn in the polynomial L~">(x) of
degree n have the sign ( -1) n. (This differs from condition (a) in the definition
of §2.2.) We also write L~ >(x) = Ln(x).
0

Reference is here made to Lagrange 1, Abel 1, p. 284, Tchebichef 3, pp.

506-508, and Laguerre 1, pp. 78-81 (pp. 434-437), who, however, consider
only the case a = 0. Laguerre uses the notation fn(x) = n!Ln( -x). Hilbert-
Courant (1, pp. 79-80) also considers only the case a = 0; the function there
called Ln(x) is the same as n!Ln(x) in our notation. Concerning the general
case L~">(x) see Sonin 1, pp. 41-42.
We have the differential equations
xy" + (a + 1 - x)y' + ny = 0, y = L~">(x),

xz" + (x + l)z' + (n + ~ + 1- :~)z = 0, z = e-:rx" 12 L~">(x),

2
(5.1.2) u" + (n +_(a +_J_)/2 + 1- a _ ~)u = O u= e-xl2x<a+OI2L~">(x),
x 4x2 4 '

v" + ( 4n + 2a + 2- x 2 + i ~2 a) v = 0, v = e-x 212 xa+! L~">(x2 ).

Once again let a > -1. Then an argument analogous to that in §4.2 (2)
shows that a necessary and sufficient condition that
(5.1.3) xy" + (a + 1 - x)y' + 'Xy = 0
100
[ 5.1] ELEMENTARY PROPERTIES 101

have a polynomial solution is that;\ = n. Also, L~">(x) is the only polynomial

solution. The latter statement follows from the relation
(5.1.4) u~(x)u2(x) - u1(x)u;(x) = const.,
which holds for two arbitrary solutions u1(x), u2(x) of the third equation in
(5.1.2). Incidentally, this argument furnishes slightly more: for a > -1, the
polynomials L~">(x) are the only solutions of (5.1.3) which are analytic near
X= 0.
The analogue of Rodrigues' formula is

(5.1.5)

To determine the constant factor we apply Leibniz' formula (which leads to

(5.1.6)) and calculate the highest term of the right-hand member.
Further, we have the explicit representation

(5.1.6) L~">(x) = f (nn +a)

•=O - v
(-xr'
v!
the formula

(5.1.7)

and the expression

(5.1.8) z~a) = ( -1)"

n!
for the coefficient z~a> of xn in L~a> (x). As a generating function we obtain
LJ">(x) + L~">(x)w + · · · + L~">(x)wn + · · ·
(5.1.9)
XW )
= ( 1- w ) -a-1 exp (
- .
1-w
The following recurrence formula holds:
nL~">(x) = ( -x + 2n + a - 1)£~~\(x)
(5.1.10) n = 2, 3, 4, · · ·,

For the "kernel polynomial" K~a> (x, y) we find

r(a + 1)K~">(x, y) = ta {(v :a)}-L~">(x)L~a>(y)

1

(5.1.11)
 102 LAGPERRE AND HERMITE POLYI\Ol\1IALS [v l
The special case y = 0 is particularly important:

(5.1.12) x L" L~"''(x)

•-0
= (n +a +1)L~"''(x)
.
- (n + 1)L~+'1 (x).

(Cf. Theorem 2.5.) Finally, by means of (5.1.6), or (5.1.9), we readily obtain

(5.1.13)
L" L~"''(x)
•-0
= L~a+ll(x),

L~"''(x) = L~"'+ 1 '(x) - L~':.1 1 '(x),

(5.1.14) d:L~"''(x) = -L~':.1 1 ,.(x) = x-1 lnL~"''(x) - (n + a)L~~-\(x) ).

(2) THEOREM 5.1. Let J"' have the same meaning as in §1.71. Then

~ {(n: a)}- L~"''(x)L~"''(y)w"

1

(5.L15) !
= r(a + 1)(1- w)- exp{-cx + y) 1
~}(-xyw)-"' 12 J..{ 2 (-xyw) },
1-w 1-w
and
(5.1.16)

See Sonin 1, p. 41, Wigert 1, Hille 2, Hardy 1, Kogbetliantz 12, Watson 4.

The first formula is a generalization of (5.1.9) (y = 0). The second formula
is obtained from the first one by replacing w by -y- 1w, y-+ oo.
Direct calculation leads readily to (5.1.16) on account of (5.1.6). The
formula (5.1.15) follows from (5.1.16) when we introduce for L~"''(y) the
integral expression which results from (5.4.1), and then integrate term-by-term.
Finally an integral formula involving Bessel functions (Watson 3, p. 395, (1))
must be used.

5.2. Generalization
By means of (3.1.6) the definition of Laguerre polynomials can be extended
to arbitrary complex values of a. No reduction in the degree ever occurs
(see (5.1.8)). For n ~ 1 we have L~"''(O) = 0 if and only if a= -k, k integral,
1 ;;;? k ~ n. In this case x = 0 is a zero of precise order k, and from (5.1.6)

L~-k'(x) = ( -x)dn-
n!
k)! I:' (
•=O
n
n - k- v
) ( -x)"
v!
(5.2.1)
_ (- )k(n- k)!L<kl ()
- X I n-k X •
n.
Formula (5.1.16) remains true for arbitrary real a. From here (5.2.1) follows
agam.
[ 5.4 l INTEGRAL REPRESENTATIONS 103

5.3. Confluent hypergeometric series; relation between Jacobi and Laguerre

polynomials; second solution
(1) In the notation of Pochhammer-Barnes, the confluent hypergeometric
senes IS

(5.3.1) ( . . ) _ 1 + ~ a(a + 1) · · · (a + v - 1) x•
1F1a,-y,x- .L...J -.
· •-1 -y(-y + 1) · · · (-y + v - 1) v!

This is obtained from the ordinary hypergeometric series [(4.21.3)] by the

limiting process
(5.3.2) lim F(a, ~; 'Y; ~- 1 x).
/J-+00

We have

(5.3.3) Ln(a) ( x ) = (n + a) (
n 1F1-n;a+1;x),

and using (4.21.2), we obtain the following important relation between Laguerre
and Jacobi polynomials:
(5.3.4) L~"'(x) = lim p~a,/Jl(l - 2~- 1 x).
/J-+00

This holds uniformly in every closed part of the complex x-plane. Concerning
further properties of the confluent hypergeometric functions see Whittaker-
Watson 1, Chapter XVI. Compare equation (B) in Whittaker-Watson 1,
p. 337, with our third equation in (5.1.2).
(2) From (4.23.1), by a limiting process similar to that used in (5.3.4), we
obtain as a second solution of the first equation (5.1.2)
(5.3.5)
For non-integral values of a the functions (5.3.3) and (5.3.5) are evidently
linearly independent. The same is true if a is a ne.gative integer less than -n,
since in this case (5.3.5) is an infinite series. However, if a is an integer not
less than -n, these solutions are identical. (If -n ~ a ~ 0, use (5.2.1). In
case a = g, g ~ 1, it is necessary to multiply (5.3.5) through by a - g before
letting a---+ g.)
The representation (5.3.3) makes possible an extension of the definition of
L~"'(x) to arbitrary values of n.

5.4. Integral representations

THEOREM 5.4. The following representation of Laguerre polynomials in terms
of Bessel functions holds:

(5.4.1) e-xx" 12 L~"'(x) = ~ 1"" e-ttn+at 2 J ,.l2(tx)!)dt, n = 0, 1, 2, ···;a> -1.

n. v

For this representation, the reader may consult E. Le Roy 1, pp. 379-384,

!
···------·.--.I
 104 LAGUERRE AN"D HERMITE POLYI'\OMIALS [V)

Erdelyi 1. The same representation is valid if a ;;;;; - 1, provided n a > -1. +

From this, (5.2.1) follows again.
Several proofs of this formula may be given. The special case n = 0, that is,

(5.4.2) e-xxo: 12 = 11XJ e-t to: 12 J o: (2(tx )! l dt,

can be obtained by expanding J o:(z) as in (1.71.1) and integrating term-by-term.

(The formula is due to Sonin; cf. Watson 3, p. 394, (4).) The general formula
then follows from calculation of the generating function of both sides of (5.4.1);
here (5.1.9) and (5.4.2) must be used.

FIG. 6

Another proof, from a more general point of view, can be given as follows.
We shall try to satisfy the second equation (5.1.2) by an integral of the form

(5.4.3) z = z(x) = J e-ttn+o:/ 2 Jo:l2(tx)!j dt

with a proper path of integration. Substituting this expression in the left-hand

member of the equation mentioned, we obtain

(5.4.4)
Je- tn+o:/ [tJ~l2(tx)!j +
1 2 (tx- 1 )!(x + !)J~l2(tx)!J

+ ( n + ~ + 1 - :~) J o: (2(tx)! l Jdt.

In view of (1.71.3) the expression in the square brackets becomes
(5.4.5)

so that the entire expression will be

(5.4.6) J~i[e-tt+o:/2+lJo:l2(tx)!j]dt.
Hence (5.4.3) is a solution of (5.1.2) .provided that
[5.5] HERMITE POLYNOMIALS 105

(a) the path of integration is a closed contour, and so chosen that

e-ttn+o:/2+lJ al2(tx)!} resumes its initial value, or
(b) the path of integration is an arc, and the expression in (a) vanishes at its
end-points.
For the interval 0 ~ t < + oo, condition (b) is satisfied provided n +a+ 1 > 0.
Assuming first that a > -1, we notice that the function x-o:12z is analytic near
x = 0, so that (see the remark concerning (5.1.3)) z = const. e-xxo: 12 L~o:'(x).
The constant factor can be determined by comparing the "lowest terms," that is,
the coefficients of xo:12 . The restriction a > -1 can then be removed by means
of analytic continuation.
Another remarkable representation is obtained by choosing the contour as
in the figure. Condition (b) is again satisfied, and the same argument as before
yields z = const. e-xxo: 12L~o:'(x). The normalization of the integrand requires
the determination of to: at a certain point. We agree to take arg t = 0 on the
rectilinear part of the contour with St < 0 (in its ·limiting position). Then
the "lowest term" becomes, if a > -1 and a :;t. 0, 1, 2, ... ,

r(a
X
o:/2
+ 1)
f e-ttn+o: dt =
r(a
X
o:/2
+ 1)
(1 - e_ 2,.io:) 100 e- t+o: dt
0
1

- 2 .
- Slll 1rae
ill'(!-o:) r(n +a
r( a+ 1) X ,
+
1) o:/2

so that on account of (5.1.7), we have

. )-1 ill'(o:-!)
(5.4.7) e-xxo: 12 L~o:'(x) = (2 sm 1ra e
1(0+) e-ttn+o:/2J o:l2(tx)!} dt.
n.1 +oo
This formula can be extended to arbitrary non-integral values of a.
Further integral representations can be derived from (5.1.5) and (5.1.9) in a
manner analogous to that used in the case of Jacobi polynomials (cf. (4.4.6),
(4.4.9)). We have for instance, x :;t. 0,

e-xxo:L~o:l(x)
2~i Je-tt+o:(t- x)-n- dt,
1
(5.4.8) =

where the contour encloses t = x, but not t = 0.

5.5. Hermite polynomials
(1) These are defined by the conditions

(5.5.1) L:oo e-x 2

Hn.(x)Hm(x) dx = 7r!2nn! Onm, n, m = 0, 1, 2, ....

The coefficient of xn in the nth polynomial is positive.

See the bibliography in Hille 1. Our notation agrees with that of Hille.
See also Hilbert-Courant 1, pp. 77-79 where the same notation is used. In
P6lya-Szego 1 (vol. 2, pp. 94, 294, 295, problem 100), H n(x) is written for
( -1) n(2n 12n!)- 1H n(2-!x) in terms of our notation.
 106 LAGUERRE AND HERMITE POLYNOMIALS [V]

(2) The derivation of the following properties of Hermite polynomials presents

no difficulty:

(5.5.2)
Jy" - 2xy' + 2ny = 0, y = Hn(x),
lz" + (2"-n + 1 - x 2)z = 0, z = e-x 212Hn(x),
(5.5.3) e-x H n(X) =
2
( -1 t (!!:_)n
dx
e-x
'
2

(5.5.4) Hn(x) = [fl (-1)" (2xt- 2"

n! .-o v! (n - 2v) ! '

(5.5.5) H2m(O) = ( -1)m ( :?!,

2
H~m+l(O) = (-1)"' ~~: :;!!,
(5.5.6) lim x-n H n(x) = 2n,

Ho(x) + H1(x) w + H2(x) w2 + ... + Hn(x) wn + ...

(5.5.7) 1! 2! n!
. 2
= exp (2xw - w ),
Hn(x) = 2xHn-l(x) - 2(n - 1)Hn-2(x),
(5.5.8)
n = 2, 3, 4, · · · ; Ho(x) = 1, H1(x) = 2x,

(5.5.9) ·t (2"v!)-1H.(x)H.(y) = (2n+ln!)-1 Hn+l(x)Hn(y) -

v-o X -
Hn(x)Hn+l(y).
y
We notice the following "individual" properties:
(5.5.10) H:(x) = 2nH n-l(x),
From (5.5.7) we obtain

(5.5.11) t•-o (n)H.(x)Hn_.(y)

v
= 2n 12 Hnl2-!(x + y)),
and, by Cauchy's formula,

(5.5.1 2) H,.(x)
- -- -
n.1
_ --;
1
27r~
f w-n-1 exp (2 xw - w2) dw,

where the contour encloses the origin.

5.6. Relation of Hermite polynomials to those of Laguerre

(1) Hermite polynomials can be entirely reduced to Laguerre polynomials
with the parameters a = ±!, for we have
(5.6.1) H2m(x) = ( -1)"'2 2
mm!L~-!l(x\ H 2m+1 (x) = ( -1)m2 2 m+ 1 m!xL~i'(x 2 ).
These formulas are in some respects the analogues of (4.1.5). Their proofs
[ 5.6) RELATION OF HERMITE AND LAGUERHE POLYNOMIALS 107

are similar to those given there. Note the fourth equation in (5.1.2) (a = ±t),
and the second equation in (5.5.2).
Combining (5.6.1) with (5.3.4), we obtain a representation of Hermite poly-
nomials as limits of Jacobi, and in consequence of (4.1.5), of ultraspherical
polynomials. This can be ascertained from (4.7.23) and (5.5.7). In fact, we
have

(5.6.2) exp (2xw - w 2) = lim (1 - 2: w + w 2)->. ,

>--~ A A
so that

(5.6.3)

From (5.6.1) the explicit representation (5.5.4) follows readily when we use
(5.1.6). From Theorem 5.1 analogous expansions for H,.(x) can be. derived
(see Watson 5). The generating function (5.5.7) follows ·from (5.1.16), while
the generating function arising from (5.1.9) for a = ±tis of a different nature
(see Problem 24).
By using (1.71.2) and (5.4.1), we obtain the integral representations

e-"' Hn(x) = (-1)[n 121 2n+1 1r-! 1~ e- 12 tn cos (2xt)dt,

2
n even,
(5.6.4)
2
e-"' H n(x) = ( -1)[n121 2n+1 1r-! 1~ e-12 tn sin (2xt) dt, n odd.

(2) Conversely, Laguerre polynomials can, to a certain extent, be reduced to

Hermite polynomials. We have (Uspensky 1, p. 604, (14))

(
5.6.5 )
(a) (
Ln X
) 1t 1r-! r (n + a + 1)
= r(a + t)
( -
(2n)!
1-1
+
1
(
1- t
2) a-t
H2n
(
X
t )
t dt, a> -
1
2·

This can be readily shown by using the explicit formulas (5.1.6), (5.5.4), and
(1.7.5).
(3) We conclude these formal considerations with the following remark con-
cerning the identities (4.21.7), (5.1.14), and (5.5.10) on Jacobi, Laguerre, and
Hermite polynomials, respectively. If the orthogonal polynomials IPn(x) l
associated with a distribution da(x) have the property that I p:(x) J is, save for
constant factors, a system of the same kind (that is, associated with a certain
distribution d[j(x)), then I Pn(x) l is (save for trivial linear transformations)
one of the three special systems mentioned before (classical polynomials).
(W. Hahn 3, Krall 1.) A similar statement holds if we replace I p:(x) J by
lP~k'(x)J (Krall2, W. Hahn 4).
Another problem of a similar nature has been considered by Bochner (1).
He determines all sets of polynomials IPn(x) l, where Pn(x) is of precise degree n,
108 LAGUERRE AND HERMITE POLYI\01\iiALS (V l

satisfying a differential equation of the form

(5.6.6) fo(x)y" + ft(x)y' + [h(x) + ;\.]y = 0, Y = Pn(x), A = An•
Bochner obtains, in addition to the classical polynomials, certain polynomials
related to J n+ 1(x), n integral, as possible solutions, as well as polynomials of
the trivial type axn +
bxm, where a and b are constants.

5.7. Closure
Here we prove the analogue of Theorem 3.1.5 for Laguerre and Hermite
polynomials. The main difficulty of these cases is due to the fact that the
orthogonality interval is infinite. Using the cu~tomary notation (§1.1), we
have the following theorem:
THEOREM 5.7.1. The system
(5.7.1) e-z/2 X a/2 X n , a > -1, n = 0, 1, 2, ·. · · ,
2
is closed in L (0, + oo); the system
-z2f2 n
(5.7.2) e x , n = 0, 1, 2, · · ·,
is closed in L 2 ( - oo, + oo ).
This statement is equivalent to the closure of the systems I e-"' 12 x,. 12L~a> (x)}
212
and le-"' H n(x)), respectively. Theorem 5.7.1 remains true, of course, if we
2
replace e-"' 12 bye-"' and e-"' 212 by e-"' , 'respectively. The idea of the following
proof is due to J. von Neumann (see Hilbert-Courant 1, pp. 81-82).
(1) We start with the remark that for a > -1 the system
1)a/2
(5.7.3) (logy yn, n = 0, 1, 2, · · ·,

2
is closed in L (0, 1). This is a consequence of Theorem 3.1.5, p = 2, since
(log (1/y)t is integrable in [0, 1].
Now let e-"'12 x,. 12f(x) belong to L 2 (0, +
oo ). Then (log (1/y)),.12j(log (1/y))
belongs to I}(O, 1), and it can be approximated in mean by functions of the
form (log (1jy)),. 12 p(y), where p(y) is a polynomial. Thus, corresponding to
every E > 0 a polynomial p(y) can be determined such that

(5.7 .4) 1"" e-"' xa lf(x) - p(e-z)} 2 dx < E.

Hence all that remains to be shown is that if m is a non-negative integer, there

exists for every o > 0 a polynomial p(x) such that

(5.7.5) 1"" e-"'x,.le-m"' - p(x) }2 dx < o.

(2) For this purpose we use (5.1.9), writing
f 5.7 I CLOSURE 109

m w
(5.7.6) w=-- ·m=--
m + 1' 1- w'
and choosing
N

(5.7.7) p(x) = (1 - w)"+1 L L~"'(x)wn.

n-O

Then

(5.7.8) 1
00
e-x x" le-mx - p(x) ) 2 dx = (1 - w) 2"+2 1 0
00
e-x x"{ £ L~"'(x)wn} dx,
n=N+l
2

which, in _view of (5.1.1), is equal to

(5.7.9) (1 - w)2a+2r(a + 1) f (n +n a)w2n.

n=N+l

Term-by-term integration is permitted here since the series

t
n.,n'=N+l
1
0
00

e-xxa I L~"'(x) II L~~'(x) I dx·wn+n'

is convergent. The expression (5.7 .9) becomes arbitrarily small when N is

sufficiently large, and this establishes the statement.
212
(3) Let e-x f(x) belong to L 2( - oo, +
oo ). Then both functions

-y/2 -d(y!) ± f(- y!)

(5.7 .10) e Y
2
belong to L 2 (0, +
oo ). Therefore, by the preceding result (first taking a = - t,
then a = +!), for every E > 0 there exist polynomials Pl(y), P2(Y) which satisfy
the inequalities

100 {e-y/2y-d(y!) +/(-yl)- e-y/2Y-1Pl(y)r dy < E,

(5.7.11)
)}2 (t y <
or
1 00 { e-y/2 y -J.f(yt)
--- - f(-
2
yl)
------ - e-y/2 y i P2 ( y E,

(5.7.12)

so that

(5.7.13)
 110 LAGUERRE AND HERMITE POLYNOMIALS (v J
For another proof based on the theory of integral equations, see Weyl 1, in
partieular pp. 58-61, 64. See also Hamburger 1, pp. 200-205.
(4) The argument used in (2) leads to the following result:
THEOREM 5.7.2. The system
(5.7.14)
a > -1, n = 0, 1, 2, . ·. ,
is closed in L(O, + oo); the system
(5.7.15) e
-z2 n
x , n = 0, 1, 2, . ·. ,
is closed in L(- oo, + oo ).
We use again Theorem 3.1.5, p = 1. By means of (5.1.1) and Schwarz's
inequality we find

(5.7.16) 1oo e-"'xo: I L~o:'(x) I dx = O(no: 12 ).

For later purposes (§9.5 (1)) we note the following theorem:

THEOREM 5.7.3. The functions
(5.7.17) fn(x) = cf>(x)xn, n = 0, 1, 2, .. ·,
where

(5.7.18) cf>(x) = {" x,

-z (j
e x,
0 <X<
X~
1,
1,
a > -1, {3 arbitrary and real, form ~ closed system in L(O, + oo ).
We apply Theorem 3.1.5 with da(y) = y- 1cf> (log (1/y)) dy, 0 < y < 1, and
p = 1. Furthermore, we need a bound for

1oo cp(x)jL~o:>(x)ldx = 0(1) £ e-:r.xo:IL~o:>(x)ldx+ J""e-"'x(jiL~o:>(x)jdx.

1

The first integral in the right-hand member is O(no: 12 ); for the second integral,

r
Schwarz's inequality yields the bound

{f"" e-"' X
2
(j-o: dx Y{i"" e-"' xo: [L~o:)(x)f dx = O(no: 12 ).

Concerning further formal properties of t.he Laguerre and Hermite poly-

nomials, see Bateman Manuscript Project, vol. 2, Chapt.er 10, pp. 188-19G.
Cf. also Problems and Exercises, 67, 68, 72-80.
For problems related to Laguerre polynomials and applications of Laguerre
polynomials see Szego 26, Askey-Gasper 3, Askey 9, Peetre 1, and Roosenraad
1. For Hermite polynomials see de Bruijn 1.
 CHAPTER VI

ZEROS OF ORTHOGONAL POLYNOMIALS

In §3.3 it was proved that the zeros of orthogonal polynomials are all real,
distinct, and lie in the in~erior of the orthogonality interval. \Ve shall now
present a further and more detailed investigation of the location of these zeros.
Starting with certain theorems valid under very general conditions imposed on
the weight function, we proceed to the zeros of the classical polynomials and
point out various methods used in their investigation. A particularly important
tool in the latter connection is Sturm's theorem (§1.82), which in various cases
leads to rather exact information concerning the zeros of polynomials satisfying
certain linear differential equations of the second order.
No claim of completeness is made for the present survey on the zeros. Con-
cerning the literature about zeros of Laguerre and Hermite polynomials, we
refer to \V. Hahn's "Bericht" (2).
The methods of this Chapter are quite elementary. In particular, no sys-
tematic use is made of the asymptotic properties of orthogonal polynomials of
a special and general kind (Chapters VIII and XII), from which important
information concerning zeros can also be derived.

6.1. Density of zeros

(1) THEOREM 6.1.1. Let da(x) be a distribution on the finite segment [a, b],
and let IPn(x) l denote the associated orthonormal set of polynomials. Let [a', b']
be a subinterval of [a, b] such that f~: da(x) > 0. Then if n is sufficiently large,
every polynomial Pn(x) has at least one zero in [a', b'].
For the proof we shall usc the Gauss-Jacobi mechanical quadrature (§3.4).
Let p(x) be an arbitrary 1r,. which is not greatrr than 0 in [a, b], except possibly
in [a', b']. Assuming that the polynomial Pn(x) has no zeros Xv in [a', b'],
and taking 2n - 1 ~ m, we obtain
b n

(6.1.1)
l p(:r) da(x) = L Avp(.rv)
v=l
~ 0.

Hence, when we apply the theorem of Weierstrass (Theorem 1.3.1 ), it follows

that

(6.1.2) ib f(x) da(x) ~ 0,

where f(x) is continuous in [a, b] and not greater than 0 in [a, b], except pos-
sibly in [a', b']. If we deJ6.ne
111
112 ZEROS OF ORTHOGONAL POL YJ\Ol\1IALS [VI)

0 in a ~ x ~ a' and b' ~ x ~ b,

(6.1.3) f(x) =
{ (x ·- a')(b' - x) in .a' ~ x :5; b',
we reach a contradietion.
(2) THEOREM 6.1.~~- Theorem 6.1.1 remains valid for infinite intervals [a, b]
provided the zero of the greatest modulus of Pn(x) is o(n).
29
This remark is due to W. Hahn (1, pp. 215-217). If we write
max [ x. - a' [[ b' - x.[ = Mn, v = 1, 2, · · · , n,
the assumption means that Mn = o(n2 ). We now choose
I

where Tk(x) denotes T.chebichef's po(ynomial (1.12.3). If we next assume

that [a', b'] contains no zeros, we have

-1 ~ M~\x.- a')(b' - x.) ~ 0,

so that p(x.) ~ 1. It then follows that

1~' p(x) da(x) ~ lb p(x) da(x) = ~ X.p(x.) ~ ~X. = lb da(x).

Now T~(x) is increasing for X ~ 1, so that Tk(~ + 1) > T~(l)~ = e~ for

~ > 0; thus we have in [a', b']

where Cis a positive eonstant independent ~f n. Hence n 2M;.X = 0(1), which

is a contradiction.
Concerning the distribution of the zeros for large values of n, see Theorem
12.7.2.

6.11.. Distance between consecutive zeros

Here and in the subsequent sections we consider distributions of the type
w(x) dx.
(1) THEOREM 6.11.1. Let w(x) be a weight function on the finite interval
[a, b], bounded from zero: w(x) ~ J.L > 0. Let xx > X2 > · · · > Xn be the zeros
of the associated orthonormal polynomial Pn(x) in decreasing order. 30 On writing
(6.11.1) x. = !(a + b) + !(b - a) cos e.' 0 < e. < 71", v = 1, 2, ... 'n,

29 He states (without a satisfactory proof) that if either a orb is finite, the subsequent
argument succeeds even with o(n 2 ).
ao Of course, each x. depends on P and n, Xv = x.,. .
 [ 6.11 J DISTANCE BETWEEN CONSECUTIVE ZEROS 113

we have
log n
(6.11.2) e,+1 - e. < K -- ,
n v = 0, 1, 2, · · ·, n; eo = 0, en+t = 71'.

Here the constant K depends only on J.L, a, and b.

See Krawtchouk 2; Erdos-Turan (written communication). Erdos-Turan

require the existence of f~ I w(x) l- dx instead of w(x) ~ J.L > 0. Their proof
1

(cf. 2) is based on the study of the distribution of the interpolation points

for which the associated fundamental polynomials (Chapter XIV) satisfy
certain conditions. The following proof for the special case w(x) ~ J.L > 0
has been prepared by B. Lengyel; he eliminated all references to the theory of
interpolation from the argument of Erdos-Turan.
(2) Let v be a ·fixed integer, 0 ~ v ~ n, and 'Y = (e. e.+ 1)/2. We then +
define p(x) = PI Ha b)+ + Hb - a) cos e1 by

(si~Y('Y (sin IN(-y - e)~~J) m

2
( 6 11 3) 2 (X) = + e)/21 ) 2m+
· · P N sin I (-y + e)/21 N sin {('Y - e)/21 '

where N and m denote certain positive integers. The single terms in the right-
hand member represent the same trigonometric polynomial of degree m(N - 1),
taken alternately with arguments 'Y +
0 and 'Y - e. Therefore, the sum is a
cosine polynomial of the same degree m(N - 1). If m(N - 1) ~ 2n - 1,
Theorem 3.4.1 can be applied.
We then have
(6.11.4) (e.+1 - e.)/4 ~; -y/2 < (-y + 11')/2 ~ 71'- (e•+1 - e.)/4;
whence for all values of k, 1 ~ k ~ n,
(6.11.5) 0 < (e•+1 - e,)/4 < ('Y + ek)/2 < 71' - (e•+1 - e.)/4,
so that

(6.11.6) . 'Y )-1 ( e-+1 -

(sm - +2-- < sm 4
Vk
ll • )-1
v.
ll
, k = 1, 2, · · ·, n.

The same inequality, with ~ instead of <, holds for I sin ('Y - ek) /2 1-\ since
I 'Y - ek I ~ (e.+1 - e.);~:. Thus,

(6.11.7) , ) =
p (.xk
< . (N sm. e.+1 4- e.)-2m, k =I, 2, · · ·, n.

By using (3.4.1) and (3.4.5), we obtain

bp(x)w(x) dx (N e 1 - e··)-2m }arb w(x) dx.

(6.11.8)
l ~ sin •+
4
On the other hand, the value of p(x) for e= 'Y is not less than !, so that

------ ·-------
 114 ZEROS OF ORTHOGONAL POLYNOMIALS l VI l

if we write Xo = Ha + b) + Hb - a) cos 1', we obtain from a later result (cf.

Theorem 7. 7)

(6.11.9) 1& p(x)w(::r) d:r ~ !J. 1& p(x) dx ~ !J.cn- p(x 2

0
) ~ ! /J.cn- 2,

where C is a positive constant depending on a and b only. Comparing (6.11.8)

and (6.11.9), we now have

! !J.Cn -2
r
;£ ( N sm
· () HI - (),
-)
-2m 1& w(.r) d:r,
a
4
or

By substituting N == [n/log n], m = [log n], we see that the condition

m(N - 1) ;£ 2n - 1 is satisfied for large values of n, and (6.11.2) follows
immediately.
The same inequality (6.11.2) holds without essential change if w(x) ~
!J.(l - xt(I +
x),s, !J. > 0, where a and {3 ~re greater than -1. In this case,
the last remark in §7.71 (4) must be used. The constant K in (6.11.2) now
depends on !J., a, b, a, and {3.
(3) We ·note the following simple result:
THEOREM 6.11.2. Let w(x) be a we-ight function on the interval [-1, +I],
and suppose
(6.11.11) -1;£x;£+1,
where A and Bare positive constants. If x. = cos e. , 0 < e. < 1r, v = 1, 2, · · . , n,
~tand for the zeros of the orthonormal polynomial Pn(x) associated with w(x), in
decreas~·ng order, we have

(6.11.12) v = 0, 1, 2, · · ·, n; Oo = 0, On+I = 1r.

This remark is due also to Erdos-Turan (written communication). The

proof can be based on Theorem 3.41.1. We denote by X. the Christoffel number
corresponding to x. . Then

1~·+
1
(6 .ll.l 3 ) A(O•+I- e.);£ w(cosO) sin Ode= 1~~~ w(x)dx ;£ x. + X,+ 1,

v = 0, 1, 2, · · · , n; Xo = An +I = 0.
On the other hand,
2
(x) = (cosO) = (sin {n(O - 0.)/2}_) + (sin {n(O + 0.)/2} )
2

P P nsin {(0- o.)/21 nsin {(0 + o.)/2l

·---,...----..,----------..,--------··-······
 [ 6.12] VARIATION OF ZEROS WITH PARAMETER 115

is a 11"n-l in X = cos e. Furthermore, p(x.) = p(cos e.) ~ 1, so that (cf. (1.6.5))

1
X. ~ t
k-I
Akp(xk) = p(x)w(x) dx ~ B {" p (cos 0) dO
1+
-1 }o
(6.11.14)
= B 1+,.. (sin. {n0/21 )
_,.. n sn1 {0/21
2
dO = 211"B.
n
On combining this with (fU 1.13), the statement is seen to be true.
Erdos-Turan proved also (cf. 2) that if 0 < A ~ w(x) ~ B, - 1 ~ x ::: + 1,
and 0 < E < 71"/2, then, wii~h the notations of Theorem 6.11.2,

(6.11.15)

provided E ~ e. ~ 71" - E. Here K1 and K2 depend on A, B, and E.

6.12. Variation of the zeros with a parameter

A. Markoff proved (4) an important statement concerning the dependence of
the zeros of Pn(x) on a parameter r which appears in the weight function
w(x) = w(x, r).
(1) THEOREM 6.12.1. Let w(x, r) be a weight function on the interval [a, b]
depending on a parameter· r such that w(x, r) is positive and continuous for
a < x < b, rx < r < r2 . Also, assume the existence and continuity of the partial
derivative w.(x, r) for a < :c < b, rx < r < r2 , and the convergence of the integrals

(6.12.1) lb x• wT(x, r) dx, v = 0, 1, 2, ... , 2n - 1,

uniformly in every closed inte·rval r' :S r ~ r" of the open segment r 1 , r2 . If

the zeros of Pn(x) = Pn(x, r) be denoted by Xx(r) > x2(r) > · · · > Xn(r), the vth
zero x. ( r) (!or a fixed value of v) is an increasing function of r provided that wT / w
is an increasing function of x, a < x < b.
The integrals (2.2.1) for the moments c. [da(x) = w(x, r) dx] converge uni-
formly in r' ~ r ~ r", and the relations (2.2.1) may be differentiated with
respect tor; v = 0, 1, 2, . · . , 2n - 1. Let a < a' < b' < b. For a' ~ x ~ b',
r' ;£ r ~ r", the function w(x, r) has a positive minimum; whence the deter-
minants Dn--1 are uniformly bounded from zero· if r' ~ r ~ r" [(2.2.11)].
According to (2.2.6) the coefficients of Pn(x), therefore also the zeros x.( r)
(which are all distinct), possess continuous derivatives for r 1 < r < r2.
Let p(x) be a fixed 11"2n-I. Apply Theorem 3.4.1 with da(x) = w(x) dx.
The Christoffel numbers ;,. = X.( r) are obviously functions of r with a con-
tinuous derivative [(3.4.3)]. Differentiating (3.4.1) with respect to r, we obtain

(6.12.2)

------ ·-------
116 ZEHOS OF ORTHOGONAL POLYNOMIALS [VI]

Now we substitute
2
(6.I2.3) p(x) = fpn(x) 1 whence
X - Xv '

so that
2
(6.I2.4)
1 b
w,. ( x, T
) { Pn(x) 1 dx = X. ( T ) { Pn1 ( x. )
X - Xv
I2 x.'(T ) ,
since p' (x 1J = 0 if J.L ;:C v. The left-hand member can be written in the form

l
2
(6.I2.5) b{-WT (x, T
)-wT(x.,r)
(
( )}{Pn(x)l dx,
) w x, T
W Xv 1 T ,X - Xv

the second term being zero because of the orthqgonality. The difference
wT(x, r) wT(x., r)
(6.I2.6)
w(x,r) w(x., r)
has the same sign as x - x. according to the assumption. This establishes the
statement. 31
(2) We point out the following eonsequence:
THEOREM 6.I2.2. Let w(x) and W(x) be two weight functions on [a, b], both
positive and continuous for a < x < b. Let W(x)/w(x) be increasing. Then if
{x.l and {X.1 denote the zeros of the corresponding orthogonal polynomials of
degree n in decreasing order, we have
(6.I2.7) x. < x.' v = I, 2, · · · , n.
Defining w(x, r) = (I - r)w(x) + rW(x), 0 ~ r ~ I, we see that
-1
wT(x, r) W(x) - w(x) _1 T
(6.I 2 .S) w(x, r) = (I - r)w(x) + rW(x) = r I - r + rW(x)jw(x)
is an increasing function of x, 0 < r < 1. We also have w(x, 0) = w(x),
w(x, I) = W(x).
Various applications of these results will be given in §6.21.

6.2. Location of the zeros of the classical polynomials

The discussion of this question, given in §3.3 for the general orthogonal
polynomials, was based on the orthogonality property. In the particular cases
called classical polynomials (§2.4) there exist various other approaches which
are interesting from the point of view of method. It is assumed that a > - I
and {3 > - I in the Jacobi case, and a > - I in the Laguerre case; furthermore,
n ~ 2.
3 1 This proof docs not differ essentially from the original one due to A. Markoff, although

the present arrangement is somewhat clearer.

-----,---r---------r--·----..--..·--.
 [ 6.2 l LOCATION OF ZEROS OF CLASSICAL POLYNOMIALS 117

(I) First of all, if we are dealing with the classical polynomials, a more precise
form can be given to the argument indicated in §3.3 (4). In fact, for the
polynomials in question the number of the sign variations in (3.3.5) for x = a
and x = b can be readily calculated. 32 We use (4.1.I) and (4.1.4) for Jacobi
polynomials, and (5.1.7) and (5.1.8) for Laguerre polynomials. In addition to
the reality and distinctness of the zeros, the statement as to their location
follows from the same argument.
In the Jacobi case the value of

sgn {P~'::f> (x) l · :x {p~a,,s) (x)}

at the zeros of P~a,,s>(x) can also be determined by means of (4.5.7) (instead of

the method in §3.3 (4) generally given). In the Laguerre case, (5.1.I4) can be
used. The situation is especially simple for Hermite polynomials; we have
but to apply the first equation in (5.5.IO).
(2) By Rolle's theorem the formulas (4.3.I), (5.1.5), (5.5.3) of Rodrigues'
type furnish the statement again. We must bear in mind that the derivatives
2
of (I - xr+a(I + xrH, e-xxn, and e-x of the orders 0, 1, 2, · · · , n - 1 vanish
at x = ±I, x = 0, + oo :• and x = ± oo, respectively.
(3) The statement ~lso follows from the differential equations (4.2.1), (5.1.2),
and (5.5.2). To this end we first show that the zeros of Jacobi polynomials
are different from -1, +I, and from one another. Differentiating (4.2.I) k
times, we have
(I - x 2 )y<HZ> + [!3 - a -· (a + i3 + 2k + 2)x]y<HI>
+ [n(n + a + i3 + I) - k(k + a + i3 + I)]y<k> = 0.
Were y to vanish for x = +I or x = -I, it would follow from (4.2.I) that
y' = 0, whence from the equation just obtained (k = I) y" = 0, and so on;
that is, y = 0. (The coefficient of y<Ho is different from zero for x = ± 1.)
Therefore, each of the zeros of P~a.,s> (x) is different from ±1. Moreover, the
zeros are all simple since (4.2.I) combined with y = y' = 0, x ;:e ±I, yields
2
(1 - x )y" = 0, or y" = 0. Using the equation for y', we find y"' = 0 in
the same way, and so on; that is, y = 0 again.
Similarly, we show that the zeros of L~a> (x) are simple and different from 0,
and the zeros of H n (x) are sim pie.
We next apply the following theorem due to Laguerre (P6lya-Szeg6 1, vol. 2,
pp. 59, 244-245, problem 118).
Let f(x) be a 11"n and Xo one of its simple zeros. Then any circle through the
points
(6.2.1) ' ( ) f'(xo)
x0 and Xo = Xo - 2 n - 1 f" (xo)
32
Professor P6lya has kindly called to my attention the fact that the same can be done
for the general orthogonal polynomials by using the determinant representation (2.2.6).

------·--------
118 ZEROS OF ORTHOGONAL POLYNOMIALS [VI]

contains some zeros of f(x) in both domains bounded by it, unless all the zeros
lie on the circumference of this circle. The same is true if a straight line replaces
the circle. 33
The proof is as follows. Let f(x) = (x - x0 )g(x), and let x 1 , x 2 , ... , Xn-I
denote the zeros of g(a:). Then g(xo) = f'(xo) and g'(xo) = !f"(xo), so that
I I g'(xo) f"(Xo)
(6.2.2)
Xo - Xx
+ Xo---
- X2
+ ... + .'ro - I Xn-1 =
()
g Xo
= 12
f' (xo)
•

Hence (6.2.I) becomes

(6.2.3) _I_ ( - I - + I + ... + I ) = I

n - I Xo -· Xx Xo - X2 Xo - Xn-1 Xo - X~ .
The linear transformation X = (xo - x)- 1 carries the points x1 , x2, · ·. , Xn-I,

and X~ into certain points Xx' x2' ... ' Xn-1 and X~ . Then we have

(6.2.4) _I_ (X1 + X2

n --I
+ · · · + Xn-x) =X~,
so that any straight line through X~ separates the points Xx, X2, ·. · , Xn-1
from one another, unless they all lie on this straight line. Referred back to the
x-plane, this yields the theorem.
For the Jacobi polynomials y = p~a,{J>(x) we obtain from (4.2.I)
2 .
1
Y I X
(6.2.5) - if y=O
y" -- a - {3 + (a + {:J + 2)x '
so that
, 2(n - I)
(6.2.6) Xo = Xo - a +I {3 +I .
I - Xo I+ Xo
Let x0 be a zero of y with the greatest imaginary part. If there were any non-
real zeros, we should have 3-(xo) > 0, and

(6.2.7) 3 (a+ XoI)> 0,

I-
3 (- {3 + I)> 0'
I+ Xo
whence S(x~) > 3(xo). Therefore x~ lies in the half-plane 3(x) > 3(xo),
and a circle can be dr:awn through x0 and x~ which contains no zeros. (The
zeros cannot all lie on this circle since they would then have to coincide with
x0 , an impossibility.) Hence all the zeros are real. Now let Xo be the greatest
zero, x0 ;:e ± 1. If we had Xo > I, (6.2.6) would give x~ > Xo. Considering
an arbitrary circle through xo and x~ , we are again led to a contradiction.
In the case of Laguerre polynomials,
33 If f"(x 0) = 0, we have x~ = oo, and a straight line through x 0 must be considered.
[ 6.2 J LOCATION OF ZEROS OF CLASSICAL POLYNOl\IIALS 119

,
, 2(n - 1)
(6.2.8) Xo = Xo- •
1 _a+ 1
Xo

From ,S(xo) > 0 we obtain, as before, ,S(x~) > ,S(x0 ). From x0 < 0 we find
'
Xo < Xo.
34

In the case of Hermite polynomials,

, n- 1
(6.2.9) Xo = Xo- - -
Xo '

so that ,S(x~) > ,S(xo) if ,S(xo) > 0.

( 4) Laguerre's theorem also furnishes certain bounds for the zeros. Let x 1
be the largest and Xn the smallest of the zeros of p~a,fl>(x). Then (a+ 1)/(1 - x 1)
- ({3 1)/(1+ +
xx) > 0 and (cf. (6.2.6))

< ~
(6.2.10) -1 X
n _
X
1
-
a +2(nI _- {31)+ I < Xx ,
1 - Xx 1 + Xx
so that

(6 .2 . 11 ) ~ +1 _ {3 + ]~ > 2(n - 1), {3- a+ 2n - 2

1- Xx 1 + x:x 1 + Xx Xx > {3 a + +
2n '
or for {3 ~ a
n- 1
(6.2.12) xx> n a -+ .
ln the ultraspherical case Xn = -xx, so that

(6.2.13)
a_+_I _
1 - Xx
~_+_I
1 + Xx
> n_-_1
Xx ' Xx > (n
n
+ 2a + 1
1 )! ·

(If n = 2, the sign > is to be replaced by =.) This bound is better than the
preceding one. Both bounds have the form 1 - (a 1)/n
2
O(n- ). Simi- + +
larly, an upper bound can be obtained for Xn.
An analogous argument yields the bounds
(6.2.14) Xo > ~~n +a - 1,
for the largest zeros Xo of L~a> (x) and of H n(x), respectively. (For n = 2 the
second inequality becomes an equation.) These are very rough estimates
(Theorem 6.32).
(5) Another proof of the reality and simplicity of the zeros (also furnishing
a < x. < b) can likewise be based on the differential equation by using the
considerations of §6.7. It must be remembered that the polynomials in ques-
tion are the only polynomial solutions of the corresponding differential equa-
tions (see §4.2 (2), §5.1 (1)).
34 From (6.2.8) we also find that the least zero is <a + 1.
120 ZEHOS OF ORTHOGONAL POLYNOl\fiALS [VI]

(6) In this connection, we refer to a very elementary method due to Laguerre

(3), which furnishes certain upper bounds for the zeros of the classical poly-
nomials .. Since g"(xo) = !f"'(xo), we have from (6.2.2)

1 + 1 + . .. + 1 = _ (~) g' (x0 )

(xo - X1) 2 (xo - X2) 2 (xo - Xn-1) 2 dxo g(xo)
(6.2.15)
_ {g'(xo) 12 - g(xo)g"(xo) _ 3{f"(xo) 12 - 4f'(xo)f"'(xo)
- {g(xo) 12 - 12{f'(xo) 12
and accor-ding to Cauehy's inequality
n-1 {n-1 1 }2
(n - 1) L: (xo -1 x.)
•-1
2
~ L: -
•-1
-
Xo - x.

or
(6.2.16) 3(n- 2){f"(xo)l 2 - 4(n- 1)f'(xo)f"'(xo) ~ 0.

This condition is necessary for each zero of a polynomial with real and distinct
zeros.
In the case of Legendre polynomials,
(1 - x~)f" (xo) = 2xof' (xo),
(1 - x~)f"'(xo) = 4xof"(xo) - (n - 1)(n+ 2)f'(xo)
= 2 - n - n
2
+ (6 + n + n )x~f'( Xo ) ,
2
2

1- Xo
so that
3(n - 2)4x~ - 4(n - 1)[2 - n - n 2 + (6 + n + n )x~] 2
~ 0;
whence

(6.2.17) ·1 Xo I ~ (n - 1) {
n + 2 }t = 1-
5/2
- 2 + .. ·.
n(n + 2)
2
n
The "true" constant, as n --t oo, in the second term is ji/2 = 2.891592 · · ·
(instead of 5/2), where j 1 is the least positive zero of Jo(x) (see (6.3.15)).
In the case of Hermite polynomials, from (5.5.2)
f"(xo) = 2.xof'(xo), f"'(xo) = 2(2x~ - n + 1)f'(xo),
3(n - 2)x~ - 2(n - 1)(2x~ - n + 1) ~ 0;

whence
< 2t(n - 1) _ 9
(6.2.18) I Xo I = (n + 2 )~ - (2n + 1) t _ 2 (2n + l) -t + ... .

·---·----·····.
[ 6.21 l ZEROS OF CLASSICAL POLYNOMIALS 121

This bound is better than that obtained by Sturm's method (cf. §6.31 (4)),
although according to (6.32.5) the "true" order of the second term is n- 116

6.21. Inequalities for the zeros of the classical polynomials

A. ::\Iarkoff's theorem (§6.12) furnishes .several remarkable inequalities for
the zeros of the classical polynomials.
(I) In discussing the zeros of Jacobi polynomials, we again enumerate the
zeros x. = cos e. in decreasing order:

(6.21.1) +I>xx>xz> ··· >xn> -l; 0<01 <02< ··· <On<11".

THEOREM 6.21.1. Let (±. = x.(a, {3) l denote the zeros of the Jacobi polynomial
p ~a' {3) (X) in decreasing order. Then

(6.21.2) ax.< 0 ax. 0

v = 1, 2, · · ·, n.
aa ' a{3 > '
In the ultraspherical case a = {3 we have

(6.21.3) ax. 0 v = 1, 2, · · · , [n/2].

aa < '
Applying Theorem 6.12.1 to w(x, r) = (1 - x)a(l + x) 13 with a = r, {3 fixed,
or {3 = r, a fixed, we obtain the inequalities (6.21.2) (cf. A. Markoff 4; Stieltjes
6, p. 76). In the first case we have, in fact, wT/w = log(l - x), which is a
decreasing function of x. The proof is similar in the second case.
The inequality (6.21.3) for the ultra.spherica.l case is due to Stieltjes (6, p. 77).
For the negative zeros the opposite inequality holds. This inequality does
not follow directly from Theorem 6.12.1, since for w(x, r) = (1 - x 2f the ratio
2
wT/w = log(l - x ) is not monotonic. However, it follows immediately from
(6.21.2) by using (4.1.5).
[The proof of Stieltjes for (6.21.2) and (6.21.3) is entirely different from that
of Markoff and is based on the differential eq ua.tion (see below §6.22). Markoff
also attempts a. direct approach to (6.21.3) through a. general theorem, but his
proof is incorrect. In his notation (4, p. 181) the function

f(y) = (y - e) V(y, ~)
aV(y, ~)
a~
2 1
is equal to y(log 1/(1 - y in the ultmspherica.l case. This function ap-
))-

proaches+ oo a.s y----. +O and- oo for y----. -0. Therefore, a.lthoughf'(y) < 0,
nothing can be said about the sign of the ratio (f(y) - f(x;))/(y - x;). Inci-
dentally, the condition aV(y, ~)ja~ > 0 of l\hrkoff is not satisfied in the ultra-
spherical case a.t y = e = 0;
In the general case, f(y) > 0 for y > e, and f(y) < 0 for y < e. This fact is
compatible with the decreasing property only if the denominator of f(y) becomes
122 ZEROS OF ORTHOGONAL POLYNOMIALS [VI]

0 as y ---t e. The function f(y) is, however, in any case discontinuous at y = e,

so that in the general case the same criticism. applies as in the special case
1

mentioned before.]
Apparently, Stieltjes was in possession of the general theorem of §6.12 (see
6, p. 79, section 5, and the remark on p. 88).
(2) THEOREM 6.21.2. Let the parameters a and {3 of the Jacobi polynomial
p~a.f3>(x) be subject to the conditions

(6.21.4)
Then we have for the zeros (notation as before)

(6.21.5) ~v- 1 < 0 < 2v _

2n +1 7r · -• = 2n +
1 7r' v = 1, 2, · · ·, n,

with equality only in the special cases a = -!, {3 == +!and a= +!, {3 = -!,
respectively.
For Legendre polynomials, that is, for a = {3 = 0, this result is due to Bruns
(1). The general case is due to A. Markoff and Stieltjes. For the proof we
observe that according to (6.21.2) the maximum and minimum of x. = cos o.
are attained in the special cases mentioned above. Now \Ve use (4.1.8).

(3) THEOREM 6.21.3. In the ultraspherical case

(6.21.6) - .l
2
<
=.....
rv = <
{3 = +.! 2

we have the inequalities

1)7r : $ p7r
(6.21.7) ( p - - 2 _5,() -- v = 1 2 .. · [n/2]
n- •- n+1' ' ' ' '
w£th the equality sign valid only in the special cases a = {3 = -! and a = {3 = +!,
respectively.
The first proof of these inequalities, which are more precise than the corre-
sponding inequalities (t).21.5) of "Bruns's type," is due to Stieltjes (6). Mar-
koff's proof is not correct (see above). For the proof of Stieltjes see §6.22.
Refer also to §6.3 (2) and (3). Corresponding inequalities for the negative
zeros can readily be obtained from the symmetry relation x. + Xn+I-• = 0.
We base our proof on Theorem 6.21.1. According to (6.21.3), the maximum
and minimum of x., v ;~ [n/21, are attained if a = {3 = -!and a = {3 = +!,
respectively. Now the zeros of the polynomials (4.1. 7) are

(6.21.8) COB (v - !) ~ and·

7r
cos v n + 1 , v = 1, 2, · · ·, n.
n
(4) In the case of ~~guerre polynomials we have w(x, r) = e-xxa, with a = r,
and wr/w = log x increasing. Hence the zeros of Laguerre polynomials are
[ 6.22] PROOF OF STIELTJES 123

increa~ingfunctions of the parameter a, a > -1. Thus, on account of (5.6.1),

we obtain the following theorem:
THEOREM 6.21.4. If
(6.21.9)
the zeros x. of L ~a) (x), arranged in increasing order, have the bounds
1:2 < . < 2
(6.21.10) c;;v = Xv = 71v •

Here~. and 'T/v denote the vth pos1'tive zeros of the Hermite polynomials H2n(x) and
H2n+I(x), respectively.

6.22. Proof of Stieltje,s for the monotonic variation of the zeros of the
classical polynomials
Stieltjes (6, pp. 73-77) gives a proof of Theorem 6.21.1 along the following
lines. Substituting x = x. in (4.2.1), we have

! y" + ! a + 1 + ! ~ + 1 = __1 _ + ... + 1

2 y' 2 Xv - 1 2 Xv + 1 Xv - X1 Xv - Xn
(6.22.1)
+ ! a + 1 + ! {3 + 1 = O.
2 x.- 1 2 x. 1 +
Differentiation of this equation with respect to a yields

(Xv -
1
X1) 2
(ax. _
Oa Oa
ax1) + (Xv -1 X2) 2(ax.
Oa
_ ax2) + ...
Oa

6 22 2 1 (ax. OXn) + 1 a + 1 ax.

( · · ) + (Xv - Xn) 2 oa - oa 2 (Xv - 1) 2 oa
+ 1 {3 + 1 ax. 1 1 _
0
2 (Xv + 1) 2 oa - 2 Xv - 1 - 1
or

(6.22.3) v = 1, 2, · · ·, n,

where

a., = 1 + ... + 1 + 1 +
(x. - X1) 2 (x. - Xv-I) 2 (x. - Xv+I) 2
(6.22.4)
1 1 a+1 1 !3+1
+ (x. - Xn) 2 + 2 (x. - 1) 2 + 2 (x. + 1) 2 '
and
1
(6.22.5) p ¢ J.L.
124 ZEROS OF ORTHOGONAL POLYNOMIALS [VI]

The matrix (av 11 ) is positive definite since

. ..
K = L: L: a.
v=l 11=1
11 u.u 11
(6.22.6)

Stieltjes now uses the following theorem: If A = (a. 11 ) is a positive definite

matrix with a.11 < 0, l' ¢ J.L, then the reciprocal matrix (A)- 1 has only positive
elements.
[We can assume that a.v = 1, v = 1, 2, . ·. , n, so that K = E - L, where E
is the unit form, and the coefficients of L are non-negative. Then the absolute
value of L is less than 1 if E = · 1, and the reciprocal form of K can be writ-
ten as follows:
(6.22.7) 1
(.K)- = E + L + L 2 + L + · ... 3

All the forms of the right-hand member have non-negative coefficients, and the
coefficients of E + L are also positive. 35]
By virtue of this theorem, the statement follows immediately from (6.22.3).
The proof for the seeond inequality of (6.21.2) is similar. The ultraspherical
case (6.21.3) can either be treated by means of (4.1.5), or directly handled (cf.
Stieltjes, loc. cit.). The same method applies to Laguerre polynomials (Theo-
rem 6.21.4). In this case we have

a,.• = .,..----1---::, + .. . + 1 + 1 + ...

(x. - X1) 2 (x. - Xv-I) 2 (xv - Xv+\) 2

(6.22.8) + 1 +a+1
(xv - x,.) 2 2x; '

with (2x.)- 1 on the right-hand side of the equations corresponding to (6.22.3).

6.3. Sturm's method; Jacobi polynomials

Sturm's method (see §1.82) leads very simply to certain inequalities for the
zeros of Jacobi polynomials (see Szego 20, Buelll). In this way we not only
confirm some of the results of §6.21, but we are also able to improve them to a
considerable degree. "T
e assume in this section that

(6.3.1) <
_!.2 = < +!.2!
a = :...._!.
2
<
= {3 <= +!.
2!

3s This argument is different from that of Stieltjes, which is likewise very simple and

elementary. The present argument, however, furnishes similar theorems in more compli-
cated cases.
 [ 6.3 I STURM'S METHOD; JACOBI POLYJ'\OMIALS 125

excluCling, in general, the case ci = {3 =

2
! ; we arrange the zeros x. = cos o. of
p~a,f3> (x) in decreasing order:
r

(6.3.2) +1 > X1 > X2 > ••• > Xn > -1; 0 < 01 < 02 < ••· < On < 7r.

(1) THEOREM 6.3.1. Under the condit£ons mentioned we have

(6.3.3) 0 0 7r
v - v-1 <n + (a+ {3 + 1)/2' p = 1, 2, ... 'n + 1.
2 2
This holds for a = {3 = i w£th the = sign instead of <. Here we define

(6.3.4) Oo ={ O if a > - !, and On+I = { 7r if ·f3 >

-01 1J a = -!, 27r- On if {3 =
Wenoticethatfora = {3 =-!,a= {3 =+!,a= -{3 =!,a= -{3 = _.l
2'

0. = (v- !) ~ 7r 7r p - !
(6.3.5) n'
P--
n + 1' P--
n+!' n-+2 17r'

v = 0, 1, · · ·, n + 1,
respectively.
Inequality (6.3.3) follows immediately from Theorem 1.82.1 by compa.ring
(4.24.2) with

(6.3.6) d2 ~ + ( + a + {3 + 1)2 = 0
d612 n 2 v .

We consider the solution v = sin {(n + (a+ {3 + 1)/2) (0- 0._1)}. For a > -!,
the condition corresponding to (1.82.2) is satisfied at xo = Oo = 0 (and similarly
for {3 > -!at Xo = On+I = 1r).

(2) THEOREM 6.3.2. Under the conditions mentioned we have

637
v + (a + {3 - 1) /2 . v
( · · ) n + (a+ {3 + 1)/2 7r < 0• < n + (a+ {3 + 1)/2 1r, v = I, 2' · · ·' n,
whereas in the ultraspherical case a = {3 = X - !
(6.3.8)' 0v > v + a/2 - ! v - (1 - X)/2 v =I, 2, ... , [n/ 2].
n + a + 21 7r = n + 1\'\ 7r,

The bounds (6.3.7) follow from (6.3.3) by addition and by using (4.1.3) (see
Buell 1, pp. 311-312). In case a = - ! , the factor v of the upper bound can be
replaced by v- !, while if {3 = -!,the factor v +(a+ {3- 1)/2 of the lower
bound can be replaced by l' + (a + {3)/2. In the cases (6.3.5) the same bounds
hold, at times with = replacing <. These inequalities are more precise than
(6.21.5) provided a + {3 > o·. In the case of Legendre polynomials (a = {3 = 0)
they are identical with (6.21.5), that is, with the inequalities of Bruns (1).
The inequality (6.3.8) holds only for the zeros 0 < o. < 1r/2; it becomes an

·------··-···-
126 ZEROS OF ORTHOGOXAL POLYXO:\IIALS [VI I

equation for a = {3 = -!or a = {3 = +!. It is more precise than the lower

estimate of (6.21.7).
For the proof of (6.3.8) we observe that according to (6.3.3), the sequence

(6.3.9) o' = () -
v ,,
v ++a/2+- i 7r, 1 v = 0, 1,2, · · · ,[(n+ 1)/2],
n a 2

is decreasing. Now for n odd, we have o;n+1)/2 - 0. For n even it suffices to

show that 8~12 > 0. This follows from (6.3.3), since On/2 + On/2+1 = 1r.
A similar argument ean be used to improve the left inequality in the ,general
case (6.3.7) provided the zeros considered all lie in a certain preassigned part
0 ;£ o. ;£ c of the interval [0, 1r}. For further proofs of (6.3.8) see (5) and
§6.6 (2). I

(3) In this connection we prove the following theorem:

THEOREM 6.3.3. Let n ~ 2. Under the condition <a - {J < +!the
sequence
(6.3.10)

of the zeros of p~a 'a) (cos 0) is convex, that is, o. - 0._ 1 is increasing.

This follows by applying Theorem 1.82.2 to (4.7.11), 0 < X < 1. In fact,

the coefficient of u in (4.7 .11) decreases monotonically. In the cases a = {3 = ±!
the differences o. - 0•.-1 are constant. Here again Oo = 0 if a > - !, while
Oo = -81 if a = - !. If n is even, the last term of (6.3.10) lies in [7r/2, 1r],
and then (1.82.4) is used. For a > -! condition (1.82.5) is satisfied.
From Theorem 6.3.3: a similar convex property for the sequence {x.} can
easily be derived (cf. Hille 4).
By means of Theorem 6.3.3 the upper estimate of (6.21.7), more precisely

(6.3.11) -!~a < +!; v = 1, 2, · · · , [n/2],

can be proved in another way (Szeg6 20, pp. 5-6, 8). For, let -! < a < +!.
The sequence

(6.3.12)
II
o. = o. - n
V
+ 1 7r, v = 0, 1, 2, · ·. , [n/2] + 1,
is convex; it therefore attains its maximum either for v = 0 or for v = [n/2] + 1.
Now o~' = o~'n/21+1 = 0 if n is odd. If n is even, we must bear in mind that
()n/2 + ()n/2+1 = 0.
II II

(4) Finally, by means of Sturm's method, we derive certain inequalities which

involve the zeros of Bessel functions. In some respects these are more precise
than the preceding inequalities, although not so simple.
THEOREM 6.3.4. Let a = {3 = X - !, 0 < X < 1. Denoting by i1 < j2 < ia < ···
the positive zeros of Bessel's function J a(x), we have
[ 6.31 J LAGUERRE AND HERMITE POLYKOl\1IALS 127

)v
(6.3.13) 0
" <n+ X'
For X = 0 and X = 1 the sign < in (6.3.13) has to be replaced by the sign =.
The statement follows by comparing (4.7.11) with (see (1.8.9))

+ {cn + >,)2 + X(1 -

2
d v X)} v = 0,
(6.3.14) d() 2 02
!
v =OJal(n + X)O}, a= X-~.
The estimate (6.3.13) of o. is the best possible in the sense that for a fixed v
and for n arbitrary the factor j. cannot be replaced by a smaller one since·
(Theorem 8.1.2)
(6.3.15) lim no. = lim n()vn = j • .
n-+oo n-+OO

Incidentally, for 0 < o. ~ 1rj2, a similar lower bound for o. can be obtained,
namely,
(6.3.16)
where k is a positive numerical constant (Szego 20, p. 9). Then (6.3.15) follows
from (6.3.13) and (6.3.1EI).
(5) Finally, another remarkable property of the zeros o. = Ovn of P~~> (cos 0)
can be proved by a proper application of Sturm's theorem. On substituting
() = ~/(n + X) in (4.7.11) we obtain
( 1 ) d2u + {ll + X(1 - X) } u = 0
6.3. 7 d~2 (n + X) 2sin 2 {~/(n + X)} ·

If 0 < X < 1, then (n + X) 2 sin 2 IU (n + X) l increases with n provided~ is fixed

and 0 < ~ < (n + X)1r. Thus (n + X)Ovn increases with n if vis fixed. 36 From
these considerations the estimate (6.3.13) follows again if (6.3.15) is known.
Indeed, (n + X)Ovn < limn-+oo (n + X)Ovn = j • .
As another application we can give also a new proof of (6.3.8) since

(6.3.18) (n + X)Ovn ~; (2v - 1 + X)Ov,2v-l = (2v - 1 + X)7r/2

for n ~ 2v - 1. Cf. Problem 32.

6.31. Sturm's method; Laguerre and Hermite polynomials

Suppose a > -1.
(1) THEOREM 6.31.1. Let x. = Xvn = Xvn(a), p = 1, 2, ... ' n, be the zeros
of L~">(x) in increasing order. Then
(j./2)2
f6.31.1) v = 1, 2, ... , n.
x. > n + (a+ 1)/2'
Here j. has the same meaning as in Theorem 6.3.4.
•• By the separation theorem (Theorem 3.3.2) 9," decreases as n increases for fixed "'
 128 ZEROS OF ORTHOGONAL POLYNOMIALS [VI]

When we compare the third equation in (5.1.2) with

(6.31.2) U" + (n + (a + 1) /2 + 1 ....:. a 2

) U = 0
x 4x 2 '

which has the solution

(6.31.3) U = X! J a {2x! (n + a ~ 1Y}

(cf. (1.8.10)), the statement follows immediately. Condition (1.82.2), x' = 0,
is satisfied in the present case.
An upper bound of ~~. of a similar kind can alsq be easily obtained. Let w
be a positive constant such that w < 4n + 2(a + 1). Compare the same
equation (5.1.2) as before with

(6.31.4) v" + {~~ + (a+

x
1)/2 - w/4 +1-
4x2
a2} v = 0,

where 0 < x ~· w. Then

(6.31.5) X < (j./2)2

• n + (a + 1)/2 - w/4'
if the expression on the right-hand side is not greater than w. (For a fixed v, this
· is the case, provided n is large enough.) The constant (j./2) 2 in the inequalities
(6.31.1) and (6.31.5) is the best possible in the sense explained in §6.3 (4). For
a fixed v, we have, for the zero x. = Xvn,
(6.31.6) lim nx.n = (j./2) 2 •
n-+oo

The same results can be obtained by use of (1.8.9) and the fourth equation
in (5.1.2). Condition (1.82.2), x' = 0, is again satisfied.
(2) Both equations (5.1.2) mentioned furnish upper bounds for the zeros
if we use Theorem 1.82.3 and take into account the fact that the correspond-
ing solutions vanish at x = + c.c. The bound which is obtained from the lourth
equation is slightly better. It is given in the following theorem:

THEOREM 6.31.2. The largest zero of L~a>(x) satisfies the inequaWy

(6.31.7) Xn < 2n +a+ 1 + ((2n +a+ 1) + i- a 1! 4n. 2 2

r-v

(3) Introducing x = (n + (a + 1)/2}- ~ in the third equation

1
(5.1.2),
we obtain

(6.31.8)

Since the coefficient of u increases with n, it follows that, for a fixed v, the
expression
[ 6.31 1 LAGUEHRE AND HERMITE POLYKOMIALS 129

(6.3{9) In+ (,x + 1)/2}x. = In+ (a+ 1)/2}x.n

decreases with increasing n. The limit as n ---t oo is (j./2) 2 (according to
(6.31.6)). An interesting; consequence of thi~ decreasing property is
(6.31.10) (n + (a+ 1)/2}x.n ~ lv + (a+ 1)/2}x•• , n = v, v + 1,
Applying (6.31.7) and recalling (6.31.1), we obtain the following result:
THEOREM 6.31.3. Let a> -1 ;for the zeros x.n of L~a>(x), arranged in increas-
ing order, the fallowing estimates hold:
(j./2)2
n +(a+ 1)/ 2: < x.n < lv +(a+ 1)/2}
(6.31.11) . 2v + a + 1 + I (2v + a + 1) + i - 2
a
2
}
1

n +(a+ 1)/2
v = 1, 2, ... , n; n = 1, 2, ....
In particular, for the least •~ero X1n we have

(6.31.12) (ji/2)
2
+ 3)
< (a + 1) (a
n + (a + 1)/2 < XIn = 2n + +1 '
a n = 1, 2, 3, ....

Here j. has the same meanin.g as in Theorem 6.3.4.

2
If vis large, (j./2) ::::: 1rV/4 (cf. (1.71.7)), while the coefficient in the right-
2
hand member of (6.31.11) is '"'-'4v • For v = 1 we do not need (6.31.7) since
x 11 can be calculated explicitly. In fact, xu = a + 1; whence (6.31.12) follo,vs.
E. R. Neumann (2, p. 2t3) obtains for a = 0 an inequality similar to (6.31.11)
in a different manner. He finds in this case

(6.31.13) v = 1, 2, ... , n; n = 1, 2, 3, · .. ,

where i < C.n < 4. In view of the well-kno"·n estimate j. > (v - i)7r (cf.
(8.1.4) and Prohlem 32), we can derive the following inequalities of the type
mentioned from (6.31.11):
2
(6.31.14) (37r /16) < C.n < 4.

(The upper bound 4 can not be diminished; cf. (8.9. 15); also Problem 33.)
W. Hahn (1, pp. 228-238) generalizes and extends Xeumann's method to arbi-
trary real values of a.
A part of these results is more precise thau those occurring iu the literature
(see W. Hahn 2, pp. 228-230).
(4) The corresponding considerations for Hermite polyuomials are very
simple. To begin with, the second equation (.5.5.2) iurnishes the upper bound
(2n + 1)! for the zeros. This is not as good as the bound (6.2.18). Further-
more, assuming n ~ 2, we obtain the convexity of the sequence of the zeros
 I30 ZEROS OF ORTHOGO~ AL POLYNO:\HALS [VI 1
(6.31.I5) Xon < X1n < X2n < · ••
of Hn(x), where Xon == 0 if n is odd, and Xon = · -Xln if n is even. In all cases,
X1n , X2n , • • • denote the positive zeros in increasing order. (See W. Hahn 1,
p. 244.)
On comparing the same equation with Z" + (2n + l)Z = 0, we find 37

(2: ~!I)! 1r,

(6.31.I6) Xvn >
{
(2n ~ I)! 1r, v = I, 2, · · ·, [n/2].
(This follows also from (6.31.I) for a = ±!.) Now let w be a fixed positive
number, w < (2n +
1)!. Then

(6.31.I7)

(2n +I - w 2)! 1r,

v = I, 2, 3, · · · , [n/2],

provided the right-hand members are not greater than w. For a fixed v we
see that the constants (v - !)1r and v1r are the best possible.
By introducing x = (2n +
I)-!~, the differential equation mentioned is
transformed into

~~: + {I + o- (}z = 0,
- (2n
2

(6.3I.l8)
z = exp { -(2n + 1)- (/21Hn{(2n + 1)-;~}.
1

The coefficient of z inereases with n; hence (for fixed v) (2n + I);xvn decreases
as n increases. Therefore, we have (cf. (3)) (2n + I)!x•n ~ (4v + I)!x•. or 2.
(4v + 3)!x•,2•+1, respeetively. Thus

v -+ !I)! 7r )
(2n 14v + I)!'
(2n + I
(6.31.I9) < x.n <
v 4v + 3
(2n + I)! 7r (2n + I)!'
v = I,2, .. ~,[n/2].

l I(
For the least positive zero X1n, we obtain (x12 = 24 , x13 = (3/2)!)

7r/2 5/2 )!
(6.31.20) (2n + I)! < Xln ~ 2n +
I !'
7r . ( 2I/2 ) n f;; 2.
(2n + I)! 2n +
I '
7
3 In this and subsequent formulas the upper line eonesponds to the case n even, while
the lower line to the ease n odd.
! 6.32 I THE LARGEST ZEROS I3I

The' upper bounds are more precise than those resulting from (6.3l.I9) for
v = 1.
From (6.31.20) we can derive bounds for the minimum distance dn between
consecutive zeros. The convex property mentioned above easily furnishes
dn ·= X1n - Xon , that is, dn = 2Xln if n is even, and dn = X1n if n is orld. It
follows that
( 10!

(6.31.2I) 7r
~----,-, < dn l(2n
:=::;;
I)!'+
(2n + I)! - (2I/2)!
n ;;; 2.
(2n I)l'+
In every case we have
7r < (2I/2)!
(6.31.2_2)
(2n +I)!< dn = (2n +I)!"
Concerning the extensive literature on this subject, we refer to Laguerre (2,
p. I05), Korous (1), Wiiman (1), A. Brauer (1), Hille (4), and Winston (1).
Hille obtains the same lower bounds as in (6.31.20) and (by a suitable choice
of w in (6.3l.I7)) the upper bounds

l
(~~n7r~ I)!{~+ ~[I- (2n ~ IYJ}-!,
(6.31.23)
x,. < (2n: J)l {! +! [ 2
1 - (2n : SJT'·
These bounds are better than those resulting from (6.31.20), except for
n ~ 6. The results of the other authors are less precise than the preceding
38
inequalities.
6.32. Sturm's method; the largest zeros of Laguerre and Hermite polynomials
(I) Let a > -I, and enumerate the zeros x. = x.n of L~a>(x) or Hn(x) in
decreasing order:
(6.32.I) X1 > X2 > Xa > ··· .
Our purpose is to derive inequalities and asymptotic relations for X1n , as well
as for Xvn for fixed values of v as n --t oo .
THEOREM 6.32. Let i11 < i2 < ia < · · · be the real zeros of Airy's function
A(x) (§1.8I, i1 > 0). If I a I ~ l, a > -I, the following inequalities hold for the
zeros {x.} of L~a>(x):
(6.32.2)
as An exception is the lower bound in (6.31.20) for the special values n = 2 and n = 3
in which Wiman's expression furnishes the exact values. For n = 3 the upper bound of
Wiman is the same as in (6.31.20).
132 ZEROS OF ORTHOGONAL POLYNOMIALS [VI]

whereas Jor the zeros Ix·.} of H n(x) :

(6.32.3) ;r. < (2n + 1)! .:._ 6-!(2n + 1)-t i •.
Furthermore, we have for a fixed v
(6.32.4) X~ = (4n + 2a + 2)! - 6-!(4n + 2a + 2)-i Ii. + Enl,
(6.32.5) x. = (2n + 1)!- 6-!(2n + 1)- 1{i. +En},
in the Laguerre and H e:rmite cases, respectively, where limn-oo En = 0.
These remarkable results possess an extended literature. (Zemike 1, W.
Hahn 1, p. 227. See also Korous 1, Bottema 1, Van Veen 1, and Spencer 1.')
In what follows, Sturm)'s method is used to prove (6:32.3). A similar argument
can be applied for the proof of (6.32.2) (cf. the fourth equation in (5.1.2)).
Formulas (6.32.5) and (6.32.4) follow from a certain asymptotic expansion of
Hermite polynomials due to Plancherel and Rotach as well as from corre'Spond-
ing expansions for Laguerre polynomials; (6.32.4) holds for arbitrary real a.
We discuss these formulas in Chapter VI.II (cf. §8.9 (3)). They show that the
constant i. in (6.32.2) and (6.32.3) is the best possible if v is fixed and n is
arbitrary.
We notice that the expressions
\ (4n + 2a + 2)~ - 6-!(4n + 2a + 2)-*id 2
,
(6.32.6)
(2n + li - 6-!(2n + 1)-til
are upper bounds for the zeros of L~">(x) and Hn(x), respectively, Ia I > 1.4)
=
a > -1. Here the constant

(6.32.7)

cannot be replaced by a smaller one. These bounds are more precise than
those previously given.
Alternative forms of (H.32.4) and (6.32.5) are
.:r! = (4n)t - 6-li.(4n)-i + o(n-1),
(6.32.8)

re:-;pectively. Also, the first formula can be written as follows:

(6.32.9) x. = 4n- 2·6-li.(4n)l + o(nl).

(2) 'Vriting hn = (2n + 1)!, we substitute x = hn - ~in the second equation
(5.5.2) and obtain
2
d z
(6.32.10) de + (2hn~ - 2'
~ )z = 0.

Next we compare this equation with

t 6.32 I THE LARGEST ZEROS 133

2
d Z
(6.32.11) de + 2hn~Z = 0,

which has the solution Z = A I (6hn)l ~ l; here A (x) is Airy's function defined
in §1.81. We can then apply Theorem 1.82.1 in [- oo, + oo]; condition (1.82.2)
is satisfied at~ = - oo (ef. the last remark in §1.81). Thus we have (6 hn)-ti"
< hn - x" , which establishes (6.32.3).
(3) We add a sketch of a direct proof of (6.32.5) based on Sturm's method.
Let the real variable~ be subject to the condition I ~ I ~ 2hnEn , where 0 < En < 1;
we shall dispose of En later. Then

(6.32.12)

We now compare (6.32.10) with

d2t
(6.32.13) de + 2hn~(l ± Enk = 0,

where the signs + and - correspond to ~ ~ 0 and~ ~ 0, respectively. Using

the notation (1.81.1), we eonsider, for - 2hnEn ~ ~ ~ 0, the solution

(6.32.14)

with

(6.32.15)

This solution vanishes for~ = - 2hnEn . On the other hand, for 0 ~ ~ ~ 2hnen,
we shall consider the solution

At~ = 0 it has the same value and the same derivative as (6.32.14) on account
of (1.81.1). According to Sturm's theorem, H n(hn - ~) oscillates more rapidly
in the interval __: 2hnEn ~ ~ ~ + 2hnEn than the function t = t(~) represented by
(6.32.14) and (6.32.16).
The only neg:;ttive zero of t(~) is ~ = - 2hnEn . We now calculate the positive
zeros of (6.32.16), that is, the values of ~ for which

(6.32.17)
l\(6hn)t(l- En)t~~
k\ (6hn) (1 -En)!~~
11
=
(1- En)t l(-Xn)
1 +En k( -Xn) •

If En is small and X n large, the right-hand member is nearly -1, and the vth
zero in question is, for a given v, near to (6hn)-t(l - En)-ti". If Xn is large
and positive, we obtain from (1.81.3) and (1.81.5)
134 ZEROS OF ORTHOGONAL POLYKOMIALR [VI]

(6.32.18)

Now let vo be a fixed positive integer and E an arbitrary positive number.

We choose
-4/3
(6.32.19) En = h n w,
where w is a fixed positive number, so large that 2 · 6tw > i"o + 1, and the left- 1

hand member of (6.32.18) is less than E. For sufficiently small vah1es of E,

and a:s n--+ oo, the first vo zeros of (6.32.17) then have the form
(6.32.20) v = 1, 2, ... ' vo,

where I 8 I is arbitrarily small with E.

(At any rate let I a I be less than 1.)
"r e see that for large n, the expressions
(6.32.20) are less than 2hn En. • Hence,
when we apply Sturm's theorem in - 2hn En ~ ~ ~ +2hn En, we have
(6.32.21)
This latter relation combined with (6.32.3) establishes the statement (6.32.5). I
i

6.4. Theorem of Polya-Szego on trigonometric polynomials wit~

monotonic coefficients
THEOREM 6.4. Let ao > Q1 > ... > am > 0. Then the functions
f(t) = ao cos mt + a1 cos(m - 1)t + · · · + am-1 cost + am, .
(6.4.1) g(t) = ao cos(m + ~)t + a1 cos(m - ~)t + · · · + am-1 cos(3t/2)
+ am cos(t/2),
have only real and simple zeros; there is, respectively, exactly one zero in 'each of
the intervals

(6.4.2) p ,-2
--"Tr
1
<t...-Jl + 21
., - - " T r and
p,-~
- - - 7r < t < J.t.+~
- - 7r
m+~ m+~ m+1 m+1'
wherr p, = 1, 2, ... , 2m, and J.L = 1, 2, · · · , 2m+ 1, respectively.
The first part of the statement is due to P6lya (3, p. 359); his proof dses the
principle of argumrnt (Theorem 1.91.1). The following proof furnishes P6lya!s
result :again and, in addition, the inequalities (6.4.2) (Szeg6 20, pp. 9-11). It
is based on Fcjer's fundamental theorem (Frjer 1; ::;ee P6lya-Szego 1, vol. 2,
pp. 78, 269, problem 17) which asserts that the sine polynomials ·
crn(t) = sin(t/2) + sin(3t/2) + · · · + sin(n + ~)t,
(6.4.3)
n = 0, 1, 2, ... ; 0 <t < 2-rr,
are non-negative.
If J(t) and {j(t) denote the conjugate functions of f(t) and g(t), respectively,
[ 6./i l FEJER'S GENERALIZATION 135

we have
m
(6.4.4) -3e-i(m+i)llf(t) + i](t) l = -3e-i(m+Otlg(t) + ig(t) l = La~< sin(,~L + !)t,
~<-o

which is positive for 0 < t < 21r, according to Abel's transformation (1.11.4).
Therefore,
j(t) sin(m + !)t - ](t) cos(m + !)t > 0,
(6.4.5)
g(t) sin(m + 1)t - g(t) cos(m + 1)t > 0, 0 <t < 27r;
whence

(6.4.6) sgn !(~-

! 1r) = sgn g (~-' - ! 1r) = ( -1)1<+
1.
m+! m+1
This sho:ws the existence of at least one zero in each of the intervals (6f4.2).
On the other hand, the functions (6.4.1) cannot have more than 2m and 2m:.+ 1
zeros, respectively, in (0, 21r].

6.5. Fejer's generalization of Legendre polynomials

(1) Starting from the representation (4.9.3) of Legendre polynomials, Fejer
(9) defines the "Legendre polynomials F n(x) associated with a given seqJence
ao , a1 , a2 , · · · " in the following way:

Fn(cos 0) = 2aoan cos nO+ 2alan-1 cos (n- 2)0 + · · ·

(6.5.1) 2a(n-1)/2 a(n+l)/2 COS 01 if n <i>dd,

+ {an/2 2
1
if n even.

The classical Legendre polynomials P n(x) are obtained if

1·3 · · · (2n - 1)
(6.5.2) ao = go =: 1; a n-- gn-- 24 1 n= 1,2,3,'···,
. . . . 2n

the ultraspherical polynomials P~\x) if (§4.9 (4))

(6.5.3) ao = 1; .an
= (n +X-
n
1) ' n = 1, 2, 3,' · · ·.

Various properties, well-known in these special cases, can be extended to the

general polynomials F n(x) by imposing proper restrictions on the sequence Ian!.
These restrictions concern certain properties of monotony and asymptotic
behavior.
(2) THEOREM 6.5.1. The zeros ofF n(x) are real and simple and lie in the inter-
val - 1 < x < + 1, provided an > 0 and the sequence
(6.5.4)
 13()
ZEROS OF ORTHOGONAL POLYNOMIALS
(VI]
is increasing. More precisely, each interval

(6.5.5) v-! v+!

n-+1 7r < 0 < n+1
- 7r,
v = 1, 2, · · ·, n,
contains exactly one zero of Fn(cos 0).

See Szego 20, pp. 15-17. Under the condition mentioned the coefficients of
(6.5.1) are decreasing, and the statement follows immediately from Theorem
6.4 if we write n = 2m or n = 2m+ 1, according as n is even or odd and 20 = t.
The condition in question is satisfied for Legendre polynomials' and in the
u1traspherical case for 0 < :\ < 1.
The inequalities (6.5.5) are not so precise as the Bruns inequalities (6.21.5).
However, they hold for a comparatively general class of polynomials.
(3) THEOREM 6.5.2. Let the sequence {an}, an > 0, be completely monotonic,
that is, for all the differences, 39

(6.5.6) t/ an = an - (;) an+l + (~) an+2 - ··• + (-1)k an+k f:; 0,

k, n = 0, 1, 2, · · · .
Then the zeros x. = cos o. , 0 < o. < 1r, of F n(x) are not only real and lie in
[-1, +1], but they also satisfy the inequalities (6.21.7) of Stieltjes:

(6.5.7)
v = 1, 2, · · · , [n/2].

Here the signs of equality hold if and only if Fn(x) is Tchebichef's polynomial of
the first (see below) or of the second kind, respectively.

See Fejer 17, pp. 311-312. According to an important theorem of Hausdorff

(1) the class of completely monotonic sequences [an}, an > 0, is identical with
the class of sequences which can be represented in the form

(6.5.8) an = 1 1
C da(t), n = 0, 1, 2, · · ·,

where a(t) is a non-decreasing function, not constant, with 2a(t) = a(t + O)

+ a(t - 0) for 0 < t < 1. For sequences of this kind the condition of Theorem
6.5.1 is satisfied (Schwarz's inequality). The ultraspherical polynomials p~XJ(x),
0 < >-. < 1 are obtained [(1. 7.2)] if

(6.5.9) a. ~ (n +~ - 1
) ~"_,sin x,. /.' t+'-'(1 - t)-' dt, n = 0, 1, 2, · · · ·

Tchebichef's polynomials of the first kind are a limiting case of (6.5.9) since
ag This definition of the differences of various orders is not the same as in (2.8.4).
 FEJI ATION 137

I
(6.5.10) ao = 1, t-
1
dt = ~' n = 1, 2, 3, · · ·,
so that

(6.5.11) lim X- 1 Fn(cos0) =- :os nO, n = 1, 2, 3, · · ·.

x-o
Tchebichef's polynomials of the second kind arise if an = rn, 0 < r ;:£ 1, that is,
a(t) has only one point of increase in 0 < t ;:£ 1.
Fejer's argument is as follows:

(6.5.12) Fn(cos 0) = 1 {:t

o
1
[I
}o k-o
t'< un-k cos (n - 2k)o}da(t) da(u).

An elementary transformation of the integrand gives

(t - u) (t+! - un+!) 0 + 2tu(tn + un)sin2 0 sin (n + 1)0

(6.5.13) cos n
t2 - 2tu cos 20 + u 2 t2 - 2tu cos 20 + u 2 sin 0
so that

(6.5.14) f'n (cos 0) =An (0) cos nO+ Bn(O) sin (n. + 1)0 ,
smO
where An(O) and Bn(O) are positive functions in 0 < 0 < 1r provided a(t) has
at least two points of increase. From this it follows that

sgn Fn {cos (v - !) ~} = -sgn Fn {cos v n : } = (-1)"+\

(6.5.15) 1
v = 1 2 .. · [n/2]
' ' ' '
which establishes the statement.
The last part of this argument is similar to that used in the proof of
Theorem 6.4.
(4) Fejer considers (20, pp. 40-45) another remarkable generalization of
Legendre polynomials. He starts from the representation (4.9.5}. Let
f3m l 0, and

Gn(cos 0) = f3o sin (n + 1)0 + {31 sin (n + 3)0 + · · ·

(6.5.16)
+ {3m sin (n + 2m + 1)0 + · · · .
This series converges for 0 < 0 < 1r (see §4.9 (2)). Legendre polynomials are a
special case, as well as the more general functions (sin 0) 2 x-rp~x)(cos 0),
X > 0, X ~ 1, 2, 3, · · · (see (4.9.22)). In these cases the sequence f3m = ! m(X)
(using the notation of (4.9.22)) is completely monotonic. In fact, we have

ex) ex) · cxl{ nX 1 X(n + 2X) 1 } _ cx)

(6.5.17) fm - f m+! = f m n +X m + 1 + n +X n +X+ m + 1 - 1m 'Ym·
138 ZEROS OF ORTHOGONAL POLYNOMIALS (VI]

The sequence hm l is completely monotonic and, on account of a well-kn(}wn

formula, we have

(6.5.18) .U.
Ak+!f(X)
m
= AkJj(X) _!(X)
l m
.U. m+!
~
l = £....
•=0
(k) v
A"f(X) Ak-P
.U. m .U. 'Ym+•·

Hence the statement follows by induction. 40

Fejer shows (loc. cit.) that Gn(cos 0) has at least one zero in each interval

1)7r 7r
(6.5.19) ( v-2-<0<v-+
n n 1
, v = 1 2 .. · [n/2]
' ' ' '
provided
(6.5.20)
His proof is based on the positiveness of certain special trigonometric poly-
nomials. We shall prove the following theorem:
THEOREM 6.5.3. The funct£on Gn(cos 0) has at least one zero £n each £nter-val
(6.5.19) provided !,B, l is a completely monotonic sequence.
This condition is more restrictive than that of Fejer. The proof is, how-
ever, very simple. Using (6.5.8), with f3n and {3(t) in place of an and a(t),
respectively, we obtain

(6.5.21)
G,.(cos 0) = 1{to1
tm sin (n +2m+ 1)0}d{3(t)

t 2t d{3(t) . t
(1 - t) d{3(t) .
= }o f-.::... 2t cos 20~ t 2 sm 0 cos nO+ }o 1 - 2t cos 20 + t 2 sm (n + 1)0.

From this point the statement follows in the same way as in the proof of
Theorem 6.5.2.

6.6. Recapitulation; additional remarks on ultraspherical polynomials

(1) We have obtained the following inequalities for the zeros x. = cos o.,
0 < o. < 1r, (arranged in decreasing order) of the ultraspherical polynomial
P~"·")(x), a = X - !, provided 0 < X < 1:
(a) Inequalities (6.5.5), derived from the representation of P~"·"\cos 0) as a
cosine polynomial:
v-! v+! v = 1, 2, · · ·, n.
(6.6.1) n + 1 < 0" < n + 1
7r 7r'

(b) Inequalities of the Bruns type:

v-! v
n-+2 n-+2
(6.6.2) 1 7r < o. < 1 7r, v = 1, 2, · · ·, n,

4o For A. = 1/2 this follows directly from (4.9.9).

[ 6.6 J RECAPITULATIO~; ADDITIO:;-;AL RK.\IARKS 139

v+A-1 v
(6.6.3) n + A 7r < o. < n +A 7r, v = 1, 2, ... ' n.

Inequalities (6.6.2) follow from (6.21.2), which was proved by A. Markoff and
by Stieltjes in two different ways (cf. (6.21.5)); (6.6.3) is a special case of the
more general inequalities (6.3. 7) proved by Sturm's method. The inequalities
(6.6.2) are more precise than those of (6.6.1) and those of (6.6.3) with A < !;
the opposite is true if A > ~.
(c) Inequalities of Stieltjes' type:

(6.6.4)
v - ! < o. < -v- 7r
---- 7r v = 1, 2, · · ·, [n/2].
n n+ 1 '
These follow from (6.21.3) and were proved by Stieltjes. They can also be
readily derived from (6.21.2) (which is due to A. Markoff and to Stieltjes).
Fejer obtains them from (4.9.19) or (4.9.22) (cf. Theorems 6.5.2 and 6.5.3).
An alternative proof for the upper bound is due to Szego (Sturm's method,
§6.3 (3)). The upper bound is better than that in each of the preceding in-
equalities; the lower bound is better than that in (6.6.2), and is better than
that in (6.6.3) provided A < !.
(d) Szego's lower bound:
v - (1 - A)/2
(6.6.5) 0" > n +A 1r, v = 1, 2, · · · , [n/2].

This follows by Sturm's method in two different ways (cf. 6.3 (2) and (5)).
For a third method see (2) below. This lower bound is more precise than
any of the'preceding ones.
(2) By combining the integral representation (4.82.3) with the argument
used in the proofs of Theorems 6.4, 6.5.2, 6.5.3,
(e) Fejer obtains (19, p. 208)

(6.6.6) v-(1-A)/2 0 v+A-! v=1,2,···,[n/2].

n + A 7r < "< n + 2A 1r,
The lower bound is the same as in (6.6.fi). The upper bound is lcF;s or greater
than that in (6.6.4) according as A < 1 or A > ~·
For the proof we substitute the bound in (6.6.5) for 0 in (4.82.3), 0 < 0 < 1rj2,
and find that
(6.6.7) sgn P (X)
n
(
cos 0) = ( - 1)" t;gn '-''
c)C J iH0-7r/2! (l
le - t e2i0)-X l .
It happens that the last sign is constant if t varies in 0 < t < 1. Indeed, the
argument of the expression in the braces lies between 0 ancl - A(7r/2 - 0).
Thus (6.6.7) becomes ( -1)"+ 1 •
Upon substituting the upper bound of (6.6.6) in (4.82.3), we have
sgnP~x)(cosO) = (-1)"sgnS!(1- te
20
(6.6.8) ; )-xl = (-1)",
which establishes the statement.
I40 ZEROS OF ORTHOGONAL POLYNOMIALS t VI l

To recapitulate: the lower bound (6.6.5) ~s the best of all lower bounds given
here; while in the case of the upper bounds, either (6.6.4) or (6.6.6) is best
according as X > ! or X < !. Here we did not refer to inequalities which in-
volve zeros of Bessel functions.

6. 7. Electrostatic interpretation of the zeros of the classical polynomials

Stieltjes gave (4, 5; 6, pp. 75-76; cf., also, Schur 1) a very interesting deriva-
tion of the differential equations of the classical polynomials, which is closely
connected with. the calculation of the discriminant of these polynomials (cf.
§6. 71) and can be interpreted as a problem of electrostatic equilibrium.
(I) PROBLEM. Let p and q be two given positive numbers. If n unit "masses,"
n ~ 2, at the variable points x 1 , X2 , x 3 , • • • , Xn fn the £nterval [- I, +I] and
the fixed masses p and q at +I and -I, respect£vely, are cons£dered, for what posi-
tion of the po£nts x 1 , X2 , Xa , • • · , Xn does the expression
n
(6.7.I) T(:r1, X2, · • ·, Xn) = T(x) = IT (I -
K-1
x.Y(l + x.)q II
,.,J.'-1,2,· · ·.n
I x. -X~< I
•<~<

become a maximum?
Obviously, log (T-1) can be interpreted as the energy of the system of electro-
static masses just defined. They exert repulsive forces according to the law
of logarithmic potential. The maximum position corresponds to the condi-
tion of electrostatic equilibrium. A maximum exists because T is a continuous
function of Xr, X2,. • · , Xn fbr -I ~ x. ~ +I, v =I, 2, · · ·, n. It is clear
that in the maximum position the x. are each different from ±I and from one
another. In addition, this position is uniquely determined. To show this, let
us suppose that (cf. Popoviciu 2, p. 74)
>xn> -I,
(6.7.2) I I I
+I > X1 > X2 > > Xn > -I
are two positions of this kind; we write
(6.7.3) y. = (x. + x~)/2, v = I, 2, . · · , n.

Then

I y. - y I' I = Ix. - x~< I + Ix~ - x~ I >= I X. - XI'

I! I x.I - xlI' I! ,
(6.7.4) 2
II ± y. l ~ II ±x. I! II ± x~ 1!,
so that T(y) ~ !T(x)l!\T(x1 )l!, the equality sign being taken if and only if
x. = x~ . This establishes the uniqueness.
THJ<JOREM 6.7.1. Let p > 0, q > 0, and let !x.l, -I ~ x. ~ +I, be a system
of values for which the expression (6.7.I) becomes a maximum. Then the \x. l
are the zeros of the Jacobi polynomial P~a,f3) (x), where a = 2p - I, {3 = 2q - 1.
('6.7 J ELECTROSTATIC INTERPRETATION OF ZEROS 141

From this fact the uniqueness of the maximum position follows again. For
a maximum we have the conditions aT j(ax.) = 0, or
1
x.- X!
+
(6.7.5)
+ 1 + ---'-P_ + q = 0.
X. - Xn X. - 1 X. +1
If we introduce the polynomial f(x) - (x - x 1)(x - x 2) ••• (x - Xn), this
becomes
1 f"(x.)
(6.7.6)
2 ]'( x:) + x. ~ 1 + x. ~ 1 = 0'
or
(1 - x;)f"(x.) + \2q- 2p- (2q + 2p)x.lf'(x.) = 0.
The last equation means that (1 - x )f"(x) + \.B - a - (a + ,B + 2)x lf'(x)
2

is a 7l"n which vanishes for all the zeros of f(x); whence this expression is equal
to const. f(x). By comparing the terms in xn we obtain for the constant factor
the value -n(n + a + ,B + 1). The resulting differential equation reduces to
(4.2.1), so that according to Theorem 4.2.2, f(x) must be a constant multiple
of P~a,f3)(x).
See also Problem 37.
(2) The zeros of Laguerre and Hermite polynomials admit a similar inter-
pretation.
THEOREM 6.7.2. Let us cons£der the pos£t£ve mass p at the fixed po£nt x = 0
and un£t masses at the var£able po£nts x 1 , X2, . • • , Xn £n the £nterval [0, +co]
such that_ the "centro£d" sat£sfies

(6.7.7)
where K is a preass£gned positive number. Then the maximum of
n
(6.7.8) U (X! 1 X2 1 • • • 1 Xn) = II x:
•-1
II
P,J.I-1,2,· · ·,n
I x. - XJ.I I
"<J.I

is attained if and only if the !x.l are the zeros of the Laguerre polynomial L~a)(cx),
where a = 2p - 1, and c = K- 1(n a). +
THEOREM 6.7.3. Let us consider a un£t mass at each of the variable points
x1 , x 2 , •• • , Xn in the interval [- co, +
co] such that the "moment of £nertia"
satisfies
(6.7.9)
where Lis a preassigned positive number. Then the maximum of
 I42 ZEROS O.F ORTHOGONAL POLYNOMIALS (VI l

(6.7.IO) V(xr,xz,···,xn)= IT
P,Jt=l.2,· · ·,n
lx.-x,.l
"<I'

is attained 2j and only if the l x. l are the zeros of the Hermite polynomial H n ( e' x),
e' = (2L)- 1(n - I) 1.

The existence and uniqueness of the position of maximum is clear in both

cases. The corresponding .<:. are all different from one another; in the first
case they are positive. It is clear, furthermore, that for the maximum position
the sign of equality holds in (6.7.7) and (6.7.9). Hence, if p is a proper
"multiplier," we have

I + __I_ + ... + I + E_ = ~'

X. - Xr X. - Xz x. - Xn x. n
(6.7.11)
x.j"(x.) + (2p - ~ x.) f'(x.) = 0

in the first case, and

I I I 2p
X" - Xr
+ X, - X2
+ ... + x. - Xn
--
n
x.,
(6.7 .I2)
4
f"(x.) - P x.j'(x") = 0
n
in the second. In both cases we have written

f(x) = (x - Xr)(x - x2) · · · (x - Xn).

If we replace x by ex, withe a proper constant factor, these conditions can easily
be reduced to the first equation in (5.1.2) and to the first equation in (5.5.2),
respectively.. Therefore,

f(x) = const. L~a) (ex), a = 2p - 1, e = 2 pn-1 ,

in the first. case, while f(x) = const. Hn(e'x), e' = (2p/n)! in the second case.
The constants e and e' can be determined from the conditions (6.7 .7) and
(6.7.9), in which the equality signs now hold. vVe observe that according to
(5.1.6) the sum of the zeros of L~a)(x) is equal to n(n a); according to (5.5.4) +
the sum of the squares of the zeros of H n(x) is equal to n(n - I)/2.
Cf. also Problem 38.

6.71. Discriminants of the classical polynomia1s

The maximum problems treated in the preceding section are closely related
to the calculation of the discriminants of the classical polynomials (Hilbert 1,
Stieltjes 4, 5). The following method is due to I. Schur (2) (cf. Popoviciu 2).
(I) Let !Pn(x) l be a sequence of polynomials satisfying the recurrence
formula
[,6.71 J DISCRIMINANTS OF CLASSICAL POLYNOMIALS 143

n = 2, 3, 4, ... ;
(6.71.1)
po(x) = 1;

We suppose that ancn ':/: 0. Denoting by (x.nl the zeros of Pn(x), we show that
n n
(6.71.2) Ll n = II Pn-1 (x •n ) = ( -1) n(n-1)/2 II (a.n-2•+1 c.•-1) , n = 1, 2, 3 1 • • ••
•=1 •=1
1
Suppose n ~ 2. The coefficient of xn- in Pn-1(x) is .. a 1 a2 · · · an- 1 , so that

Using the recurrence formula, we obtain

(6.71.4)

which establishes the statement.

(2)

(6.71.5)
THEOREM

D~a,/3) =
6.71.

Tn(n-1)
The discriminants of P~a,/3)(x), L~a)(x), Hn(x) are
n
II llv-2n+2(11 + a)"-1(11 + {3)"-1(n +II+ a+ f3t-•,
-
•=1
n
(6.71.6) (a) -
Dn _ II II•-2n+2(II + a )"-1 1
•-1

(6.71.7) D n -_ 23n(n-l)/2 II II•

n

1
•-1
respectively.

We start from the familiar expression (cf., for instance, 0. Perron 4, vol. 1,
p. 259, (12), (13); p. 260, (16))

Dn(a ,/3) = J
l
ln(a ,/3) j2n-2 II (Xm- X,.n
)2
P,JJ-1,2,· · ·,n

(6.71.8) •<I'
n
= ( _ 1r(n-1)/2 (l~a,/3) j n-2 II p~a,/3)' (x.n),
•=1

where z~a ,/3) has the same meaning as in (4.21.6), and ( x.n l denotes the zeros
of p~a,/3)(x). The discriminants D~a) and Dn admit a similar representation.
According to the first formula in (4.5.7), we have
144 ZEROS OF' ORTHOGONAL POLYI\Oi\iiALS [VI]

(6.71.9) (1 - x 2) :X (P~a.~)(x)) =
2
(;n+;~~+.B.B)'p~~~)(x) if P~a.~)(x) = 0,

so that

n
' II (1 - Xvn
2 )-lp(a,{J)(
n-1 Xvn
)
•=1
(6.71.10)

n
· (P~a.~)(1)P~a.~)( -1) )-1 II P~~~)(xvn).
v=1

The last factor can be calculated by means of (6.71.2), so that on account of

(4.21.6), (4.1.1), (4.1'.4), and (4.5.1), we obtain (6.71.5).
The expression (6.71.6) can be calculated in the same way by using (5.1.14);
(5.1.8), (5.1.7), and (5.1.10); or even more simply, from (6.71.5) by using the
limiting process of (5.3.4). Indeed, if (xvn = Xvn(.B)) denotes the zeros of
p~a.~)(x) in decreasing order, we have, for fixed v and n,

(6.71.11) lim ,8(1 - Xvn) = 2~vn,

~-00

where {~vn) are the zeros of L~a)(x) in increasing order. Therefore,

D~a) = (l~a))2n-2 II (~vn _ ~l'n)2

P,JJ=l,2 1 • • • ,n
•<I'

(6.71.12)
- (l~a)J2n-2lim (,B/2f(n-1) II (xl'n- Xvn)2
{3-+oo JJ=l,2 1 •
P 1 • • 1 n
•<I'

_ (l~ a) )zn-z lim (,B/ 2t<n-1) ( z~a.~) J-zn+z D~ a,~),

~-00

which establishes the statement.

The discriminant Dn can also be obtained either directly, or from (6.71.5)
by using (5.6.3), or from (6.71.6) by using (5.6.1). The first method is the
simplest. By using (5.5.6), (5.5.10), (5.5.8), and (6.71.2), we find (6.71.7).

6.72. Distribution of the zeros of the general Jacobi polynomials

(1) Let a and .B be arbitrary real numbers, n ~ 1, and let P~a.~)(x) denote
the generalized Jacobi polynomials defined in §4.22. Then (6.71.5) still holds.
It follows from (4.1.1) or (4.21.2) that x = +1 is a zero of p~a.~l(x) if and
only if
(6.72.1) a=-1,-2, ... ,-n.
 [_6.72] ZEROS OJ.' .JACOBi POLYNOMIALS 145

(The multiplicity of this zero is I a I; cf. (4.22.2).) Similarly, x = -I is a

zero if and only if (cf. (4.1.4))
(6.72.2) fJ = -I, -2, ... , -n.
Finally, it follows from (4.21.6) that x = oo is a zero, if and only if
(6.72.3) n + a + fJ = -I, -2, ... , -n.

If such values of a and fJ are excluded, the zeros of P~a,f3>(x) are different from
±I and oo; in addition, (6.71.5) shows that they are distinct. (This follows
also from (4.2.I); cf. §6.2 (3).) Let N1, Nz, N 3 be the number of zeros in
-I < x <+I,- oo < x < -I, and +I < x < + oo, respectively. We shall
now determine these numbers as functions of a and fJ.
Hilbert .(1) calculated the number N1 N2 + +
N 3 of the real zeros. A remark
of Stieltjes (5, p. 444) indicates that he obtained the numbers N 1 , N2 , N 3 three
years before Hilbert's paper. The later results of Klein concerning the number
of the zeros of the general hypergeometric function (1, pp. 562-567) readily
lead to these numbers. (Cf. also Shibata 1, Fujiwara 1, Sen-Rangachariar 1.)
By use of Klein's .symbol

0 if u ~ 0,
(6.72.4)

l
E(u) = u [u_]

we can formulate the following theorem:

I
if u
if
> 0, u non-integral,
u = I' 2, 3, ... '

THEOREM 6.72. Let a, fJ be arbitrary real values, and set

X = X (a, fJ) = E (! (I 2n + a + fJ + I I - ( a I - i fJ I + I) J ,
(6.72.5) Y = Y(a, fJ) = E(!( -l2n + a + fJ + I I + I a I - I fJ I + I)),

Z = Z(a, fJ) = E( !(-I 2n + a + fJ + I I - I a I + I fJ I + I)).

If we exclude the cases (6.72.I), (6.72.2), and (6.72.3), the numbers of the zeros of
p~a,f3>(x) in -I< x <+I, - oo < x < -1, +I < x < + oo, respectively, are

n! a) (n! fJ) >

l
2[(X + I)/2] if (- I t ( 0,
(6.72.6) N1 = N1(a, fJ) =
2[X/2] +I if (-It ( n! a) (n! fJ) < 0,

+ I)/2] > 0,

l
2[(Y if (2n+:+fJ)(n!fJ)
(6.72.7) Nz = Nz(a, fJ) =
2(Y /2] + I if (2n+:+fJ)(n~fJ) < 0,
 14G ZEROS OF ORTHOGONAL POLYNOMIALS [VI]

2[(Z + 1)/2] if (2n +na +iS) (n ~a)> 0,

(6.72.8) Na = Na(a, {3) =
2[Z/2) + 1 if ( n
2
+1~ +iS) (n ~a) < 0.

We notice that the numbers 2f(X + 1)/2], 2[X/2] + 1 can be characterized,

respectively, as the even or odd of the numbers X and X + 1, so that N 1
is either X or X + 1. We also see that the conditions in (6.72.6) are
equivalent to p~a ~)(1)P~a.~)( -1) > 0 or < 0, respectively. For instance, if
sgn p~a.~)(1)P~a.~)( -1) = ( -1)x or ( -1)x+\ then N 1(a, !3) = X or X + 1,
respectively. Similar remarks hold for N 2 and N 8 •
I

{3

E(n + 1 +a) E(n + 1) = n

~
X
Q-
E(n + 1 + {3)
x
,(?
X
./
~
0
E(-n) = 0

E(- n- a)= 0
FIG. 7

It is sufficient to calculate N1. On account of (4.22.1) we obtain N 2 by

replacing a by -2n - o: - !3 - 1 in (6.72.6), while N 8 is obtained from N 2 by
interchanging a and jS. The following proof is based on the continuity of the
zeros as functions of a and jS.
(2) For convenience, we introduce the notation M(a, !3) for the function of
the right-hand member of (6.72.6) so that we must prove N 1 (a, {3) = M(a, !3).
If the point (a, !3) varies, the function N 1(a, {3) can change only if (a, !3) crosses
one of the straight lines (6.72.1) or (6.72.2). We first show that the function
M(a, !3) has the same property.
An easy calculation furnishes the value of X(a, !3) in the seven regions
bounded by the a- and iS-axes and by the straight line 2n + a + J3 + 1 = 0
(see figure 7). The only discontinuities of the function E(u) are at the points
[ 6.72] ZEROS OF JACOBI POLYNOMIALS 147

u = 1, 2, 3, · · · ; hence, apart from the straight lines (6.72.1) and (6.72.2), a

jump of the function M(a, {3) is possible only in the triangle where X(a, {3) =
E(n + 1 + a + {3) if n + 1 +a + {3 coincides with an integer k, 1 ~ k ~ n.
Let a = ao and {3 = f3o be negative non-integers, n + 1 + a0 + {30 = k, and
let the integers p and q be chosen such that
(6.72.9) -p < ao < -p + 1, -q < f3o < -q + 1, 1 ~ p ~ n, 1 ~ q ~ n.
Then we necessarily have k = n +2 - p - q. We investigate M(a, {3) for
n + 1 + a + {3 = k ± E, 0 <E< 1, where I a - ao I and I {3 - f3o I are suffi-
ciently small. Obviously,

sgn (-1r(n ~ a)(n ~ {3) = (-1t(-l)v-1 (-1)q-1 = (-1)k;

furthermore, X(a, {3) = E(n + 1 + a + {3) = E(k ± E) = k or k - 1, respec-

tively. In other words k = X or k = X + 1, respectively. Then (see the
remark concerning Theorem 6.72) M(a, {3) = kin both cases, so that no change
occurs in M(a, {3).
(3) We now show that when (a, {3) crosses one of the lines (6.72.1) or (6.72.2),
the jumps in N 1 (a, {3) and M(a, {3) are the same. This willprove the statement
(6.72.6), since, for a > 0, {3 > 0, we have N 1 (a, {3) = n and X(a, {3) =
M(a, {3) = n.
Because of the symmetry in a and {3, it is sufficient to consider the case
a = - k ± E, E > 0, 1 ~ k ~ n, {3 not an integer, and to discuss the location of
the zeros near x = + 1.
From (4.21.2) we obtain for a = -k + E
2kk! (n- k)! p<a,f3)( )
(6.72.10 ) (n +a+ {3 + 1) · · · (n +a+ {3 + k)(a + k + 1) · · · (a+ n) n X
1
= Co(E) + C1(E)(x - 1) + ••• + Ck-1(E)(x - 1)k- + (X - 1/
+ Ck+l(E)(x - 1)H1 + · · · +. Cn(E)(x - 1r.
Here the coefficients are real rational functions of E, regular for E = 0, and

(6.72.11) eo(O) = c1(0) = · · · = ck-1(0) = 0.

Furthermore,
1
2k{(n)}- (a + 1) (a + 2) · · · (a + k) .
co(E) = k (n +a+ {3 + 1)(n +a+ {3 + 2) · · · (n +a+ f3 + k)'
(6.72.12) ' k-1 k
{(n )}- 1 (k - 1)! .
1
co(O) = ( - ) 2 k (n + {3)(n + {3- 1) · · · (n + {3- k + 1) ¢ O.

By means of a simple consideration from the theory of analytic functions (see

below), we now obtain the following result. If ois an arbitrarily small positive
number (o < sin 1r/k fork > 1), then for sufficiently small values of E > 0, the
 148 ZEROS O.F ORTHOGONAL POLYNOMIALS [VI]

function P~a,f3)(x), a = -k +
E, has exactly k distinct zeros in the neighborhood

of x = +
1. More precisely, if 'Y/1 , TJz, TJa , • • • , 1Jk denote the numbers satis-
fying the equation
(6.72.13) sgn co'( 0) + 1J k
= 0,
these roots lie in the circles
(6.72.14) X = 1 + (E I c~(O) i) 11 k(1Jv + z), I z I < o, II = 1, 2, • • · 1 k.
By replacing E by - E (that is, for a = - k - E), the same res.ult holds with
the circles

(6.72.15) x = 1 + (E I co(O) I) Crv + z),

I 1/k I

I z I < o,
where r1 ' fz ' ••• ' rk are the roots of
(6.72.16) -sgn Co'( 0) +r k = 0.

rv
The root corresponding to a realrJv or is obviously real.
To prove the previous statement we introduce x = 1 + (E I c~(O) I l'ky in
(6.72.10) and obtain

;!
2 p

Ec~(O) + ~! c~'(o) + · · · + d")(O) + · · · + c1(E)(E I c~(O) l) 11 k y

(6.72.17) + · · · + Ck-1(E)(E I c~(O) l)<k-O/kyk-1 + E I c~(O) I yk
+ Ck.ti(E)(E I c~(O) i)(k+1)/kyk+1 + ... + Cn(E)(E I c~(O) i)n/kyn = 0.

If this be divided through by E I c~(O) I , the "principal terms" are sgn ~(0)
+ yk. Now we can apply Theorem 1.91.2 (Rouche's theorem); whence the state-
ment follows immediately.
We are interested especially in the number of real zeros x < 1 near x = 1. + +
From the preceding result we see that this number increases or decreases by
one unit if we replace E by -E, according as ( -1)kc~(O) is positive or negative.
With reference to (6.72.12) this is equivalent to the condition that

(ntJS)<o or >O,
respectively.
(4) On the other hand, we discuss the jump of M(a, J3) as (a, iS) crosses
the line a = -k with !3 non-integral. First suppose J3 > 0, so that X(a, !3)
= E(n + +
1 a). For a = ~k ± E, E > 0, we have

(6.72.18) X(a, J3) = E(n +1- k ± E) =

n +i - k,

\ n - lc,
respectively, and
 [ 6.72 J ZEROS OF .JACOBI POLYNOMIALS I49

' (6.72.I9) ~n- ( I) n + (n a) ·(n +,B)

n n
=s~-
( I}n (n +a) = 1(
n
-l)n+k-\
( -I)n+k
1

respectively, which in both cases is (-I)x(a.~). Hence (cf. the remark to

Theorem 6.72) M(a, .B) changes from n + I - k ton - k, a loss of one unit.
In this case

Now let .B be negative and non-integral. Then n + I + a + .B is non-

integral near a = -k, so that X(a, {3) remains constant in this neighborhood.
More precisely,

(6.72.20) X(a, ,B) = n +I- k + (.8] if n +I- k + .B > 0,

1 0 if n +I - k + .B < 0.
In the first case we have for a = - k ± E, E > 01

(6.72.2I) ( If n
)
( ( )= 1
++a
.B ( n
-If( -I)k-I( -I)r~l+I
1

sgn - n n (-If( -I)k( -1)[~1+\

respectively, which are (-I)x<a.~HI and (-I)x(a,~), respectively. Hence (cf.
the remark to Theorem 6.72), M(a, .B) changes from X(a, .B) + I to X(a, .B),
a loss of one unit. In this case, again

Let us now consider the second case in (6.72.20). Then

(6.72.22) = sgn (-If ( n ~ a) ( n t .B) ~i (.B + I)(.B + 2) · · · (.B + n - k)

= -sgn ( n +.B)
k or sgn ( n +.B)
k 1

respectively, for a= -k ± E, E > 0. For instance, if

M(a, ,B) changes from 0 to I, a gain of one unit. The opposite is true if

This establishes Theorem 6. 'Z2.

I 50 ZEROS O.F ORTHOGONAL POLYNOl\HALS [VI]

6. 73. Distribution of the zeros of the general Laguerre polynomials

The discriminant formula (6.71.6) is valid for general Laguerre polynomials
L~a)(x), a arbitrary and real, n ~ 1. On account of (5.1.7), x = 0 is a zero of
L~a)(x) when and only when

(6.73.I) a= -I, -2, ·. · , -n.

(Its multiplicity is I a I; cf. (5.2.I).) If such values of a are excluded, the zeros
of L~a)(x) are finite and different from 0; in addition, we conclude from (6.71.6)
(or from the first differential equation (5.1.2)) that they are distinct. Let
n1(a) and nz(a) denote the number of positive and negative zeros, respectively.
By using Theorem 1.91.3 (Hurwitz's theorem) and (5.3.4), we see that if {3 is
large, p~a,/3)(x) has at least n 1(a) zeros in [-I, +I], at least nz(a) zeros in
[+I, +oo], and at least n - n1(a)- nz(a) zeros which are not real. There-
fore, using the notation of the prev:ious section, we see that n 1(a) = N1(a, {3)
and nz(a) = N 8 (a, {3) if {3 is large; that is,
(6.73.2) n1(a) = lim N1(a, {3); nz(a) = lim Na(a, {3).
B-+~ !3-+~

Obviously, if a > -I, then n 1(a) = n, and nz(a) = 0.

Now suppose a < -I, a ¢ -2; -3, . ·. , -n. From (6.72.5) we obtain
n + [a] + I if a > - n,
lim X(a, {3) = E(n + a+ I) = {
P-+~ 0 if a < -n,
since the argument of E in the formula for X(a, {3) is not a positive integer;
furthermore,
lim Z(a, {3) = E( -n) = 0.
!3-+~

Now

n n +a _
( -It+[al+I if a > -n
'
sgn ( - I) ( ) - {
n I if a < -n.
Thus, N 1(a, {3) =n + [a] + I in the first case, and N1(a, {3) = 0 in the second.
Therefore,
n + [a] + I if a> -n,
(6.73.3) n1(a) =
{ 0 if a < -n.
Furthermore,
(6.73.4) n~(a) = 0 or I,
according as
 [6.8) THEOREM OF HEINE-STIELTJES 151

THEOREM 6.73. Let a be an arbitrary real number, a¢ -1, -2, ... , -n.
' The number of the positive zeros of L~a)(x) is n if a> -1; it is n + [a]+ 1 if
-n < a < -1; it is 0 if a < -n. The number of the negative zeros is 0 or 1.
This result can also be obtained by a direct method similar to that used in
§6.72. In fact, the numbers n 1 (a), nz(a) can change only if a passes one of the
integers -1, -2, · · · , - n. If a decreases through an odd value of this kind,
a positive zero is lost and a negative zero gained. If the passed value is even,
a positive and a negative zero are lost. For a < - n, there are no real zeros if n
is even and one real zero (which is negative) if n is odd. (Compare Lawton 1,
W. Hahn 1.)

6.8. Polynomialb which satisfy a second:..order linear homogeneous differential

equation with polynomial coefficients; theorem of Heine-Stieltjes
Heine (3, vol. 1, pp. 472-479) has studied the following problem:
PROBLEM. Let A(x) and B(x) be given polynomials of degrees p +
1 and p,
respectively. To determine a polynomial C(x) of degree p - 1 such that the
differential equat1:on
2
d d
(6.8.1) A (x) dx~ + 2B(x) d; + C(x)y = 0

has a solution which is a polynomial of a preassigned degree n.

Heine asserts that, in general, there are exactly

(6.8.2) <T = <Tnp = (n +~ - 1)

determinations of C(x) of this kind.
The hypergeometric equations (4.2.1) and (4.21.1) are of this type with
p = 1. Lame functions satisfy an equation of the same type with p ~ 2.
These cases were the starting points of Heine's investigations on this subject.
Stieltjes (3) discusses only a special case of (6.8.1) which, however, is of
primary importance. He obtains the following result:
THEOREM 6.8. Let A (x) and B(x) be given polynomials of precise degree
p + 1 and p, respectively, and let the highest coefficients of A(x) and B(x) have the
same sign. If the zeros of A (x) and B(x) are real, distinct, and alternating with
one another, there are exactly <r polynomials C(x) of degree p - 1 such that the
differential equation (6.8.1) has a solution which is a polynomial of degree n.
Here <r has the meaning (6.8.2).
The proof of Stieltjes uses a part of Heine's assertion, namely, that <r is an
·upper bound for the number of polynomials C(x) in question. He obtains,
however, not only the existence but also a characterization of the <r solutions
mentioned, in the following way: Then zeros of these solutions are distributed
152 ZEHOS OF ORTHOGONAL POLYI\OMIALS [VI l

in all possible ways in the p intervals defined by the p +

1 zeros of A (x). (The
number of such distributions is obviously a-.) The solutions are obtained by
means of a maximum problem similar to those treated in §6. 7.
The following proof of Stieltjes' theorem uses the latter's idea based on a
maximum problem, but Heine's elimination process (which furnishes the upper
bound a-) is replaced by certain elementary considerations related to Sturm's
theorem (see §1.82). Our proof is, consequently, independent of Heine's work.

6.81. Preliminary remarks

We assume with Stieltjes that
(6.81.1) A (x) = (x - ao)(x - a1) · · · (x - ap), ao < a1 < ··· < ap,
and

(6.81.2) BA((x)) = ~ + x -PI a1 + ... + x -Pp ap ' p. > O,v = o; 1,2, ... ,p.
x x - ao
This is equivalent to the assumption that the zeros of A (x) alternate with
those of B(x) and that the highest coefficients of A (x) and B(x) have the
same sign.
Let C(x) be a given polynomial. Then (6.8.1) cannot have two polynomia1
solutions y and z linearly independent of each other. Otherwise, we should
have for x r!: a.
A(x)(y'z - yz')' + 2B(x)(y'z - yz') = 0,
(6.81.3)
y'z - yz' = const. exp {- J~(~; dx} = const. gI x - a.i-
2
P'.

Since the last product approaches oo as x ~ a., this leads to a contradiction

unless y' z - yz' = 0.
Now let y be a polynomial solution, y ¢ 0. We show that y r!: 0 at x = a•.
If the contrary were true, substitution of x = a. in (6.8.1) would give y' = 0.
Differentiation of (6.8.1) k times results in a differential equation of order
k + 2 in y, which -has the form
A(x)y<k+ 2>+ {kA'(x) + 2B(x) Jy<k+I> + · · · = 0;

the further coefficients are again polynomials. Now from (6.81.2) we see that
B(a.) = p,A'(a.), so that kA'(a.) +
2B(a.) r!: 0. Thus y = y' = y" = · · · =
y<k> = 0 would imply y<k+I> = 0. In a similar way we can show that all the
zeros of y are distinct.
Next we prove that the zeros of y lie in the interval [ao, ap]. Upon writing
y = j(x) = const. (x - x1)(x - x2) . · · (x - xn), we have, according to (6.8.1),

(6.81.4)

or, using (6.2.2) and (6.81.2),

[ 6.82] A MAXIMUM PROBLEM 153

(6.81.5)
+ Xk -
1
Xn
+ t•-o p.
Xk - a,
= 0.

'Ve assume that some of the zeros of f(x) lie outside of [a 0 , ap]. Let Xk be one
of these zeros such that the segment [ao, avl and the remaining zeros all lie in a
closed half-plane with Xk as a boundary point, the segment itself lying in the
open interior of the half-plane. :rhen the complex vectors

Xk- Xm, Xk- a.

lie in an angle not greater than 1r for all values of v and m, m ~ k, and the same
is true of the reciprocals of these vectors. The vectors Xk - a, are directed into
the interior of this angle. But then we see that (6.81.5) is a contradiction.
(For this argument, see P6lya 1.)
Let n1 , n2 , · · · , nv be the number of zeros of y in [a 0 , ad, [a1 , a2], · · · ,
[av-I, av], respectively. Then we say that y is of the type ln1, n2, · · · , nv}·
Here n1 + n2 + na + · · · + nv = n, and (J represents the number of all possible
types. It is our intention to show the existence of exactly one polynomial
solution of each type, corresponding to (J different determinations of the poly-
nomial C(x) of degree p - 1.
6.82. A maximum problem
Following Stieltjes (loc. cit.), we first show that a polynomial solution of each
type exists.
Let x1 , x2 , · · · , Xn be variable points, each different from the a, and dis-
tributed in [ao , ap] so that in each interval [a,_1 , a.] there lies a certain pre-
assigned number, say n., of these points, L~=I n. = n. Letting each Xk vary
in a fixed interval, we c·onsider the maximum of the product

(6.82.1)
W = II IxK
"=1,2,· · · ,n
- a. II · · I x"
IP• X,p.=-1,2,· ,n
- xl' I·
·-0,1,·. ·,p "<~'

The existence and positiveness of this maximum are clear. In the maximum
position, the points Xk are different from the a. and from one·another, and they
are of a preassigned type. Furthermore, we have aWjaxk = 0; whence (6.81.5)
again follows. If f(x) denotes the polynomial w.ith the zeros Xk, the latter
equation means that A (x)f"(x) + 2B(x)f'(x) vanishes for x = xk and hence is
divisible by f(x). If this ratio is denoted by -C(x), (6.8.1) follows. It is
clear that log (W-1), apart from the constant terms P~<P• log I a~< - a. ~-~ , 11- ~ v,
is the "energy" of the system of masses p, concentrated at a. and of unit masses
concentrated at the Xk. (Cf. §6.7 (1).)
Incidentally, the argument used in §6.7 (1) shows that the system (xd with
the maximum property is uniquely deterrp.ined. This is not the same as the
154 ZEROS OF ORTHOGONAL POLYNOMIALS [VI)

uniqueness of the solution of a given type since (6.81.5) is not equivalent to the
maximum property. It is, however, easy to show (cf. (6.22.6)) that (6.81.5)
is equivalent to a relative maximum of W.

6.83. Uniqueness
Let C(x), y and D(x), z be two solutions of our problem of the same type,
C(x) ~ D(x). Suppose both polynomials y and z have positive highest coeffi-
cients. By combining (6.8.1) with the corresponding equation for z, we find
the relation

(6.83.1) d (y'z - yz') + 2 i: p, (y'z - yz') + C(x) - D(x) yz = 0.

dx .-o x - a. A (x)
Introducing H = II~-o I x - a. I2P', we obtain for x ~ a.

(6.83.2) d (H( , ') l _ D(x) - C(x) H

dx y z - yz - A(x) yz ·

Suppose the function (D(x)- C(x)JIA(x) is non-negative in the fixed interval

a,_1 < x < a.. Then between two consecutive zeros a and (3 of y, a < (3, in this
interval, z must change its sign at least once. Otherwise, yz would be perma-
nently positive or negative, and therefore, H(y'z - yz') increasing or decreasing
in a < x < (3. For x = a + E, E > 0, however, the last expression has the
same sign as y'z or yz, and for x = (3 - E, E > 0, the same sign as y'z or -yz.
Hence it passes from positive to negative values if yz > 0, and from negative
to positive values if yz < 0. Either case contradicts the property of monotony
of H(y'z - yz') mentioned before.
Under the previous assumption concerning ID(x) - C(x)} I A (x), the function
z must change its sign also in [a•-1, 'Y] and in [o, a.] if 'Y and o are, respectively,
the first and last zeros of y in [a.-1, a.]. Otherwise, y and z would each have
a constant sign in those intervals; moreover, y and z would have the same
sign in a given interval because they are of the same type. Then H(y'z - yz')
would be an increasing function. It vanishes for x ~ a,_ 1 + 0 and also for x ~
a. - 0, so that y'z - yz' must be positive in [a,_11 'Y] and negative in [o, a.].
Now for x =/',we have sgn (y'z- yz') = sgn y'z, which is the same as sgn (y'y)
at x = 'Y - E, E > 0, that is, negative. This is a contradiction. The same argu-
ment can be used at x = o.
Finally, we remark, under the assumption made about (D(x) - C(x) JI A (x),
that the polynomial z must vanish in [a,_1 , a.], even if y has no zeros there.
Otherwise, because of (6.83.2), the function H(y'z - yz') would be monotonic.
This is impossible since H vanishes for x = a•-1 and for x = a• .
. To recapitulate: this argument would furnish at least one more zero for z
in [a,_1 , a.] than for y, which is impossible. Therefore, (D(x) - C(x)} I A (x)
must be negative in some points of a•-1 < x < a.. By interchanging C(x), y
with D(x), z, it is seen that the same function must also be positive somew~ere
in a,_ 1 < x < a. ; whence D(x) - C(x) must have at least one variation of sign
[ 6.9) LEGEXDRE FVXCTIO:'\S OF SECOND KIND 155

in [a.-1, a.]. Since this is true for v = I, 2, ... , p, the function D(x) - C(x)
has at least p variations of sign in [ao, ap]. However, this is impossible since
D(x) - C(x) is of degree p - 1.

6.9. Zeros of Legendre functions of the second kind; generalization

(I) In connection with Fejer's second generalization of Legendre poly-
nomials (§6.5 (4)) we consider the function

Hn(cos 8) = f3o cos (n + I)8 + (31 cos (n + 3)8

(6.9.I)
+ · · · + f3m cos (n + 2m + 1)8 + · · ..
Here f3m l 0, so that the series converges for 0 < 8 < 1r. This is the conjugate
series of the function Gn (cos 8) defined by (6.5.I6). The function Qn(cos 8)
of (4.9.I6) is a special case; then the sequence f3m is given by· (4.9.5) and is
completely monotonic. Now we prove the following theorem:
THEOREM 6.9.1. Let f3m > 0 and let lf3m} be a completely monotonic sequence.
The function Hn (cos 8) defined by (6.9.I) has at least one zero in each of the intervals

(6.9.2) v = 0, I, 2, · · · , n.

More precisely, Hn (cos 8) has an odd number of zeros in each of these inter-
vals. For the proof we again use Hausdorff's representation and obtain for
(6.9.I)

(6.9.3)
1{%;
1

0
tm cos (n +2m + I)()} d(3(t)

= ecos
}o
(n + I)8 - t cos (n - 1)8 d[3(t),
I - 2t cos 28 + t2
where (3(t) is a function of the same type as a(t) in (6.5.8). Upon substituting

v v +!
8 = - -1 1r and 8 =n + .1. 1r,
n+2 2

we find
(-I)"(~os (8/2)- t cos (38/2)),
(6.9.4) cos (n + I)8 - t cos (n - I)8 = { +
( -I)"+I{sin ( / ) t sin ( / )},
82 38 2

respectively. Both expressions in the braces are positive; this establishes the
statement. (For v = 0 and v = n we must take into account that
lim Hn(cos8) = (-Ir+I lim Hn(cos8) > 0
8-++0 8-+1!"-0

(+ oc if L::...o f3m is divergent).)

156 ZEROS OF ORTHOGONAL POLYNOMIALS [VI)

(2) THEOREM 6.9.2. The function Qn(cos 8) (§4.62 (3)) has exactly n + 1
zeros in 0 < () < 1r, which lie in the intervals (6.9.2).
In this special case of Theorem 6. 9.1 there cannot be more than one zero
in each of the intervals (6.9.2); otherwise, Pn(cos e) would have more than n
zeros by Sturm's theorem.
Inequalities (6.9.2) for the zeros of Qn(cos e) are due to Stieltjes (8, p. 252),
whose proof is, however, different from that just given. Fejer obtains (20, pp.
51-52) less precise inequalities in a manner similar to that used above; his
conditions concerning the sequence I.Bm l, ho\vever, are more· general.

(3) THEOREM 6.9.3. Legendre's function of the second kind Qn(x) (§4.61 (I))
has no zeros in the complex plane cut along the segment [ -1, + 1], except x = oo ,
which is a zero of multiplicity n + 1.
This theorem is due to Hermite (3) and Stieltjes (9) (cf. also Hermite-
Stieltjes 1, vol. 2, pp. 80-104, no. 267-274). The following argument is a slight
modification of the second proof of Stieltjes.

- 1
FIG. 8

We start from (4.62.10). This function Qn(x) has n + 1 zeros in the interior
of [-1, +I], which alternate with the zeros of Pn(x) (cf. Theorem 6.9.2). Let
2
+I = Xo > X1 > · · · > Xn > Xn+I = -1 denote the zeros of (1 - x )Pn(x) in
decreasing order. Then

(6.9.5) sgn Qn(x.) = ( -lr, 11 = 0, 1, 2, ... , n + 1.

Furthermore, Qn(x) is a solution of (4.2.1), a = .B = 0, so that Qn(x)/Pn(x)
is increasing in the interval -1 ;£ x ;£ +I (cf. (4.2.6)); it becomes infinite at
each x•.
The curve in the figure encircles the points x = ± 1 and avoids the zeros of
P n(x) by means of semi-circles. Now we shall show that the variation of
arg (Qn(x)/Pn(x) J along nhis curve is equal to 21r(2n + 1). Then, according
to Theorem 1.91.1 (principle of argument), the function Qn(x)/I>_n(x) has
exactly 2n + 1 zeros exterior to this curve. But since it has a zero of multi-
plicity 2n + 1 at x = oo, the statement will follow.
Let E be a sufficiently small positive number. As x encircles + 1 in the nega-
tive (clockwise) direction from 1 + E to 1 - E, the variation of the argument in
question is approximately (cf. (4.62.7)) the variation of arg (log 1/(x - 1) l,
a quantity which tends to 0 with E. If x describes the segment from x. - E
to x.+I + E, 11 = 0, 1, 2, ... , n, along the "lower border" of [ -1, + 1], we have
[ 6.10 J FURTHER RESULTS 157

(6.9.6) Y = Qn(X - iO) = i1r + Qn(X)

P n(x) 2 P n(x)"
Then y describes a straight line 3Y = 1r/2 in the direction of decreasing abscissas.
The variation of the argument of y is +1r.
In the neighborhood of x., v = 1, 2, · .. , n, the function Qn(x)/Pn(x) differs
only by a bounded term from
Qn(x.) 1 h Qn(x.)
(6.9.7) - ,-) , w ere - ,- < O.
Pn(x. X- Xv Pn(x.)
Therefore, the half-circle in the lower half-plane around x. will be carried over
into a curve which approximates a large semi-circle in the lower half-plane; the
argument of '!j then increases by +1r.
Finally, if x = -1 is encircled. in the negative sense, from -1 + E to -1 - E,
the variation in arg y is again a quantity which tends to zero with E.
To recapitulate: While x moves along the lower border from 1 + E to -1 - E,
the total increase in the argument of y i-s (n + 1)7r + n1r = (2n + 1)1r. The
same is true on the upper border as x varies from -1 - E to 1 + E. This es-
tablishes the statement.
6.10. Further results
(1) Let (O.nl denote the zeros of Pn(cos e) in the interval [0, 1r], ordered in
un increasing way. Turan (1) proved that the sequence x.n - x •. n-1, where
x.n =cos e.n, is increasing as v runs from 1 to [!(n- 1)}. S~ego proved (in a
correspondence with Turan, 1946) the same fact for the differences e•. n-1 - e.n.
Cf. Szego-Turan I.
(2) Concerning the topics treated in §§6.8-6.83, see also Makai 3.
(3) The argument of §6.9 (3) leads to a more general result concerning the
number of the zeros of Qn(x) - aPn(x) in the complex plane cut along the
8egment [ -1, + 1} where a is a given complex constant (Hermite and Stieltjes,
Joe. cit.). This number is again = 2n + 1 if -1r/2 <Sa < +1r/2, and =n if
Sa f;; +1r/2 or Sa ~ -1rj2.
Indeed, in the case -1r/2 < 3a < +7r/2 no essential change in the reasoning
is needed. Now let Sa > +1r/2. If x describes the segment from x. - E to
x.+ 1 + E, v = 0, 1, 2, · · ·, n, along the lower border of [ -1, + 1], the variation
of the argument of y - a is -1r. There is no change in the contribution of the
semi-circles around x. and of the whole upper border. Thus the total increase
in the argument of y - a is - (n + 1 )1r + n1r + (n + 1 )1r + n1r = 2n1r.
Let Sa = +1r/2. On the segment from x. - E to x•+1 + E there is a unique
point~ such that Qn(~)/Pn(~) =~a. We must use an indentation of the contour
into the lower half-plane and take into account that for 8 > 0
Qn(x - iO)) = (Qn(X - iO)) _ io f.!!_ Qn(X - iO)} +
( Pn(X) z=t-ia Pn(X) x=t [dx . Pn(X) z=t
158 ZEROS OF ORTHOGONAL POLYNO~JIALS [VI)

the imaginary part of the right-hand side is 1r 12 - 8' < 1r /2 where 8' > 0. The
argument is similar if ~a = -1r /2.
( 4) Makai-Turan (I) have proved the following theorem of the Picard-
Landau type. Let Hn(z) be Hermite's polynomial. There exists a positive
(absolute) constant A such that every equation of the form
H 0 (z) + HI(z) + 'Y Hn(z) = 0
has a solution in the strip I 3 z I ~ A; n ~ 2, 'Y arbitrary real or complex.
The exact value of A has been determined by Schmeisser I. The extremal
polynomials are of third degree. The corresponding problems for quadrinomials
and more general equations remain open.
(5) Szego 20 improved the right-hand estimates in Theorem 6.4 when
2ao -a1 > a1 -a2 ~ a2 -a3 ~ · · · ~ an-1 -an~ an~ 0. Both sides can be im-
proved when (2k -l)ak-1 ~ 2kak > 0, k ~ I. See Askey-Steinig I.
(6) The results mentioned in §6.10 (I) have been extended to ultraspherical
polynomials in Szego-Turan I.
(7) For the positive 8-zeros of the Legendre polynomials Pn(cos8) and
also for the positive z~ros of the Hermite and Laguerre polynomials, written
in increasing order, the second differences of the respective sequences of con-
secutive zeros are all positive, as an immediate consequence of the Sturm
theory (cf. Theorem 6.3.3). In L.Lorch-P.Szego I, 2, it is conjectured that
all higher differences are also positive, but this remains unresolved. Sub-
stantiating numerical evidence is cited there and in Davis-Rabinowitz I. The
latter present also similar evidence connected with P~ (cos 8). See also Lorch-
Muldoon-P.Szego I, 2.
(8) If 7rn(x) is an arbitrary polynomial of degree less than or equal to n
and if

is its Laguerre expansion, then the number of sign changes of 7rn(x) for x > 0
is at least as great as the number of sign changes of the sequence of the differences
b0 , (bo- b1), (bo- 2bl +b2), · · ·.
See Turan 2.
 CHAPTER VII

INEQUALITIES
No inequalities, except trivial ones, are known for general orthogonal poly-
nomials. However, inequalities involving an unspecified constant can easily
be derived under certain conditions concerning the weight function w(x).
Still more precise estimates can be obtained if w(x) is monotonic, and a great
number of special inequalities follow from this added restriction.
Another very extensive class of inequalities can be derived for the classical
orthogonal polynomials, and in the present chapter we intend to enumerate and
compare the various methods used to obtain these inequalities. Aside from inte-
gral and series representations, the main tool is differential equations. As regards
the latter, we remark that there is a special method for deriving inequalities for
the solutions of certain differential equations (cf. Theorem 7.31.1). In recent
years this method has been used i'n several special problems (not only for po'ry-
nomials), with slight variations, primarily by G. N. Watson and S. Bernstein;
it originated, however, in an idea of Sonin. 41
At the end of this chapter we use the above mentioned inequalities in dealing
with certain extremum problems which involve polynomials of a fixed degree.
The selection of the material treated in this chapter has been influenced by the
needs of later chapters, especially by those of Chapters IX, XIV, and XV.
Historically, the major part of the inequalities for classical polynomials arose
from the discussion of the corresponding expansion problems.
We shall postpone till Chapter VIII the asymptotic l')alculation of certain
maxima (which can also be expressed in terms of inequalities), since they
require more intricate asymptotic consideratidns. However, we have found it
necessary in the present chapter to anticipate certain asymptotic results of
Chapter VIII.
7.1. Rough bounds for orthogonal polynomials
In this section we make essential use of the representation of positive func-
tions discussed in §10.2. However, this does not play a role in the further
course of Chapter VII.
(1) Let w(x) be a weight function on the interval [-1, +I] for which the
integral

(7.1.1) 1-: 1
2
(1 - x )-! Jog w(x) dx

exists in Lebesgue's sense. (This implies that w(x) cannot vanish on a whole
u Cf. Sonin 2, pp. 23-24. I owe this reference to Professor J. A. Shohat.
159
 I60 1.~\EQUALITIES [VII]

segment.) Let D(j; z) = D(z) be the analytic function associated with f(8) =
w(eos 8) I sin 8 I in the sense of §I0.2 (2).
The conformal representation x = Hz + z- 1) maps the unit circle I z I < I
(or I z I > I) onto the x-plane cut along the segment -I ~. x ~ + 1. For
z = eill we have x = cos 8. (Cf. §1.9.)
THEOREM 7 .1.1. Let IPn (x) l be the orthonormal set of polynomials associated
with a weight function w(x), -I ~ x ~ +I, for which (7.1.I) exists; then
(7.1.2) I 1rJD(z)pn(x)zn I < (I - I z 12 )-!, Iz I < I,
where x = Hz + z- 1
) is an arbitrary point of the cut plane.
For, (cf. (I0.2.9))
+1
I =
1_
1
2
IPn(x) \ w(x) dx·

(7.1.3) I1+1r·
= 2 _,.. IPn(cos 8)) 2 w(cos8)1 sin 8 I d8

- r~~o ~ 1:""1 Pnf!(z + 1

Z- )
2
}zn j I D(i)
2
1 d8, z = re i8

Now if f(z) = L::-o CmZm is regular in I z I < I, we have, according to Cauchy's

inequality,

(7 .1.4)

This establishes the statement.

The bound in (7.1.2) becomes infinite if x lies on the segment [-I, +I};
for all other values of x it furnishes a first appraisal of the magnitude of Pn(x)
under a rather general condition. This information is comparatively precise
because we shall prove (cf. Theorem I2.1.2) that, for a fixed x, the left-hand
member of (7.1.2) 'tends to 2-! as n ~ oo. It is rather remarkable that no use of
the orthogonal property has been made in deriving (7.1.2); only the normaliza-
tion of Pn(x) is employed.
(2) THEOREM Let w(x) be bounded from zero, that is, w(x) ~
7.1.2. 11- > 0.
Then, if x is not on the segment [-I, +I],

(7.1.5)
where (x - I)! is chosen so that I x + (x - I)! I > 1. The constant A depends
2 2

on x and 11- but not on n; A is uniformly bounded in the exterior of any closed curve
which contains [-I, +I] in its interior.

·we have in this case I D(z) I > I 11-(I - i)/2 I! (cf. (I0.2.IO)), so that from
(7.1.2)

--------------------------------· - - - - - - · - - · · · .
[ 7.1 l BOVXDS FOR ORTHOGOXAL POLYXO~IIALS I6I

(7.1.6) I z I <I,
follows.
(3) In the same case w(x) ~ 11- > 0, we can also easily obtain bounds for
Pn(x) on the orthogonality interval -I ~ x ~ +I itself. In fact, if I z I < I,
we have, in view of (7.1.6), (7r~J-/2)t(l - I z 1)! I Pn(x)zn I < (I - I z 12)-t. Let
2

x be on the segment [-I, +I], x = Hz+ z- 1), where I z I = 1. Then Pn(x)zn

is a 1r2n in z, and we have for -I ~ x ~ + I, I z I = I, r < I,
I Pn(x) I = I Pn(x)zn I < max I Pn(~)(' I
lll=r-1

max ! Pn(~)rn I < r- 2 n(7r~J-/2)-t(I - r2)-1•

2
= r- n
l!!=r

Here~= Hr + r- On putting r = I - I/n, for n ~ 2, -I ~X~ +I,

1 2
).

we have
(7.1. 7)
For -I < x < +I, we can reduce the exponent in (7.1.7) from I to ! (cf.
(7. 7I.28)).
(4) The same elementary method gives an idea of the magnitude of Jacobi's
polynomial if n is large. In this case w(x) = (I - x)a(I + x)fl, a > -I,
{3 > - I and (cf. (I0.2.I3))

(7.1.8) D(z) = 2-<a+fJ+l)! 2 (I - z)aH(I + z)fJH.

Then by (7.1.2)
7rt2-<a+{J+I)/ 2 I I - Z jaH ! 1 + Z ifJH I Pn(x)zn I < (I - I Z 12)-t,
(7.1.9)
x = Hz+z- 1),1zl < 1.
Now assume -I ~ x ~ +I; we obtain, as in (3),
(7.l.IO) I Pn(x) I < cr-2"(I - r 2 )-~ max II - r 1-a-t II + r 1-fl-!,
i!l=r
where C depends only on a and {3, and 0 < r < I. We choose again r 2 = I - I/ n.
Discussing the right-hand member of (7.l.IO) for I r I = r, ·m(r) ~ 0, and for
I r I == r, 91(t) ~ o, we obtain
(7.1.11) I Pn(x) I < C'nmax(a+l,{J+l, t>, -I~ X~+ 1.
Here C' depends only on a and {3. The "true" exponent is max(a + !, {3 + !, 0)
(cf. (7.32.2)).
For later purposes we give a formulation of (7.1.9) in terms of the Jacobi
polynomials p~a,fJ>(x) (cf. (4.3.4)). If xis exterior to the segment [-I, +I],
we have
(7.1.I2)
a>- I,{3 > -I,n--4 oo.
 IG2 INEQUALITIES [VII]

This holds uniformly in the exterior of any curve containing the segment
[ -· 1, +I] in its interior. Inequality (7.1.12) follows, of course, immediately
from the asymptotic formula (8.21.9).
(5) The following theorem is rather useful in obtaining bounds for ortho-
normal polynomials:
THEOREM 7.1.3. Let w(x) and w(x) be two weight functions on the segment
[-1, + 1], w(x)/w(x) = k(x). Assume k(x) ~ k > 0, and let k(x) satisfy the
Lipschitz c.ondition
(7.1.13)
If IPn(x)} and IPn(x)} are the orthonormal polynomials associated with w(x)
and w(x), respectively, we have
(7.1.14)

This theorem is due to Korous (3). The proof follows from the identity
( (3.2.3))

Pn(x) = 1: 1

Pn(t) {~ p.(x)p.(t) }w(t) dt

k
= fc: Pn(x) + 1+1
_1 Pn(t) {n-1
~ p.(x)p.(t) } w(t) ( 1-- k(x)
k(i))
dt

·w(t) k(x~ = :c(t) dt,

where kn has the same meaning as in (2.2.15), and kn has the corresponding
meaning for Pn(x). Now, according to Schwarz's inequality

fc: _1 p,(t)pn(t)w(t) dt ~ {1+1

k = 1+1 _ [pn(t)] 2 w(t) dt {1+1
1
}!
_ [jj,(t)] 2 w(t) dt
1
}!
1
= {/_: [k(t)r1w(t)[pn(t)] 2 k-l, dty ~
k~-l
~
= 1+1 tPn-l(t)pn(t)w(t) dt ~ 1+1 I Pn-1(t) I IPn(t) I w(t) dt ~ 1,
-1 ~

11~' p.(t)p._,(t)W(t) at I ~ {1~' lp.(t))'w(t) at}' ~ k -•,

which establishes (7.1.14).
We mention two important special· cases which follow immediately from
(7.1.14) by use of the bounds of the Legendre and Tchebichef polynomials (con-
cerning the fir;;t case, see (7.21.1) and (7.3.8)):
(a) If w(.-r) iR poRitin and satisfies a Lipschitz condition I w(xl) --- wCx2) I
< Xl.rl- x2l,wc han~
[7.2] :\10NOTONIC WEIGHT FUNCTIONS 163

(7.1.15)
-1<x<+l.
Here the positive constants A and A' are independent of x and n.
2
(b) If w(x) = (1 - x )-! k(x) where k(x) is positive and satisfies a Lipschitz
condition I k(xi) - kCx2) I < XI XI - x2l, we have
(7.1.16) I Pn(X) I <A, -1<x<+1,
where A is independent of x and n.
For other elementary considerations of a similar nature, see Shohat 4, pp. 165-
166 and Jackson 6, pp. 893-898. See, also, §7.71 (6).

7.2. Monotonic weight functions

THEOREM 7.2. Let w(x) be a weight function which is non-decreasing in the
interval [a, b], b finite. If (pn(x)) is the set of the corresponding orthogonal poly-
nomials, the functions Iw(x))! I Pn(x) I attain their maximum in [a, b] for x = b.
See Szego 3. A corresponding statem·ent holds for any subinterval [xo , b]
of [a, b] where w(x) is non-decreasing.
The proof is based on the identity

w(b)lpn(b)) 2 - w(x)lpn(x)) 2 = 2 [b w(t)pn(t)p~(t) dt + [b 1Pn(t)) 2

dw(t),

which follows from (1.4.4). It suffices to show that this expression is non-
negative in a ~ x ~ b. Denoting by x1 < X2 < : · · < Xn the zeros of Pn(x)
in increasing order, we have Pn(t)p: (t) > 0 for t > Xn and Pn(t)p~(t) < 0 for
t < XI • Therefore, the statement is trivial for Xn ~ x ~ b and follows from

1b w(t)pn(t)p:(t) dt = 1b w(t)pn(t)p~(t) dt - 1"' w(t)pn(t)p~(t) dt

= - 1"' w(t)pn(t)p~ (t) dt
for a ~ x ~ x1 • Here we used the monotony of w(x) only in x ~ t ~ b.
Now let x. ~ x ~ x.+l, 11 = 1, 2, . ·. , n - 1, n ~ 2. If we introduce the
new weight function
2
W(x) = w(x) I (x - xi)(x - x2) · ·. (x - x.) ) ,
the corresponding orthogonal polynomial of degree n - 11 will be, save for a
positive constant factor,
Pn(X)
Qn-•(x) = (X -
)
X1 (X - X2) • • • (x - X.
),

with the zeros x.+ 1 , x.+2, · · · , Xn. In fact, if p(x) is an arbitrary 11"n-•-I,
164 INEQUALITIES [VII]

1b W(x)qn_.(x)p(x) dx = 1b w(x)pn(x)(x- x1)(x - x2) · · · (x- x.)p(x) dx = 0.

But W(x) increases monotonically for x. ~ x ~ b, so that the preceding argu-

ment furnishes, for x. ~ x ~ x., 1 ,
2
w(b)lpn(b) 12 - w(x)lpn(X) 12 ~ 0.
2
W(b) /qn_.(b) 1 - W(x)lqn_.(x) 1 =

In addition we conclude that the equation w(b)lpn(b) 12 = w(O IPn(O 12 holds

if and only if ~ < xi and w(t) vanishes in (a, ~] (this condition has, of course,
no signifieance if a = ~),and is step-wise constant in[~, b]. Furthermore, Pn(t)
must vanish at the points of increase of w(t). (The weight function w(x) cannot
vanish for x ~ xi .)

7.21. Applications
On putting a = -1, b = +1, and w(x) = 1, we obtain the important in-
equality
(7.21.1) I P n(X) I ~ 1,
for the Legendre polynomials Pn(x), Pn(l) = 1. If n > 0, the equality sign
holds only for x = ± 1.
Another interesting case is a = -1, b = +I, and w(x) = I x l2k, k > 0. On
account of (4.1.6) we have I x lk I p;o.k-!>(2i - 1) I ~ P~o,k-!>(1) = 1, so that

(7.21.2) I (1 - x)/2) a/Hi I p~a ,o)(x) I ~ 1, -1 ~ X ~ +1, a ~ - !.

The equality sign hold!-l only for x = -1. In the interval 0 ~ x ~ 1 this is,
for large n, less precise than the first inequality (7.32.6).
In case a = 0, b = + oo, and w(x) = e-", we obtain for the Laguerre poly~
nomials
(7.21.3) X~ 0,
the equality sign holding only for x = 0 if n > 0.
With regard to these special cases, see Szeg6 2 and 3.

7.3. Legendre polynomials

A second proof of (7 .21.1) and various other important inequalities can be
obtained by means of the differential equation for Legendre polynomials.
(1) THEOHEM 7.3.1. Let n ~ 2. 'l'he successive relative maxima of I Pn(x) I,
when x decrease8 from 1 to 0, form a decreasing sequence. More precisely, if
p, 1 , p, 2 , , J.I.Cn/2J denote these maxima corresponding to decreasing values of x,
we have
(7.3.1) 1 > J.LI > P,2 > ··· > J.l.[n/2) •

From this (7.21.1) follows again.

------------------------·----·------····
[ 7.3 J LEGENDRE POLYXOMIALS 165

If n is even, we have

J.Ln/2 = J Pn(O) I= 1·3 .. • (n - 1).

2·4 · · · n
For the proof let
(7.3.2)
2
Then we havef(x) = IPn(x) ) if P:(x) = 0, or if x = ±1. Therefore,
~
2
(7.3.3) max 1Pn(x)) max f(x).
-l~z~+l -l~z~+l

Now, on account of (4.2.1),

n(n + 1)f'(x) = 2P~(x)ln(n + 1)Pn(x) - xP~(x) + (1 - x 2 )P~(x))
(7,3.4) I I I 2
= 2Pn(X)·XPn(X) = 2x(Pn(x)),

so that f(x) is decreasing for x < 0 and increasing for x > 0. This establishes
the statement.
(2) THEOREM 7 .3.2. Let n ~ 2. The successive relative maxima of
(sin e)t 1 P n(cos e) 1 when e increases form o to 1r/2, form an increasing sequence.
From (4.24.2) we obtain for a = {3 = 0
d2 u
(7.3.5)
de2 + ¢(e)u = 0; u = u(e) = (sin e)! Pn(cos e),

Introducing
(7.3.6) f(e) = (u(e)) 2 + ~(e)(tt (e))2,
1

we have
(7.3.7) f'(e) = 2u (e)lu(e)
1
+ ~(e)u"(e) + 1~ (e)u (e)) 1 1
= ~~(e)lu 1 (e)(

Now ~(e) is an increa:-;ing function in [0, 1rj2], so thatj'(e) > 0, andf(e) is also
2
iner<'nsing. Butf(e) = i·u(e)) ifu 1 (fJ) = O;thisprovcsthetheorem.
(3) An important. application of Theorem 7.3.2 is
THEOREM 7.3.3. We have
(7.3.8) o ~; e ~ 1r.
Here the constant (2/7r)! cannot be replaced by a smaller one.
The first proof of nn irwqunlity of the type (7.3.8), with a constant A instead
of (2/7r)i, is due to Sticlt.jcs (8, p. 241). Further proofs have been given by
Gronwall (1, p. 221) and Fejer (9, pp. 280-291). The following proof is due to
S. Bernstein (2, p. 236); it was the first leading to the precise constant (2/7r)}.
Let n be even. From Theorem 7 .3.2
1GG INEQUALITIES [VII]

(7.3.9)
with the .sign of equality holding if e = 1r /2. l\ ow let n be odd. Then for
o~e~1r

(sin e)! I Pn(cos e) I < max lf(e) l t

0
(7.3.10) ~ 0 ~,.
= lf(7r/2)l!= !i+(n+~) )- 1 1P~(O)I,
2

where f(e) is defined by (7.3.6). Using the notation (4.9.2), we have 42

I Pn(O) I = gn/2 < (2/7r)!n-!,
(7.3.11)
I P~(O) I = (n + 1)g(n+l)/2 < (2/7r)!(n + 1)!,
according as n is even or odd. (In the second case we can use (4.7.31).) Now
I± + (n + !/ l-\n + 1) 1 < n-!; therefore (7.3.8) follows.
That (2/7r)! is the best possible constant is easily seen hy considering./ Pn(O) I,
n even. Besides this we have (cf. (7 .32. 9), a = {3 = 0)
(7.:3.12) max (sin e)! I P,. (cos e) I "'' (2/7r )} n-!, n ~ oo.
0~8~,..

7.31. Theorem of Sonin; Bessel functions

The argument used in the preceding section can be generalized in various
ways. For instance, the following important theorem holds.
THEOREM 7.31.1. Let y = y(x) satisfy the differential equation
(7.31.1) y" + ¢(x)y = 0,
where¢ (x) is a positive function having a continuous derivative of a constant sign
in xo < x < Xo . Then the successive relative maxima of I y I , as x increases
from x 0 to Xo ,form an increasing or decreasing sequence according as cp(x) decreases
• 43
or tncreases.
If we write
(7.31.2)
we have, in fact, f(x) = 2
(y(x) ) if y'(x) = 0, and
(7.31.3) f'(x) = 2y'(x)ly(x) + ..J;(x)y"(x) + !..J;'(x)y'(x)) = ..J;'(x)ly'(x) )2 •
That is, sgn f'(x) = -sgn ¢'(x); whence the ~-.;tatement follows.
42 The sequence (mi! 2gml is increasing, so that mi! 2grn < limm_. 00 m 112 gm = 71"- 1/ 2 •
43 Professor P6lya has kindly pointed out to me the following generalization of this

theorem: Let y(x) satisfy the differential equation

lk(x)y'l' + q,(x)y = 0,
where k(x) > 0, q,(x) > 0, and both functions k(x), q,(x) have a continuous derivative. Then
the relati1•e maxima of I y I form an increasing or decreasing sequence according as k(x)q,(x)
is decreasing or increasing. This was obtained independently by Butlewski I.
[ 7.32] JACOBI POLYNOMIALS 167

As an illustration, we apply this result to (1.8.9) for k = 1, x > 0. In this

case we have
1 2
(7 .31.4) ¢(x) = 1 +4 -2 a
X

so that ¢(x) is decreasing if a? < -!- and increasing if a 2 > -!-. In the latter
case x must be taken so large that ¢(x) > 0. We then conclude that the
relative maxima of x'l J a(x) I form an increasing sequence if a 2 < l, and a
decreasing sequence if a > -!-. In the first case x > 0, and in the second
2

X> (a2 - 41 )' .

According to (1.71.7), there are infinitely many such maxima, and they
tend to (2/ni. Thus, we have the theorem: .
THEOREM 7.31.2. If J a(x) denotes Bessel's functz'on of order a, we have

(2/?r·)'
(7 .31.5) sup lx' I J a(x) ll =
.:~o 1 finite and > (2/?r· )' if a>!.

For a = ±~we can use formulas (1.71.2). For a < -~we have x'J a(x) ~ oo
as x -~ +O [(1.71.1)]. The second statement holds in this case provided the
least upper bound in question is taken in an arbitrary interval [xo , + oo] with
Xo > 0.
See Szego 17, pp. 40-41, and compare similar theorems in Watson 3, pp.
488-489. See abo §7.8.

7.32 Jacobi polynomials

(1) The consideration of §7 .3 can easily be extended to ultraspherical poly-
nomials. A treatment of this case will be given in §7 .33. First, however,
we discuss general Jacobi polynomials P~a .P> (x). In applying the previous
methods to P~a .P> (x), the main difficulty lies in the fact that there is no special
point x = ~ interior to [ -1, + 1] at which P~a,P>(x) and its derivative have
values which are easily calculated, as in the case of the ultraspherical, and
especially of the Legendre, polynomials at x = 0. This is the reason that here
44
we must anticipate certain comparatively simple results from Chapter VIII.
These are the following:
(a) The formula of the Mehler-Reine type [(8.1.1)],

limn-a p~a,p) (cos

n-oo
=)
n
= (z/2)-a J a(z),

which holds uniformly for I z I~ R, R fixed.

u After completing the manuscript I received a paper of Korous (3) in which some of
the results of §7.32 arc derived by use of the differential equation· of Jacobi poly no~
mials but without using the asymptotic formulas (8.1.1), (8.21.10).
168 INEQUALITIES [VII]

(b) Darboux's formula l(8.21.10)],

p~a,P>(cos e) = n-! k(e) cos (Ne + ,) + O(n-!),

k(e) = 1r-!
( sin
())-a-! (
cos
e)-P-i ,
2 2
N = n + (a + {3 + 1)/2, 'Y = -(a+ !h/2, E ~ e~ 1r- E,

Here E is a fixed positive number; the bound for the error term holds uniformly.
In both cases, a and {3 are arbitrary real numbers.
Concerning the results of this section and subsequent ones, cf. Kogbetliantz
19, p. 125, S. Bernstein 2, and Szego 17.
(2) THEOREM 7.32.1. Let a> -1, {3 > -1,

(7 .32.1)

We have

(7 .32.2) max I P~a,P>(x) I = {(n+q)

n ,.._, n
q
ifq =max (a, {3) ~- t,
-l~x~+I I p~a,P)(x') I ,.._, n-! if q = max (a, {3) < - t.
Here x' is one of the two maximum points nearest xo .
The symbols ,.._, refer to the limiting procedure n ~ oo. In the second case
-1 < Xo < +1; then we use (8,21.10). The subsequent argument furnishes
the more precise result that the maximum of I P~a,P>(x) I in the interval 0 ~ x ~ 1
is of order nmax(a,-l/ 2). A similar result holds for -I ~X~ 0.
For the proof we generalize the argument of §7.3 (1) as follows. Let n ~ 1,
and let

r.
n(n + a + {3 + 1)f(x)
(7.32.3)
= n(n +a+ f3 + 1) IP~a.P>(x) ) + (1 2
- x2 ){! P~a.P>(x)
Then by using (4.2.1), we obtain

(7.32.4) n(n +a+ {3 + 1)f'(x) = 2la- {3 +(a+ {3 + 1)x){! p~a,P>(x)r.

Thusf'(x) can change its sign only at x = xo.

Weseethatthecondition -1 < Xo < +1 isequivalentto (a+ t)(f3 + !) > 0.
Now let a > -! and {3 > -!. Then the sequence formed by the relative
maxima of I P~a,P>(x) I in -1 ~ x ~ Xo and by the value of this function at
x = -1, is decreasing, while the sequence of the maxima in xo ~ x ~ +1 and
of the value of the function at. x = +I. is increasing. Therefore, I P~a,P>(x) I
attains its maximum in [ -1, + 1] at one of the end-points.
[ 7.32] JACOBI POLYNOMIALS 169

In case a ~ -!and -1 < {3 ~ -!,the linear function a - {3 (a+ {3 + l)x +

is non-negative, so that the sequence of relative maxima in question is increasing
+
in [ -1, 1], save for the case a = {3 = - !, in which it is stationary; the situ-
ation is opposite in the case {3 ~ - !, -1 < a ~ - !. Finally, let -1 < a<-!,
-1 < {3 < -!, so that again -1 < Xo < +
1; then the sequence of maxima is
+
increasing in [ -1, xo] and decreasing in [xo , 1], so that the absolute maximum
of I P~"·P>(x) I in [ -1, +I] is attained at the point of maximum nearest Xo on
the left or on the right.
See Problem 39.

(3) THEOREM 7 .32.2. Let a and {3 be arbitrary and real, and c a fixed positive
constant, n ~ oo. Then

(7.32.5) '/,f en-1 =< e =< 7r/2 '

'2f
<
. 0 =
< -1 .
e =en
See S: Bernstein 2, pp. 225-232 where Sonin's theorem is applied, but where
the proof is perhaps slightly more complicated than ours. Szego (17, p. 77)
uses the asymptotic formula (8.21.17) which is, of course, a more complicated
tool than (8.1.1) and (8.21.10) used below.
The bounds in (7 .32.5) are precise as regards their orders in n. They follow
also, as mentioned, from the more complicated asymptotic formula (8.21.17)
of "Hilb's type." By use of (4.1.3) we can obtain similar bounds for the
intervals 7r/2 ~ e ~ 7r.
We notice the useful inequalities
o-a-!O(n-!),
(7.32.6) p~a,P>(cos e) =
{ O(na),

(7.32.7)

which follow from (7 .32.5). Concerning the second bound in (7 .32.6), and
concerning (7 .32. 7), see §7 .32 (2).
We observe that e-a-!n _, "'--' n a if e "'--' n -I; thus it suffices to prove (7.32.5)
for a special value of c. Apply Theorem 7.31.1 ·with [(4.24.2)]

X = e, y = Un(e) = (sin~) aH (co::; ~YH p~a,p)(cos e),

(7 .32.8)
cp(e) = i - a2 + t__=-_.{j_2_ + {n + a+ {3 + 1}2·
. 2e 2e 2
4 sm - 4 cos --
2 2
First, let o = o(a, {3) be a fixed positive number, sufficiently small. Then cp(e)
is positive and decreasing in 0 < e ~ o if a < i. It is positive and increasing
2

in kn- ~ e ~ o if a > i; here k is a fixed number, k > (a 2 - t)\ and n is

1 2
170 INEQUALITIES [VII]

sufficiently large. Thus in both cases, the function ¢(0) is positiv:e and monotonic
in kn- ~ e ~ o, where k = k(a, {3), o = o(a, {3), and n is sufficiently large. The
1

same holds for a = i-, {3 ~ i-. Therefore, the sequences of the relative maxima of
2 2

I un(e) I in the interval kn -21 ~ 8 ~ o2are increasing and decreasing, respectively,

for large n, according as a < i- or a > i-. According to (8.1.1) and (8.21.10),
we find in both cases un(e) = O(n-~). (We have (sin 0/2)a+!(cos 0/2)!9+!
'""'"' ea+'.) This furnishes the first part of (7.32.5) with c = k. The second part
follows immediately from (8.1.1).
2 2
In the case a = {3 = t, excluded before, ¢(0) is constant. Then we know
p~a,P>(cos e) explicitly (cf. (4.1.7), (4.1.8)).

(4) THEOREM 7.32.3. Let un(e) have the same meaning as in (7.32.8), and
Mn =max I Un(e) I when 0 < e ~ 7r/2. We have

lim n'Mn = J1r -!

if-!~ a~ +t,
(7 .32.9)
n-+oo lfinite and > 7r_, tf a> 2·
1

Here {3 is greater than - 1.

Cf. S. Bernstein 2, pp. 225-232; Szego 17, pp. 79-80. Cf. Theorem 7.31.2.
The preceding argument needs only a slight modification. We have to discuss
the maximum of n' I un(e) I foro~ e ~ 1r/2 if-!~ a~ +!,and fore = n- z,
1

0 ~ z ~ c, if a > 1. The first is ""' 1r-! as n ~ oo (according to (8.21.10));

the second is

""' n'
0~~;c {(2: )a+!na(z/2)-a IJ a(z) 1} 0~~~~ 1(z/2)' I J a(z) IJ
=

(according to (8.1.1)). For sufficiently large c this is independent of c and

greater than 1r-! (Theorem 7 .31.2).
(5) Finally, as an application of (4.21.7) we point out the following generali-
zation of Theorem 7 .32.2:
THEOREM 7 .32.4. Let a and {3 be arbitrary and real, and c a fixed positive
constant, n ~ oo. Then

f(!!_)k p<a.fi>(x)} = ~e-a-k-!O(nk-!) f -I< e =1r,

ten= < ;2
(7.32.10) j d n 2k < e =en
< -1 .
\ X x=c<Js8 O(n ·+a) tf 0 =

From this we find, uniformly in x, -1 ~ x ~ + 1,

(7.32.11) (;xy P~a.P>(x) = O(nq), q = max (2k + a, 2k + {3, k - !).

7.33. Ultraspherical polynomials

In the ultraspherical case the preceding considerations can be simplified.
(1) By the same argument·as that in §7.32 (2) we find that
[ 7.331 ULTRASPHERICAL POLYNOMIALS 171

is increasing or decreasing in 0 ~ x ~ 1 according as X > 0 or X < 0, X non-

integral; we assume n > 0 in the first and n > - 2X in the second case. Thus
we obtain the following theorem:
THEOREM 7.33.1. We have

n + 2X- 1) if X > 0,
(7 .33.1) max I P~">(x) I = {( n
-1~.:~1 I P~">(x') I if X < 0, X non-integral.
Here x' is one of the two maximum points nearest 0 if n is odd; x' = 0 if n is even.
In the first case (4.7.3) has been used (cf. also Theorem 7.4.1). In the second
case we obtain, if n is even,

(7 .33.2)

whereas, for n odd,

max IP~">(x)l < lf(O))!= /n(n+2X))-!Ip~>->'(O)I
-1~.:~1

(7.33.3)
= I2X lln(n + 2X))-! (X I t (n ~)}~1 2) \.
Both bounds (7 .33.2) and (7 .33.3) are "'"' 21 ->- I r(X) l-1n"-·1 as n ~ oo ; the first
bound is attained for x = 0, the second bound is precise in the asymptotic sense.
(2) By use of (4.7.11) we obtain in a manner similar to that in §7.3 (2), (3),
the following:
THEOREM 7.33.2. Let 0 < X < 1. Then we have for 0 ~ e~ 7r

(sin e)" I P~">Ccos o) 1

(7.33.4) if n is even,
if n is odd,
and
(7.33.5)
1
Here the constant 2 -"l r(X) )- cannot be replaced by a smaller one; an has the
1

same meaning as in (4.9.21).

In (7.33.4) the sign of equality holds only fo~ even n and e = 1rj2. Now
an "'--' Ir(X) )- n"-\ and an < I r(x) l-ln>-.-I (since In 1-"an l is increasing45 ); more-
1
172 1NBQUAL1TIES [VII]

over, p. (1 >..) (n + A) )-i(n+ 2

+
1)" < n"-\ 45 so that equation (7.33.5)
follows.
Less precise (but more general) inequalities can be obtained from the general
result (7.32.5) for a = {3 = A - !. We have ·
rr"O(n"- 1), cn- 1 ~ o~
l
1r/2,
P~'->(cosO) =
(7.33.6) O(n2"-1), 0 ~ 0 ~ en-\
A arbitrary and real, A ~ 0, -1, -2, ... ; c > 0.
(3) We point out an interesting special case of (7.33.6), namely,. A = ~
(cf. Szego 16). We have P:(x) = p~i~ 1 (x) (cf. (4.7.14)), so that
10-!0(n!), cn- ~ 0 ~ 1r/2,
1
1
(7.33.7) Pn(cosO) = 2
O(n) , 0 ~ 0 ~ cn-
1
•

The first bound can be used for the whole interval 0 < 0 ~ 1r/2 [cf. (7.32.6)].
According to (7.33.1) the inequality
(7.33.8) I P:(x) I ~ n(n + 1)/2, -1 ~ x ~ + 1,
holds, the equality sign being taken if n = 0, 1, or n > 1, x = ±1.
By using the first identity in (4. 7 .27), we find

(7.33.9) (1-x)PnX
2 1 ( )
= n(n + 1)
n+ ( ) ))
(Pn-1X -Pn+1(X .
2 1
Thus we conclude from the first bound in (7.33.7) (which now holds for
0 < 0 2 1r /2, cf. the previous remark) the following:
THEOREM 7.33.3. If Pn(x) denotes Legendre's polynomial, we have for 0 < 0 < 1r
(7.33.10) Pn-1Ccos 0) - Pn+1(cos 0) = (sin o)!O(n-!).
The bound of the factor 0 (n -!) is independent of 0.
This result, without the factor (sin 0)!, is due to Stieltjes (cf. Hermite-Stieltjes
1, vol. 2, pp. 174-177; Fejer 9, pp. 295-298). The present form of the theorem
is implied in the previous more general results of S. Bernstein, Kogbetliantz,
and Szego; cf. Szego 16.

7.34. Bounds for integrals involving Jacobi polynomials

THEOREM 7 .34. Let a, {3, Jl. be real numbers each greater than - 1. Then as
n -~ oo (concerning the second part of the statement see below)

46 In view of the concavity of log x, we have

(1 - A) log (n - 1) + A log n < log (n + A - 1),
{ l - A) log (n 2 ) -r A log l(n + 1) 1 <log 1{1- A)n + A(n + 1)
2 2 2
1.

------------------ ·---·--·
[ 7.34] lN'tEGilALS iNVOLVING JACOBi POLYNOMiALS 173

2J.L <a- ~~
(7.34.1) 2J.L =a-~~
2J.L >a-~-
See Szego 17, pp. 84-86, where the existence of the limits of the corresponding
ratios is proved, and the limits are calculated. The proof of the second part
of (7.34.1) requires a more complicated apparatus [(S.21.18)]; here we prove
only that

(7.34.2) Un = n! 1 1
(1 - x)"' I P~a,f3>(x) I dx = O(log n), and Un ~ oo.

This is sufficient for later purposes (cf. §9.41 (5)).

We use (7.32.5); in fact, we have

}ot (1 - x)"' I p~a,f3>(x) I dx = 0(1) }o e·' rlp.+l I p~a,f3>(cos o) I do

2

(7.34.3)
= 0(1) 1n- 1

02P.+lna dO+ 0(1) 1: 2

0 P.+lo-a-!n-! dO
1
2

= O(na-2p.- 2 ) + O(n -!)I 0(1) + O(na- p.-~)}. 2

If 2J.L -- a+ ~ = 0, the last term must be replaced by O(log n).

On the other hand,

A 1n- 1

02P.+l I P~a,f3>(cos 0) I dO,

(7.34.4) 1 1
(1 - x)"' I P~a,f3>(x) I dx > 2J.L <a- ~'

1
1r/2
A I P~a,f3>(cos 0) I dO, 2J.L > a - ~'
/4

where A is a proper positive constant. According to (8.1.1), the first bound is

:::A 1 1
(z/n) 2p.+l?t(z/2)-a I Ja(z) I n-1 dz I"V na-2p.-2•

According to (8.21.10), the second bound is

12
r-v n-! t. l cos (NO + 'Y) I dO r-v n-!.
J"'4
In case 2,L =a-~' a> -!,we have
1 1
(1 - x)"' I P~a,f3>(x) I dx > A' 1wn-
1
02P.+l I P~a,f3)(cos 0) I dO,

where w is a fixed poi:litive number, and A' is independent of n and w. From

this inequality, according to (8.1.1),
174 INEQUALITIES [VII]

follows. The last integral becomes arbitrarily large with w; this furnishes the
second part of (7.34.2).

7.4. Fejer's generalization of Legendre polynomials

(1) THEOREM 7 .4.1. Let (an l be a sequence with positive terms. Then the
"Legendre polynomials" Fn(x) (6.5.1) associated with the sequence (anl satisfy
the £nequalities
(7.4.1)
The sign of equality holds only if n = 0, or n > 0 with x = -1 or x = + 1.
This leads to a new proof for (7.21.1) and for the first part of (7.33.1).
(2) The inequalities (7.3.8) and (7.33.5), of the Stieltjes type, can likewise
be extended to the polynomials Fn(x), though with certain larger constants.·
We prove the following:
THEOREM 7 .4.2. Let
an > 0, dan = an - an+l > 2
0, d an = an - 2an+l + an+2 > 0,
(7.4.2)
n = 0, 1, 2, ... ,
and
(7.4.3)
Then for the "Legendre polynomials" F n(x) associated with the sequence Ian l,
we have
(7.4.4) 0. < () < 71"; n = 0, 1, 2, .. · .
See Fejer 9, pp. 291-295; Szcgo 11, p. 179. Szego obtains a larger bound
under a more restrictive condition. The inequality (7.4.4) and the present
proof are new.
Under the conditions (7 .4.2) the function f(z) is regular for I z l < 1 and con-
tinuous for I z I ~ 1, I z - 1 I ~ o, where ois an arbitrarily small positive number.
Indeed, we see that limn_.ooan = a ~ 0 exists. If a = 0, we use a well-known
case of Abel's inequality (1.11.6); if a > 0, we write an = (an - a) a. +
Now from (6.5.1),
[n/2] [n/2]
(7 .4.5) F n( COS () ) = Z-n/2 ""''
L.J akan-kZ k +Z n/2 ""''
L.J akan-kZ-k , z = e2i6 ,
k-0 k-0

where the sign L:' indicates that for even n the last term k = n/2 has to be
multiplied by!. Hence,

(7.4.6) I Fn(cos 0) I~ 21 [I:} a~;an-kll,

k-0
z = e2i6 .
 [7.5] RECAPITULATIOJ'\ 175

By virtue of (1.11.6) we obtain

(7.4.7) m = n - [n/2], z = e2'·e .

Now, writing
(7.4.8)

we have, according to a theorem of Fejer-Szego (1),

(7.4.9) I f(z) I ~ I Pl(z) I ~ I p2(z) I ~ ··· , 1 z I ~ 1, z ~ 1.
But

n even,
(7.4.10)
n odd,
so that the modulus of this sum is in both cases not greater than 21 f(z) I ;
whence (7.4.4) follows.
We give here a sketch of the proof of (7.4.9). It suffices to show that I f(z) I
~ I Pl(z) I ' or I f(z) I ~ I f(z) - ao I' or m[j(z)] ~ ao/2 for I z I < 1. Now
limn_=-,an = a ~ 0 exists. If f(z) is replaced by f(z) - a(1 - z)-\ it is seen
that we can assume a = 0 from the start. Then for I z I < 1
00
2
f(z) == L Ll anln + 1 + nz + (n - 1)z2 + · · · + zn}
n-o

+ 1) + f
1
:::-2 "£
n-O
2
Ll an(n
n-O
2
Ll anl (n + 1)/2 + nz + (n - 1)i + ·· · + znl
00

=: ao/2 +L
n~O
2
Ll anl (n + 1)/2 + nz + (n- 1)i + · · · + zn).
But for a real (}
ml (n + 1)/2 + neie + (n - 1)e2i 6 + ... + eniel
n
= L: ci + cos o + cos 2o + . . . + cos vo)
•=0

= f sin (v + i)O = J(sin I (n + 1)0/21) 2

•=O 2 sin I(} /21 2 sin {(} /21

(cf. Fejcr 1; see also P6lya-Szego 1, vol. 2, pp. 78, 269, problem 17).
Fejer (9, pp. 295-298) also gives an extension of Stieltjes' theorem on
Pn_ 1(x) - Pn+l(x) (cf. Theorem 7.33.3) to the polynomials Fn(x).
7 .5. Recapitulation
In the last sections we gave various derivations of the important inequality
(7.21.1):
 176 INEQUALITIES [VII]

(a) from the orthogonal property, by use of the general Theorem 7.2;
(b) from the differential equation (§7.3 (1));
(c) from the trigonometric representation (4.9.3).
Moreover it follows also
(d) from Laplace's integral representation (4.8.10); in fact, if- 1 ~ x ~ + 1,
Ix + (x 2
- l)! cos 1> I - Ix + i(1 - x 2)' cos¢ I
cos2 ¢ l t ~ { x 2 - x 2) l ~
2
- {x + (1 - x
2
) + (1 - 1.

7.6. Laguerre and Hermite polynomials

(1) THEOREM 7.6.1. Let a be arbitrary and real. The sequence formed by the
relative maxima of I L~" >(x) I and by the value of this function at x = 0, is decreasing
for .'X < a + !, and increasing for x > a + !. The successive relative maxima of
I H,.(x) I form a decreasing sequence for x ~ 0, and an increasing sequence for
X ~ 0.

Indeed, the function

(7.6.1) n{L~")(x) 1 +X {:x L~")(x)

2
r
is decreasing for x < a + ! and increasing for x > a + !. The function
(7.6.2)

is decreasing for x < 0 and increasing for x > 0. Both statements follow by
differentiation as in §7.3 (1); we use the first differential equation in (5.1.2)
and (5.5.2), respectively.
(2) THEOREM 7.6.2. Let a be an arbitrary real number. The successive relative
maxima of
(7.6.3)
forrn an increasing sequence provided x > Xo. In the first case
0 if a 2 ~ 1,
(7.6..4) Xo = a2 - 1
{
2n +a+ 1
In the second case
< 4,
~ { ~,,' - tl'
2 f a2 = 1
(7.6.5)
"' '/,f a2 > 1
4·

In the first case we take n so large that 2n + a + 1 > 0.

Sonin's theorem 7.31.1 applies to the functions ·u aud v occurring, respectively,
in the third and fourth equations of (5.1.2). The larger zero 'Yn of the coefficient
[ 7.6] LAGUERRE AND HERMITE POLYNOl\1IALS 177
of u is an upper bound for the zeros of u (cf. Theorems 1.82.3 and 6.31.2). 46
The differential equation shows also that x = 'Yn is the last point of inflexion of u;
thus 'Yn is at the same time an upper bound for the points at which the relative
extrema of u are attained. Similarly, if 'Y:
denotes the larger zero of 4n +
2a +
2 - x + 2
(i - a )x-\ we find in ('Y:)! an upper bound for the points at
which v attains its relative extrema. (The bound of Theorem 6.31.2 is 'Y: .)
Now if Xo is chosen according to (7.6.4) and (7.6.5), we have
n + (a+ 1)/2 + 1 - a
2
_ i > O, n + (a+ 1)/2 1 - a2 . 0
x 4x2 x2 2x3 <
forxo < x < 'Yn,
(7.6.6) 1 2
4n + 2a + 2 - x +4 -
X
a > 0, -1

for Xo < X < 'Y n , 1

respectively. This establishes the statement.

THEOREM 7 .6.3. The successive relative maxima of
(7 .6. 7) e-x 212 IH n(X) I
form an increasing sequence for x ~ 0.
Here the second equation in (5.5.2) can be used.
(3) The bounds analogous to (7.32.5) are readily obtained by means of the
asymptotic formulas (8.1.8) and (8.22.1), which correspond to (8.1.1) and
(8.21.10), respectively. It is convenient to use the fourth equation in (5.1.2).
Then 4n +2a +
2 - x (i - a 2)x-1 is positive and decreasing in 0 < x ~ o
+
if a ~ i; it is positive and increasing in kn- 1 ~ x ~ o provided a 2 > i, k >
2

2
(a - i)/4 and n is sufficiently large. Here o = o(a) is a sufficiently small posi-
tive constant. Therefore, as in §7.32 (3), we obtain the following:
THEOREM 7.6.4. Let a be arbitrary and real, c and w fixed positive constants,
and let n ~ oo . Then
:t
2J en -1 =< x =< w,
(7.6.8)

These bounds are precise as regards their orders in n; they follow also from
the more complicated formula (8.22.4) of Hilb's type.
For a ~ - !, both bounds hold in both intervals, that is,
x-at2-iO(nat2-l)
(7.6.9) L~a>(x) = { a ' 0 <X ~ w, a ~ - !.
O(n ),
4
GIf this coefficient is constantly negative for x > x0 , lui has no zeros and no maxima.
Similarly for v. These cases can be excluded.
178 INEQUALITIES [VII]

On the other hand,

(7.6.10) 0 ~X~ w, _.!
2.

And generally, with a arbitrary and real,

(7.6.11) a = max (!a - t, a), 0 ~X~ w.

Finally, we obtain the following analogue of Theorem 7.32.3:

THEOREM 7.6.5. L et M n = maxo <x~"' e-x/2X a/2+l I L(a)(
n
)I
X • Then
-!
, n -af2+iM., = 7f if-!~ a~+!,
(7.6.12) I1m
n -oo
1 finite and > 7f -!
f a>
1,
1
2·

The proof is very similar to that in §7.32 (4); of course, (7.6.12) also follows
directly from the deeper formula (8.22.4) combined with (7.31.5).

7.7. Problem of Lukacs

(1) This problem (Lukacs 1) deals with a more precise form of the mean-value
theorem

(7 .'7 .1) A ~ b _1 a 1b
a f(x) dx ~ B,

where A = mina~x~b f(x), B = maXa~x~b f(x), provided f(x) is restricted to the

set of all1rn with a fixed value of n.
THEOREM 7.7. Letf(x) be an arbitrary 1fn with the minimum A and maximum B
in [a, b]; then

(7.7.2) A + B- Tn
A
~ -1
-
b- a
lb
a
B- A
f(x) dx ~ B - - - - ,
Tn

where
J (m + 1) 2
ifn =2m,
(7.7.3)
Tn = l (m + 1)(m + 2) if n =2m+ 1.

The number Tn cannot be replaced by a smaller one.

This result is the analogue of an older theorem due to Fejer which deals
with the analogous question for trigonometric polynomials of a fixed degree n,
b -- a = 21r. In this case Tn = n +
1. The proof of Fejer's theorem can
be based on Theorem 1.2.1; however, various other methods have been used
(cf. P6lya-Szego 1, vol. 2, pp. 83, 277-279, problem 50).
It suffices to prove the first inequality of (7.7.2); the second one follows when
we replace f(x) by - f(x). In addition, there is no loss in generality in assuming
A == 0. It is readily seen then that Tn is the greatest possible value of f(x) if x
varies in [a, b], and f(x) ranges over the class of the 1fn which are non-negative in
 [ 7.7] PROBLEM OF LUKACS 179

[a, b] and satisfy

(7 .7 .4) b- a
1 1b
a f(x)' dx = 1.

Let now max f(b) = Mn for the same set of7rn; we shall prove that Tn = Mn.
It is clear that Tn ~ Mn. On the other hand, if x 0 is an arbitrary point in [a, b],
we see, by means of a linear transformation, that

(7.7.5) f(xo) ~ Mn - -
1
Xo- a
1x.o· f(x) dx,
a
1
f(xo) ~ Mn b--=-
Xo
1b
x0
j(x) dx.

Multiplying the firf5t. inequality by Xo - a, the second by b - Xo, and adding, we

find

(7.7.6) 1
f(xo) ~Mnb-a 1b
a f(x)dx=Mn.

(2) First method of calculating Mn. See P6lya-Szego 1, vol. 2, pp. 96, 297,
problem 108.
Assuming a = - 1, b = +
1, we use Theorem 1.21.1 and represent the poly-
nomials in (1.21.1) as linear combinations of certain convenient polynomials.
For arbitrary and real u., v. (subject only to the normalization condition (7. 7.8)),
we write

if n =2m,

(7.7.7)
f(x) = (1 - x){ta r
u.P~ 1 ' 0 > (x) + (1 + x){ta v.P~ 0 ' 1 > (x) r
if n =2m+ 1.
Because of the orthogonality of Jacobi polynomials, we have, in the notation of
(4.3.3).,

+ m-1
m
L h(O,O)
• u.2 L h(l,l) 2
• v. if n =2m,
(7.7.8) 2 = 1+1
_ j(x) dx =
1
•-0
m
•-0
m
L h(l,O)
• u.2 +L h(O,l) 2
• v. if n =2m+ 1.
•-0 •-0
But

{ta u.P~o,o)(l)r if n =2m

(7.7.9) f(l) =
2{ f: v.P~0,1)(1)r
•-o
if n =2m+ 1,

so that
180 INEQUALITIES [VII]

m
2 L {h~O,O)} -1 { p~O,O) (1)} 2 if n =2m,
•-0
(7.7.10) m
4 L {h~0,1) J-1{P~0,1)(1) }2 if n =2m+ 1,
•-0

which is attained for u. = const. {h~ 0 ' 0 >}- 1 P~o,o>(1), v. = 0, and u. = 0, vp =

const. {h~ 0 ' 1 >)- 1 P~ 0 ' 1 l(l), v = 0, 1, 2, · .. , m,respectively. Thecorresponding
polynomials are obviously the "kernel" polynomials K~a ,{j\x) ( cf. (4.5.3)) for the
cases a = {3 = 0 and a = 0, {3 = 1, that is, constant multiples of P~1 ' 0 >(x) and
P~ ' >(x). In other words, we have f(x) = const. {P~l.Ol(x) } , and f(x) =
1 1 2
const. (1 + x) {P~1 ' 1 >(x) }2, respectively. From (7.7:10) we now obtain (7.7.3)
by direct calculation or, more easily, by mathematical induction, reasoning from
m tom+ 1.
(3) Second method of calculating M n· See Lukacs, loc. cit. 47 This method is
based on certain mechanical quadrature formulas related to considerations
similar to those in §3.4.
Let n = 2m, x0 = 1, and let the zeros of (1 - x) p~l.Ol(x) be denoted by
Xo , X1 , · · · , Xm • If j(x) is a 11"n and L(x) the Lagrange interpolation polynomial
of degree m which coincides with f(x) at Xo , x1 , · · · , Xm, then
f(x) - L(x) = (1 - x) P~1 ' 0 >(x)p(x),
where p(x) is a proper 11"m-1 . Therefore,

1-~
1

1-~
1

1-~
1

(7 . 7.11) f(x) dx - L(x) dx = (1 - x)P~1 ' 0 >(x)p(x) dx = 0,

so that as in (3.4.1)

(7 . 7.12)
11+1
2 _ f(x) dx = ~ "A.j(x.),
m
1
where the coefficients "A. do not depend onj(x). Upon writing

(1- x)
p<l,o>(x)}2
{
_m__ if v > 0,
(7.7.13) j(x) = X - Xp
{
(P;!·o)(x) }2 if v = 0,
we show as in §3.4 (2) that the numbers "A. are positive. Now in view of f(x) ~ 0
and of (7.7.4), we obtain from (7.7.12)
1
(7 . 7.14) 1 ~ "Aof(1), f(1) ~ Ao ;
this is the precise bound of f(1), attained when an only when f(x.) = 0, v =
1., 2,. · ·, m; that is, when f(x) = const. {P;!' 0>(x)} 2. In order to find Ao, it is
convenient to write f(x) = 'Y P~1 ' 0 >(x) in (7.7.12) (cf. (3.4.3)), where 'Y is a con-

Lukacs uses this second method in 1; however, the first method was also in his possession
47

(r.f. 1, p. 296).
[ 7.71] GENERALIZATIONS; APPLICATIONS 181

stant; then, on account of (4.5_.3), a = {3 = 0,

I
2
!+
-1
1
j(x) dx =I
2
JH P~1 ' 0>(x)
-1
dx =
m
_'Y_
+1
JH K~o.o>(x)
-1
dx = __2__ = 1.
m +1
Consequently, A.() = 'Y P~ ' > (1) = (m + I t
1 1 0

Now let n = 2m + 1, Xo = 1, Xm+1 = -1, and denote the zeros of the poly-
2 1 1
nomial (1 - x )P~ ' >(x) by Xo, X1, · • • , XmH. The same argument as before
leads to
11+1 m+1
(7.7.15)
2 _1 f(x) dx = ~ A..j(x.), A..> 0,

wherej'(x) is an arbitrary 11'n. The maximum in question is again A.il\ which is

attained when and only when f(x.) = 0, v = 1, 2, ... , m + 1; that is, f(x) =
const. (1 + x) {P~ ' >(x) } • In order to find A.o, we writej(x) = 'Y (1 + x) P~1 ' 1 >(x),
1 1 2

so that on account of (4.5.3), a = 0, {3 = 1,

1
2 _ f(x) dx = ~ _
1+1
1
1+1 (1 + x)P~ ' >(x) dx
1
1 1

(7.7.16)
+1
= ~ (1 + x)K;.:>· (x) dx = ~ = 1, 0
m+ 2 1 -1 m+ 2
and we obtain
1
A.il = 2-y(m + 1) = (m + l)(m + 2).
7 .71. Generalizations; applications
(1) Let da(x) be an arbitrary distribution on the finite or infinite segment
[a, b], (pn(x)} the associated set of orthonormal polynomials, and x 0 an arbitrary
but fixed point. Then if p(x) is an arbitrary 11'm with

(7.71.1) lb l 2
p(x) f da(x) = 1,

we have
m
(7.71.2) I p(xo)
2
1 ~ L
v=O
I p.(xo) 1
2
,

with the sign of equality if and only if p(x) = const. I:~!:o p.(x0 ) p.(x). This
was proved in §3.1 (3).
In certain special cases we can calculate the maximum of (or some upper
bounds for) the right-hand member of (7.71.2) if Xo runs over a certain interval.
The bounds obtained hold uniformly in that interval for the set of all p(x)
which are 11'm and satisfy (7.71.1).
In what follows we consider various distributions of the form da(x) = w(x) dx;
p(x) denotes an arbitrary 71',. which ~ati~fics (7.71.1).
(2) Let a = - 1, b = + 1, w(x) = l.
182 INEQUALITIES [VII]

THEOREM 7.71.1. Let p(x) be an arbitrary ?rm subject to the condition

(7.71.3) 1-: 1
I p(x)
2
1 dx = 1.

Thenwehavefor- 1 ~ Xo ~ +1
2-'(m + 1),
(7.71.4) I p(xo) I ~ { A( 2 )_* !
1 - Xo m.
Here A is an absolute constant.
The first bound is precise; it is attained for x0 = ± 1 if p(x) is a proper 7rm.

These inequalities follow from (7.71.2) when we use (7.21.1) and (7.3.8).
Now let a= - 1, b = + 1, w(x) = (1 - x)a (1 + x)fl.
THEOREM 7.71.2. Let a > - 1, (3 > - 1, and p(x) an arbitrary ?rm subject to
the condition
[+I
(7.71.5) J_ 1
(1 - x)a(1 + x)tl I p(x) 2
1 dx = 1.

Then
o-a-i O(m') •f
o -1 < 0 =
em= < 7/2
r,
(7.71.6) p(cos 0) =
{ O(maH) •f 0< 0 <
o ==em. -1

Here c is an arbitrary but fixed positive number, and the constants in the 0-terms
depend only on a, (3, and c. Similar bounds hold in the interval [1rj2, 1r].
For the proof we notice that in this case Pn(x) rv n' P~a,fl)(x) [(4.3.3)], and
according to (7.32.5),
m
L v{P~a,fl)(cos 0) } 2 = L vO(v 2a) +L v0-2a-1 0(v- 1)
•=1 •O<c v06;c ·

= oco-2a-2) + o-2a-10(m) = o-2a-10(m),

if em- 1 ~ 0 ~ 1r /2. For the same sum we obtain the bound L:';';..1 vO(v a) =
2

O(m2 a+2) if 0 ~ 0 .::;; cm-1 •

By use of (7 .32.2) certain precise bounds can likewise be derived.
Let a = 0, b = + co, w(x) = e-x. Then, according to (7.21.3),

(7.71.7) e-"' 012 1 p(xo) l ~ (m + 1)!, Xo ~ 0, laC() e-x I p(x) 12 dx = 1.

This bound is attained if Xo = 0 and p(x) is a proper ?rn .

The case a = 0, b = +co, w(x) = e-:r xa, a > -1, can be treated by a
method similar to that used in the Jacobi case discussed above (cf. (7.6.8)).
(3) Now let [a, b] be a finite interval and f(x) an arbitrary ?rn which is non-
negative in [a, b] and satisfies the condition
r 1.111 GEKERALIZATIONS; APPLICATIONS 183

(7.71.8) lb f(x)w(x) dx = 1.

We intend to determine the maximum of If(xo) I, where x 0 is a fixed real or

complex value.
By Theorem 1.21.1 we can write

{ta + (x- a)(b _ {'%

l
u.p.(x)Y x) v.q.(x)r if n =2m,
(7.71.9) f(x) =
(x- a) {ta u.r.(x)Y + (b- x) {ta v.s.(x)Y if n =2m+ 1,

where {p.(x)}, {q.(x)}, {r.(x)}, {s.(x)} are the orthonormal sets of polynomials
associated with the weight functions
(7.71.10) w(x), (x - a)(b - x)w(x), (x - a)w(x), (b - x)w(x), a ~ x ~ b,
respectively. The third arid fourth sets are special cases (x0 = a, x0 = b) of
the "kernel" polynomials (Theorem 3.1.4); the second set can be calculated by
means of Theorem 2.5. In both cases, n = 2m and n = 2m + 1, we have for
the real numbers u. , v.
b m m

(7.71.11)
l f(x)w(x) dx =

so that according to Cauchy's inequality

L u~ + ·-0
•=0
L v; = 1, Vm =0 if n = 2m,

!f(xo) I~

(7.71.12)
max {ta I 2
p.(xo) / , I Xo - a II b - Xo I '%: I q.(xo) 1 2
} if n = 2m,

max {1 Xo - aI ta I r.(xo) 12, ·1 b - Xo I ta l s.(xo)

2
1} if n = 2m + 1.
In case a ~ x 0 ~ b, the absolute value signs can be omitted. The right-hand
side of (7.71.12) represents the maximum required.
(4) Let a = - 1, b = + 1, w(x) = (1 - x)a(1 + x)tl, a and (3 > - 1. Then
the four sets of polynomials mentioned in (3) are, respectively, constant mul-
tiples of
(7.71.13) {P~a,/l)(x)}, {P~a+l,/l+I)(x)}, {P~a,/l+I)(x)}, {P~a+l,/l)(x)}.

In the special case Xo = 1, for the maximum of /(1) we obtain (cf. (4.5.3)):
m
L {p.(1) }2 = K~a,tJ)(l) if n =2m,
v=O
(7.71.14) m
2 L {r.(1) }2 = 2K~a,tJ+I)(1) if n =2m+ 1.
•-o
Therefore, the following theorem holds:
184 INEQUALITIES [VII]

THEOREM 7. 71.3. Let f(x) be an arbitrary 1r n which is non-negative in [- 1, + 1]

and satisfies the condition
+1
(7.71.15)

Then
1_
1
f(x)(l - x)a(l + x)fJdx = 1, a > - 1, (3 > - 1.

/(1) ~
2-a-{J-1 r(m + a + 2)r(m + a + (3 + 2)
if n =2m,
(7.71.16) r(a + l)r(a + 2)r(m + l)r(m + (3. + 1)
2-a-{J-1 r(m + a + 2)r(m + a + (3 + 3)
if n =2m+ 1.
r(a + I)r(a + 2)r(m + I)r(m + (3 + 2)
2
These bounds are precise; both are ""' m "+2 as m -7 oo.

Cf. P6lya-Szego 1, vol. 2, pp. 96-97, 298, problem 110. Upon permuting a
and (3, we obtain the corresponding bounds for f(- 1). In general, we find,
under the same ~onditions as in Theorem 7.71.3 (cf. Theorem 7.71.2),
0- 2 a-lO(m) if cm-1 ~ 0 ~ 7r/2,
(7.71.17) f (cos 0) =
1O(m 2
a+ 2) if 0 ~ 0 ~ cm- 1•
The bounds for f(cos 0) are similar in 1r/2 ~ 0 ~ 1r. Further, a bound of thEl
form O(mc) holds uniformly in 0 ~ 0 ~ 1r, where c = max (2a + 2, 2(3 + 2, 1).
The constants of all these 0-terms depend only on a, (3, and c.
(5) By means of Theorem 1.21.2 we can treat the following problem. Let
f(x) be an arbitrary 'Trn, non-negative for x ~ 0, and

(7.71.18) 1C() e-xxaf(x) dx = 1, a> - 1.

What is max f(O)?

We write (cf. (5.1.1))

I-ta u,{r(a + l)(v ~a)}-! L~a>(x) r

+ x!~ v.{r(a + 2)(v + ~ +l)}-t L~a+I>(x) r
if n =2m,
(7.71.19) f(x) =
~~ u.{r(a + l)(v ~a)}-; L~a\x) r
+ x\~ v.{r(a + 2)(v +: +
1
)}-! L~a+I>(x) r
if n =2m+ 1,
 [ 7.71 1 GE NERALIZATTOl\S; APPLICATIO:-\S 185

where the complex numbers u., v. satisfy the condition

m m
(7.71.20) L: u.l + •=0
•=0
I L: v. 2
I
2
1 = 1.

(In the second sum Vm = 0 if n = 2m.) Now in both cases we have (cf. (5.1.7))

j(O) = I~ u. {rca+ l)(v ~ a)}-t L~Q)CO) r

~~{rca+ l)(v ~ a)}- IL~Q>(o)} 2 1
C7.71.21)

= !rCa+ 1)}-
1
f: (v +a)=
•=O v
!r(a + 1)}- 1 (m +a+
m
1),

and this is the required maximum.

If a = 0, we obtain

(7.71.22) f(O) ~ [n/2] + 1,

provided f(x) is a 7rn, non-negative for x ;S; 0. We can readily prove the more
general inequality
(7.71.23) e-xf(x) ~ [n/2] + 1,
where f(x) is subject to the same condition as in (7.71.22). To this end, let Xo
be an arbitrary positive number. Applying (7 .71.22) to
1
f(x + Xo) {1C() e-x f(x + Xo) dx}-
which satisfies the required condition, we find

+ xa) dx}- ~
1
f(xo) {1C() e-x f(x [n /2] + 1.
Whence

e-x0 f(x 0 ) ~ ([n/2] + 1)e-xo lr.<J e-xf(x + Xo) dx

= ([n/2] + 1) 1C() e-xj(x) dx ~ [n/2] + 1.

xu

See also Problem 42.

(6) Certain bounds for the orthonormal polynomials IPn(x) l can be derived
from the preceding results, provided the weight function w(x) satisfies an
inequality of the type
C7 .71.24) a~ x ~ b;
186 I.l';EQUALITIES [VII]

or
(7.71.25) w(x) ~ JJ.(x - a)"(b - xl, a < x < "b, a > - 1, (3 > - 1;
or
(7. 71.26) x > 0, a> - 1.
In the first and second cases a and b arc finite.
For instance, under the condition (7.71.24) we obtain

(7. 71.27)

Consequently, aceording to Theorem 7.71.1,

Am,
(7.71.28) I P'n(Xo) I < · a < Xo <b.
1 A'[(xo- a)(b- xo)r 1m!,
Here A and A' are positive constants depending only on a, band JJ..
A similar argument applies to the cases (7.71.25) and (7.71.26).
If we assume that 1D(x) satisfies a Lipschitz condition, the bounds m and m!
in (7.71.28) can be replaced by m! and 1, respectively. (Cf. (7.1.15).)
(7) Finally, by use of Theorem 7.32.4, we obtain the following generalization
of (7.71.28). Let w(x) ~ JJ. > 0, a = - 1, b = +
1, and k ~ 0, an integer.
Then
(sin 0)-k-!O(nk+!)
(7.71.29) p<;> (cos e) = ' 0 < 0 < 7r.
1O(n2k+l),
The bounds of the 0-terms depend only on JJ. and k.
For the proof we use an argument similar to that used in proving Theorem
7.71.2.
7.72. A problem of Tchebichef
(1) PROBLEM: Let w(x) be a weight function on the interval [a, b], and let W(x)
be a given real-valued function, defined on the same interval, and for which the
integrals

(7.72.1) k = 0, 1, 2, ·. · , n,

exist. Let f(x) be an arbitrary polynomial of fi.xed degree n, not identically zero,
and non-negative in a ;£ x ;£ b. To detc;mine the maximum and the minimum of
the ratio

(7.72.2) ~C f(x) W(x) dx : 1b f(x)w(x) dx.

Sec Tehcbiehcf 7. Fir:-;t let a and b b(~ finite. By using the representation
(7.71.9) again, we easily find that the quantities in question are the maximum
[ 7.72 J A PlWBLEi\1 OF TCHEB1CHI•;F 187

and minimum of the following quadratic forms in 1u.l and Iv.l:

lb {t u.p.(x)r W(x) dx + lb {~ v.q.(x)r(x - a)(b - x)W(x) dx

(7.72.3) if n 2m,
+ lb {"'
=
r }2 (b
l b{
~
m
u.rv(x) / (x - a)W(x) dx a ~ v,8,.(x) - x)W(x) dx

if n =2m+ I,
under the condition L~=o u; + L~=o v= = I. (In the first case Vm = 0.) Here
lp.(x)), lq.(x) l, lr.(x) l, ls.(x) l have the same meaning as in (7.71.9).
Let now a be finite and b = + oo. Then we have to consider the maximum
and minimum of the form

1C() f~ u.p.(x)ywcx)dx + lC() f(n~) v.q.(x)Ycx-

1 121
C7.72.4) a)w(x)dx

under the condition u; + L L:

v: = 1. Here lp.(x) land Iq.(x) l are the ortho-
normal sets associated with w(x) and (x - a)w(x), respectively, x ~ a.
In case a = - oo , b == + oo, we must consider the form
+r.<J f[ [ n/2]
(7.72.5) f -c<J
n/2]
l ~ u.p.(x)
}2
W(x) dx, L: u: =
•=0
I,

where lp.(x) l is associated with w(x) in [- oo, + oo ].

Thus, in all these cases, the problem in question if-5 reduced to the determina-
tion of the greatest and least characteristic values of a certain quadratic form.
In dealing with the sum of two quadratic forms in lu.l and lv.l, respectively,
we determine the greatest characteristic value of the single forms, and the
greater of these values is the maximum in question. A similar remark applies
to the minimum. The actual application of this method is difficult, however,
and certain mechanical quadrature formulas (see below) are often preferable.
Similar considerations apply if the integrals (7. 72.2) are replaced by Stieltjes
integrals.
(2) Leta=- I, b =+I, W(x) = xw(x). It suffices to determine the maxi-
mum and minimum of

(7.72.6) J_r+l (p(x)l

1
2
xw(x)dx: _
1+! /p(x)) w(x)dx,
1
2

if p(x) is an arbitrary 7rm, not identically zero with real coefficients. Having
2
done this, we must replace w(x) by (I - x )w(x), (I ± x)w(x), respectively; sec
below. Let x 0 , x 1 , . • . , .Xm be the zeros of the orthogonal polynomial Pm+I(x)
associated with w(x); according to (3.4.I), we find for the ratio (7.72.6), the
representation
m m
(7.72.7) r:
v=O
x.(p(x.) l x. :
2
L: x.(p(x.) 1
v=O
2
,

where X. denote the Christoffel numbers. Therefore, the maximum and mini-
mum in question coincide with the greatest and least zero of Pm+I(x), respec-
 188 II\'EQUALITIES [VII]

tively. (Cf. §3.4 (3).) If p denotes the greatest zero of p(x), it is seen from
(7. 72.3) that the maximum of (7. 72.2) is, in this special case,
max CPm+l , qm) if n =2m,
(7.72.8)
max (rm·H, §m+I) if n =2m+ 1.
The result for the minimum is similar.
(3) Here the general discussion of Tchebichcf ends (cf. 7, p. 395). We can
prove, however, that the expressions (7.72.8) are Pm-11 and rm+l, respectively,
so that the following theorem holds:

THEOREM 7. 72.1. Let w(x) be a weight function on the interval [- 1, + 1].

Let f(x) be an arbitrary 'Trn, not identically zero, and non-negative in [- 1, + 1].
Then the maximum of

(7.72.9) +! 1+!
_ f(x)xw(x) dx : _ f(x)w(x) dx
1 1 1

is the greatest zero of Pm+I(x) if n =2m, and the greatest zero of Pm+2( -1)Pm+1(x) -
Pm+I( -l)pm+2(x) if n = 2m + 1. Here !Pn(x) l is the .c;et of the orthonormal
polynomials associated with w(x) w the interval [ -1, + 1].
According to Theorem 2.5,

Pm(x) Pm+2(x)
2
(1 - x )qm(x) = const. Pm( -1) Pm+2( -1) ,
Pm(l) Pm+l(l) Pm+2(1)
(7.72.10) p.,(x) Pm+l(x)
(1 + x)rm(x) = const.
Pm(-1) Pm+I(-1)

Pm(x) Pm+l(x)
(1 - x)sm(x) = const.
Pm(1) Pm+I(l)

First, let ~ 0 > ~~ > ... > ~m be the zeros of Pm+I(x) in decreasing order. We
show that the first determinant in the right-hand member of (7.72.10) is non-
zero 0 if ~ 0 < x < 1. Indeed, according to (3.2.1)
Pm(X) Pm+!(X) Pm+2(X) Pm(X) Pm+I(X)
Pm( -1) Pm+l( -1) Pm+2( -1) = Am+2 Pm( -1) Pm+!( -1) -Pm+l( -1)

Pm(l) Pm+I(1) Pm+2(1) Pm(l) Pm+I(l) Pm+I(l)

h(x) 1 X

h( -1) 1 -1
h(l) 1 1
where h(x) = Pm(x)!Pm+I(x). Now, by using (3.3.9), we see that the last
determinant is positive in ~o < x < 1, since

--------------.,.-----
 [7.8 J FURTHER RESULTS 189
(x - ~.)-1 1 X

( -1 - ~.)-1 1 -1 - 2(1 - x 2 )
( 2
1 - ~.)(x- ~.)
> 0.
o - ~.r~ 1 1
On the other hand, h(x) decreases from + oo to - oo between ~ 1 and ~0 , and
from + oo to h(1) between ~o and 1 (cf. the proof of Theorem 3.3.4). Further-
~ore, h(- 1~ < 0, h(1) > 0. Thus the greatest zero of rm(x), or h(x)- h(- 1),
IS greater than the greatest zero of sm(x), or h(x) - h(I), x < 1.
(4) Tchebichef discusses in detail the case
(7.72.11) a = - 1, b =+ 1, w(x) = 1, W(x) = x.
(Problem of the "centroid," see loc. cit., p. 399; cf. also Szego 13, pp. 627-629.)
Save for constant factors, we now have
= Pm+l(x);
Pm+l(x) qm(x) = P~l,l)(x) = const. P~+ 1 (x),
(7.72.12) rm+l(x) = P~Cl_t~(x) = const. (Pm+ 1 (x) + Pm+2 (x)\(l + x)- 1 ;
Sm+l(x) = P~~C~)(x) = const. {Pm+l(x) - Pm+2(x)\(l - x)- 1•
This yields the following theorem:
THEOREM 7.72.2. Let f(x) be an arbitrary 'Trn, not identically zero, and non-
negative for- 1 ~ x ~ 1. Then the maximum of
+1 xf(x) dx : 1+1
(7.72.13)
1 _
1
_ f(x) dx
1

is the greatest zero of P m+I(x) if n = 2m; if n = 2m + 1, it is the greatest zero of

Pm+I(x) + Pm+2(x).
The distance from the maximum value to 1 is,...._, n-2 as n -7 oo. The minimum
is obviously the corresponding negative value.
For other refinements of the mean-value theorems, the reader is referred to
Tchakaloff 1. Cf. Problem 43. Concerning other extremum problems for poly-
nomials and connected inequalities, see Geronimus 2, 3, 4, and Shohat 2.
7.8. Further results
(1) From Theorem 1.82.i) we conclude the following refinement of Sonm's
Theorem 7.31.1. Let y = y(x) satisfy the differential equation (7.31.1) and let
y(x) have an infinite set !xml of zeros ordered in the inerea~ing way: x1 < x~ <
x 3 < · · · . The function .-p(x) should be positive, eontinuous, 1.tnd decreasmg.
J,rt p he a fixrd positive number. Then the integrals

{'•·I ly(x)l~' rh
... rv

are inereasing.
A similar statement holds if q;(x) is inereasing.
For p -7 0 this reduces to the assertion of Theorem 1.82.2. Taking the pth root
of the integral, for p ~ oc we obtain the assertion of Sonin's theorem. Hence,
190 Ir\'EQtTALITIES [VII]

the assertion above Is a common generalization of both theorems (Makai 2).

In Makai 2 one finds various remarkable speeial eases of this useful theorem.
(2) Let JJ.,·,n be the suceessive relative maxima of 1/\ (x )I when x decreases
from +1 to -1. We have (Theorem 7.3.1):
I > 1-'l.n > IJ.2,n > ··· > /J.h.n where h = [ni2].
::\f ow for H fixed r, we have (Szego 23)
(7.8.1) n~r+l.

These inequalities also hold for the relative maxima of 1P~"·ll\x) 11 pn<a.ll)(l)
when a={3>- t, Szasz 2, and from this for a>(3=- 1 by use of (4.1.5).
The same result for a> (3 > - t
is probable, but still open. Somewhat
surprisingly these inequalities fail for

p~O,-l)(x) = Pn(X) + Pn-l(x)

2

and graphical evidence suggests that the inequalities (7.8.1) are reversed
for this function. See Askey-Gasper 4 for a problem which would be solved
by these inequalities.
For the orthonormal Hermite functions defined by
Hn(X)e-x'/ 2
M;;(x) = 1l"l/4( 2 nn!)ll2'
Szasz 3 proved inequalities similar to ( 7 .8.1) and used them to prove
I M;;(x) I ~ ~(x) = 1!"-114.
Stronger inequalities in which the right-hand side goes to zero in n are
known. See Askey-Wainger 1 and Muckenhoupt 3. They also obtain refine-
ments similar to that given in Problem 40 and similar inequalities for Laguerre
polynomials.
(3) Many inequalities, involving in particular the classical orthogonal
polynomials, have been investigated by Karlin-Szego 1. Problem 70 (Turan's
inequality) is a specia~ case. See Gasper 5 for a Turan type inequality for
Jacobi polynomials.
(4) Theorems 7.31.2, 7.32.3, and 7.6.5 have been sharpened for a> - 1 by
Lorch 3.
(5) The inequalities (7.71.12) have been refined in Schoenberg-Szego 1.
(6) Turan 3 considers the problem of maximizing the Markoff-type functional

fo "'[1r~(x) re-xdx
fo"'[1rn(x)] 2e xdx

over polynomials 1l"n of degree ~ n. He shows that the exact maximum IS

1
2 sin 1r I (4n + 2)
 CHAPTER VIII

ASYMPTOTIC PROPERTIES OF THE CLASSICAL POLYNOMIALS

The consideration of the asymptotic properties of the orthogonal polynomials
!Pn(x) l, n -7 oo, leads to two fundamental problems: the asymptotic behavior
of' the polynomials in question outside the orthogonality interval, especially
in the non-real domain, and the asymptotic behavior on the orthogonality
interval itself. In general, the second problem is deeper and more difficult than
the first one. In our treatment we start with a discussion of Legendre poly-
nomials, obtaining various important asymptotic formulas for them. We
intend not only to give a survey of results, but also to point out the various
methods used. The extension to ultraspherical and general Jacobi polynomials
will also be indicated. The asymptotic investigation of Laguerre and Hermite
polynomials, in general, requires new considerations, although essentially the
same methods as before can also be applied to these cases.
The simplest special case,

the case of Tchebichef polynomials of the first kind, furnishes a good illustration
of the characteristic features of our results. If xis located outside the interval
[- 1, + 1], we can take I z I > 1, and we then see that
n -7 oo.

On the interval [- 1, + 1], we write z = ei 0, T n(x) = cos nO. Here the poly-
nomials have an oscillatory behavior.
These results need only a slight modification for Legendre, and even for Jacobi,
polynomials as long as x ~ ± 1. A new difficulty will, however, arise in the
vicinity of the end-points ± 1, which are in some respects exceptional. This
is ultimately due to the fact that the coefficient of dO in
(1- x)'"(l + x)fJdx = -(1 -cos O)a(1 +cos O)fJ sin OdO

vanishes, in general, or becomes infinite at 0 = 0 and 0 = 1r. When this occurs,

functions of the type cos nO are not suitable for the approximation of the poly-
nomials in question in the neighborhood of x = ± 1. For this purpose we shall
use certain Bessel functions.
Usually, the problems and results for Laguerre and Hermite polynomials are
similar. But it is rather curious that, in the corresponding asymptotic expres-
sions, the quantity n! appears instead of n. In the general Laguerre case,
Bessel functions are needed near x = 0, whereas for the Hermite polynomials
the origin x = 0 does not play an exceptional role. In both cases new difficulties
arise due to the fact that the interval of integration is infinite. For the expan-
191
I92 ASY:\IPTOTIC PROPERTIES OF CLASSICAL POLYX0:\11ALS [VIII]

sion problem it is of great importance to have asymptotic formulas that hold in

intervals which become infinite as n --+ oo.
8.1. The formulas of Mehler-Reine type
These important formulas are elementary in character, and we shall discuss
them briefly before entering into the general considerations of §§8.2I-8.23.
(I) THEOREM 8.1.1. Let a and {3 be arbitrary real numbers. Then

limn-a P~a.~) (cos:.) = lim n-a P~a,f3) (I -

2
(8.1.I) z ) = (z/2)-aJa(z),
n-oo n n-oo 2n 2
where J a(z) has the same meaning as in (1.71.I). This formula holds uniformly
in every bounded region of the complex z-plane.

For Legendre polynomials, a = {3 = 0, formula (8.1.I) is due to Mehler (3,

p. I40) and Heine (3, vol. I, p. I84). Concerning further literature we refer
to Watson 3, p. I55. In case a = {3 = 0, a very simple proof follows from the
first integral of Laplace [(4.8.IO), (1.71.6)]. For a = ± t the function in the
right-hand member of (8.1.I) is a constant multiple of z-1 sin z and cos z, re-
spectively (cf. _(1.71.2)). Formula (8.1.I) is trivial for the elementary cases
(4.1.7) and (4.1.8).
The proof can be based on (4.21.2). In fact, we have for the (v + I)st term
of (4.21.2) if x = cos(z/n), z and" fixed and n--+ oo, the following asymptotic ex-
pression:
I r(n +a+ {3 +" + I) r(n +a+ I) ( . 2 z )"
v!(n-v)! r(n+a+f3+I) r(v+a+I) -sm2n
(8.1.2)
~ v!r(v ~aa·+ I) (-f)".
Here we exclude the case of a negative integer a. Passing to the limit under the
summation sign is valid because of the existence of a dominant for the total sum
which is readily derived. Indeed, we have, if n is large enough,
n-a r(n +a+ {3 +"+I) r(n +a+ I)
(n - v)! r(n + a + {3 + I) 2•n2•

~ n-,a n"(2n + a + {3)" r(n ~ ~ + I) = O(I),

n. •n •
uniformly in v, 0 ~ " ~ n. The argument needs only a slight modification if a
is a negative integer.
Formula (8.1.I) gives a complete characterization of the function P~a,f1l(cos e)
for () = O(n-1). As an important consequence we note the following:

THEOREM 8.1.2. Let x 1n > X2n > · · · be the zeros of P~a,{J)(x) in [- I, +

I]
in decreasing order (a, {3 real but not necessarily greater than - I). If we write
[ 8.1 l FORMULAS OF MEHLER-HEINE TYPE 193

Xvn = cos Ovn' 0 < O.n < 'TT', thenfor a fixed v,

(8.1.3) lim nOvn = j.,

where j. is the vth positive zero of J ,.(z).

(2) Relation (8.1.1) enables us to derive some properties of Bessel functions
from the corresponding properties of Jacobi or Legendre polynomials. We use
the symbol (a)--+ (b) to indicate that in writing x = cos (z/n), a certain formula
(a) is transformed into another formula (b) by the limiting process n --+ oo.
Then we have the following relations:
(4.1.7) and (4.1.8) --+ (1.71.2),
(4.2.1) --+ (1.71.3),
(4.22.2)--+ J_z(z) = (- 1) 1Jz(z), l an integer,
(4.24.2) --+ (1.8.9),
each of (4.8.6), (4.8.10), (4.9.3) --+ (1.71.6) in the special case a = 0, and
(4.9.19) --+ (1.71.6) in the general case.
See also Problem 44.
Ji..,rom Theorem 1.91.3 (Hurwitz's theorem) the reality of the zeros of z-a J ,.(z)
follows for a > - 1. From (6.6.5) and (6.6.3) (or (6.6.2)) we obtain for the
positive zeros j. of J ,.(z), f.. = a + i,
(8.1.4) 11 = 1, 2, 3, · · · , - i ~ a ~ + t.
The upper bound can be replaced by (v + a)7r if - i ~ a ~ 0 (cf. (6.6.6)).
Furthermore (cf. Watson 3, p. 49, (1); cf. (7.31.5)),
(8.1.5) (7.33.1) --+rca + I)(z/2)-" 1 J ,.(z) 1 ~ 1, z > 0, a ~ - i,
(8.1.6)
The expansion (4.9.17), (8.21.5), and the inequality (8.21.6) of the remainder
term furnish the important formula (Stieltjes 8, p. 242):

(~)!I: (1·3 · · · (2v- 1) ) cos (z- (v + ih/2) + Ep(z),

2
Jo(z) =
'TT'Z v=O 2 · 4 · · · 2v 22• z•
(8.1.7) t
IEp(z)l ~
-
(.3.) (1·32...· 4 ·(2p• • 2p-1))2 2
'TT'Z
_1_,
2 P zP
Z > 0.

Thus the error Ep(z) is numerically less than the first neglected term (replacing
cos by 1). For p = 1 we obtain the special case a = 0 of (1.71.7) (with a
numerical constant in the bound of the remainder).
(3) THEOREM 8.1.3. Let a be arbitrary and real. Then for an arbitrary com-
plex z
(8.1.8) limn-a L~")(z/n) = z-" 12 J ,.(2zt),

uniformly if z is bounded.
194 ASYlVIPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII]

This formula is of a type similar to (8.1.1) and yields similar results. The
proof can be given along the same lines as there. In the special cases a = ± i,
we again obtain trigonometric functions. From this case an analogous formula
for Hermite polynomials can be derived. See Problem 45.
Both formulas (8.1.1) and (8.1.8) can be extended to an asymptotic expan-
siOn. Concerning the case of Laguerre polynomials, a = 0, see Moecklin 1, p. 28.

8.21. Asymptotic formulas for Legendre and Jacobi polynomials

From the point of view of the asymptotic problem, the Legendre polynomials
Pn(x) represent the simplest non-trivial case. We start with an enumeration
of some classical results concerning the behavior of Pn(x) as n -7 oo. The
proofs, based on various methods, are given in subsequent sections. In what
follows, E denotes a fixed number with 0 < E < 1r /2, so that the interval [E, 1r - E]
lies wholly in the interior of [0, 1r]; p is a fixed positive integer.
(1) THEOREM 8.21.1 (Formula of Laplace-Heine; Heine 3, vol. 1, p. 174).
Let x be an arbitrary real or complex number which does not belong to the closed
segment [- 1, +1]. Then as n -7 oo,
(8.21.1)

Here (x 2 - 1)- 114 , (x 2 - 1) 112 , and lx +

(x 2 - 1) 112 ) n+ 112 are real and positive if x
is real and greater than 1. This formula holds uniformly in the exterior of an arbi-
trary closed curve which encloses the segment [- 1, +
1], in the sense that the ratio
tends uniformly to 1.
THEOREM 8.21.2 (Formula of Laplace; Heine 3, vol. 1, p. 175).
Pn(cos e) = 2112 (7rn sin e)- 112 cos I (n + !)e - 7r/4) + O(n- 312
),
(8.21.2)
0 < () < 7r.
The bound for the error term holds uniformly in the interval E ~ () ~ 1r - E.

THEOREM 8.21.3 (Generalization of Laplace-Heine's formula). Let x be in

the complex plane cut along the segment[- 1,
1
+
1]; let x =Hz+ z- ), I z I> 1.
Then
1
p ( ) n ~ 1·3 · · · (2v- 1) -2•( 1 -2)-•-t
( . 1. ) " X =
82 3
gnZ f=t g. (2n- 1)(2n- 3) · · · (2n- 2v + 1) Z - Z

+ O(n-p-! Iz n.
Here g. has the same meaning as in (4.9.2), that is,
1·3 · · · (2v - 1)
go= 1; g.= "= 1, 2, 3, ....
2·4 · · · 2v
Formula (8.21.3) holds uniformly in the same sense as in Theorem 8.21.1.

------------··-----·----····
[ 8.21 J LEGENDRE AND JACOBI POLYNOMIALS I95

THEOREM 8.21.4 (Darboux's generalization of Laplace's formula; Darboux 1,

p. 39).

p n(cos e) = 2gn L
p-1 ( )
g.---.- · l· 3 . . . 2" - I______ .
v=O (2n- I)(2n- 3) • • · (2n- 2v + 1)
(8.2I.4)
. cos I (n - " + !)e - (v + ~]71"/2) + O(n-p-l)
~~n~~ '
O<e<1l".
Here gn has the same meaning as in Theorem 8.21.3. The bound for the error term
again holds uniformly in E ~ e ~ 1r - E.
THEOREM 8.21.5 (Stieltjes' generalization of Laplace's formula; see Stieltjes
7, 8).
4 2·4 · · · 2n
Pn(cos 0) - ; 3 . 5 ... ( 2n +I)

~h cos ((n +" (+ .!)e- (v + !h/2)

(8.2I.5) £.......•
-o )t
2mne~
+ R pe,
()

Here h. has the same meaning as in (4.9.I8), that is,

ho =I; h -
_ I· 3 · · · (2v - 1)
-
! ~ · · · (v - !) --
• 2 · 4 · · · 2v (n + V(n + D· · · (n + " + ! ) '
"= I, 2, 3, ....
We have
4 2·4 · · · 2n M
(8.21.6) < -1r 3 · 5 (
· · · 2n + I )- hp ( 2 sm e) P+,1 '
.

M = max (J cos e1-\ 2 sin e).

The factor M is between I and 2. Thus the error is numerically less than twice the
first neglected term (replacing cos by I).
THEOREM 8.21.6 (Formula of Hilb; Hilb 1).
(8.21.7) Pn(cos e) = (ejsin e)tJ0 ( (n +!)e) + O(n-l),
uniformly for 0 ~ e ~ 1r - E. More precisely, for the error term we have the
bounds
e!O(n-~) if cn- 1 ~ e ~ 7r - E,
(8.2I.8)
e20(1) if 0 < e~ en-t,
where c is a fixed positive constant.
(2) Some of these results can be extended to Jacobi polynomials. The
extension of (8.2l.I) is due to Darboux (1):
 19G ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII]

THEOREM 8.21.7. Let a and {3 be arbitrary real numbers. Then

p~a,f3)(x) ,.__, (x _ 1)-a/2(x -t-1)-f3/2((x + l)t + (x _ l)t)a+f.
(8.21.9)
. (27rn)-t(x2 - 1)-t {x + (x2 - l)t l n+t,
where xis outside of the closed segment [ -1, +
1]. Thz:s formula holds umformly
in the same sense as in Theorem 8.21.1. The determination of the multivalued
functions occurring in this formula is obvious.

The extension of (8.21.2) is also due to Darboux (1); this is the important
formula to which we referred in §7 .32:
THEOREM 8.21.8. Let a and {3 be arbitrary real numbers. Then
p~a,f3)(cos e) = n-tk(e) cos (Ne + 'Y) + O(n-~),
(8.21.10) k(e) = 71'-!
e
( sin 2
)-a-! ( cos
e)-{3-! ,
N = n + (a+ {3 + 1)/2,
2
'Y = -(a+ !h/2, 0 < e < 71'.
The bound for the error term holds umformly in the interval [ E, 7r - E].
The extension of Theorems 8.21.3 and 8.21.4 to Jacobi polynomials is readily
achieved. However, the law of the coefficients is, in this case, rather com-
plicated.

THEOREM 8.21.9. Let a and {3 be arbitrary real numbers. There exists a se-
quence of analytic functions ¢.(z) =¢.(a, {3; z) which are real for real z and regular
for I z I > 1 and I z I = 1, z ~ ± 1, such that
p-1
(8.21.11) z-n P~a,f3)(x) = L ¢.(z)n-•-! + O(n-p-!);
v=O
x = Hz+ .Z- 1), J z 1· > 1,
uniformly for I z I ~ R, R > 1.
Furthermore,

(8.21.12) p~a,{3)(cos e) = 2m {
einO L
p-1
¢.(ei 8)n-•-!
}
+ O(n-p-!), 0 < e < 71',
•=0

uniformly for E ~ e ~ 7r - E.

These extensions attain, in the ultraspherical case, the following more pre-
cise form:

THEOHEM 8.21.10. Let x = Hz + z- I z I >

1
), 1, and f.. > 0 or f.. < 0, f.. ~
-1, -2, -3, · · ·. Then

(I-) n ~ (1 - f..) (2 - f..) · · · (v - f..)

P n (x) = anZ .L.J a. ( )( ) ( )
(8.21.13) v=o n + f.. - 1 n +f.. - 2 · · · n +
f.. - "

--·--------
[ 8.21 l LEGENDRE AND JACOBI POLYKOMIALS 197

Furthermore,

P n(l-l( cos ll) _ ~ (1 - A.)(2- A.) · · · (v -A.)

v - 2an£....... a.-:---:---:-:-- ---:-
v=O (n +A - 1) (n +A.- 2) · · · (n +A.- v)
(8.21.14)
. cos l(n- "+ A.)e- (v -t_~h/~ + O(nl--p-l) 0 < 0 < 1r.
(2 sin e)•+>- '
Here a. has the same meaning as in (4.9.21). Regarding the uniformity, the same
remark holds as in the previous theorem.
The special case of an integral value of A. is discussed in §8.4 (5).
An extension of Theorem 8.21.5 to ultraspherical polynomials is the fol-
lowing:
THEOREM 8.21.11. Let 0 < A. < 1. We have

P n(A)( COS ll) . '\ r(n + 2A.) ~ r(v + A.)r(v -A+ 1)

V = (2/ 7r ) Sin /\7r ---:---:--- £.......
r(A.) •=o v!r(n +"+A.+ I)
(8.21.15)
. cos ( (n + " + A.)e - (v + A.)7r /2) + R (e) 0 < e < 7r,
(~sin e)•+>- p '

where

(8.21.16) I R 1,(e) I < (2/7r)sin A7r r(n + 2A.) r(p + A.)r(p- A.+ 1) .M .
r(A.) p!r(n + p +A.+ I) (2sme)P+>-
Here M has the same meaning as in Theorem 8.21.5.
(3) Finally, we mention the following formula of "Hilb's type" (cf. Szego
17, p. 77; Rau 2, pp. 691-692).
THEOREM 8.21.12. Let a > -1, and let {3 be arbitrary and real. Then we
have

( . e)"( e)2 p<a {;)(

13
sm 2 cos n • cos e) = N_"r(n +a+ 1) (I.
. --;nr---- e sm e )!J"(N)
e
(8.21.17)
( e!O(n-~) if en- 1 ~ e ~ 7r - E,

+ l
e"+ZO(n ") if 0 < e ~ en-r,
where N has the same meaning as in (8.21.10); e and E are fixed positive numbers.
Obviously, the remainder term is always e~O(n-l). If we use (4.1.3), a
similar formula can be obtained in the intervals E ~ e ~ 1r - en- and
1

1r - en - 1 ~ e < 1r provided {3 > -1. In view of (I. 71. 7) this leads to the
following important result:
THEOREM 8.21.13. Let a > -1, {3 > -1. We have, with the same notation
as in (8.21.10),

·-----------------·-----····
198 ASY~IPTOTIC PHOPEHTIES OF CLASSICAL POLYNOMIALS [VIII]

P~a,f3>(cos 0) = n-tk(O) {cos (NO+ 'Y) + (n sin 0)- 0(1)),

1

(8.21.18)
< <
en-1 =0=1r-cn.
-1

Here cis a fixed positive number.

This formula (Szego 17, p. 77) is more precise than (8.21.10); for a = {3 -
A - ~. 0 < A < 1, it follows from Theorem 8.21.11, p = 1. The restriction
a > -1, {3 > -1 is not essential (cf. Szego, loc. cit.). See also Obrechkoff 2.
(4) Analogous formulas hold for Jacobi's functions of the second kind Q~a,tl>(x)
and, particularly, for Legendre's functions of the second kind Qn(x), if x is
in the cut plane, as wrll as for Q"(cos 0) if 0 < () < 1r. Here we point out
only the analogue of Laplace':-:; formula (8.21.2):
Tm~ORI~M 8.21.14. For 0 < () < 1r
(8.2l.HJ) Qn(cos e) = 7r;(2n sin e)-t cos I (n + ~)e + 7r/4l + O(n-j).
'I' his holds uniformly in the interval [E, 1r - E].
8.22. Asymptotic formulas for Laguerre and Hermite polynomials
Similar, but slightly more complicated, formulas hold for Laguerre and
Hermite polynomials. In what follows n --t oo; we denote by E and w fixed
positive numbers, E < w, by p a positive integer.
(1) THEORI<~M 8.22.1 (Fejer's formula; Fejer 3). Let a be an arbitrary real
number; we have
1-'n(a) (,x) = 1r-t ex/2 x-a/2-l n a/2-l \ nx )! _ .....
cos 12( rv""/2, _ ""/4)
,
(8.22.1)
X> 0.

The bound for the remainder holds uniformly in [E, w].

THEOREM 8.22.2 (Perron's generalization Fejcr's formula; Perron 2, p. 78,

of
(49)). Let a be an arbitrary real number; we have for x > 0
L~">(x) = 12
7r-!ex 12 x-" 12 - 1n" -l cos (2(nx)t- a7r/2- 7r/4)

·{I:.-o A.(x)n-" 12 + O(n-p 12 )}

(8.22.2)
+ 1r-! e x/2
x-a/2-l n a/2-l sm
• \ nx )! - a1r /2 - 1r /4)
12(

·{2=: B.(x)n-"
•=0
12
+ O(n-p 12 )},
where A.(x) and B.(x) are certain functions of x independent of nand rp,gular for
x > 0. 'l'he bound for the rp,mainder holds uniformly in [E, w].

\Ve notice that Ao(x) = 1 and Bo(x) = 0.

[ 8.22 J LAGUEJWE A:\D HEIL\fi'I'I-: POLY:\0\IIALS 199

THEOREM 8.22.3 (Perron'~ formula in thr complex domain; loc. cit.). Let a
be an arbitrary real number. Then
1 -} ex/2( -x)-a/2-! n a/2-·! exp (2( -nx )t)
L (na(x) -_ z1T'
(8.22.3)
·f~ Cv(:r)n-" 12 + O(n-p 12 )},
where Cv(x) is again independent of n; it is regular in thr com7Jlrx plane cut along
the positive part of the real axis. Formula (8.22.:3) holds if x is in the cut plane
mentioned; (- x)-a/Z-i and (- x)l must be taken real and positive if x < 0. The
bound for the remainder holds wuformly in every closed domain w1'th no points in
common with x ~ 0.
Here \Ve have Co(x) = 1.
THEOREM 8.22.4 (Asymptotic formula of Hilb't> type). For a > -1 we have

e-x/2Xa/2L~a)(x) = N-a/2 :r_~n_±~.±-~) Ja(2(Nx)ll + O(na/2-'1),

(8.22.4) n.
N = n + (a + 1) /2, x > 0,
the bound holding uniformly in 0 < x ~ w. More precisely, the following bounds
are valid:
x5/40(n a/2-~) • en-1 =< x =< w,
"f
(8.22.5)
xa/2+20(n a)

In case a = 0 the last bound is to be replaced by x 2 log (x- 1n- 1); in (8.22.5)
cis a fixed positive number.
Evidently, the remainder term (8.22.5) is equal to x 514 0(na 12·-'l) throughout
0 <X~ w.
As a consequence of (8.22.4) we obtain the following analogue of (8.21.18),
which is more precise than (8.22.1).
THEOREM 8.22.5. Let a > -1 and cn- 1 ~ x ~ w; thrn
L~a) (x)
(8.22.6)
= 71

Here c and w arc fixed poxz'tivr constants.

We observe that N! - nl = O(n-l) where N hm; the sn,nw meaning as in
(8.22.4).
(2) Substituting a= -1-~ in (8.22.4), we find, by using (5.6.1) and (l.7l.2),
a formula of Hilb's type for Hermite polynomials. This is containC'd in the
more general theorem:
THEOREM 8.22.6 (Asymptotic expansion for Hermit<~ polynomials). For a
1·cal x
 200 ASY:\1PTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII]

212 p-1
A.-;. e-x Hn(x) =cos (N!x- mr/2) L u.(x)N-•
1

·-0
(8.22.7) p-1
+ N-! sin (N!x- n1rj2) L v.(x)!V" + O(n-p),
•=0

N = 2n + 1,
where
r(n +
1)
or r(n + 2) -!
A.n = r(n/2 + I) r(n/2 + 3/2) N '
according as n is even or odd. The coefficients u.(x) and v.(x) are polynomials
depending on v; they contain only even and odd powers of x, respectively. The
bound for the error term holds uniformly in every finite real interval. whether it
contains the origin or not.

For arbitrary n, we have A.n = (r(n + 1)/r(n/2 + 1))\1 + O(n- 1) l, and

uo(x) = 1, vo(x) = x 3/6, so that

r(n/2 + 1) -x2!2H ( ) (N! / )

r(n + l) e nX = COS X - n1r 2
(8.22.8)
3
+ ~ N-! sin (Ntx- n1rj2) + O(n- 1).

THEOREM 8.22. 7 (Asymptotic expansion for Hermite polynomials in the

complex domain). The expansion (8.22.7) holds in the complex x-plane if we
replace the remainder term by exp \N!I S(x) llO(n-p). This is true uniformly
for I x I ;£ R where R is an arbitrary fixed positive number.

(3) Finally we deal with another type of asymptotic formulas requmng a

more elaborate consideration.

THEOREM 8.22.8 (Formulas of Plancherel-Rotach type for Laguerre poly-

nomials). Let a be arbitrary and real, E and w fixed positive numbers. We have
(a) forx = (4n + 2a + 2) cos 2 ¢, E ;£ ¢ ~ 1r/2- En-! ,

e-x/2L(a)
n
(X ) = ( - l)n( 7r Sin ,~,.)-! -a/2-i a/2-i
• '+' X n
(8.22.9)
\sin [(n +(a +1)/2) (sin 2¢- 2¢) + 37r/4] + (nx)-!0(1)1;
(b) for x (4n + 2a + 2) cosh 2 ¢, E ;£ ¢ ~ w,

e-x/2 L~a) (x) = H -lt(7r' sinh¢)-! x-a/2-i na/2-l

(8.22.10)
·exp \(n +(a+ 1)/2)(2¢- sinh 2¢)}\1 + O(n- 1)};
(c) for x = 4n + 2a + 2 - 2(2n/3) 1t, t complex and bounded,

·-------------------·-···· .
[ 8.23 l REMARKS ON PRECEDIKG RESULTS 201

(8.22.11)
where A(t) is Airy's function defined in §1.81.
In all of these formulas the 0-terms hold uniformly.
The corresponding formulas for Hermite polynomials (Plancherel-Rotach 1)
are given by the following:
THEOREM 8.22.9. Let E and w be fixed positive numbers. We have
(a) for x = (2n + l)t cos ¢, E ~ ¢ ~ 1r - E,
e-x2f2 Hn(x) = 2n/2H(n!)t(7rn)-l(sin ¢)-t
(8.22.I2) 1
·{sin [(n/2 + t)(sin 2¢- 2¢) + 37r/4] + O(n- ) } ;
(b) for x = (2n + I)t cosh ¢, E ~ ¢ ~ w,
e-x2f2 Hn(x) = 2n/2-i(n!)t(7rn)-t(sinh ¢)-t
(8.22.I3) 1
· exp [(n/2 + t)(2¢- sinh 2¢)] {I+ O(n- ) l;
(c) for x = (2n + I)t - 2-t3-!n-1t, t complex and bounded,
(8.22.14) e-x 212 Hn(x) = 3!7r-i2n/2+i(n!)tn-1112 {A(t) + O(n- 1) }.
In all these formulas the 0-terms hold 1tniformly.
Note that (8.22.I2) holds uniformly in the vicinity of x = 0.

8.23. Remarks on the preceding results

(I) Of all the formulas enumerated in §8.2I, formula (8.21.I) has the simplest
character. It can be proved by various methods. An extension to an asymp-
totic series is given by (8.21.3). Corresponding formulas hold for Jacobi poly-
nomials (§8.2I (2)). The following simple consequence of (8.21.9) is important
for various purposes:

(8.23.I) n-'> oo;

here x is in the cut plane. The right-hand member is >I and represents the
sum of the semi-axes of the ellipse with foci at ± 1 and passing through x.
We compare (8.23.I) with the following formula for Jacobi's functions of the
second kind:

(8.23.2)

Here y is again in the cut plane and the right-hand member is < 1. (Cf.
(8. 7l.I9) .)
We shall give also several proofs for the classical formula (8.21.2) of Laplace.
It can be similarly generalized in various directions.
Darboux's formula (8.21.4) is the most important illustration of the method
due to him (1). This method furnishes asymptotic formulas for the coefficients
202 ASYMPTOTIC PROPERTIES OF CLASSICAL POL YI\0~1 IALS [ VI II ]

of power series whose singularities on the circle of convergence have a certain

simple character. The same method yields (8.21.3), as well as similar expan-
sions of Jacobi polynomials. (Cf. §8.4.)
The significance of the formula of Stieltjes is due to its unrestricted validity
in the interval 0 < e < 1r (although the bound for the remainder given by
(8.21.6) becomes infinite if e ~ 0 or e ~ 1r). In view of this fact, it can be used
in various cases not only in a fixed interval in the interior of [0, 1r], but even in
the vicinity of the end-points where the formulas of Laplace and Darbonx
fail in general. For p = 0 it holds in the sense that Pn(cos e) = R0 (8). Then
(8.21.6) furnishes inequality (7.3.8) but with a larger factor on the right side
(with 2(2/7r)! instead of (2/7r)!). From the formula of Stieltjes an arbitrary
number of terms in (8.21.4) can readily be derived. However, it seems diffi-
cult to obtain the general law of the coefficients of (8.21.4) in this manner.
The importance of Hilb's formula also lies in its unrestricted validity in the
neighborhood of e = 0, with the additional advantage that the remaindrr term
tends to 0 uniformly in this neighborhood. ·It furnishes the .1\Iehler-Heine
formula (8.1.1) immediately and yields Laplace's formula (8.21.2) by means
of (1.71.7). The bounds (8.21.8) are a slight improvement over Hilb's result.
Our proof ( §8.62) is essentially the same as that of Hilb. Szego (15) gives
an asymptotic expansion in terms of Bessel functions of increasing order and
generalizes Hilb's result. Analogous formulas hold for Legendre's function of
the second kind. Szego also obtains (15, p. 450) an analogue of (8.21.1) which
holds in the cut plane, arbitrarily near its boundary. This formula involves
Bessel functions with imaginary arguments.
Another formula of a type similar to (8.21.7) has been given by Watson (2)
with a numerical estimate of the remainder. It involves Jo(z) and Yo(z).
Theorem 8.21.12 is the extension of Hilb's formula to Jacobi polynomialR.
The proof given in §8.63 follows the same line of argument as that in §8.62.
The proofs of Theorems 8.21 .1-8.21.14 are based on the following methods:
(a) Explicit series or integral representations;
(b) Darboux's method;
(c) method of Liouville-Stekloff (method of the integra-differential equation);
(d) method of steepest descent.
A short survey of these methods will be given at the proper place:-;.
(2) Fejcr's proof of (8.22.1) is baRed on the grnerating function (5.1.9),
which in this case has the essential singularity w = 1 on the circle of con-
vergence I w I = 1. This argument is of a charactrr ,<;imHar to Darboux's
method. The more complicated type of singularity in thiR ca~e naturally
requires a more careful diRcussion; it is carried out by Fcjer by an elementary
method similar to the second mean-value theorrm of the integral calculus.
In Perron's first proof of (8.22.3) (in the special case p = 1, see 1), complex
integration is used. He obtains the complete rxpansions (8.22.2) and (8.22.3)
by using certain general asymptotic results concerning confluent hypergeometric
functions.
r s.2a1 REMAkKA ON PRECEDING RESULTS 203

Further proofs of Fejer's formula (partly giving more exact bounds for the
remainder and holding on certain segments the end-points of which tend to +O
and + oo) have been given by Rotach (1), Szego (10), and Kogbetliantz (14).
They use either the method of steepest descent or similar arguments. Fejer's
i()rmula is contained in (8.22.4); the latter result follo:ws from certain general
asymptotic theorems of Hilb's type due to Wright (1, p. 261, however only
for fixed x). We give a proof for (8.22.4) by using the Liouville-Stekloff
method (§8.64).
The special cases a = ±! are equivalent to Hermite polynomials (see
(8.22.7)); in these cases, Fejer's theorem was known previously by Adamoff (1).
Adamoff obtains the remainder term with certain numerical bounds.
We derive (8.22.7) by using the method of Liouville-Stekloff. Uspensky's
formula (5.6.5) then leads immediately to a corresponding asymptotic expan-
sion for Laguerre polynomials involving Bessel functions. We indicate this in
§8.66. Concerning this expansion see Wright, loc. cit.; its first term is (8.22.4).
From this expansion Perron's formulas (8.22.2) and (8.22.3) follow readily. A
second proof of these formulas can be based on the method of steepest
descent (§8.72).
The first term of the asymptotic expansion mentioned for Hermite poly-
nomials furnishes (8.22.8). Adamoff's formula is less precise. (On the other
hand it contains numerical constants.) Comparison of (8.22.8) with Us-
pensky's formula (1, p. 597, (6)) indicates that it is convenient to work with
N = 2n + 1 instead of N = 2n.
A very detailed asymptotic investigation of Hermite polynomials is due to
Watson (1, second paper).
We mention the following simple consequence of Theorems 8.22.3 and 8.22.7:
Let x be in the complex plane cut along the non-negative real axis. Then

(8.23.3) n-+oo.

Here ( -x); is taken real and positive if x < 0. However, if x is non-real,

(8.23.4) n-+oo.

Theorems 8.22.8 and 8.22.9 arc closely related to the important results of
Planchcrel-Rotach (1). These authors deal exclusively with Hermite poly-
nomials and use the rnethod of steepest descent; they obtain a complete asymp-
totic expansion in all three cases of Theorem 8.22.9. Their argument has been
applied to Ln(x) by Moecklin (1). We shall derive (§§8.73-8.75) only the
principal terms of these expansions, however, for general Laguerre polynomials
L~a> (x), by using the method of steepest descent. Our argument is based on
the generating function (5.1.16) and on the asymptotic expansion of Bessel
functions ~n the complex domain.
 204 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYI\0:\rJALS [ VIli]

The formulas (8.22.9)-(8.22.11) describe Laguerre polynomials in the

"oscillating region," in the region beyond this, and in a certain vicinity of the
largest zero, respectively. The same is true for (8.22.12)-(8.22.14).
Van Veen (1, 2) derives the asymptotic series corresponding to (8.22.12)
with numrrical estimates. Schwid (1) applies the method of Liouville-Stekloff
(in the more precise form dur to Langer (1)) to the asymptotic investigation of
Hermite polynomials.

8.3. "Elementary" proof.of the formulas of Laplace-Heine and Laplace

(1) 'Ve start from the representation (4.9.4) and first prove (8.21.1). Let
x = Hz + 1
z- ), I z I > 1; then
n n
(8.3.1) '"" g",gn-mZ n-2m = g,z n
P n(X ) = L._, '"" gn-m g"'z -~m .
L..., ·--·
m=O tn=O gn

We next show that

·
(8.3.2) hm Ln ( gn-m
n....,.oo m=O
···--
gn
- 1) g"'z •
-2m
= 0,

uniformly for I z I ;?; R, R > 1. Indeed, the expression in the brackets tends to
zero if m is fixed and n ----+ Cl). On the other hand, it is easy to find a dominant;
we have, for instance,

0~ (gn-m
gn
_ 1) gm ~ gn-mgm •
gn
Now (n + 1);gn is bounded from zero and from infinity, and
(n + 1)! 5 1
0 ~ m ~ n.
(n - m + 1)!(m + 1)! - '
(2) We can prove (8.3.2) in another way, by use of certain very elementary
properties of the sequence I gn}. Let 8 > 0 be arbitrary, and let M be a positive
integer, such that
00

L R-2m < 8.
m=M+l

The numbers gn-m/gn - 1 are positive and increase with m; we therefore have,
if n > M,

The last expression tends to 0 as n ----+ Cl) since gn/ gn-I ----+ 1. On the other hand,

t
m=M+l
(gn-m _
gn
1) gmR-2m < t m=M+l
gn_tngm R-2m'
gn

--- --·-···"'··
[ 8.3] FORMULAS OF LAPLACT~-REINE AND LAPLACE 205

Now gm/gm~i is increasing, so that

- gn-:-rr.gn,_ ;;£ 1 or S; 1
gn-m-tlgm-1

according as m ;;£ (n + 1)/2 or m ~ (n + 1)/2. Consequently, gn-mgm attains

its maximum, as m varies from 0 ton, form = 0 or m = n; hence gn-mgm ;;£ gn,
so that

t
m-M+l
(Yn-m-
gn
1) gmR-2m ;;£ t
m-M+l
R-2m < f
m=M+l
R-2m <D.

This establishes (8.3.2).

Therefore the expression
n
(8.3.3) (gnzn)- 1Pn(X) - L gmz-?.m
m=O

tends to 0 as n --? oo, uniformly for I z I S; R, R > 1. The last sum tends to
2
(1 - z- )-!; whence (8.21.1) is readily derived.
(3) For the proof of (8.21.2) we use (4.9.3). We have for 0 < e < 1r
[n!2] }
P n(cos e) -_ 2 gnJl
(I)
{ e
-in8 ~
'"' gn-m
- gme 2im8
m=O gn
(8.3.4)
m f -in8 [n/2]
= 2gnJt)C """' ( ·----
~ gn-m - 1) gme 2im0 } +2 r
"' -in8 [n/2]
g,,UL)C """' fJ,,e 2im8 } '
~
l -o gn l fu=O

where .L:' has the same meaning as in (7.4.5). The sequence (gml tends
monotonically to 0, so that for 0 < e < 1r
00

""" gme 2im8 = (l - ez;o)-l = ei(7r/4-8t2>( 2 Sin

~
. e)-; .
m=O

This series converges uniformly for E ;;£ e ;;£ 1r - E.

Now, if 8 is an arbitrary positive number, we determine the positive integer !vi
so that

(8.3.5) M" > M' > M.

The numbers gn-m/gn - 1 being positive and increasing with m if n > 2.M, we
have, according to Abel's inequality,

(8.3.6)

where K is a fixed constant. On the other hand,

20G ASY:\IPTOTIC PROPERTIES OF CLASSICAL POLY~O:\rJALS [VIII J

L'
[n/2] ( Yn-m _ 1) gme2im8 I
(8.3.7)
I m=M+1 Yn

~ (Yn-[n/2] -
Yn
1) max
M<l'~(n/2]
fJ ;m(J2im81 < D
I m-1'
1
Yn-[n/2] •
Yn
Since Yn/Y2n is bounded, we see that the first 2:' in the right-hand member of
(8.3.4) tends to zero. Hence we have
(8.3.8)
where on--+ 0 uniformly in E ~ e ~ 1r - E. This is the formula of Laplace with
a remainder term o(n-i).
(4) Although they do not lead to the remainder term O(n -I) stated in (8.21.2),
these elementary arguments are important since they use only some very
simple properties of the sequence IYn ). At the same time they yield certain
asymptotic formulas for Fejer's polynomials Fn(x), introduced in §6.5, valid
in the cut plane and in -1 < x < + 1, respectively, provided certain condi-
tions regarding the sequence Ian) are satisfied. We have the following:
THEOREM 8.3. Let (am) be a positive sequence, am--+ 0, and let am/am-l j 1.
Then the following asymptotic formula holds:
00

(8.3.9) Fn(x) "-' anZn .L: amz-2m, n-+oo,

m-0

where xis in the cut plane, x = Hz + z- I z I >

1
), 1. Ij-in addition am/a2m
remains bounded, we have

(8.3.10) Fn(cos e) = 2anm{eino t

m-0
ame-2imB} + o(an), n--+ oo,

where 0 < e < 1r. Both formulas are valid uniformly in the same sense as (8.21.1)
and (8.21.2), respectively.
The generalization (8.3.9) of the Laplace-Heine formula is new; concerning
the generalization (8.3.10) of Laplace's formula, see Szego 11, pp. 186-187.
We observe that the series in (8.3.9) is convergent, and the series in (8.3.10)
is uniformly convergent in E ~ e ~ 1r - E. With regard to the latter fact we
note that cx.m/am-1 ~ 1, so that am is decreasing. As an application of Theorem
8.3, we obtain the analogue of (8.21.1) and (8.21.2) (with a less precise estimate
of the remainder in the second case) for the ultraspherical polynomials P~>..\x)
provided >-. > 0.

8.4. Darboux's formula proved by Darboux's method

(1) We shall prove formula (8.21.4), as well as others, by means of an impor-
tant method due to Darboux (1), and we begin with an illustration ~f this
method. Supposing 0 < e < 1r, let us consider the generating function of
[ 8.4 J DARBOUX'S FORMULA 207

Legendre polynomials (cf. (4.7.23))

m the neighborhood of ei6 :

h(w) = (1 - we-i6)-! (1 - iiB)-} {1 - e2•6e2i6- 1

(1 - we-io)}-l
(8.4.2)
= (1 - e2i6)-l f: g. ( e'2.:2i6- 1)" (1 -
•=O
we-iO)•-l.

A similar representation holds in the vicinity of e-io. Denoting the Lth partial
2
sums of these expansions (stopping at the terms v = L) by siu(w) and si >(w),
respectively, let us consider the difference
(8.4.3)
We see immediately that the Lth derivative H<L>(w? possesses continuous
boundary values in I w I ~ 1. Thus if we expand H(w) in a power series about
w = 0, the coefficients of H\L) (w) tend to 0. This simple remark shows that
the coefficients dn of H(w) satisfy the condition
(8.4.4)

Each of the terms of the finite sums siu(w) and si2 >(w) has only one singu-
larity on the ·unit circle. The vth term of si1 >(w) contributes to the coeffi-
cient of wn in h(w) (and therefore to Pn(cos e)) an expression of the form

(8.4.5)

The sum si2 >(w) contributes the conjugate of (8.4.5). Both terms are O(n_"_').
For a fixed value of p the coefficient of wn in H(w) is of higher order than n-p-i,
provided L is sufficiently large.
By use of the same argument, the follmving general theorem can be obtained:
THEOREM 8.4. Let h(w) be regular for I w I < 1, and let it have a finite number of
singularities
(8.4.6)
i<J>,
e ,e
i<i>2
, ··· ,e
i<l>l
, ei<Pa ,....
_.r.
ei<J>p , a~ {3,

on the unit circle I w I= 1. Let

00

(8.4.7) h(w) = L c~k)(l - we-•<~>krk+•b\ k = 1, 2, ... 'l,

•=0
in the vicinity of ei<Pk, where bk > 0. Then the expression

(8.4.8) f: j: c~k> (a" +n vbk)( -ei<~>kr

v=O k=l

·---···-----·····.
208 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII ]

furnishes an asymptotic expansion for the coefficient of wn in h(w) in the following

sense: if Q is an arbitrary positive number, and if a sufficiently large number of
terms is taken in the sum L~o in (8.4.8), we obtain an expression which approxi-
mates the coefficient in question with an error equal to O(n-Q).
A simple discussion shows that it suffices to stop at the term v = p - 1 of
the sum, where p is a positive integer such that
(8.4.9)

Proper logarithmic singularities of h(w) can also be admitted.

(2) In the case of Pn(cos e), this asymptotic expansion becomes

(8.4.10)

The general term is O(n -·-!);thus, if we stop at v = p - 1, the error is O(n -p-!).
This agrees with formula (8.21.4). It is also clear that the bound for the
remainder holds uniformly in E ~ e ~ 1r - E.
The same method can be used for the proof of the expansion (8.21.3), which
corresponds to the Laplace-Heine formula (8.21.1). (Actually, this case is
simpler than the preceding one.) Indeed, let I z I > 1; then
h(w) = (1 - zw)-!(1 - z- 1 w)-;
(8.4.11)

whence (8.21.3) readily follows.

(3) The infinite series which corresponds to Darboux's formula (8.21.4) is
convergent in the ordinary sense and represents Pn(cos e) provided 2 sine > 1,
that is, 1r /6 < e < 57r /6. In fact, the representation (8.4.2) holds uniformly
near w = 0 if
(8.4.12)
See 8.92 (4).
(4) Darboux's method also applies to the general Jacobi, and in particular
to the ultraspherical, polynomials and leads to the Theorems 8.21.9 and 8.21.10.
Another method of deriving the expansions of Theorem 8.21.9 will be indi-
cated in §8.71 (4) and (5).
(5) Finally we observe that the expansion (8.21.14) stops at v = A - 1 if A
is a positive integer. Then we obtain the exact representation

P~x>(cos e) = 2an I: a.(.

•=O
(1 - A)( 2 - A) ... (v (- A)
n +A - 1)(n +A- 2) · · · n +A- v
).

(8.4.13)
cos \ (n - v + A)e - (v + Ah/21
(2 sin e)•+A

n = 0, 1, 2, · · · A= 1, 2, 3, · · · an __ (n + nA - 1).
[ 8.5) PROOF OF FORMULA OF STIELTJES 209

In fact, the difference (8.4.3) (with L = A - I) is in this case a rational func-

tion which has no singularities for any w (including w = oo) and which van-
ishes for w --+ oo. Hence it must be identically zero.
The analogous representation of p~>-> (x) for I z I > I, A a positive integer, is
slightly more complicated; it can be readily derived from (8.4.I3).

8.5. Proof of the formula of Stieltjes

(I) Stieltjes' formula (4.9.I7), (4.9.I8) has been derived in §4.9 (3) from the
integral representation (4.8.I7). This argument furnishes for the remainder
RP(e) of (8.21.5) the representation

(8.5.I)

Here (g. has the same meaning as in Theorem 8.21.3)

ei(6-1r12>
(8.5.2) z = (I - t) 2 . .
sme
According to Stieltjes we have

1lor
1r
-1 • 2v
g. = 1r sm <P d<P,
(8.5.3)
(I - z)-! - ~ d,~..'I'·
2
v _ -1 zP sin P <P
L....t g.z - 1r •
v=O 0 I - Z Slll 2 <P

The last formula is obvious first under the assumption that I z I < I; it then
can be extended to the whole strip 0 ~ ffi(z) ~ ! without restriction. Now,
writing (I - t) sin2 <P = r, we find

II - z sin
2
<P 1
2
= II - r/2 + (ir/2) cote 1
2
= [sine - r/(2 sin e)] 2 + cos 2 e.
The minimum of this expression is cos2 e or (2 sin e)- 2, according as 2 sin2 e ~ I
or 2 sin2 e ~ I, so that

(8.5.4)

where M has the same meaning as in (8.21.6). Thus

IRp(e) I ~ ~
7r
(2 sin er!
}o
e t(I - o-to - t)P gp(2 sin erp M dt
(8.5.5)
2 r(n + I)r(p + !) M
= :;;. gp r(n + p + -!) (2 Sin e)P+i 1

which is equivalent to (8.21.6).

The analogous formula of Theorem 8.21.11 for the ultraspherical polynomials
p~>->(cos e), 0 < A < I, results from (4.82.3) (cf. Szego 17, pp. 57-60). It is
the expansion (4.9.25) completed with an estimate of the error if we stop at the
210 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII)

term v = p - 1. The error is again less than twice the first neglected term in
which cos is replaced by 1. The proof is the same as before; we use for the
quantities a, of (4.9.21) the representation (cf. (6.5.9))

(8.5.6) a. = 1r
-1 •
sm "A1r
1" I tan ~ !2>--1 sm
• 2•
~ d~.

The same remark as in §8.4 (3) applies to the infinite series corresponding to
Stieltjes' formula and to its generalization. (Cf. (4.9.17), (4.9.25).)
8.61. Method of Liouville-Stekloff; formula of Laplace
We shall now derive the formula of Laplace from the differential equation
(7.'3.5). The essential idea is the transformation of this equation into an
integral equation of Volterra's type which permits a successive improvement of
the asymptotic formula in question. The idea is very old, and appears in the
investigations of Liouville on differential equations of the Sturm-Liouville type.
Stekloff (1) applied this method to the asymptotic discussion of certain classical
polynomials.
Recently, Langer (1, 2, 3) employed this method systematically and improved
its efficiency considerably. He generally considers "singular" cases like (4.24.2),
or any of the equations (5.1.2), in the neighborhood of e = 0 and x = 0, respec-
tively, and obtains general asymptotic formulas of "Hilb's type." He takes up
also applications of this method in the complex domain.
(1) We write the differential equation (7.3.5) in the form

(:0yl (sin e)! p n(cos e) l + (n + !l (sin e)! p n(cos e)

(8.61.1) !
(sin e) Pn(cos e)
=-
4 sin2 e
Interpreting this relation as a non-homogeneous equation for (sin e)!Pn(cos e),
we can apply (1.8.12); the corresponding homogeneous equation has the funda-
mental system \cos (n + !)e, sin (n + !)el, so that with certain constants
eo , C1 , C2 ,

(sin e)! Pn(cos e) = c1 cos (n + !)e + C2 sin (n + !)e

(8.61.2) __+1_1 n
1
2 Bo
6
sin \(n 4+ . !)(e- t)l (.
sm2 t
sm t)!p n (cos t) dt .

If we assume e0 =1r /2, 0 < e < 1r, the last integral and its derivative vanish

for e = /2.1r This remark enables us to determine c1 and c2 . We find

(sin e)! Pn(cos e) = An cos I (n + !)e - 7r/4l
(8.61.3)
- - 1- t sin l (n +. ~)(e - t) l (sin t)! Pn(cos t) dt,
n +! J~r/2 2
4 sm t

------·-···.
[ 8.61 J FORMULA OF LAPLACE 211

where
if n is even,
(8.61.4) An = {:n: 1
n-+21 Y<n+l)/2
if n is odd.
.

This is the Volterra equation mentioned above.

If () is confined to the interval [E, 1r - E] and M n denotes the maximum of
the absolute value of the left-hand member of (8.61.3), we have

(8.61.5) M <A +2
_7r__ l'rln
n=· n n +14'2'
Sin E

Therefore, if n is sufficiently large,

M n < 2An = O(n-}),
or
(8.61.6) (sin e)! Pn (cos e) = An cos {(n + !)e - 11"/41 + O(n-!),
which readily furnishes Laplace's formula.
(2) Successive application of (8.61.3) leads to an expansion of Darboux
type, namely, to the formula

(sine)!Pn(cos e)= cos (n + !)e {

L: A.(e)n-•-t
P-1 }

•-0
(8.61.7)
+ sin (n + !)e { I: B.(e)n-•-!
p-1 }
+ O(n-p-!), E ~ () ~ 7r - E1
•-0

where A.(e) and B.(e) are certain functions analytic in 0 < () < 1r and inde-
pendent of n and p. The explicit determination of these functions, that is, the
identification of (8.61.7) with the formulas of Darboux or Stieltjes, seems,
however, to be rather difficult.
For the proof we use mathematical induction. Assuming (8.61.7), we obtain
from (8.61.3),
(sin e);Pn(cos e)

=An cos {(n + !)()- 7r/4l - _1_18

n+! 1r12
sin {(n + !)(()-
4 sin2 t
t)l

·{cos (n + !)t I: A.(t)n-•-!

•=0

+sin (n + !)t L B.(t)n-•-!

p-1

v=O
}
dt + O(n-p-!).

Now 2 sin {(n + !)(e- t) l cos (n + !)t =sin (n +!)()+sin {(n + !)(e - 2t) l,
and integration by parts yields
212 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS (VIII J
8

1 12
sin ( (n + !) (0 - 2t) l
A (t'
: : t dt
4 sm

= cos (n + !) o{I:
p=O
aiO)n-~'} +sin (n + !) o{I:
p-0
biO)n-~'lJ + O(n-p),

where aiO) and b~'(O) are certain functions of the same type as Av(O) and Bv(O).
The integrals 'involving Bv(t) can be dealt with in the same way. This leads
by use of Stirling's series (cf. P6lya-Szeg6 1, vol. 1, pp. 29, 193, problem 155),
to a representation of the form

(sin olPn(cos 0) = cos (n + !) o{t A;l)(O)n-v-!}

v-o

+sin (n + !) o{t B; 0 (0)n-v-l}

v-0
+ O(n-p-J).
Comparing this with (8.61.7), we see that

Jl = 0, 1' 2, ... ' p - 1.

This establishes the statement.
We have
(8.61.8) Ao(O) = (2/7T} cos (7r/4) = -t,
1r Bo(O) = (2/7r)l sin (7r/4) = 1r-!.

The same method can be used for the asymptotic evaluation of Qn(cos 0),
(see Problem 18), as well as of the ultraspherical polynomials. In the first
case we obtain Theorem 8.21.14. The application of the method to general
Jacobi polynomials is more difficult because no explicit values for these poly-
nomials are known at the point 0 = 1r/2 (or at any other fixed point in 0 <
0 < 7r).48

8.62. Method of Liouville-Stekloff; formula of Hilb

(I) We again deduce an integral equation for Pn(cos 0), different from
(8.61.3) and involving Bessel functions. Writing (7.3.5) in the form

(~Y I (sino)! Pn(cos 0) l + {4~2 + (n + !) 2

}(sino)' Pn(cos 0)
(8.62.1)

we apply (1.8.12). The corresponding homogeneous equation is (1.8.9) (a = 0,

k = n + t) with the solutions

(8.62.2) o' Jo( (n + t)Ol and o' Yo( (n + !)Ol.

4s Cf., however, Korous 3.

 [ 8.62] FORMULA OF HILB 213

Consequently, with certain constants Oo , c1 , c2 , we have

(sin o)'Pn(cos 0) = c~o'Jo((n + !)O} + c2o'Y0 ((n + !)O}
8
o' { -! Jo{(n + !)O} Yo{(n + !)t} - Y 0 {(n + !)O}Jo{(n + !)t}
(8.62.3) + n +! }o 0 t J~( (n + !)t} Yo( (n + !)t} - Y~( (n + !)t}Jo( (n + !)t}

.(:e2 - 4 s:n2 t) (sin t)' p n(cos t) dt.

According to (1.8.14)

(8.62.4) J~( (n + !)t} Yo( (n + !)t} - Y~ ( (n + !)tlJo( (n + !)t} = - 2 ) • (

1rn+ 21 t
2 2
Furthermore, C - (sin t)- is analytic at t = 0; that is, we can take 00 = 0,
and we then obtain
(sin o)'Pn(cos 0) = c~o'Jo((n + !)O} + c2o'Yo((n + !)O}

- ~ o' 1
8

t'[Jo((n + !)O}Yo{(n + !)t}- Yo((n + !)O}Jo((n + !)t}]

.(~2 - 4 s:n2 t) (sin t)' p n(cos t) dt.

If this equation is divided through by o',
and 0 approaches 0, the last term
tends to 0, whereas the left-hand member tends to 1. Hence (cf. (1.71.10),
(1.71.1)) c2 = 0, c1 = 1, and we find for 0 < (} < 1r
sin
( - 0-
o)' Pn(cos 0) = Jo( (n + !)O} + B}ot
1r
[Jo( (n + !)0} Yo( (n + !)t}
(8.62.5) { } )'
- Yo( (n + !)O}Jo( (n + !)t}]t-
1
ci~ ty- 1 (si~ t Pn(cos t) dt.

This is the integral equation required.

(2) First assume 0 < nO ~ 1. Then according to (1.71.1) and (1.71.4)
Jo{(n + !)O}Yo{(n + !)t}- Yo{(n + !)O}Jo{(n + !)t}

= Jo( (n + !)O} [~log ( (n + !)t}Jo( (n + !)t} + 0(1) J

(8.62.6)
- Jo{ (n + !)t} [~log {(n + !)O}J0 ( (n + !)O} + 0(1) J
2 o· o
= --J0 {(n + !)O}Jo{(n + !)t}log- + 0(1) = 0(1) log-+ 0(1).
7r t t
Therefore, the integral of the right-hand member of (8.62.5) is (by (7.21.1))

0(1) t t log~t dt +
}o
0(1) t t dt = 0(1)0 }ot
}o
2
t log! dt + 0(0 2) = 0(02).
t
 214 ASYMPTOTIC PROPERTIES' OF CLASSICAL POLYNOMIALS [VIII]

(3) On the other hand, let nO ~ 1, () ~ 1r - E. Then, according to

(1.71.10) and (1.71.11), the contribution of the interval 0 < t;;;;
n-1 to the inte-
gral previously considered, is
1

O((ne)-!Jln- ti Yo(nt) ldt + O((ne)-!lln-l tiJo(nt) ldt

= Ol (ne)-~l·n- = O(e-!n- 1) = (ne}- 1 0(etn-~) = O(e'n-~).
2

Finally, we find for the contribution of the interval n- 1 ;"£ t ;"£ e, according
to (7.3.8),

8.63. Method of Liouville-Stekloff; extension of Hilb's formula to

Jacobi polynomials
By use of complex integration, Szego (17) has extended Hilb's formula and
the corresponding asymptotic expansion mentioned in §8.23 (1), to ultraspher-
ical, and even to general J·acobi, polynomials. Following Rau (2), we deduce
the principal term of this general expansion by means of the Liouville-
Stekloff method and obtain formula (8.21.17). The bounds for the remain-
der are better than those in Szego 17, p. 77, (47), and in Rau 2, pp. 691-692,
(29), (30).
(I) Let a> -1. We write (4.24.2) in the form

~()~ + {t ()2 a2 + N2}u = {/32- to + (i- a2) (~ - 1 e)} u,

4 cos 2 - 4 sin 2 2
(8.63.1) 2

u = (sin ~)'·H (cos ~YH p~a,fl)(cos e), N =n+ (a + /3 + 1)/2.

Again applying (1.8.12), we obtain, because of (1.8.9),

(sin ~YH (cos ~YH P~a.fl>ccos e) = c1e'JaCNe) + c2o'J-aCNe)

e' t -! J a(Ne)J--a(Nt) - J -a(Ne)Ja(Nt)
+ N )eo t J: (Nt)J -aCNt) - J:_a(Nt)J a(Nt)
(8.63.2)
.
/3
2
- 1t
{4 cos 2
+ (1- _ 2)
' a t2
(I _ 1 )}
4 sin2
t
2 2
· ( sin t)aH(cos t)flH p~a,fl>(cos t) dt.
2 2
Here and in what follows .J-a(z) must be replaced by Y a(z) if a is an integer.
We refer again to the identity (1.8.14), by which
 [ 8.63] EXTENSION OF HILB'S FOR:\1ULA 215

(8.63.3) .!.a, (Nt ) J_a (Nt ) - J_a

, (Nt)Ja(Nt) = 2 sin
1rNta1r .

(If a is an integer, sin car is to be replaced by -1.) Therefore,

o-! ( sin (J)aH ( cos (J)flH p~a,fl>(cos 0) = c1 J a(NO) + c2J -a(NO)

2 2
(8.63.4) (8 (J a(NO)J-a(Nt) - J _a.(NO)Ja(Nt) }t!j(t) ( sin t)aH ( cos t)flH
+)eo
2 2
· P~a,fl}(cos t) dt,
where j(t) is regular in 0 ~ t < 1r, and independent of n.
The last integral is convergent for Oo = 0; as () ~ +O it becomes (n fixed)

0(1) 1 8
(OaCa + 0-at)t~tH dt· = O(Oa) 1
8
t dt + 0(0-a) 1 8
t 2aH dt = O(Oa+2).

This is true whether a is an integer or not [(1.71.10)], except for a = 0. Then

we obtain

0(1) 1(log~ +
8
2
log}) t!t! dt = 0 (o log}).

Dividing (8.63.4) by oa, 0 ~ +O, we find [(1.71.1)] a relation of the form

2-a-!p~a,fl)(1) + 0(02) = c1{r~~ ~):) + 0(0~} + c20-aJ-a(NO) + 0(0 2
).

(The last term must be modified for a = 0.) Hence if a ~ 0,

c2 = 0 and c1 = 2-!N-a r(n +a+ 1)(n!)-1.
The same result holds when -1 < a < 0 if we take into consideration the
fact that the "principal term" of o-aJ -a(NO) is 0- 2a. Thus, for 0 < 0 < 1r,

()-!(sm
. (J)aH(
2 (J)flHp<a,fl>(
cos 2 n cos 0) -- 2-!N-a r(n +a+
n! 1) J a(NO)

8 63 5
( · · ) + 1 8
IJ aCNo)J-aCNt) - J -aCNo)J aCNt) lttf(t)(sin ~)aH (cos ~YH
· P~a,fl>(cos t) dt.
(2) Now let n ------+ oo. We find bounds for the last integral in a manner
similar to that in §8.62. First, let 0 < nO ~ 1. Then for a ~ 0, the integral
is (cf. the second bound in (7.32.5))

0(1) 1I8
(nOt(nt)-a + (nO)-a(nt)a}t!taHna dt = O(naoa+ 2).

In case /x. = 0, we reason as in §8.62 (2) and obtain the same bound, that is,
216 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII ]

O(fl). On the other hand, let n- 1 ~ () ~ 1r - E. The contribution of the

interval 0 < t ;;;; n- 1 is (cf. the second bound in (7.32.5))

0(1) 1n- 1

(nO)-! II J -a.(Nt) I + I J a.(Nt) llt'ta.-Hna. dt

= 0(1) (no)-!n -a.- 2na. = oco-!n-I) = (no)- 10(o'n-j) = O(o'n-j).

(Here J -a must be replaced by Yo if a = 0.) For t ~ n -\ we use the first

bound in (7 .32.5) and obtain

0(1) 1 8
_
1
(no)-'(nt)-'t'n-l dt = O(o'n-!).

8.64. Method of Liouville-Stekloff; asymptotic formula of Hilb's

type for Laguerre polynomials
By use of the fourth equation in (5.1.2) this method readily leads to (8.22.4).
The third equation could likewise be used, but the calculation would then be
slightly more complicated. Concerning an extension of (8.22.4) to an asymp-
totic expansion (at least for a > -!) see §8.66.
(1) Let a ,> -1. Writing the equation in question as

v" 4 - a2)
+ (4N + -1 x2 v=
2
x v '·
(8.64.1)
N = n + (a + 1)/2,
we can apply (1.8.12) and (1.8.9). Hence with certain constants Xo, Ct, c2 ,
e-'" 212xa.HL~a.>(x 2 ) = c1x'Ja.(2N'x) + c2x!J_a.(2N'x)
+ _t_ 1'" Ja.(2N'x)J_a.(2N!t)- J_a.(2N'x)Ja.(2N't) e-t2t2t+2r<a.>(t2) dt.
2Nt xo J:(2N't)J_a.(2N't) - J!_a.(2N't)Ja.(2N't) n

Once again we use (8.63.3) and obtain

e-'" 212xa.L~a.>(x 2 ) = c1J a.(2N'x) + c2J-a(2N'x)
+ . 7r {'" IJ a.(2/tx)J-a.C2N't) - J -a.C2N'x)J a.(2N't) l e-t 212t+aL~a.> (t 2) dt.
2 Sln a1r ]xo

If a is an integer, J -a.(z) must be replaced by Y a.(z) and sin a1r by -1.

Let x 0 = 0. For a fixed n, as x ~ + 0, the last term is

0(1) 1'" (xa.Ca. + x -a.ta.)ta.+a dt = O(xa.H).

(If a = 0, this bound must be multiplied by log (1/x).) Therefore, as in §8.63,

c2 = 0 and L~a.>(o) = c1Na. 12 lf(a + 1) l- 1 ; whence
(8.64.2) C1 = ~a./ f(n
2
+a+ 1)(n!)-1,
 [ 8.64 I FORMULA FOR LAGUERRE POLYKOMIALS 217

Consequently,

(8.64.3)
+ 2 sm.a1r
.7r 1"' IJ a.(2N'x )J-a.C2N't)
0

- J -a.C2N'x)Ja(2N't) l e- 1212ta.+aL~a.> (t 2) dt.

(2) As n ~ oo, the remainder term can be estimated in the same way as
in the previous cases. However, we avoid here the use of bounds of the type
(7.32.5). (In §7.6 (3) we derived such bounds as a consequence of (8.1.8)
and of Fejer's formula (8.22.1); this is almost (8.22.4), which is just what we
must prove now.) In the following proof we apply only the elementary
formula (8.1.8) of the Mehler-Reine type, in particular only the second bound
in (7.6.8).
First, let 0 < x ~ n-!. Then we have L~a.>(t 2 ) = O(na.) for 0 ~ t ~ x.
It then follows that the integral term in (8.64.3) is

(8.64.4) 0(1) 1"' (na. 12xa.n-a. 12Ca. + n-a.12x-a.na. 12ta.)ta.+ana. dt = O(xa.Hna.);

for a = 0 this bound must be multiplied log (x -ln -!).

Now let n -! ~ x ~ w\ w
where is a fixea positive number. Let M n be the
212 2
maximum of e-"' xa. I L~a) (x ) I in this interval. Then the contribution of the
part 0 ~ t ~ n -! of the integral term is, a ~ 0,

(8.64.5)

The same result holds if we have a = 0. The contribution of the other part
n-! ~ t ~ xis

(8.64.6) 0(1) 1"' n-ix-!n- C't Mn dt

1 3
= 0(1)n-!x 3Mn = Mn·o(1).

Taking account of (8.64.3), and using the same argument as in §8.61, we find
that

(8.64. 7)

(This is, of course, identical with the first bound in (7.6.8).) Therefore, in
view of (8.64.5), (8.64.6), and (8.64.7), we obtain for the remainder term, if
n-! ~ x ~ w\
O(x -!na./ 2- 914 ) + O(l)n-!iO(x-!na/ 2- 1)
= O(x -!n a.t2-9t4) + O(xst2n a.t2-!) = O(xst2na.t2-'<).

·----·--····-'
218 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNO:\HALS [VIII]

8.65. Method of Liouville-Stekloff; Hermite polynomials

(1) The integral equation (8.64.3) assumes a particularly simple form in case
a = ±~, that is, for the Hermite polynomials. Using (5.6.1) and (1.71.2),
and writing n =2m, a= - t, and n = 2m+ 1, a = +!,respectively, we obtain
e-x 212Hn(X) = An COS (N!x- nrr/2)
(8.65.1)
+ N-! lox sin (N!(x - t) }t2e- 1212 H n(t) dt,
where
(8.65.2)
according as n is even or odd, and N = 2n + 1. However, it is more con-
venient to deduce this directly from the second equation in (5.5.2).
\Ve prove (8.22.7) by mathematical induction. The statement is true for
p = 0, replacing both sums I:.-;::0 by 0. In fact, if M n denotes the maximum
1
212
of e-x Hn(x) I in a fixed real interval, we find from (8.65.1) that
1

(8.65.3) Mn ~An+ O(n-!).Mn.

Then Mn = AnO(l).
(2) Now assuming (8.22.7) for an arbitrary p, we obtain from (8.65.1)
e-x 212Hn(x) =An COS (N!x- nrr/2)

+ AnN-! {x sin (N!(x - t) l {cos (N! t - nrr/2) ~ t2 uv(t)N-v

(8.65.4) }o v-o
+ N-! sin (N! t - nrr /2) 1: t Vv(t)N-v} dt + An O(n-p-!).
v-0
2

The second term of the right-hand member contains expressions of the fol-
lowing type:

AnN-v-! lox sin (N\x- t)l COS (N!t- nrr/2)tk dt

= (2k + 2) - 1 An N-v-! xk+l sin (N! X - nrr /2)

+ !AnN-v-! lox l sin (N!(x - 2t) + nrr/2l dt,

(8.65.5)
AnN-v-I lox sin (N!(x - t) l sin (~ t - nrr/2)t1 dt

= -(2Z + 2)-1 AnN-v- 1 x1H cos (N!x- nrr/2)

+ !"AnN-v-1 lox l COS (N!(X- 2t) + nrr/2l dt,

where k is even and lis odd (k ~ 2, l ~ 3). Integration by parts furnishes

[ 8.66 l LAGUERRE POLYNOMIALS 219

L" tk sin {N'(x - 2t) + n1r/2l cjt = !~txk cos (N' x - n1rj2)

- !kN-l l~ tk- 1
cos {N'(x - 2t) + n1r/2l dt,
(8.65.6)
1"' 1
t cos IN'(x - 2t) + n1r/2l dt = !N-!x 1 sin (N'x - n1rj2)

+ !ZN4 1"' t 1 1
- sin {N'(x - 2t) + n7r/2l dt;
the second formula holds also for l = 1, and

(8.65.7) 1"' sin IN'(x- 2t) + n7r/2l dt ={~-!cos (N!x _ n1rj ),

2
according as n is even or odd.
This consideration leads to a formula of the type
212
e-"' Hn(x) = Xn{cos (N'x- n1rj2) 'tu;l)(x)N-v
v-o

By repeated application of the same argument

e-"'212 Hn(x) = Xn {cos CN' x - n7r/2) t u~ 2>(x)N-v

•=0

+ N-l sin (N' x - n7r/2) t v; >(x)N-v +

v-0
2
O(n-p-1)}.

2
We readily see that u?> (x), u; >(x); v;l) (x), v; >(x) are polynomials of the same
2

type as Uv(x); Vv(x), respectively, and Uv(x) = u;l)(x) = u; 2>(x), v.(x) = v?>(x) =
v?>(x), v ~ p - 1; whence (8.22.7) follows.
The proof of Theorem 8.22.7 can be given along these same lines.

8.66. Application to Laguerre polynomials

(1) The asymptotic expansion (8.22.7), combined with the formula (5.6.5)
of Uspensky, readily furnishes an asymptotic expansion of Hilb's type for the
Laguerre polynomials L~">(x), at least for a > -t. We shall give only an
outline of the proof.
Substituting (8.22.7) in (5.6.5), we obtain an asymptotic expansion of which
the general term is, apart from trivial constant factors depending on n,
220 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII]

Here the uv(x) and the Vv(x) are even and odd polynomials, respectively, and
both are independent of n. The expression (8.66.1) is a linear combination of
terms of the form

(8.66.2)

where k and l are non-negative integers, k even and l odd. If we expand

eT/Z r kJZ m
• a power senes
• a b ou t r = x, we ob tam
• f or t he fi rst mtegra
• 1 terms of
the type

with q integral, and q ~ 0. This can be expressed in terms of Bessel functions

[(1.71.6)]. A similar method furnishes for the second integral (8.66.2) terms
of the type

q again being integral, and q ~ 0. These can also be expressed in terms of

Bessel functions (combine the second formula (1.71.5) with (1.71.6)).
If we stop the expansion of the first expression (8.66.2) at a certain term, the
remainder appears in the form

(8.66.3)

wheref(t) = Cmrm + Cm+ITm+r + · · · is an integral function with a zero of order

mat r = 0; herem is an arbitrary integer. The remainder is of similar form
for the second integral in (8.66.2). Writing
g(t) = (1 - t 2 )a-~f[x(I - t2 )],

we see that the functions g(t), g'(t), g"(t), . ·. , g<m-r>(t) vanish at t = ± 1;

they are all xmO(I), where 0(1) is uniformly bounded in -1 ~ t ~ +I, and in
a fixed finite interval a ~ x ~ b containing the origin or not. Integrating by
parts, we find for (8.66.3) a bound of the form O(n-K), where K is arbitrarily
large with m, uniformly in x, a ~ x ~ b.
Similar remarks hold for the second remainder.
(2) The first term of this expansion furnishes (8.22.4). We also obtain
readily an extension of (8.22.4) to the complex domain.
Now assume 0 < E ~ x ~ w. Then applying (1.71.8), we find Perron's ex-
pansion (8.22.2). (Cf. Uspensky 1, pp. 608-610.) There is no difficulty in
deriving the complex formula (8.22.3) of Perron in the same way.

--------····
[ 8.71 l LEGENDRE POLY~O:\IIALS AXD RELATED FUNCTIONS 221

In all these considerations we assumed a > -!. An extension of Perron's

formulas to arbitrary real values of a is possible by means of the second formula
in (5.1.13). Concerning a second proof of Perron's formulas (by use of the
method of steepest descent), see §8.72.

8. 71. The method of steepest descent; Legendre polynomials and

related functions
(1) This method can be used for approximating integrals of the form

(8.71.1)

extended over a certain arc or closed curve, where F(t) = e1 w and g(t) are
given analytic functions regular in a certain part of the complex t-plane, and
n ~ oo. According to Cauchy's theorem, a defqrmation of the contour is
possible. In many cases it is convenient to make it pass through some of the
points to at which f' (to) = 0 (saddle point); in addition, the direction of the
contour at to (critical direction) must be determined according to the condition
that (assuming !"(to) ~ 0) the expression
(8.71.2) nf" (to) (t - to) 2/2
is real and negative if t is sufficiently near t0 • Then

Hence under proper conditions concerning the behavior of f(t) on the comple-
mentary part of the path of integration, the neighborhood
(8.71.3) 0 < c < !,
#
of the saddle point furnishes the "principal" part of the integral as n ~ oo.
Its contribution is of the form
)!
(R.71.4) enf<to> g(to)n -!
1 +n6

-n6
exp (-a/) dp "" enf<to> g(to)
(
.!!.._
an
; a > 0, n ~ oo.

(Additional difficulties arise if f" (to) = 0; concerning a case of this type, cf.
§8.75.)
The term "method of steepest descent" arises from the following consider-
ations. Let t = u + iv. Then representing u, v, mfj(t)] as cartesian coordi-
nates in the ordinary euclidean space, we obtain a surface with a saddle point
at t = t0 , and the curve with the critical direction is the "steepest" curve on
the surface through this point.
There is, of course, considerable freedom in the choice of the contour; only
its direction through the saddle point is restricted. However, the exact "calcu-
lation of the saddle points to from !'(to) = 0, and particularly that of the corre-
sponding critical directions, might well be a complicated task in certain cases
222 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII)

For this reason, the following observation may greatly simplify matters. In-
stead of the critical direction itself, any other direction through the saddle point
can be taken for the path of integration, provided it forms an angle of less than
1r/4 with the critical direction. Then the constant a in (8.71.4) becomes com-
plex with a positive real part. Geometrically, this condition means that in
passing through the saddle point on the surface, no higher level is reached in
the neighborhood of this point than at the point itself.
For our purposes, we shall prefer a contour satisfying the last condition, and
along which 9?(f(t)] varies monotonically. Then the discussion of the integrand
exterior to the neighborhood (8.71.3) becomes comparatively simple.
Concerning the history, further details, and important applications of this
method, the reader is referred to Watson 3, pp. 235-236.
(2) As a first illustration let us consider the Legendre function of the second
kind, that is, the special case a =. {3 = 0 of (4.61.1). Let x = cos 0 - iO,
0 < 0 < 1r. Then

(8. 71.5) Q~o,o) (cos 0 -

'
iO) = Qn (cos 0 - iO) =l J(! 2
t -
1 )" ~t (
2 t - cos 0 cos -
The original path of integration, -1 ~ t ~ + 1, can be deformed into the upper
half of the unit circle described in the negative sense. The saddle point con-
dition is

(8.71.6) !!:_ (!
dt 2 t - cos 0 2 (t - cos 0) 2
!
t2 - 1 ) = t2 - 2t cos 0 + 1 = 0 whence t = e±iB
' '

and we see that the path of integration passes through the saddle point t = e•8•
·o
If t = e' we have

(dtd)
2 2
(8.71.7) (1 t - 1 ) 1
2t- cos 0 = i sin 0 '
so that near this point
2
(8.71.8)
1 t 1 - •o (t - ei8 ) 2
2 t - cos 0 = e + 2i sin 0 + ··· ·
Along the critical direction, e-i<B+r/ 2) (t - ei8 ) 2 must be real and negative; that is,
8
(8.71.9) arg (t - ei ) = 0/2 + 37r/4 or 0/2 - 7r/4.
Now, the angle between this line and the tangent to the unit circle at the point
i8 •
e IS
(8.71.10) a rg 1l ei(8/2+3r/4) •• ei(B+r/2)} = 1r /4 - 0/ 2,
and I 7r/4 - 0/2 I ~ 7r/4 - E/2 if E ~ 0 ~ 1r - E. We can therefore use the
circle I t I = 1 as the path of integration. 49
4
9 The critical direction is given by that of the bisector of the acute angle between the
tangent at the point ei 6 and the horizontal direction.
[ 8.71] LEGENDRE POLYNOMIALS AND RELATED FUNCTIONS 223

Substituting t = ei"', 0 ~ cp ~ 1r, we find that

2
1 t - 1 / sin cp
(8.71.11)
1 2 t- cosO.= ((coscp- cos0) 2 + sin2cp}i
= { (cos ~i~cpcos oy + 1 r-i
is an increasing function of cp for 0 ~ cp ~ 0, and a decreasing function for
{} ~ cp ~ 1r. We next consider the contribution of the arc
(8.71.12) -na :::;; p:::;; + na ,
where ois a properly chosen positive number. On this arc we have

ei<J> - eiB = eio (exp [in-! p] - 1) = eio {in-! P + (in~; p)2 + ... }'
and, in view of (8.71.8),
2 2~
1- t t - 1 = eiB - e
--:--;-- n -1 p2 11
l
+ c1 n -! p + C2 (n -l p)2 + ... },
2 -COS 0 22Slll 0

provided o < t. Here c1, c2, · · · are certain functions of{} independent of n, for
which Cm = O(Am) holds uniformly in E:::;; {} ~ 1r -E, and in m; A = A(E).
Now

1 t - 1
2

( 2 t - cos {}
)n = e
inB {
exp - 2i sin {}
ei
8

P
2} w
e '
where

8
- ~
L...J 1- ( -.-.- )" n 1-• p2.( 1
ei + c1 n-! p + .. ·)".
•-2 v 22 sm {}
Let o < !; then if M is an arbitrarily large positive integer, Wand ew can be
reduced to a finite sum plus a remainder which is O(n-M). This yields the re-
lation

(8. 71.13)
(1
-
t2
2 t - cos 0
- 1)n = e
in8
exp - { ei8
2i sin {}
p
2}

· {1 + u1(p,O)n-! + U2(p,O)n-1 + ua(p,O)n-1 + · · · }.

The series in the braces is an asymptotic expansion. If only m terms of this
are taken, the error is less than an arbitrarily large power of n- 1 provided m is
sufficiently large; here u.(p, 0) is a polynomial in panda function of 0, analytic in
E :::;; {} :::;; 7r -E. In addition u.(- p, 0) = ( -1 u.(p, 0). r
224 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII]

(3) Now multiply (8.71.13) by

1 __d_t__ _ !
iei"' dcp _ 1 iei8ein-!p n-1dp
2 cos fJ - t 2 cos fJ - ei<l> 2 cos() - eiB ein-!p
i8
== - ~ n-1 {1 c~n-tp c~(n- 1 p) 2
sm 2 0 + + + · · ·l dp,
where l c~ l is a sequence similar to l em l ; this does not change the essential
character of equation (8. 71.13). We obtain, as the contribution of the arc
(8.71.12),

_ ei<n+I)B n-! 1-n6 exp{- eiB /}

(8.71.14) 2 sin() +n6 2i sin()
·P + VI (p, fJ)n -! + v2(p, fJ)n + va(p, fJ)n-l + · · · l dp,
-I

where l v.(p, fJ) l is a sequence of polynomials similar to l u.(p, fJ) \. The last
series is again an asymptotic expansion of the same type as (8.71.13).
At the end-;points of the arc in question the modulus of the integrand of
2
(8.71.14) is O(e-cp\ that is, O(e-cn \ c > 0; the same is true of the contribution
of the complementary arc becanse of the monotonic character of the function
(8.71.11). Thus,
Qn(cos fJ - iO)
i(n+l)B 1+n6 { • i8 }
= e--.- n -!
2 Sln 0 -n6
exp -
2'/, 0
~ /
Slll
(1 + vi (p, fJ)n -! + v2(p, fJ)n + · · · l dp
-I

+ O(e-cn2\
The terms corresponding to the odd powers of n-! vanish after integration.
Upon completing the interval of integration, we obtain

i(n+I)8 { i(8-7r/2)}-!
(8.71:15) _ -
- e - n-! e 1r
1 + O(n-i)
2 sin fJ 2 sin fJ

= ( ~
2n sm fJ
)! exp !i[(n + !)fJ + 7r/4]l + O(n- 1),
the bound for the remainder holding uniformly for E ~ fJ ~ 1r -E.
This method leads to a complete asymptotic expansion of Qn(cos fJ - iO) of
the type (8.61.7) for E ~ fJ ~ 1r - E, but seems to be too difficult for use in
finding the general law of the coefficients.
From (8.71.15) we obtain the corresponding asymptotic expansion for
Qn(cos fJ + iO) by merely replacing i by -i. From this, and (4.62.8), we
easily derive Laplace's formula. Formula (8.61.7) can also be derived in this
 [ 8.72 l PERRO.K'S FOR:'vWLAS FOR LAGUERRE POLYNOl\IIALS 225

way. At the same time we obtain a similar asymptotic expansion for Q~o,o) (cos fJ)
= Qn(cos fJ) with the principal term (8.21.19).
(4) The analogous considerations for the Jacobi polynomials P ;_a.#) (cos fJ)
are immediate. Here a and tJ are arbitrary and real, n ----; oo, and () again
satisfies the conditionE ~ fJ ~ 1 r - E. According to (4.4.6) we have

p ;_a.f3) (cos fJ) = 29{ {~

27r~
f (~ t2·- 1
2 t - cos ()
)n
(8.71.16)
1- t
( 1 - cos ()
)a (1 +1 +cos () ){3 cos () -
t dt }
t '
where the contour is the same as that used in (2). 50 The additional factor

(8.71.17) 2 (
7r~
1- t
1 - cos ()
)a (1 +1 +cos () ){3
t

does not cause any new difficulty and furnishes fort = ei8

(8. 71.18) 2 ( .
1ri sm
o)-a (cos o)-13 exp[da(fJ-
. .
7r)/2 + {:JfJ/2\].
2 2
Therefore, we find (8.21.10) and an expansion of the type (8.21.12).
(5) The same method can readily be applied to the functions Qn(x), P n(x),
or more generally to Q~a ,!3) (x) and P~a ,!3) (x), where xi::; arbitrary real or complex
but not on the segment [ -1, +
1]. This leads to (8.21.9) and also to th~
expansion (8.21.11). In the case of Q~a.f3J(x), we start from (4.61.1). It.is
convenient to replace the half-circle u::;ed above by the circular arc through
±1 and z = x - (x - 1)!, I z l < 1, (xi::; in the cut plane), which was intro-
2

duced in §4.81 (1). The resulting integral iH the same as (4.82.4). Applying the
method of HteepeHt descent, we obtain a formula of the type
(8.71.19) (x- 1)a(x + 1) 13
Q~a,/3J(x) ""'n-i{x- (x 2 - 1)!r+1 cf>(x),
2
where I x - (x - 1)! I < 1, and cf>(x) iH independent of n, and regular and non-
zero in the cut plane.

8. 72. Method of steepest descent; Perron's formulas for Laguerre polynomials

As a further application of this method we again prove the expansions (8.22.2)
and (8.22.3). The present proof is based on the integral representation (5.4.1),
which is valid for an arbitrary real a, provided n is sufficiently large.
(1) Let x be arbitrary but nonzero. We start from the asymptotic expan-
sion (1.71.8) of the Bessel function J a(z), z complex. The contribution of the
segment 0 ~ t ~ 1 in (5.4.1) is (n!)- 10(1). We can therefore confine our at-
tention to values of t ~ 1, HO that (1.71.8) may be applied. Substitution of
this expansion into (5.4.1) leadH to integrals of the type
50
We must avoid the points t = ±1 by means of small semi-circles.
226 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [VIII ]

~! loo e-tt+a12 (tx)-m 12- 114 c~s (2(tx) 1 - a7r/2- 7r/4l dt, m even,
(8.72.1)
~! 1oo e-tt+a12 (tx)-m 12- 114 sin (2(tx) 1 - a1rj2- 7r/4l dt, m odd,

m = 0, 1, 2, · · ·.
(Here the range of integration has again been completed to. 0 ~ t < oo .) The
remainder term will be of the form

(8.72.2) ~! 1oo e-tt+a/ 2 -q exp (2t1 J9{( -x) 1 Jl dt,

with q a fixed positive number which can be chosen arbitrarily large; the deter-
mination of ( -x) 1 is the same as in Theorem 8.22.3. If x is a fixed positive
number, this is obviously O(na/ 2-q). If x is complex, the discussion of this
remainder term requires greater care. Subsequent considerations furnish a
bound in this case also.
(2) For the sake of convenience we shall first discuss the integral

(8.72.3)
1
I
1 00
n+I -n
e~t t exp [t 1~] dt = n ~
1oo (e 1-t t)n exp [n! t1~] dt,
n. o n. o
(~ ~ 0, arbitrary complex) to which the integrals (8.72.1) and (8.72.2) can be
reduced; here n ~ + oo, but n is not necessarily an integer. In the last integral
the saddle point is "essentially" t = 1, and the positive real axis corresponds
61
to the critical direction. If we write, as in (8.71.12),
(8.72.4) t = 1 + n-! p, -n : :;
~
p:::;; +n, ~

we obtain, for 0 < o< 1/6,

(e 1·-tt)n = exp {n(l - t) + n log [1 + (t - 1)]l

(8.72.5) = e-p 212 exp { n 'f. (-1),_ 1 (n-! p)"}

~ •=3 v

= e -p
212
{1 + u1(p)n-1 + u2(p)n- 1 + · · ·L
where the u.(p) are polynomials in p independent of n. This is an asymptotic
expansion similar in character to that in (8.71.13). Furthermore,
exp [n! t! ~] = exp l n! ~ + n 1~([1 + (t - 1)]! - 1) l
(8.72.6)

where the c. are certain numerical constants. This furnishes exp (n 1 ~ + p~/2)
61 This case is not a direct application of the method indicated in §8.71 (1) since the
1 2
integrand has the form [F(t))n[G(t))n 1 •
[ 8.73] LAGUERRE POLYNOMIALS 227

multiplied by an expression similar to that in the braces in (8. 72.5). The

coefficients corresponding to the u.(p) are in this case polynomials in p and ~.
We also see that the contribution of the range of integration complementary to
(8.72.4) is O(exp [ -cn ~]), c > 0. Therefore, we have in the same sense as in
2

(8. 71.14)

loo (e t) nexp [n1t 1~] dt

1
-t

= n-! exp [n
1
~] 1:: 6

exp [ -//2 + p~/2] /1 + v1(p, ~)n-! + v2(p, ~)n- 1 + · · ·l dp

= n-! exp [n
1
~] 1:oo exp [ -//2 + p~/2] /1 + v1(p, ~)n-! + v2(p, ~)n- 1 + · · ·l dp,
where the v.(pt ~) are polynomials in .p and ~. Now if q is a non-negative integer,
we have, in general,

1.:oo exp [- //2 + p~/2] pq dp = e~ 218 J~oo e-p 212 (p + ~/2)qdp,

and the last integral is a 7rq in ~. (For q = 0 we get (21r) 1 e~ 218 .) According
to Stirling's formula an expansion of the type

exp (n ~
1
+ ~ /8) /1 + v1(~)n-! + v2(~)n- + · · ·l
2 1

results for (8.72.3), in which the v.(~) are polynomials.

By applying this result to the expressions in (8.72.1), (replacing n by
n + a/2 - m/2 - 1/4), we obtain the required expansions. The special case
~ > 0 yields the required bound for the remainder term (8.72.2).

8. 73. Method of steepest descent; Laguerre polynomials for

1 < x < (4- TJ)n
Here and in the next two sections we derive formulas (8.22.9), (8.22.10), and
(8.22.11) by the method of steepest descent. vVe notice that in the first case
the condition x = (4n 2a + +
2) cos2 cp, E ~ cp < 1r/2 - m-i, means that x
satisfies the inequality xo < x < (4 - TJ)n, where E, Xo, and 'YJ are fixed positive
numbers, E < 1rj2, 'YJ < 4, n large. The parameter a is arbitrary and real.
(1) We start with formula (5.1.16) (cf. the remark at the end of §5.2), in
which we replace x by ~2 and w by -w2/4; thus
oo
2: r<"'l(t2)
" '" c-w2/4)n = 2"'e-w2'4C~w)-"'e"' .. i'2 J .. ce-i .. ,2~w).
(8.73.1)
n=O r(n +a+ 1)
Consequently,

(8.73.2)

·----------·
228 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYXOMIALS [ Vlll]

The integration is extended over a contour enclosing the origin. We choose

it as a circle with center at the origin and radius to be determined later. The
function (8.73.1) is real if w is real.
Since~ = x!,
(8.73.3) ~ = ln COS cp, ln = (4n + 2a + 2)\
Then ~ is bounded from zero. According to (1.71.9),' as I w I ----; co, we have
uniformly in 0 ~ arg w ~ 1r,
e'"ri 12 Ja(e-i" 12 ~w) = (27r~w)-ieEwll + O(C'Iwr')l
(8.73.4) ! 1
+ (27r~W)- exp [ -~w + (a + !hi](l + O(C'I w !-) l.
Hence from (8. 73.2), for ~ = ln cos cp, w = lnz,

(8 73 5)
• •
(-l.)n
4
r (n
L~")(f)
+ a + 1) = 2"t-"(2 t)-iz-2 n-a-i2m{__!_
<; 7r<; n a~ 27ri
G + _!_ H
27ri
+ x} '

where

G = J z-i exp l - i Z!i + Z!z cos cp - ! Z! log z l dz,

(8. 73.6) H = e<a+D,.; f z-i exp l- i Z!i - Z!z cos cfJ - t Z! log z l dz,

K = oc~-lr;;l) f I exp l- t Z!i ± Z!z cos cp- ! Z! log zlll dz I.

Here the integration is extended over the upper half-circle I z I = 1, andlog z
is zero if z is 1. In the last integral we choose the plus or minus of the ambigu-
ous sign according to which gives the integral the larger value. If now we set
(8.73.7) f(z) = - iz 2 + z cos cp - ! log z,
we find the saddle points of the first integral from the equation f'(z) = -z/2
+cos cp- (2z)- 1 = 0, that is, z = e±i<~>. Since f"(e;"') = sin cp e-•<4>+" 12\ we have
in the neighborhood of e;"',
(8.73.8)
Therefore, the critical direction can be found in a manner similar to that used
in §8.71 (2).
(2) On the circle z = e;.;,
\8.73.9) 9{ lf(e"") l = - i cos 21/; + cos if; cos cp
is increasing if 0 ~ if; ~ cp, and decreasing if cp ~ if; ~ 1r. Consequently, it
suffices to consider the contribution of the arc
\8.73.10) 1/1 = cfJ + r;.'p, -n
~
~P~+n,
~
[8.7:3 I LAGCERRE POLY~OMIALS 22fJ

where ois a fixed positive number, o< 1/6. ~ow we have

lr1 + ·z-1 + (il-;;2!p) + . . .} '

1 2

z = eif = eit/> ~ n p .

(8.73.11) dz = e;"'il-;:1{ 1 + il-;; 1p + (il;1

;)
2
+ ... } dp,
1
·z-1 {1
z-e it/> =e it/> ~nP + il-;;
- p + .. ·}.
2!
Whence from (8.73.8)
(8.73.12) f(z) = f(e;"') - t sin 4>ei<4>-" 12 >l-;;2/(1 + c1l-;; 1p + c2Cf;. 1p) 2 + ... l,
where the c, are functions of 4>, independent of n and p, and Cm = O(A m) uni-
formly in E ~ 4> ~ 1r - E, and uniformly as tom; A = A (E). (We note that
this condition for 4> is more general than that in (8.22.9).) Hence, we have in
the same sense as in (8.71.14), the following asymptotic expansion:
G = e-i4>! 2 exp (l~f(e;"') le'"'il-;: 1

(8.73.13)

where c1I , c1" , c2I , c2" , cIll

2 , · · · are constants. Th e pnnc1pa
. . I term f urms
. h es

G = e-i4>! 2 exp ll~f(e;"') I e;"'il-;;1 (27r)i(sin 4>)-!e-i<<~>t 2 - .. 14 >II + O(Z-;; I

2
)

= ( _21r -)i exp 1- i(2n +a+ 1)4> + 37ri/41Z-;;1

(8:73.14) sm 4>
·exp l ~ l~ cos 4>
2
+ ! l~ + !il~ sin 4> cos 4> Ill + O(Z-;; I, 2
)

since the integrals with odd powers of p vanish. The bound for the error holds
uniformly forE ~ 4> ~ 1r - E.
The integral in H iH obtained by replacing 4> by 1r - 4>, so that

(8. 73.15)
H = ( . 7r
2
8111 4>
Y exp ((a+ i)7ri- i(2n +a+ 1)(7r- 4>) + 37ri/41

.z-;: 1exp \i l~ cos 2 4> + t l~ - !il~ sin 4> cos 4> Ill + O(Z-;; )}.
2

Taking the absolute values of the integrands in G and H, we obtain, E ~ 4> ~

1r/2 - m·-i,

(8.73.16)
We easily see that, except for terms of higher order, H = -G. Therefore,
230 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII J

~ )
1

(8.73.17)
w{2_2_. G + 2.]__ H + K} =
~ ~
(
rmncf>
Z-;1 exp It Z! cos2 cf> + t Z!}
·(sin[! Z! sin <1> cos <1> - (2n + a + 1)<1> + 3r/4] + O(C1Z; 1)},
1 1
since Z-;2 = O(C Z-;; ). Introducing this into (8.73.5), we obtain

L~a)(~ )
2

( - .!)n = -1cs·n,~..)-lt-a-!2a+1z-2n-a-J (.! z2 2,~..'f' + l.l2l

4 r(n +a+ 1) r 1 'f' <; n exp 2 n cos 4 n

·(sin[! Z! sin <1> cos <1> - (2n +a+ 1) <1> + 3r/4] + O(C1Z;1)}.
Now

ln = 2n1 exp (a! 1) (1 + O(n~2)},

4nr(n +a+ 1) = r 122n+!nn+a+!e-n{l + O(n-1)}.
And since n- 1 = O(C1Z;1), we have
L~a) (~2) = (- 1t(r sin <t>)-!~-a-!nat2-1t4

·exp { -n - (a+ 1)/2} exp {t f + n + (a+ 1)/2}

·/sin [t z; sin <I> cos <I> - (2n +a+ 1) <I> + 3r/4] + oc~- Z-;; )}' 1 1

e-FI2L~a)(f) = ( -1r(r sin cf>)-!~-a-!na/2-l

(8.73.18)
·/sin [(n + + 3r/4] + oc~- Z; )}. 1 1
(a+ 1)/2) sin 2cf>- (2n +a+ l)cf>
This is identical with (8.22.9).
We observe that in the application of this result to Hermite polynomials
certain simplifications are possible. For a = ±! the 0-terms in (1.71.9) and
(8.73.4) vanish identically, so that Kin (8.73.5) can be cancelled, and ~can be
arbitrarily near zero. Therefore, (8.22.12) follows readily by use of (5.6.1).

8.74. Method of steepest descent; Laguerre polynomials for

(4 + 11 )n ~ x ~ An
We start again from (8.73.2) and integrate along a proper circle about the
origin. Using a notation analogous to that in the previous section, we assume
that
(8.74.1) ~ = ln cosh cf>, E ~ cf> ~ W1
[ 8.74] LAGUERRE POLYNOMIALS 231

where
1
G1 = f z-i exp 1l - i ln2 z2 + ln2 z cosh cp - ! ln2 log z l dz,

(8.74.3) H~ = e<a+i),.; Jz-i exp l-il!i- Z!z cosh cp- !Z! log z} dz,

K 1 = O(C'r;;') JI exp l-il!i ± l~z cosh cfJ - !Z! log zlll dz 1.

In each of G~ and H~, we integrate along both arcs

-7r/4 ~ arg z ~ + 7r/4,

of the circle I z I = e-"', while in G~ and H~ (which have the same integrands as
G~ and H~, respectively) we take 7r/4 ~ arg z ~ 37r/4; in K 1 we take the arc
0 ~ arg z ~ 1r of the preceding section. Then (8. 73.4) can again be used
[cf. (I. 71. 9) ].
In this case we discuss

(8.74.4) f(z) = -iz 2 + z cosh cp - ! log z.

The condition f'(z) = 0 furnishes z = e±"', and we have f 11 (e-"') = (e 2"' - 1)/2
= e"' sinh cp. Therefore, the circle I z I = e-<~> pttsses through the saddle point
with the smaller modulus and has the critical direction at that point. For the
second integral we obtain the saddle points -e±"', from which it follows that the
circle I z I = e-4> can be used here again. Obviously, for z = e-HN·,

(8.74.5) mU(z) l = -ie- 2"' cos 21/1 + e-"' cosh c/J cosy;+ efJ/2
decreases as if; increases from 0 to 1r.
On writing if; = r;:'p, -n~ ~ p ~ -t n~, 0 < o < 1/6, we obtain
z _- e-<~>+if -_ e-<~> { 1. +
2
·z-' p +
~ n
Cir;;l
---,-·
p) + •. •
}'
2.

(8.74.6) dz = e-4> il;;-1 { 1 + il;;- 1 p + (il;'t)- +

2
· ·.} dp,

z-e -<~> =e -<~> ~.1·n_, p{ 1 + iZ;;-'

- 2! P + ··· }·'
whence

(8.74.7)

Here the coefficients c, , c2 , · · · (and also c,I , c,II , c2I , c2II , c2Ill , · · · beI ow ) are
analogous to those in §8.73. Hence, in the same sense as before, we have the
asymptotic expansion
232 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII ]

. (I + Z~ (c~p + c~' /) + l~ (c~/ + c~' / + ct/) + ...

1 2
l dp
(8. 74.8)
= e"' 12 exp /l!J(e-"')le-"'iZ~\27r}(sinh¢)-!e"' 12 {I + O(Z~ 2 )l
2
= ( . h1l' )! e< n+a+l)</> iZ;-. exp (Z! cosh 2 ¢ - lZ! e2"' j (I + O(l;-. ) 1.
2 1 2
sm ¢
The "principal parts" of G~ and H~ are given by the arcs

respectively. If we replace z by ei ..z in H~, we see immediately that these

principal parts are identical. Consequently, we have G~ = H~ except for terms
which are of higher order than the remainder term in (8.74.8); furthermore, G~
and H~ are of higher order than the same remainder term. Therefore, because
of C tn 1 = O(l~ ), we have
1 2

(8.74.9)
2~iG~ + '2~iH~ +2m (2~ia~) + 2W (2~iH~) + K'
= (11' si~h <Py e< 2 n+aH><~>z~ 1 exp (l! cosh 2 ¢ - il!e
2
"'l (I + O(l~2) j,
so that
L~a\~ )
2

( -Jc)n = -1(. h,~..)-!t-a-! 2 az-2n-a-J

4 r(n + a+ I) 11' sm "~' ., n
(8.74.10)
2
·exp(~ - il!e
2 2
"' + (2n +a+ I)¢}(I + O(tn )};
or
e-t2t2 L~a> (~2) = !( -Ir(7r sinh ¢) -! Ca-! n at2-t
(8.74.II) 1 l
·exp {[n +(a+ I)/2](2¢- sinh 2¢)j {I+ O(n-) .
Returning to the variable x, we obtain (8.22.IO). From this result (8.22.I3)
follows immediately.
8.75. Method of steepest descent; Laguerre polynomials for
x = 4n O(n1) +
(I) First, let t be real and bounded. We write as before
X = ~2, ~ = ln - (6ln)- 1t, W = lnZ 1

(8.75.I) L(a)(t2) {I I }
( - 1)n n ., = 2aCa(27r~)-!Z~2 n-a-!2m -.G" + -.H" + K" ,
4
r (n + a + I) . 211'2 211'2

where
 [ 8.75] LAGUERRE POLYNOMIALS 233

G" = J z-!exp U!J1(z)- 6-tz~tzl dz,

(8.75.2) H" = e<a+!),.i J z-! exp U!J2(z) + 6-tz~tz l dz,
K" = O(l-;. )
2
JI exp U!f.(z) =F 6-tz~tzlll dz I, ., =I, 2,

with
!()
1 z - - 1
4 z2+ z - 11
2 og z,
(8.75.3)
f 2 (z) - - 1
4 z2 - z - 1 1og z.
2

The integrals in (8.75.2) are extended over the upper half of a proper curve
symmetric to the real axis for which I z I and I z \- 1 are bounded.
The "saddle point condition" f~(z) = -z/2 +
I - (2z)- 1 = 0 furnishes
z = I, and we notice thatj~'(I) = 0 andj~"(I) = -1. Therefore, this saddle
point is of a different character from the preceding ones.
We first integrate along the segment
(8.75.4)

where ois a fixed positive number, o < I/6, then along the segment symmetric
to this one with respect to the imaginary axis, and finally along the arc of a

-I +I
FIG. 9

circle with center at z =:= 0 which connects the ends of the segments mentioned
(see Fig. 9). For certain constants c4 , C5, . . • , we have

(8.75.5)
f1(z) = ft(l) + ~! (z - I) j;" (1) + · · · 3

= £ - l-;. + C4(z-;.Jp) 4 + C0(l-;. 1p) 5 + · · ·.

2
/

Furthermore, if r is the radius of the arc of the circle, then

(8.75.6) mlft(rei"') l = - ir2 cos 21/; + r cos 1/; - t log r

is decreasing for 0 ~ 1/; ~ 1r, sinee r cos if; < 1. \Y e now obtain the following
expansiOn
 234 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII ]

G" = exp Iil~ - 6-tzit l .6t~ie 2 .... 1 ~

(8.75.7) ·ln& exp (- / - pe2.... 13t} ( 1 + (c~p + c~1p4) z-;;i

+ (c~/ + c~/ + c~'p 8 )l-;;! + · · ·l dp.
This is an asymptotic expansion in the usual sense. At the end-point p = n•
33
we have the bound O(exp [-cn ]), c > 0, for the integrand, which is at the same
time a bound for the contribution of the remainder of the contour. (The slight
modification of the circle around z = -1 is immaterial.) Thus
G" = exp (i l~- 6-tz~t l ·6i~Je 2....13
(8.75.8)
.·{1" exp (- p' - pe"''"t) dp + O(r;;!)}.
In the same way we see that the principal part of I-I" is due to the small segment
(8.75.9)
We find

!2(z) = !2(- 1) + ]_ (z + 1) 3h
11
(- 1) + · · ·
(8.75.10) 3!
= 3/4 - i1r/2 - z-;; 2/ +
Therefore,

H" = - e(a+i).-ie-i.-/2 ex p {3 z2n-

4 2 z2
i7r n- 6-izint}
(8.75.11)
1
· 6 r. 1 e''"{1" exp (- p' + pe'' t) dp + O{l:1)},
13

and
(8.75.12)

Now we can readily show that H" = - G" except for terms of higher order.
Consequently,

L~a) (~ )
2
( - .!)n
4 f(n+a+1)
= -~ ?a+t 6t .;ca-t z-2n-a-7/6
7r.., n
e p
X
I !!l2
\4n
- 6-i zi tl
n
(8.75.13~

·{s[e'•"' 1" exp (-p'- pe''"'t)dp]+ O(r;;')}.

The imaginary part in the braces is Airy's function A (t) (Problem 2). Sub-
stituting the approximate values of 4 n r(n + a + 1), l-;; 2 n-a- 716 , and ca-l
= z-;;"'- 1(1 + O(l-;;413 )}, and observing that
~ /2
2
(8.75.14) = z~;2- 6-tz~t + O(z-;;i),
[ 8.8] DIFFERENTIATION OF ASYMPTOTIC FORMULAS 235

we obtain
e-E
212
L~a\~ 2) = (-1)n7r- 1 2-a-!3!n-!!A(t) + 0(tn"1)},
which is (8.22.9) with the less precise remainder term O(n-t).
(2) Direct calculation of a further term in (8.75.5) leads to

(8.75.15) _ 1 j, (4)(1)64/3 Sri/3 _ 1

64/3 eSri/3 .
C4 - ! 1 e - 'g'
4
Furthermore, from (8.75.4)
(8.75.16)
Consequently, the expression in the braces in (8.75.8) can be written as follows

i"" exp (-/ _ pe2,i/3t)dp[1 +(-! 6!e2ri/3P + t 64/3e8ri/3/)tn"1 + O(tn"4'3)].

The corresponding more precise form of (8.75.13) can be obtained by replac-
ing the expression in the braces by

3 [ e2ri/31"" exp (- / - pe2ri/3 t) dp J

J
00

- ! 6t z-;;-1 3 [ e4ri/31 exp (- / - pe2ri/3 t)p dp

+ t 64t3z-;;1 3 [ elori/3 i"" exp ( _ / _ pe2,i/3 t)/ dp J+ O(tn"4/3)

= A(t) l!6!A'(t) + t6413 A< 4>(t)jl-;;1

+ + O(Z:13)
= A(t) + !6-1z:-1t2A(t) + ocz-;;-413).
Here the differential equation (1.81.2) has been used. Now (8.75.14) can be
also written in the more precise form

so that
e-t
212
= exp [-Z~/2 + 6!z~t]!I- !6-ttn"1t2 + O(tn"413 )}.

From this, (8.22.9) readily follows with the remainder term O(Z:13 ) = O(n-1).
(3) The case of a general complex t can be settled by a slight modifica-
tion of the argument in (1).
The corresponding asymptotic formula for Hermite polynomials follows
immediately by means of (5.6.1).

8.8. Differentiation of certain asymptotic formulas

Differentiation of an asymptotic formula with respect to a parameter occurring
in that formula is in general not permitted. In some of our previous formulas,
236 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII 1

however, the permissibility of differentiation is readily established. We must,

of course, modify the remainder in a proper way. We shall discuss the formulas
(8.21.18) and (8.22.6), involving Jacobi and Laguerre polynomials, from this
point of view, using the important identities (4.21.7) and (5.1.14).
(I) First let us consider (8.21.18) which contains Darboux's formula (8.21.10).
We shall prove that

:o IP~a,fl) (cos 0) l = ni k(O) 1- sin (NO + 'Y) + (n sin 0)-10(1)},

(8.8.1)
<
.a > - 1,{3>- 1,en-1 ;£0=1r-cn,
-1

with the same notation as in (8.21.10). We note that k'(O) = k(O)(sin 0)- 10(1).
In the proof, we write for the left-hand member of (8.8.1),

-t sin O(n + a + {3 + l)P~~~ .flH> (cos 0)

1

- -t sin O(n + a + {3 + 1) (n - 1)-i k(O)

1
1
· (sin; cos;) - ! cos (NO + 'Y - 1r/2) + (n sin o)- 0(1)},

which establishes the statement. Evidently, the remainder term of (8.8.1)

can be replaced by o-a-JO(n-!<) if en - 1 ;£ 0 ;£ 1r - E, and by O(n -l) if
E ;5_ 0 ;;£ 7r - E.
From (8.8.1) we derive the following important formulas:
P~a,fl)(cos 01) - P~a,fl)(cos 02)
01 - 02
= n-ik(01) cos (N01 + 'Y) -cos (N02 + 'Y)
01 - 02
+ 01a-!O(n-1) + (7r- 02)-fl-~O(n-i),
(8.8.2) P~a,fl)(cos 01) - P~a,fl)(cos 02)
cos 01 - cos 02
= n-t k(0 1) cos (N01 + 'Y) - cos (N02 + 'Y)
COS 01 - COB 02

Similar formulas hold with k(02) instead of k(01) in the right-hand member.
For the proof we apply the mean-value theorem to
¢(0) = p~a ,fl) (cos 0) - n -i k(O) cos (NO + ')').
Thus
[ 8.9 l APPLICATIONS; ASYMPTOTIC PROPERTIES OF ZEROS 237

¢'( T2) fh < Tl < 82,

---.-,
.sm r2 81 < T2 < 82 •
1 1
But, we have ¢'(r1) = n k(r1)(n sin ri)- 0(1), which is 81"-JO(n- 1) or
(1r - 82 )-tH O(n- 1), according as r1 ;;£ 1r/2 or r 1 ~ 1rj2. Moreover,
k(81) cos (N81 + 'Y) - k(82) cos (N82 + 'Y)
81 - 82

= k(81 ) cos (N81 + 'Y) - cos (N82 + 'Y) + cos (N82 + 'Y) ~_(81) - k(8 2).
81 - 82 81 - 82
The latter ratio is k'( r) = 01"- 10(1) +
(1r - 82)-tHO(I), 81 < r < 82 • A
similar argument leads to the second formula in (8.8.2).
If both 81 and 82 are confined to an interior interval E· ;;£ 8 ;;£ 1r - E of [0, 1r],
the remainders in the formulas (8.8.2) are simply O(n -i).
(2) Next we consider (8.22.6), which contains Fejer's formula (8.22.1).
We write (8.22.6) in the form
L~">(x) = k(x)n" 12 -l (cos [2(nx) 1 'Y] + +
(nx)-10(1)1,
-l x/2 -a/2-1
(8.8.3) k( X ) = 7r e X ', 'Y = -(a+ !h/2,
a > - 1, en -1 <
= x <= w,
where c and ware fixed positive numbers. By means of (5.1.14) we now obtain

(8.8.4) d(~!) (L~">(x)l 12

=·2k(x)n" +1( - sin [2(nx) 1 + 'Y] + (nx)-~O(l)l,
noting that d/k(x) l /d(x 1) = k(x)O(x- 1), and that (n - 1)! - n 1 = O(n- 1).
The mean-value theorem furnishes, if both X1 and X2 belong to the interval
en-1 <
= x <
= w,

(8.8.5) = k(x1)na/2-l cos [2(nx1) 1 + 'Y}- c~s [2(nx2) 1 + !_]

Xi -Xi
+ x1at2-iO(n"'2-1) + x;at2-iO(n"'2-l), a > -1.

The proof is similar to that used in the case of Jacobi polynomials.

8.9. Applications; asymptotic properties of the zeros of Jacobi and

Laguerre polynomials
(I) We shall first point out some consequences of the formulas (8.21.18)
and (8.8.1), which will be important for the discussion of interpolation with
Jacobi abscissas (Chapter XIV). Setting c = 1 in (8.21.18), and using the
same notation as before, we have the result:
238 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII J

THEOREM 8.9.1. Let a> -1, {3 > -1, and let 0 < 01 < 02 < ... <On< 1r
be the zeros of P~a,fJ> (cos 0). Then

(8.9.1) Or= n- jJnr

1
+ O(l)l,
with 0(1) being uniformly bounded for all values ofv = 1, 2, ... n; n = 1, 2, 3, . · ..
Furthermore,
(8.9.2) 0 < o. ~ 71"/2,

in the sense that the ratio of these expressions remains between certain positive
bounds depending only on a and {3.
Let p be a fixed number, 0 <p< 1rj2. Substitute in (8.21.18)
(8.9.3) NO = (v - !)1r - 'Y ± p, v > 0, v an integer, 0 < 0 ~ 1rj2.
Then the first term on the right, that is, ± ( -1)' n-!k(O) sin p, is the principal
term provided the remainder o-a-J O(n-!) is less than n-! k(O) sin p. This
is so if v and n are sufficiently large, v ~ M = M(a, {3, p). The same is true
for the formula (8.8.1). We have also 0 > 0 if M is properly chosen. Further-
more, let (v - !)1r - 'Y +
p ~ N1rj2, so that 0 ~ 1rj2. Then for the values
(8.9.3)

(8.9.4) sgn P~a,fJ>(cos 0) = (-1)' and (-1)·+~,

respectively. Hence P~a,fJ>(cos 0) has at least one zero between the bounds
given by the values (8.9.3) and, since (8.8.1) shows that it is monotonic in this
interval, it has only one zero there. Also, we see that in the same interval

(8.9.5)

and in the "complementary intervals" of [0, 1rj2],

(8.9.6)
The statements (8.9.5) and (8.9.6) hold in the sense that the ratio of the ex-
pressions in question remains bounded from zero and infinity, uniformly in 0
if 0 lies in the intervals mentioned; here n is sufficiently large, v ~ M(a, {3, p),
and {v - !)1r - 'Y +
p ~ N1rj2.
Thus p~a,fJ>(cos 0) has no zeros in the "complementary intervals."
In the interval 0 ~ 0 ~ N-1 ( (M - !)1r - 'Y - p} we have a bounded number
of zeros which have the form n - 1 (j~c +
En), where j1c denote the positive zeros of
J .. (z) and En ~ 0 (cf. (8.1.1) and (8.1.3)). This furnishes (8.9.1) for 0 <
o. ~ 1rj2. Taking into consideration the formula (4.1.3), the full statement
(8.9.1) follows.
In addition, we find, by use of (8.9.5) 1 that

(8.9.7)
[ 8.9 l APPLICATIONS; ASYMPTOTIC PROPERTIES OF ZEROS 239

provided the conditions mentioned are satisfied. The extension to the total
interval follows again by means of (8.1.1). Combining this result with (8.9.1),
we obtain (8.9.2).
For the zeros o. from a fixed interval [a, b] in the interior of [0, 1r], that is,
0 < a < b < 1r, (8.9.1) can be written in the more precise form
(8.9.8) Ov = N-1 ( (11 - !)7r - 'Y + K1r + Enl,
where K is a fixed integer (depending only on a, {:3, a, b), and En~ 0. In this
case, (8.9.2) attains the simpler form
(8.9.9) 1 P~a ,p), (cos o.) I ~ n!.

(2) The analogous statements for Laguerre polynomials follow from (8.22.6),
(8.1.8), and (8.8.4). We use here the same notation as before.
THEOREM8.9.2. Leta> -1,andletx1 < X2 < · · · <xnbethezerosofL~a>(x);
then we have for the zeros x. from a fixed interval 0 <x ~w
(8.9.10) 2x! = n-!(111r + 0(1)1.
Moreover,
(8.9.11) I L n(a)'( Xv ) I ~
-a/2-i a/2+i
Xv n ,-....., II
-a-J a+l
n .
Let p be a fixed positive number. Introducing in (8.22.6)
(8.9.12) 2(nx)! = (11 - !)1r - 'Y ± p, v > 0, 11 an integer, 0 < x ~ w,

we obtain values of opposite signs provided 11 ~ M = Jlr!(a, p), and n is suffi-

ciently large. Then we see from (8.8.4) that L~a> (x) is monotonic between
the corresponding values of x, and hence L~a> (x) has precisely one zero in the
corresponding interval. Taking (8.1.8) into account, we find (8.9.10) as in (1).
Again, from (8.8.4) and (8.1.8), 0 < x. ~ w,
I L~a>'(x.) I ~ x~!k(x.)na 12
+t.

For the zeros x. from a fixed positive interval e ~ x. ~ w, e > 0, we obtain

(8.9.13) I L n(a)'( Xv ) I ,..._, n a/2H .
(3) Let a be arbitrary and real. The preceding results (especially formula
(8.22.11)) enable us to discuss the largest zeros of L~a> (x~. Let i 1 < i 2 <
i 3 < ... be the zeros of Airy's function A (t); then the only zeros of L~a> (x) in
the interval
(8.9.14) x = 4n + 2a + 2 - 2(2n/3)tt, t real and bounded,

are those corresponding to the values t = i.. (Near t = i. we have precisely one
zero if n is sufficiently large; this follows from the uniform validity of (8.22.11)
for bounded complex t, by use of Theorem 1.91.3 (Theorem of Hurwitz).)
On the other hand, the upper bound (6.31.7) of the zeros belongs to this interval.
240 ASYl\IPTOTIC PROPERTIES OF CLASSICAL POLYKOMIALS [VIII ]

Consequently, if we arrange the zeros x1 > x2 > x3 > . -. of L~a> (x) in decreasing
order, we have
(8.9.15) Xv = X,n = 4n + 2a + 2 - 2(2n/3)t(i. + En), Jim
fl-00
En = 0, 'II fixed.

This is identical with (6.32.9). An analogous result holds for Hermite poly-
nomials, namely (6.32.5). It follows immediately from (8.22.14). This
formula was proved in §6.32 by use of Sturm's method.

8.91. Applications; asymptotic properties of the maxima of Laguerre and

Hermite polynomials
Another application is the discussion of the magnitude of the Laguerre and
Hermite polynomials for non-negative and for real values of x, respectively.
(1) THEOREM 8.91.1. Let a be arbitrary and real, a > 0, 0 < 7J < 4. We
have for n --} oo

I if a ~ x ~ (4 - 71)n,
(8 .91 . 1) max e-:r/2 X (a+1l/21 L(a)(
n X
)
l'o.J
{na/2
+~
na/2 •
if x ~a;

1 ?)
(8 .9 ... max e
-:r/2 af2+i I L(a)( ) I"-' {na/2-i if a~ X~ (4 - rJ)n,
X n X na/ 2_1112 .-.r
"J x ~a.

These maxima are taken, respectively, in the intervals pointed out in the right-
hand members.
These asymptotic formulas play an important role in the discussion of
Laguerre series (Chapter IX). 52
A more exact characterization of these maxima when n ~ oo is also possible.
·we can replace r v by by introducing into the right-hand members the
l'o.J

constant factors
(2/7r)t(4 71- 1 - 1) 1, 1r -\12) 1 max A (t);
(8.91.3)
1r -!(4 71 - 1)t, 1r -
1
(18) 8 max A (t),

respectively. This follows readily from the subsequent proof. We note that
these factors are independent of a and a. The first maximum of A(t) is posi-
tive, and is actually the greatest value of I A (t) I for all real values of t. (Cf.
Theorem 7.31.1 and (1.81.2).)
(2) Let t = j 1 be the first maximum point of A (t), 0 < j1 < i1 . Now choose
two values t', t" such that 0 < t' < t" < j1 and denote the corresponding

&2Kogbetliant?. (22, p. 144; 23, pp. 39, 51-53), states erroneously the appraisal L~a) (x) =
O(exf2z-af2-lf4naf2-l/4), x ~a. The error is made in 22, p. 154, where a certain bound valid for
H n(x), if I x 1 < cn112 , is appiicd to an arbitrary Hm(x), m ~ n. Unfortunately, the main
rcr,ults of the paper 23 (all!! partly also those of 24) arc based on this erroneous statement.
[ 8.91 l ASYMPTOTIC PROPERTIES OF THE MAXIMA 241

values of x in (8. 9. 14) by x' and x", respectively, so that x' > x" > x 1n if n
is sufficiently large. Then A (t') < A (t"), so that by (8.22.11)
(8.91.4)
Therefore, x' and x" both cannot'be on the left of the last extremum point of
e-x L~a>(x), whence this extremum point lies in the interval (8.9.14).
12

The same consideration applies to e-x 12x)o.L~a>(x), where A is an arbitrary but

fixed real number. Indeed, x)o. = (4n/ ( 1 + O(n-j) j. Thus, a formula analo-
gous to (8.22.11) holds for e-x 12 x)o.L~a>(x) with the additional factor (4n/ in
the right-hand member.
(3) In the interval a ;;£ x ;;£ (4 - 71)n formula (8.22.9) can be applied.
Since
X
(a+l)/2 ( · ,~,.)-!
Sin '+' X
-a/2-i
n
a/2-i • ,~,.)-! a/2-i
= X i ( Sin '+' n rv n
a/2 ( t
CO '+'
,~,.)!
,

X af2+i (sin </>)-!X- a/2-i n a/2-i = (sin </>)-I; n a/2-i'

the maxima in question are rvn af 2 and n a/ 2- 1, respectively.
In order to calculate the maximum in the interval x ~ a, we use Theorem
7.6.2. The sequence of the relative maxima of
(8.91.5)
is increasing for x > Xo , where xo is a certain non-negative number depending
on a, and n is sufficiently large. Therefore, ori account of (1), the absolute
maxima of the functions (8.91.5) for x ~a are attained in the interval (8.9.14).
According to (8.22.11) these maxima are ""' n<a+l)f 2 n-t and na 12 +in-l, respec-
tively. Between a and Xo, Fejer's formula (8.22.1) must be used. This
furnishes the complete proof.
(4) We also readily prove the following more general theorem:
THEOREM 8.91.2. Let a and A be arbitrary and real, a > 0, 0 < 71 < 4. Then
for n ~ oo

(8.91.6)
where
_ jmax (A - !, a/2 - t) if a ;;£ x ;;£ (4 - 71)n,
~a.
(8.91.7)
Q - lmax (A - t, a/2 - t) if x
In the first case, that is to say, in the interval a ;;£ x ;;£ (4 - 71)n, we again
apply (8.22.9). Now
. ,~,.)-! -a/2-i a/2-i
x)o.( sm '+' x n ""' n 'lo.-1;( cos '+',~,.)2'1o.-a- !(si· n '+',~,.)-! •
This expression attains its maximum in the interval E ;;£ </> ;;£ 1rj2 - m-l
for <J> = E or <J> = 1r/2 - En-l according as A ~ a/2 + t or A < a/2 + i.
The maxima are rvn'lo.-! and n'lo.-!(n-!) 2)..-a-! = na/ 2-t, respectively. This
242 ASYMPTOTIC PROPERTIES OF CLASSICAL POLYNOMIALS [ VIII ]

establishes the first part of the statement. We also see that the maximum is
attained near x = ( 4 - 71)n if A ~ a/2 + i, and near x = a if A < a/2 + t·
In the interval (4 - 71)n ~ x ~ 4n + O(n\ we have
max e-"' 12x"l L~a)(x) I"-' n"!.-(a+I)/ 2 max e-"' 12x<a+I)/ 2 \ L~a)(x) 1.
Because of (8.91.1), this furnishes the second part of the statement. The
maximum is attained near x = 4n if A - t > a/2 - i, and near x = a if
A - i < a/2- t·
(5) By use of (5.6.1), or directly from (8.22.12) and (8.22.14), we obtain the
corresponding results for Hermite polyno~ials.
THEOREM 8.91.3. Let A be arbitrary and r~al, a > 0, 0 < 71 < 2. Then
(8.91.8)
where
max (A/2 - 1/4, -1/4) if a ~ I xI ~ [(2 - 71)n]!,
(8.91.9) S= {
max (A/2 - 1/12, -1/4) if I xI ~ a; x real.
We have, for instance (Theorem 7.6.3),
Hn(x) I~ (2nn!)!2t3t7r-!n-1112 max A(t).
212
(8.91.10) max e-"' 1

Here x and t range over all real values. (Cf. Hille 1, p. 436, (30).)

8.92. Further results

( 1) Gatteschi (1, 2) investigated various asymptotic formulas involving the
classical polynomials and their zeros, and replaced the (usually unspecified)
constants occurring in the 0-terms by numerical values. As an illustration we
mention the following remarkable refinements of Theorem 8.21.6.
We have
Pn (cos e) = (fJ/sin fJ)!Jo{ (n +!)e) + u
where
\u\ < 0.09 fJ 2 if 0 < () ~ 1r / (2n ),
\u\ < 0.63 ()!n-a if 7r/(2n) < () ~ 1rj2.
The following formula is even more informative:

(si~ey Pn(cosfJ) = Jol(n+!)e)-

24
(n(J+!)Jd(n+!)e) +u'

where
\u'\ < 0.03() 4
if 0 < () ~ 7r/(2n),

\u'\ < 0.25 ()~n-~ if 7r/(2n) < () ~ 1rj2.

These formulas can be )J.sed for the asymptotic evaluation of the "first" zeros
with specified constants in the error terms. See also Tricomi 4.
 [8.92) FURTHER RESULTS 243

(2) Further information concerning the asymptotic behavior of the Jacobi

and Laguerre polynomials and their zeros can be found in Bateman Manuscript
Project, vol. 2, Chapter 10, pp. 196-202; also Tricomi 5, pp. 219-224, Thorne 1,
and Erdelyi-Swanson 1. See also the literature quoted at these places.
Recently, Erdelyi 3 has obtained two new asymptotic formulas for the
Laguerre polynomials L~al(x) valid for x ~ a(4n + 2a + 2) and for x ~
b(4n + 2a + 2) where a and b are fixed, 0 < b < a < 1. These formulas
involve Bessel functions and Airy's function, respectively, and hold on the
above ranges which overlap and cover the entire real axis. This is an important
result.
Erdelyi only proved these asymptotic formulas for a~ 0. They were ex-
tended to a> -1 by recurrence relations by Muckenhoupt 1.
(3) The formula (8.2l.i) of Laplace-Heine has been used in the estimation
of the smallest eigenvalue of the truncated Hilbert matrix (1/( i +j + l))f:J=o·
See Szego 27 and Widom-Wilf 1. The size of this eigenvalue gives a measure
of the degree of difficulty in finding the inverse of this matrix.
(4) The remark in §8.4 (3) is not quite correct. The infinite series corre-
sponding to Darboux's formula (8.21.4) is convergent for 1r/6< 8 < 57r/6,
but it converges to 2Pn(cos0) rather than Pn(cosO). See Olver 1. While
asymptotic series in the sense of Poincare must converge to the "right" function
if they converge this is not true of more general asymptotic expansions; so
great care must be exercised when using them.
 CHAPTER IX

EXPANSION PROBLEMS ASSOCIATED WITH THE

CLASSICAL POLYNOMIALS
The formal expansion of a function in terms of general orthogonal poly-
nomials having been defined in §3.1 (cf. (3.1.3)), we shall now turn our atten-
tion to expansions in terms of the classical polynomials. In this connection we
shall deal with the following problems:
Expansion of an analytic function in series of Jacobi, of Laguerre, and of
Hermite polynomiah;; discussion of the domain of the convergence.
Expansion of an "arbitrary" function in series of Jacobi,. of Laguerre, and
of Hermite polynomials; discussion of equiconvergence and summability
theorems.
In the main, our principal concern will be with ·the second problem, and we
agree that "arbitrary" function shall mean a function restricted only by certain
conditions of integrability, or continuity, and by conditions involving the
existence of certain integrals.
Two series L::-o Un and .Z::-o
Vn are called equiconvergent if the series
.Z::-o (un - Vn) or, more generally, if
00

L (un
n=O
- Avn), A r£ 0,

is convergent. We shall try to find simple trigonometric (Fourier) expansions

equiconvergent with a given type of polynomial expansion. Such a procedure
will enable us to reduce the discussion of the polynomial expansion to the dis-
cussion of trigonometric series under very general conditions.
Of the various methods of summability, we shall be primarily interested in
that of Cesaro. A series L::-o Un is called (C, k)-summable, k > -1, with
the sum s if
. s~k)
lim <k> = s,
n-+oo Cn
where
00 00

(1 - r)-k-l
n=O
L Un rn = L s~k) rn'
n=O

Obviously, k = 0 corresponds to convergence in the ordinary sense. If k' > k,

it is readily shown that (C, k)-summability involves (C, k')-summability with
the same sum.
244
 I 9.1 I RESULTS 245

A necessary condition for (C, k)-summability, k ~ 0, is un = O(nk).

Cesaro summability of any order k implies Abel summability; that is, the
existence of the
00

lim
r-I-0
.L:
n=O
'Un rn

with the same sum.

9.1. Results
(1) The formal expansion of a function f(x) m a Jacobi series is (compare
(4.3.3))
00

f(x) ""' L anP~a,fJ) (x),

n=O
(9.1.1)

The expansion of f(x) in a Laguerre series is (cf. (5.1.1))

00

f(x) ""' L anL~a) (x),

n=O
(9.1.2)
r(a + l)(n ~ a) an = l+oo e-x xaf(x)L~a>(x) dx;
and in an Hermite series it is (cf. (5.5.1))
00

f(x) ""' L anH n(x),

n=O
(9.1.3)
71J2nn!an = L:oo e-x f(x)Hn(x) dx.
2

In all these cases f(x) is a measurable function, and the existence of all
integrals occurring is required.
In what follows we assume a > -1, {3 > -1 in the Jacobi case and a > -1
in the Laguerre case.
(2) In this connection the following results may be stated:

THEOREM 9.1.1 (Expansion of an analytic function in a Jacobi series). Let

f(x) be analytic on the closed segment [-1, +1]. The expansion of f(x) in a
Jacobi series is convergent in· the interior of the greatest ellipse with foci at ± 1,
in which f(x) is regular. The expansion is divergent in the exterior of this ellipse.
Using the notation (9.1.1), we have the following representation for the sum R of
the semi-axes of the ellipse of convergence
(9.1.4) R = lim inf I an ,-l/n.
n-oo
I

i
I
I

I
I

i
_l__
 246 EXPANSION PROBLEMS FOH THE CLASSICAL POLYNOMIAI}S (IX]

THEOREM 9.1.2 (Equiconvergence theorem for Jacobi series in th!e interior

of the interval -1, +1). Letf(x) be Lebesgue-measurable in [-1, +~],and let
the integrals
+1

(9.1.5)
1
r+I
_
1
(1 - x)"(1 + x)P lf(x) Idx,
J_
1
(1 - xr"
12 1
- (1 + xl 12-t lf(x) I dx
exist. If sn(x) denotes the nth partial sum of the expansion of f(x) a Jacobi inl
i

series, and Sn(cos 0) the nth partial sum of the Fourier (cosine) series oJ
(9.1.6) (1 - COS O)"'f2+i(l + COS 0)PI 2Hf(cos 0),

then for -1 < x < + 1,

(9.1.7) lim {sn(x) - (1 - x)-a/ 2-t(l
n-oo
+ xrM 2
- 1 sn(x) l = 0,

uniformly in -1 + E ~ x ~ 1 - E, where E is a fixed positive number, 1 E < 1.

THEOREM 9.1.3 (Summability theorem for Jacobi series at the ehd-points
x = ± 1). Let f(x) be continuous on the closed segment [ -1, + 1]. The ~xpansion
of f(x) in a Jacobi series is (C, k )-summable at x = + 1, provided k :::j. a + ! .
This is in general not true if k = a +
! . An analogous statement hJlds for x
= -1, a being replaced by {3.

THEOREM 9.1.4 (Generalized summability theorem for Jacobi seri~s). Let

f(x) be Lebesgue-measurable in [ -1, +1] and continuous at x - +L! Then if
we assume the existence of the integral ·

/_~
1

(9.1.8) (1 - x)"'(1 + x)f3 If(x) I dx,

the Jacobi series is (C, k)-summable, k > a + !, at x = +1, provided t~at in the
case !

(9.1.9) {3 > -!, a + ! < k < a + {3 + 1,

the following additional "antipole condition" is satisfied: the integral

(9.1.10) L: (1 + x)P 12-t I f(x) I dx

exists. (Fork ~ a+ {3 +
1 no antipole condition is necessary.) Fork 'I~ a+ !
or fork > a +
!, but without the antipole condition, the statement is not true.
i
I

THEOREM 9.1.5 (Equiconvergence theorem for Laguerre series for Ix > 0).
Let f(x) be Lebesgue-measurable in [0, + oo ], and let the integrals

(9.1.11) i·t x., I f(x) I dx,

--+----
[ 9.1 I RESULTS 247

exist. lf the condition

(9.1.12) ioo e-:~:f2xaf2-13/12 If(x) I dx = o(n -!), n~ oo,

is satisfied, and if sn(x) denotes the nth partial sum of the Laguerre series of f(x),
we have, for x > 0,

(9.1.13) lim {sn(x) - 1r-

1
{"'t+s f(l) sin { 2~'(x' - r) l dr} = 0,
n-oo }:>:t-o X - T

where o is a fixed positive number, o < x 1. This holds uniformly for every fixed
positive interval E ~ x ~ w, o < El.
The same equiconvergence theorem (9.1.13) is valid if the integrals (9.1.11) tfXist
and condition (9.1.12) is replaced by the following:

(9.1.14)
1 00

e-"'12 xa12- 1 1 f(x) I dx is convergent;

i'o e-"'xa,-2 lf(x) l2dx = o(n-1), n ~ oo.

THEOREM 9.1.6 (Equiconvergence theorem for Hermite series, x arbitrary

and real). Let f(x) be Lebesgue-measurable in [- oo, + oo], and let the integral

(9.1.15) J~a If(x) I dx

exist for every a > 0. If the condition

(9.1.16) n ~ oo,

is satisfied, and if sn(x) denotes the nth partial sum of the Hermite series of f(x),
we have, for an arbitrary and real x,

(9.1.17) lim {sn(x) - 71"-1 1:>:+/J f(t) sin {(2n)l(x - t) l dt} = 0,

n-oo :>:-& X - t

where o is a fixed positive number. Moreover, (9.1.17) holds uniformly in every

finite interval.
The same equiconvergence theorem (9.1.17) holds if the integral (9.1.15) exi$ts
and we replace condition (9.1.16) by the following:

1 00

e-'"
212 1
x- {lf(x) I+ Jf(-x) J} dx is convergent,
(9.1.18)
100
2 4
e-"' x- 11 f(x) 1
2
+ If( -x) 1 2
} dx = o(n-
3
)., n ~ oo,

THEoREM 9.1.7 (Summability theorem for Laguerre series at x = 0). Let

248 EXPAI\SIO.N PROBLEMS .FOR THE CLASSICAL POLYNOMIALS [ IX]

f(x) be Lebesgue-measurable in [0, + oo] and continuous at x = 0. lf we assume

the existence of the integral

(9.1.19) loo e-x/2Xa-k-i if(x) I dx,

the Laguerre series of f(x) is (C, k)-summable at x = 0 with the sumf(O) provided
k > a+ !. This statement is not true fork ~ a+!.

9.11. Remarks
(1) Theorem 9.1.1 is well-known in the Tchebichef and Legendre cases. 53
The determination (9.1.4) of the ellipse of convergence is analogous to the well-
known Cauchy-Hadamard formula.
(2) The importance of Theorem 9.1.2 is due to the possibility of applying it
to convergence and summability ·problems for Jacobi series similar to those
which are found in the classical theory of ordinary Fourier series. This theorem
has been proved in the special case a = {3 = 0 (Legendre series) by Haar (2)
and W. H. Young (1) with conditions concerning f(x) somewhat different from
those arising from our general theorem in the special case a = {3 = 0. W. H.
Young considers also series proceeding in terms of Legendre polynomials, with-
out being Fourier expansions in the ordinary sense. The proof of Theorem
9.1.2 given in §9.3 is due to Szego (17, pp. 88-92). Recently, Obrechkoff (2)
has treated the same problem, using as his main tool Darboux's formula in the
more precise form (8.21.18).
Summability theorems for interior points were investigated earlier (in the
ultraspherical case) by Adamoff (2) and (in the ultraspherical as well as in the
general Jacobi case) by Kogbetliantz (1, 2, 3, 7, 18, 19). Concerning properties
of these expansions analogous to those treated in Riemann's theory of trigo-
nometric series (in particular theorems of uniqueness) see Kogbetliantz (6, 20)
and Zygmund (1).
In setting up the expansion in Jacobi series, we must require the existence of
the first integral (9.1.5). Using an obvious notation, we readily see that

(9.11.1)

according as a ~ -! or -1 < a ~ - t. That the condition of the existence

of the second integral cannot be improved upon for a > - t follows when we
consider the special function f(x) = (1 - x)"' with 11- = -a/2 - 3/4 (cf. §9.3
(4)). The existence of both integrals (9.1.5) follows from that of

(9.11.2)

63 The Legendre case is usually attributed to F. Neumann (cf. Whittaker-Watson 1,

pp. 322-323).
I 9.11 I REMARKS 249

Instead of (9.1.6) the function

(9.11.3)

may also be considered (cf. Szego, loc. cit.). The change, necessary in (9.1. 7)
for this function, is at once apparent. Both functions (9.1.6) and (9.11.3) are
integrable in -1r ~ 0 ~ +1r. The difference (9.1.7) may be written as

where the notation (4.5.2) has been used. Comparison of sn(x) = sn(a, {3; x)
with sn('Y, o; x), where 'Y and oare arbitrary, is also possible.
(3) Theorems 9.1.3 and 9.1.4 have an extensive literature. Gronwall (1, 2)
has investigated the special case a = {3 = 0 (Legendre series). His proofs have
been simplified to a considerable extent by Lukacs (2), Hilb (1), and Fejer (8).
The ultraspherical case has been considered in great detail by Kogbetliantz
(2, 4, 19; 21, pp. 70-73) (cf. also Obrechkoff 1). The method used in §§9.4-9.42
is new and comparatively simple. It is based on a peculiar relationship between
the Jacobi polynomial P~"H+I.tll(x) and the "kernel" of the kth Cesaro mean
of the Jacobi series which corresponds to the parameters a and {3. (In the case
of Laguerre series the corresponding relation is trivial (cf. §9.6).) The case
of a Legendre series (if we consider the series at the end-point x = + 1) is equiva-
lent to a Laplace series; whence the term "antipole condition". At the end-
point x = -1 a similar theorem holds; in this case the summability index k
must ~xceed {3 + !, and an "anti pole condition" must be satisfied near x = + 1.
Theorem 9.1.3 furnishes the convergence of the Jacobi series at the end-point
x = + 1 provided f(x) is continuous in [ -1, + 11 and -1 < a < -! (cf. also
Rau 1; Szego 17, §20; Lorch 1, 2).
In the case of Legendre series the "kernels" of the Cesaro means of the second
order are non-negative (Fejer 4). This fact has, of course, important conse-
quences. Concerning similar theorems in case of ultraspherical expansions,
cf. Kogbetliantz 19.
In the special case of Legendre series Kogbetliantz (16) gave important
refinements of Theorems 9.1.3 and 9.1.4. He studied, among others, the C-
summability of proper order at the end-points x = ± 1 if f(x) becomes infinite
of a certain order at the antipole.
As regards certain older (very complicated) results on the convergence of ultra-
spherical expansions at the end-points x = ± 1, see Adam off 2.
(4) The equiconvergence problem for Laguerre and Hermite expansions
has been treated (the first only in the specia.l case a = 0) by Rotach (1), and
(in all cases) independently by Szego (10). For the special Laguerre case a = 0
the conditions of Szego are more restrictive than those of Rotach, and the same
250 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [ IX ]

is true in the Hermite case. The conditions formulated in §9.1 are, however,
slightly more general than those of Rotach.s 4
That the condition of the existence of the second integral (g.l.ll) cannot be
improved upon is shown by considering the expansion of the functionj(x) = xp.,
p. = -a/2 - 3/4, in a Laguerre series (§9.5 (6)).
Neither of the two sets of sufficient conditions formulated in Theorem 9.1.5
contains the other. Indeed, the function f(x) = ex12x-a 12-i satisfies (9.1.14),
but not (9.1.12). On the other hand, if

(9.11.5)
·-x/2 af2j( )
e X X =
{1, m2 ~ X. < m2 + 1, ·
0, otherw1se, m = 1, 2; 3, · · · ,
then (9.1.12) is satisfied, but not the second part of condition (9.1.14).
Condition (9.1.12) implies, of course, that the integral on the left side exists.
From the condition (9.1.12) the first part of (9.1.~4) follows. In fact, if we write

(9.11.6) u(x) = Ioo e-t{2taf2-13/121 j(t) I dt,

we find from (1.4.4) that

(9.11.7) 1"' 1 1
x du(x) = w u(w)- u(1) -!1"' x-1u(x)dx

is bounded asw ~ oo, since u(x) = o(x-1).

A sufficient condition for the validity of (9.1.13) is that

(9.11.8) 0 > 0, X~+ oo.

Then (9.1.14) is satisfied. On the other hand, for j(x) = ex12x-a 12H conditions
(9.1.11) are satisfied, but not (9.1.12) and (9.1.14), and (as we are going to show
in §9.5 (7)) the Laguerre series is divergent for x > 0.
A sufficient condition for the validity of (9.1.17) is that

(9.11.9) 0 > 0, X~ oo.

212
Then (9.1.18) holds. An analogous "Gegenbeispiel" here is f(x) = xex .
From Theorems 9.1.5 and 9.1.6 there follow the usual theorems on the con-
vergence and the summability of Laguerre and Hermite expansions. Indeed,
the integrals occurring in (9.1.13) and (9.1.17) are essentially the partial sums
of order [ni] of a Fourier series (cf. (1.6.4)).
As early as 1907 Adamoff (2) obtained a convergence theorem for Hermite

64 Rotach's second condition b~ on p. 8 makes the first part of b; superfluous. His

first set of conditions is equivalent to (9.1.11) plus (9.1.14) (for a= 0), whereas his second
set is more restrictive than (9.1.11) plus (9.1.12). In the theorem on p. 6, the second
2
I
condition b2 must be corrected to read "fie-• 14 IJC±z) z312 dz exists." (I owe this to a
written communication from Mr. Planchercl.) This, of course, implies b1. Here the
first set of conditions is more restrictive than (9.1.18), and the second set is more restrictive
than (9.1.16). (Rotach's notation differs from ours; we must write z = 2! x.)
[ 9.2) JACOBI, LAGCERRE, AXD HER\IITE SERIES 251

expansions. There is an extensive further literature on this subject; E. R.

~eumann (1), Galbrun (1), \Yigert (1), Hille (1, 2, 3), Cramer (1), Uspensky (1),
Korous (1, 2), Stone (1), :\Hintz (1), and Kowallik (1) have all given direct
treatments of Laguerre and Hermite expansions of an "arbitrary" function.
(Concerning the expansion of an analytic function intG Hermite series see \Vat-
son 1 (first paper).) These authors have obtained convergence and summability
results but. no equicom·ergence theorems, except Korous (loc. cit.). The con-
ditions which all these authors use at x = cc are more restrictive than those in
our theorems. For instance, in case of Laguerre expansions, the convergence of

(9.11.10) j"" e-x xaf -llf(x) I dx

12 2

is required by Korous; for Hermite expansions, convergence of the integrals

(9.11.11)

are required by Uspensky and Kov~·allik, respectively. In addition, we mention

the earlier treatment of Laguerre and Hermite expansions by means of the
theory of integral equations (:\1yller-Lebedeff 1, eyl1). 'r
Concerning Kogbetliantz 22, 23, 24, see the footnote in §8.91. Concerning
Theorem 9.1.7, see Szego 10 and Kogbetliantz 10. The condition regarding
(9.1.19) is satisfied if
(9.11.12) 0 > 0, X -} + oo.
On the other hand, the series is not (C, k)-summable for the function f(x) =
ex12xk-a' k > a + 1/2 (cf. §9.6 (3)).
Gibbs' phenomenon in the case of Hermite expansions has been studied by
Jacob (2).
9.2. Expansion of an analytic function in Jacobi, Laguerre, and Hermite series
(1) THEO~EM 9.2.1. Assume a> -1, {3 > -1, and let P~a,fJJ(x), Q~a,fJJ(x),
and h~a,fJJ have the same meaning as in Chapter IV (see §4.61 and (4.3.3)); then

(9.2.1) ta (h~a,,S) l-1 p~a,{J) (x)Q~a,{J) (y) = ~ (y - 1~-~YX + 1)-fJ

where x lies in the interior of, andy in the exterior of, an arbitrary ellipse with foci
at ± 1. This expansion holds uniformly if x and y belong to closed sets which are,
respectively, in the interior and exterior of the ellipse mentioned.
The functions (y - 1)a(y + 1)fJQ~a,fJJ(y) are single-valued and regular in the
complex y-plane cut along [ -1, + 1].
This important formula is well-known in the special case a = {3 = 0 (Heine 3,
p. 78). The general formula is obtained from the identity (4.62.19) by letting
n approach 00. We write X = Hz + Z- 1), y = Hr + r- 1); 1 < i zl < Ir I. By
252 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [IX]

use of (8.23.1) and (8.23.2) we find the "remainder term" of (4.62.19) to be

2-a-tl r(n + 2)r(n +a+ {3 + 2)
2n +a+ {3 + 2 r(n + a + 1)r(n + {3 + 1)
(9.2.2)
. P~+~)(x)Q~a,fl)(y)- p~a,fl)(x)Q~~q> (y) = O(n)O[(I z I + E)"]O[(I r l-1 +E)"]
x-y
as n-----+ oo, with E > 0 arbitrarily small. This tends to 0 as n-----+ oo provided E is
sufficiently small.
(2) Now let f(x) be regular if xis in the interior of the ellipse I z I = R > 1.
On multiplying (9.2.1) by (7r'i)-\y - 1)a(y + 1)tlf(y) and integrating over
the ellipse I t I = R - E, 0 < E < R/2, we obtain an expansion of f(x) in a Jacobi
series whicn is uniformly convergent for I z I ~ R - 2E. The usual term-by-
term integration over the segment [ -1, + 1] identifies this expansion with the·
expansion (9.1.1), and for the coefficients a,. we obtain the representation

(9.2.3)
a,.= {rih~a.tl)}-1 f (y- 1)a(y + 1)tiQ~a,tl)(y)f(y)dy,
n = 0, 1, 2, ... ,

where the integration is extended over the ellipse I t I = R - E in the positive

sense.
(3) By means of (8.23.1) we can discuss formal series of the type
(9.2.4) aoPaa,fl>(x) + a1P~a,t1>(x) + ll2p~a.tl>(x) + ... + a,.P~a.tl>(x) + · .. ,
which are not necessarily Fourier expansions in the ordinary sense. Let R
have the same meaning as in (9.1.4), and assume R > 1. Then (9.2.4) has as its
domain of convergence the ellipse of Theorem 9.1.1. In the interior of this el-
lipse it represents an analytic function.
(4) Expansions in terms of Jacobi's functions of the second kind, which are
the analogues of Laurent expansions, can also be readily discussed.
THEOREM 9.2.2. Assume a > -1, {3 > -1, and let F(y) be regular at y -
oo with F( oo) = 0. Then

(y - 1)-a(y + 1)-tiF(y) = boQaa,fl>(y) + b1Q~a,fl>(y) + b2Q~a,fl>(y)

(9.2.5) ( fl)
+ · · · + bnQna, .(Y) + • • · •

This expansion is convergent in the exterior of the smallest ellipse with foci at ± 1,
in the exterior of which F(y) is regular. It is divergent in the interior of this ellipse.
The sum of the semi-axes of this ellipse is given by
11
(9.2.6) p = lim sup j b,. 1 ".
n-oo

Obviously, p ~ 1. In case p = 1 the statement needs a slight modification.

The following representation holds for the coefficients:
[ 9.3 J PROOF OF THEOREM 9.1.2 253

(9.2.7) n = 0, 1, 2, · · · ,

where the integration iR extended over the ellipse I z I = p + E, x = Hz+ z- 1),

E > 0, in the positive sense. For the proof we again use (9.2.1).
Formal series proceeding in terms of Jacobi's functions of the second kind can
also be discussed.
(5) The boundary of the convergence domain of a Laguerre series
(9.2.8) aoLaa>(x) + a1L~a>(x) + a2L~a>(x) + · · · + a,.L~a>(x) +
can be characterized by the condition m((
-x)i l = const. Therefore, this
boundary is a parabola with its focus at the origin. The series is convergent in
the "interior" of this parabola and divergent in its "exterior". The analogue
of Cauchy-Hadamard's formula holds. The proof is based. on (8.23.3).
For an Hermite series
(9.2.9)
the corresponding condition is I.S(x) I = const., which defines a strip with the
real axis as axis of symmetry (cf. (8.23.4)). The series is convergent and di-
vergent in the "interior" and in the "exterior" of this strip, respectively. The
analogue of Cauchy-Hadamard's formula holds again.
Concerning the expansion in an Hermite series of a given analytic function
regular in the strip I .S(x) I ~ a, a > 0, see Watson 1 (first paper).
9.3. Proof of Theorem 9.1.2
(1) First let us replace f(x) by a polynomial p(x). Then s,.(x) = p(x) if n
exceeds the degree of p(x). Furthermore, (1 - x)-a/ 2- 1(1 + x)-tl12- 1s,.(x) ----+ p(x)
as n ---+ cx:::, according to elementary tests for the convergence of Fourier series
(see for instance Zygmund 2, p. 25). Now the integral

(9.3.1) 1: 1
(1 - tt(l + t)b I f(t) - p(t)l dt,
fa= min (a, a/2- 1/4),
lb= min ({3, {3/2 - 1/4),

can be made arbitrarily small by proper choice of p(x) (cf. Theorem 1.5.2). It is
therefore sufficient to show that the difference (9.11.4) admits an estimate
+1
(9.3.2)
1
0(1) _
1
(1 - tt(l + t)b 1 f(t) I dt + o(1),
where both bounds hold uniformly in x, - 1 + E ~ x ~ 1 E, and 0(1) is

independent of f(x).
(2) In subsequent considerations we use Darboux's formula (8.21.10) for
p~a.tl>(cos 8), as well as the second formula (8.8.2) for the difference ratio. In
the latter case we assume that both arguments 81 , 82 lie in an interval which is
entirely in the interior of [0, 1r].
According to (4.5.2), we have
254 EXPAKSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [IX]

K~a,tl) (x, t)
(9.3.3) = '\ p<a,fl)(')
p<a,tl)(t)
n+1 - p<a,fl)(
n+1 X ) _ p<a.tl)()
p<a,tl)(t)
n -
p<a,fl)( )}
n X
1\n { n X t -x n+1 X t -x '

where
(9.3.4) An = 2-a-tl-1 {n + 0(1)}.
Writing x = cos 8, t = cos~' and using the notation of (8.21.10), we find for
-1 <X< +1, -1 < t < +1,
K~a,fl>(x, t) = 2-a-fl-lk(())k(~){cos (N() + -y) cos[(N +1)~+-r]- cos[(N+ 1)8+ -y]
cos~- cos()

_ cos [(N + 1)8 + -r] cos (N~ + -r) - cos (N8 + -y) + 0(1)}
cos~- cos()

= 2 -a-fl-2k(())k(~) {cos [N(~ + 8) + ~ + 2-y] + cos [N(~ - 8) + ~]

cos~- cos()

(9.3.5) _ cos [N(~ + 8) + 8 + 2-y] + cos [N(~ - 8) - 8] + 0( )}

1
cos~- cos()

= 2-a-tJ-2k(())k(~) {sin [(N + 1)(~ + 8) + 2-y]

. ~ + ()
sm--
2
+ sin [(N + !)(~ - 8)] + 0( 1)}.
sin~ - ()
2
Now assume -1 + E ~ x ~ 1 - E. Then
12
+ t)flj(t)K~a,tl)(x, t) dt
1

1 -•
-1+•/2
(1 - t)a(1

= ["-'1 2a+t1+1(sin ~ya+l (cos ~ytl+l f(cos ~)K~a,fl) (x, t) d~,

where cos 1J = 1 - E/2. Next replace K~a,fl>(x, t) by (9.3.5). By Riemann's
lemma (Zygmund 2, p. 18), the result will be

.,._'1 sin (2n + 1) - -

~ - ()
1-•/2

(9.3.6) 2~ 1 f(cos ~) .
sm--
~_
2
()
2
d~ + 0(1) 1 I
. -1+•/ 1/(t) dt+ o(1),
2

as n----+ co.

The bound of the term 0(1) is independent of f(x). When we replace a and {3
by -!, andf(t) by (1 - tt 12+1(1 + t)ti!Hif(t), repeated application of Riemann's
lemma yields
 [ 9.3 J PROOF OF THEOREM 9.1.2 255

= -
1
27T'"
1. -, (1- cos~t' +i(l + cos~)PtHlf(cos~)
2
sin (2n + 1) - -
2 d~
~- ()

• ~-()
s m --
2
(9.3.7) + 0(1)
1 12
-• 1/(t) I dt + o(1)
-1+•/2
1
1 . _, sin (2n + 1) ~-
- -()
=-2 (1-cosO)a/Hl(1+coso)M2+l
7r 1"
f(cos~) 2 d~
•
s m --
2
~-()

+ 0(1) 11
-•
-1+•/2
12
jj(t) I dt + o(1).

Consequently, the part of (9.11.4) which corresponds to - 1 + E/2 ~ t ~

1 - E/2 has the form (9.3.2).
(3) We consider now the expressions

Ip<a.p) (t) p<a,p) ( ) I

(9.3.8) O(n) !p~a,p) (x) I
1 1

-•/2
n+ 1 -
t - X
n+l X (1 - t)a(1 + t)P lf(t) I dt,

O(n) Ip~-!.-!> (x) I

(9.3.9)
l
.
1 1

-•/2
lp<-l.-!>(t)
n+1 -
p<-l.-iJ( )
n+1
t -
X
X
(1 _ t)a/2-1(1 + f)P/2-ljj(t) I dt,

and the corresponding integrals extended over -1 ~ t ~ -1 + E/2. The first

integral is

O(n)O(n-i) [I
11-•/2
{jP~+~>(t) I+ /P~+~l(x) /}(1- tt(1 + t)P lf(t) /dt

From (7 .32.6) and (7 .32. 7) it follows that the first term of the right-hand mem-
ber is

{" 2 1
0(1) }o ~-a-i~ a+ 1/(cos ~)I~= 0(1) _ 1+1 (1- t)a/ -i(l + 2
tl 12- 1 lf(t) I dt,
1
or 1s
25G EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [ IX ]

0(1) }or~ ~ a+l If(cos ~) I d~ = 0(1) 1~

2
_ (1 - t)a(1
1
+ t)P I f(t) I dt,
according as a ~ -! or a ~ - t. In view of (4.1.7) the expression (9.3.9) is

O(n)O(n- 1 )0(~-!) e
11-•/2
(1 - t)a/ 2- 1(1 + t)M-i I f(t) I dt
~

1
= 0(1) _ (1 - t)a 12 - 1(1
1
+ t) 1312 - 1 1J(t) 1 dt.
The integrals corresponding to -1 ~ t ~ -1 + E/2 are similarly treated.
(4) Finally, we consider the Jacobi series of the function (see, in the special
case a = {3, Kogbetliantz 19, p. 184, (62))
(9.3.10) f(x) = (1 - x)",
and we show that for proper values of 1-' the first integral in (9.1.5) exists, but
that the second does not. Moreover, the Jacobi series is divergent in -1 < x <
+L Here a> -t.
To this end we note that, according to (4.3.1) and (4.3.3),
+1
{h~a,{j) }-1
1_
1
(1 - t)!'+a(l + t)P p~a,{j)(t) dt

= {h,.
(a,{j) -1 ( -1)" JJ.(JJ. - 1) · · · (JJ. - n
} -- 2 r
+ 1)
"n.
(9.3.11)

-~<-1
+ + 1)
n2"f(-JJ.) 2
::::::::: -a-P n n+~<+a+P+1 -14-a-1 r(
- 2 n 1-' a

as n -----+ co.

If we assume that 1-' +a> -1, 1-' ~ 0, 1, 2, · .. , and take 0 < () < 1r, the general
term of the Jacobi series has the form n - 214-a-J cos (NO + 'Y). The required
values of 1-' are thqse which satisfy the condition
(9.3.12) -1 - a < 1-' ~ - a/2 - 3/4.
9.4. Proof of Theorem 9.1.3; preliminaries
(1) We first derive the following important identity:

p<a+k~.P>( ) = 2a+H1 r( + 1) f(n + {3 + 1)

n X a f(n + a + {3 + k + 2)
(9.4.1)
.t
•=0
r(n
r(n·
+
v +a+ {3
+
k
v +a {3 2)
2) C~k~.{h~a.P> ~-1 p~a,P>(l)P~a.P>(x),
+ +
+ +
where h~a .PJ and C~k> have the usual meaning (see (4.3.3) and the introduction to
[ 9.4 J PROOF OF THEORE:\I 9.1.3; PRELL\IIXARIES 257

Chapter IX). This is a generalization of (4.5.3) (last identity), to which it re-

duces when k = 0.
For the proof we calculate

(9.4.2)
r+1 P~a+H .tll (x)P}a,fl) (x)(1
1- 1
. - xt(1 + xl dx.
1

According to (4.3.1), this last expression is equal to

<2.:( j_:1 p~a+H1.tl) <x) (:x)" t (1 - xr+a<1 + xr+tll dx

= _!_ (+1 (1 - x)"+a(l + xy+t~ (!!:_)" {P~a+H1,fl>(x) l dx;
2•v! 1-1 dx
or, by (4.21.7) and (4.3.1), it is equal to
1 1
2 •v! 2- (n +a+ {3 + k + 2)(n +a+ {3 + k + 3) · · · (n +a+ {3 + k + v + 1)

(-1)"-· r(n+a+f3+k+v+2)
- 2n+• v! ( n - v)! r( n a {3 + + + +
k 2)
1
+ (1 - x)-k- 1
1-1 dx
(.!__)"-• {
(1 - x) n+a+k+1(1 + x) n+tl} dx

=
(k + 1)(k + 2) · · · (k + n - v) r(n + a+ {3 + k + v + 2)
2n+•v!(n- v)! r(n +a+ {3 + k + 2)

·1+1 (1 - x)-k-1-n+•o - x) n+a+k+1(1 + x) n+tl dx,

-1

which, if (1.7.5) is taken into account, furnishes the statement.

(2) On substituting the explicit expressions for P~a,fl>(1) and h~a,fl> (see (4.1.1)
and (4.3.3)), we obtain

p~a+k+l.tl)(x) = r(n + {3 + 1) t r(n + v +a+ {3 + k + 2)

(9.4.3)
r( n + a + {3 + k + 2) •=0 r( n + v + a + {3 + 2)
·C(k) (2 + + + 1) r(v +a·+ {3 + 1) p<a.tl)(x).
r(v + {3 + 1)
R
n-• v a 1.1 •

Since this is an identity in a, {3, k, we may replace a by a + k + 1 and k by

-k - 2. Whence the inversion formula
p<a,fl)(x) -
n
r(n
- r(n +
+ +
{3
a+ {3
1)
+ +
+r(n
1) •=0 r(n
t + +
v +a+ {3
+
v +a+ {3
k
1)
3)
(9.4.4)
.c<-k-2)(2 + + + k + 2) f(v +a+ {3 + k + 2) p<a+k+l,fi)(X)
R
n-• v a 1.1r(v + {3 + 1) •
258 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [IX]

follows. Consequently,
n
S~k\x) =m-o
L C~k2m(h~a,fl>
.
}-1 p~a,t1>(1)P~a,fl>(x)

2-a-fl-1 n

(9.4.5) = r(a + 1) ~ G.(n, k)(2v +a+ {3 + k + 2)

. r(v + a + {3 + k + 2) p<a+k+l,fl) ( )
f(v + {3 + 1) • X '

with

(9 4 6) G ( k) = ~ c<k) c<-k-2) (2 + a + + 1) r(mr(m++v +v +a +

a+ {3 + 1)
{3 + k + 3)'
R
. . , n, ~ n-m m-• m 1.1

The expression S~k> (x) represents the numerator of the nth Cesaro kernel of
order k.

9.41. Continuation; the Lebesgue constants of order k

(1) For the proof of Theorem 9.1.3, according to Theorem 1.6 (Reily's the-
orem), it is sufficient to show that the sequence of the "Lebesgue constants"

(9.41.1) L~k) = { c~k)} - 1

1
i:
(1 - xt(1 x)tl Is~k) (x) Idx,
+
is bounded if and only if k > a + 1/2. Since fork > a +t (cf. (7.34.1), first
and third case),

[:1 (1 - x)a(1 + x)tl j p~a+H1,fl) (x) Idx

( .41. ) = 0(1) 1\1 - xt I p~a+H 1 .tl>{x)j dx P~·a+k+I>(x)ldx
9 2
+ 0(1) 1\1 - x)tlj

= O(nk-a-1) + O(n-i) = O(nk-a-1),

we have from (9.4.5)
n
(9.41.3) L~k> = O(n-k) L IG.(n, k) I v2k+l.
•=0
2
The last factor v k+l must be replaced by 1 for v = 0.
(2) First let k be a non-negative integer. Then in (9.4.6) we must consider
only the terms v ~ m ~ v k + +1, so that G.(n, k) can be written in the form
G(n - v - 1), where 55

66 If v + k + 1 > n, certain additional terms not permitted by (9.4.6) occur in G(u).

However, these terms vanish if we substitute u = n - " - 1, except if " = n and m =

n + k + 1. Now the contribution of this term to G(u) = G(-1) is O(m)(m + v)-k-2 =
0 (n-k- 1 ). This multiplied by the factor v2k+ 1 = n 2k+ 1 , which occurs in (0.41.3), furnishes
a total contributionO(n-k)O(nk) = 0(1).
[ 9.41] CONTINUATION; THE LEBESGUE CONSTANTS OF ORDER k 259

1
G(u) = ·~ ( -1) m-• (u + k + 1 - m + v)
m-• k
(9.41.4)
·(k + 1) (2m+ a+ l3 + 1) ~m + v +a+ 13 + 1)
m - v r(m + v + a + 13 + k + 3) ·
This is a 'Irk in u, containing v and k as parameters. ~ewton's formula shows us
(cf. Markoff 5, p. 15; notation as in (2.8.4)) that

(9.41.5) G(u) = t (u)

p=O p
t:l G(O);

and from (9.41.4) it follows that

t:lG(O) =•!1(-1)m-•(k +1-m+ v)(k + 1)

m-• k - p m- v
. (2m+ a+ 13 + 1) r(m· + v +a + 13 + 1)
r(m + v +a+ 13 + k + 3)
1
= 1 t " ! ( -l)m-•(2m +a+ 13 + 1) tm+•+a+f1(1 _ t)k+l dt
(k-p)!}o m-• (m-v)!(p+1-m+v)!
.
=
1 ll{ -2 •+P+1
L (-1)m-•-l tm+•+a+tl(1 - t)k+l
-----:----:-:
(9.41.6) (k- p)! 0 m-•+1 (m - v - 1)! (p + 1 - m + v)!
•+~I m-• r+•+a+fl(1 - t)k+I }
+ (2v +a +13 + 1) 11
!=:, (-1) (m _ v)!(p + 1 _ m + v)! dt

= 1
(k- p)!(p + 1)! }o
t{ -2(p + l)t2•+a+fl+I( 1 _ t)k+p+I

+ (2v + a + 13 + 1)t2•+a+fl(l - t)k+P+2}dt

= O(v-k-p-2) + O(v )O(v -k-p-3) = O(v -k-p- 2), v ~ 1.

The last integral formula also shows that t::.kG(O) = 0, v ~ 1; in particular,

G(u) = 0 for k = 0, v ~ 1. Both facts hold also for v = 0, as analytic con-
tinuation with respect to a + 13 shows. This settles the case k = 0 in view of the
remark in the last footnote. Fork > 0,
k-1
(9.41.7) G.(n, k) = L (n - v)P O(v -k-p- 2),
p-0

to which (according to the same remark) the term O(n -k-I) must be added if
2
v = n. In this case the factor (n - vY, in the case v = 0 the factor v-k-p- ,
must be replaced by 1. Fork > a + ~'this provides, as desired,
k-1 n-1 k-1
L ~k) = O(n -k) L
nP + O(n -k) L L
(n - vY v-k-p- 2v2k+ 1
p=O •=1 p=O
(9.41.8)
+ 0 (n-k) {
k-1
"" -1..-p-2 2k+I
LJn n
+ n -k-1 n 2k+I r=
~
'
O(J)
.
p-0
2GO EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [IX]

(3) We now consider the case k > a + !, knot an integer. Then, according
to the previous result,
(9.41.9) .n = 0, 1, 2, ... ,
'vhere k' = [k] + 1, and A is a proper constant inJependent of n.
r
Let(]'> k'. If we write Un(x) = lh~a,p) 1 p~a,/l)(l)P~a,fl)(x), the definition
(9.4.5) furnishes
~ ~

L s~k') (x)rn = (l - r)-k'- 1 L Un(x)rn.

n~ n~

Thus,
~ ~ ~

L s~u)(x)rn = (1 - r)k'-u(l - r)-k'- 1 L Un(x)rn = (1 - r)k'-u L s~k')(x)r\

n~ n~ n~

SO t hat S (u)
n
( X ) = "'n
L.Jm=O
c<u-k'-1)
n-m
s<k')
m
( X ) an d
n n
< IC(u)
I..Jn(u) = l-1 ""c<u-k'-1) c<k') L(k') <A {C(u) ~-1 ""c<u-k'~1) c<k')
n .L...J n-m 1n m = n J .L...J n-m m •
m=O m=O

Consequently,
(9.41.10) (]' ~ k', n = 0, 1, 2, ....
This holds in particular if (]' = k + 1, k + 2, .. · .
(4) Now we prove the identity

r(n t_v +a+ 13 + ~ + 2) c~k) = !'!~~±-~ + 13 + 2k + 3) £ (-] )P

P

r(n + +a + 13 + 2) - r(2n +a + 13 + k + 3)
V p=O
(9.41.11)
k+p) (k+p) k(k- 1) ... (k- p + 1)
· ( p Cn-• (2n + ~-+{3-+ k + 3) · · · (2n +a+ 13 + k-+ -P-+ 2)'
0 ~ v ~ n.
The series on the right converges absolutely if n ~ 1. (For p = 0, the last
fraction is to be replaced by 1.) For the proof we use the following trans-
formation of the left-hand member:

r(2n +a+ 13 + 2k + 3) _ _ 1 t tn-•H(1 _ t) n+•+a+fl+k+1 dt

r(n + v + a + 13 + 2) (n - v)! r(k + 1) )o
r (2n + a + 13 + 2k + 3) 1
r(n + v +a+ 13 + 2) (n- v)!r(f+"l)

•£ (-J)P (k)P )ot

p=O
tn-•+k+p(1 -t) n+•+a+fl+l dt.

Calculation of the last integra.! completes the proof. Term-by-term integra-

tion is permitted, since the terms are, save for a finite number of exceptions,
all of the same sign.
[ 9.41] CONTINUATION; THE LEBESGUE CONSTANTS OF ORDER k 2GI

Consequently, from (9.4.1) we have the representation

p~a+k+l,fJ) (x) = 2a+fJ+ 1r(a + 1)

(9.41.12)
. r(n + {3 + l)r(2n +a+ {3 + 2k +3)
r(n +a+ {3 + k + 2)r(2n +a+ {3 + k + 3) p-0
-1)P t ( (k +
p
p)
k(k - 1) ... (k - p 1)
-;-::::------,--___:_,-~---,-:...:..-~~:..._...,::---__
+ S (k+ P) (x) 1
(2n +a+ {3 + k + 3) · · · (2n +a+ {3 + k + p + 2) n
or
s~k) (x) = { 2a+fJ+l r(a + 1)
1
. r(n + {3 + 1)r(2n + a + {3 + 2k + 3) }- p<a+k+1,fJ) (x)
r(n +a + {3 + k + 2)r(2n +a+ {3 + k + 3) "
(9.41.13) - i: (-1 y (k + p)
p-1 p

k(k - 1) ... (k - P + 1) s<k+p> < )

(2n + a + {3 + k + 3) · · · (2n + a + {3 + k + p + 2) " x ·
On account of (9.41.2), this gives
(9.41.14) L~k) = O(n-k)O(n'x+ 1)0(nk-a-1) + O(n-k)An = 0(1) + O(n-k)A,. 1
where

An =
f (k + P) (2n +a+ 1 k(k - 1) · · · (k - p 1) 1+
{3 + k + 3) · · · (2n +a+ {3 + k + P + 2)
p- 1 p
(9.41.15) 1+1
• _ (1 - x)a(1 + x)fl 1 s~k+p) (x) I dx.
1

Now, according to (9.41.10),

An~ A f (k + P)
p=1 p
k(k - 1) .. · (k - p + 1) I
1 c<k+p)
(2n +a+ {3 + k + 3) · · · (2n +a+ f3 + k + p + 2) n

(9.41.16) = 0(1) ~I(:) I ~ k)

(n

(n + k + 1) · · · (n + k + p)
(2n + a + {3 + k + 3) · · · (2n + a + f3 + k + p + 2)
_ k ~
- O(n) ~
I (k) (a +
p
1 (k + 2) · · · (k + p + 1)
{3 + k + 5) · · · (a+ {3 + k + P + 4)
= O(nk)
'
since (n + k + l)/(2n +a+ {3 + k + l + 2) decreases as n increases provided
l > a + {3 - k + 2. This completes the proof of the first part of the statement
in Theorem 9.1.3.
262 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [ IX ]

(5) Finally, let k = a + ~. Using the previous notatiori, we have

L (J..:) > {2a-r-B-1 r( + 1) + + I)r(2n + a + (3 + 2k + 3) }-1

r(n (3
n a + + k + 2)r(2n + a+ (3 + k + 3)
r(n a+ (3
·{C~J..:)l- 1 1: (1- x)"(l + x)Pip~a+k+LB>(x) ldx- {C~k)l- 1 An
1

from (9.41.13). The first term on the right is~ n"+1n-kn-! log n = log n (ac-
56
cording to the second part of (7.34.1) ); the second term is bounded, according
to the result of (4). That is,
(9.41.17) L~k) > A log n, A > 0, k =a+~'
so that the expansion of a continuous function in Jacobi series is, in general, not
(C, k = a+ !)-summable.

9.42. Proof of Theorem 9.1.4

(I) Let f(x) be continuous at x = 1, let f(l) = 0, and assume k > a + !.
First we discuss the integral
+1
(9.42.1)
1i
_
1
f(x) I (1 - x)"(l + x)P J p~a-':-k+ 1 .,9)(x) I dx
as n - t ex;. Denoting byE an arbitrary positive number, E < 1r /2, we decompose
the interval 0 ~ e ;;£ 1r, X = COS e, into
(9.42.2)
The corresponding integrals I, II, III, can be estimated as follows (cf. (9.41.2),
(7.32.6), and (7.32.7)):

I = max i f(cos e)
o;;;o;:;;.
i 1 1

cos •
(1 - x)"(l + x)P i P~a+k+ 1 .P>(x) j dx

= max If (cos e) I O(nk-a- 1 ),

0~0~<

II = O(n-!) = o(nk-a-1),

1~.1 f(cos e) I (7r - e) 2,9+10(n-!) de = O(n-!) = o(nk-a-1),

1~f i !(cos e) 1 (1r -

2
e) P+lo(np) de
III=
= O(nk-a-
1
) 1~. I!(cos e) I (7r - e)~B+1 de,
i~f lf(cos e) I (7r- e) 2,9+1(7r- e)-P-!o(n-!)de = O(n-!) = o(nk-a-1),

66 If we use only (7.34.2), we obtain L',,k>-+ oo as n-+ oo; this is sufficient for our purpose.
t 9.42 l PROOF OF THEOREM 9.1.4 263

according as - 1 < {:J ;§; - ~, - ~ < {:J ;§; k - a - 1, or {:J > k - a - 1. In

the last case we use the antipole condition of Theorem 9.1.4. Since E is arbi-
trarily small, it appears that in all cases the integral (9.42.1) is o(nk-"- 1 ).
(2) Next we introduce the constants
1

(9.42.3)
M~k) = IC~k)}- 1 ~-~ lf(x) I (I- x)"(l + xll s~k)(x) I dx,
n = 0, 1, 2, ... ,
where ~J(x) has the i:iame meaning as in (9.4.5). We obtain the following
analogue of (9.41.3);
n
M~k) = O(n-k) L I G.(n, k) I o(/k+
v=O
1
).

2 1
The last factor o(v k+ ) must be replaced by 1 for v = 0. Therefore, if k is any
positive integer, we have as in (9.41.8),
k-1 n-1 k-1
M~k) = O(n-k) L nP + O(n-k) L L (n - v)Pv-k-p- 2o(v2k+ 1)
p=O v=1 p=O
(9.42.4)
+ O(n-k) {~ n-k-p-29(n2k+l) + n-k-lo(n2k+l)} = o(I).
The same holds if k = 0, for then G(u) = 0 (see the remark in the footnote of
§9.41 (2)).
(3) For non-integral k we use (9.41.13) again. Let E be an arbitrary positive
number, and let no be so chosen that, fork' = [k] 1, +
(9.42.5) if n ~no.

Then for n ~no, (]' > k' we find, as in §9.41 (3),

n
M n<u) =
< {c<u)
n
l-1 ~
L..J
c<u-k'-1)
n-m
c<k')
m
M<k')
m •
m=O

We decompose the latter sum into the sums L:~~~o 1 and L::.=no. In the sec-
ond part (9.42.5) can be applied so that
no-1 n
M~u) ;§; {c~u) l-1 L c~u::,~'-1) c~~') M~~') + d c~u) l-1 L c~u::,'-1) c;~').
·m=O tn=no

The second term of the right-hand member is less than E. Therefore,

(9.42.6)

where A is a positive constant depending on E and independent of (]'. This

holds in particular if (]' = k 1, k + +
2, . . . . vVe must replace C~u-k'- 1 ) by
c~u.=-::+~) if(]' < k' +
1, which occurs when (]' = k +
1, but not when (]' ~ k 2. +
Consequently, we obtain, as in (9.41.14),
2G4 EXPAI'\SIOK PROBLEMS FOR THE CLASSICAL POLYNOMIALS [IX]

M~k) < O(n -k)O(na+l)o(nk-,a-1) + O(n-k) f:

p=l
(k + P)p

! k (k - 1) · . · (k - P + 1) I
(9.42. 7)
(2n +---;-+ "-+f+-ar-:-:-~2n +- ~ -+ " -tk--+- -p-+ 2)
·\EC~k+p) + AC~k+p-k'-l) }.
(For p = 1 we must replace C~k+p-k'-l> by C~~~;;-~~- 1 > .) The first term of the
right-hand member is o(1). The second term can be decomposed into two parts.
The first part has the form EO(I), according to §9.41 (4). To estimate the
second part we use the fact that

c~k+p-k'-l) = r(n + k + P - k'Lr(_k_± -~± 1) < B (---k + P_)k'+l

--C~k.f.p)-- r(n + k + p + 1 )r(k + p - k') n +k+p
where B > 0 depends only on k. For the second part we therefore obtain the
bound (cf. (9.41.16))

0(1) f, l(k)l------- (n + k + 1) ... (n -t_~±_cl______ -- -

p=l _P (2n +a+ {j + k + 3) · · · (2n +a+ {j k + + +
p 2)
·( k+p
n+k+p
)k'+l
Decomposing this sum into the sums L~=l and L~P+l, where Pis an arbitrary
positive integer, we obtain the bound

0(1) ~
p (
n
k+
+k~ P
)k'+l
+ 0(1) t
p=P+l p
__
l(k)l-----~-±-~_: ~ p +___!1_ ______ .
(a + {j + k + 5) · · · (a + {j + k + p + 4)
The first term tends to zero as n ---+ oo. The second term is arbitrarily small if
P is sufficiently large. Therefore M~k) = o(I) as n ---+ oo.
Concerning the case k = a +
!, compare §9.41 (5).
(4) REMARK. The continuity at x = +
1 can be replaced by the more general
condition

(9.42.8) l" I !(cos e) - J(1) 1 de = o(o), o- +O.

Only the estimation of the integral I (cf. (9.42.2)) mm;t be slightly modified.
We have, by use of (7.32.5), ifj(I) = 0,

+ O(n-i) }n{~
1

I= O(n"+H 1) {n- e2"Hjj(cos e)jde ea-k-i !f(cos e)jde.

}o 1

In both integrals we integrate by parts (cf. Fejer 8, p. 280). Let

[9.42] PROOF OF THEOREM 9.1.4 265

1I 8

/(cos t) I dt = F(O).

Then we :find

I = O(na+k+1)n-2a-1F(n-1) + O(na+k+1) ln-1 ()2"F(O) d()

+ O(n-;) \ Ea-k-iF(E) + nk-a+;F(n- 1

)} + O(n-;) 1~ 1 ()a-k-!F(O) d()
= o(nk-a-1) + o(nk-a-1) + O(n-;) + o(nk-a-1) + O(nk-a-1) max {0-1F(O)}
0<8~·

= o(nk-a-1) + O(nk-a-1) max {()-1F(O)}.

0<9~·

(5) Finally, we show that the assertion of Theorem 9.1.4 does not hold in
general if the "antipole condition" is not satisfied. We consider the function
(cf. §9.3 (4))
(9.42.9) f(x) = (1 + x)'"'.
Its expansion at the point x = 1 is

f
n-o
a)1+1 (1- x)"(1 + x)'"'Hp~a,P>(x)dx
{h~a,,9)}- 1 (n +
n -1
(9.42.10)
= f (-1)n\h~a,,9)}-1 (n +n a)1+1 (1- x)'"'+P(1 + x)"P<.f·">(x)dx.
n-O -1

According to (9.3.11), up to a fixed constant nonzero factor, the principal part

of the general term of (9.42.10) is

(9.42.11)

But (9.42.10) cannot be (C, k)-summable if k ~ a - {3 - 2J.L - 1 = X. Indeed,

0()

(9.42.12) (1 - r)-k- 1 L: (-1r c~>->rn ~ (1 - r)-k- 1(1 + r)->.- 1

n-O

Darboux's method (§8.4) yields for the coefficient of rn in the power series
expansion of this function, the principal term
(9.42.13) AC~k> + B( -1rc~>->,
where A and B are fixed constants, different from zero.
Now let (9.1.9) be satisfied. If we take
(9.42.14) -{3 - 1 < 1J. ~ !(a - {3 - k - 1),
the Jacobi series exists, the integral (9.1.10) is divergent, (since we have
!(a - {3 - k - 1) < -{3/2 - 3/4), and the series (9.42.10) is not (C, k)-sum-
mable.
266 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [IX]

9.5. Proof of Theorems 9.1.5 and 9.1.6

(1) We start from the representation

(9.5.1)

of the nth partial sum of the Laguerre series of the function f(x), where the
"kernel" K~")(x, t) has the meaning in (5.1.11). \Ve write the formula given
there in the more convenient form
(t)
f(a + 1)K~")(x, t) = n +1 L (a) ( )L(a-1) ( )
n+l X n+1 t - n+1
L(a-1) ( )L(a)
X n+l

(nta) x-t

(9.5.2)

This follows from (5.1.11) if for L~"\x) we substitute L~+> 1 (x) - L~~1 )(x),
1

and similarly for L~")(t) (cf. (5.1.13)).

First we assume that the integrals (9.1.11) exist and that the condition (9.1.12)
is satisfied. Then the first integral (9.1.14) exists (cf. the remark in §9.i1 (4)).
Let f(x) be a polynomial p(x); then the statement (9.1.13) is true. Therefore,
according to the closure property pointed out in Theorem 5.7.3, it suffices to
show that the difference in (9.1.13) admits an estimate of the form
1
(9.5.3) 0(1) 1 ta I f(t) I dt + 0(1) 1oo e- 112
t" 12-: 1 I f(t) I dt + o(1), n--+ oo,

where a = min (a, a/2 - 1/4), and the bounds 0(1) and o(1) hold uniformly in
x, E ~ x ~ w. Furthermore, both factors 0(1) are independent of f(x).
(2) Let us consider the contribution to sn(x) of the interval 0 ~ t ~ E/2. In
accordance with the first formula (9.5.2) and (7.6.9), for a ~ ~ this is

(9.5.4) = 0(1) le t" 12 H I f(t) I dt + 0(1) 1· t" 12 -i If(t) I dt

= 0(1) le t" 12 -i IJ(t) J dt.

If - ~ ~ a < ~'we use (7.6.9) and (7.6.10); if a < -~'we use (7.6.10). In the
first case the result (9.5.4) remains valid, while in the second case we obtain
0(1) SO t" I f(t) I at.
[ 9.5] PROOF OF THEORE:\18 9.1.5 AND 9.1.6 2G7

(3) We next consider the contribution of the interval E/2 ~ t ~ 2w, and we
apply (8.8.3) and (8.8.5) to the f3econd formula (9.5.2). Here the variables are
confined to a fixed positive interval; so the remainders in (8.8.3) and (8.8.5)
depend only on n. We find (notation as in (8.8.3))

K (a) ( t) _ 1-a a/2-l (a-1)/2-f k(x)k(t)

n x, - n n n xi + ti

·{ti cos [2 (nx)i + 'Yl sin [2(nt)i + 'Y] - sin [2(nx)i + 'Y)
(9.5.5) ti - xi

_ xi sin [2 (nx)i + 'Yl cos [2(nt)i + ~l =~~s [2(nx)i + 'Yl + O(l)}

The mean-value theorem allowf3 us to replace n' = n + 1 by n. Suppose,

for instance, that cp(m, t) = cos (2mit + 'Y). Then
[cp(n', t) - cp(n, t)] - [cp(n', x) - cp(n, x)] a2 cp
=-
(n'-n)(t-x) am at
taken at a proper place in, l, where fii is between n and n', and lis between x
and t. This readily furnishes

K (a)( t) _ k(x)k(t) if [ ( )i + ]sin [2(nt)i + 'Y]- sin [2(nx)i + 'Y]

n x, - xi + ti x \cos 2 nx 'Y ti - xi

(9.5.6) _ sin [2 (nx)i + 'Y] ~os [2(nt)~_+ 'Yl - cos [2(nx)i + 'Yl + O(l)}
ti - xi

= ~xilk(x)\2Cisin (21:i(ti ~xi)\ +0( 1).

t - X

Thus, according to Riemann's lemma, if E < 1 < w,

(9.5.7)

+ 0(1) 1 t" if(t)! dt+ 0(1) loo e-

1
112
t" 12 -i if(t) \ dt + o(I).

(4) In the interval 2w ;£ t ;£ 3n (n large), we have, according to the first

statement (8.91.2),

(9.5.8)
Consequently, from the first formula (9.5.2) it follows that
2()8 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [ IX ]

(9.5.9)

which is equal to

(9.5.10) 0(1) 1 00

e- 112 ta/ 2-t I f(t) I dt.

In the interval 3n ~ < + oo, Theorem 8.91.2 (x ~ a, >-. - a/2 + 1/12),

tells us that
(9.5.11) e-t/2ta/2+1/121 L~a)(t) I = O(na/2-l).
Therefore, from (5.1.11), on account of (9.1.12), we have

1~ e-Ita f(t)K~a) (x, t) dt

I { I L~a)(t) I + I L~~1(t) I} dt
00

(9.5.12) = O(n1-a)na/2-l1 e- 1ta-1 I f(t)

a..

= O(n1-a)na/2-l na/2-l roo e-1/2 ta/2-13/12 I f(t) I dt = o(1).

}an
(5) If the condition (9.1.12) is replaced by the requirements of (9.1.14), a
treatment of the interval 2w ~ t ;:;:; 3n, the same as that given in (4), applies.
In 3n ;:;:; t < + oo we use Schwarz's inequality:
O(ni-a/2) roo e-t ta-1 I f(t) II L~a)(t) I dt
}3n
(9.5.13)

The last integral is O(na) (see (5.1.1)), and this establishes the statement.
(6) Let
(9.5.14) f(x) = x~~o

(Blumenthal 1, pp. 32-33). We shall show that for proper values of IJ. the
first integral in (9.1.11) exists, but that the second does not; furthermore the
Laguerre series is divergent for x > 0. Here a > - !.
The coefficient a,. of the corresponding Laguerre series is given by

r(a + l)(n ~ a)a.. loo e-:r:xa+~~oL~a\x)dx

=
(9.5.15)
= .!_
n!
1
0
00

at(!!)"{e-:r:xn+a}dx
dx
[ 9.51 PROOF OF THEOREMS 9.1.5 AND 9.1.6 2()9

(see (5.1.5)). On integrating by parts, we have

(9.5.16) r(.u + a + 1) r(n - JJ.)

all = r( -IJ.) r(n + a + 1)"
Here we assume 1J. + a > -1, 1J. ~ 0, 1, 2, · ·. . Then an""' n-,.-a-\ and in
view of Fejer's formula (8.22.1), the principal term of the Laguerre series
behaves like
(9.5.17) n-p.-a/2- 614 cos {2(nx) 1 - a1r/2 - 7r/4}
for x > 0. This series is therefore divergent (cf. Problem 47) when and only
when 1J. + a/2 + -i ~ !. If a > -t and
(9.5.18) -a - 1 < 1J. ~ -a/2 - 3/4,
the first integral (9.1.11) exists, but the second does not, and the series 1s
divergent for x > 0.
(7) We consider also the function
(9.5.19) f(x) = e:z 12x"',
and we show that for a proper value of 1J. (particularly for IJ. = - a/2 + 1/4) the
integrals (9.1.11) exist, but the conditions (9.1.12) and (9.1.14) are not satis-
fied; in addition, the Laguerre series is divergent for x > 0. Here a > -1,
a+J.L> -1.
The integral

(9.5.20) r(a + 1)(n ~a) an= la> e-x12 xa+,.L~a>(x)dx

can be calculated by means of the generating function (see (5.1.9))

n-o
:t r(a + 1) (n +n a) anrn = (1 - r)-a- 1 {"" e-:z<Hr/( 1-r)) xa+~~o dx
)o
(9.5.21) = r(a + IJ. + 1)(1 - r)-a-1 (!2 + 1-
_r_)-a-p.-1
r
= r(a + IJ. + 1)2a+p.+l(1 - r)"'(1 + r)-a-p.-1.
Thus, Darboux's method (§8.4) gives for an the principal term

(9.5.22) n-too,
where A and B are fixed constants, A ~ 0, B ~ 0. We have A = 0 if
p. = 0 1 2, ....
If x '> '0, this shows, on account of Fejer's formula (8.22.1), that anL~a>(x)
does not tend to zero if p. = -a/2 + 1/4.
(8) Finally, we discuss the Hermite series
(9.5.23)
Writing y = x 2 , we readily see that (see (5.6.1))
00 00

L
m=O
~mH2m(x) and L ~m+1H2m+1(x)
m-o
 270 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [ IX ]

are the expansions of

{f(x) + f( -x) l /2 = {f(yi) + f( -yi) J/2 and
1
X- 1f(x) - f( -x) J /2 = y-i lf(yi) - f(- yt)} /2

into a Laguerre series with the parameters a = -!and a = +!,respectively.

Applying Theorem 9.1.5 to these functions with a = -! and a = +!, respec-
tively, we obtain the conditions (9.1.11), (9.1.12), and (9.1.14) in the fol-
lowing form:
1 1
1 y-i lf(±yi) I dy = 21 if(±x) I dx exists;

1 00
12
e-" Y-! I f(±yi) I dy = o(n-i), or 1
00

e-x 212 x-& I f(±x) j dx = o(n- 1);

too e-"/2y-1 I f(±yi) I dy = 21oo e-z2f2x-1, I f(±x) I dx exists;

2
loo e- 11 y-1 I f(±yi) 12 dy = o(n-1), or 1oo e-:r x-4 1 f(±x) j2 dx = o(n-3).
This establishes (9.1.17), provided x belongs to an interval not containing
the origin.
In order to accomplish the proof, if x lies in an interval of the form l-E,
+ E], we have only to show that

(9.5.24) e-t2 f
•-0
(2.v!7ri)-1H.(x)H.(t)- 7r-1sin {(2n)i(x- t)} = 0(1)
X - t

uniformly if both x and t belong to [ -E, +E]. The first member of this dif-
ference is, according to (5.5.9),

(9 .5 .25 ) (2n+In!7ri)-1e-t2{Hn+ 1(x) Hn(t!_::_ Hn(x! _ Hn(x) Hn+1(t) - Hn+1(x)}·

x-t x-t
Now, by using the the notation of Theorem 8.22.6, we obtain
-,...-;, 1e-:r 212 H n(x) = cos (NiX - n7r /2) + O(n-!)'
A-;; 1 e:._~:r~~~ Hn(x) - e-~~~~HnCD = ~os (Nix-:-. n1r/2)- cos (Nit- n1rj2) + 0(1).
x-t x-t
The second formula follows by an argument similar to that used in §8.8 (cf.
the first formula in (5.5.10)). Its left member can also be written in the form

>.:;,1e-o:2;2 Hn(x) - Hn(t) + O(l).

X- t

Replacing Ni by (N + c)i, c a fixed constant, we find the error committed in

the right-hand member to be 0(1).
These asymptotic formulas furnish for (9.5.25), by use of Stirling's formula,
!9.6] PROOF OF THEOitE\I !U.7 271

(2n+ln! 11"~)- 1 AnAn+lex 2- 12 {sin (Nix - n71"/2)

cos (Nit- n71"/2) - cos (Nix - n71"/2)

X - t

_ cos (N~x _ n71"/ 2) ~~n (Nit - n71"/2) - sin (Nix - n11"/2! + 0( 1)}
x-t
_ -1 sin !Ni(x - t)} + 0( 1) _ --1 sin I (2n)i(x - t) l ()
- 71" X - t - 71" ---- X - t +0 1.

9.6. Proof of Theorem 9.1.7

(1) The Cesaro means of the Laguerre series at x = 0 can be represented in a
particularly simple form. For, from (5.1.7),

(9.6.1)

Hence, by applying (5.1.9), the Cesaro means of order k are foun~ to be

(9.6.2) IC~k)r(a + 1) }-1 1oo e-ttaf(t)L~a+k+l)(t) dt. , ..

Assume k > a + !, and subdivide the interval [0, + oo] into [0, E], [E, w],
and [w, + oo ]. Then we find for (9.6.2)

0( n -k) max I f(t)

O;:ijt;:ije
I 1e
0
ta I L~a+k+l) (t) I dt

(9.6.3) + O(n-k) j"' e-tta I f(t) II L~a+k+1 )(t) I dt

+ O(n-k) 1 00

e-tta I f(t) II L~a+k+l) (t) I dt.

By use of (7 .6.8) the first integral becomes

(9.6.4)

The second integral is O(n<a+k+l)t 2- 1) = o(nk). Now, using Theorem 8.91.2,

with a replaced by a + k + 1, and A = k + !, we see that A - ! =
k > (a+ k + 1)/2 - i; whence the third integral is

(9.6.5) 0(1) 1 00

e-l/ ta I f(t)
2
I ck-} nk dt = O(nk) 1 00

e-tt 2 ta-k-A I f(t) I dt.

Thus (9.6.2) can be represented in the form

(9.6.6) 0(1) max I f(t)

O;:ijl;:ije
I + 0(1) 1"" e-
"'
112
ta-k-} I f(t) I dt + o(1).
Here the bounds of the terms 0(1) are independent of E and w. If it is assumed
that f(O) = 0, the s~atement is established.
272 EXPANSION PROBLEMS FOR THE CLASSICAL POLYNOMIALS [ IX ]

(2) Let k = a + t, and apply Theorem 1.6 (Helly'~-:> theorem) to the linear
operations

= I C~k> r(a r- 1e- 1 t"j(t)L~a+k+ 1 > (t) dt,

1
Un(f) + 1) 1

(9.6.7)
U(f) = j(O).
It suffices to show that the second condition in (1.6.10) is not satisfied. As a
matter of fact, if this condition is not satisfied, then a continuous function
f(x), 0 ~ x ~ 1, exists for which
lim Un(f) = U(f)

does not hold. Extending the definition of this function to .r > 1 by means
of the condition j(x) = 0, we obtain the "Gegenbeispiel" required. (By use
of the remark made in connection with Theorem 1.6, a continuous function
f(x) can be constructed for which (9.6.2) is unbounded when n--+ oo .)
Now if n is a positive constant independent of n,

(9.6.8)
{C~k>r(a + 1)}-1 1 1
e-tta I L~a+k+1 >(t) I dt

where A is positive and independent of n and n. The last expression is, ac-
cording to (8.1.8),

r-.J jo zaz-<a+k+1' 12
1 Ja+k+1(2i) 1 dz = lo z-t I Ja+k+1(2z~ I dz.
This integral becomes arbitrarily large with n.
(3) Integral (9.1.19) exists if
(9.6.9) f(x) = O(elt;12 xk-a-f-a), 0 > 0, X --+ + oo.
On the other hand, there is no difficulty in proving that the Laguerre series
L::-o anL~a>(x) of f(x) = elt;12xk-a is not (C, k)-summable at x = 0. Here the
condition k > a + ! is satisfied. Indeed, from (9.5.21) we obtain, p. = k - a,

(9.6.10) (1 - r)-k-1 ~ anL~a)(O)rn = ~~~! g 2k+1(1 - r)-a-1(1 + r)-k-1.

Darboux's method furnishes for the coefficient of rn in the expansion of this
function the number
(9.6.11) C > 0, n--+ oo.
This establishes the statement.
We also notice that the Laguerre series is (C, k)-summable at x = 0 with
an arbitrary k > a + 1/2, provided j(x) is continuous at x = 0, and
(9.6.12) j(x) = O(elt;12 x- 116 ), x--+ + oo.
Again, the Laguerre series of the special function f(x) = elt;12i is not (C, k)-
summable at x = 0 with any k > a + 1/2.
[9.7] FURTHER RESULTS 27.3

9.7. Further results

(1) Using a generalized translation operator which can be defined by means

of (4.10.21) it is possible to use Theorem 9.1.3 to show that the Jacobi series
of a continuous function is uniformly ( C, k) summable fork> max( a+ t, {3 +f)
when a,{3 ~ - t, or a +!3 ~ -1, and a,{3 > -1. See Askey-Wainger 4 and
Gasper 3, 4.
(2) For a = {3 ~ - t Kogbetliantz 19 proved the positivity of the ( C, 2a + 2)
means of the formal reproducing kernel. A simple proof was given in Askey-
Pollard I. For the general Jacobi case it is likely that the ( C, a + {3 + 2) means
are positive when a,{3 ~- t. This has been proven for max({3 -1, -{3) ~
a~ {3 + 1 and max({3- 2, 3- {3) ~a~ {3 +2. See Askey 8.
(3) Theorem 9.1.2 can be used with the Carleson-Hunt theorem (Carleson
1, Hunt 1) to obtain almost everywhere convergence of Jacobi series when
the function satisfies the conditions of both theorems.
( 4) Slightly stronger equiconvergence theorems for Laguerre and Hermite
series were obtained by Muckenhoupt 4. Again almost everywhere convergence
theorems follow from these results. These improvements and the mean con-
vergence results to be given below use Erdelyi's relatively recent asymptotic
formulas in an essential way. These estimates are described in 8.92 (2).
( 5) If f_\ I/( x) I Pdx < oo, if f( x) is expanded in a series of Legendre poly-
nomials, and if Sn(X) denotes the partial sums of this series, then Pollard I
proved that

for 4/3 < p < 4 and these bounds are best possible (Newman-Rudin 1). This
theorem has been extended to Jacobi series (Pollard 2, 3, M uckenhoupt 5),
and Laguerre and Hermite series (Askey-W ainger 1, M uckenhoupt 2, 3). A
slight simplification in the proof was given in Askey 6, and this result and
the positivity of some Cesaro mean was used in this paper to solve a problem
of LP convergence of Lagrange interpolation at the zeros of Jacobi polynomials.
( 6) There are a number of other interesting positive summability methods,
e.g. the analogues of Poisson-Abel summability (Bailey 1); Gauss-Weierstrass
summability (Bochner 2, Karlin-McGregor 3); and the de Ia Vallee-Poussin
method which uses Bateman's formula (4.10.23). Horton 1 used this formula
to prove that the de la Vallee-Poussin method is positive and variation
diminishing.
(7) Muckenhoupt-Stein 1 have introduced a number of the deeper functionals
from classical Fourier series into the study of orthogonal polynomials and
were able to construct a theory of H P spaces associated with singular Cauchy-
Riemann equations and prove an analogue of the Marcinkiewicz multiplier
theorem. Another proof of this theorem by mapping the theorem for Fourier
series to Jacobi series was given by Askey-W ainger 2 and Askey 3. Another
extension of the Marcinkiewicz theorem was proved by Bonami-Clerc 1. They
were motivated by work of Coifman-G. Weiss 1.
( 8) Hirschman 1 has constructed a theory of variation diminishing trans-
formations associated with orthogonal polynomial expansions which paraillels
Schoenberg's classical theory.
 CHAPTER X

REPRESENTATION OF POSITIVE FUNCTIONS

In the present chapter we deal with an extension of Fejer's representation of
non-negative trigonometric polynomials, given in §1.2, to certain general classes
of non-negative functions. In particular, we are interested in the discussion
of this representation if the given function is subjected to certain continuity
conditions. Extensions of Fejer's theorem in this direction are important for
the investigation of the asymptotic behavior of the general orthogonal poly-
nomials associated with a distribution on a finite real interval or on the unit
circle (Chapter XI). It seemed convenient to separate these considerations
from the subject proper.
Concerning the results of this chapter see Szego 6, 7, 8, 9. See also Grenan-
der-Szego 1, 1.12-1.1.3.

10.1. Fatou's theorems

THEoREM 10.1.1. Let f(O) be integrable in Lebesgue's sense, and let

2- L+"' f(t) 1
2
1 1 - r
(10.1.1) f(r, 0) = 1r _,.. -
2r cos (t - 8) + r 2 dt
be the corresponding Poisson integral. Then we have almost everywhere m
-7r ~ () ~ +7r,
(10.1.2) lim f(r, 0) = f(O).
r-1-0

See Zygmund 2, p. 54, §3.442.

THEOREM 10.1.2. Let
(10.1.3) F(z) = Co + C1Z + C2i + · · · + CnZn + · · ·
be regular for I z I < 1, and let the integral

(10.1.4) ;7r i:"' I F(re' ) 1

8 2
dO

be bounded for r < 1. (This condition is equivalent to the convergence of

2
(10.1.5) I Co 1
2
+ I C1 1
2
+ I C2 1
2
+ " •+ I Cn 1 + •" .)
Then we have almost ev'erywhere in - 1r ~ 0 ~ + 1r

18
(10.1.6) lim F(re ) = F(e 18).
r-1-0

Furthermore, F(/ is measurable, and I F(e 18) 12 is integra_ble in Lebesgue's sense.

8
)

The Fourier series of F(ei8 ) is obtained by writing z = e'8 in (10.1.3).

274
[ 10.2) GENERALIZATIO!\ OF FEJER'S REPRESENTATIOK 275

This theorem is due to Fatou (1). vVe say that F(z) is of the class H 2 • A
function F(z) which is regular and bounded in I z I < 1 belongs to this class.
Concerning the more general classes H&, see F. Riesz 2 and Smirnoff 2. In
particular, if F(z) is of the class H1 , the boundary values F(ei8) exist (almost
everywhere), I F(ei ) I is integrable in Lebesgue's sense and Cauchy's theorem
8

can be applied on I z I = 1. (Cf. F. Riesz, loc. cit., p. 94, c); Smirnoff, loc. cit.,
pp. 337-338.)

10.2. Generalization of Fejer's representation

Concerning this section, see Szego 7.
(1) Let g(O) be a non-negative trigonometric polynomial not vanishing
identically. According to Theorem 1.2.2, there exists a polynomial h(z) of the
same degree, uniquely determined by the following conditions:
2
(a) g(O) = lh(z) 1 , wherez = / 8 ;
(10.2.1) (b) h(z) is different from zero in I z I < 1;
(c) h(O) is real and positive.
We obviously have

(10.2.2) log g(O) = 2m{log h(z) l,

The function log h(z) is regular for I z I ;;;; 1 except at the points z = ei8 which
correspond to the zeros of g(O), at which both functions log g(O) and log h(z)
become logarithmically infinite; log h(O) is real. The function 2m (log h(z)}
is regular and harmonic for I z I < 1 and has absolutely integrable boundary
values log g(O). Applying the mean-value theorem of Gauss, we obtain

(10.2.3)
2
1+"'
111' _,.. logg(O)dO = 2m{logh(O)} = 2logh(O),

so that

(10.2.4)

The last expression is called the geometr'ic mean of. the function g( 0).
(2) The relation (10.2.2) enables us to extend this consideration to arbitrary
non-negative functionsj(O) (instead of g(O)), defined in [ -11', +11'] and integrable
in Lebesgue's sense, provided ®(f) > 0. This last condition is equivalent to
the existence of the integral

(10.2.5)
1 +,.. log j(O)
_,..
dO = (
}o~/(8)~1
log j(O) dO + (
}!(8)>1
log j(O) dO.

The existence of the second integral on the right follows from the integrability
of j(O). Thus, the condition ®(f) > 0 is equivalent to the existence of the
first integral on the right-hand side. This is a restriction on the "nearness"
276 REPRESENTATION OF POSITIVE FUNCTIONS [X]

of f(O) to 0. A consequence of this condition is that j(O) can vanish only on a

set of measure zero.
We now introduce the harmonic function u(rei8 ) defined by.: Poisson's integral
of! logj(O)
+,.. 1 - r2
(10.2.6) u(re )
iB
=
1
1
411' _,.. log j(t) 1 - 2r cos (0 - t) + r2 dt, O~r<l.

In the case of a continuous function j(O) > 0_, we know that (see for instance
Zygmund 2, p. 51)
(10.2.7) lim u(rei8 ) = ! log j(O),
r-1-0

uniformly in 0. In the general case considered above, this is true only with
the exception of a set of measure zero and without uniformity in general (see
Theorem 10.1.2). If now u is completed to an analytic function u + iv = k(z)
with the condition that k(O) is real, then k(zj is uniquely determined. Writing
D(z) = l<•>, we obtain the analogue (generalization) of the function h(z) con-
sidered before. This function D(z) = D(f; z) has the following properties
(Szego 7, p. 237):
(a') D(z) is of the elass H2 (§10.1); almost everywhere in -11' ~ 0 ~ +11',
(10.2.8) lim D(t/8) = D(ei8) exists, and j(O) = I D(ei8 ) 2
1 ;
r-1-o

(b') D(z) ~ 0 in I z I < 1;

(c') D(O) is real and positive.
We have again (D(0)} 2 =®(f); this is obvious from (10.2.6). Furthermore,
we show that for an arbitrary continuous function F(O) of period 211',

(10.2.9)

If D(z) = do + d1z + d2i + . · . , according to Schwarz's inequality, we have

{J_:"' II D(ei 8 2
) 1 - I D(rei
8 2
) 1 1 dO r
~ L:" (I D(ei
8
) I - I D(rei8) 1) 2 dO· L:" (I D(e
18
) I + ID(rei8) 1) 2 dO

for r-? 1 - 0. Here (a') is used. (See also Smirnoff 2, p. 338.)

·-----..,...---------..,...--· -----·--······
 [ 10.3 I FURTHER STUDY OF POSITIVE FUNCTIONS 277

It should be observed that the function D(z) is not uniquely determined by

the conditions (a'), (b'), and (c'). For instance, we can multiply it by
exp (-(1 + z)/(1 - z)}. For this point see Szego 7, p. 241.
(3) By means of (10.2.6) we can obtain the following explicit representation
of D(z) in terms of j(O):

(10.2.10) D(z) = D(J; •~) = exp { -1 1+"' log j(t) 1 + ze -it l

. dtf, lz I< 1.
41r -... 1 - ze-4t
If j(O) is an even function:, the expansion of D(z) around z = 0 has real coeffi-
cients. Let JI(O) and !2(8) be arbitrary functions satisfying the same condi-
tions as j(O), and let p be a.rbitrary and complex, p ~ 0. Then
(10.2.11) D(JI; z)D(h; z) = D(Jd2; z); (D(J; z) }P = D(f'; z).
As an example we mentilon the casej(O) = g(O) considered in (1). We have
(10.2.12) D(g; z) = h(z), D(g-1 ; z) = {h(z) }-1•
In the second formula we assume that g(O) is positive.
A further example is
f(O) = 2'Y+a(1 -· cos o)'Y(1 + cos o)'; D(f; z) = (1 - z)'Y(1 + z)';
(10.2.13)
'Y > -!, 0 > -!.
10.3. Further studly of the representation of positive functions
The derivation of an asymptotic formula for the polynomials orthogonal on
the unit circle requires some further properties of the representation defined
before. In particular, we shall deal with the behavior of the function D(z) on
the unit circle I z I = 1. In this connection certain restrictive conditions on the
function f(O) are necessary.
(1) First let us consider again the case of a non-negative trigonometric
polynomial g(O) not identically zero, and let g(O) = I h(z) 12, z = ei8, be the
normalized representation defined in Theorem 1.2.2. Then, by (10.2.10), we
have for 0 ~ r < 1,

(10.3.1)
iB { 1 1+"'
1 + reiCB-t> }
D(g; z) = h(z) = h(re ) = exp 7r · _,.. log g(t) _ reiCB-e> dt ·
4 1
Whence
sgn h(rei8)

(10.3.2) = I h(re
iB) ~-1 (
h 1·e
iB) [i 1+"' ()1- 2r2r cos
= exp'\47r _., log g t
sin (0 - t) d}
(0- t) + r2 t

=
fi 1+" 2r sin (0 - t)
exp '\47r _,.. [log g(t) - log g(O)] 1 - 2r cos (0 - t) + r2 dt} .
Therefore, for all values of 0 with the exception of the zeros of g(O),

-------....,...------------
278 REPRESEXTATIOX OF POSITIVE FUXCTIOXS [X)

(10.3.3) =
4
1+""
exp { i7r _,.. log g(t) cot 0
2
t dt}

= exp { .7r 1+""

2
_,.. [log g(t) - log g(O)] cot 0 t dt} .
4 2
The first integral is taken in the sense of Cauchy's principal value; the second
integral is absolutely convergent.
(2) This consideration can easily be extended to a positive continuous func-
tion j(O) which satisftes certain conditions sufficient for the existence of the
integrals corresponding to those in (10.3.3). Since j(O) is continuous, defining
D(z) as in §10.2 we now have, uniformly for all values of 0,
2
(10.3.4) lim I D(rei8) 1 = j(O).
r-+1-0

If the integral

(10.3.5) )__~+,..,. Ilog j(t) - log j(O) I cot I 0

2
t Idt
exists, then
t
1
+,.. (} -
(10.3.6) _,.. log j(t) cot - - dt
2
exists in the sense of Cauchy's principal value. We then show the existence
of the boundary values of

sgn D(re i8) exp { - i 1+"' logf()t 2r sin (O

(0 - t)
t) dt}
=
41r _,.. 1 - 2r cos - +r 2

(10.3.7) }
£ 1+"' 2r sin (0 - t)
= exp { 47r _,.. [logj(t) - logj(O)] 1 - 2r cos (0 - t) + r2dt .

Indeed, if Eisa fixed positive number, we have

.1
hm
r-+1-0 18-t 1>e
[logj~t) -- logj(O)]
1
_
2r sin (0 - t)
2r cos
(O _ t) +r 2 dt

0
= ( [log j(t) - log j(O)] cot t dt.
}18-tl>e 2
On the other hand, 1 - 2r cos (0 - t) + r2 > 2r (1 - cos (0 - t)}; so

r [log j(t) -
I}18-tl~e log· j(O)] 2r sin (0 - t)
1 - 2r cos (0 - t) + r2
dt I
Isin (0 - t) I
~ 1 18-tl~e llogj(t) - logj(O) 11 - cos (0 - t) dt,

and the last integral is arbitrarily small with E.

 [ 10.3 1 FURTHER STUDY OF POSITIVE FUNCTIONS 279

Therefore, under the condition mentioned,

(10.3.8) lim D(rei8) = D(e;8)
r-1-0

exists, and a representation analogous to (10.3.3) holds for sgn D(e;8),

(10.3.9)
sgn D(e;
8
) = exp { i7r
4
1:. logj(t) cot 0
2
t dt}

{ . 1+...
2
= exp 411" -..- [log j(t) - log j(O)] cot 0 t dt} .
2
8
In this case we have j(O) = D(e; ) • 2
I 1

The condition (10.3.5) is satisfied if the function j(O) fulfills the Lipschitz-
Dini condition
(10.3.10) I f(O + o) - f(O) I< L I log ol-1- \
L and X being fixed positive constants. This is assumed in the present section.
Then (10.·3.8) exists uniformly in 0, and D(e;8) is continuous.
The preceding exposition could be considerably simplified by using the
theory of conjugate functions (see Zygmund 2, pp. 50, 54, 55, §§3.321 and 3.45).
(3) Let m be a positive integer. Then there exists a positive trigonometric
polynomial g(O) of degree m such that
(10.3.11) lf(O)- (g(0)}- 1 I < P(log m)_1_ \
where Pis a constant, depending on the minimum and maximum of j(O) and
on L and X. This follows by applying Theorem 1.3.2 to the function (j(O) l-\
which satisfies the Lipschitz-Dini condition
I (j(O + o)}- 1
-- (j(0)}- 1 1 < (minj(0)}-2L I log 0 l-1->-.
If D(z) and h(z) denote the functions defined in §10.2 which correspond, respec-
tively, to j(O) and g(O), we can show that
(10.3.12) I D(z) - ( h(z)} - 1 I < Q(log mr·>-
uniformly for I z I ~ 1. The constant Q depends on the minimum and maxi-
mum of j(O) as well as on L and X.
It suffices to prove this for I z I = 1. An analogous inequality for the differ-
ence of I D(z) I and I h(z) ~-- 1 is trivial (even with (log m)- 1->-). We therefore
need a bound only for
sgn h(e;8) (sgn D(e;8) - sgn [h(e;8)r1 }
(10.3.13)
= exp {4. 1+"" (log j(t) -
2
11" -..- log [g(t) ]-1 } cot 0
2
t dt} - 1;

or, what amounts to the same thing, for

-·----·---r--,----------
 280 REPRESENTATION OF POSITIVE FUNCTIONS [X]

1
~ 0-t
(10.3.14) -r (logj(t) - log [g(t)r1 } cot - - dt.
2
To this end, we use Theorem 1.22.1. We have
(10.3.15) I !7(0 + o) - g(O) I ~ om(max g(O) l,
so that

(10.3.16) I log g(O + o) - log g(O) I~ om m~x gt~.

mmg 0
Let E = E(O, m, X) denote the set of t-values defined by the condition
I 0 - t I ~ m- 1 (log m)->-, and let E' be the complementary set with respect to
[- 1r, +1r]. On writing (10.3.14) in the form

(10.3.17)
1 (logf(t) - logj(O) -log [g(t)r 1 +log [g(0)]- 1 } cot 8
2
t dt

( 1 0-t
+ }s' (logj(t) - log [g(t)J- l cot - - dt,
2
and using (10.3:10) and (10.3.16), we obtain for the first integral

0(1) Lllog I0 - t 11-l->. I0 - 1

t l- dt + O(m) 110 - t 11 cot
0 tjdt
2
= O[(log m)->-] + O(m)O[m-1 (log m)->-] = O[(log m)->-].
On the other hand, (10.3.11) yields the bound

O[(Iog m)- ->.]

1
1, I cot 0
2
t I dt == O[(Iog m)->.]

for the second integral. This establishes the statement.

10.4. "Local" properties of the representation of positive functions

In this section we prove certain theorems on the representation and approxima-
tion of positive functions, important for the purposes of Chapters XII and XIII.
(1) We have the following theorem:

THEOREM 10.4.1. Let j(O) be integrable in Riemann's sense, and let it have
the form

f( e) = •i>( e) I (z - z1)"' (z - z2)" 2 (z - z1)" 1 I,

i8
(10.4.1) • • • z = e ,

where 0 < A ~ 4>(0) ~ B, and Zv = e;e, are distinct points on the unit circle,
(fv> 0, v = 1, 2, · · · , l. Let j(O) be differentiable at the fixed point 0 = a,
a = eia =/= z., v = 1, 2, ... , l, and let the following be bounded near 0 = a:

j(O) - j(a) - f' (a) (0 - a)

(10.4.2)
(0 - a) 2

--·-----,- ------.,..----------,...---·· --------------·-

[ 10.4 J LOCAL PROPERTIES OF THE REPRESENTATIOX 281

lj D(J; z) denotes the analytic junction corresponding to j(O) in the sense of

§10.2, there follows the exi~:tence of the limits
lim D(J; re.:a) = D(J; e.:a) = D(J; a),
r-1-0
(10.4.3)
lim D'(J; re.:a) = D'(J; e'a) = D'(J; a).
r-1-0

This is true under more general conditions (see Zygmund 2, pp. 52, 53, §3.44,
and pp. 62, 63, example 1.3). The following elementary argument is based on
the formula (10.2.10).
If we integrate with respect to t only over a fixed arc not containing a, the
corresponding limits obviously exist. If t is near a, we can write
(10.4.4) log j(t) == c + d(e-.:e - e-ia) + 0(1)(t - a) 2,
where c and d are certain constants. But if I z I< 1, then

(10.4.5) -1
2 1r
1+"' {c + d(e-it- e-.:j} 11 +- zeze--ot-it. dt
_,..
= c - de-.:a.

Hence it suffices to show that if E > 0 is small enough, the function

(10.4.6) [+• 0(1)(t - a) 21 + ze-i.e dt

J-• 1 - ze--ot '
as well as the derivative of this function with respect to z, is arbitrarily small
if z = re.:a, r --+ 1 - 0. This is true for the derivative since
+• (t - aY dt 1+• (t - a)
2

1-• 11 - re•<a-t) 1
.-::---. 2
=
1 - 2r cos (a --• t) + r dt
2

1 1+• (t -
(10.4.7)
<- )2 dt. a
4r -• . a- t
sin2 - -
2
The argument is even simpler for the function itself, for in the last denominator
we then have I sin {(a - l0/2} I instead of sin2 {(a - t)/2}.
(2) Now we prove the following theorem:
THEOREM 10.4.2. Let j(O) be £ntegrable in R£emann's sense, and assume
0 < A ~ j(O) ~ B 2j -1r ~ 0 ~ +1r. Also suppose j(O) differentiable at the
fixed po£nt () = a and the expression (10.4.2) bounded near 0 = a.
If E is an arbitrary pos£ti've number, there exist pos£tive trigonometric polynom£als
g1(0) and g2(8) such that
(10.4.8) J1(8) ~ j(O) ~ J2(8), fi(a) = h(a),
where !I (0) = ( g1 (0) }-\ !::( 0) = ( g2( 0)} -\ and

(10.4.9)
 282 REPRESENTATION OF POSITIVE FUNCTIOKS [X]

Let m and M be the lower and upper bound of (10.4.2) for -1r ~ 8 :::; +1r.
The function f1 ( 8) is greater than a positive number depending only on m, A,
f(a), and f'(a); similarly h(O) is less than a number depending only on M, f(a),
and f'(a).
We observe that (10.4.8) implies
(10.4.10)
so that the integral in (10.4.9) exists; furthermore, (10.4.9) implies

(10.4.1l) J_:"logf2:8) ; ~o!f1Ce) de< E',

2
Sin~

where E' is arbitrarily small with E.

For the proof we apply Theorem 1.5.4 to the functions

(e) = {f(e:)} - l - c - d sin (8 - a)

P . 8-a '
S1n2 - -
2
(10.4.12) d = -f'(a){f(a)}-2,
q(O) = f(O) - f(a) - f'(a) sin(O - a),
. 8-a
S1n 2 - -
2

which are both integrable in Riemann's sense. Therefore, given o> 0, there
exist certain trigonometric polynomials P(O) and Q(O) such that
+...

(10.4.13)
p(O) ~ P(O),
1-... {P(O) - p(O)} dO
+...
< o,
q(e) ~ Q(e),
1-..- {Q(O) - q(O)} dO < o.
Here max P(O) and max Q(O) are less than certain constants depending on
m, A, f(a), f'(a), and on M and f'(a), respectively. Writing

8
(10.4.14) g1(B) = c + d sin (8 - a) + P(e) sin 2
2
a,

1
we find that (!(8) }- ~ g1(8), or f1(8} ~ 'f(O), f1Ca) = f(a), and that

(10.4.15) L:" f(~)2 ; ~ (:) f-:"

sm -
1
de= f(O)f1(e) 1
sm -
8 -
1
g (e~ ~ {~(e~}- de~ B o. 2

2 2
Heref1(8) is greater than a positive constant depending on m, A,j(a), andf'(a).
On the other hand, considering the continuous function
[ 10.4] LOCAL PROPERTIES OF THE REPRESENTATION 283

G(o)

~ ~(a)+ /'(a) sin (8 - a) + Q(O) sin' ~r- c- d sin (8- a)

(10.4.16) . 20-a
sm - -
2
1
{R(O)} - - c - d sin (fJ - a) R- 1 - c - d sin (0 - a)
= > '
. 20-a . 20-a
sm. - - s1n - -
2 2
we have f(O) ~ R(O) < .R, where R is a positive constant depending on M,
f(a), and f'(a). Now according to Theorem 1.3.1, a trigonometric polynomial
S(O) can be determined sueh that
1
(10.4.17) R- - c - d sin (O - a) < S(O) < G(O) < S(O) + o.
. 20-a
sm - -
2
If we write

(10.4.18) g2(0) = c + d sin (0 - a) + S(O) sin2 0 2

a,

we find that

(10.4.19)
1
Furthermore, g2(0) < (j(O) l- • On account of the last inequality in (10.4.19)
and (10.4.13) we have

r j,(O; 8-
sm - -
!~dO~ r~_o + 0 SI~O~
2
r () .
sm -
2 0 -

2
-
a}- 1

-
( )

f 8 d8

1
+ .. {g2(0) r· 1
- {g2(0) + 0 sin2 0 2 a }-
+ 1 -.- • 2 0 - a
sm - -
2
dO

(10.4.20)
+. R(O) - f(O) 1+ . {g2(0)} -1 - {g2(0) + osin2 0 2 a} -1
<
1-.- 0 -- a
• 2
sm - -
·
2
dO + -.- 0 - a
.
sm - -
2
2
dO

·------r---------.....--· ____________..._
 284 REPRESENTATION OF POSITIVE FUNCTIONS [X)

The last integral is arbitrarily small with o (since g2 (8) > R- 1).
Combining (10.4.15) and (10.4.20), and taking o sufficiently small, we ob-
tain the statement above.

(3) THEOREM For the analytic functions D(f1 ; z), D(f; z), D(j2 ; z),
10.4.3:.
corresponding to the functions f1(8), f(8), h(8) of Theorem 10.4.2 in the sense of
§10.2, the following inequalities hold:
(10.4.21) I D(f; a) - D(f. ; a) I < E', I D'(f; a) - D'(f.; a) I < E',
where a = eia, v = 1, 2; here 1
E is arbitrarily small with E.

The symbols D(J; a), D'(J; a) have the same meaning as in (10.4.3).
According to (10.2.10) we have for I z I < 1, v = 1, 2,

1 1+r
log D(f; z) -· log D(f.; z) = 71' _.., {log f(t) - log f.(t) l _ ze-it dt,
1 + ze-it
4 1
(10.4.22) '
D'(f; z) D'(f.: z) 1 (+... -it
z D(f; z) - .z D(f.; z) = 271' )_... {log f(t) - log f.(t)} (1 ~ ze-it)2 dt,
and for z = reia = ra, r ~ 1 - 0,

log D(f; a) - log D(f.; a) =

1+...
i71' _.., {log f(t) - log f.(t)) cot a
- -- t dt,
4 2
(1 0.4.23) a D' (j; a) _ a D' (f.; a) = _ _!_1+ . . log f(t) - log f.(t) dt.
D(f; a) D(f.; a) 811' -r . t - a
sm2 - -
2
Both integrals are absolutely convergent and arbitrarily small with E (see
(10.4.11)); D(J; a) is a. fixed number different from zero.
(4) THEOREM 10.4.4. Letf(8) be afunction satisfying the conditions of Theorem
10.4.1. Given an arbitrary positive number E, there exist positive trigonometric
polynomials g1 (8) and g2(8), such that on putting
1
f1(8) = {g1(8)} - I (z - z1)(z - z2) · · · (z - z1) I", z=e,
iB
(10.4.24)
f2( 8) = { g2( 8)} -\
we have
(10.4.25)
and

(10.4.26)

Here (f is the least even number greater than max C(f1, (12, · · · , (11); max h(8) is
1
bounded from above, while min {g1 (8) }- is bounded from below, and both bounds
are independent of E.

-·----------------- ·-~----,--------.,--·---------·---"
[10.4] LOCAL PROPERTIES OF THE REPRESENTATION 285

For the functions D(j1 ; z), D(f; z), and D(j2; z), corresponding, respectively, to
!1(8), f(O), and !2(8), a statement. similar to that in Theorem 10.4.3 holds.
REMARK. We can choose for (f any even number which is greater than
max (0'1 , 0'2, • • • , 0'1); in particular, we can choose an even number divisible
by 4. This is important for certain later purposes (cf. §13.5 (2)).
Let the function ](8) be identical with f(O) except in certain closed intervals
around e., v = 1, 2, · · · , l, in which ](8) = 1. These intervals are chosen so
small that they do not overlap, they do not ·contain a, and in each of them
f(O) ~ 1. Furthermore, let
< ~.
(10.4.27)
1-r
+"' log ](e) - log f(O) de
•
sm -
2 () -

2
-
a 4

Then f(O) ~ ](8) for each 8, and f(O) = ](8) in a certain neighborhood of a
which can be chosen independent of E provided E is small enough. The func-
tion ](8) = ](E; 8) satisfies the conditions of Theorem 10.4.2 and depends on E,
although it has an upper bound independent of E. The same is true for the
upper bound M of the ratio corresponding to (10.4.2).
We now determine the trigonometric polynomial g2(8) such that, for !2(8) -
{g2(8)}-\

+ .. log !2(8) - ~og ](e) de < ~.

(10.4.28)
1_,. •
sm - 2
8
2
-
a 4

We observe that max !2(8) is less than a number independent of E. (Here we

use the independence of M of E.)
On the other hand, let f(O) be identical with f(O) except in certain closed
non-overlapping intervals around e., v = 1, 2, · · · , l, not containing a, and
such that in these intervals
(10.4.29) J(z - z1)(z - z2) · · · (z - z1)l" ~ (z - z1)" 1(z - z2)" 2 · · · (z - z1)"
I I,
1

z = ei8• In these intervals we define

i8
(10.4.30) j(O) = 4>(8) J(z - z1)(z - z2) · · · (z - z1)i", z = e .

Furthermore, assume
< ~.
(10.4.31)
1 +.. log f(O) - log /(8) dO
-r
sm -
• 2 () -
2
-
a 4

Then f(O) ~ f(O), and f(O) = f(O) near 8 = a. Moreover,

i8
(10.4.32) f(O) i(z - z1)(z - z2) · · · (z - z1)i-", z = e ,
 280 REPRESENTATION OF POSITIVE FUNCTIONS [X]

satisfies the conditions of Theorem 10.4.2 and depends on E, although it

has a positive lower bound independent of E; the same is true for the (not
necessarily positive) lower bound 'Itt of the ratio corresponding to (10.4.2).
We determine a positive trigonometric polynomial g1 (()) such that
1
(10.4.33) \gl(e) l- ~ j(e) I (z - Z1)(z - z2) ... (z - zl) r", i8
z = e ,
with equality for () = a, and
+ .. log lf(e) J(z- z1)(z- z2) · · · (z- zl)r"l -log (gl(e) l- 1 d
(10.4.34)
1
-.- • 2 () - a
sm -
2
-
() < E-.
4

Then, using the notation (10.4.24), we obtain

c1o.4. 35) 1+.- log j(e) --: log f1Ce) de < 4~.
-.- • () - a
s1n2 - -
2
Addition of the inequalities (10.4.27), (10.4.28), (10.4.31), and (10.4.35) es-
tablishes (10.4.26). We observe that min (g1(e) l-1 is greater than a positive
constant independent of E.
The assertion concerning D(f1 ; z), D(f; z), and D(f2 ; z) follows as in Theorem
10.4.3.
(5) THEOREM 10.4.5. Let f(e) be an even function which satisfies the conditions
of Theorem 10.4.1; assume that 4>(()) = 4>(- e), and that all non-real "zeros" z. of f(e)
occur in conjugate pairs with the same "multiplicity." Also, let 0 < a < 1r.
Then the functions f1(e) and f2(e) of Theorem 10.4.4 can be chosen as even func-
tions, and we have instead of (10.4.26)

( 10.4. 36) ( .. lo~ f2(e) - log f)1~e) d() < E.

}o cos () - cos a
An analogous supplement can be made to Theorem 10.4.2. Previous proofs
need only a slight modification. Instead of the first ratio in (10.4.12), consider
{f(e)) - 1 - c - d (cos() - cos a)
(10.4.37) (cos() - cos a) 2
c = {!(a))-\ d sin a = f' (a) {!(a)) - 2 •

(10.4.37)
(cos() - cos a) 2
is bounded from 0 and oo if 0 ;:£ () ;:£ 1r.

-·---·---·---------- ·-------r-----------,r---------------······
----------------

CHAPTER XI

POLYNOMIALS ORTHOGONAL ON THE UNIT CIRCLE

A weight function on a given curve having been assigned, we may extend the
definition of polynomials orthogonal on a real interval to the more general
complex domain. The corresponding polynomials are then orthogonal with
this given weight function on the specified curve in the complex plane
(Chapter XVI).
Of all the special instances, that of a circle is most interesting, and in the
present chapter we shall consider the polynomials orthogonal on the unit circle
with a given weight function. It will be seen that these polynomials possess
properties which are in some respects simpler than those derived for poly-
nomials orthogonal on a real interval. Moreover, there exists a relation between
the case of the circle and that of a real, finite interval which enables us to
apply certain results of this chapter to polynomials orthogonal on a real interval.
Concerning §§11.1-11.4 see Szego 4. See also Grenander-Szego 1, Chapter 2.

11.1. Definition; preliminaries

(1) Let f(O) be a non-negative func,tion of period 211", integrable on [ -11", +11"]
in Lebesgue's sense, and assume
(h
(11.1.1) )_" f(e) de > o.
We introduce the Fourier constants

(11.1.2) Cn = 1 {~" j(O)e-inB de, n = 0, ± 1, ±2, · · · .

211" J-7r
Obviously, C-n = Cn, so that the matrix of "Toeplitz' type"
(11.1.3) v, J.L = 0, 1, 2, · · · , n,
is Hermitian.. The corresponding Hermitian form

(11.1.4) Hn = L L Cv-p.Up.il• = -2111" 1+" j(O)

n n
v=O p.-0 -1r
I
2
Uo + U1Z + UzZ + · · · + UnZn 1
2
dO,

where z - ei8, is positive definite and has the positive determinant

(11:1.5) v, J.L = 0, 1, 2, · · · , n.
7
(2) DEFINITION. If we orthogonalize the system5

&r Cf. the last remark in §2.1 (4).

287
288 POLYNOMIALS ORTHOGONAL ON THE UNIT CIRCLE [XI)

(11.1.6) z=e ~., n =0 ,1, 2,···,

we obtain a system of polynomials
(11.1. 7) ~(z), c1>1(z), cf>2(z), · · · , cl>n(z), · · ·
with the following properties:
(a) cl>n(z) is a polynomial of precise degree n in which the coefficient of z" is real
and positive;
(b) the system l cl>n (z) l is or.thonormal ; that is,

(1_1.1.8)
2
11+,. -
71' _,. f(e)cpn(z)cj>m(z) de = Onm, z = eiB ; n, m = 0, 1, 2, · · · .
Moreover, the system (cl>n(z) l is uniquely determined by conditions (a) and (b).
If f(e) is an even function, that ·is, if f(e) = f(- e), the coefficients of cl>n(z)
are real.
(3) We show (see §2.2 (2)) that

Cn-1 Cn-2 Cn-3 C-1

CoZ - C-t

.........................................
Cn-1 Z - Cn-2 Cn-2 Z - Cn-3

ct>o(z) = nc;t' n = 1, 2, 3, · · ·.
The coefficient of zn in cl>n(z) is
(11.1.10)
The analogues of the representations (2.2.10) and (2.2.11) can also be readily
derived.
(4) We pass now to considerations corresponding to those of (1), (2), and
(4) of §3.1.
THEOREM 11.1.1. Let F(ei 6 ) be a given measurable function for which

(11.1.11)

exists. The weighted quadratic deviation

(11.1.12) 1 1+.-
271' ~. f(e) I F(z) - p(z)
2
1 de, z = e ,
i6
[11.2] EXAMPLE 289

where p(z) ranges over the set of all 11"n, is a minimum if p(z) is the nth partial
sum of the Fourier expansion

(11.1.13) 1 (+"
2; J-1r
i6
Pn = f(e)F(z)4>n\z) de, z = e ;n = 0, 1, 2, • • ••

As a ready consequence, there follows Bessel's inequality

(11.1.14) I Fo 1
2
+ I F1l
2
+ I F2l
2
+ · ·· + I Fn
2
1 + · · · ~ 2~ i:"f(e) IF(e' 8) 12de.

In addition, Parseval's formula (that is, (11.1.14) with the equality sign) holds
if one of the following sets of conditions is satisfied:
(i) F(z) is regular and bounded for I z I < 1.
(ii) f(e) is bounded and F(z) is of the class H2 (see §10.1);
Concerning a more general condition, see Smirnoff 2, p. 363.
A consequence of Theorem 11.1.1 is the following:
THEOREM 11.1.2. The polynomial K-;; 14>n(z) minimizes the integral

(11.1.15) iB
z = e ,

if zn + a 1zn- + · · · + an ranges over the set of all 11"n with the highest term zn.
1

The minimum is K-;; 2 •

11.2. Example
An important special case in which the system (4>n(z) l can be calculated
explicitly, except for a finite number of terms, is
(11.2.1) f(e) = {g(e)l-\
where g(e) is a positive trigonometric polynomial of degree m.
THEOREM
·e
11.2. Let f(e) be defined by (11.2.1), and let g(e) = I h(z) I2, z = e',
be the normalized representation of g( e) defined in Theorem 1.2.2. Using the
notation of (1.12.4), we have
(11.2.2) 4>n(z) = zn-mh*(z) = znh(z- 1), n=m J m+1 ' m+2 J ····
Evidently, condition (a) of the definition in §11.1 (2) is satisfied. In order
to show the orthogonality, let p(z) be an arbitrary 11"n_1 • Then, if z = e' 8,
we have, according to Cauchy's theorem,

-1 1+" f ()e 4>n (z)- 1 J

p(z) de = -2 {h( z)-(
h z-1))-1 zn-(
h z-1) p(z-1)dz
--;-
211" -r 11" lzl=-1 tZ

=-
1 fr zn-1 p(z--,1)
dz = 0.
27ri 1~1-1 h(z)
290 POLYNOMIALS ORTHOGONAL ON THE UNIT CIRCLE [XI]

In addition

2~ i:"f(O) I<Pn(z) l
2
d8 =
2
1
71" J:" 2
lh(z) l- lznh(z- 1) l
2
de = 1.

The simplest case is f(O) = 1. Then

(11.2.3) n = 0, 1, 2, .. · .
Concerning other cases in which an explicit calculation of <Pn(z) is possible,
we refer to Szego 4, pp. 187-188; and 12, pp. 245-247. See also (11.5.3) and
(11.5.4). '

11.3. A maximum problem

Because of the similarity between this problem and that treated in §3.1 (3),
we can omit details.
(1) THEOREM 11.3.1. Let p(z) be an arbitrary 11"n subject to the condition

(11.3.1) z = eiO .
For a fixed value of a, the maximum of I p(a) )2 is attained if
(11.3.2) p(z) = dsn(a, a) l-tsn(a, z), IE I = 1,
where
n
(11.3.3) Sn(a, z) = L ;p.(a)q,.(z).
•-0

The maximum itself is sn(a, a).

These "kernel polynomials" sn(a, z) can be used for the representation of the
partial sums of the expansion (11.1.13) in form of integrals (see (3.1.11)).
(2) THEOREM 11.3.2. For a ~ 0 the polynomials (11.3.2) satisfy the following
identity:
(11.3.4) - ) n Sn c--1
Sn ( a, z ) = ( az z, , a--1) .
Furthermore,
n
(11.3.5) Sn(O, z) == L cp.(O)cj>.(z)
v=O
= Kn Zn i/>n(z- 1) = Kn cj>: (z

where Kn has the same meaning as in (11.1.10); finally

n
(11.3.6) Sn(O, 0) = L I cp.(O)
v=O
1
2
= K! = Dn-d Dn .

The last formula also holds for n = 0 if D_1 = 1.

Introducing p*(z) = r(z) or p(z) = r*(z), we have
[ 11.3 l A MAXTMU::\1 PROBLE::\f 2!) 1

(11.3. 7) z = eiB ,

and, for a ~ 0,
(11.3.8)

This yields sn(a, a) = Ia l

2
nsn(a-r, a·-
1
), that is, (11.3.4) for z = a, and also
(11.3.9)

where e is a proper constant with I e I = 1. (The symbol * refers to the

variable z.) On combining this with the first result, we obtain (11.3.4). In
the limiting ca~.;;e a = 0, the identity (11.3.5) arises, from which, for z = 0,
(11.3.6) follows.

(3) THEOREM 11.3.3. Let log j(e) be integrable in Lebesgue's sense. 58 Then
the following limits exist:
00

(11.3.10) lim sn(a, a) =

n-+oo
L
v=O
I cp.(a)
2
1 , Ia I < 1,
00

(11.3.11) lim Sn(a, z) =

n-+oo
L
v=O
cj>.(a)cj>.(z), lal<1, lzl<1,
(11.3.12) lim Kn = K > 0, lim z-n cj>,.(z), I z I> 1,
n-+0

(11.3.13)
lim cl>n(z) = 0,
n-+oo
I z I < 1.

We first consider the special case j(e) s >

0, assuming I a I < 1. Let
J-L
p(z) have the same meaning as in (11.3.2). Then we have, because of Cauchy's
inequality (see (7.1.4)),

I p(a)
2
1 ~ - -1
-- -1 1+"" p(z) de I
2
1

(11.3.14) 1 - I a 12 271" _,.

~ _J-L_I_I~ -1 1+"" j(e) I p(z) 1de =

-1
2 J-L i8
-1
z = e ;
1 - a 2 71" _,.

consequently, the same inequality holds for n ---+ a:;, Thus (11.3.10) and
(11.3.11) are established in this case. For a = 0, (11.3.6) and (11.3.5) show
that the limits (11.3.12) exist; (11.3.13) follows from the convergence of
(11.3.10).
In order to prove the statement generally, we first observe that the maximum
in the problem of Theorem 11.3.1 is attained for a polynomial p(z) which is
different from zero for I z I < 1; here again I a I < 1. In fact, if Zo is a zero of
p(z), I zo I < 1, then

bB cr. §10.2 (2).

292 POLYNOMIALS ORTHOGONAL ON THE UNIT CIRCLE [XI]

-
1+. f(O) Ip(z) 1_- z;; z 12 dO = 1;
__!__
271" _,. z - Zo
I p(a) I2 <
I p(a)
1 - zoa;
a- Zo
i ,
1
2
z = e .
i9

Now assume a = rei<P, 0 ~ r < 1, z = e;9 ; according to (1.11.3),

1
1
= 2 71" 1:,. f(O) I p(z) 1
2
dO ~ : + ~ 2171" [:,. f(O) I p(z)
2
1
1 _ 2r c~s Co: 4>) + r2 dO
- 1
1 - -r exp { -1
2::: -
+r
1+.-log f(O) 1 -
21r _,.
1 - r
2r cos (0 - 4>)
2

+r 2
dO}

·exp { -1 1+.-
21r -..
log lp(z) 1
2 1 r )
1 - 2r cos 0 - 4>
C
2

+r 2
}
d0.

The last integral is Poisson's integral of the harmonic function 2~ log [p(z)],
which is regular for I z I < 1. The last exponential factor is therefore equal to
exp 12m log (p(a)] l = I p(a) 12 ;
this establishes the boundedness of max I p(a) 12 = sn(a, a).
The further formulas follow as before. Later (§12.3 (6)) we shaH calculate
the limits (11.3.10)-(11.3.12).
11.4. Algebraic properties
(1) Let a be fixed, I a I < 1. The above considerations show that the zeros
of sn(a, z) lie in I z I ~ 1. The same is true, of course, for I a I ~ 1. From
(11.3.5) we find that the zeros of 4>n(z) lie in I z I ~ 1.
We now prove the following more informative statement:
THEOREM 11.4.1. For I a I < 1 the zeros of sn(a, z) lie in I z I > I ,for I a I > 1
in I z I< 1, and for I a I = 1 on I z I = 1. The zeros of 4>n(z) lie in I z I < 1.
Let z0 be an arbitrary zero of sn(a, z). If we put

(11.4.1) fl(O) = f(O) I Sn(a, _z) 12, z = e ,

i9
z- Zo

and consider all the linear functions p(z) with

i9
(11.4.2) z = e ,

it is clear that max I p(a) 12 is attained for p(z) = const.. (z - zo). It therefore
suffices to discuss the case n = 1. From (11.1.9)
s1(a, z) = 4>o(a) 4>o(z) + 4>I(a) 4>1(z)
(11.4.3)

with the zero

C1 ii - Co
(11.4.4) z = ---- ····- .
Co a- C1

---------·--·-·-·--··---···
[ 11.4 l ALGEBRAIC PROPERTIES 293

This establishes the statement concerning sn(a, z) since I c1 I< c0 (se~ (11.1.2) ).
The statement concerning 4>n(z) follows from (11.3.5).
(2) THEOREM 11.4.2. We have the identity

(11.4.5) sn(a, z) = i: 4>v(a)4>.(z)

v=O
= cf;~4>:+1(z) - ~-;:.-;;(a)4>n+l(z)
1 - az
and the "recurrence formulas"

(11.4.6) KnZ4>n(z) = Kn+14>n+l(z) - 4>n+l(0)4>:+1(z),

(11.4.7) Kn4>n+l(z) = Kn+IZ4>n(z) + 4>n+l(0)4>:(z).
The first identity corresponds, in some respects, to the Christoffel-Darboux
formula (3.2.3). The proof can be given along the same line as the proof in
§3.2 (3). As in the case of a real interval, we can characterize sn(a, z) by the
equation
1 [+"
(11.4.8)
2
7r )_, f(O)sn(a, z)p(z) dO = p(a), z = ei9 ,

which holds if p(z) is an arbitrary 7rn . But

1+" f(O)
__!__ ~~+l(a)4>:+1(z) - _4>n+l(a)4>n+:~! P(Z) dO
21r _, 1 - az

= p(a) 1+" f(O)

__!__ ~~a)4>:+1(z) - -~~~-l~_f!_)4>n+_1C~! dO
(11.4.9) 27r -rr 1 - az
+ -- -
+ -21r 1
1_,
" - ------ - * --- p(z) - p(a)
.f(O) 14>~+1(a)4>n+l(z) - 4>n+l(a)4>n+l(z) l -- - - _- · dO,
1 - az
i9
z = e .
The last integral vanishes, for if we write p(z) - p(a) = (z - a)r(z), we have

(11.4.10) ,e
z = e .

Therefore,

(11.4.11)

where c is independent of <:. Interchanging a and z and taking the conjugate

complex values of both sides, we see that c is also independent of a. Writing
z = a = 0, we obtain
n
(11.4.12) 2
Kn+l - l4>n+l(O) 1
2
= C L
v=O
l4>v(O)
2
j ,

so that c = 1, by (11.3.6).

-----·-·----····
 294 POLYNOMIALS ORTHOGONAL ON THE UNIT CIRCLE [XI]

Comparison of the coefficients of an in (11.4.5) leads to (11.4.6). By taking

the reciprocal polynomials of both sides of (11.4.6), and eliminating 4>:+1 (z),
we find (11.4.7).

11.5. Relation to polynomials orthogonal on a real interval

(1) THEOREM 11.5. Let w(x) be a weight function on the interval -1 ~ x ~ + 1,
and let
(11.5.1) j(e) = w(cos 0) I sin () ).

Further let IPn(x) l and I qn(x) l be the orthonormal sets of polynomials which
belong respectively to w(x) and (1 - x )w(x) in -1 ~ x ~ + 1, and (<Pn(z) l the
2

orthonormal set associated with f(O) on z = e' 9• Then, on writing x = Hz+ z- 1),
we have for n ~ 1

Pn(x) = (27r)-! {1 + 4>2n(O)}-! (z-n cP2n(z) + Zn cP2n(Z-1)}

K2n

See Szego 6, pp. 204-206. The second equation follows from the first one,
and similarly, the fourth follows from the third, by means of (11.4.7). The
function f(O) is even; so the polynomials <Pn(z) have real coefficients.
The constant factors in these equations are different from zero (see (11.3.6)).
These formulas, except the second one, hold also for n = 0.
(2) The right member of the first equation represents a 7rn in x; the property
of orthogonality can be expressed in the form
+,.

1_ .. Pn(cos 0) cos vO·w(cos 0) I sin ()I dO = 0;

or, for z = ei9 ,

v = 0, 1, 2, . · · , n - 1.

This is the case because

and Ci>2n(z) = cP2n(z). Moreover,

r 11.s1 POLYNO;v1JALS ORTHOGONAL OX A HEAL INTERVAL 295

For the third formula the proof is similar, except that it is convenient now to
express the orthogonality in the form

(v + 1)0 sm
1 +,.
-r
qn ( cos 8) sin
----;-----
Sin 0
. 2 O·w ( cos 0) I sm
. 0 IdO = 0
· '

v=012 ... n-1

' ' ' .' .
A new proof for the formulas (2.6.2) and (2.6.3) can be derived from the first
and third formulas in (11.5.2) by use of (11.2.2).
If in the last two equations of (11.5.2) we replace n by n - 1, we can express
1
z-"¢2n(z) and z-"+ ¢zn-J(z) as linear combinations of Pn(x) and (l - X2 ) 112qn-l(x),
1
where x = Hz+ z- ). (See Szego 18, pp. 9-11.) These relations enable us to
calculate the polynomials ¢n(z) associated with
(11.5.3) f(O) = I (1 - z)..,(1 + z) 8 2
j = 2..,H(1 - cos Of(1 + cos 0) 8
, z = e; 9,
in terms of Jacobi polynomials. vVe find
z-n ¢2n(z) = AP~..,--!.8-!l I Hz+ z- l + B(z -
1
) z- 1 )P~~i!.H!l I Hz+ 1
Z- ) l,
(11.5.4) z-n+ 1 ¢2n-1(z) = CP~-y-!.8-!ll Hz + l 1
Z- )

+ D(z - z- 1 )P~~i!,H!ll Hz - z- 1) l,
where A, B, C, Dare proper real constants.
 CHAPTER XII

ASYl\1PTOTIC PROPERTIES OF GENERAL ORTHOGONAL

POL YN0:\1IALS
The following sections deal with the asymptotic properties of polynomials
orthogonal on the unit circle, or on a real, finite interval, when the degree
n of these polynomials becomes infinite. In both cases the weight function
will be restricted merely by certain conditions of continuity and boundedness.
T\vo important problems appear in connection \Vith polynomials orthogonal
on the unit circle. These are (a) the asymptotic behavior exterior to the unit
circle, (h) the behavior on the unit circle itself. For a weight function iden-
tically unity, the system in question is I zn j. This latter instance is typical to a
certain extent.
The corresponding problem'-> for polynomials orthogonal on a finite segment
are (a') the asymptotic behavior in the complex plane cut along the given seg-
ment, (b') the corresponding question on the segment itself. (See Chapter
VIII.) \Ve give the following illustration as characteristic:

Problems (a) and (a') are simpler, and relative to them comparatively general
results will be obtained. It is only recently that problems (b) and (b'), which
are much more difficult than (a) and (a'), have been treated by S. Bernstein
and G. Szego. We remark that the weight function conditions in case (b) are
more restrictive than those in case (a). A similar comment may be made
concerning (a') and (b').
The results of Chapter X are applied in discussing the above problems. Our
investigation is first concerned with questions (a) and (a'). As regards (b')
we may state that S. Bernstein's main result is obtained in a new way which
is shorter than his original argument. Then there follows Szego's older method
as applied to (b').
12.1. Results
(1) Let G denote the class of functions f(O) ~ 0, defined and measurable in
+1r], for which the integrals
[ - 1r,

(12.1.1) J_:" \log f( e) I dO

exist with the first integral supposed positive. With such a function f(O) we
associated in §10.2 a uniquely determined analytic function D(f; z) = D(z),
regular and nonzero for j z I < 1 with D(O) > 0. The conditions for the class G
imply the existence of the "geometric mean" ®(f) = (D(O) l of f(O).
2

296
[12.1] RESULTS 297

THEOREM 12.1.1 (Asymptotic formula for the polynomials orthogonal on the

unit circle, considered for z exterior to the unit circle). Let f( 0), belonging to the
9
class 0, be a weight function on the unit circle z = e; • If l<t>n(z)} denotes the
corresponding orthonormal set of polynomials, then in the exten:or of the unit circle
(12.1.~~) I z I> L
This holds uniformly for Iz I ~ R > 1.
THEOREM 12.1.2 (Asymptotic formula for the polynomials orthogonal on the
segment [-1, +1], considered for x exterior to this segment). Let w(x) be a
weight function on the interval -1 ~ x ~ +
1 such that w( cos 0) I sin 0 I = f( 0)
belongs to the class G. If D(f; z) = D(z) denotes the analytic function corresponding
to f( 0) in the sense mentioned, the orthonormal polynomials I Pn (x) l, associated with
w(x), possess the asymptotic formula
(12.1.3)
Here x is in the complex plane cut along the real segment [- 1, +1], and x =
Hz +z- 1), where I z I > 1. Formula (12.1.3) holds uniformly for Iz I ~ R > 1.
(2) In order to obtain the deeper asymptotic formulas = 1 andvalid on I z·l
on -1 ~ x ~ +1, respectively, we must impose certain further restrictions on
f(O) and w(x).
THEOREM 12.1.3 (Asymptotic formula for the polynomials orthogonal on the
unit circle, considered for z on the unit circle). Let f(O) be a positive weight
function on the unit circle, which satisfies the Lipschitz-Dini condition
(12.1.4) I f(O + o) - f(O) I< L I logo I - 1->-,
where Land Xare fixed positive numbers. Then we have, for I z I= 1,

(12.1.5)

where limn-+oo En(z) = 0, uniformly for Iz \ = 1. More precisely,

(12.1.6)

the positive constant C depends on L, X, and the minimum and maximum of f(O).

THEOREM 12.1.4 (Asymptotic formula for the polynomials orthogonal on the

segment [ -1, + 1], considered for x on this segment). Let w(x) be a weight
function on the interval - 1 ~ x ~ +
1, x = cos 0, such that the function
w(cos 0) I sin 0 I = f(O) sat?:sfies the conditions of Theorem 12.1.3. Putting

(12.1.7)

we have uniformly on the segment -1 ~ x ~ + 1 or 0 ~ 0 ~ 1r, x = cos 0,

(12.1.8)
298 ASYMPTOTIC PROPERTIES OF GENERAL POLYNOMIALS [XII]

The constant factor in the 0-term depends only on L, X, and the minimum and
maximum of f(O).

(3) Finally, we prove two theorems similar to Theorems 12.1.3 and 12.1.4
under conditions possessing a kind of "local" character.
THEOREM 12.1.5 (Asymptotic formula for the polynomials orthogonal on the
unit circle for I z I = 1, and with weight function subject to a "local" condition).
Letf(O) satisfy the conditions of Theorem 10.4.1. Then (12.1.5) holds for z = a =
eia, in the less informative form
(12.1. 9) En--+ 0.
From this we obtain the following theorem of "local" character, which cor-
responds to Theorem 12.1.4:
THEOREM 12.1.6 (Asymptotic formula for the polynomials orthogonal on the
segment [-1, + 1] for x on this segment, and with weight function restricted
by a "local" condition). Let w(x) be integrable in Riemann's sense, and let it
have the form
(12.1.10) w(x) = t(x) Ix - X1 I Tt I x - X2 I T
2
• • • Ix - xz I TZ,
where 0 < A ~ t(x) ~ B, and -1 ~ x 1 < X2 < · · · < xz ~ 1, r. > 0, v = 1, 2,
Further let w(.x) be differentiable at the fixed point x = ~' where -1 < ~
59
... , l.
+
< 1, and~ rf x., v = 1, 2, · · · , l, and let
w(x) - w(~) - (x - ~)w'(~)
(12.1.11)
(x - ~)2
be bounded if x is near to ~.
Then (12.1.8) holds in the less informative form
(1 - ~ ) 1 (w(~;)) 1 pn(~) =
2
(2/-rr} cos (na + 'Y(a) l + En,
(12.1.12)
~ = cos a, 0 < a < 1T', lim En = 0,
n-+oo

where 'Y(a) has the same meaning as in (12.1.7).

12.2. Remarks
(1) Theorems 12.1.2, 12.1.4, and 12.1.6 follow readily from 12.1.1, 12.1.3,
and 12.1.5, respectively. We observe that D(z) = D(z) in (12.1.3); that is,
D(z) is in this case a "real" function.
In Theorems 12.1.3-12.1.6 the function D(z) has boundary values at the point
considered (in Theorems 12.1.3 and 12.1.4 even continuous boundary values
on the whole unit circle I z I = 1). This follows from the considerations of
Chapter X.
The important function 'Y(O), defined by (12.1.7), is completely determined
6g For x 1 = -1 it suffices to assume r1 ~ -1/2; similarly for xz = +1.
[ 12.2 l REMARKS 299

save for an integral multiple of 21r. If we choose (see (10.3.9))

1 [+" 0 - t
(12.2.1) 'Y(O) = 7r )_, [log j(t) - log f(O)] cot - - dt,
4 2
then 'Y(O) is continuous. In the cases occurring in Theorems 12.1.4 and 12.1.6,
'Y(O) can easily be expressed in terms of the weight function w(x). Let
(12.2.2) f(O) == w(cos 0) I sin 0 I = W(cos 0).
From (10.3.9) we obtain (seeS. Bernstein 2, p. 132),

'Y(O)
4
lh 0 - t
= -17r _, Pog W(cos .t) - log W(cos 0) l cot---
2
dt

= _.!_ {1r l!og W(cos t)- log W(cosO)l {cot 0-~-t +cot O_±_t}dt
47r }o 2 2
(12.2.3)
= -
1 11r (log W(cos t) - log W(cos 0) l · ~nO dt
27r 0 cos t - cos 0
= _.!_ (+1 log W(~)
27r J-1 ~
- log
- X
~~x] (11 -- ~x2)i d~'
2
x = cos 0.

In this case
(12.2.4) 'Y(- 0) = -')'(0).
For the functions I (f(0)) 1<t>n(z) l, z = ei 9, n = 0, 1, 2, ... , which form an
orthonormal system in the usual sense (see the Definition in §11.1 (2)), we
obtain from (12.1.5) the simple asymptotic expression zneir<e> = /I nHr<e> l.
(2) Theorem 12.1.1 is a direct consequence of Theorem 12.1.3 provided con-
dition (12.1.4) is satisfied. In fact, the function z-n<Pn(z) - ID(z- 1) l-1 is regular
for I z I > 1 and continuous for I z I ~ 1. In this special case, (12.1.2) follows
in the more informative form
(12.2.5)
uniformly for I z I ~ 1.
(3) Concerning Theorems 12.1.1 and 12.1.2, see Szego 6. Under more re-
strictive conditions than those in Theorem 12.1.2, Faber (4) proved
(12.2.6) lim I Pn(X) l11 n = I Z I , Iz I > L
n-+oo

This less informative statement suffices for various applications, for instance,
for the purposes of §12.7 (2) and (3). Theorem 12.1.3 is new, while 12.1.4 is
due to S. Bernstein (2). The proofs of Theorems 12.1.3 and 12.1.4, given in
§12.4 and 12.5 (2), respectively, are based essentially on an idea of S. Bernstein,
used in his original proof, and on another idea similar to that used in connection
with the method of LiouviHe-Stekloff (§8.61). As mentioned, this arrange-
ment seems to be simpler than S. Bernstein's original line of argument. The-
300 ASYMPTOTIC PROPERTIES OF GENERAL POLYNOMIALS [XII]

orem 12.1.5 1 and Theorem 12.1.6 which is its consequence, are due to Szego (8).
The conditions required in the present treatment are slightly more general than
those in the paper just cited.
S. Bernstein assumes (2, p. 132, (18))
I W(x + o) - W(x) I <L I log 0 l-1->.; -1 ~ X ~ +1, -1 ~ X+ 0 ~ +1,
instead of (12.1.4). These conditions are, however, equivalent since
log {I cos fh - cos fJ2I- 1}
log II o1 - o2 l-1}
is bounded from 0 and oo if fJ1 and fJ2 are arbitrary in [0, 1r], I fJ1 - fJ2 I ~ !. 60
12.3. Proof of Theorem 12.1.1; applications
(1) Before we proceed to this proof, let us consider the special case f(O)
1
(g(fJ) }- , where g(fJ) is a positive trigonometric polynomial of degree m. Let
h(z) have the same meaning as in §10.2 (1). According to the second formula
in (10.2.12) we have D(f; z) = {h(z) }-1. On the other hand, by (11.2.2)
(12.3.1) 4>n(z) = znh(z-1) = zn(D(z-1) } - \ n ~ m.
The limit relation (12.1.2) can therefore be replaced by one of equality in this case
provided n ~ m.
(2) Now let f(O) again be an arbitrary function satisfying the conditions of
Theorem 12.1.1. Let p(z) = zn + a1zn-1 + · . · + an be an arbitrary 7rn with
the highest term zn. According to Theorem 11.1.2, the minimum P.n(f) of

(12.3.2)
1 1+,.
27r _,. f(O) I p(z) 12 dfJ, z = i9
e ,
1
is K.n - 2, attained for p(z) = K-;: 4>n(z).
LEMMA. Let f(O) satisfy the conditions of Theorem 12.1.1, and let P.n(f) have
the previous meaning. Then
(12:.3.3) lim P.nCf) = @(f),
61
where @(f) is the "geometric mean" off( fJ).
If p(z) is any one of the polynomials considered, then zp(z) is a 7rn+1 with the
highest term zn+l, and I zp(z) 12 = I p(z) I if z = ci • Thus Ji.n+1(f) ~ P.nCf).
2 9

(This follows also from (11.3.6).) Consequently, limn ..... oo P.n(f) = p.(f) exists,
and p.(f) ~ 0. We must show that p.(f) = ®(f). ·
(3) Using the inequality for the arithmetic and geometric means, we have
'9
for z = et,
60 If I 81 - 82 I = o, o ,;:;; 7r/2, is given, the maximum and minimum of I cos 81 - cos 82l
are 2 sin (o/2) and 2 sin 2 (o/2), respectively.
ot Cf. §10.2.
[ 12.3 l PROOF 01<' THEOREM 12:1.1; APPLICATIONS 301

1 [+" { 1 [+1( }
dO ~ @(f) exp
2 2
27r )_, f(O) I p(z) 1 7r )_, log I p(z) 1 dO
2
(12.3.4)
= @(f) exp {
2
1
71" L:" log I p*(z)
2
1 dO} ~ @(f) I p*(O)
2
1 = @(f),

according to Jensen's theorem (see, for example, Titchmarsh 1, p. 125). There-

fore, p.,.(f) ~@(f), and p.(f) ~@(f).
On the other hand, let T(O) be a non-negative trigonometric polynomial of
degree k, not vanishing identically, and let T(O) = I P(ei 9) I 2 be the correspond-
ing normalized representation in the sense of Theorem 1.2.2. Then @(T) -
(P(O) \ • The highest coefficient of !P(O) l-1P*(z) is 1, so that
2

(12.3.5)
®(f) ~ p.(f) ~ J.Lk(f) ~ 2~ i:" f(O) I !P(O) l-1P*(z) 12 dO

= (@(T)\-
1
2
1
71" £:" f(O)T(O)dO, z = ei9 .

By Weierstrass' theorem, the inequality

(12.3.6)

holds for an.arbitrary positive and continuous function T(O) which has the period
21r. F'or the special case in which f(O) is positive and continuous, the lemma
follows from this by writing T(O) = lf(O) l- 1•
(4) Now in the general case assume first f(O) ~ p. > 0, and let E be an arbi-
trary positive number. By Theorem 1.5.3;. we can find a trigonometric poly-
nomial 4>( 0) such that 4>( 0) ~ p. and
1 [+"
(12.3.7) 2; )_, If(O) - 4>(0) I dO < €.
We then have
(12.3.8) log 4>( 0) - log f( 0) ~ p.- 1 I 4>( 0) - f( 0) I;
1
hence ®(4>) ~ @(f)e"'- '. In (12.3.6) we write T(O) = (4>(0) l-1 • Then

(12.3.9)
1
1271" £:" f(O) T(O) dO - 11 ~ J1. - 1 1
2 71" £:" If(O) - 4>(0) I dO < J1.- 1€,
so that
(12.3.1 0)

whence, since E is arbitrary, p.(f) = @(f).

In the general case we use the obvious inequality p.(f) ~ p.(f + t:) =
@(j + t:), E > 0, from which p.(f) = @(f) follows again. This establishes the
proof of the lemma.
 302 ASYMPTCTIC PHOPEWfiES OF GE::\ERAL POLYNOMIALS [XII]

(.5) For the proof of Theorem 12.1.1 let us now consider the function
(12.3.11) D(z)¢:(z) ·- 1 = [D(O)Kn - 1] + dnlZ + dn2Z 2 + · · · ,
which is regular for I z I< 1. If r < 1, we have

(12.3.12)
1
2rr 1:" 8
I D(rei )¢: (rei
8
) -
2 1
1 1 dO = rr
2
1:" I D(rei8 ) 12 I¢: (rei 8 ) 12dO

1
+1)- 2ffi { rr
2
i:" 8 8
D(rei )¢:(re; ) do}

The third term is evidently -2ffi[D(O)¢:(o)] = -2D(O)Kn. As r ~ 1 - 0,

we obtain, by use of (10.2.9),

r~~o 2
1
71' 1:" I
8
D(r/ )¢: (rei
8
) -
2
1 1 dO =
1
2 71' 1:" f(O) I¢: (ei
8 2
) 1dO +1- 2D(O)Kn

= 2171' 1+" f(O) I¢n(et) I dO + 1 -

-r
·e 2
2D(O)Kn =2 - 2D(O)Kn

= 2 - 2( @(f) l t l J.Ln (f) \-; j

or in another form,
2 2 2 2+
I D(O)Kn - 1 1 + I dnl 1 + I dn2 1 + I dna 1 •••
(12.8.13) '
= 2- 2(@(!)} (J.Ln(f)}-!.

In consequence of Cauchy's inequality this yields, for Iz I < 1,

I dnlZ + dn2i + dnai + · · · 12 ~ (I dn1J 2 + Jdn2l 2+ Idnal 2+ ·· ·) 1 ~ ~~: !2

(12.::U4)
2
~ 1 ~!lzl2[2- 2l®(f)}'lJ.Ln(f)r-i],

and as n ~ oo, the last expression tends to 0 uniformly in z for I z I ~ r < 1.

2
The same is true for I D(O)Kn - 1 1 . Theorem 12.1.1 now follows at once
from (12.3.11).
(6) As an application we calculate the limits occurring in Theorem 11.3.3.
The second limit in (11.8.12) is given by (12.1.2); the special case z = oo yield~
(12.3:.1.5) lim Kn
n-+oo
= n-+ooz-+oo
lim lim lz-" ¢n(z)} = lD(O)} ·l = l ®(f) r·'.
The same formula (12.1.2) furnishes, for I z I < 1,
(12.3:.16)
n-+oo n-+oo

and a similar formula holds for ¢:+I(a), l a l < 1. Sinec ¢n+I(a) ~ 0 and
<PnH(z) ~ 0 (sec (11.3.13)), we have from (11.4 ..5)
[12.4] PROOF OF THEORE:\I 12.1.3 303

00
1 1 1
(12.3.17)
•
hm Sn(a, z) = L -
¢.(a)¢.(z) = - - = - .
n-.oo •=o 1 - az D(a) D(z)
For instance,

(12.3.18) Iz I < 1,

and in particular (see (11.8.6))

(12.8.19) ~ I¢. (O) 12

L.., =
l'
1m Kn
2 l' Dn-1
= n-+oo 1
1m -D = D( ) I 1
1
= ru(f) ·
P=O n-+oo n 0 2 ~

From the last equation (12.8.15) follows again.

12.4. Proof of Theorem 12.1.3

(1) Let g(O) be a positive trigonometric polynomial of degree m, determined
as in §10.8 (8), and let D(g; z) = h(z), as there, so that D(g- 1 ; z) = lh(z)\-1•
Let l ifin(z) l be the orthonormal set of polynomials associated with the weight
function jg(O) l- on the unit circle. We have, by Theorem 11.2,
1

(12.4.1) ifin(z) = znh(z - 1), n ~ m.

In close relationship with an idea of S. Bernstein (2, p. 158, (75)), we shall
now express the polynomial ¢m(z), associated with f(O), in terms of the poly-
nomials if;.(z), corresponding to lg(O) l- 1 :

¢m(z) = ~a. if;.(z) = am ifim(z)

m
+ 21 [+"
7!" )_,.
{m-1
jg(t) \- ¢m(f) ~ if;.(f)if;.(z) dt
1 }
(12.4.2) = am ifim(z)
r+r [lg(t) \-
+ 21 J-r 7!"
1
-
{m-1
f(t)]¢m(f) ~ if;.(f)if;.(z) dt
}
+ 21 [+"
7!" )_,.
{m-1
f(t)¢m(f) ~ if;.(f)if;.(t) dt,
} r= e .
it

The last term vanishes because of the orthogonality of ¢m(z).

Let K"' and K: = h(O) be the highest coefficients of ¢m(z) and iflm(z), respec-
tively; then a, = K,K;,-;- = Km!h(O) l- • Furthermore, by §12.8 (2), we have
1 1

l D(O) \ 2 = @(f) ~ /J.m(f) = K-;;.

2
~ 2~ l:" J(O) llh(O) l- h*(z)
1
12 dO

2~ L:" I D(z)h(z) 1dO,

2 2
= lh(O) l-
and because of (10.8.12) this equals
Ih(O) l-2 11 + O[(log m)->-Jl - I D(O) \ 2 + O[(log m)->-],
so that
(12.4.3) Km = !D(O) l-1 + O[(log m)->.],
304 ASYMPTOTIC PROPERTIES OF GENERAL POLYNOMIALS t XII]
(2) We shall next try to find a bound for
(12.4.4) max I ¢m(z) I = M = M(m).
Jz J=1

Using (10.3.11), we find from (12.4.2) that

(12.4.5) M ~ 0(1) + O[(Iogm)- 1 ->-]·M·~,~~ l:,.l%1/t.(f)ljt.(z)ldt, f =/ 1

•

Because of (11.4.5), the sum under the integral sign can be represented in
the form

(12.4.6) 'I: 1/t.(f)l/t.(z)

v=O
= h(flh(z) - fi*Tf)h*(z)
1 - fz
We now show that

(12.4.7) I:" I h(f)h(z\-=- r:)h*(z) Idt = O(log m), r= e '

it

uniformly in z for I z I = 1. Indeed, the numerator is a 7rm in z, which vanishes

for z = r, Therefore, the theorem of S. Bernstein (seeM. Riesz 1, especially p.
357) furnishes O(m) for the integrand. Thus the. contribution of the arc I r- z I
~ m- is 0(1), while the complementary arc 1 r - z 1 > m- 1 supplies
1

0(1)

Returning to (12.4.5), we obtain

1 11-zJ>m-1
II
dt
- rz_ I = O(log m).

M ~ 0(1) + O[(log m)->-]M,

so that M = 0(1). Hence, in view of (12.4.3), (12.4.1), (10.3.11), and (12.4.7),
equation (12.4.2) yields

(12.4.8) ¢m(z) = ll + O[(log m)->-]lzmh(z- 1) + O[(log m)- 1->-]0(l)O(log m).

The assertion of Theorem 12.1.3 now follows because of (10.3.12). The con-
stants of all 0-terms of this section depend only on L, )1., and the minimum
and maximum of f(O).

12.5. Asymptotic formulas for the polynomials on a finite segment; proof

of Theorems 12.1.2 and 12.1.4
Theorems 12.1.2 and 12.1.4 follow from Theorems 12.1.1 and 12.1.3, respec-
tively, almost immediately by using (11.5.2). It suffices to use the first formula.
(l) If x and z have the same meaning as in Theorem 12.1.2, we have
lim z- 2n¢2n(z) = ID(z-1)\-1 ; lim ¢2n(z-1) = 0;
n--+oo n--.oo
(12.5.1)
lim ¢2n(O) = 0;
n-+oo
lim
n--+oo
K2n = K > 0.
 [ 12.6 ] THE ASYMPTOTIC PROBLEM UNDER LOCAL CONDITIONS 305

Here we took into account (12.1.2), (11.3.13), and (11.3.12). If (11.5.2) is

taken into consideration, this establishes Theorem 12.1.2.
(2) F'or the proof of Theorem 12.1.4, let n be an arbitrary integer, m = n - 1,
and let g(B), as well as h(z), be determined as in §10.3 (3). Then by (12.1.5)
and (12.1.6),

11+.-
¢,.(0) = 211" -.- ¢n(z)dB
11+.-
= 11" -.- zn[lD(z)\-1
2
- h(z)]dB
(12.5.2) +
+ 2~ l.. zn h(z) dB + O[(log n)->'], z = ei8 .

The first integral on the right is O[(log n) ->-], while the second vanishes, since
h(z) is a 11"n-l. Therefore,
(12.5.3)

The sequence lK2nl is bounded from 0 (K~ ~I ¢o(O) 12 ~ lmaxf(B) l-\ accord-
ing to (11.3.6)), so that
Pn(x) = (211")-lll + O[(log n)->-]l2ffilzn[D(z)r + O[(log n)->-]l
1

(12.5.4)

x = cos () z = e''8
' '
which is identical with (12.1.8). The assertjon concerning the constant of the
0-term is immediate.
The same result is obtained from the second formula in (11.5.2).

12.6. The asymptotic problem under "local" conditions ; proof

of Theorems 12.1.5 and 12.1.6
In this section essential use is made of the approximations given in Theorems
10.4.4 and 10.4.5.
(1) First the following problem will be considered:
PROBLEM. Let "J\, J.L, and a be arbitrary complex numbers, and let f(B) be an
arbitrary weight function on the unit circle. We intend to determine the maxi-
mum of
(12.6.1)
when p(z) ranges over the set of all 11"n which satisfy the condition

(12.6.2) z = ei8 .

Concerning the special case )1. = 0, J.L = 1, see §11.3.

We write

·--------·---------------------------------·--·-·-··-····-·
30G ASYI\IPTOTIC PROPERTIES OF GENERAL POLYNOMIALS [XII]

where (<Pn(z) l is the orthonormal set associated with f(B). Then I Uv 12 .L::=o
= 1, and with the notation of (11.3.3), according to Cauchy's inequality,

(12.6.3) I Xp(O) + J.Lp(a) 1 = I~ Uv l X<t>v(O) + J.L<t>v(a) l r ~ ~ I X<t>v(O) + J.L<t>v(a) I~

2

= I X 1 Sn(O, 0) + 2ffij>,J.LSn(O, a) l + I J.L 1 Sn(a, a).

2 2

The expression in the right-hand member is the desired maximum.

(2) For clarity we now write sn(f; a, z) instead of sn(a, z) and use an analogous
notation for the orthonormal polynomials <Pn(z) = <Pn(f; z), associated with f(B),
and their highest coefficients Kn = Kn(f). Then the preceding solution of the
maximum problem showR immediately that for the functions /1(8), f(B), /2(8) of
Theorem 10.4.4,
2 2
I X 1 sn(fl ; 0, 0) + 2ffi [XJJ.Sn(fl ; 0, a) l + I J.L 1 sn(fl ; a, a)

(12.6.4) ~ I X 1 sn(f; 0, 0)
2
+ 2ffi ( XJ.LSn(f; 0, a) l + I J.L 1 sn(f; a, a)
2

~ I X~~ Sn(f2 ; 0, 0) + 2ffi(XJ.LSn(f2 ; 0, a) l + I J.L 1 Sn(f2 ; a, a).

2

In particular, we have for arbitrary a

(12.6.5)
Furthermore, we find
2
I Sn(f; 0, a) - Sn(/2 ; 0, a) 1

(12.6.6) ~ ( Sn(f; 0, 0) - sn(f2 ; 0, 0) ll sn(f; a, a) - Sn(/2 ; a, a) l

~ (Sn(fl ; 0, 0) - Sn(/2 ; 0, 0) \( Sn(fl ; a, a) - Sn(/2 ; a, a)\.
Then, by virtue of (11.3.5) and (11.3.6),
2
I Kn(f)<t>! (f; a) - Kn(/2)<1>: (!2 ; a) 1

2 2
~ l[Kn(/1)] - [Kn(/2)] \lsn(fl; a, a) - Sn(/2; a, a)\.
Here <t>! denotes the polynomial reciprocal to <Pn . If Ia I = 1, the same
inequality holds for <l>n , that is

Now, if n is sufficiently large,

(12.6.8)

(see Theorem 11.2). Next, according to (12.3.15),

(12.6.9) !~ Kn(f) = (@(f)l-t = exp {-

4~ 1:" logf(B) dB};
[ 12.6 ] THE ASYMPTOTIC PHOBLEM UNDER LOCAL CONDITIONS 307

and similar relations hold for /I , f 2 • By Theorem 10.4.4,

l @(f1)l-1 - l @(/2)\-1

~ exp {- Lr log f,(B) dD}{exp (2: r [log f,(B) - log f, (8) 1dD) - I}

< l @(f)l-\e• 12" - 1).

Thus

• ....s~p I ¢n (J ; a) - {@(f)}f
I~~ I
@(! ) an lD(f2; a) l-112
2

12
~ (e. " - 1) lim sup l sn(f1; a, a) - sn(/2; a, a) l,
n->oo

or (see Theorem 10.4.4)

lim sup I ¢n(f; a) - an(D(f; a) l- 1 1
2
n->oo
(12.6.10)
< e" + (e• 12" - 1) lim sup l sn(f1; a, a) - sn(f2; a, a) l,

where e" is arbitrarily small with e. This reduces the proof of the statement to
the discussion of the difference
sn(f1 ; a, a) - sn(f2 ; a, a).
(3) Let us use the abbreviation D([2 1 ; z) = h(z). We find from (11.4.5)
and Theorem 11.2 that
. ) _ h({i)h(z) - (az)n+ 1 h(a)n(z- 1)
(12.6.11) Sn (f2, a, Z - _
1 - az

provided n exceeds the degree of lf2(0) l- 1 = g2(0); consequently using !'Hospi-

tal's rule, a = ei"', we obtain

sn(f2; a, a) == (n + 1) I h(a) 1 2
- 2ffi l ah(a)h'(a) l
(12.6.11')
== lf2(a) l- {n
1
+ 1 + 2ffi [a ~(~2 / :;]}.
(4) The discussion of sn(f1 ; a, a) i~ slightly more complicated. From
(10.4.24) it follows that

(12.6.12)

Now let p(z} range over the set of the 11"n satisfying the condition
1 [+" 2 i8
(12.6.13) 11" )_,. /1(0) I p(z) 1 dO = 1, z = e ,
2

which can be written in the form

308 ASYMPTOTIC PROPERTIES OF GENERAL POLYNOMIALS [XII]

(12.6.14)
i8
z = e .
Therefore, by Theorem 11.3.1, if l' = ul/2,
1
Sn+l'(g1 ; a, a) ~max I (1 - i1a)(1 - Z:!a) · · · (1 - iza) I" I p(a) 12
(12.6.15)
= I (1 - i1a)(1 - i2a) · · · (1 - iza) I" sn(fl ; a, a).
Making use of the previous result, we find that
sn(f1; a, a) ~ I (1 - i1a)(l - ~a) · · · (1 - z1a) 1-u
(12.6.16)
{n + l' + 1 + 2ffi [a ~(~;1 /:~]},
1

·g1(a)

provided n + l' exceeds the degree of g1(0). Now from (12.6.12) we have
1
a D'(g1 ; a) = a D'(f1; a) _ ~ -az. ±,
D(g1-I; a) D(f1; a) 2 •=1 1 - z.a'
(12.6.17)
2ffi [a D'(g11; a)] = 2ffi [a D'(f1; a)] - ~l.
D(g1 1 ;a) D(f1;a) 2
Consequently, the important inequality

(12.6.18)

holds provided n is sufficiently large. Since f 1 (a) = /2(a) = f(a), we have

lim sup (sn(f1; a, a) - snU2; a, a)}
n--+oo

(12.6.19) ~ lf(a)l-12ffi {a D'(f1; a) - a D'(/2; a)}

D(f1; a) D(f2; a)
~ 2 lf(a)}-11D'(/I; a)_ D'(f2; a)
D(f1; a) D(f2; a)
I·
But this expression iH arbitrarily small with e, which establishes the proof of
Theorem 12.1.5.
(6) Under the assumptions of Theorem 12.1.6, the function
(12.6.20) f(O) = w(cos 0) j sin 0 I
satisfies the conditions of Theorem 12.1.5 or of Theorem 10.4.1. Indeed, we
have, if x. = cos o.' 0 ~ o. ~ 71', ei " = r.'
8

(12.6.21)
f(o) = 2-T~-T 2 - .. · - q -
1
t( cos o) J (i - 1) IT
v=l
(z - r .r· (z - f .r·[,
z = ei8.,
[ 12.7 1 APPLICATIONS 309

whence
(12.6.22)
We also observe that f(B) is differentiable at B = a and that the ratio (10.4.2)
is bounded near B = a.
As in §12.5, we use Theorem 11.5, particularly the first formula in (11.5.2),
and find that K2n = K2n(f) tends to a positive limit and limn--.oo <f>2n(O) = 0. Thus
Pn(~) = Pn(cosa) = (2/1l·)!p +en}ffi[an(D(f;a)}-
1
], limen= 0.
n--+oo
By use of the function -y(c~) discussed in §12.2 (1) we obtain
ffi(an\D(f:a)}- 1] = \ D(f; a) \- 1 ffi(anei'Y<a>l = lf(a)l-t cos Ina+ -y(a)\.
This establishes Theorem 12.1.6.
The validity of the asymptotic formula (12.1.9) has been extended lately
hy G. Freud (3).

12.7. Applications
(1) By means of Theorem 12.1.2 we can readily derive certain asymptotic
formulas for the highest coefficients of the orthonormal polynomials IPn(x) \.
THEOREM 12.7.1. Let w(x) be a weight function on the interval -1 ~ x ~ +1,
satisfying the conditions of Theorem 12.1.2, and let
(12.7.1) Pn(x) = kn~Xn + kn1Xn- 1 + kn2Xn-2 + ... ' n = 0, 1, 2, ... '
be the associated orthonormal system. Then as n ---t co ,

2~ 1:
1
(12.7.2) kno'""' 1r-l2n exp {- log w(x) (
1
~xx2 )l}'
and
kn1""' -11' -! 2n-1 1+1 X log w(x) ( 1 ~X
-1 X2
)!

(12.7 .3)
·exp
{
-~;
1+1 d }
-1 log w(x) (1 _xx2)i .

Coneerning these formulas, see Shohat 2, p. 577. If the right-hand member

of (12:1.3) vanishes, we read (12.7.3) as follows: limn--.oo 2-nkn1 = 0. There is no
difficulty in deriving corresponding formulas for the later coefficients knv, v fixed.
(See Problems 54, 55, 56.) From (12.7.2) we readily derive an asymptotic
formula for Dn/Dn-1, where Dn·is the determinant of Hankel's type defined
by (2.2.7) (see (2.2.15)). The first asymptotic investigation of these deter-
minants is due to Szego (:L, p. 517). The coefficient kno was denoted by kn in
Chapter II (see (2.2.15)).
The proof follows immediately from (12.1.3). If
D(z) = do + d1z + d2i + · · · ,
we have
310 ASYMPTOTIC PHOPERTIES OF GENERAL POLYNOMIALS [XII]

.
I
n~
2-n (1 + (1 - x-2)!)-n (k
. nO +k nl X
-1
+ kn2 X-2 + · · ')
2

= (27r)-'(do + d1z-1 + · · ·)-1 = (27r )_, (do1 - ~ x--l + ... ) .

2d6
Now we use Titchmarsh 1, p. 95 anrl. (10.2.10). We note that on account of
the mean-value theorem of Gauss,

exp {- 41 1+.. log Isin 0 IdO} =

11" -.- exp { i 1+..
-
4
11" -.- 11 -
i I dO}
log - ---
2

= exp (-~log~) = 2t, z= e .

iO

Combining Theorem 12.7.1 with the formulas (3.2.2), we obtain for the
coefficients An, Cn of the recurrence formula (3.2.1),
(12.7 .4) lim An= 2, lim Cn = 1,
n-+oo n-+oo

provided the conditions of Theorem 12.7 .1 are satisfied. If we compare the

terms with xn-l in (3.2.1), we obtain

(12.7.5)

Thus, under the same conditions,

(12.7.6) lim Bn = 0.
ft-+00

The formulas (12.7.4) and (12.7.6) are noteworthy from the point of view
of a classical theorem of Poincare on recurrence formulas (see Blumenthal 1,
p. 16).
(2) For the further applications we need only the less informative form (12.2.6)
of our asymptotic formula.

THEOREM 12.7.2 (Distribution of the zeros). Let the weight function w(x)
satisfy the conditions of Theorem 12.1.2, and let

(12.7 .7) X1n , X2n , · • • , Xnn

denote the zeros of the orthogonal polynomial Pn(x). Let Xvn = cos Ovn , 0 < Ovn < 1r.
If Ji'(O) is an arbitrary Riemann-integrable function, we have

(12.7.8) lim F(Oln) + F(02n) + · ·· + F(Onn) = 1r-1 { .. F(O) dO.

n-+00 n }o
Using the terminology of Weyl, we say that the values l Ovn l are equally
distributed in the interval [0, 1r] (see P6lya-Szego 1, vol. 1, p. 70). This
result is proved under slightly more restrictive conditions by Szego (1, p. 531).
[ 12.7 1 APPLICATIONS 311

It is rather remarkable that the asymptotic nature of the distribution is inde-

pendent of the weight function. As a consequence of (12.7.8) we obtain the
following result: Let [a, b] be a subinterval of [-1, -j-1], and let a = cos a,
b = cos /3, 1r ~ a > {3 ~ 0. If N = N(n, a, b) denotes the number of the zeros
of Pn(x) in the interval [a, b], we have

(12.7.9) lim N(n, a, b) =a -{3.

n-~ n 71"

Thus the "density" of the zeros Xvn towards the end-points of the interval
[ -1, + 1] becomes large.
For the proof of Theorem 12.7.2 we write (12.2.6) in the following form (see
(12.7.2)):

n-~
= lim n -
n-~
1
t
v=1
1

log \1 - ~
X
\

(12.7.10)
= log 2zx I = 1r
-1 r,. log I1 - coso
}o x 1
dO.
1

Here Iz I > 1. The last formula follows from Gauss's mean-value theorem,
smce
r + r- 1
~ cr - z)(r - z!
1 _
1
= I I

1 z + z- 1 2xz '
Thus,

(12.7.11) !~~ n- ~ ffi {log (1

1
- co:Ovn)} = 1r-
1
1" 0
ffi {log(l -co: )} dO,

so that (see Titchmarsh 1, p. 95)

(12.7.12) !~~ n
. -1
£; (cos Ovn)
n
' k
= 1r
-1
1"
0
k
(cos 0) dO, k = 0, 1, 2 .....

This establishes the statement (see P6lya-Szego 1, loc. cit.).

(3) Further applications are the following theorems:
THEOREM 12.7.3 (Expansion of an analytic function in terms of orthogonal
polynomials). Let the weight function w(x) satisfy the conditions of Theorem
12.1.2, and let IPn(x) l be the associated orthonormal system of polynomials. Let
f(x) be an analytic function regular on the segment [- 1, + 1], and let
f(x) "'foPo(x) + f1P1(x) + · · · + fnPn(x) + · · · ,
(12.7.1:3) +1
f,. =
1 _ f(x)pn(x)w(x) dx,
1
n = 0, 1, 2, · · ·,

be its Fourier expansion. Let R be the sum of the semi-axes of the largest ellipse
with foci at ±1 in the inte1·ior of which f(x) is regular. Then the Fourier expan-
sion (12.7.13) is convergent (with the sum f(x)) in the interior and divergent in
312 ASYMPTOTIC PROPERTIES OF GENERAL POLYN0:\1IALS [XII]

the exterior of this ellipse. The convergence is uniform on every closed set lying
in the interior of the ellipse. Moreover
(12.7.14) lim inf lfn ~- 1 /n = R.
n-+oo

THEOREM 12.7.4. Let the weight function w(x) satisfy the conditions of Theorem
12.1.2, and let
(12.7.15) fopo(x) + f1P1(x) + hP2(x) + · · · + fnPn(x) + · · ·
be an infinite series proceeding in terms of the orthonormal polynomials Pn(x)
associated with w(x). Let
(12.7.16) lim inf lfn ~- 1 /n = R.
n-+oo

Then the series (12.7.15) is convergent in the interior of the ellipse with foci at ±1,
whose semi-axes have the sum R; it is divergent in the exterior of this ellipse. The
series represents an analytic function which is regular in the interior of the ellipse
and has at least one singular point on the ellipse itself. The expansion associated
with this function is identical with (12.7.15).
Thus the "ellipse of regularity" coincides with the "ellipse of convergence";
(12.7.14) and (12.7.16) are the analogues of the Cauchy-Hadamard formula (see
(9.1.4)). Concerning these theorems, see Szego 1, p. 538; 6, p. 193. Theorem
12:1.4 must be modified in an obvious way if R ;:?. 1, or if R = oo.
As a consequence of (12.2.6), we see that the domain of convergence of series
of type (12.7.15) is always an ellipse I z I = canst. To prove Theorem 12.7.3,
we use Theorem 1.3.5. We can find a 7rn-1, say p(x), such that
(12.7.17) I f(x) - p(x) I < M(R-1 + E) n.
Here E > 0 is arbitrarily small, and M = M(E) is independent of n. Thus,
1+1
so that
f,. =
1 _
+1
1
f(x)pn(x)w(x) dx = _ !f(x) -
1
p(x) }pn(x)w(x) dx,

lfn \
2
·~ 2 1 2
M (R- + E) n i: 1
2
(pn(x) } w(x) dx 1-: 1

w(x) dx;

whence lim supn-+oo lfn \11 n ~ R- 1• This furnishes the statement of Theorem
12.7.3 concerning the convergence. Now the last relation must be an equality,
for otherwise, the expansion would be uniformly convergent in the interior of
an ellipse larger than that of the ellipse of regularity, and f(x) would be regu-
lar therein. This argument also furnishes the divergence in the exterior of the
ellipse of regularity.
Theorem 12.7.4 can also be established without difficulty. It is quite re-
markable that the domain of convergence of expansions of type (12.7.13) and
(12.7.15) is independent of the weight function w(x) (compare Theorem 9.1.1).
 CHAPTER XIII

EXPANSION PROBLEMS ASSOCIATED WITH GENERAL

ORTHOGONAL POLYNOMIALS

We shall now prove four theorems of the equiconvergent type in the sense
of Chapter IX (see the introduction). Two of them, Theorems 13.1.2 and
13.1.4, deal with expansions of a preassigned function on a finite segment in
terms of polynomials orthogonal on this segment. The other two Theorems,
13.1.1 and 13.1.3, are concerned with the expansion of the boundary values of
an analytic function, regular in the interior of the unit circle, I z I < 1, in
terms of polynomials orthogonal on the unit circle. In all cases the expansions
in question are compared with trigonometric and power series expansions, and
the weight functions are subject to conditions similar to those in the asymptotic
theorems of Chapter XIL The function developed is very general and merely
satisfies certain integrability conditions.
The basic idea of the method used in the proof of Theorems 13.1.1 and
13.1.2 is due to Szego (see 9 where 'only the case of a segment is considered).
Our present treatment of these theorems is slightly different from and more
general than in Szego 9. The two other theorems are new. It is noteworthy
that no direct use is made of the asymptotic results of the previous chapter.
The methods, however, are very closely related.
After having finished the manuscript, I came into possession of three im-
portant papers of Korous (3, 4, 5).
In 3, Korous deals with the expansion problem of Theorem 13.1.2. His con-
ditions are of "local" character but less restrictive than those of Theorem
13.1.2. Also·, his method. is entirely different from that used by Szegi;i in 9 or
from that of the present treatment.
In 4 and 5, Korous proves two other equiconvergence theorems generallzing
the Laguerre series of Theorem 9.1.5.

13.1. Results and remarks

(1) THEOREM 13.1.1 (Equiconvergence theorem on the unit circle I z I = 1
if the weight function is subject to "local" conditions). Det f( e) be a weight
function on the unit circle which satisfies the conditions of Theorem 12.1.5
( = 10.4.1). Let F(z) be an analytic function regular in I z I < 1 and of the
class H2 (§10.1).
If sn(z) denotes the nth partial sum of the expansion of the boundary values
F(z), I z I = 1, in terms of the orthonormal polynomials (<Pn(z) l associated with
f(O), and if sn(z) is the nth partial sum of the ordinary power series expansion of
F(z), we have
313
314 EXPANSION PROBLEMS FOR GENERAL POLYNOMIALS [XIII l

(13.1.1) lim (sn(a) - Sn(a)} = 0.

n-+oo

Here a eia has the same meaning as in Theorem 12.1.5.

THEOHEM 13.1.2 (Equiconvergence theorem on a finite real segment if the

•
weight function is r,;ubject to "local" uonditions). Let w(x) be a weight functio'n
on the segment [ -1, +1) subject to the conditions of Theorem 12.1.6. Let <I>(x)
be an arbitrary real-valued function, measurable 1·n Lcbesgu,e's sense, for which
the integrals
+1
(13.1.2)

exist.
1 _
1
I<Nx:) )2 w(x) dx,

If sn(x) and s:(x) denote the nth partial sums of the expansions of <I>(x) in terms
of the orthonormal polynomials \Pn(x) l associated with w(x) and of the Tchebichef
polynomials I cos ne l, cos e = x, respectively, we have

(13.1.3) lim (sn(~)- Sn(~)\ = 0.

n-+oo

Here -1 < ~ < + 1, and ~ has the same meaning as in Theorem 12.1.6.

It is remarkable that in both cases a wide class of expansions displq,y the

same convergence behavior. The expansion into a series of cos ne occ~1rring
in Theorem 13.1.2, is, of course, the ordinary cosine expansion of <I>(cos fJ). A
comparison of sn(~) with other special expansions can also be readily made.
In making such comparisons, existence of certain other integrals must be re-
quired. By the use of theorems thus obtained, the customary convdgence
and summability theorems of the classical Fourier expansions can be 1easily
extended to the general expansions in question. Theorem 13:1.2 can be 1easily
extended to an arbitrary finite segment instead of the seg:rp.ent [-1, +1].
·These theorems hold under the conditions of "local" character of Theorems
12.1.5 and 12.1.6.
(2) The following theorems correspond to the conditions of "S. Bernstein's
type" occurring in Theorems 12.1.3 and 12.1.4.

THEOREM 13.1.3 (Equiconvergence theorem of S. Bernstein's type on the

0
unit circle). Letf(O) be a weight function on the unit circle z = ei which satisfies
the conditions of Theorem 12.1.3 with ;>.. > 1. Let F(z) be an analytic fu~ction,
regular and bounded for I z I < 1. Employing the same notation as in Theorem
13.1.1, we have

(13.1.4) lim (sn(z) - Sn(z) l = 0, lz 1: ~ 1,

n-->oo

uniformly in the whole closed unit circle Iz I ~ 1.

0
The function F(z) has integrable boundary values F(ei ) (for almost all: 0).
 [13.2] :YIAXI:VIU:W PROBLEM 0:\ UXIT CIRCLE 31.5

THEOREM 13.1.4 (Equieonvergence theorem of S. Bernstein's type on a finite

real segment). Let w(x) be a weight function on the interval -1 ~ x ~ + 1,
x = cos e, which satisfies the conditions of Theorem 12.1.4 with A > 1. Let
<P(x) be an arbitrary bounded function, which is meas'IJi"able in Lebesgue's sense.
Employing the same notation as in Theorem 13.1.2, we have
(13.1.5) lim (Sn(x) - Sn(x) l = 0, -1 <X< +1,
n-+oo

uniformly in the interval [ -1 + E1 1 - E], 0 < E < !.

In the proofs of Theorems 13.1.3 and 13.1.4 essential use is made of Theorems
12.1.3 and 12.1.4, respectively.

13.2. A maximum problem on the unit circle

(1) Let f(e) and F(z) have the same meaning as in: Theorem 13.1.1. We
write G(e) = (27r)- 1f(e)F(ei 0), so that

(13.2.1) J_:r IG(e) Ide

is convergent. In what follows we may vary f(e), but G(e) is supposed to'be a
fixed function for which (13.2.1) exists.
PROBLEM. Let A and f..l· be arbitrary complex numbers, and let Ia I = 1. We
intend to determine the maximum of

i8
(13.2.2) z ==:= e ,

for p(z) ranging over the set of all 7rn which satisfy the condition

2~ 1-:r j(e)
2
(13.2.3) I p(z) 1 de = 1,

Concerning the special ease A = 1, f..l. = 0, see §11.3. Compare also with §12.6
(1). We write again

(13.2.4) p(z) = Uo</>o(z) + Urcf>r(z) + · · · + Uncf>n(z),

where lct>n(z) l has the usual meaning. Then L:'..o I u. 12 = 1, and

Ap(a)
I + JJ. 1 +.-
-• {i(e)p(z) de! =
12 J

11
?; u. Acf>v(a) +
n {
JJ.
l+r
_., G(e)¢.(z) de
} !
2

(13.2.5) +..
I ·
n
2

~ ~I Acp.(a) + JJ. 1-r G(e)cp.(z) de

z=~:e.
i8

Here, as in Theorem 13.1..1, sn(a) is the nth partial sum of the expansion. of
316 EXPANSION PROBLEMS FOR GENERAL POLYNOMIALS ; [XIII]

1
21rG( e) !f( e) )- in terms of the polynomials lct>n(z) l, and

(13.2.6) Hn = ~
n 11+r
_,.. G(e)q,.(z) de 12 1
. i8
z = e .

The right-hand member of (13.2.5) is the maximum in question.

(2) For clearness, as in §12.6 (2), we now write sn(f; a, z), sn(f; a), ¢n(f; a),
H nCf) for the expressions involved in the previous considerations, wh~ch are
associated with the weight function j(e). We note that sn(f; a) is tihe nth
partial sum of the expansion of 21rG(e) If( e) }- 1 in terms of the polyriwmials
ct>nU; z) associated with j(e). \Ve add that in all subsequent conside~ations,
even if j(e) is replaced by some other weight function, we shall keep th~e func-
tion G(e) usPd before.
Applying Theorem 10.4.4 again, we obtain
2 2
I A l snCfr; a, a) +
2ffi{A.Usn(fr; a)} +
I JJ.I HnCfr)
2
(13.2.7) ~ I A l sn(f; a, a)
2
+
2ffi {A.Usn(f; a)} I JJ. I H ;..(f) + I
2
~ I A l sn(f2; a, a) +
2ffi{A.Usn(h; a)} 2
I JJ.I Hn(f2). +
In particular, we have
(13.2.8) snCfr ; a, a) ~ sn(f; a, a) ~ sn(f2 ; a, a),
(13.2.9)
2
I snCfi a) - Sn(f2 i a) 1 ~ (H nCf) - H n(f2) }{sn(f; a, a) - Sn(f2 ; a, a)}
(13.2.10) .
~ Hn(f){snCfr;a,a)- snU2;a,a)}.

13.3. Proof of Theorem 13.1.1

(1) Bessel's inequality enables us to obtain

(13.3.1)

Whence (12.6.19) shows that

2
lim sup I sn(f; a) - Sn(h; a) 1
n-+oo

(13.3.2)
~ 2
1
71" L:" \ 0 2
F(ei ) 1 j(e) de·li~-s~p {sn(f1; a, a) - sn(h; a, a)\

~ I
__!_ [+r I F(eio) 12 j(e) de. 2 {!(a) ~-1 D'(f1 i a) - D'(f2 i a)
21r )_, D(f1; a) D(h; a)
I·
The right-hand member is arbitrarily small with the positive number E occ~rring
in Theorem 10.4.4. ·.
(2) Next we discuss the behavior of the partial sum sn(f2 ; a) as n H oo.
1
As in §12.6 (3), we use the abbreviation D({; ; z) = h(z). We find thalt if n
 [ 13.3 J :PROOF OF THEOREM 13.1.1 :317

is sufficiently large (see (12.6.11)),

Sn(f2; a) = r+r G(61)sn(f2; z, a) de = [+r G(e) ~(a)~ - (az) ~+I h(a)h(:z) de

(13.3.3)
J-1r )_, 1 - az
= J:..l+•·j(e)F( ) h(a)h(z) - (az)n+r h(a)h(z) d = ,6
2r -7r z 1 - ~
- e, z e .
Comparing this latter expression with

(13.3.4) Sn(a) = -
1 1+7r F(z) 1 ~ ( -)n+I
az _ de, z = etO
-1r -
1

2r az I

we obtain

sn(f2; a) - Sn(a) ==
1
2r
j+r
-1r
F(z) j(e)h(a)hG)_ -
az 1-
1 de

(13.3.5) +7r -
_ __!__ ( F(z)j(e)h(a)h(z)_- 1 (az)n+r de, z =:. e•o.
2r J-1r 1 - az
The second integral approaches zero as n ---+ oo, since
j(e)h(a)h(z) - 1
(13.3.6) i
z =:e ,
i()
1 - az I

has a limit at e= a. The first integral can be written as follows:

(13.3.7)
I ·o
z ='e'.

since (see the remark at the end of §10.1)

(13.3.8) ( F(z) 1 - h(a)/h(z) dz = 0.

Jlzl-1 a- z
By virtue of Schwarz's inequality,
2 2
4r lim sup I sn(f2; a) - Sn(a) 1

r r
n-+oo

~ IF(z) I' d9 I~' (9),~ ~9~' lf<aW' IJ,(9W' d9

sm - -
2

~ (4f(a)}-
1
1:1r I F(z) 1•zde 1-:1r f2 :e~;!~de
sm - -
2

~ {4f(a) }- maxf2(e) [:" jF(z)j 2de

1
1:7rlogf2 _(e~e-~o!j(e) de, z=e.
I i(}

sm - -
2

·--·-·--··- ······
 318 EXPANSION PROBLEMS FOR GENERAL POLYNOMIALS [XIII]

Combining this result with (13.3.2), we find lim SUPn-+oo I sn(f; a) ~ Sn(a) I
to be arbitrarily small with the positive number E occurring in Theorem 10.4.4.
This establishes the statement.

13 . 4. A special case of Theorem 13.1.2

(1) In this section we consider the special case
(13.4.1)
+
where p(x) is a 7rz which is positive in the interval [ -1, 1]. The corresponding
orthogonal polynomials (for 2n > l) have been calculated in §2.6. \fe shall
verify the validity of Theorem 13.1.2 in this special case. To this: end it
suffices to assume the existence of the integral
+1
( 13.4.2) _
1
I cfs(x) I (1 - i)-t dx,
1

a condition which is more general than that required in Theorem 13.1.2. As

in that theorem we have ~ = cos a, -1 < ~ < +1, 0 < a < 1r.
By virtue of the Ch:ristoffel-Darboux formula (3.2.3)

Sn(~) = kn
kn+1
1+1 <P(x) Pn+r(x)pn(~) -
-1 X-~
Pn(X)Pn+1(~) w(x) dx;

or

== (" <P(cos O) Pn+r(cos e)pn(cos a) - Pn(cos e)Pn+t(cos a)

(13.4.3)
}o cos e- cos a
·(p(cos e)} - 1de.
Assume now 2n > l. Use formuias (2.6.2) and write (2/7r)!ein 8 h(ei8 ) ~ un(e)
+ ivn(e). Then
Pn(cos e) = Un(e),
(13.4.4) "8 .
Pn+1 (cos e) = 9?(e' [un(e) + ivn(e)]j = Un(e) cos e- Vn(e) sin ~I
so that
Pn+l(cos e)pn(cos a) - Pn(cos e)PnH(cos a)
cos e- cos a
= (un(e) COS e- Vn(e) Sin e}un(a)- Un(e) (un(a) COS a- Vn(a1 Sin a}
cos e- cos a .
(13.4.5)
' sine - sin a
= Un(e)un(a) - Vn(e)un(a) - - - -
cos e- cos a
Un(e)vn(a) - Vn(e)un(a~ .
+ - - -cos- e-
- - - - - + sm a.
cos a :

·-------,--------------r------·------------·----~------------------~
 [ 13.4 l A SPECIAL CASE OF THEOREM 13.1.2 319

In view of Riemann's lemma, this leads to

kn+l ( )
kn Sn COS a

(13.4.6)
= sin a (" <I>( cos e) Un(e)vn(a) - Vn(e)un(a) ( (cos e)} - 1de + o(1)
}o COS e- COS a p I

Now
1
(p(cos e)}- (un(e)vn(a)- Vn(e)un(a)}
- -(p(cos e)}- 3(un(e)
1
+ ivn(e)}lun(a)- ivn(a)}
(13.4. 7) - ~ (p(cos e)} - 131 ein(B-a) h(ei8)h(eia)}
7r

__ ~ ()( { in(8-a) h(eia)}

7r '-' e h( e•"B) .
Repeated application of Riemann's lemma shows that h(e'.a) (h(ei 8) }-1 can be
replaced by 1, and
.
sma(cose-cosa)
-1 .
+- s m --
=Sma {2sm-
. a e. a- e -1

2 2 }
1
by (2 sin (a- e)/2}- • Thus the right-hand member of (13.4.6) assumes the
form

! ( r <I>( cos e) sin n(e - a) de + o(1)

7r }o . e- a
sm--
2

= ! 1+" <l>(cos e) sin (n + ~)(e - a) de+ o(1).

7r -'II e- a
•
sm--
2
Finally kn/kn+1 = ! when n is sufficiently large (see (2.6.5)). This establishes
the statement.
(2) In the same special case we intend to calculate Kn(~, 0, where Kn is the
kernel polynomial defined by (3.1.9). We have, by (3.2.4),

(13.4.8)

Now, by (13.4.4), for a su1ficiently large n,

sin a p:(cos a) := -u~(a),
(13.4. 9) sin a P:+1( cos a)
= -u:(a) cos a + v:(a) sin a+ Un(a) sin a + Vn(a) cos a,
320 EXPANSIOK PROBLEMS FOR GENERAL POLY:r\OMIALS [XIII]

and kn/kn+1 = !. Thus,

2 sin a Kn(cos a, cos a) = (-u~(a) cos a+ v~(a) sin a

+ Un(a) sin a + Vn(a) cos a)un(a)

+ u~(a)iun(a) cos a- vn(a) sin a)
= sin a 3i[un(a) - ivn(a)][u:(a) + iv~(a)Jl
+ Un(a)3{[un(a) + ivn(a)]eiaj;
hence

2 sin a
+ Wn
. (a ) l2 Kn ( ) . + iv:(a)}
{u:(a)
I_(_)_
Un a
cos a, cos a = sm a3 ( )
. (
+ Wn
Un a a)

s·
+ !. ~m + {[un(a) + ivn(a)] ,·a}
l0\
2
2
a Un(a) + ivn(a) e ·
h) I 12

. ia h'(eia)J [ h'(a)J
= n - 3 [ ~e h(ei<>) = n - 9? a h(a) '
we get the important formula

(13.4.10)
1r(p(~) )- Kn(~, ~)
1
= n +! - 9? [a r(;1 J
·
+ (z sm a
)-10\ [
-0 a
2n+l h(a)J
h(a) ,
t
, = em; a, a =
ia
e , 0< a < 1r.

This holds for sufficiently large values of n.

13.5. Preliminaries for the proof of Theorem 13.1.2

(1) Let w(x) and cf>(;c) have the same meaning as in Theorem 13.1.2. Write
G(x) = w(x)<P(x), so that the integral
[+1
(13.5.1) J_ IG(x) Idx
1

is convergent. In subsequent considerations we may vary w(x), but G(x) is

supposed to be a fixed function for which the integral (13.5.1) exists.
(2) In what follows we apply Theorem 10.4.5 with j(e) = w(cos e) I sin e I·
This function satisfies the conditions of the theorem mentioned (see §12.6 (5)).
We define the functions w1 (x) and w2(x) by
(13.5.2) j.(e) = w.(cos e) I sin e I, v = 1, 2.
Obviously, ~ = cos a, 0 < a < 1r,
(13.5.3) 0 ;£ w1(x) ~ w(x) ;£ w2(x),

·--.--,---------------..,._·--·-----·-·····
[ 13.5 I PRELIMI:KARIES FOR THE PROOF OF THEOREM 13.1.2 321

Moreover, w2(x) is of the form (13.4.1), while w1(x) = k(x)u(x) = (r(x) }2u(x).
Here u(x) is of the form (13.4.1), and r(x) is a polynomial of degn~e al/4. (See
the remark to Theorem Jl0.4.4.)

(3) PROBLEM. Let A and JJ. be arbitrary complex numbers, and assume that
- 1 < ~ < 1. + We intend to determine the maximum of

(13.5.4) IAp(~) + }J. £: 1

G(x)p(x) dx r
when p(x) ranges over the set of all 'Trn which satisfy the condition
+1
(13.5.5)
1
_
1
I p(x)
2
1 w(x) dx = 1.

This is the problem corresponding to that of §13.2. We substitute again

(13.5.6) p(x) = ·uopo(x) + Urp1(x) + · ·· + UnPn(x),

where lPn(x) l is the orthonormal set ~f polynomials associated with w(x).
Then L.:::=o I u. 1 = 1, and according to Cauchy's inequality
2

(13.5.7)
~ ~ I Ap.(~) + JJ.1:1 G(x)p.(x) dx 12
= I A 1 Kn(~)
2
+ 2ffi[XJ,J.Sn(~)] + IJJ.I 2Hn.
Here Kn(~) = Kn(w; 0 has been written for the "kernel" Kn(~, 0, and

(13.5.8) Sn(~) = sn(w; ~) = ~ p.(~) £: 1

G(x)p.(x) dx = £: G(x)Kn(~,
1
x) dx,

~ {1:
1
(13.5.9) Hn = Hn(w; ~) = G(x)p.(x) dx Y.
We notice that sn( ~) is the nth partial sum of the expansion of <I>(x) =
Iw(x) }-1G(x) in terms of the polynomials Pn(x) associated with w(x).
The right-hand member of (13.5.7) is the maximum in question. Hence
2
i A 1 Kn(W1; 0 + 2ffi[XJJ.Sn(W1;+ I JJ.I 2Hn(w1) ~)]
2 2
(13.5.10) ~ I A Kn(w; ~) + 2ffi[XJJ.Sn(w; ~)] + I JJ.I Hn(w)
1

2 2
~ I A Kn(W2 ; 0 + 2ffi [XJJ.Sn(W2 ; ~)] + \ JJ.i H n(w2).
1
 322 EXPANSION PROBLEMS FOR GENERAL POLYNOMIALS [XIII]

In particular
(13.5.11) Kn(w1 ; ~) ;;;; Kn(w; ~) ;;;; Kn(w2 ; ~),
(13.5.12) H n(w1) ;;;; H n(w) ;;;; H n(w2).
Furthermore,

Sn(w; ~) - Sn(W2; ~) ~ (Hn(w) - Hn(w2)) (Kn(w; ~) - Kn(W2; ~))

2
I 1

(13.5.13)

if we observe Bessel's inequality. We next show that the last difference is

arbitrarily small with f, uniformly in n.

13.6. Pr.oof of Theorem 13.1.2

The main tool of this proof is the special equiconvergence theorem of
§13.4 (I) and the representation (13.4.10) of the kernel. The latter yields
immediately the representation

7rf(a)Kn(w2; ~) = n+ ~ + m[a ~(~2 : :;]

(13.6.1)

+ (2 sm
.
a
)-10< [ 2n+1
'-' a
D(f2; a)
D(f2; a)
J,
1
since (g2(a) l- = h(a) = f(a). We derive, on the other hand, an inequality
for Kn(w1 ; ~) analogous to that in (12.6.18). Using the notation introduced
in §13.5 (2) and writing ~al = l', we obtain as in §12.6 (4),
Kn+l'/2(u; ~) ;;;; (r(~) ) Kn(W1; ~) = k(~)Kn(W1; ~).
2
(13.6.2)

(13.6.3)

+ (2 Sin
. )-1()( [ 2n+Z'+1 D(g1 ; a)
a '-' a ( )
1
J,
D g11 ;
a
where, as before, ~ = cos a, a = eia.. Now we can use (12.6.17). Moreover
from (12.6.12)

(13.6.4)

---,---·---·-------r-----
 [ 13.7 l PROOF OF THEOREM 13.1.3 323

On comparing this with (13.6.1), we have

Kn(w1; ~) - Kn(w2; ~)

~ (1rf(a)} -t {m [a. D'(f1; a) _ a D'(f2; a)]

D(h; a) D(f2; a)

_!!_(~)__]}
(13.6.6) + (2 sm
·
a
)-10< [
'"' a
2n+1 D(JI; a)
D(f1; a)
_ a2n+1
D(f2; a)

~ (1rj(a)}-t1D'(f1; a)_ D'(f2; a) I

D(f1; a) D(f2; a)

+ (27rf(a) sm a}
• _1 I D(JI; a)
-D~)
_ D(h; a)
D(h; a)
I·
The left-hand side is non-negative, while the right-hand ::;ide is arbitrarily small
with E.
Returning to (13.5.13), we notice that sn(w2 ; ~)is the nth partial sum of the
1
r-
expansion of !w2(x) }- G(.x) = lw2(x) 1w(x)<I>(x) in terms of the polynomials
associated with w2(x). Since I w2(x) }-1w(x) ~ 1, and since the second integral
in (13.1.2) exists, the result of §13.4 (1) can be applied. Thus sn(w2 ; ~) =
sn(w2 ; cos a) can be replaced by the nth partial sum of the corresponding
Fourier expansion with an error o(1), n ~ oo. Also ( w2(x) }- 1w(x) = 1 for
x = ~~ so that the partial sum can be replaced by that of <I>(x), which is sn(~).
This completes the proof of Theorem 13.1.2.
REMARK. It is easy to show that the difference

(13.6.7)

occurring in (13.5.13), is also arbitrarily small with E. If this fact were used,
it would suffice to obtain an upper bound for the first difference in (13.6.6).
This remark furnishes a slight variation of the preceding argument.

1:3:.7. Proof of Theorem 13.1.3

(1) It is sufficient to discuss the statement for Iz I = 1. The partial sums
in question are

(13.7.1) -
1 1+,.. F(t)sn(S', z)f(x) dx and -
1 1+,.. F(s) 1 - (fzt+ 1 dx,
_ - s
·
= e'%,
27r _,.. 27r _,.. 1 rz

- - ,- - - ·---.....,..-----------------·---------·--- ......
 324 EXPANSION PROBLEMS FOR GENERAL POLYNOMIALS [XIII]

where sn(s, z) has the usual meaning. We introduce the difference of the
"kernels"

(I3.7.2) I - (fz)"H
Lln(s, z) = sn(s, z)f(x) - _ ,
I - rz
and show that

(I3.7.3) !~ )_"[+>< F(s)lln(s, z) dx = o, r= ix

e :
uniformly on the unit circle I z I = 1.
(2) First, as a consequence of the Lipschitz-Dini condition (10.3.10) assumed
for j(e), we prove that a similar condition (with the exponent -X instead of
-I - X) holds for D(f; z) = D(z); that is,
(13.7.4) I D(ei<BH)) - D(e' 8) I < L' /logo 1-\
where L' is a positive constant. Let m be an arbitrary integer. By use of
(10.3.I2) we obtain
jD(ei<BH))- D(e;e) I< 2Q(logm)-A + /lh(ei(BH))l-1- \h(e;e)l-1/.
The last term, according to Theorem I.22.2 is equal to oO(m); whence the
bound O[(Iog m)-A) +oO(m) follows. Putting m "-' o- 1 I log 0 1-A, \Ve have
(log m)-A rv I log o ,-~, and om ,....__,I log o 1-A, which establishes the statement.
This furnishes, for the function 'Y(e) = 31log D(ei8 ) l defined by (12.1.7), also
the relation
(13.7.5) I 'Y(e + o) - 'Y(e) I < L" /log o 1-\
where L" is a po~itive com;tant. TheHe results hold for an arbitrary X > 0.
By the same argument the following more general inequality follows:
(I3.7.6) I D(z1) - D(z2) I< L' I log I z1 - z2jj-A,
where z1 and z2 are arbitrary in the unit circle I z I ~ 1.
(3) Let I z I = I r I == I, z ~ r.
From (I1.4.5) and Theorem I2.1.3 it is seen
that

An(s, z) = ID(f)D(z) l ~1 -I (fz)_"H(D(r).D(Z) l-1 I D(s) 12

- rz

- I - (fz_)"H + /I - fz l-1O[(log n)-A]

(I3.7.7) I - rz
= D(() (D(z) ~- - I _ (fz)nH D(f) (D(z) l_-1 - I
1

I - rz I - rz
+ II - fz /-1O[(log n)-A].
Now, according to Cauchy'i-i theorem,
[ 13.7 l PROOF OF THEOREM 13.1.3 325

(13.7.8)

since the integrand is a function of r which i:;; of the class li 1 (see the remark
at the end of §10.1). Here we u~ed the condition X > 1. And by Riemann's
lemma,

r
(13.7.9) ~ = eix .

(Concerning the uniformity cf. below.) Therefore we obtain, if e is an arbi-

trary positive number, E' = E(z, n, e) is the set on I r I = 1 with I r - z I ~
en-\ and E' the complementary set, the relations

1 F(s)An(S,z)dx = -1 E'
F(s)fD(s)(D(z)i_
l 1 - tz
1
_ ~- (fzr+ 1 D0(15(z)~- -l}dx
1 - SZ
1

+ o(l) + O( (log n)-'l 111 - fz l-1 dx

= 0(1) ,[,II - fz l-1 I log I z - r II-A dx

+ o(l) + O( (log n)-Al 111- fz l- 1 dx

= O((logn) 1-Al + o(l) + O((logn)-AlO(logn)
= o(l),
On the other hand, the numerator of (11.4.5) is bounded in the present case
so that Theorem 1.22.2 yields (see §12.4 (2)) sn(s, z) = O(n), and hence also,
An(s, z) = O(n). This holds uniformly for I z I = I r I = 1. Therefore, we
have

1E'
F(s)An(s, z) dx = O(n)en- 1 = eO(l), ,.
~
ix
= e '

where 0(1) is independent of e and n. This establishes the statement.

(4) The uniform validity of (13.7.9) needs some explanation. According to
(13.7.4) (see the remark at the end of (2)),

QiiliPCz)l~=-_I = K(s)
1- rz
s
is a function of which is of the class H 1 • If Pn(r) denotes the nth Cesaro
mean of the partial sumE: of the power series expansion of F(r), we have
 32G EXPAI'\SION PHOBLE;\IS FOR GEJ'\EHAL POLYNOMIALS [XIII]

where E1 is the set on I r I = 1 with I r - zI ~ e, and E~ the complementary

set. The integral o,·er 1!.\ as n ------7 cx:; is

0(1) 1I1
F(s) - Pn(S") Idx = 0(1) 1-:1r IF(s) - Pn(S") I dx

= o(l).

The integral over E~ (~:ince F(r) - Pn(r) is uniformly bounded) is equal to

0(1) r' IK(f) Idx,

JEt
that is, it is arbitrarily small with e.

13.8. Proof of Theorem 13.1.4

(1) We show that
+1
(13.8.1) !~;
1_ <I>(~)A: (~, x) (1 -
1
~ 2) -! d~ = 0,

uniformly in -1 + e ~ x ~ 1 - e; here A:(~, x) is the difference of the

"kernels"; that is, according to Christoffel's formula

A:(~, x) = (1 _ ~2); w(~) ~ Pn+1(x)pn(~) - Pn(x)pn+t(~)

kn+l X- ~

(13.8.2) . 1 Jsin (2n + 1) ~ sin (2n + 1) 0~1

- 21r il . e + ct> + . e - ct> '
sm - 2- sm - - )
2
~ = cos ct>, x = cos e.
The symbol kn is used for the highest coefficient of Pn(x).
(2) We first notice that from (11.5.2) for n ~ 1

kn = (27r)-1{1 + c/> 2n(O)}-! 2n(K2n + c/>2n(O) l

K2n
(13.8.3)
= (27r)-!{1 + c/>2n(O)}! 2n K2n
K2n
follows so that on account of formulas (12.4.3) and (12.5.3) (see also Theorem
12.7.1)

(13.8.4)

From (12.1.8) we conclude, using the notation (12.2.2), that

 [ 13.8 l PROOF OF THEORE:\I 13.1.4 327

A:(~, x) = A:(cos<t>, cos,9)

1
= 1r- (x- ~)- (W(x) l-!(W(~) l!(cos [(n + 1)e + -y(e)] cos [n<t> + -y(</>)]
1

- cos [ne + -y(e)] cos [(n + 1)</> + -y(<t>)Jl

(13.8.5) ! + ~ + sin
_ __!_ sin (2n
l . +e
I) (2n
. <1>-
+ I) 'L~ !)
21r <1>
sm- - sm -
-e
2 2
+ Ix - ~ l-1O[(log n)-A]
= An(</>, e) - Bn(</>, e) + I X - ~ 1-! O[(log nr-AJ.
The first term can be written in the form

An (</>, e) . <I>+ --e sm

= (21r ) -1{2 sm- . <I>- f r cos e)l-!IW(
-- - e')-11W( l cos .<1> )l!
2 2
·(cos [n<t> + (n + 1)e + -y(<t>) + -y(e)] +cos [n<t>- (n + 1)e + -y(<t>)- -y(e)]
-cos [(n + 1)</> + ne + -y(</>) + -y(e)]- cos [(n + 1)</>- ne + -y(<t>)- -y(e)]l
1
= {21rsin <I>~ e sin~> e}- (W(cose)l-t(W(cos<t>)l!
2

·{sin [ (2n + 1) ~~ ~ e + J 2e
-y(<t>) + -y(e) sin 4>_

(13.8.6)
+sin [ (2n + 1) <P-- - e+ -y(<t>) - -y(e) Jsin - +- e}
<1>
2 2

1 {sin [ (2n + 1) <I> ~

e + -y(</>) + -y(e) J
= 71" (W(cos e) l-t (W(cos </>) l; -· . <I>+ e
2
sm--
2

+ si_n[ (2n +~ ~~ + ~(q,) -~(e) J}·

_ _e
sm "~',~,.-
2
(3) The integrals
:,.. IW(cos e) l-! I W(cos </>) l! cos [-y(<t>) ± -y(e) l ---:__!
[ -.. . <I> ± e d<t>,
sm--
2
(13.8.7)
sin [-y(<t>) ± -y(e)]
. <t>±e
sm--
2

--,-----
 328 EXPAXSIO?\ PROBLEMS FOR GENERAL POLYNOMIALS [XIII l

exist (see (12.2.4), (12.11.4), and (13.7.5)), so that by Riemann's lemma,

(13.8.8) ~i!:!c I:·· <P(cos cf>) IAn(c/>, e) - Bn(cf>, e) l dcp = 0,

uniformly in e. Let 1J bP an arbitrary pof;itive number, E = E(e, n, TJ) the

set I cos cf> - cos e I ;;;: r1n~\ and E'' the complementary set. We have

1-:71' cf>(cos c/>)A~(cos cf>, cos e) dcp

~ J. <!>(cos .P) IA" (.p ,8) - H"(.P, 8) Id.p + 0[ (log n) _,1 /.1 cos .p - cos 8[-·'d.p

(13.8.9) +(
)E'
<P(cos ¢)A~(cos cf>, co~ e) dcp

= -1, <P(cos cf>) IAn(c/>, e) - Bn(cf>, e) l dcf> + o(1)

+ O[(logn)~:~Jllcoscp- cosel- 1
dcf> + ( <i>(coscp)A~(cos¢, cose)dc/>.
E )E'
Now

I An (c/>, e) I ;;£ -1 {max W(x)}! { sin I'Y(c/>) + 'Y(e)}

1r min W(x) 2n + 1+
2 sin cf> + e
2

(13.8.10) + ~~J.!(cf>) - 'Y(e) l }

. ¢-
2sm-
e
2
= O(n) + 0(1) Icf> - e l- I log Icf>
1
- e II--A,
IBn(cf>, e) I ~ 7r--
1
(2n + 1)'

so that the first term in the right-hand member of (I :~.8.9) is

n ~ oo,

where the bound of 0(1) in the terrn TJ0(1) is independent of 'YJ· The third
term is O[(log n)-"]O(Iog n). Finally, in the last term we apply Theorem
1.22.3. Since the polynomial Pnn(x)pn(~) - Pn(X)Pn-tl(~) is uniformly bounded,
-1 ~ x ~ +1, -1 ;;£ ~ ~ +1, we have A~(~, x) = O(n), uniformly in x and
~' confined to an interval of the form [ -1 + e, 1 - e). Thus, for the term
in question, the hound 1Jr,~~ 0(n) = TJO(l) ean he obtained. Here 0(1) is again
1

independent of 'YJ· This completes the proof of Theorem 13.1.4.

 ,_ _ _ _ _ ____L _ __

CHAPTER XIV

INTERPOLATION
In this chapter we shall consider certain problems of interpolation related to
the theory of orthogonal polynomials. In particular, we are interested in inter-
polations whose abscissas are the zeros of the orthogonal polynomials Pn(x)
associated with a given distribution of the type da(x) or w(x)dx. We dear with
the ordinary Lagrange polynomials and with the "step polynomials" (Treppen-
polynome) introduced by Fejer. This topic is naturally very closely related
to that of the next chapter on mechanical quadrature.
Concerning the subject matter of this and of the next chapter, see the
recent monograph of Feldheim (4).

l.4.1. Definitions; problems

(1) Let [a, b] be a finite or infinite interval, and let
(14.1.1) X1n ~ a, Xnn ~ b
denote a set of n distinct points of this interval. Let l(x) be a 7rn, not iden-
tically zero, vanishing at x: = Xvn, v = 1, 2, · · · , n; it is determined save for a
constant nonzero factor. When there is no ambiguity, we shall write xv instead
of Xvn . The polynomials
l(x)
(14.1.2) lv(x) = l'(xv)(x - Xv)' v = 1, 2, · · ·, n,

are called the fundamental polynomials of the Lagrange interpolation cor-

responding to the set Sn. They have the property
(14.1.3) v, J.L = 1, 2, ... , n.
Let !1, h, · · · , fn be arbitrary values. Then the expression
n
(14.1.4) Ln(X) = L
v=1
fvlv(x)

represents the uniquely determin8d 7rn-1 which assumes the value fv at x = Xv.
This is the nth Lagrange polynomial corresponding to the abscissas Sn . It is
readily seen that
(14.1.5) l1(x) + l2(x) + ··· + ln(x) = 1.
(2) Now let Sn, n = 1, 2, 3, ·. · , be a sequence of sets of abscissas. If f(x)
is a given function defined in [a, b], we can consider the sequence of the cor-
responding Lagrange polynomials Ln(x), n = 1, 2, 3, . · · , defined by (14.1.4)
withfv = f(xvn). Various convergence and divergence properties of this sequence
have been studied under proper conditions of continuity concerning f(x). In
329
330 INTERPOLATION [XIV]

what follows we shall be interested exclusively in the case in which the abscissas
Sn are the zeros of the orthogonal polynomials associated with a preassigned
distribution. Different types of "convergence" can then be considered; for
instance:
(a) ordinary convergence: limn-oo Ln(x) = f(x);
(b) convergence in mean: limn-oo f~ I Ln(x) - f(x) 1 dx = 0;
2

(c) generalized convergence in mean, arising from (b) by replacing the ex-
ponent 2 by p, where pis positive;
(d) quadrature convergence: limn_oo f~ ILn(x) - f(x) )dx = 0.
The last type is partieularly important from the point of view of Chapter XV.
Of course, a certain flxed weight function can be introduced into the integral
conditions.
(3) Let a and b be finite. If only the continuity of f(x) is assumed, the
behavior of the Lagrange polynomials is rather irregular. Faber (2; see also
Fejer 11, pp. 450-453: and Marcinkiewicz 1) proved that for a given arbitrary
sequence I Sn l there exists a continuous function f(x) such that the sequence
ILn(x) l is not uniformly convergent. S. Bernstein (4) has even proved the
existence of a continuous function for which Ln(x) is unbounded at a preassigned
point Xo. According to Helly's theorem (§1.6), this is equivalent to the un-
boundedness of the sequence of "Lebesgue constants"
n
(14.1.6) L
v=l
llv(Xo) I, as n --7 oo.

In the case of the special sequence

rr
Xvn = <·os(2v - I) n, v = 1, 2, · · · , n,
2
a = -1, b = +1, that is, for the zeros of Tn(x), much more is known. GrUn-
wald (1) and Marcinkiewicz (2) proved the existence of a continuous function
f(x) for which the sequence of Lagrange polynomials corresponding to these
Xvn is everywhere divergent, even everywhere unbounded.
(4) In order to obtain convergent sequences of interpolation polynomials,
it is necessary to introduce additional restrictions concerning either: (a) the
functicm f(x), particularly, restrictions concerning its modulus of continuity
(see Theorem 1.3.2), or (b) the interpolation polynomial, such as conditions
concerning its derivative, and the like.
We introduce the polynomials

l"(xv) } 2
hv(x) = { 1 - l'(xv) (x - Xv) llv(x))
(14.1.7)
= Vv(x) llv(x) )2 ,
Qv(x) = (x - Xv) !lv(x) )2,
called the fundamental polynomials of the first and second kind of the "Hermite
[ 14.1 l DEFINITIONS; PROBLEMS

interpolation" corresponding to the Ret Sn . These 7rzn-l are completely deter-

mined by the conditions
hv(xl') = Ov~> , h~(x") = 0; f)v(x") = 0, f)~(x") = Ov~> ,
(14.1.8)
v, J.L = 1, 2, ... , n,
and for any given set of values fv, f~ ,
n n
(14.1.9) W n(X) = L
v-1
fv hv(x) +L
v-1
f~ f)v(x)

represents the uniquely determined 7r2n-l for which

(14.1.10) W n(Xv) = fv, w:(xv) = j~, v = 1, 2, ... , n.
(5) Again, let Sn, n = 1, 2, 3, · · · , be a sequence of sets of abscissas. If
f(x) is a given function having a derivative in [a, b], we can take fv = f(xvn),
f~ = f'(xvn) and consider the corresponding "Hermite polynomials" 62 W n(x),
n = 1, 2, 3, · · · , defined by (14.1.9). If only the continuity of f(x) is known,
we may choose f~ arbitrarily; for instance, we may take f~ = 0. Iff~ = 0, we
call W n(x) the step polynomials corresponding to I Sn l. In the more general
case I!~ I < A, where A i~: a consta.nt independent of v and n, they are called
generalized step polynomials.
The simple and generalized step polynomials Wn(x) display a more regular
behavior as n ----7 oo than do the ordinary Lagrange polynomials Ln(x). They
coincide with the given function f(x) at the same points as do the corresponding
Lagrange polynomials, but they satisfy certain additional restrictions concerning
their first derivatives. Their degree is 2n - 1 instead of n - 1. We shall
show that for certain sequences I S-n l the step polynomials (even the generalized
step polynomials) are uniformly convergent if f(x) is an arbitrary continuous
function.
The step polynomials and their generalizations have been introduced and
investigated by Fejer (10, 11, 13, 16). The trigonometric analogue for f~ = 0
had been previously considered in the simplest case of equidistant abscissas by
.Jackson (4, p. 145, Theorem VI).
For the fundamental polynomials (14.1.7) the following important relations
hold:

(14.1.11) ·,n n

~= Xv hv(x)
v=l
+L v=l
f)v(x) = X.

An important com;equence of the first identity may be pointed out for the step
polynomials, that is, when f~ = 0. Let the set I Sn l be such that
(14.1.12) a ~ x ~ b, v = 1, 2, ·. · , n.
6 2 We write "Hermite polynomials" in quotation marks in order to avoid confusion

with the Hermite polynomials of Chapter V.

332 INTERPOLATION [XIV]

Then (14.1.9) implies

(14.1.13) min f.~ Wn(x) ~ maxf., a~ x ~ b, n = 1, 2, 3, · · ·.

We observe further that (14.1.12) is equivalent to the fact that the linear
functions

(14.1.14) l" (x.)

v.(x) = 1 - l'(x.) (x - x.)

do not vanish in the open interval a < x < b, or, what amounts to the same
thing, that the "conjugate points"

(14.1.15) l' (x.)

x. + l"(x.)' v = 1, 2, · · ·, n,

lie outside of this interval. (See Fejer 13, 16.)

14.2. Fundamental polynomials of the Lagrange interpolation

63
(1) THEOREM 14.2.1. Let da(x) be an arbitrary distribution on the interval
[a, b], \pn(x) I the associated orthonormal polynomials, and l 1 (x), l2(x), · · · , ln(x)
the fundamental polynomials (14.1.2) of the Lagrange interpolation corresponding
to the set of zeros of Pn(x). Then we have

(14.2.1) lb l.(x)lll(x)da(x) = X. Ovp, v, fJ. = 1, 2, · .. , n,

where X. are the Christoffel numbers defined by (3.4.1).

This follows immediately from (3.4.1), since l.(x)lll(x) vanishes at the zeros of
Pn(x) if v r! f.J.. For v === fJ. the zero x. is the only exception, and we have !Z.(x.) )2
= 1.
Also, by (14.1.8),

[b h.(x)da(x) = X., [b h:(x)da(x) = [o xh:(x)da(x) = 0;

(14.2.2)
1 b

f).(x)da(x) = 0, 1° f)~(x)da(x) = x.' 1° xf):(x)dx = x.x.;

v = 1, 2, ... , n.
(2) We obtain atl an important consequence the following result:

THEOREM 14.2.2. Let Kn(Xo, x) have the same meaning as in (3.1.9). Then
the following identity holds:
n-1 n
(14.2.3) Kn-J(Xo' x) = L
v=O
p.(xo)p.(x) = L
v=l
x; 1 l.(xo)l.(x).

6
3 Erdos-Turan 1.
[ 14.3) CONVERGENCE IN MEAN OF LAGRANGE POLYNOMIALS

Indeed, let p(x) be an arbitrary 11"n-I . The polynomial Kn_ 1 (xo, x) is uniquely
determined by the condition (3.1.12). But by (14.1.4) and (14.2.1),

1b {t1 X~ 1 l,(xo)l,(x)} {~ p(xl')ll'(x)} da(x)

n n
= L
v=l
p(x,)l~(xo) = L
v=l
p(x,)l,(xo) = p(xo),

and this establishes the statement.

On setting Xo = x = x, in (14.2.3), we obtain a new proof of (3.4.8).
(3) In view of (14.2.1), (14.1.4) shows that

(14.2.4) j" I Ln(X)

2
1 da(x) = ~ X,lf, j 2
•

Furthermore, let f(x) be an arbitrary complex-valued function for which the

integrals f~f(x)x" da(x), v = 0, 1, · .. , n - 1, exist. Then (14.2.3) yields the
identity

(14.2.5) f.; Jro f(x)p,(x) aa(x)

n-1 I

a
1
2
= ~ x~l
n llo f(x)l,(x) aa(x) I·
a
12

Thus, according to Bessel's inequality (see (3.1.5)),

(14.2.6)

provided the last integral exists.

14.3. Convergence in mean of Lagrange polynomials

64
(1) THEOHEM 14.3.1. Let da(x) be an arbitrary distribution on the finite inter-
val [a, b], and let lPn(x) l be the corresponding set of orthonormal polynomials. For
the complex-valued function f(x) let the R.iemann-Stieltjes integrals

(14.3.1) 1° I f(x) 1
2
da(x), n = 0, 1, 2, . · · ,

exist. Then if Ln(x) denotes the Lagrange polynomial of degree n - 1 which

coincides with f(x) at the ze?·os of Pn(x), we have

(14.3.2)

The eonvergenee in mean in the usual sense follows from (14.3.2) for any f(x)
integrable in Riemann',.., scrL'ie in the case where da(x) = w(x)dx, w(x) ;:;;, fJ. > 0.
For the proof we use (14.2.4), (14.2.6), and the later Theorem 15.2.3 on
quadrature convergence. Now
64 Cf. Erdos-Turan 1 and Shohat 8.
 334 lNTERPOLATIO:K [XIV]

- 2m ~ f(x.) 1b f(x)l.(x) da(x)

(14.3.3)
~ lb lf(x) 12
da(x) + t1 X. lf(x.) 12

2{t, A,/J{x,) I'}' {t, A~'/ t

+ f(x)l,(x) da(x) /'}'

~ 1b lf(x) 1 2
+ ~X. IJ(x,) 1
da(x) 2

+ 2{~ X. lf(x,) Y{ib lf(x) 1 y.I

2 2
da(x)

Hence, using Theorem 15.2.3, we obtain

(14.3.4)

Next let e be an arbitrary positive number, and p(x) a polynomial for which
(Theorem 1.5.2)

(14.3.5)

If Ln(f;x) denotes the Lagrange polynomial of degree n - 1 corresponding to

f(x), we have
(14.3.6) f(x) - Ln(f;x) = f(x) - p(x) - Ln(f - Pix),
provided n exceeds the degree of p(x). This establishes the statement.
(2) In case da(x) = (1 - x )-tdx, a = -1, b = + 1 (that is, for the Tchebichef
2

abscissas of the first kind) Erdos and Feldheim showed that we may even assert
the validity of the following theorem:
65
THEOREM 14.3.2. Let p be an arbitrary positive number, and let f(x) be a
continuous funct2"on. Then for the Lagrange polynomials corresponding to the
Tchebichef abscissas of the first kind,
+1
(14.3.7) !~
1_ I
11
f(x) - Ln(x) IP (1 - x~-! dx = 0.

It suffices to show this for even integral values of p. In the proof which is
based on induction with respect to the even integer p essential use is made of the
property of the Tchebiehef abscissas formulated in Probl-em 57 (see below).

B& Erdos-Feldheim 1; cf. also Feldheim 2 and 3, pp. 33-36.

-~-------------------------
[ 14.4 ) LAGRAI\GE POLYl\0\IIIALS FOR JACOBI ABSCIRSAS 335

Feldheim points out (2, p. 330) that (14.3.2) is not true in general if da(x) =
2
(1 - x )!dx, a= -1, b = +1 (that is, for the Tchebichef abscissas of the second
kind), and if the exponent 2 is replaced by p = 4. For the same abscissas it is
also not true that
[+1 2
~~~
(14.3.8)
1 )_
1
I f(x) - Ln(x) 1 dx = 0.

In both cases the superior limit of the integrals in question may be + oo if f(x)
is a properly chosen continuous function.
(3) From Theorem 14.3:.1 the mean convergence in the sense of (14.3.8) fol-
lows immediately for the Jacobi abscissas, that is, for the zeros of the Jacobi
polynomial P;,aJj)(x), provided max (a, {3) ~ 0 (cf. Erdos-Turan 1); here f(x)
is au arbitrary continuom', (or even Riemann-integrable) function.
Recently, A. Hollo (1) proved this mean convergence for the Jacobi ab-
scissas with max (a, /3) < ~ provided f(x) is continuous. The bound ! is
the precise one, on account of the last result of Feldheim mentioned in (2).
Hollo also investigated the validity of

(14.3.9) !~~ 1: 1

I f(x) - Ln(f) I dx = 0,

where f('£) is continuous, and showed that (14.3.9) holds for max (a, /3) < l
The latter bound is again the precise one, at least in the sense that for
max (a, {3) > i and for a proper continuous f(x), the statement (14.3.9) does
not hold. (This follows from the second part of Theorem 15.4.)

14.4. Lagrange polynomials for Jacobi abscissas

For the following discussion Theorem 8.9.1 will be found invaluable, while
the bounds of P~a,/3J(cos 8) obtained in §7.32 will also be used.
(1) Assume a > -1, {3 > -1, and let X1 > X2 > · · · > Xn denote the zeros
of the Jacobi polynomial P~a,/3J(x) in decreasing order. Here Xv = cos Ov,
0 < ev < 7r. '1'hen we may assert the following theorem:
THEOREM 14.4. Let f(x) be continuous in [ -1, + 1] with the modulus of con-
tinuity w(o). 66 'l'hen the Lagrange polynomt-'als wZ:nct-'dt-'ng with f(x) at the zeros
of P~a,/3J(x) converge uniformly tof(x) in every interval[-!+ e, 1 - e], where
0 < e < !, provided that u.1(o) = o( I logo I - 1). The same holds in the interval
[ -1 + e, 1] if either a ~ --~and w(o) = o( I logo I -I) or fJ. - J ~ a < fJ. + ~
67

and f(x) has a continuous derivative of order fJ. with modulus of cont,z:nuity w11 (o)
satisfying the condition w11 (o) = o(oa-l'+l), fJ. = 0, 1, 2, ....
In case a < - } the Lagrange polynomials are conve1 ent at the point x = + 1
if f(x) is an arbitrary continuous function.
There ext-'st functions contt-'nuous t-'n [ -1, + 1] whose Lagrange polynomials are

GG Cf. Theorem 1.3.2.

67 For,.,. = 0 the equality ~i1gn, that is, the e1.1~e a = -1/2, is excluded.
33() INTERPOLATION {XIV l
divergent (unbounded) at a preassigned point x 0 , - 1 < x 0 < + 1 ; the same is true
at the point Xo = + 1 provided that a ~ - ~.

Similar statements hold in the interval [ -1, 1 - e] and at the point x = -1

if we replace a by {3. The convergence is uniform in the whole interval [ -1, +I]
provided max (a, /3) ~ -~and w(o) = o( jlog o I - 1), or fJ. - q ~ max (a, {3) <
fJ. + ~ and f(x) has a continuous derivative of order fJ. with modulus of con-
•
tmmty •
w~' (~) · ·/l)-1'+!) , fJ. = 0, 1, 2, . . . . A gam
u = o ( omaxi'a · ti1e
l
equa1'1ty s1gn
' '
m
case fJ. = 0, that is, the case max (a, [3) = -~'is excluded.
Compare Fejer 13, pp. 22, 24, 27; Shohat 5, p. 146. The present results are
more precise than those obtained in these papers.
(2) We shall start with a discussion of the "LebeHgue constants"
n
(14.4.1) L
v=1
IZ.(xo) I,

where -1 < xo < + ll. Let o be a fixed positive number, o< 1 - I Xo I·

Then p~a,/l\xo) = O(n-·!), so that

""
£...J IZ"(Xo ) I = 0( n -!) ""
L....J I Pn(a,/l)'(x,.. ) l-1
lx.-xol>~ lxv-xol>~
(14.4.2 )
n
= O(n-!) L Ip~a,/l)'(x,) l- 1
.
v=1
According to (8.9.2)
n n n
(14.4.3) L I P·~a,/ll'(x,) ~-I = 0(1) L vaHn-a- 2 + 0(1) L vll+~n-tl- 2 = O(n!).
v=1 v=1 v=1

Consequently,
(14.4.4) L
lx.-xol>~
IZ.(xo) I = 0(1).

On the other hand, assume I x. - Xo I ;£ o. For a fixed v we have (see (8.8.2))

-l p~a,/ll(x,) - p~a,/ll(xo)
(14.4.5) Z.(xo) = O(n ) ---- --- ·· · = 0(1),
x.- Xo
so that if Xo = cos Oo , 0 < Oo < 1r,
L
lxv-xol;:;~
IZ.(xo)l =
(14.4.6 )
= O(n-!)O(n-!) L Ie. - Oo l- + 0(1),
1

n- <16.-0ol;:;~'
1

where o' is a fixed positive number. According to (8.9.1) the last expression is
O(log n). The same bound holds for (14.4.1), and does so uniformly if t.he con-
dition -1 + e ~ Xo ~ 1 - e is satisfied.
We can also show that
 [ 14.4 J LAGRANGE POLYNOMIALS FOR JACOBI ABSCISSAS 337

(14.4.7)

if n tends to + oo over a proper sequence of integers. It suffices to choose n so

+
that I cos(Ne y) [ ~cos E1 where N and 'Y have the same meaning as in Dar-
boux's formula (Chapter VIII), and e = ! min (eo, 1t - e0 ). This is obviously
possible since from

(m + !)?t - e < Neo + 'Y < (m + !h + E,

m integral, the inequalities

(m + ~) 1t + e < (N + 1) eo + "}' < (m + j) 1t - e

follow. Now the formula of Darboux coupled with the previous argument
and (8.9.1) gives us the desired proof.
(3) As a consequence of the last result, we may conclude by means of Theorem
1.6 (Helly's theorem) the existence of a continuous functionf(x) whose Lagrange
polynomials are unbounded at the interior point x = x 0 • On the other hand,
let f(x) have the modulus of continuity w(o) = o( [log o [ - 1), and let Ln(f; x)
be the Lagrange polynomial of degree n - 1 corresponding to f(x). Approxi-
mate f(x) by a ?tn-1 = p(x) such that
(14.4.8) f(x) - p(x) = o[(log n)-1],
(Theorem 1.3.2). Then (see (14.3.6))

[ Ln(f; x) ·- f(x) [ = [ Ln(f - Pix) - \f(x) - p(x)) [

(14.4.9) 1
= o\ (log n)- ) O(log n) = o(1).

(4) Assume 1 - o ~ Xo ~ 1. On putting J.L = n +1 - v, J.L = O(n), we

have, by Theorem 7.32.2 and (8.9.2),

L
I x,-xo I>~
[Z.(xo) [ == 0(1) [ P~a.~)(xo) [ L [P~a.~)'(x.) [- 1

(14.4.10)

a = max (a, ,!).

We now pass to the determination of an upper bound for

L ll.(xo) I == L I p~a.~)(cos eo) It p~a.~)'(cos e.) 1-l

lx.-xol~~ cos eo- cose.
(14.4.11)
a+J -a-2
0 (1) I p n
(a.~) ( ) [ "' v n
== cos eo L...J I e5 - e; I'

(since (e~ - e )/(cos eo - cos e) is bounded).

2
In what follows we use both
bounds (14.4.11). We have by (4.21.7),

------------
 338 INTERPOLATION [XIV]

I p<a.fJ)( e) p<a,fJ)( e)'

I n cos o - .. cos -~ / 1 p~a.{J)'(eos e.) -l
(14.4.12) cos eo - cos e. 1

where r is between eo and e• .

Let eo = n·-J~' and: consider first the case ~ = 0(1). Let v = 0(1); then
(14.4.12) becomes O(Jl)na·Hva+ln-a-l = 0(1). On the other hand, if vis larger
than a sufficiently large fixed positive number, the second expression in the right-
hand member of (14.4.11) becomef-l (see the second bound in (7.32.5))
(14.4.13) 0(1)na L va+ln-a-
2
e.- 2
= 0(1) L va.:..! = O(na+!), 0 (log n), 0(1),
according as a > - ! , a = - } , or a < - ?t.
Now let~ be "large" and~ - v1r = 0(1), so that the number of these values of
vis bounded. Then we obtain for (14.4.12) (see the first bound in (7.32.5))
. (14.4.14) 0(1)r-a-~n-!vaHn-a-·l = 0(1)(v/n)-a-!n-!va+~n-a-l = 0(1).
Finally, assume that both ~ and ~ - v1r are "large". Then the second ex-
pression in the right-hand member of (14.4.11) becomes (see the first bound in
(7.32.5))

(14.4.15)

where the summation is extended over v1r ~ U2, U2 < v1r ~ 3U2, v1r > 3U2,
respectively. Here we have to take into account the fact that the ratio

I
e - e1 - + o(1) )
V7r
v27r2
2
1

has a positive lower bound. In the second Hum I~ - v1r I is larger than a fixed
positive constant. v\Te find now

(14.4.16)

according as a> -.},a= -~'or a< -~.

Recapitulating, we obtain for the Lebesgue constant (14.4.1), the bounds
O(naH) and O(log n), uniformly in -1 + e ~ x0 ~ 1, according as a > -!or
a ~ - ~. A similar result is found for the interval - 1 ~ xo ~ 1 - E by re-
placing a by {3. In the whole interval -1 ~ x0 ~ +1 we obtain the bounds
O(n-rH) and O(log n) for 'Y > -~and 'Y ~ -!,respectively, where we have
'Y = max (a, /3).
Now an argument similar to that in (3) must be used. Assume first that
 [14.5] STEP POLYNOMIALS IN CLASSICAL CASES 339

-1 + E ~ Xo ~ 1, and let f(x) satisfy the conditions of Theorem 14.4.

According to Theorem 1.3.3 a ?rn-1 = p(x) exists such that
n-~'o(n-:-a+r~) = o(n-a-~),
(14.4.17) f(x) - p(x) =
1o [(log n)- 1
], -1 ~ x ~ +1,
according as a > -! or a ~ - ~. This establishes the statement concerning
the convergence in the interval [ -1 + E, 1] .. The proof is obviously the same
for [- 1, 1 - E] and for [-1, + 1].
(5) There remain to be discussed the "Lebesgue constants" for x 0 = +1,
that is, the expressions
n n
(14.4.18) L
v=l
ll.(1) I "-' na L
v=l
(1 - x.)- 1 I P~a,/3)'(x,) l- 1 •

The positive zeros.x. furnish a contribution.--..., na L

(v/n)- 2 vaHn-a- 2 = L va-~,
which is O(naH), O(log n), 0(1), according as a > -!,a = -~,a< -~. The
contribution of the negative zeros is
'"" I p<a.,/3)/(
n a L....J n x. ) ~-1
"-' n a L....J
'"" v/3H n -{3-2 ,....., n aH .

And now, one more step, na,mely, the application of Helly's theorem, yields the
desired proof of Theorem 14.4.

14.5. Preliminary discussion of the step polynomials in the classical cases

We calculate the linear functions v.(x) occurring in (14.!.7) for the Jacobi,
Laguerre, and Hermite abseissas, respectively. We assume a > -1, {3 > -1
in the first and a > -1 in the second case.
(1) From (4.2.1) we obtain in the Jacobi case, since ~(x.) = 0,

(14.5.1) l"(x.) a - {3 + (a + {3 + 2)x.

=
l'(x.;i 1 - x;
Whence
( 14 .5 _2) v.(x) = 1 - x[a - {3 + (a + {3 + 2ix~ ! v
(a -_{3)x. + (a + p_±})x;.

In particular,

v.( _ 1) == (1 + a) (1 + x.) - /3(1 - x.),

1- x.
(14.5.3)
.- (1 + /3)(1 - x.) - a(1 + x.)
v. (+ 1) ·- 1 + x. .

The zeros x. are everywhere dense in [-1, + 1] (Theorem 6.1.1) if n is large.

Thus v.( -1) is non-negative for each v and n if and only if {3 ~ 0. Similarly,
v.( +1) ;;;; 0 if and only if a ~ 0. Since v.(x) is linear, we· obtain the following
result:

~---·---·---- ·---,--..,-----------------·--------..----- ...

340 INTERPOLATION [XIV]

THEOREM 14.5. The fundamental polynomials h.(x) of the first kind associated
with the Jacobi abscissas, are non-negative in -1 ~ x ~ + 1 for all values of v and
n, if and only if
(14.5.4) -1 <a~ 0, -1 < {3 ~ 0.
(2) In tpe Laguerre case (cf. (5.1.2), first equation)

(14.5.5) v,.(x) = ~.(x. - a) + x(a + 1 - x.);

v.(O) = x. - a.
x.
Here v.(x) changes its sign for all values of a if v and n are properly chosen.
In the Hermite case (cf. (5.5.2), first equation) we obtain
(14.5.6) v.(x) = 1 - 2x.x +2x;.
14.6. Step polynomials and "Hermite polynomials" for Jacobi abscissas
We again assume a > -1, {3 > -1 and use the same notation as in §14.4.
THEOREM 14.6. Let f(x) be continuous in [ -1, +1]. The generalized step
polynomials (14.1.9) [f. = f(x.), If: I < A] converge uniformly to f(x) over every
interval [ -1 + E, 1 - E]. The same holds over the interval [ -1 + E, 1] provided
a < 0. The step polynomials are in general divergent at x = + 1 if f(x) is merely
continuous and a ~ 0.
The "Hermite polynomials" (14.1.9) [f. = f(x.), f: = f'(x.)] converge uniformly
to f(x) in [-1 + E, 1] if a < t and f(x) has a continuous second derivative, or if
J.L/2 ~ a < (J.L + 1)/2 andf(x) has a continuous (J.L + 1)st derivative with modulus
of continuity w11 +I(o) salisfying the condition w11 +l(o) = o(o2 a-11 ), J.L = 1, 2, 3, ....
Similar statements hold in [ -1, 1 - E] and at x = -1 if we replace a by {3,
and in the whole interval [ -1, +1] if we replace a by max (a, /3). (Cf. Shohat 5,
pp. 138-139; Szego 14.)
(1) We start with the discussion of the convergence for -1 < Xo < +1.
Here again Theorem 8. 9.1 is virtually indispensable.
If x = x. , the numerator of v.(x) in (14.5.2) is

1 - x.[a - {3 + (a + {3 + 2)x.] + (a - {3)x. + (a + {3 + 1)x;

(14.6.1) 2
= 1 - x. > 0.
Therefore, v.(xo) is positive if I x. - xo I is sufficiently small, that is, I x. - xo I
:S o. The same is of course true for h.(x0 ). Furtherrp.ore, v.(xo) has for these
v a positive lower bound which is independent of o. We therefore obtain, on
account of the first identity in (14.1.11),

(14.6.2)

and from (14.1.7) we :see that

[ 14.6] STEP AND HERMITE POLYNOMIALS FOR JACOBI ABSCISSAS 341

Here the bound 0(1) is independent of o.

vVe shall now find a bound for the corresponding sums if I x. - x0 I > o. From
(14.5.2) and (8.9.2) we have
LI h.(xo) I = 0(1) (1 - x;)- 1\Z.(x ) ) 2
L 0
lx.-xoi>B lx.-xol>~

= 0(1)\P~a.~)(xo) ) 2 L (1 - x;)-1\P~a.~)'(x.)) - 2
lxv-xol>~
(14.6.4) n n
= O(n-1) :E (v/n)-2v2a+an-2a-4 + O(n-1) L (v/n)-2v2~+3n-2~-4
v~l v=l

Moreover, we notice
L
lx.-xol>~
I f).(xo) I = 0(1) L \l.(xo) )2
lx.-xol>~

= 0(1) \P~a.~)(xo) ) L 2
\P~a,~)'(x.)) - 2
(14.6.5) lx.-xol>~

n n
= O(n -1) L v2a+a n-2a-4 + O(n-1) L /~+an -213-4
v=1 v=1

= O(n - 1).
This yields the convergence of the generalized step polynomials of a continuous
function if -1 < xo < +1. Indeed, according to (14.1.9) and (14.1.11),
n n
I wn(Xo) - f(xo) I ~ L I f(x.)
v=1
- f(xo) I I h.(xo) I + A L
v=1
I f).(xo) I

~ max If(x.) - f(xo) I

lx.-xol;:io• ·
L h.(xo) + 0(1)
lx.-xol;:i~
L
lx.-xol;:i~
I f).(xo) I
(14.6.6)
+ 2 max IJ(x) I L
lx.-xol>~
I h.(xo) I + 0(1) L
lx.-xol>~
I f).(xo) I

max I f(x.) - f(xo) I 0(1) + o0(1) + O(n-1) + O(n-1).

I x,-xo I~;~

The factors 0(1) in the last expression are independent of o.

(2) Now assume a < 0 and 1 - o ~ xo ~ 1. The second formula in (14.5.3)
shows that v.(xo) and h.(xo) are again positive and v.(x0 ) ·is bounded from zero
if I x. - xo I ~ o, provided o is sufficiently small. (We have v.(xo) ~ v.( +1)
if a - {3 + (a + {3 + 2)x. > 0.) Then the analogues of (14.6.2) and (14.6.3)
follow immediately. In (14.6.4) and (14.6.5) a slight modification is necessary
due to the fact that in this case P~a.~)(xo) = O(na), a = max (a, - !). (The
formulas corresponding to (14.6.4), (14.6.5) hold for arbitrary a > - i; this
remark is used in (3).) Since a < 0, the conclusions of (1) remain valid; this
establishes the convergence of the generalized step polynomials if a < 0 andf(x)
342 INTERPOLATIO~ [XIV]

is continuous. (Of course, the same is true for the "Hermite polynomials" if
f'(x) is bounded.)
(3) We postpone the discussion of the step polynomials at x = + 1 if f(x) is
continuous and a ~ 0, and we pass on to a discussion of the "Hermite poly-
nomials" for 1 - o ~ Xo ~ 1 with a arbitrary but greater than -1. First we ob-
serve that the numerator of (14.5.2) vanishes for x = x. = +1; thus we have
for I x. - Xo I ~ o,

(14.6.7) I v.(xo) I < (1 - x;)-1 e(o),

where e(o) ---? 0 if o---? 0. Now,

(14.6.8)

By use of the notation and argument of §14.4 (4), we obtain for the sum above:
(a) O(n ) if~ = 0(1), v = 0(1);
2

2 1 2 2
(b) 0(1) L v a- (v/n)- = O(n a), O(n2 log n), O(n2), according as a > 1,
a = 1, a < 1, if~ = 0(1), and vis "large";
2 2
(c) O(I)(v/n)- = O(n ) if~ is "large" and~ - v1r = 0(1);
(d) O(I)r2a-l L
v2o+a(~2 - v27r2)-2 (v/n)-2 = L~ L~ L~ if both~ and + +
~ - v1r are large. The summations in the last three sums are extended over
the same values of vas in (14.4.16). We have

L~ = O(C2a·-5) L v2a+a(v /n)-2 = O(Ca n2) = O(n21'

v ,.;;;~12

(14 ·6 ·9) L~ = 0(1) L (~- v7r)-2(v/n)- 2 = O(C2n 2) = O(n2),

~/:2<v,.;;;aU2

O(n 2 log n),

according as a > 1, a = 1, or a < 1. To these cases the bounds

(14.6.10) L I h.(xo) I = e(o)O(n2a), e(o)O(n2 log n), e(o)O(n2)

jx.-xo ;;;6
1

correspond.
In order to obtain the analogous bounds for l).(xo), we cancel in (14.6.8) the
factor (I - x;)- 1 ,....._, (v/n)- 2 • Thus, we readily obtain

(14.6.11) L I f).(xo) I = oO(n2a), oO(log n), oO(l),

jx.-x 0 j;;;6

according as a > 0, a := 0, or a < 0.

The corresponding bounds for I x. - Xo I > oare O(n 2a) (cf. the remark made
in (2)). Thus, (14.1.9) and the first formula in (14.1.11) give the result
[ 14.6] STEP AND HERMITE POLYNOMIALS FOR .JACOBI ABSCISSAS 343

2
O(n a) }
I Wn(xo) - f(xo) I ~,,.~,~~ 6 lf(x.) - f(xo) Ie(o) O(n log n)
2

{
O(n2)
(14.6.12) {O(n2a) }

+ ,~,~~ lf'(x.)
1 6 I o O(log n) +max lf(x) I O(n2a)
0(1) .

+max lf'(x) I O(n2a); a = max (a, - !).

In the first term we have the alternative a > 1, a = 1, or a < 1, in the second
term a > 0, a = 0, or a < 0. The 0-expressions are independent of f(x) and o.
We now apply the usual argument. Let Wn(f; x) be the "Hermite poly-
nomial" corresponding to f(x), and p(x) an arbitrary '1!"2n_1 • Then
(14.6.13) W n(f; Xo) - f(xo) = W n(f - p; Xo) - lf(xo) - p(xo) }.
Under the condition mentioned in Theorem 14.6 we can determine p(x) (cf.
Theorem 1.3.3) such that
2
f(x) - p(x) = o(n- ), f'(x) - p'(x) = o(n-1 ) if a < !,
(14.6.14) f(x) - p(x) = o(n-2a-1), f'(x) - p'(x) = o(n-2 a),
if !JL ~ a < !CJL + 1), J.' = 1, 2, 3, · · · .
Thi~ establishes the statement concerning the "Hermite polynomials."
(4) Finally, we discuss (ef. 14.5.3))

(14.6.15) :t Ih.(l)l :t ·~ g__j-2)_Q.-=~=--

•-1
==
1 + x.
•=1 .
~~~- +~~ lll.(1) } 2•

The part of this sum defined by x. ~ 1 - ois, for a ~ 0,

(14.6.16) "-~n2a L (1 - x.)-21P~a,!9),(x.) l-2 "-' n2a L (v/n)-4v2a+3n-2a-4
"'•~1-6 x.!1;1-6

which is of the order n a or 1, according as a > 0 or a < 0. This shows

2

that the step polynomials (and also the generalized step polynomials) of a
continuous function are in ge.neral divergent at x = + 1 if a > 0. (The con-
vergence for a < 0 has be-en proved in (2).) The possibility of divergence in
case a = 0 follows by choosing f(x) = 1 - x. The corresponding step poly-
nomial is in fact

~ h.(1)f(x.) + (3) ~ (~ ~ ~:
2
2
= (1 { Z.(1) \
n
(14.6.17)
= (1 + (3) L v~l
(1 + x.)-1{P~o.!9)'(x.) ~-2
""
""'L...J (vn
I )-2 v2!9+a n -2!9-4 ,....._,,
1
344 INTERPOLATION [XIV I

so that it cannot tend to f(1) = 0. Still simpler is the proof of the divergence
of the step polynomials of f(x) = (1 +
,6)(1 - x) - a(1 x) if a > 0, since +
f(1) = -2a < 0. .

14.7. Step polynomials for Laguerre abscissas

Assume a > -1, and let x1 < X2 < · · · < Xn denote the zeros of the Laguerre
polynomial Ln(a)(x). We prove the following theorem:
Let f(x) be continuous for x ~ 0 and f(x) = O(xm) if
68
THEOREM 14.7.
x---? + oo; herem is an arbitrary but fixed positive number. The generalized step
polynomials (14.1.9) (f.= f(x.), If: I <A) converge uniformly tof(x) over every posi-
tive interval e ~ x ~ w. The same holds in the interval 0 ~ x ~ w provided a < 0.
The step polynomials are in general divergent at x = 0 if f(x) is continuous and
a~ 0.

For the proof we shall need considerations similar to those in the Jacobi case
(§14.6). In particular, we shall use Theorem 8.9.2. Some modifications in the
argument are necessary due to the fact that the zeros are unbounded. The
mechanical quadrature appears as an important new tool (cf. (15.3.5)).
(1) Assume 0 < e ~; xo ~ w. If v.(x) has the same meaning as in (14.5.5),
the values v.(xo) and h.(xo) are positive, and v.(xo) is bounded from zero and
infinity provided I x. - x0 I is sufficiently small. Thus for small o,
(14.7.1)

(14.7.2) L: I o.(xo) I =
Jx.-x 0 J~6
o0(1) L:
Jx.-xoJ~6
h.(xo).

Here 0(1) is independent of o.

If x. is small, v.(xo) = O(x;-1); if x. is large, v.(xo) O(x.). Therefore (cf.
(7.6.8))
L(a)( ) }2
L
Jx.-xoJ>6
Ih.(xo) I = 0(1) L x;- 1{ (a)' n Xo
x.<xo-6 Ln (x.)(xo - x.)

(14.7 .3)

n
= O(na-!) L x;- 1 {L~a)'(x.) l-2
•=1

But combining (15.3.5) with (3.4.5), we find

(14.7.4) ~ .-11 L<a)'( )l-2 _ r(n + 1)r(a + 1~

f=t a.• \ n x. - r(n +a+ 1) '
which yields the bound O(na-!)O(n-a) = O(n-!) for (14.7.3).
es Cf. Shohat 6, p. 139; Szego 14, p. 597.
 [ 14.8 I LAGRANGE POLYKO:\IIALS FOR GENERAL CLASSES OF ABf-lCISSAS 345

More generally, we find from (1.5.3.5) and (3.4.1), if m is a positive integer,

m ~ 2n - 1, that

(14.7.5) :t x':'-I{L~,a)'(x.) 1-z = r(n +I'(nl)r(m

v=l
+a)+ 1).
+a+ 1
Hence the :same argument as before leads to
(14.7.6) L
lx.-xo I >tl
x~' I h.(xo) I= O(n4 ),

m fixed. We have also

(14.7.7)

Equations (14.7.1), (14./'.2), (14.7.6), and (14.7:7) establish the uniform·con-

vergence of the step polynomials in the interval e ~ x 0 ~ w (cf. (14.6.6)).
(2) Now assume a < 0 and 0 ~ x 0 ~ o. If o is sufficiently small, and
I x. - xo I ~ o, both v.(xo) and h.(xo) are positive, and v.(xo) is bounded from
zero. Hence (14.7.1) and (14.7.2) are again valid. In (14.7.3) only the bound
of IL~a)(xo) l must be changed. According to (7.6.11) this will be O(n2a)
2

where a = max (~a - t, a). Therefore,

(14.7.8)

and the same bound holds for the sums in (14.7.6) and (14.7.7). Since the ex-
ponent 2a - a = max (- !, a) < 0, these sums tend to zero. From this the
uniform convergence in 0 ~ x ~ w follows.
(3) The case xo = 0, a ~ 0, can be readily disposed of by choosing f(x)
x - a. We have
n n
,~,
"'""
L....J f(x.)h.(O) = ,C... (x. - a) 2 U.(O) l 2
v=l •=1
(14.7.9)

Since this expression is positive, it cannot tend to f(O) = -a if a is positive.

If a = 0, the last expression in (14.7.9) is 1 (cf. (14.7.5), m = 1), andf(O) = 0.

14.8. Lagrange polynomials for certain general classes of abscissas

(I) Let X1n > X2n > · · · > x,.,. denote the zeros of the nth orthogonal poly-
nomial Pn(x) a:s:sociatccl with the weight function w(x) in the interval -1 ~
x ~ +I. We consider two classes A, B of weight functions characterized by
the following conditions:
A. There exists a positive number iJ. such that

(14.8.1) w(x) ~ JJ.,

B. There exists a positive number iJ. such that

-,------
 346 INTERPOLATION [XIV I

(I4.8.2) -I<x<+I.
By using this notation we prove the following statement:
THEOREM I4.8. Let f(x) be defined in -I ~ x ~ 1. Let !Ln(f; x) l denote +
the sequence of the La17range polynomials coinciding with f(x) at the zeros Xvn
of the orthogonal polynomials Pn(x) associated with the weight function w(x),
-I ~ x ~ + 1. Then limn-oo Ln(f; x) = f(x), uniformly in the interval [-I, +I],
provided thatw(x) belongs to A andf(x) has a continuous derivative in [-I, +I].
The same conclusion holds if w(x) belongs to B and w(o) = o(o!). Moreover
limn-oo Ln(f; x) = f(x), uniformly in the interval [-I + e, I - e], where
0 < e < I, provided that w(x) belongs to A and w(o) = o(o!).
Compare Shohat 7, Grunwald-Thran 1. Here, as before, w(o) is the modu-
lus of continuity of f(x) in [-I, +I].
(2) We show that for the fundamental polynomials of Lagrange interpolation
O(n), -I ~ x ~ +I,
(I4.8.3) i; ll'.(.r) I= { O(n!), -I~ x ~+I,
Q (n !) , - I + E ~ X ~ I - E,

where w(x) belongs to A in the first and third case, and to Bin the second case.
From (I4.8.3) the statement follows by reference to Theorems 1.3.2 and 1.3.3.
Let X be fixed, Ev = sgn l,.(x). We write in the case A
n n-1
(I4.8.4) p(t) = L
~~1
e,.l.(t) = L
~~o
c.Pv(t),

where P.(t) is the vth Legendre polynomial. Then

p(x) = 'i:, ll.(:r;) I = 'f c.P,(x) ~ {'f v +c;

v=1 v~o .~o
12
}!{'f
v~o
(v + !)[P.(x)]2 }!
(I4.8.5) = {f-: 1

(p(t)]
2
dtY{~ (v + t)[P.(x)] y
2

~ il-4 {i: 1

w(t)[p(t)]
2
dtrf~ (v + !)[P.(x)] 2
r.
Now, according to (I4,.2.4),

1-: ~X.[p(x.)]2 = t;x.e; =~X.= 1: w(t)dt.

1 1
2
(I4.8.6) w(t)[p(t)] dt =

Statement (I4.8.3) is readily derived by using (7.21.I) and (7.3.8).

The only essential modification of the proof in the case B is that we write
n n-1

(I4.8.7) p(t) = L
Jl-==1
e.l.(t) = L
Jl=0
d. T.(t),

where T.(t) is Tchebichef's polynomial of the first kind.

-,----
[14.9) FURTHER RESULTS ON INTERPOLATIOr\ 347

14.9. Further results on interpolation

We indicate briefly a few more recent results on interpolation.
(I) We employ the notation of §14.8. Let the weight function w(x) be con-
tinuous with a positive minimum in (-I, +I] and let e > 0. Then
max ll.(x)l ~I as n ~ oo

where the maximum is taken in the interval [-I + e, I - e] and the limit
relation holds for any sequence v = v(n) for which the corresponding zero
x. = Xm is in [-I + e, I ·- e]. (Erdos-Lengyel!.)
(2) The result of GrUnwald (1) and Marcinkiewiez (2) mentioned in §I4.I (3)
has been deepened by Erdos-GrUnwald and Erdos. We consider the zeros of
T n (x) as the set of abscisE.as.
Erdos-Grunwald (1) ., have shown the existence of a continuous function
f(x) = f(cos 8) the Fourier series of which is uniformly convergent and at the
same time the sequence of the corresponding Lagrange polynomials Ln (x) is
everywhere divergent, even everywhere unbounded.
Erdos (2) has shown that if x 0 = cos (p1r/ q ), p and q odd, there exist continuous
functions f(x) for which Ln(xo) ~ oo. Erdos-Turim (1) have shown previously
that this cannot hold for any other points xo; see also Erdos 2, p. 3I3.
(3) Important results have been obtained by Erdos for the "normal" sets of
abscissas introduced and investigated by Fejer. They are characterized by the
property that the "conjug~ate points" (I4.l.I5) lie outside of the interval [a, b]
of interpolation.
Fejer showed that for a normal set x. - x._ 1 tends to zero as n ~ oo. ErdoH-
Turan (2) have sharpened this result and Erdos (1) proved that

x. -- x._ 1 -- ?!: (I - x.2)-t

n
+ 0(. n -~)
provided that x. = Xm is restricted to a fixed interval (-I + e, I - e].
Erdos (1) solves the following remarkable extremum problem. We consider
all normal sets lxml, v and n fixed. (We follow the notation (I4.l.I).) What
is the minimum and maximum of xm? They are given by z. and -Zn-• where
z. = Zm are the zeros of Pn(z) + Pn-1(z).
(4) GrUnwald (2) and Webster (1) have studied a certain type of "sum-
mability" for Lagrange polynomials similar to a procedure of '\V. Rogosinski for
the partial sums of Fourier series. The abscissas used are the zeros of the
Tchebichef polynomials of the first and second kind, respectively.
GrUnwald (3) gives a survey of divergence properties of the Lagrange poly-
nomials. This paper contains also a discussion of the convergence of the
Lagrange and Hermite interpolation polynomials under the condition that the
abscissas form a normal set.
(5) Babizs-Turim (1, 2) and Suranyi-Turim (1) have investigated various
properties of certain polynomials of interpolation connected with ultra-
spherical abscissas. The novel feature of this investigation is the study of inter-
348 INTERPOLATION [XIV I

polation polynomials for whichf(x) andf"(x) are prescribed.

(6) Erdos (3) has investigated the "Lebesgue function"
n
L:
·=l
lUx) I,

see (14.1.6). Denoting the maximum of this function m the interval - 1

~x ~ 1, by Mn, he proved the following inequality

Mn > (2/11'-) log n -c.

This is a rather deep result. It is known that for the zeros of Tchebichef
polynomials
Mn < (2/11'-) logn +c.
Here c is a positive absolute constant.
( 7) The more diffieult analogous problem concerning the fundamental
polynomials ~.(x) of Hermite interpolation has been solved by Erdos-Turim
( 4), who prove that

_ 1~a;+1 ~I b.(x) I~
n 1(2; logn -c2loglogn ) .
n
This is "asymptotically best possible" as the case of Tchebichef abscissae
shows.
The paper Erdos-Twran 4 also contains a proof that
n 2
max L ll.(x) I > -
-1;ax;a+1 •=1 7r
log n - c3 log log n,

which is weaker than the result cited in ( 6) above, but preceded it in time.
(8) If Ln(x) is the Lagrange interpolation polynomial of a continuous
function f(x) with interpolation at the zeros of p~a,P>(x), a,fJ ~ - t, then

!~U.: J_
1
(14.9.1) ILn(x) -f(x) I P(l -x)a(l +x)Pdx =0
1

when p<min(4(a+1)/(2a+1),4(fJ+1)/(2fJ+l)') =A(a,fJ) and for p>

A(a,/3) there is a continuous function f(x) for which (14.9.1) fails. See Askey 6.
(9) By simple methods of interpolation Egervary-Turan 1 have proved the
following beautiful identity

1 _ [Pn(x) )2 = ±
•=1
1 - x: {
1 - x.
' Pn(x)
Pn(x.) (x- x.)
}
2
'

where the x. are the zeros of the Legendre polynomial Pn(x). From this they
derive a new proof of 1lhe inequality (7.21.1). By similar reasoning one can
obtain a new proof of the inequality ( 7 .21.3). See Egervary-Turan 2.
 CHAPTER XV

MECHANICAL QUADRATURE
The reader will recall that the Gaust;-Jacobi mechanical quadrature was
studied in §3.4. In the present chapter we turn to other mechanical quadra-
ture problems which are also connected with the theory of orthogonal poly-
nomialH.

15.1. Definitions
(I) Let [a, b] be a finite or infinite interval, and let
(15.1.1) Sn : X1n < X2n < • •• < Xnn ,
denote a >3et of n distinct points in [a, b]. Furthermore, let
(15.1.2) An : A1n , A2n , • • • , Ann
be a Het of real numbers. If f(x) is an arbitr~ry function defined in [a, b], we
write

(15.1.3) Qn(f) = L"

v=l
Avnf(Xvn).

We call the numbers x.,. the abscissas and the numbers X.n the Cotes numbers
of the "mechanical quadrature" Qn(f). Having been given the sequences of
corresponding sets I Sn l and IA,.l, n = 1, 2, 3, · · . , >Ve are interested in the
convergence properties of the sequence
(15.1.4)
a>3~;ociated
with a given function f(x).
As in Chapter XIV we write x. and X. instead of Xvn and Avn when there is
no ambiguity.
(2) An important speci1~l case is the following. Let the set I Sn l be an
arbitrary set of distinct numbers in [a, b], and let u(x) be a given non-decreasing
function. vVe shall define the Cotes numbers x. by requiring that

(15.1.5) Qn(f) = ib f(x) du(x)

shall hold if f(x) is an arbitrary 11"n-l • Obviously, uhder such a condition,

(15.1.6) X. = ib Z.(x) du(x), v = 1, 2, · · ·, n,

where Z.(x) denote the fundamental polynmhials (14.1.2) of the Lagrange inter-
349
350 MECHANICAL QUADRATURE [XV]

polation corresponding to the abscissas I Sn }. In this case Qn(f) is called a

quadrature of the interpolatory type.
The Gauss-Jacobi quadrature is seen to be a special case of the interpolatory
type when the abscissas I Sn l are the zeros of the orthogonal polynomials
associated with the distribution du(x). Another remarkable case of the inter-
polatory type is u(x) = x with ISnl arbitrary. The integral (15.1.5) is then
the ordinary integral of f(x) over the interval [a, b], which is assumed to be
finite in this case.
We shall use the previous notation throughout the whole chapter.

15.2. A general convergence theorem on mechanical quadrature;

theorem of Stekloff-Fejer
(1) THEOREM 15.2.1.69 Let [a, b] be a finite interval, and let the system I Sn}
(in [a, b]) and IAnl be arbitrary; let Qn(f) be defined by (15.1.3). Denote by u(x) a
non-decreasing function 1 and assume the "quadrature convergence"

(15.2.1)

for an arbitrary polynomial f(x). Then a necessary and sufficient condition for
the validity of (15.2.1) for an arbitrary continuous function f(x) is the bounded-
ness of the sequence of the "Lebesgue constants"

(15.2.2) I An1 I + I An2 I + ' ' ' + I Ann I'

This theorem is an immediate consequence of Theorem 1.6 (Helly's theorem).
It is to be noted that condition (15.2.1) holds for an arbitrary polynomial f(x)
if the quadrature is of the interpolatory type (cf. §15.1 (2)).
(2) As an application we shall prove the following important theorem of
Stekloff and Fejer:
70
THEOREM 15.2.2. Let [a, b] be a finite interval, and let the Cotes numbers
IAn l be non-negative. If the quadrature convergence (15.2.1) for an arbitrary
polynomial f(x) is assumed, the same convergence can be stated for an arbitrary func-
tion for which the Riemann-Stieltjes integral in the right-hand member of (15.2.1)
exists.

The expression (15.2.2) remains bounded in the present case, since

n n
(15.2.3) L I Avn I = ·-1
.~1
L Avn = Qn(1)

has a limit as n ---? oo. Therefore, (15.2.1) holds for a continuous f(x). The

eg Cf. P6lya 4, p. 267, Theorem I.

1o Cf. Stekloff 2, pp. 176-179; Fejer 16, p. 291; P6lya 4, p. 282, d); Shohat 7, pp. 474-476.
[ 15.2] THEOREM OF STEKLOFF-FEJER 351

extension to Riemann-integrable functions is made by means of Theorem 1.5.4.

Theorem 15.2.2 holds without change if the Cotes numbers {Anl are non-
negative only for sufficiently large values of n.
(3) A remarkable special case of Theorem 15.2.1 is obtained by considering
an arbitrary quadrature of the interpolatory type (defined by a monotonic
non-decreasing function u(x)), and by choosing for the set {Snl the zeros of the
orthogonal polynomials IPn(x) l associated with a preassigned distribution
da(x). Here a(x) is in general different from u(x). Then (15.2.1) holds for an
arbitrary polynomial f(x).
The Gauss-Jacobi quadrature is derived as a special case by taking a(x) =
u(x). The Cotes numbers are then identical with the Christoffel numbers of
§3.4 and are all positive. Applying Theorem 15.2.2, we obtain the following:
THEOREM 15.2.3. Let da(x) be an arbitrary distribution on the finite interval
[a, b], and let Qn(f) be the corresponding Gauss-Jacobi mechanical quadrature
(that is, Xvn are the zeros of the orthogonal polynomials Pn(x) associated with da(x)
and Xvn the corresponding Christoffel numbers). Then the "quadrature con-
vergence"

(15.2.4)

holds for an arbitrary function f(x) for which the Riemann-Stieltjes integral in the
right-hand member exists.
(4) Another remarkable special case arises upon choosing u(x) = x with
a(x) arbitrary. Such a choice permits us to assert the following theorem:
THEOREM 15.2.4. 71 Let Xv be the zeros of the orthogonal polynomial Pn(x) asso-
ciated with the arbitrary distribution da(x) on the finite or infinite interval [a, b].
If we define the mechanical quadrature Qn(f) of the interpolatory type by the re-
quirement in §15.1 (2) with u(x) = x, the corresponding Cotes numbers can be
represented as follows:

(15.2.5) v = 1, 2, · · ·, n,

with

(15.2.6) Kn(x) - [b Kn(X, t) dt = [b {t Pv(x)pv(t)}dt = v-0t Pv(x) 1b Pv(t) dt.

a v~o a

Here we use the symbols kn and Kn(Xo, t) in the sense explained in (2.2.15) and
(3.1.9); Xv and Av stand for Xvn and Avn , respectively.
For the proof we write (15.1.6) as follows:

(15.2. 7)
.., _ { 1 ( ) (
,..., - Pn Xv Pn+l Xv
) ~-1 [b Pn(t)Pn+l(Xv)t - XvPn+l(t)pn(Xv) dt '
a -

11 Szego 17, p. 94.

352 MECHANICAL QUADRATURE [XV]

Comparison of this formula with equation (3.2.3) establishes the statement.

According to (3.3.6) we have
(15.2.8) sgn Av = sgn Kn(x.).
In particular, if Kn(x) ~ 0 in [a, b], the Cotes numbers are non-negative. Then
the Stekloff-Fejer theorem can be applied.
The polynomial Kn(x) may be characterized by the following condition:

(15.2.9) ib Kn(x)p(x) da(x) = ib p(x) dx,

where p(x) is an arbitrary 11"n. In case da(x) = w(x) dx, we see that Kn(x) is
the nth partial sum of the expansion of {w(x) l- 1 in terms of the poly-
nomials Pn(x).
(5) In the next sections we consider the special cases previously mentioned,
namely:
(a) u(x) = a(x) (cf. (3)),
(b) u(x) = x (cf. (4));
here the abscissas {Sn l are the zeros of the orthonormal polynomials associated
with the distribution da(x) with the additional specialization that these poly-
nomials are. the classical ones.

15.3. Cotes-Christoffel numbers in the case u(x) = a(x) (Gauss-Jacobi

quadrature) for the classical abscissas
(1) We use the representation (3.4.7) of the Christoffel numbers A. and
arrange the zeros x. in decreasing order in the Jacobi and Hermite case and in
increasing order in the Laguerre case. Concerning the following results see
Winston 1. The representation (15.3.5) has already been used in §14.7.
By use of (4.3.4) and (4.21.6), the second formula (4.5.7) furnishes for the
Jacobi abscissas
A _ 2a+,B+l r(n +a+ 1)r(n + {3 + 1) ( 1 _ 2)-l{p<a.tJl'( ) l-2
(15.3.1) p - r(n + 1)r(n +a+ {3 + 1) x. n x. '
v = 1,2, ... ,n;a > -1,{3 > -1,

and, in particular, for the ultraspherical abscissas (cf. (4.7.1) and (1.7.3))

A.= 22-2>-1T"{r(A)\-2 ri~ ~ ~~) (1- x;)-l{p~>-J'(x.)\-2,

(15.3.2)
v = 1, 2, · · · , n; A > -t, A ~ 0.
2
For A = 0, that is, in the "Mehler case" w(x) = (1 - x )-!, we find
11"
(15.3.3) A.= -. v=1,2, .. ·,n,
n
[ 15.3 l COTES-CHRISTOFFEL NUMBERS IN THE CASE u(x) = a(x) 353

that is, A1 = A2 = · · · = An. For A = 1, that is, in case w(x) = (1 x 2);,

we obtain
7r ( 2) 7r • 2 V1r
(15.3.4) Av = n + 1 1 - x. = n + 1 sm n + 1' v = 1, 2, · · ·, n.
For the Laguerre abscissas we have from (5.1.10) and (5.1.14)

(15.3.5) Av = r(;(! ~ t 1
) x;-1{L~al'(x.) \-2 , v = 1, 2, · · · , n; a > -1,

whereas for the Hermite abscissas, from the second identity of (5.5.10),
1
(15.3.6) A. = 1r!2n+ n!{H:(x.) l-2, v = 1, 2, · .. , n.

(2) In the Jacobi case the argument of §7.32 (2) shows that, for a> -!and
{3> -!, that part of the sequence A1 , A2 , · · . , An corresponding to the zeros
x. > Xo (cf. (7.32.1)) is increasing, and that part corresponding to x. < xo is
decreasing. The opposite is true if a < -!, {3 < -!. If a> -!, {3 < -!,
or a < -!, {3 > -!, the whole sequence in question is increasing or decreasing
according as the first or second pair of relations holds.
In the ultraspherical case this argument furnishes
(15.3.7) if A > 0,
and
(15.3.8) if A < 0.
The symmetry relation (Problem 11) yields the analogous statement for the
other A. . The values (15.3.3) correspond to the case A = 0. In the Legendre
case we have (15.3.7).
In the Laguerre and Hermite cases we use the argument of §7.6 (1). In the
Laguerre case the sequence A1 , A2 , · · · , An is increasing for x. < a + !, and
decreasing for x. > a + !. In the Hermite case we have (x1 > X2 > · · · )
(15.3.9)

(3) In the Jacobi case, if x. = cos o. , and o. belongs to a fixed interval in the
interior of [0, 1r], Darboux's asymptotic formula (cf. (8.21.10) and (8.8.1)) yields
2a+tl+l 7r ( • o.)2a+l ( o.)2tl+l
(15.3.10) Av = Avn ,. . . _, n sm
2 cos
2
Here, for a - {3 = -! the symbol ,. . . _, can be replaced by =, according to
(15.3.3). The same is true for a = {3 = +! (see (15.3.4)) if we replace n by
n + 1. On the other hand, if v is fixed and n - ? oo, we obtain, by use of equa-
tion (8.1.1),
(15.3.11) "1\v -_ "1\vn ,. . . _, 2 a+tl+l(.)v j 2 )2a { J'a (.)
= )v
~-2 -2a-2
n '
354 MECHANICAL QUADRATURE [XV]

where J. is the vth positive zero of J a(z). (Here the symboli'V can be replaced
by = if a = {3 = -!.) Therefore, on account of the monotonic property
proved in (2), we have uniformly in v,
(15.3.12) if 0 < 0. ~ 1r - e,

according as a ~ - !, or a ~ - !. Here e is a fixed positive number.

By use of the argument and notation in §7.32 (3), we readily see that for
increasing v, v ~ vo,
0 )-2a-1 ( 0 )-2tl-1
(15.3.13) ~(0.) ( sin~ cos~ ;\.

is increasing or decreasing according as a ~ l or a ~ !, and 0 < o. ~ o,

2 2

where o is a sufficiently small positive number. (If n is sufficiently large,

~(0.) > 0, v ~ vo .) Thus (15.3.13) has uniformly the "order" n if 0 < o. ~ o.
In view of (15.3.10) and (15.3.11) for 0 < o. ~ 1r - e this yields (see (8.9.1))
" 02a+1 -1
(15.3.14) 1\v ,.....,_, • n I'J v2a+1 n-2a-2 ' 0 < 0. ~ 1r - e;
that is to say, the ratio of these expressions is uniformly bounded from zero and
infinity if v and n are arbitrary, 0 < o. ~ 1r - e. The expression in the right-
hand member of (15.3.14) attains its greatest value for v r...~ nor v ,.....,_, 1 according
as a ~ -! or a ~ - !. This again furnishes (15.3.12).
Similar results can be obtained if o. is confined to the interval e ~ o. < 1r.
(4) In the Laguerre case, if e ~ x. ~ w, we obtain (cf. (8.22.1) and (8.8.4))

(15.3.15) "
1\v = "
1\vn ~ 'Ire -x • x.a+! n -! '

where e and w are fixed positive numbers. On the other hand, if v is fixed and

n --? oo, we find from (8.1.8)

(15.3.16)

where J. has the previous meaning.

Applying an argument similar to that in the Jacobi case to the fourth equa-
tion in (5.1.2), and taking into consideration the argument of §7.6 (2), we find
that the sequences

+ 2a + 2- x. + 1 ~x~a)- e-x·x~+J{L~al'(x.)\ 2 ,
1

(4n
(15.3.17) 2
1 - 4a x, -a-!"
( 4n + 2a + 2 - x. + 4
x.
)
e x. Av

are monotonic as v increases provided a ~ l and 0 < x. ~ w. In case a > l

2 2

the same sequences con~:>ist of two monotonic parts (v ~ vo). On account of

(15.3.15) and (15.3.16) we obtain for the second sequence (15.3.17) the uniform
"order" n! so that
'
[ 15.4 l QUADRATURE OF INTERPOLATORY TYPE WITH u(x) = x 355
"\
(15.3.18) 1\v ,.....,_, x.a+in-i' 0 < x. ~ w.

Because of (8.9.10) we may write

(15.3.19) 0 < x. ~ w.

The condition x. = 0(1) is equivalent to v = O(n!).

15.4. Quadrature of interpolatory type with u(x) = x for Jacobi abscissas

(1) In this section we prove the following theorem:
THEOREM 15.4. Assume a > -1, {3 > -1, and let the mechanical quadrature
Qn(f) be defined by the following conditions:
(a) P~a,/3)(x.) = 0, v = 1, 2, · · ·, n,
(15.4.1)
(b) Qn(xk) = f. AvX~
v~1
= 1+
-1
1
xk dx, k = 0, 1, 2, · · · , n - 1.

Assume max (a, {3) ~ 3/2. Then for an arbitrary function f(x) which is con-
tinuous in [ -1, +
1],
+1
(15.4.2) !~ Qn(f)
1
= _ f(x) dx
1

holds. If the numbers a, {3 are such that max (a, {3) > 3/2, there exist continuous
functions f(x) for which (15.4.2) is not true.
Notice the difference between this statement and that of Theorem 15.2.3.
Apart from the special case max (a, {3) = 3/2, a ~ {3, this theorem has been
proved by Szego (17, pp. 102-108). The present proof is simpler; it is based
on the bounds (7 .32.5) of P~a •13\x), and on Theorem 8. 9.1 concerning the zeros
of p~a ·13\x). The notation of §14.4 is used.
(2) First assume a ~ 3/2, {3 ~ 3/2. We must show the boundedness of the
sum (15.2.2) if A. = Avn is defined by (15.4.1). We discuss ;\. only in case
0 ~ x. < 1, or 0 < o. ~ 7r/2. For the remaining values, (4.1.3) can be
used. According to (8. 9.2)

(15.4.3) "\
1\v ,.....,_,
. V a+J n -a-2
,i . p<a,/3)(
n
cos 0 - coso.
COS
0)
Sln
• OdO .

We decompose the last integral into five parts I, II, III, IV, V, corresponding
to the intervals

0./2 ~ 0 ~ 30./2,
(15.4.4)

2
and we note that (0 - o!)/(cos 0 - cos o.) is bounded in all these integrals.
Since 01 = O(n-1),
.j,)(j MECHANICAL QUADRATURE [XV]

(15.4.5)

Further, applying (8.21.18), we obtain

8,12 -! k(O)
II = [ ~ cos (NO+ -y) sinO dO
If2 cos - cos o.
(15.4.6)
8,/2 o-a-! n -!

The first integrand can be written as follows:

+ 0(1)
1 o. -
If2
2 2
0
dO.

o-a+!
(15.4.7) n-! 2 ~(o., O) cos (No+ -y),
o. -
2
0
where ~(0., 0) and its partial derivative with respect to 0 remain bounded in
the interval considered (uniformly in v). Therefore, integration by parts gives

(15.4.8) O(n-J) [ - 2 - 2
]8
0-aH ,12
+ O(n-J) 18.121 -a0 { -0-a+! -- ~(o., 0)
} IdO.
o. - o 8If2 8If2 a 2 2
o. - o
Here the symbol [f(O)]~ means f(b) + j(a). A simple discussion gives for the
first term of (15.4.8) (cf. (8.9.1))
(15.4.9)
and for the second term
-J 18,/2 { o-a+! o-a-! o-a+J }
(15.4.10) O(n )
8If2
-2-2
o. - 0
+ o. -
2
0
2 + (0. 2
- 0)
2 2 dO.

We readily see that (15.4.10) can be combined with the second integral of II.
On putting 0 = O.x in this int~gral, we obtain
8,/2 o-a-! !! -a-!
(15.4.11) n -J
1
!12
2
0. - 0 2
d0 = o.-a-J n -J
1-
X
X
2
d x.

The lower limit of integration is Ot/(20.) ,......, 1/v, so that the last integral is
O(va-!), O(log v), or 0(1) according as a > ~'a = ~'or a < !. Consequently,
(15.4.12)
. as a > 2,
accordmg 1
a -- 1
2, or a < .!
2•
Now by (8.8.2),

III= r8,!2 n-!k(O.) cos (NO+ -y)- cos (NO.+ -y) sin OdO
(15.4.13) }8,/2 cos 0 - cosO.
+ 0(1)0;a-l n1 o!.
Here n-!k(O.) = O(n-!o;a-!) = O(v-a-!na), and the same bound holds for the
[ 15.4] QUADRATURE OF INTERPOLATORY TYPE WITH u(x) = x 357

second term. Furthermore,

. No- o.
-2- . ( 0 - o. )
1 38,/2

,/2
Sin
- - - - sm N - - -
•
sm -
0 -
2
-
o. 2
+ NO. + -y +
sm--
sin 0
o.
. 0
dO

(15.4.14) (38,/2 sin N ° 2 °• ( 0 _ 0 )

= ) 8 • 12 •
0 _ 0.-- sin N -~ +NO. + 'Y dO+ 0(1)
sm - - -
2 .
N(O 0)
1
a8,!2 •
= ! sin (NO.+ -y) ~m - • dO+ 0(1),
,/2 . 0 - o.
Sln -2-

since I sin 0 - sin (0 +

o.)/2.1 < I (0 - o.)/2 j, and o.{sin (0 o.)/2\- 1 is +
bounded. The term with cos (NO. +
-y) vanishes, since the integrand is an odd
function of 0 - o. . The denominator in the last integral can be replaced by
(o - o.)/2. Hence (15.4.14) is bounded, and
(15.4.15) III = O(v-a-!na).
Now, as in II,
3'1r/4 6-a+! 13'1r/4 o-a-! -J
(15.4.16) IV =
1 8,/2
n-! T------ 2 ~(0., 0) cos (NO+ 'Y) dO
o. - 0
+ 0(1) 38,/2
~;- dO,
(J - o.
and the first integral is
o-a+! ]3'1r/4
O(n -~) [ r---:~
o - o. 38,/2
+ O(n -J) 137r/41 --aoa { o------
38,12
0-a+!
o.2
-
}I
~(0., 0) dO 2

{
1 -:---2 + ~-2
3~r/4 -a+! -a-!
= O(n-J) + O(n-J)o;:-a-J + O(n-J)
(15.4.17) 8 ,/2 0 - 0. 0 - 0.

On putting 0 = O.x, we have

_ 137r/4 o-a-! _, -a- 13'1r/(48,l X -a-! -a- "'
(15.4.18) n! 2--:~dO = n '0. ! ~-dx = O(v 1n ),
38,/2 0 - o. J x2 - 1
so that
(15.4.19)
Finally, the second mean-value theorem gives (compare (4.21.7) and also
§7.32 (2))
358 MECHANICAL QUADRATURE [XV]

V =-(cosO.- cos [3·11-/4])- 1 t' P~a,fJJ(cosO)sinOd6

}a.-/4
(15.4.20)
= O(n- ) / P~'+1 1 'fJ- 1 l(cos 0') - P~'+1 1 'fJ- 1 l(cos 37r/4) /
1

= O[nmax ({J-2,-J)J = O(n-!)'

Here we used the a~>sumption {3 ~ 3/2. Hence
X. = O(l)va+~n-a-2(v-2na + v--a-!na + n-l)
(15.4.21) ! '
= 0(1)va+~n-a- (v-a-!na + n-) = O(l)(vn- + va+~n-a- 1 ).
2 2

In all cases we obtain

(15.4.22) L jX.j=0(1),
0~"'·<1

and this establishes the boundedness of (15.2.2) and the first part of the
statement.
(3) Now assume a ~ {3, a > 3/2. We assume that o.lies in a fixed interval
[a, b), 0 < a < b < 1r; then v ,. . .,_, n, and the number of zeros o. satisfying this
condition is also ,.....,_,n. Let e and w be fixed positive numbers, e <min (a, 1 r - b).
Decompose the integral in question into the parts I, II, III, IV, V, corresponding
to the intervals
0 ~ 0 ~ w/n, w/n ~ 0 ~ 0. - e, 0. - e ~ 0 ~ 0. + e,
(15.4.23)
0. +e~ 0 ~ 1r - w/n, 1r - w/n ~ 0 ~ 1r.

Then,

I= (1- coso.)- 1 1"''" p~a,fJ)(cosO)sinOdO

(15.4.24)
+ 0(1) 1"''" Ip~a,fJJ(cos 0) I 00 2
dO.

The first integral can be calculated according to (4.21.7), and we obtain (cL
(7.32.5))

I = (1 - COS 0v )
-1 2 {J)(a-1,/J-1)( )
n+l 1 -
p<a-1,{J-1) (
n+l
W)}
COS -
n+a+f3 n
(15.4.25) + O(na- 4
)

The bound of the term O(n a- 2) is independent of w. In the same way we find
p<fJ,a)(, O)

(15.4.26)
V = ( -1) n-1
i w/n
cos
"
cos 0 +coso.
sin 0 dO
[ 15.5 l AN ALTERNATIVE :VII~THOD IX THE VLTRASPHEl{fCAL CASE 359
2
where O(nfJ- ) has a bound independent of w. (For {3 = 0 the first term
vanishes.)
Furthermore,

(15.4.27)
II+ IV= O(n-!) ie.-.
win
o-a+!do + O(n-;) ("-win
}e,+•
(1r- o)-tJ+!do

= w-a+~O(na- 2 ) + w-fJHO(ntJ-- 2) + O(n-!).

Here the Harne remark applies to the 0-terms aH above.
Finally, according to (8.8.2),

III= n- 1k(O.) (e·+• coH (NO f._2.~-cos (No.+ -y) sin OdO + O(n-!)
}e .-• cos 0 - cos 0,
(15.4.28)
e.+• s1n
. No-
--
o.
= O(n-!)
1 •-.
.
. . 0 - o.
Sin --- -
2
2 dO+ O(n-!) .

The la:-;t integral is O(log n) (cf. .Zygmund 2, p. 172); whence

(15.4.29) III = O(n-! log n).

Now assume a > {3, a > 3/2; then the previous results give

I+ II+ III+ IV+ V = na-2 {~ l + w-aH0(1)}

(15.4.30) r(a) 1 - coso.

where 0(1) is independent of w. Hence, by (15.4.3)

(15.4.31)

In the more complicated caHe a = {3 > 3/2 we find

I + II + III + IV + V
(15.4.32)
= na-2 {_2_
r(a)
1
1 - coso.
__ + -~- --~~~~
r(a) 1 + coso.
+ w-aHO(l)} + o(na-2),

where again 0(1) iH independent of w. This furnisheH the same result (15.4.31).
15.5. An alternative method in the ultraspherical case
In the ultraspberical eaHe a = {3 the convergence theorem (15.4.2) follows
from the following theorem:
THEOHEM 15.5. Let -1 < a = {3 ~ j. Then the Cotes numbers X., defined
by (15.4.1 ), are positive provided that n is chosen sufficiently large. In the ca.<ws
-1 < a = {3 ~ 0 and ~ ~ a = {3 ~ 1 the statement is true for all values of n.
360 MECHANICAL QUADRATURE [XV]

In view of the Stekloff-Fejer theorem (Theorem 15.2.2; cf. the last remark
1.'1 §15.2 (2)) this leads indeed to another proof of the convergence theorem of
§15.4, even for Riemann-integrable functions. It is very probable that X. > 0
for all values of n pro'vided -1 < a = {3 ~ j.
Concerning this section, refer to Szego 17, pp. 95-99, 109-110, where the
positiveness of X. is proved for - ~ ~ a = {3 ~ 0 and for certain other ::;pecial
cases.
(1) We prove X. > 0 by showing that the function Kn(x) = K~>-l(x) of §15.2
is positive for -1 < x < +1; here da(x) = (1 - x 2)>--! dx with X = a ! +
(cf. (15.2.8)). We can assume that n is even, since the last integral in (15.2.6)
vanishes if n is odd, and this assumption will be made throughout this section.
According to (4.7.15), (4.7.14), and (4.7.3), we have (cf. (1.7.3))

K<;)(x) = 2a-2>-. _______r(2~--- L X+ m __ p;,;-l(x)

(15.5.1) r(X + t)r(X - ~) (2X + m - 1)(m + 1)
m = 0, 2, 4, · · · , n.
Let X < 2. From a remark made in connection with (15.2.9), we have by
Theorem 9.1.2, for -1 < x < +1,

(15.5.2) . K<>-J(
1Im n
2
X ) = (1 - x )!->-,

uniformly in - 1 e+ ~ x ~ 1 - e.
(2) First assume 0 < X < t. Then

(15.5.3) m = 2, 4, 6, · · ..

In addition, p;,;-l(cos e), as a cosine polynomial, has non-negative coefficients

(cf. (4.9.19)). Therefore K<,;l(cos e) attains its minimum for e = 0, and
this minimum decrea~es as n increases. Now

" i>-) p<>-l(I) _ 22-2>-- _ r(2X)

m-o~.... "' m - r(X + !)r(X - !)
L (---- 1 + _ 1 )(2X+ m- 1)
m=0,2,4.-·· 2X + m- 1 m +
1 m
2-2>-- r(2X - 1)
2
= rex+ !)r(x - t)
(15.5.4)
m=o~ .... {(2X + n7- 2) + (2X ~~ 1 - 1)}
_ 22 -2>--
-
r(2X- 1) _
r(X + t)r(X- !)
f. (2X + mm -
m=O
2) = o,

and this establishes the statement for 0 < X< t. If X = !, we have d';:l = 0,
[ 15.5] AK ALTERNATIVE :\1ETHOD r;-.; THE ULTRASPHERICAL CASE 361

m == 2, 4, 6, · · o , and K~l(x) = 1. In the limiting case A ~ 0, we find (cf.

(4.7.8))
•
hm Kn(X) ( cos e) = 2- - -2 L...
"" ( -1- - -1-) cos me,
(15.5.5) x-.o 1r 1r m - 1 m +1
m = 2, 4, · · ·, n.
Therefore, this case can be discussed as before. (See Szego 17, p. 96.)
(3) Next, assume that -t <A < 0. In this case d;:l > 0. According to
(4.9.19), the coefficients of the trigonometric polynomial p;:l(cos e) are non-
negative except for the highest coefficient, that is, that of cos which is me,
negative, m > 0. Therefore, the coefficient of an arbitrary term cos me,
m > 0 even, in K~l(cos e) appears as a sum of the form -uo + u 1 + u2 +
•• + .uz with non-negative uo, u 1 , ... , u 1 where l = (n - m)/2. If m is
o

fixed, n ~ oo, l ~ oo, this expression tends to the coefficient of cos me in

(1 - x2)!-x = I sin e l1- 2x. But this is negative (cf., for example, P6lya-
Szeg6 2, pp. 31-32), HO that the same iH true for the partial sums
-uo + u1 + u2 + ·. · + Uz. Consequently, K~l(cos e) is again of the same
type as in the previow; case, and it attains its minimum for e = 0. The terms
d;:l P;:\1) are negative, except for m = 0. This establishes the Htatement.
(4) In case 1 < A ~ i, we start with the following identity:
(15.5.6) (1 - x2 )K<;:l(x) = K<;:- 1\x) + O"xP;;+21l(x), n even,
where O"x is a proper constant. We readily show (we can write x = 1, or com-
pare the highest terms, cf. (4.7.9)) that O"X < 0. For later purposes we give
the value of O"X :
21-2X r(2A) n + 2
(15.5.7)
O"x = 1 - A r(A + t)r(A - !) 2A + n - 1.
To prove (15.5.6), let p(x) be an arbitrary 'lrn. Then, by (15.2.9),

(15.5.8) 1:
1
{(1 - x 2)x-J [K~;- 1 ) (x) + O"xP<;:+ll (x)] - 1\p(x) dx = 0,

where O"X mw;t be determined so that the right-hand member of (15.5.6) vanishes
for x = 1.
According to the re:mlt in (2), the coefficient of cos me, m > 0, in the right-
hand member of (15.5.6) is again non-positive, and the minimum is attained
fore= 0, or X= 1; whence (1 - i)KC::l(x) > 0, -1 <X< +1.
This argument needs only a slight modification for A = 1. (See Szego 17,
pp. 96-97.)
(5) We now prove K~l(x) > 0, -1 < x < +1, for sufficiently large values
of n, provided t < A < 1, or i < A ~ 2.
We may assume n even, 0 < x < 1. First let A < 1. By (7.33.6), we have
fore ~ cn-1
(15.5.9)
 3(i2 MECHANICAL QUADRATURE [XV]

Now from (15.5.2), n even, we obtain

(15.5.10) K;:>(eos e) ~ (sin e) 1- 2x -
m~n+2,n+4.· ··

The last sum is (cf. (15.5.1)) equal to

(15.5.11)
1
since n- > m- 1 and A < 1; this furnishes
KC:) (cos e) > Hsin e) 1- 2A > 0,
provid~d n i~-; :-:uffi<'iently large and e> 1
, where cis a proper po:-;itive eon-
r:n-
Rtant.
This argument also holds for A = I with the following modification. We
have (cf. (4.7.2))

K~1>(cos e) = ~ L --~ ... ~i_n_(~ _±J)~ = ._!_ - ~ L __!__ s~_nj~ + 1)e.

1r m+1 sm e sin e 1r m+1 sin e

Here m is even; in the first :-;urn m ~ n, in the second :-;um m > n. By (1.11.6)
wefind •
(15.5.12)
provided e > cn- 1 , c > 0.
In the case ~ < f.. < 2 we usc (15.5.6) and (15.5. 7); taking the previous
results into account, we find
KC:- 1
\ cos e) > Hsin e) 1 - 2(X- 1)' O"X
<X-1)( COS, e) = e1-XO(nX-2),
P "+2
(15.5.13)
-1
' e >en
which again shows that K<,;> (cos e) is positive provided e > c'n- 1, where c'
is a proper poRitive constant. For A = 2 we uRc (15.5.12). Since 0"2 =
1
-2/7r + O(n- ), we obtain fore> c'n- 1 ,
2
sin e K~ >(cos e) > (1 - 2/7r)(sin e)-1
2

(15.5.14)

(6) Finally, we assume e> 0, e= O(n-1), n even. According to (4.7.1),

we have, A = a + ! ,

K~:) (cos~)= r(:; L (2~ +~--±i)(m"+ 1) r~!~~!\) p~a,a) (cos~),

where m = 0, 2, 4, .. · , n. By (8.1.1) we obtain for a > 0:

(15.5.15)
K (X)
n ( X)
COS• -
n
2
= ---~
r(a)
1
-a
2 2 ""
n a X _ a -2x- L..J
n
{(mx)a- J (mx)}
-
n
1
a
n

+L m2a-1 o(l) + 0(1)' m = 2, 4, · · ·, n.

[ 15.5] AN ALTERNATIVE METHOD IN THE ULTRASPHERICAL CASE 363

If n ~ oo, the first term of the right-hand member is

::::= 21-a(r(a) J-1n2ax-2afa(x),
uniformly in x, 0 < x ~ xo , where

(15.5.16)

The second term is equal to o(n2 a), a > 0.

2
Obviously, limx-.+O x- afa(x) exists and is positive. \Ve shall :-;how that the
integral function !a(x) > 0 for x > 0 provided 0 < a ~ ~·
The representation (Watson 3, p. 170, (3))

(15.5.17) ta-!J (t) = 2a+1 1r-i 1"" t-1 sm• tu d

t > 0,
a r(t- a) 1 (u 2 - 1)a+i u,
holds for -! < a < +!. Assuming 0 < a < !, we can integrate by parts.
We thus obtain the following represe~tation, valid for 0 < a < ~:

(15.5.18) ta-1J a(t) =

r
r~
7r·-l
- a)
j"" (tu)-1u(usin2 -tu 1)a-
1
- C!OS tu du, t > 0.

This reduces the statement to the discussion of the special case

(15.5.19) (7r/2)lfl(x) = 1z (t-1sin t - cost) dt = 1z t-1sin t dt - sin x.

This function increases for 0 < x < t1 and decreases for t1 < x < t2 , where
t1 and t2 are the least positive roots of sin t - t cost = 0; t1 < 21r < t2 . Thus
we need only prove h (x) > 0 if x f;; 21r. We have, however, for x f;; 21r

(15.5.20) 1z t-1sin t dt = 1r/2 - 1"" t-1sin t dt.

According to the second mean-value theorem the modulus of the last integral
is less than 2x- ~ 2(27r)- . The left-hand member of (15.5.20) is therefore
1 1

greater than 1r/2 - 2(27r)- > 1.

1

(7) The positivity of all the Cotes numbers A, for 0 ~a =11 ~ 3/2 has
been proven in Askey-Fitch 1, and for a,{J f;; 0, a +11 ~ l in Askey 5. Two
inequalities of Vietoris 1 for trigonometric polynomials give the positivity
of another sum which is related to a more general quadrature problem (inter-
polate at the zeros of PAa,tJ>(x) but integrate with respect to (1 - x) .,(1 x) &dx). +
For this and a summary of other positive sums related to quadrature problems
see Askey-Steinig 1.
(8) A new proof of the divergence half of Theorem 15.4 was given by Locher 1.
 CHAPTER XVI

POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE

In Chapter XI there were introduced certain systems of polynomials which
play a role with respeet to the unit circle similar to that which the orthogonal
polynomials previously discussed do for a real interval. We shall now give a
short survey of a further generalization which uses a rectifiable Jordan curve
in place of either the unit circle, or a real interval.

16.1. Preliminaries; definitions

(1) Let T be a simply connected region in the complex x-plane, containing
x = oo as an interior point; let the boundary C of T be a continuum consisting
of a finite number of rectifiable Jordan arcs. When integrating along C, the
parts of C are described in an arbitrarily fixed order; along the parts which
have the character of euts, we must integrate twice. The integrals considered
will have the form
(16.1.1) lt(x) I dx I·
Here f(x) is a Lebesgue-integrable function defined on C, and I dx I is the arc
element on C. We denote the total length of the boundary C by L, counting
the cuts twice.
A particularly important case is that of a rectifiable Jordan curve. Another
remarkable instance is that of a Jordan arc. The case of a finite interval is of
this type.
(2) Let
(16.1.2) X = cp(z) = CZ + Co + C1Z-l + c2z-2 + · · · , > 0,
C

be the analytic function, regular and univalent for I z I > 1, which maps I z I > 1
conformally onto the region '1', preserving the point at infinity and the direc-
tion therein. According to a theorem of Osgood and CaratModory (Cara-
tModory 1, p. 86), the function ¢(z) is continuous in I z I ~ 1 and furnishes a
one-to-one and continuous correspondence between the unit circle I z I = 1
and the boundary C ofT (described in the way indicated above). The function
cp(z) is uniquely determined, and the number c is called the transfinite diameter
(Robin's constant, capac£ty) of C (cf. §1 6.2 (5)).
Let C be a Jordan curve, and let Xo be a preassigned point in the interior of C.
In this case we may consider also the function
(16.1.3)
which is regular and univalent for I z I< 1, and which furnishes the conformal
364
[ 16.1 J PRELIMINARIES; DEFINITIONS 365

mapping of I z j < 1 onto the interior U of C; it carries the ongm z = 0 into

x = Xo and preserves the direction at the origin; cp(z) is also uniquely deter-
mined and continuous in I z I ~ 1.
For the interval -1 ~ x ~ + 1 we have ¢(z) = Hz +
z- 1) (cf. §1.9), while
t/;(z) has no sense in this case. For the cirele I x I = 1, x 0 = 0, we have ¢(z) =
t/;(z) = z.
We shall denote the inverse function:-; of (16.1.2) and (16.1.3) by z = <P(x)
and z = 'l'(x), respectively.
A particularly simple case occur~-1 when the boundary C of 7' (or of U) con-
sists of a finite number of analytic arcs. Then ¢(z) i:-; analytic on I z I = 1,
apart from a finite number of point~-1 corresponding to the corner.s of C. The
. J
length of any subarc of C corn~sponding to z = e' , 01 ~ e ~ e2 , can be repre-
sented in the form

(16.1.4)

(similarly for the mapping (16.1.3)).

(3) Let C be a .Jordan curve and w(x) a positive and continuous weight
function defined on C. Then the con:-;idc~ration:-; of §10.2 can be applied to
the functions w !ct>(c -io) l and w It/;(ei0 ) l which are both positive and continuous
on the unit circle z = e; , --1r ~ e ~ +1r. Substituting into the corresponding
0

analytic function!'l D.(z) and D;(z) the functions z = l<P(x)}- 1 and z = 'l'(x),
we obtain certain analytic functions Ll.(x) and A;(x) which have the following
properties:
(a) Ll.(x) is regular in T including x = oo, .:l,(x) is regular in {!;
(b) Ll.(x) ~ 0, .:l;(x) ~ 0;
(c) Ll.(oo) and .:l,(xo) arc real and positive.
Furthermore, we have
(16.1.5)

where Xe ~ x indicates an exterior approach to th, point x of C and x, ~ x

stands for an interior approach t.o x of C. Tlw crmvcrgenrP is in hoth cases
uniform with respect to x.
These considerations can be gPneralized by replneing the continuity of w(x)
by more general conditions, and also the~ curve C by more~ general point sets.
If Cis an arc (for instance if C ii' a finite segment), A;(x) is mc~aninglcss.
(4) We define the scalar product of two funetions j(x) and g(x), x on C, by
the integral

(16.1.6) (!, g) = ~ J f(xfg(x)w(x) I rlx 1.

We can then orthogonalize the system

2 n
(16.1.7) 1, X, X,···, X,
3()() POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE [ XVI )

and this procedure will lead to a set of polynomials uniquely determined by

the following requirements (cf. §§2.2, 11.1):
(a) Pn(x) is a polynomial of precise degree n in which the coefficient of xn
is real and positive;
(b) the system {pn(x)l is orthonormal, that is,

(16.1.8) 11 -
-y
l.J c
Pn(x)pm\X)w(x) I dx I= Onm, n, m = 0, 1, 2, · · ·.

If C is a finite real segment, or the unit circle, we obtain the polynomials

previously considered. In the next section we take up certain fundamental
properties of the polynomials Pn(x) which can be classified as follows: formal
properties (minimum-maximum properties, zeros), asymptotic behavior of
p .. (x) if n ~ oo and x is in the interior of C, asymptotic behavior of Pn(x) if
n ~ oo and x is in the exterior of C, asymptotic behavior of Pn(x) if n ~ oo
and xis on C.
In what follows we give only short indications of the proofs, particularly
when no essential change is necessary in the arguments used for the previous
special cases.
Concerning the definition and principal properties of orthogonal polynomials,
see Szego 6 and Walsh 1, Chapter VI. Most of these properties have analogues
for the polynomials orthogonal on the interior U of C; they are associated with
the following definition of scalar product:

(16.1.9) (!, g) = 1f f
u
j(x)g(x) dO',

where A is the area and dO' the element of area of U. These polynomials have
been investigated by Carleman (1, pp. 20-30). Here a weight function can
also be introduced.
16.2. Formal properties
(1) Let Dn > 0 be the determinant of the positive definite quadratic form
(cf. (2.2.8) and (11.1.4))

(16.2.1)

(16.2.2) k.,. = i1x· x"w(x) I dx I

(cf. (2.2.1) and (11.1.2)). The orthogonal polynomials Pn(x) can be represented
[ 16.2] FORMAL PROPERTIES 3G7

as follows (cf. (2.2.6), (2.2.10), and (11.1.9)):

po(x) = Do 1,
koo kw
ko1

ko,n-1 k1,n-1 ··· kn,n-1

(16.2.3) 1 X

=
(Dn-1Dn)·-t
£nn!
1 1( ) )
o
···
o
X - Xo (x - X1 · · · (x - Xn-1
)
2
·• II· ·.n-1 Ix.
,,.~0,1,2,·
·- x,. 1 w(xo)w(x1) · · · w(Xn-1) I dxo II dx1 I · · · I dXn-1 1.

We have (cf. (2.2.7), (2.2.Jl1) and (11.1.5))

+ 1)!-· }or... }or .,,.=0,1,2,

rr .. ·,n
1 2
(16.2.4) = 1 x - x 1
£n+ 1(n • "

· w(xo)w(x1) · · · w(x,.) I dxo II dx1 I · · · I dx,. 1.

(2) 72 Let j(x) be a continuous function defined on C. Then the partial sums
s,.(x) of the Fourier expansion

f(x) ""fopo(x) + f1P1(x) + f2P2(x) + · · · + f,.p,.(x) + · · ·,

(16.2.5)
f,. = i1 f(x)i 1n(x)w(x) I dx I, n = 0, 1, 2, · · ·,

minimize the integral

(16.2.6) ~.11f(x) 2
- p(x) 1 w(x) I dx I
if p(x) ranges over the class of al11r,.. The minimum is

(16.2.7)

This also yields Bessel's inequality

(16.2.8)
2 2 2
l!o 1 + l!1l + lh 1 + ·· · ~ i 2
jlt(x) 1 w(x) I dx I·
Let C be a rectifiable Jordan curve, and letj(x) be an analytic function regular
in the interior of C and continuous on C. Then the equality sign in (16.2.8)
72 Concerning the considerations of (2), (3), (4), cf. §3.1, §11.1, §11.3.
368 POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE [ XVI]

is to be taken; that is, Parseval's formula holds. This follows from Theorem
1.3.4.
Smirnoff (1) investigated the validity of this formula for the more general
class of functions j(x) for which
(16.2.9) f!tf;(z) )D;(z) 11/t'(z) )1
is of class H 2 in I z I < 1 (cf. §10.1); here we have used the symbols previously
introduced. Parseval's formula holds for this class if and only if the map
function t/;(z) satisfies the hypothesis

(16.2.10)
'6'
log I t/;' (re' ) I = -
1 1+~
21r -~
·t
log I tf;' (e' ) I
1- r
2

1 - 2r cos (e - t)
----2 dt,
+ r
O~r<l.

(See Smirnoff 1, pp. 164-168.) According to Keldysch and Lavrentieff (1),

there exist rectifiable curves for which the hypothesis (16.2.10) is not satisfied.
(3) Let p(x) be an arbitrary 11"n in which the coefficient of xn equals unity.
'Then

(16.2.11) mini Ia I p(x) 1 w(x) I dx I

2

is attained if and only if p(x) = (Dn/Dn-1)'Pn(x), and this minimum is Dn/Dn_ 1.

(~) Let xo be an arbitrary but fixed point in the complex plane, and let p(x)
be an arbitrary 11"n with p(xo) = 1. Then the minimum of (16.2.11) is attained
for p(x) = !Kn(Xo, Xo) l- Kn(Xo, x), where
1

(16.2.12) Kn(Xo, x) = l:Jo(xo)po(x) + P1(xo)PI(x) + · · · + Pn(Xo)Pn(x).

The minimum is {Kn(Xo, xo) l-1• (The same result can be expressed in terms
of a maximum property; see §3.1 (3) and §11.3.) The "kernel polynomials"
Kn(Xo , x) can be characterized by the condition

(16.2.13) i1 Kn(Xo, x)p(x)w(x) I dx I = p(x 0 ),

where p(x) is an arbitr:~ry 11"n (cf. (3.1.12)).

Using Kn(Xo, x), we can express the nth partial sum sn(x) of the development
(16.2.5) in the following form (cf. (3.1.11)):

(16.2.14) s,.(x) = i 1f(~)Kn(~, x)w(~) Id~ 1.

(5) The extremum problem of (3) can be generalized as in the case of an
intJerval (§3.11). For instance, the problem of the "Tchebichef deviation"
corresponding to (16.2.11) consists of the determination of the minimum of
· max I p(x) j, x on C, when p(x) is an arbitrary 11"n in which the coefficient of
xn is 1. The polynomials which solve this problem have been investigated
by Faber (3).
[ 16.3] ASYMPTOTIC BEHAVIOR OF Kn(x 0 , x) INTERIOR TO THE CURVE C 369

Let be the minimum in question. Then (cf. Faber, loc. cit.)

JJ.n

(16.2.15)
• 1/ n
11m JJ.n = c,

where cis the transfinite diameter defined in §16.1 (2). We shall see that a
similar formula holds for the minimum of the problem in (3) (cf. (16.4.3)).
Fekete (1) extended the definition of the transfinite diameter to an arbitrary
closed set in the complex plane by showing that for the corresponding minimum
values JJ.n (which have meaning if we replace the curve C by an arbitrary closed
set), the limit (16.2.15) still exists.
The general transfinite diameter is a remarkable set-function which can be
calculated in various cases (cf. for instance Szego 6, p. 254).
Concerning the extension of §3.11 (5) to an arbitrary curve, see Julia 1.
(6) The zeros of Pn(x) lie in the least convex region containing the curve C.
See Szego 6, pp. 236-241; Fejer 7. The proof can be based on an argument
similar to that of §3.3 (2) (cf. also §16.4 (1), (a)).
Concerning the location of the zeros of Kn(Xo , x), see Szego 6, pp. 241-244.
See also Theorem 11.4.1 and Problems 5 and 49.

16.3. Asymptotic behavior of Kn(x 0 , x) in the interior of the curve C

(1) In this section we assume that C is an analytic Jordan curve and the
weight function w(x) defined on C is positive and continuous there. Let Xo
be an arbitrary point interior to C; denote by z = 'lt(x) = 'lt(xo, x) the inverse
function of the map function t/;(z) defined by (16.1.3). In the present case
t/;(z) is regular and univalent in a certain circle I z I ~ P where P > 1. Let
~i(x) have the same meaning as in §16.1 (3). vVe shall prove the following
theorem:
THEOREM 16.3. Let !Pn(x)} denote the set of orthogonal polynomials defined
by the conditions (16.1.8). Then the series
(16.3.1) K(xo, x) = i"i;;(xo)po(x) + PI(xo)PI(x) + ·· · + Pn(Xo)Pn(x) + · · ·
is convergent provided Xo and x are arbitrary points in the interior of C. The
convergence is uniform both in Xo and x if Xo and x are limited to a closed set in
the interior of C. Furthermore, we have

(16.3.2)

In case w(x) = 1, see Szego 6, pp. 244-251. The assumptions of this theorem
concerning w(x) and C can be generalized (cf. Smirnoff 2, pp. 353-356). For a
Jordan arc (especially for a finite segment) these considerations have no sense.
For the special case where C is the unit circle, see (12.3.17).
As a conr:;equence of the convergence of (16.3.1) we notice that

(16.3.3) lim Pn(x) = 0,

370 POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE [XVI]

if x is in the interior of C (uniformly on every closed set in the interior of C).

(2) Let p(x) be an arbitrary 'lrn, and let Xo be in the interior of C. Then
according to Cauchy's theorem,

(16.3.4) (p(xo) } 2 = _!.

21r1.
J (p(x) } dx.
X - Xo
2

Consequently,

(
16.3.5 ) I p( xo)
2
I
1
~ 21r
fI I p(x) 1
x _ Xo
2

1
1dx I ~
JJ. -
2
1
L 1 ( (
1ro L }c I p x) I2 w(x) Idx I,
where w(x) ~ p. and o denotes the least distance of x0 from the points x on the
curve C. Taking into account §16.2 (4), we have

(16.3.6)

Hence the series K(xo, xo), and therefore (16.3.1), are convergent, as is readily
shown by Cauchy's inequality.
(3) Now we show that

(16.3.7) K(xo, Xo) = lim Kn(Xo, Xo) ~ L (.L\i(Xo)}-2'lt'(xo).

n-+oo 21r
To this end we consider the function

(16.3.8)

which is regular in the interior of C, that is, for I z I < 1; x = t/;(z), z = 'lt(x).
(The last factor is regular even for I z I ~ 1.) Let r be fixed, 0 < r < 1, and e
an arbitrary positive number. According to Theorem 1.3.4 we can find a
polynomial Q(x) such that
(16.3.9) I F(rz) - Q(x) I< e, x on C.
1
'Vriting p(x) = lQ(xo)}- Q(x), we obtain from §16.2 (4), for a sufficiently
large n,

(16.3.10) /K(xo, Xo)}--

1
~ !Kn(Xo, Xo)}- 1 ~ IQ(~) ~-
2

J
IQ(x) l2 w(x) ldx I;

whence, on allowing e to approach zero, we obtain

(16.3.11)

Now if r ---t 1 - 0, we have

(16.3.12)

which is equivalent to (16.3.7).

[ 16.4] ASYMPTOTIC BEHAVIOR OF Pn(x) EXTERIOR TO THE CURVE C 371

(4) Finally, we consider

Jn =
r-1-0
1
lim -£ 1Ilz!==r
Kn(Xo' x).L\i(X) l 'lt'(x)} -i - 2L (.L\i(Xo)} - 1 { 'lt'(xo)} t
7r
2
/ 1dz I

(16.3.13)
= i j1 Kn(Xo, x) l2 w(x) I dx I

lim
r-+1-0
7r- I r (.L\i(Xo) l- 1 1'¥' (xo) l t r
J1z1~r
Kn (xo' x).L\i(X)i 'lt'(x) ~-~ ~z
'/.Z

+ ~7r (.L\i (x~J) }- 'lt' (xo).

2

The first term equals Kn(~~o , Xo); for the second term we obtain
1
(.L\i(Xo) l- / •[!'(xo)} t · 27rKn(Xo, Xo).L\;(xo)i 'It' (xo)} -t = - 2Kn(Xo, Xo);
1
(16.3.14) -7r-

so

(16.3.15) Jn = -·Kn(Xo, Xo) + ir (.L\i(Xo)}-2'lt'(xo).

Therefore, limn-+oo J n = 0 by (16.3.7). From this it follows, if Iz I < 1, or if

xis in the interior of C, that (cf. (7.1.4))

(16.3.16)

which is the same as (16.3.2).

16.4. Asymptotic behavior of Pn(x) in the exterior of the curve C

(1) We shall introduce the assumptions concerning the curve C and the
weight function w(x) used in §16.3. We denote by z = cJ>(x) the inverse func-
tion of the map function x = ¢(z), defined by (16.1.2); in the present case ¢(z)
is regular and univalent in the closed exterior of a certain circle I z I = r where
r < 1. Let Ae(x) have the same meaning as in §16.1 (3). We may then make
the following assertion:
THEOREM 16.4. Let IPn(x) l denote the set of orthogonal polynomials defined
by the condt'tions (16.1.8). lf x is in the exterior of the curve C, we have

(16.4.1)

The ratio of these expressions as n ~ oo tends to unity uniformly in every finite

or infinite closed region exterior to the curve C.
In case w(x) = 1, see s~:ego 6, pp. 260-263.
We mention the following noteworthy consequences of (16.4.1):
(a) The zeros of Pn(x) approach the closed interior of C uniformly as n ~ oo
(apply Theorem 1.91.3 (Hurwitz's theorem)).
372 POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE [XVI]

(b) If xis in the exterior of C,

(16.4.2) lim Pn+l(x) = lim {pn(x) \ 11 n = <I>(x).

n·... oo Pn(X) n->oo
Cf. (12.2.6).
(c) The convergence domain of the expansion (16.2.5) of an analytic func-
tion, regular in the closed interior of C, is the interior of a level curve CR of the
conformal mapping x = ¢(z) (cf. §1.3 (2)). The determination of R from the
coefficients fn is analogous to that in the case of a power series (cf. Theorems
1.3.5 and 12.7.3).
(d) Let kn be the coefficient of xn in Pn(x). Then

(16.4.3) kn "'-/ (;y {Ll.(oo)\-lc-n-t,

where cis the transfinite diameter of C (cf. §16.1 (2); §16.2 (5)). This result
can easily be expressed in terms of the determinants Dn introduced in (16.2.4).
(2) The proof of (16.4.1) is based on an argument similar to that in §16.3;
here the minimum property of §16.2 (3) is used. We first show that

(16.4.4) • SUp C-2n-lk-2

I1m n
< 27r {AL.le ( 00 ) \2 •
= -L
n.... oo
To this end we introduce the polynomials of Faber's type fn(x) (Faber 1),
defined as the polynomial part in the Laurent development of the function

(16.4.5)

around x = oo. The function gn(x) is regular for I z I ~ r, and fn(x) is clearly
a 'lrn. Let Cr be the eurve corresponding to I z I = r. On applying Cauchy's
theorem to the ring-shaped region bounded by Cr and a large circle, we obtain,
if xis on C,

where the integration is extended in the negative sense. Hence

(16.4.7) for x on C,
where M depends only on C and r. The functions gn(x) are uniformly bounded
if xis on C.
Now let 'Y1 , 'Y2 , · · · , 'Y m 'be arbitrary constants, n > m. The polynomial
[ Hi.4] ASYMPTOTIC BEHAVIOR OF Pn(x) EXTERIOR TO THE CURVE 373

is of degree n, and its highest coeffi.eicnt is unity. Then, according to §16.2 (3)
and (16.4.7),

k;
2
~ (:~) -
1
2
C n+l i ]1 f,. (x) + 'Ydn-1 (x) + · · ·

(1H.4.9)
+ 'Ymfn-m(x) l2w(x) I dx I
1
2
= ( : : ) - C n+li Jlg,.(x) + 'Yign-t(X) + · · ·
+ 'Ymgn-m(x) l2w(x) I dx I + c2"0(r"),
so that

n-tooo

(16.4.10)

or

(16.4.11) lim sup c--

n->oo
2
,._
1
k-;:
2
~ 1
-£ 1JzJ=1
I )'(z) 2
1 w(x) I dz I,

where 'Y(z) is an arbitrary analytic function, regular for I z I ~ 1, with 'Y( oo) = 1.
On putting
1
() _ fLl.{¢(Rz)l}-
(16.4.12)
'Y z - Ll.( 00) l
' R > 1,
and allowing R to approach 1 + 0, we obtain the inequality (16.4.4).
(3) Now we consider

J: = 1
lim '-£ (
R-+1+0 JJzJ=U
jp,.(x)Ll.(x){<I>'(x)l-!{<I>(x)l-"- ( L)
27r
1 2
1 Idzl

(16.4.13) = iJ I p,.Cr:) 1
2
w(x) I dx I

p,.(x)Ll.(x) {<I>'(x) l-! { <I>(x) ~-n ~z +

2
- lim LR ( L )! { 1.
U->HO 27r }JzJ=U ~Z
The second term is

-~ (:~Y 21r !~~ p,.(x)Ll.(x){ct/(x)j- 1{<P(x)l-" = -2(~Y Ll.(oo)c"Hk,..

Therefore,

(16.4.14) J In = 2 - 2 (27r)!
L Lle (00 )Cn+! k,. .
In view of (16.4.4), this implies lim,._."" J: = 0, which establishes the statement
(16.4.1).
374 POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE [XVI]

16.5. Asymptotic behavior of Pn(x) on the curve C

Finally we have the following result:
THEOREM 16.5. The asymptotic formula (16.4.1) holds uniformly outside and on
the curve C proV1:ded the function Ll. (x) is regular in the closed exterior of C; more
precisely we have

(16.5.1) Pn(x) == (~)' {Lle(x)l- 1

{<I>'(x)j 1 {<I>(x)jn + O(hn),
where 0 < h < 1; this constant h depends on C and w(x).
The same form'ula holds in a sufficiently small neighborhood of C in the interior
of C.
(1) For the proof we use the polynomials F n (x) of Faber's type associated with

(16.5.2) (J.;)' {Ll.(x)\- {<I>'(x)\ 1 {<I>(x)ln = Gn(x)

1

(which is the "principal part" of the right-hand side of (16.5.1)) in the same way
as the polynomials fn(x) defined in §16.4 (2) are associated with (16.4.5). If
r
0 < r < 1 and is sufficiently near to 1, the function Gn(x) is regular for I z I ;s; r.
We have again

(16.5.3) for x on C,

where M depends only on C, r and w(x). The functions Gn(x) are uni-
formly bounded on C.
We write (cf. §16.4 (2))

(16.5.4) p(x) = (£)-!

2
7r Ll.( oo )en+! Fn(x).

This is a 'lrn with the highest coefficient unity. Hence

k;/ ~ (~)-
1
22
{Ll.( oo) \ c n+l i Jl Fn(x)
2
1 w(x) I dx I

(~)-
1
22 2
(16.5.5) = {Ll.(oo)\ c n+liJIGn(x)l w(x) ldxl + c2n0(rn)
= ~ {Ll.( oo) l2c2n+l + c2nO(rn).

(The simpler nature of this argument is due to the fact that Ll.(x) is regular on C.)
On the other hand,

(16.5.6)
1
1 = -£
C
j'2
I Pn(x) 1 w(x) I dx I =
1
I
2
GPn((x)) / 1 dz I
27r JzJ=1 n X
1
 [ 16.5 l ASYMPTOTIC BEHAVIOR OF Pn(x) ON THE CURVE C 375

so that

(16.5.7)

consequently

(16.5.8)

(2) Now let

(16.5.9)

where Ao, A1, A2, · · · , An are proper constants, Av = A(v, n); we have

(16.5.10)

so that An is real and, by (16.5.8),

(16.5.11)

From the definition of F n(x) we conclude that

CtY Lle(x){<I>'(x)}-l {pn(x)- AnFn(x)}

= (~YLl.(x) { <I>'(x)}-! {AoFo(x) + A1F1(x) + · · · + An-1Fn-1(x)}

2
(16.5.12) = Ao + A1<l>(x) + A2 {<I>(x)} + ··· + An-d <I>(x)} n-1 + )' 1x- 1

+ 'Y2X -2 + •••

where 'Yv, 'Y; are certain constants. Therefore,

I Ao 12 + I A1 12 + •· · + I An-1 12
~ 2 1
L7r
27r
1
Jzl-1
1Ll.(x){<I>'(x)} 1 {pn(x) -AnFn(x)} l2ldzl

(i6.5.13) = i 11 Pn(x) - AnF n(x) l w(x) Idx

2
I

= i 11
2
Pn(:r;) l w(x) I dx I

-
2
~n i Pn(x)F n(X)w(x) Idx I+ ?; [ IF n(x) l2 w(x) I dx 1.
376 POLYNOMIALS ORTHOGONAL ON AN ARBITRARY CURVE [XVI]

In the second term F n(x) can be replaced by x; pn(x); in the third term we use
1

(16.5.11) and (16.5.3) and obtain

(16.5.14) IXo 12 + IXtl 2 + · · · + IAn-11 2 ~ -1 + i 11 Gn(x) l

2
w(x) Idx I+ O(rn),
or
(16.5.15) I Ao I 2 + I At 12 + · · · + I An-1 2
1 = O(rn).
Now Fn(x) 0(1), uniformly if x is on C. Whence, according to Schwarz's
inequality,
(16.5.16) XoFo(x) + X1F1(x) + · · · + An-lFn-t(X) = O(n1rn 12 );
and from (16.5.9), (16.5.11), and (16.5.3),
Pn(x) == XnFn(x) + O(n;rn 12 ) = Fn(x) + O(n1rn 12 )
(16.5.17)
= Gn(x) + O(n1rn 12 ).
This establishes the sta.tement. The extension of (16.5.1) to the interior of C
also follows immediately, since (16.5.3) holds if x is in the interior of C and
sufficiently near to C.
 PROBLEMS AND EXERCISES
1. We denote by i 1 < i2 < · · . the positive zeros of Airy's function A (x)
(§1.81); then iv "'-/ v1 if v ~ oo. (Use (1.81.4), (1.81.1), and (1.71.7).)
2. Let A(x) denote Airy's function (§1.81). We have, for real values of x,

(): {
0 e27ri/3 J('"" exp (- p3 - pe 27ri/3 x ) dp} = A (x ) .
0

(Develop both sides in a power series in x; see (1.81.4), (1.81.1), and (1.7.3).)
3. Let Pn(x) denote the Poisson-Charlier polynomial (2.81.2); then
(·-1)na-n 12 (n!)!Pn(x) = Pn(x)
satisfies, for x = 0, 1, 2, · . · , the relation
Pn(x) = p,(n).
4. By use of the notation (2.2.1), (2.2.7), the "kernel" polynomials Kn(Xo, x)
(cf. (3.1.9)) can be represented as follows:
Co c1 c2 Cn 1
c1 c2 C3 Cn+l io
Kn(Xo, x) -D-1
n .........................
Cn Cn+l Cn+2 C2n -n
Xo
2 n
1 X X X 0
(Use (3.1.12) with p(x) = :r·, v = 0, 1, ·. · , n.)
5. Location of the zeros of the "kernel" polynomials Kn(Xo, x) (cf. (3.1.9)).
Let a and b be finite, and let xo be an arbitrary non-real number. Every zero
~of Kn(xo, x) lies in the area bounded by the interval [a, b] and by the circular
arc through a and b whose eontinuation passes through xo. (Cf. Szego 5, p. 244.
By (3.1.12)
~
l b I (x -- xo) Kn (xo' x) 12 x -
X-~ X - Xo
da(x) = 0.

In the conformal mapping (x - ~) / (x - xo) = x', the image of a ~ x ~ b is a

circular arc; the segment bounded by this arc and its chord contains x' = 0.)
6. Derive from (3.2.1) the representation, n = 1, 2, 3, · · · ,
A1x + B1 C~ 0 0
C~ A2x + B2 c~ 0
Pn(x) = po(x)
0 c~ 0

0 0 0
377

__________________
,
-------------

378 PROBLEMS AND EXERCISES

7. Prove the reality of the zeros of Pn(x) using (2.2.9), or using Problem 6.
8. Let X1, X2, · · · , Xn be distinct values in [a, b], and letf(x) have a derivative
of order 2n in [a, b]. If H(x) is the 7r2n-1 satisfying the conditions
v = 1, 2, ... , n
there exists a value ~ == ~(x) in [a, b] such that '
J(2n) (~)
2 2
f(x) - H(x) = ( n)! (x - xJ (x - x2) · · · (x - xn) 2•
2
(See A. Markoff 6, pp. H-8. Let x be a fixed value, x r!' Xv ; and to the function
of z given by
f(x) - H(x) \2( 2 2
f (z) - H (z) - (
X - X1 ) 2( ) (
X - X2 2 • • • X -
) (z -
Xn 2
Xv z - x2) · · · (z - Xn) ,

apply Rolle's theorem.)

9. Let f(x) have a continuous derivative of order 2n in [a, b]. By using the
notation of Theorem 3.4.1 obtain
!~W
i b
f(x) da(x) = Xd(xl) + Xd(x2) + · · · + Xnf(x,) + (2n)! k·;/.
Here ~is a proper value in [a, b], and kn is the highest coefficient of the ortho-
normal polynomial Pn(x) associated with the distribution da(x) (cf. (2.2.15)).
(See A. Markoff 6, p. 81; use the preceding problem.)
10. Let X1 , X2 , • • • , Xn be the zeros of the orthogonal polynomial Pn(x) asso-
ciated with a given distribution da(x) on the interval [a, b], and let X1, X2, · · · ,
'An denote the Christoffel numbers (3.4.3). Define the scalar product of two
functions f(x), g(x) by
n
(f, g) = L
v-1
'Avf(xv)g(xv).

The functions 1, x, x 2, • • • , xn-1 are linearly independent; by orthogonalization

we obtain the polynomials po(x), p1(x), · · · , Pn-1(x) which are the same as
those associated with da(x).
11. If da(x) = w(x)dx, w( -x) = w(x) and a + b = 0, we have for the Chris-
toffel numbers (3.4.3)
v = 1, 2, . ·. , n.

12. For the Jacobi polynomial p~-t.t>(x) the Christoffel numbers are

27r
Av =
2n + 1 (1 + Xv), v = 1, 2, · · ·, n.

(Cf. (4.1.8) and (15.3.1).)

13. In the special case p~-i.-i>(x), the numbers Yv = cos 4>v, 0 < 4>v < 1r, of
the separation theorem of §3.41 are
 PROBLEMS AND EXERCISES 379

<P~ = (n - v)n/n, v = 1, 2, ... , n - 1.

Verify the separation theorem for p~-i.-i>(x), p~t.i>(x), p~-U>(x).
14. Let X1, X2, · · · , Xn be the zeros of p~a,fJ>(x). Then

X1 + X2 + · ·· + Xn = (ft - a) =---n--,--
2n+a+tJ
(Cf. (4.21.2).)
15. New proof of (4.7.31). Combining the second part of (4.1.5) (polynomials
2
in 1 - 2x ) with the first formula in (4.22.1), we find

p~>->(x) = 2"(n +~ - 1
)x"F(-n/2, (1- n)/2, - n - X+ 1;x-~.
16. The generating functions (4.7.16) and {4.7.23) of the ultraspherical poly-
nomials p~>->(x) are identical if and only if X = !, that is, in the Legendre case.
17. The functional equation
(i - x)f'(x) = Xf( -x),
where X is a parameter, has a polynomial solution f(x) ~ 0 if and only if X =
( -1)"(n + 1) andf(x) = const. (Pn(x) + Pn+1(x)}, n = -1, 0, 1, · · · ; P -1(x)
= 0. (Write
N

f(x) = L
v--1
Cv{Pv(x) + Pv+1(x)},
where c-1 , Co , • • • , eN are constants, and use
(1 - x)IP~(x) + P:+1(x)} = (n + 1){Pn(x) - Pn+1(x) },
which follows from (4.7.27).)
18. Show that
Q'(o)-(- 1)"'2 2·4···n n even,
n - 1·3 .. ·(n-1)'

Qn(O) = (- 1)(n+0/2 2·4 · · • (n- 1) n odd.

1
3·5 · · · n
Here Q~(O) = 1, Q1(0) = --1. (Use the recurrence formula (4.62.13), (4.62.14),
(4.62.3).)
19. The "Laplace transform"

f(8) = 1"" e_,, F(t) dt,

of Laguerre's function F(t) = taL~a>(t) is

+ a + 1) -n-a-1( - 1)"
f( 8) -- r(nr(n + 1) 8 8 •

(Cf. Sonin 1, p. 42; use the method of the "generating function".)

380 PROBLEMS AND EXERCISES

20. Writing
X L~a>(x)(f(n
0
+a+ 1))-1 = f(n, a; x),
we have, for a> f3 > -1,

f(n, a; x) = r(a ~ f3) 1"' (x - u)a-~-1f(n, {3; u) du.

{Kogbetliantz 22, p. 1J56. Take the generating function of both sides and use
{5.1.16); the resulting :formula for Bessel functions can be verified by means of
(1.71.1) and (1.7.5).)
21. Writing e-"' 12xa 12L~a>(x) = fn(x), we have

(-1)n
f,,,(x) = - -
1oo Ja{(xy) i }fn(y)dy.
2 0

(Hardy 1, p. 139; use (5.1.9) and (1.71.1).)

22. Assume x ~ 0, y ~ 0, max (x, y) > 0. Then

e-:r:·- 11 f Ln(x)Ln(y) = (oo Cle-' dt.

n-o n + 1 .fmax (:r:,u)

(E. R. Neumann 1, Watson 6. Apply Theorem 9.1.5 and use the formula

(n -1- 1) 1"' Ln(t) dt = x{Ln(x) - L:(x)}

(cf. (5.1.2)).)
23. From (5.1.15) derive Mehler's formula
2
~ Hn(x)Hn(y) ( / 2)n = (1 _ 2)-1 {2xyw - (x + y']w }
2
L...t
n-o I
n. w w exp 1 - ·ur
Ao.2 •

(See Watson 6, Erdelyi 2.)

24. From (5.1.9) and (5.6.1} derive the following generating function for
Hermite polynomials:

to H~7) wn == (1 + 4w')-1(1 + 2xw + 4w 2

) exp ( 1 4~ ~~),
2

where m = [n/2]. (Cf. Doetsch 1, p. 590, (7).)

25. Assume k > -~ and let H~k>(x) denote the orthogonal polynomials cor-
responding to the weight function e-"' I x l2k in [- oo, + oo ]. Then the follow-
2

ing differential equations are satisfied:

xy" + 2(k - x']y' + (2nx - ex- 1)y = .0,

E _ {0, n even,
2k, n odd; y = H~k>(x);
2
z" + {2n + 2k + 1 - x2 + (- 1) ; - k} z = O;

(Generalizations of (5.S.2).)
 PROBLEMS AND EXERCISES 381

26. For fixed {J, n, and x we have

lim a-n p~a,{j\x) = _!_(X + 1)n.

a-+00 n! 2
(Use (4.21.2).)
27. Let (xv} be the zeros of p~a,fJ>(x) in decreasing order, a > -1, {3 > -1.
Then for v = 1, 2, . · · , n

lim Xv = -1, lim Xv = +1.

a-++oo tl-++oo
In the first case {3 and n, and in the second case a and n, are fixed. (Use Prob-
lem 26 and (4.1.3).)
28. Let (x.} be the zeros of P~'Jo.>(x), X > 0. We have for fixed n

lim Xv = 0.
'Jo.-++oo
(From (4.7.6) lim'Jo.-.oo (2X)-n P~'Jo.>(x) = xn/n!; see also (5.6.3n
29. Assume a> -1; {3 > -1_; if (xv} and (x;} denote the zeros of p~a,tl)(x)
and L~a>(x) in decreasing and increasing order, respectively, we have (a, n, and
v fixed)
lim {3(1 - Xv) = 2x;.
tl-++oo
(Use (5.3.4).)
30. For fixed nand x we have
.
lIm -nLca>( ) (1 - x)n
a n ax = ! .
a-+00 n .
(Use (5.1.6).)
31. Let jo = 0 < j1 < j2 < · · · be the positive zeros of Ja(x). Then Uv}
is a convex sequence if -! < a < +!,that is, jv+l - jv is increasing. Further-
more v-1jv is increasing. (Apply Theorem 1.82.2 to (1.8.9).)
32. With the same notation and assumptions as in Problem 31, we have

jv > (v + a/2 - 1/4)11', l' = 1, 2, 3, ....

(Put n = 2v - 1 in (6.3.13).)
33. \Ve have for the factor Cvn, defined by (6.31.13),

(ji/4)
2
< Cm < 4.
Here j 1 is the least positive root of Jo(x). These bounds are the best possible.
(Use the increasing property of v- 1jv ; see Problem 31.)
34. Assume a > - 1; denote by X1 < X2 < · · · < Xn the zeros of L~a) (x). If
a ~ 41 , t he sequence x;! - x;_l, v = 2, 3 , · · · , n, IS
. mcreasmg;
. . I'f a > ,,
1 th' .
2 1 2
IS Is
true for Xv-l > (a 2 - !) 1. (Apply Theorem 1.82.2 to the fourth equation in
(5.1.2).)
 382 PROBLEMS AND EXERCISES

35. With the notation of the previous problem,

·c•
nnn x. - .)
XP-1 ""' n --~ , max ( Xpi - t )
XP-1 = Xnt - i
Xn-1 "-' 2-j 3-i ('1·,2 - h· ) n -A '

where i1 and i2 are the least positive zeros of Airy's function A(x). (Use(8.1.8),
(8.22.1), and (8.9.15).)
36. Let X1 > X2 > · · · > X[Cn+ll/21 be the non-negative zeros of the Hermite
polynomial Hn(x). Write

Then for fixed v and increasing n, the numbers tm are decreasing. Further,
show that for 1 ~ v ~ (n + 1)/2,

where P < PPn < Q; here P and Q are two absolute positive constants. (On
writing ~ = (6hn)-1t in (6.32.10), we have
d2z
dt2 + {t/3 - (6hn)-! t2 }Z = 0.

The monotonic character of tpn follows by means of Theorem 1.82.1. Further-

more, see (6.32.3),

Now apply Problem 1.)

37. Consider n unit "masses", n ~ 2, at the variable points X1, X2, • • • , Xn
in the interval [ -1, + 1]. For what position of these points does the expreSBion
II
"·l'-1,2,•••,n
IXp- x,. I
P<l'

become a maximum? (Stieltjes 4, p. 441; the maximum position is the same

as in Theorem 6.7.1, obtained by replacing n by n - 2 and writing p = q = 1.
We have (1 - x )P~:!d (x) = const. {P n(x) - P n-2(x)} = const. (1 - x2 )P~-1(x);
2

see (4.7.27).)
38. Consider n unit "masses", n ~ 2, at the variable points X1 , X2 , • • • , Xn
in the interval [0, + oo ], such that
1
n - (x1 + X2 + · · · + Xn) ~ K
where K is a fixed positive number. For what position of these points does the
expressiOn
II
"·~£-1,2,•••,n
IXp- x,. I
P<l'

become a maximum? (Cf. Problem 37 and Theorem 6.7.2. We have xL~12 1 (x)
= const. {Ln(x) - Ln-1(x)} = const. xL~(x); see (5.1.14).)

----------------------------------------·---··-··-·······
 PROBLEMS AND EXERCISES 383

39. Assume a > -1, f3 > -1. With the notation of (7.32.2) we have, for
n ~ oo,

{r(q + 1)}-1 nq if q ~ -!,

max Ip~a,fJ>(x) I rv 11"-;n-; I a+! ~-a/2-i If3 +! ~-{j/2-i Ia + f3 + 1l(a+fJ+ll/2
-l;;!z;;!+\ {
if -1 <q ~ -!.
(Cf. (7.32.2).)
40. From (7.33.10) derive
Pn(cos 8) - Pn+l(cos 8) = 8'0(n4 ), 0 < 8 ~ 11"/2.
(Use
(1- x){P:(x) + P:+l(x)} = (n + 1)1Pn(x)- Pn+l(x)}
(see Problem 17) and (7.33.9).)
41. We have for 0 < 8 < 1r
(sin 8)' I Qn (cos 8) I< (7r/(2n)}'.
The constant (11"/2)' cannot be replaced by a smaller one. (See Hobson 1,
p. 309, where the bound (1rjn)' is obtained; compare Theorem 7.3.3, and Prob-
lem 18.)
42. Let f(x) be an arbitrary 11"n, non-negative for all real values of x, and let

i:co e_., f(x) dx = 1.

2

Then

max f(o) = 1r ---1 3 · 5 ·.· .· .(2m + 1) rv 1r -1 n ; , m = [n/4), n ~ oo.

2 4 . 2m
(Use Theorem 1.21.2, (7.71.2), (5.5.9), and (5.5.5).)
43. A mean-value theorem for polynomials. Let f(x) be a 11"2n . Then
f(b) - f(a) = (b - a)f'(~),

where ~ is a proper point in the interval

!(a + b) - !(b - a)x1 ~ ~ ~ !(a + b) + !(b - a)x1.
Here X1 denotes the greatest zero of the Legendre polynomial P n(x). (Tchaka-
loff 1; use (3.4.1).)
44. Derive the formula of Lipschitz

leo e-at Jo(bt) dt = (a + b~ 4,2

a> 0, b > 0,
from the generating function o'f the Legendre polynomials, that is, from (4.7.23)
for X = !. (Write w = e-atN, x = cos (b/N), a and b fixed, N ~ oo, separate
 384 PROBLEMS AND EXERCISES

.L::-o P m(x)wm into m ~ wN and m > wN (w a fixed positive number),. and

use (8.1.1) and (7.3.8).)
45. Derive (5.4.2) from (5.1.9) by writing x = a/N, w = e-btN, a and b fixed,
b > 0, N .- + oo. (See Problem 44; use (8.1.8) and (7.6.8).)
46. New proof of the asymptotic formula (8.22.4) of Hilb's type for Laguerre
polynomials. From the generating function (5.1.9) we obtain

e-:z:/2LCa>
n
·
( ) 1
X = -.
21rt
1(1-J exp {- -21r-w
+ao ·
1-+-
X w} (1 - W
)-a-1
W
-n-1 d
W

1
= 27rt
-. l(O+l exp {-- 1 + X e-z} (
1 - e
-z)-a-1 nzd
e z
-ao 2 1 - e-z
1 1(0+) d
-_ _
. e nz-z-lz -a-1;,:.(
z x, z '¥
)
z,
21rt -ao

where cf>(x, z) is regular in I z I < 21r. If we develop cf>(x, z) in a power series

in z, the resulting integrals can be reduced to Bessel functions (Watson 3, p. 176,
(1)). (This argument furnishes not only (8.22.4) for an arbitrary real a, but
also an asymptotic expansion of Hilb's type whose terms are Bessel functions;
see Wright 1. This is the analogue of Szego's argument used in 16 for Legendre
polynomials.)
47. Let a be real, but different from zero. The infinite series
CIO
""
L...t n -A eianl/2
n-1

is convergent if X > !, and divergent if X ~ !. (For X > 0 the convergence of

the series is equivalent to that of the integral J~ x-heia"' 112 dx.)
48. The polynomial sn(a, z) (see (11.3.3)) admits the following representation,
in which the notation of §11.1 is used:-

Co C-1 C-n+1 C-n 1

C1 Co C-n+2 C-n+1 a
Sn(a, z) - -D-1
n ...........................
Cn Cn-1 C1 Co a"
n-1
1 z z z" 0
(Cf. Problem 4.)
49. Second proof of Theorem 11.4.1 on the zeros of sn(a, z). Use the argument
of Problem 5. (If zo is the zero in question, and (z - z0 )/(z - a) = z', the image
of I z I = 1 contains z' = 0.)
50. Theorem 11.3.3 on the convergence of L~=o l4>n(z) 12, I z I < 1, is not true
if the weight function /(8) is such that log /(8) is not integrable in Lebesgue's
sense. (Let/(8) = 0 for - e < 8 < +e, andf(8) = 1 fore ~ 8 ~ 21r - e, and
apply (16.4.2). In this case l4>n(O) 111 " . - R; R > 1.)
 PROBLEMS AND EXERCISES 385

51. Letj(O) be a weight function on the unit circle, -1r ~ 0 ~ +1r, for which
f~; logf(O) dO exists. Assume p > 0, and let P.n = P.n(/; p) be the minimum of

2~ 1:" f(O) Ip(z) IP dO, z = ei8 ,

where p(z) = zn + ·· · is an arbitrary 7r'n with the highest term zn. Then

!i~ P.n(/; p) = exp { 217r 1+"" log f(O) dO} = ®(f). '
_,.

(This is a generalization of (12.3.3); see the argument used there.)

52. Let kno be the highest coefficient of the orthonormal polynomial Pn(x)
associated with the distribution da(x) on the finite interval [a, b]. Then

kno > 22n-1(b - a) -n {lb da(x)}-!.

(Cf. Theorem 12.7.1; Shohat 2, p. 575, (24). Use the extremum property of
Theorem 3.1.2, choosing p(x) = 21- 2n(b - a)nTnl2(x- a)/(b - a) - 1).)
53. Use the notation of Problem 52. Let [a', b'] be a subinterval of [a, b]
such that a(x) is constant in [a', b']. Then two positive constants A, B exist,
B > 4/(b- a), such that kno > ABn. (Cf. Shohat 2, p. 577. Use the extremum
property of Theorem 3.1.2, and choose for p(x) the "Tchebichef polynomial"
corresponding to two disjoint segments in the sense of §16.2 (5).)
54. By use of the notation of §12.7 (1) we have, under the assumption of
Theorem 12.7.1,
12
(2 ) -! ( -1Y 2n-p P/2d-1 v even
7r (v/2)! n o'

( - 1)
(P+l)/2
(2 ) -! n-P (P-1)/2d-2d
7r [(v- 1)/2]! 2 n o 1, v odd.

Here v is fixed, n .- oo. If d1 = 0, the second formula is to be read as follows:

limn-+ao 2-nn(l-p)f 2knP = 0. (See the special cases in Shohat 2, p. 577. Use
Theorem 12.1.2 and observe that on putting
2-n (1 (1 - X-2) 1} n = 1
+ 'Yn1X-2 'Yn2X-4 + + + · •· ,
we have 'Ynv "" (22pv!)- 1(- nY, v fixed, n .- oo .)
55. In case of the Jacobi polynomials

we have, v fixed, n .- oo,

12 (P-1)/2
"-' -! ( -1)" n-p+a+!l (v-1)/2 _ {3) -! ( - 1)
l nv = 7r' (v/2)! 2 n ' ( 2n-v+a+!i P/2-1
a 7r [(v- 1)/2]! n '

according as v is even or odd. In the second case we assume a -:;6. {3. (See

---·----·-----·---------------·-----------·-----------
 386 PROBLEMS AND EXERCISES

Geronimus 1, p. 380; usc the result of Problem 54; in this case we h:we do =
= ({3 - a) 2-(a+!i+l)/2.)
2-(a+!i+l)/2, dl

56. With the notation and assumption of Problem 54 we have

kn2 /
-k = - n 4
nO
+C+ En 1 lim En = 0,
where c is a constant. (Cf. Shohat 2, p. 577. Use the same argument as in
Problem 54.)
57. Let Z.(x), v = 1, 2, · · · , n, be the fundamental polynomials of the La-
grange interpolation with the zeros of T n(x) as abscissas. Then if k is even,

Here v1 , v2 , • • • , vk are distinct integers between 1 and n. (E. Feldheim 1;

l T n(x) Ik-\ x = cos fJ, is a cosine polynomial not containing terms with cos vfJ,
v < n.)
58. In the ultraspherical case a = {3, -1 < a = {3 < 0, we have for -1 ~ x
~ +1 (notation as in Problem 57)

!l1(x) } + l~(x) } + · · · + !ln(x) } ~ J a l-1.

2 2 2

(For a = -! see Fejer 13, p. 5. Use Problem 59 and the first identity in
(14.1.11).)
59. In the ultraspherical case a = {3 > -1 we have (cf. (14.5.2))

v.(x) = 1- 2(a + 1)xx. ~ (2a + 1)x~ ~ -a, _ 1 ~ x ~ +1.

1- x.
60. In the Legendre case we have

> t 2 31!'
v. (x) = an 4(2n + 1) ' -1~x~+l.

(Cf. Fejer 13, p. 23. Use (6.6.5).)

·----------------------------------·-·-.............
 FURTHER PROBLEMS AND EXERCISES
61. Prove the following identities:
P~n (x) = (2n+ l)xP~ j~ (2x 1), 1 2 -

xP~(x) = nPn(x) + (2n- 3)Pn-2(x) + (2n- 7)Pn-4(x) + · · ·,

(1- x )P~'(x) 2
= -n(n- 1)Pn(x) + 2(2n- 3)Pn-2(x)
+ 2(2n - 7)Pn-4(x) + · · · .
62. Prove:

if n = 1,

if n ::: 2.

63. Prove the formula:

! -1
+1 p (x)xn+2•dx =
n
_!_ (n
2n
+ 2v)!
(2v)!
r(v
r (n
+ i)
+ + j) 'll
ll = 0, 1, 2, ....

64. Prove the identity

! -1
+1 Pn(x)e-i>-zdx = i-n(211'/X)lJn+!(X).

(Use Problem 63 and (1.71.1).)

65. Prove that
00
<X<
f +
-oo
X-!J n+l (X )ei>-z dX =
J (211' )linP (x)
l n
0
if
if
-1
x > 1 or x
1,
<- 1.
(Use Problem 64 applying Fourier's inversion formula.)
66. Prove the identity

~
f;:o
(n) p~>->(x)
vp~>-> (l) y
n-v = (1 +2 +
xy y
2)n!2 p~>->( (1 + 2xy + y2)-l(x + y)l
p~>-> (1) '

in particular the identity

}_: (n)
v =:0 v
P .(x )yn-· = (1 + 2xy + y 2)n' 2P n1(1 + 2xy + y 2)-l(x + y) l.
(Use (4.7.23).)
67. Prove the identity
n
2:
v =:0
(n)
ll
£~a) (x) n-• -
va>(O)
v
Y - (y + 1)
n L~a) ( (y
£<a>(O)
n
+
1)-lx l
·

387
388 FURTHER PIWBLEMS AND EXERCISES

(Use (5.1.9 ). )
68. Prove the identity

.to (~)H.(x)(2y)n-• = Hn(X + y).

(Use (5.5.7).)
69. Let x be a parameter, - 1 < x < + 1. The zeros of the polynomial in z:

Po(x) + (~) P1(x)z + (~) P2(x)z + · · · + Pn(x)zn

2

are all real.

(Use Problem 66.)
70. Prove Turan's inequality
(Pn(x)j2- Pn-l(x)Pn+l(x) ~ 0, -1~x~+l.

(Turim 1, see also Szego 22, Karlin-Szego 1 and Csordas-Williamson 1.) If

a 1 and a 2 denote the first and second elementary-symmetric function of n
real numbers, we have:

(Use Problem 69:)

~1. Derive from Problem 66 the generating series (4.10.6) and, in particular,
(4.10.7). (Put y = n/z, n-? oo; use (8.1.1).)
72. Derive from Problem 67 the generating series (5.1.16). (Put y = n/z,
n-? oo; use (8.1.8).)
73. Prove the following formula of the "Rodrigues type":

e-zxa£<a> (x) = ( -1 )n tn+l

n n!
(d
dt
)n (e-lltt-a-1)
'
xt = 1.

(Use Taylor's formula and (5.1.9).-In the special case a = 0 this is due to
G. P6lya, 1941.)
74. Let (unl and (vnl betwosequences;n = 0, 1, 2, · · ·. Oneoftherelations

Un = ~(n+a)
.i...J ( -l)•v.; Vn = ~(n+a)
.i...J ( -1 )•u.
• =O n - v • =O n - v
implies the other.
75. Using Problem 74 and (5.1.6) prove

X~
n.
= t (n +a) (-l)·L~a> (x ).
v=O n - ll

76. Let Un and Vn be two sequences; n = 0, 1, 2, · · · . One of the relations

[n/21 ( -1)• [n/21 1
Un =
v
L=0 --,-
ll.
Vn-2•; Vn = L=0
v
f Un-2•
ll.

implies the other.

 FURTHER PROBLEMS AND EXERCISES 389

77. Prove the identity:

(2x)n = [~] _!_ Hn-2.(x)
n.' •.i...J
=O v.' ( n - 2)1'
v .
(Use Problem 76 and (5.5.4).)
78. Prove the following identities:
y" + 2xy' + 2(n + l)y = 0,

79. Prove that

(Use (5.5.4). Cf. P. Turan, Matematikai Lapok, vol. 5 (1954), pp. 134-137.)
80. Prove that
lim
a-> oo
a-ni 2 L~':> (a):t + a) = ( -1)"2-nl 2 (n !)- 1}[ n(2-l:r).

81. Let IPm(x) l be the orthonormal polynomials associated with the distribu-
tion da(x) in 0 ~ x < + oc. Denoting by h ~2, · · • , ~k any zeros of Pm (:r) \\'<'
have

t > 0.
(See Karlin-McGregor 1, pp. 507-509. Sinee .f(O) = 0, we have

d
e-ht dteh:f(t) = 1.0
e-"' 1 1(~1- x) · · · (~k-1- x)l- 1 1Pm(x)j2da(x) = cp(t),

Induction.)

82. Notation: and assumption as in Problem 81, Pm (0) > 0 for all m. Provp
that
t > 0.

(See Karlin-McGregor 1, loc. cit. Let rn > n. We represent Pn(x) by Lagrange':-;

interpolation formula corresponding to the abscissas ~~'~' · · · , ~~',+! ehosen as
in 3.3 (6); use Problem 81.)
83. Let IPn (x) l be the Poisson-Charlier polynomial:-: (2.81, sgn Pn (0) =
(-l)n). We have, i\ > 0, rn ~ n, j(x) as in (2.81.1),

---------------------------------------------·-··-··-- ..
390 FURTHER PROBLEMS AND EXERCISES

.
L
.r=O
e-'1\zj (x )pn (x )pm (x)

=
n
exp [a(e-'1\- l)]aHn+m>(m!/n!)! ·~=O
.
(n)
,
"
(e-'1\- l)n+m- 2• (ae'J\ )-•
(m - v)! ·

84. Let 'A> -~,'A -:;6. 0. Using the notation (4.9.21) we have

provided that l + m + n = 2s is even and a triangle with sides l, m, n exists.

The integral on the left is zero in every other case. (Cf. (4.7.15); cf. Hsii 1.)
85. Let ln(x) have the same meaning as in (2.8.1); we have then
ln(x) = ( - l)"ln(N- 1- x).

Here n and x run over the range 0, 1, · · ·, N- 1. Prove also that

ln(N- 1) = n! (N-1) n

86. Let Pn(x) have the same meaning as in (2.81.2); we have then
Pn(O) = ( _ l)"(a"jn!)l/2 = ( _ l)"eaf2(j(n))1f2.

Prove also that, writing p,.(x)/pn(O) = cn(x,a), we have

Cn(x;a) = Cz(n;a).
87. Writing Hn(x) = Hn we have
28 !R! !
f ro

-ro
e -z2 H a H s H 1' dX-1r
'
- t;2 a. /J • 'Y •
(s-a)!(s-(J)!(s-')')!
+
provided that. a+ (3 'Y = 2s is an even integer and s ~a, s;?; (3, s ~ 'Y·
In all other cases the integral is zero.
88. Prove that

L~a)
2
limn! ( - Y2 )" (n ) = e-Y.
n~ro n y
89. Prove the following identities:
L~a>(x) = y, + ne-xxay =
(e-xxa+ 1y')' 0;

e-xf 2Ln(X) = h, xh" + h' + ( n + -2-- X) h = 0.

4

·-------------·-··--·
 FURTHER PROBLEMS AND EXERCISES 391

90. Prove the identity

n
L~a+P+ll(x +y) = L L~a~k(x)LkPl(y).
k=O

91. Obtain the estimates

2J.L < ap -
2 p /2, +
2J.L =ap -2 +p/2,
2J.L > ap -2 +p/2,

where a,{3,J.L > -1 and p > 0 are fixed. (See (7.34.1).)

92. Prove that
1 sin(n + 1)8 · V6 1r 1r
--< < - __ s:;,o_s;1r--,
3 = (n + 1) sinO = 9 ' n n
1 sin(n + 1)8 1 1r 1r
--< <- -<o<-
3 =(n+l)sinO =s• n = =2'
and show that all the bounds are obtained for some 8 and n. (Use (7 .8.1) for
p~l/2,1/2) ( x) I p~1!2,1/2l( 1) .)
93. Prove that
d2n
dx2n (1 -x)--ao +x)-P > 0, -1 < x < 1, a,{3 ~ 0,
and

X> 0, a~ 0.

(Use Theorems 6.72 and 6.73. These results were conjectured by I. Jo6.)
94. Prove that

(Use (5.1.9).)
95. Prove that
n+m
Pn(x)pm(X) = L a(k, m, n)pk(x)
k= ln-ml

with a(k, m, n) ~ 0 if Pn(x) satisfies

Pl(x)pn(X) = Pn+l(x) +anPn(X) + f3nPn-1(x), n =1,2, · ··,
where an ;;s:; 0, f3n > 0, an+l ~ an, f3n+1 ~ f3n, n = 1, 2, · · · ,po(X) = 1, P1(x) =
x +a. (Askey 4. Use induction. This contains Problem 94.)
 392 FURTHER PROBLEMS AND EXERCISES

96. Prove that

0 ~X~ 1,
when n(n + 1) = k(k + 1) +m(m + 1). Also show that
Jo(x) +Jo(Y) ~ 1 +J0 (z)
when x 2 +y 2 =z 2• Gri.inbaum 1, 2, Askey 7.
97. If L;=oa; is finite, a> -1, and if

0 ~X, y < oo,

prove that

98. If ak,m,n are defined by

1
--..,..---.,....--:---:-:----:-------,----- L: a rksmtn
00

(1-r)(l-s) +(1-r)(l-t) +(1-s)(l-t) k,m,n=O k,m,n

show that

Then use Problem 84 for A. = ! and ( 4.5.4) to show that ak,m,n ~ 0. Generalize
to .fooo Lia>(x)L~>(x)L~a>(x)xae-3xdx, a~-!. (Szego 26, Askey-Gasper 3.)
99. Prove that

k =1,2, .. ·,

for all choices of plus and minus signs and any choice of integers ni. (Use
(5.5.11) and Problem 87. See Ginibre 1.)
100. Show that

2nn! to ( -1)kLk(2x 2 +2y 2) =to(~) [Hk(x) JZ[Hn-k(y) )2

and in particular that

n (n/2) (2j)l
2nL(-1)kLk(2x 2) =L .1 • 1( " .)! [Hn-2j(x))2.
k=O j=O}·}· n - 2
} ·
Extensions of this sum to p~a,o>(x) are given in Askey-Gasper 4.
101. Show that
~
[L(a)( )) 2 =r(n+a+1) (2k)!(2n-2k)! £<2al( )
2
n X 2 2nn! 6k![(n-k)!] 2r(k+a+1) Zk X.
This formula of Howell is given in Bailey 2.

----------------------------------·-------------- ..-····-·· .,
 APPENDIX

ON A SINGULAR CASE OF ORTHOGONAL POLYNOMIALS

In recent publications F. Pollaczek (1-4) has introduced certain remarkable
generalizations of the Legendre and other classical polynomials which should be
treated in this Appendix in a brief way. The polynomials of F. Pollaczek show
in many respects a singular behavior. For a short treatment of this topic we
refer to Bateman Manuscript Project 1, vol. 2, pp. 218-221. Cf. also Szego 24.

1. Definitions and formal properties

Let a and b be real parameters, a > lbl. We write

) h (()) = a cos .() + b

(1. 1 2 sm ()

and define the polynomials p n(x; a, b) = knxn + ... by the generating function
.,
(1.2)
f(x, w) = f(cos 8, w) = L Pn(x; a, b)wn
n=O

or, in another form:

(1.3) f(x, w) = (1 - 2xw + w )-J exp { (ax+ b) ml;;"'
2
1
m
wm Um-l(x) }

where Um-l has the same meaning as in (1.12.3). The polynomials Pn(x; a, b)
reduce to the Legendre polynomials in the limiting case a = b = 0.
The following identities are easy to establish:
(1.4) Pn(x;a, b)= (-1)nPn(-x;a, -b),
(1.5) Pn(1;a,b) = Ln(-·a-b), Pn(-1;a,b) = (-l)nLn(-a+b)
where Ln(x) = L;?l(x) is Laguerre's polynomial (Chapter V). The highest
coefficient kn of Pn(x; a, b) can be obtained by replacing w by wjx in (1.3) and
taking x --+ oo . We find

(1.6) k = 2n (
n
n + l(a -
2
n
1)) - ""'2"
nl<a-o
r(t(a+1))
as n--+ oo
.

The following recurrence formub holds (cf. Bateman Manuscript Project,

loc. cit.):

(1.7)
nPn(x; a, b) = [(2n- 1 + 2a)x + 2blPn-l(x; a, b)
- (n- l)Pn-2f.r; a, b), n = 2, 3, 4, · · · .
393

·---------·-·-···· ··--··
394 APPENDIX

Here P 0 = 1, P1 = (2a + l)x + 2b.

We have the important relation of orthogonality (cf. Szego 24):

~-~
1

(1. )
8
Pn(x; a, b)Pm(x; a, b)w(x; a, b) dx = (n + Ha + 1))- 1 Onm,
n, m = 0, 1, 2, · · · ,
where the weight function is defined by
(1.9) w(cos 8; a, b) = e< 2 B-.-Jh<OJ [cosh (1rh(8))]- 1•
We note that
(1.10) w(cos 8; a, b)= 2 exp l (a+ b)(1- 1r/8)) as 8-t +O.
The behavior of w is similar as 8 - t 1r - 0.
The following representation in terms of the hypergeometric function holds
(Bateman Manuscript Project, loc. cit.):
(1.11) Pn(cos 8; a, b) = einBF( -n,! + ih(8); 1; 1 - e- 2 i8 ).

2. Generalization
Let X be real, X > -!. We define the polynomials p~l (x; a, b) by the gener-
ating function

(2.1) L.. p;;l(x; a, b)wn = (1- 2xw + w )->- exp { (ax+ b) L-

2 .. wm Um-l(x) } .
n=O m=l m
For a = b = 0 we obtain the ultraspherical polynomials p;;l (x ). The case
dealt with in 1. corresponds to X = t. The polynomials P;;l (x; a, b) are orthog-
onal in -1 ~ x· ~ 1, x =·cos 8, with the weight function
(2.2) w<>-l(x; a, b) = 2 >--te< 28 -rlh<BJ (sin 8) 2 >--tJr(X
?r- 1 2 + ih(8))J 2
•

Concerning a recurrence formula and a representation in terms of the hyper-

geometric function, see Bateman Manuscript Project, loc. cit.

3. Integral representations
The following generalizations of the Laplace integral (4.8.11) and of the Mehler
integrals (4.8.6) and (4.8.7) hold (Novikoff 1):

Pn(cos 8; a, b) = ?r- e-
1 28
h<OJ cosh (1rh(8)) 1r exp { 2ih(8) log ctg ~}
(3.1)
· (cos 8 + i cost sin 8)-n-l dt
= e-oh<oJ cosh (1rh(8))
. 8 +- t}
(3.2) .;1
2
o
o
cos
{
(n +
sm-
2
t )t - h(8) log . 8 _ t
sm--
2
. (2 cost - 2 cos 8)-t dt

____________ , __ ·--····-·
 APPENDIX 395

= e<r-Olh<Ol cosh (1rh(O))

2 .. . t + ()}
sm-2-
(3.3) . :;;: [ sin (n + t )t - h(O) log . t _ ()
{ sm--
2
. (2 cos () - cos t)-l dt.
In these formulas 0 < () < 1r.

4. Infinite interval
Pollaczek (3) defines another remarkable class of polynomials P':l (x; a) by the
following generating function:
.
(4.1) L P':l(x; a)wn = (1 - weia)->-+ix(l - we-ia)->.-ix.
n =0

Here 0 < a < 1rand A. > 0. These polynomials are orthogonal in the interval
- oo < x < oo with the weight function
(4.2)
Laguerre polynomials appear as a limiting case. Indeed, replacing x by x/ a
and assuming a--+ 0 we obtain
ix 1 - weia I
lim (1 - weia)->-(1 - we-ia)->- exp -log _. \
a-->0 {a 1 - We Ia J

= (1 - w)- 2 >- exp (2x w )

1-w
so that, cf. (5.1.9),
(4.3) limP~"l(x/a;a) = £<fl(-2x), {3 = 2>.- I.
a -->0

It is also clear that the polynomials in 2. arise from P~"l(x; a) as follow::;:

(-!.4) p~hl(h(O); 8) = P':l(cos 8; a, b).
5. Asymptotic properties
(a) By means of the generating function (1.1) it is not difficult to obtain an
asymptotic expression for Pn(x; a, b) when n--+ oo. We may use Darboux'::;
method (§8.4).
First let x be outside of the closed interval [ -1, + 1]. Writing x = ! (z + z- 1 ),
z == ei , 30 > 0, we find
8

Pn(x; a, b) = I r(! + ih(O))l-1(1 - e2i8)-l+ih(8)

(5.1) . e-i"8n-l+ih(8) ( 1 + o(~)).

It is not difficult to extend this to an asymptotic expansion. NO\v let
-1 < x <+I; forming the real part of the right-hand expression in (5.1),
396 APPENDIX

the asymptotic formula for !Pn(x; a, b) arises. Finally, in view of (1.5) we

find from (8.22.3)
(5.2) Pn(1; a, b)""' !-n·-le-l<a+bl(a + b)-in-1 · exp l2(a + b)lnlj.
(b) Novikoff (1) investigated the asymptotic behavior of Pn(cos (tn-l);a, b)
where t > 0 is fixed. His principal results are as follows:
Pn(cos (tn-l); a, b) = !1r-l(a + b- t 2 )-1 exp (-!(a+ b))

. n-i exp {nl (~at b + A(t))} (1 + o(~)),

(5.3)

O<t< (a+b)l;
Pn(cos (tn-l); a, b) = ?r-l(t 2 - a- b)-1 exp (-!(a+ b))

· n-1 exp { nl ( ~a t b)} . {cos ( ~- ~

nl ( t)) + 0 ( ~)} ,
(5.4)

t>(a+b)l.
From (5.3) and (5.4) interesting conclusions can be drawn about the "extreme"
zeros of P n ( x; a, b). Let us denote the zeros of these polynomials, as in Chap-
ter VI, by cos e. where 0 < 81 < 82 < · · · < On < 1r; e. = Ovn. Then, for any
fixed value of v,
(5.5) lim nl()vn = (a
n-->ao
+ b )l.
6. Associated orthogonal polynomials
(a) We consider the system l cl>n (z) I of polynomials which are orthogonal
on the unit circle lz! = 1 relative to the weight function
(6.1) f(O) = w(cos O)isin Ol;

here w (x) has the same meaning as in ( 1. 9). These polynomials show also in
many respects a singular behavior.
The relation to the Pollaczek polynomials can be established by means of
the formulas (11.5.2):

(6.2)
z-•ql,, (z) f -(~)'{I ± ql~,;o T p,(x)

z-n+lcJ>2n-1(z)) +Hz- ~-l) (~Y{ 1 4= c/> 2;LO)r Qn-t(X).

 APPENDIX 397

We haVB Pn(x) = !n +!(a+ 1)llPn(x; a, b) [cf. (1.8)]; the system lqn(x)l i:s
orthonormal with the weight function (I - x 2 )w(x ). Moreover, K 2n denotes
the (positive) highest coefficient of cP2n(z). Between x and z the relation
x = ~(z + z- 1 ) holds.
(b) Let lzl < 1. We investigate the asymptotic behavior of the polynomials
cf>n(z) as n--+ oo. Let z r= 0, z = ei 8 , 38 > 0. We rewrite (5.1) in the form

Pn(x) = A(x)nih(Olz-"(1 +0(~)),

(6.3)
A(x) == lr(~ + ih(O))I-1(1- z2)-l+ih<Bl.
Now the polynomials l q" (x) I can be represented in terms of l Pn (x) I by using
Christoffel's formula 2.5, taking (1.5) into account. Thus
Pn-1(x; a, b) Pn(x; a, b) Pn+1(x; a, b)

L~-1 -L~

L" = Ln( -a -b), L~ = Ln( -a +b).

Denoting the latter determinant by ~" (x) we find by an easy calculation:

(1 - x 2)qn-1(x) = (Ln-1L~ + LnL~_ 1 )-l(LnL~+ 1 + Ln+1L~)-l

·(~:::YIn- 1 +!(a+ l)ll~n(x).

Using (1.5), (8.22.3), and (1.6) we obtain
(1 - x 2)qn-1(x) = ~A(x)nih(Olz-n- 1

(6.4) · {( 1 + ~: ~ E') z2 + (E- E' )z - (1- E~ E')} ( 1 + 0 ( ~)) ,

E= (a-t-b)ln-l, 1
E = (a - b )in-~.
(c) From (6.3) and (6.4) we conclude, since 1 - x 2 = - (1 - z 2 ) 2 (2z)- 2 ,
!(z- z- 1)qn-1(x) _
Pn(X) -
(6.5)

Denoting by k~ :;:= In+ !(a+ 1) Ilk" the coefficient of X" in Pn(x) and by
1 the coefficient of x"- 1 in qn-1(x), we find from (6.5) for z--+0:
[ 11 _

------- ·-----·---------------------------·--·······--···.
 398
APPENDIX

Now the first formula in (6.2) yields:

(6.7)
+E (E n1)) .
1
_ (211· )l a/ 2 (
-r(!(a+1))n -2-+0

In particular,
(6.8)

(d) We obtain from (6.2), lzl < 1, in view of (6.3), (6.5), (6.8),

cf> 2 n(z) = (~Y A (x)nih(O) { 1 + E ~ E' + ( 1 - E ~ E')

(6 _9 ) . (1 + !(E + E ))z + iE
1 2
= ;~)z- (1- !(E + E
1
))}(
1+ o(~))

(~Y + o(~))'
2
= (lrC{j(: i)a(z) n-r(z) I (a+ b)l(1 + z) +(a- b)i(l- z)l (1

cf> 2 n_ 1 (z) = (~Y A(x)nih(Olz- 1

{ 1 - E ~ E' + ( 1 + E ~ E')
(6 .1Q) . (1 + !(E +
1 2
E))z + iE = ;:)z- ( I - !(E +E
1
))}( 1+ o(~))

= (~2) (1 - z2 )a(z)
r({j(z))
n-r(z) l (a+ b)l(1 + z)- (a- b)l(1- z)l (1 + o(~))
n
where we set for abbreviation:

)- _ Ha3) + bz + Ha + 3 )z 2
~
a( z 1 -z 2 '

R( ) = !(a + 1) + bz + t(a - l)z 2

(6.11) fJZ 1 -z 2 '

_ t(a - 1) + bz + !(a + l)z 2

- (y)z - 1 -z 2

____________________________________._____ , , __ ______
, ,_
 APPENDIX 399

The argument of the r--function can be written as follows:

a 1 +z 2
z 1
(6.12) [j(z) = 21- z2 +b1- z2 +2
and the real part of this expression is >! for \z\ < 1. Hence the function 1/ r
is never zero in \z\ < 1. The only point of accumulation of the zeros of the poly-
nomials I<Pm(z) I in \zl < 1 is the point
(a + b)l
- (a - b )l
~~~7---~
(a 2 - b2 )l - a
= ~--~~--
(a+b)l+(a-b)l b
appearing only for m = 2n - 1.
The exponent 'Y(z) of n has a real part > -!. HeRce the series l:l<t>m(z)l2
is divergent for all lz\ < 1. In particular,

</>2n(O)l
<P2n-t(O)J
(6.13)
~ (~Y 1r(!(a + 1)) ,-lnl<a-li((a + b)l ±(a_ b)l) ( 1 +a(~)).
Taking (11.3.6) into account, this yields again the main term of (6.6).
(e) Recapitulating, we may point out certain properties of the Pollaczek
polynomials whieh indicate a rather singular be~avior compared with the clas-
sical polynomials.
The weight functions w(x) or f(8) in Theorems 12.1.1 and 12.1.2, respeCtively,
are such that log w(cos 8) and logf(8) are integrable. The weight function w(x)
of the Pollaczek polynomials vanishes at the endpoints x = ± 1 so strongly
that log w(cos 8) is not integrable [cf. (1.10)].
The normalized .Jacobi polynomials are at x = ± 1 of the order na+! and
ntl+l, respectively. The orthonormal Pollarzek polynomials at x = ± 1 are of
order n! exp l2(a + b)!n!j [cf. (5.2)].
The Toeplitz minima .Un (f) (12.3) associated with the weight functionf(8) tend
to a positive limit under the assumption of Theorem 12.1.1. In case of the poly-
nomials discussed in 6. this limit is zero; the weight function defines a "determin-
istic" process. We have in this ease: ~n(f) = K;; 2 ""'n-a. Let f(8) = 0 in a
certain interval, say -E < 8 < +E, 0 < E < 1r and f(8) = 1 otherwise. We
have then again .un(f)----> 0, and, more preeisely, !lnU) ""'rn, r < 1; cf. (16.4.3)
and Problem 50.
The orthonormal polynomials defined in Theorem 12.1.2 are asymptotically
of the order (x + (x 2 - 1 )l )n if x is not on the cut [ -1, + 1] ). The Pollaczek
polynomials are under the same condition of the order nK(x + (x 2 - 1 )l)n
where K is a function of x. A similar discrepancy arises if x _is on the segment
[-1, +I].
For the "largest" zeros eos 8., 0 < 8, < 1r, 8, = 8,(n ), v fixed, n----> oo, of the
.Jaeobi polynomials we have 8,(n) ""n- 1J, "·here j. is the corresponding zero of

---------------------------------·-----------··"-"'•·-··
400 APPENDIX

an appropriate Bessel function; see (6.3.15 ). The similar zeros of the Pollaczek
polynomials satisfy the relation O.(n) ""'n-! (a + b )l, see (5.5); the order of
magnitude is different nnd the constant does not depend on v.
 LIST OF REFERENCES73
ABEL, N.H.
*1. Oeuvres Completes. Vol. 2, 1881.
ACHIESER, N. I.
1. Uber cine Eigenschafl der "elh'ptischen" Polynome. Communications de Ia Societe
Mathematique de Kharkoff, (4), vol. 9 (1934), pp. 3-8.
2. Verallgemeinerung einer Korkine-Zololareffschen Minimum-Aufgabe. Ibid., (4),
vol. 13 (1936), pp. 3-14.
ADAMOFF, A.
1. On the asymptotic expansion of the polynomials Un(x) =;: eax 2 12 dn[e-ax 2 12 ]jdxn for
large values of n (in Russian). Annals of the Polytechnic Institute of St. Peters-
burg, vol. 5 (1906), pp. 127-143.
2. Expansions of an Arbitrary Function of a Single Real Variable in Series of Functions
of a Preassigned Kind (in Russian). Thesis. St. Petersburg, 1907, 191 pp.
BANACH, S.
*1. Theorie des Operations Lineaires. W arszawa-Lw6w, 1932.
BERNSTEIN, S.
*1. Le~ons sur les Proprieles Exlremales et la Meilleure Approximat1'on des Fonclions
Analytiques d'une Variable Reelle. Paris, 1926.
2. Surles polynom~s orlhogonaux relalifs dun segment jim'. Journal de Mathematiques,
(9), vol. 9 (1930), pp. 127-177; vol. 10 (1931), pp. 219-286.
3. Sur une classe de polynomes orlhogonaux. Communications de la Societe Mathe-
matique de Kharkoff, (4), vol. 4 (1930), pp. 79-93. Complement. Ibid., vol. 5
(1932), pp. 59-60.
4. Sur la limitation des val.eurs d'un polynome P,.(x) de degre n sur tout un segment par
ses valeurs en (n + 1) points du segment. Bulletin de l' Academie des Sciences
de l'URSS, 1931, pp. 1025-1050.
BLU:\IENTHAL, 0.
1. Ueber die Entwicklung einer willkiirlichen Funktion nach den Nennern des Kellen-
bruches fur I~<X> [<t>(~)/(z - md~. Inaugural-Dissertation. Gottingen, 1898.
BocHNER, S.
1. Uber Sturm-Liouualesche Polynomsysleme. Mathematische Zeitschrift, vol. 29
(1929), pp. 730-736.
BOTTEMA, 0.
1. Die Nullslellen der Hermiteschen Polynome. Koninklijke Akademie van Weten-
schappen te Amsterdam, Proceedings, vol. 33 (1930), pp. 495-503.
2. Die Nullstellen gew1'sser durch Rekursionsformeln defim'erlen Polynome. Ibid.,
vol. 34 (1931), pp. 681-691.
BRAUER, A.
1. Uber die Nullstellen der Hermileschen Polynome. Mathematische Annalen, vol. 107
(1932), pp. 87-89.
BRUNs, H.
1. Zur Theorie der Kugelfunctionen. Journal fur die reine und angewandte Mathe-
matik, vol. 90 (1881), pp. 322-328.
B'UELL, c. E.
1. The zeros of Jacobi and related polynomials. Duke Mathematical Journal, vol. 2
(1936), pp. 304-316.
73
The asterisks indicate items not dealing with orthogonal polynomials.
401
402 LIST OF REFERENCES

CARATHEODORY, c.
*1. Conformal Representation. Cambridge, 1932.
CARLEMAN, T.
1. Vber die Approximation analytischer Funktionen durch linearc Aggregate von vorge-
gebenen Potenzen. Arkiv for Matematik, Astronomi och Fysik, vol. 17 (1923),
no. 9, 30 pp.
CHRISTOFFEL, E. B.
1. Ober die Gaussische Quadratur und eine Verallgemeinerung derselben. Journal fur
die reine und angewandte Mathematik, vol. 55 (1858), pp. 61-82.
CRAME:R, H.
1. On some classes of series used in mathematical statistics. Comptes Rcndus du Sixi-
cme Congrcs des Mathematiciens Scandinaves, Stockholm, 1926, pp. 399-425.
DARBOUX, G.
1. Memoire sur l'ap1Jroximation des fonctions de tres grands nombres. Journal de
Mathematiques, (3), vol. 4 (1878), pp. 5-56, 377-416.
DIRICHLET, G. L.
1. Surles series dont le terme general depend de deux angles, et qui servent a exprimer des
fonctions arbitraires entre des limites donnees. Journal fi.ir die reine und angc-
wandte Mathematik, vol. 17 (1837), pp. 35-56.
DoETSCH, G.
1. lntegraleigenschaften der Hermiteschen Polynome. Mathematische Zeitschrift,
vol. 32 (1930), pp. 587-599.
2. Die in der Statist£k seltener Ereignisse auftretenden Charlierschen Polynome und
eine damit zusammenhiingende Differentialdifferenzengleichung. Mathematischc
Annalen, vol. 109 (1933), pp. 257-266.
Du Bois-REYMOND, P.
*1. Untersuchungen iiber die Convergenz und Divergenz der Fourierschen Darstellungs-
formeln. Abhandlungen der Akademie Munchen, vol. 12 (1876), pp. 1-103.
ERDELYI, A.
1. Ober eine lntegraldarstellung der Mk.m-Funktionen und ihre asjjmptotische Darstel-
lung filr grosse Werte von V?e k. Mathematische Annalen, vol. 113 (1936), pp.
357-362.
2. Ober eine erzeugende Funktion von Produkten Hermitescher Polynome. Mathe-
matische Zeitschrift, vol. 44 (1938), pp. 201-211.
ERDOs, P., and FELDHEIM, E.
1. Sur le mode de convergence pour l'interpolation de Lagrange. Comptes Rendus de
I' Academie des Sciences, Paris, vol. 203 (1936), pp. 913-915.
ERDOS, P., and TuR.A.N, P .
1. On interpolation. I. Quadrature- and mean-convergence in the Lagrange interpola-
tion. Annals of Mathematics, (2), vol. 38 (1937), pp. 142-155.
2. On interpolation. II. On the distribution of the fundamental points of Lagrange
and Hermite interpolation. Ibid., vol. 39 (1938), pp. 703-724.
EuLER, L.
*1. lnstitutiones Calculi lntegraHs. Vol. 2, sect. I, chap. X, problem 130. Opera
Omnia. Ser. 1, vol. 12, p. 224.
FABER, G.
1. Ober polynomische Entwicklungen. Mathematische Annalen, vol. 57 (1903),
pp. 389-408.
*2. Ober die interpolatorische Darstellung stetiger Funktionen. J ahresberich t der
Deutschen Mathematiker-Vereinigung, vol. 23 (1914), pp. 192-210.
3. Tschebyscheffsche Polynome. Journal fi.ir die reine und angewandte Mathematik,
vol. 150 (1919), pp. 79-106.
 LIST OF REFERENCES 403

4. Uber nach Polynornen fortschreitende Reihen. Sitzungsbcrichte der Bayrischen

Akadcmic dcr Wisscnschaftcn, 1922, pp. 157-178.
FATOU, P.
*1. Series trigonometriques et series de Taylor. Acta Mathematica, val. 30 (1906),
pp. 335-400.
FAYARD, J.
1. Sur les polynomes de Tchebicheff. Comptes Rend us de l' Academic des Sciences,
Paris, vol. 200 (1935), pp. 2052-2053.
FEJER, L.
*1. Sur les fonctions bornees et integrables. Comptes Rend us de l' Academic des Sci-
ences, Paris, vol. 131 (1900), pp. 984-987.
*2. Untersuchungen uber Fouriersche Re1'hen. Math~matische Annalen, vol. 58 (1904),
pp. 51-69.
3. Asymptotikus ertekek meghatdrozrisdr6l. Mathematikai es Termcszettudom:inyi
l~:rtcsito, vol. 27 (1909), pp. 1-33.
4. Uber die Laplacesche Reihe. Mathcmatischc Annalen, vol. 67 (1909), pp. 76-109.
*5. Ober tn'gonometrische Polynome. Journal fur die rcinc und angcwandte Mathe-
matik, vol. 146 (1915), pp. 53-82.
6. Uber Interpolation. N a.chrichtcn dcr Gcscllschaft der Wissenschaftcn zu Gottingen,
1916, pp. 66-91.
7. Ober die Lage der Nullstellen von Polynomen, die aus Minimumforderungen gewisser
Art entspringen. Mathcmatischc Annalen, vol. 85 (1922), pp. 41-48.
8. Uber die Summabilitiit der Laplaceschen Reihe durch arithmetische Mittel. Mathe-
matische Zeitschrift, vol. 24 (1925), pp. 267-284.
9. Abschiitzungen fur die Legendreschen und verwandte Polynome. Mathematische
Zeitschrift, vol. 24 (1925), pp. 285-298.
10. Uber Weierstrass'sche Approximation, besonders durch Hermitesche Interpolation .
.Mathcmatische Annalen, vol. 102 (1930), pp. 707-725.
11. Die Abschatzung eines Polynoms in einem Intervalle, wenn Schranken fur seine Werle
und ersten Ableitungswerte in einzelnen Punkten des Intervalles gegeben sind, und
ihre Anwendung auf die Konvergenzfrage Hermitescher Interpolationsreihen.
Mathcmatischc Zeitschrift, vol. 32 (1930), pp. 426-457.
12. Ultraspharikus polynomok osszegerol. Matcmatikai es Fizikai Lapok, vol. 38
(1931), pp. 161-164.
13. Lagrangesche Interpolation und die zugehiirigen konjugierten Punkte. Mathe-
matische Annalen, vol. 106 (1932), pp. 1-55.
14. Bestimmung derjenigen Abszissen eines Intervalles, fur welche die Quadratsumme der
Grundfunktionen der Lagrangeschen Interpolation im Intervalle ein moglichst
kleines Maximum be8itzt. Annali della Scuola N ormale Superiore di Pisa, (2),
vol. 1 (1932), pp. 3-16.
15. ili echanische Quadraturen mit positiven Cotesschen Zahlen. Mathematischc Zci t-
schrift, vol. 37 (1933), pp. 287-309.
16. On the characterization of some remarkable systems of points of interpolation by means
of conjugate pm'nts. American Mathematical Monthly, vol. 41 (1934), pp. 1-14.
17. Potenzreihen mit rnehr fach mono toner [{ oeffiz·ientenfolge und ihre Legendre Polynome.
Proceedings of the Cambridge Philosophical Society, vol. 31 (1935), pp. 307-316.
18. A hatvrinysorr6l es a vele kapcsolatos Legendre-fele tobbta(J'Ilakr6l. Mathematikai
cs Termeszettudom:inyi Ertesito, vol. 53 (1935), pp. 1-17.
19. Bestimmung von Grenzen fur die Nullstellen des Legendreschen Polynoms aus der
Stieltje.~schen Integraldarstellung desselben. Monatshefte fur Mathematik und
Physik, vol. 43 (1936), pp. 193-209.
20. Trigonometrische Reihen und Potenzreihen mit mehrfach monotoner Koejfizientenfolge.
Transactions of the American Mathematical Society, vol. 39 (1936), pp. 18-59.
 404 LIST OF REFERENCES

FEJEm, L., and SzEoo, G.

*1. Uber die monotone Konvergenz von Potenzreihen mit mehrfach monotoner Koejfizien-
tenfolge. Prace Matematyczno-Fizyczne, vol. 44 (1935), pp. 15-25.
FEKETE, M.
*1. Uber die Verteilung der Wurzeln bei gew1"ssen algebraischen Gleichungen m1"t ganzzah-
ligen Koeffizienten. Mathematische Zeitschrift, vol. 17 (1923), pp. 228-249.
FELDHEIM, E.
1. Sur l'orthogonalite des fonctions fondamentales de l'interpolation de Lagrange.
Comptes Rend us de l' Academie des Sciences, Paris, vol. 203 (1936), pp. 650-652.
2. Sur le mode de convergence dans Nnterpolation de Lagrange. Comptes Rendus de
l'Academie des Sciences de l'URSS, vol. 14 (1937), pp. 327-331.
3. Sur l'orthogonaliU: des fonctions fondamentales, et sur la forte convergence en moyenne
des polynomes d'interpolation de Lagrange dans le cas des abscisses de Tchebychef.
Bulletin de la Societe Mathcmatique de France, vol. 65 (1937), pp. 1-40.
4. Theorie de la convergence des procedes d'interpolation et de quadrature mecanique.
Memorial des Sciences Mathematiques, vol. 95, Paris, 1939.
FRIEDRICHS, K.
1. On certain inequalities and characteristic value problems for analytic fu'~'~ctions and for
functions of two variables. Transactions of the American Mathematical Society,
vol. 41 (1937), pp. 321-364.
FuJiwARA, M.
1. On the zeros of Jacobi polynomials. Japanese Journal of Mathematics, vol. 2 (1925),
pp. 1-2.
GALBRUN, H.
a
1. Sur un developpement d'une fonction variable reelle en series de polynomes. Bul-
letin de la Societe Mathematique de France, vol. 41 (1913) 1 pp. 24-47.
GAUSS, c. F.
*1. Summatio quarumdam serierum singularium. Werke. Vol. 2, pp. 9-45.
2. Methodus nova integralium valores per approximationem inveniendi. Werke. Vol.
3, pp. 163-196.
GEGENBAUER, L.
1. Uber einige bestimmte lntegrale. Si tzungsberichte der mathematisch-natur-
wissenschaftlichen Klasse der Akademie der Wissenschaften in Wien, Abteilung
Ila, vol. 70 (18'74), pp. 433-443.
2. Ober d1"e Functionen C~(x). Ibid., vol. 75 (1877), pp. 891-905.
3. Uber die Functionen C~(x). Ibid., vol. 97 (1888), pp. 259-270.
4. Zur Theorie der hypergeometrischen Reihe. Ibid., vol. 100 (1891), pp. 225-244.
5. Das Additionstheorem der Functionen C~(x). Ibid., vol. 102 (1893), pp. 942-950.
6. Zur Theorie der Functionen C~(x). Denkschriften der Akademie der Wissen-
schaften in Wien, Mathematisch-naturwissenschaftliche Klasse, vol. 48 (1884),
pp. 293-316.
7. Einige Satze ilber die Functionen C~(x). Ibid., vol. 57 (1890), pp. 425-480.
G(u)ERONIMus, J.
1. Sur le polyn6me multiplement monotone qui s'ecarte le mains de zero, dont un co-
efficient est dtmne. Bulletin de l' Academie des Sciences de l'URSS, 1929, pp.
377-389.
2. Sur l' ecart minimal quadrati que de zer:o d'un polynome. Rendiconti del Circolo
Matematico di Palermo, vol. 54 (1930), pp. 298·-313.
3. On some problems of Tchebycheff. American Journal of Mathematics, vol. 53 (1931),
pp. 597-604.
4. On a problem of M. J. Shohat. Ibid., vol. 54 (1932), pp. 85-91.
5. On some extremal properties of polynomials. Annals of Mathematics, (2), vol. 37
(1936), pp. 483--517.

________________________________________________________
, .__.
 LIST OF REFERENCES 405

GOTTLIEB, M. J.
1. Concerning some polynomials orthogonal on a finite or enumerable set of points.
American .Journal of Mathematics, vol. 60 (1938), pp. 453-458.
GRONWALL, T. H.
1. Ober die Laplacesche Reihe. Mathematische Annalen, vol. 74 (1913), pp. 213-270.
2. Uber die Summierbarkeit der Reihen von Laplace und Legendre. Ibid., vol. 75 (1914),
pp. 321-375.
GRUNWALD, G.
1. Uber D1'vergenzerscheinungen der Lagrangeschen Interpolatt'onspolynome stetiger
Funktionen. Annals of Mathematics, (2), vol. 37 (1936), pp. 908-918.
GRuNWALD, G., and TuRA.N, P.
1. Uber Interpolation. Annali della Scuola Normale Superiore de Pisa, (2), vol. 7
(1938), pp. 137-146.
HAAR, A.
*1. Zur Theorie der orthogonalen Funktionensysterne. I. Mathematische Annalen,
vol. 69 (1910), pp. 331-371.
2. Reihenentwicklungen nach Legendreschen Polynomen. Ibid., vol. 78 (1917), pp.
121-136.
HAHN, w.
1. Die Nullstellen der Laguerreschen und Hermiteschen Polynome. Inauguraldisserta-
tion, Berlin. Schriften des Mathematischen Seminars und des Instituts fi.ir
angewan.dte Mathematik der Universitat Berlin, vol. 1 (1933), pp. 213-244.
2. Bericht ilber die Nullstellen der Laguerreschen und der Hermiteschen Polynome.
Jahresbericht der Dcutschen Mathemp.tiker-Vereinigung, vol. 44 (1934), pp. 215-
236. Nachtrag. Ibid., vol. 45 (1935), p. 211.
3. Uber die Jacobischen Polynome und zwei verwandte Polynomklassen. Mathe-
matische Zeitschrift, vol. 39 (1935), pp. 634-638.
4. Uber hOhere Ableitungen von Orthogonalpolynomen. Ibid. 1 vol. 43 (1937), p. 101.
HAMBURGER, H.
1. Beitrage zur Konvergenztheo!ie der Stieltjesschen Kettenbrilche. Mathematische
Zeitschrift, vol. 4 (1919), pp. 186-222.
2. Uber eine Erweiterung des Stieltjesschen Momentenproblems. Mathematische An-
nalen, vol. 81 (1920), pp. 235-319; vol. 82 (1920), pp. 120-164, 168-187.
HARDY, G. H.
1. SummaNon of a serie<; of polynomials of Laguerre. Journal of the London Mathe-
matical Society, vol. 7 (1932), pp. 138-139, 192.
HARDY, G. H., LITTLEWOOD, J. E., and P6LYA, G.
*1. InequaliNes. Cambridge, 1934.
HAUSDORFF, F.
*1. Momentprobleme fur ein endliches Intervall. Mathematische Zeitschrift, vol. 16
(1923), pp. 220-248.
HEINE, E.
1. Mittheilungilber Kettenbrilche. Journal fUr die reine und angewandte Mathematik,
vol. 67 (1867), pp. 315-326.
2. Die Fourier-Besselsche Function. Ibid., vol. 69 (1869), pp. 128-141.
3. Handbuch der Kugelfunct1'onen. Vols. I, II. 2d edition. Berlin, 1878, 1881.
HELLY, E.
*1. Ober lineare Funktionaloperationen. Sitzungsberichte der mathematisch-natur-
wissenschaftlichen Klasse der Akademie in Wien, Abteilung Ila, vol. 121 (1912),
pp. 265-297
HERMITE, c.
1. Sur les polyn6mes de Legendre. Rendiconti del Circolo Matematico di Palermo,
vol. 4 (1890), pp. 146-152. Oeuvres. Vol. 4, pp. 314-320.

·----------·----·-·-·
40() LIST OF REFERENCES

2. Sur les polynomes de Legendre. Journal fi.ir die reine und angewandte Mathe-
matik, vol. 107 (1891), pp. 80-83. Oeuvres. Vol. 4, pp. 321-326:
3. Sur les racines de la fonction spherique de seconde espece. Annales de la. Facultc
des Sciences de Toulouse, vol. 4 (1890), 10 pp. Oeuvres. Vol. 4, pp. 327-336.
HERMITE, C., and STIELTJEs, T. J.
1. Correspondance d'Hermile et de Stieltjes. Vols. I, II. Paris, 1905.
HILB, E.
1. Ober die Laplacesche Reihe. Mathematische Zeitschrift, vol. 5 (1919), pp. 17-25;
vol. 8 (1920), pp. ~79-90.
HILBERT, D.
1. Ober die Discriminante der im Endlichen abbrechenden hypergeometrischen Reihe.
Journal fi.ir die reine und angewandte Mathematik, vol.103 (1888), pp. 337-345.
HILBERT, D., and CouRANT, R.
1. M ethoden der mathematischen Physik. Vol. 1. 2d edition. Berlin, 1931.
HILDEBRANDT, T. H.
*1. On integrals related to and extensions of the Lebesgue integrals. Bulletin of the
American Mathematical Society, vol. 24 (1918), pp. 113-144, 177-202.
HILLE, E.
1. A class of reciprocal functions. Annals of Mathematics, (2), vaL 27 (1926), pp.
427-464.
2. On Laguerre's serie~:. I, II, III. Proceedings of the National Academy of Sci-
ences, vol. 12 (1926), pp. 261-265, 265-269, 348-352.
3. Bemerkung zu einer Arbeit des Herrn Muntz. Mathematische Zeitschrift, vol. 32
(1930), pp. 422-425.
4. Ober die Nnllstellen der Herm1'teschen Polynome. Jahresbericht der Deutschen
Mathematiker-Vereinigung, vol. 44 (1933), pp. 162-165.
HoBsoN, E. W.
1. The Theory of Sphe1·ical and Ellipsoidal Harmonics. Cambridge, 1931.
HOLL6, A.
1. A mechanikus quadraturar6l. Thesis. Budapest, 1939, 23 pp.
JACKSON, D.
1. On functions of closest approximation. Transactions of the American Mathematical
Society, vol. 22 (1921), pp. 117-128.
2. Note on a class of polynomials of approximation. Ibid., vol. 22 (1921), pp. 320-326.
3. A generalized problem in weighted approximation. Ibid., vol. 26 (1924), pp. 133-154.
4. The Theory of Approximation. American Mathematical Society Colloquium Publi-
cations, vol. 11, 1!}30.
1>. Series of orthogonal polynomials. Annals of Mathematics, (2), vol. 34 (1933), pp.
527-545.
6. Certain problems of closest approximation. Bulletin of the American Mathematical
Society, vol. 39 (1933), pp. 889-906.
7. The summation of se1·ies of orthogonal polynomials. Ibid., vol. 40 (1934), pp. 743-752.
8. Pormal propertie.~ of orthogonal polynomials in two var·iables. Duke Mathematical
Journal, vol. 2 (1936), pp. 423-434.
JACOB, M. M.
1. Sullo sviluppo di una funzione di ripartizione in serie di polinomi di Hermite.
Giornale dell'Jstituto Italiano degli Attuari, vol. 2 (1931), pp. 100-106, 356-368.
2. Sur le phenomene de Gibbs dans les developpements de series de polynomes d' Hermite.
Comptes Rend us de I' Academic des Sciences, Paris, vol. 204 (1937), pp. 1540-
1543.
JACOBI, c. G. J.
1. Ueber Gauss' neue Methode, die Werthe der lntegrale niiherungsweise zu finden.
Journal fi.ir die reine und angewandte Mathematik, vol. 1 (1826), pp. 301-308.
Gesammelte Werke. Vol. 6, pp. 3-11.
 LIST OF REFERENCES 407

2. Uber die Entwickelung des A usdrucks (aa - 2aa' (cos w cos </> + sin w sin </>
·cos (0 - 0')) + a'a')- 112 Ibid., vol. 26, 1843, pp. 81-87. Gesammelte Werke.
Vol. 6, pp. 148-155.
3. Untersuchungen iiber die D1'jJerentialgleichung der hypergeometrischen Reihe. Ibid.,
vol. 56 (1859), pp. 149-165. Gesammelte Werke. Vol. 6, pp. 184-202.
JORDAN, CAMILLE.
1. Cours d'Analyse de l'Ecole Polytechnique. Vol. 3, 2d edition. Paris, 1896.
JoRDAN, CHARLEs.
1. Sur une ser-ie de polynomes dont chaque sornme partielle represente la meilleure
approximation d'un degre donne suivant la methode des moindres carres. Pro-
ceedings of the London Mathematical Society, (2), vol. 20 (1920), pp. 297-325.
JuLrA, G.
1. Sur les polynomes de Tchebichef. Comptes Rend us de I' Academie des Sciences,
Paris, vol. 182 (1926), pp. 1201-1202.
KACZMARZ, S'I'., and STEINHAus, H.
*1. Theorie der Orthogonalreihen. Warszawa-Lw6w, 1935.
KELDYSCH, M., and LAVRENTIEFF, M.
1. Sur la representation conforme des domaines limites par des courbes rectijiables.
Annales Scientifiques de I' Ecole N ormale Superieure, (3), vol. 54 (1937), pp. 1-38.
KLEIN, F.
1. Ueber die Nullstellen der hypergeometrischen Reihe. Mathematische Annalen,
vol. 37 (1890), pp. 573-590. Gesammelte Abhandlungen. Vol. 2, pp. 550-567.
KoGBETLIANTZ, E.
1. Sur les series de jonctions ultraspheriques. Comptes Rendus de l'Academie des
Sciences, Paris, vol. 163 (1916), pp. 601-603.
2. Sur la sommation des series ultraspheriques. Ibid., vol. 164 (1917), pp. 510-513,
626-628, 778-780; vol. 169 (1919), pp. 54-57.
3. Surles developpements de Jacobi. Ibid., vol. 168 (1919), pp. 992-994.
4. Surles series ultraspheriques. Ibid., vol. 169 (1919), pp. 322-324.
5. Nouvelle observations sur les series ultraspheriques. Ibid., vol. 169 (1919), pp.
423-426.
6. Sur l'unicite des developpements ultraspheriques. Ibid., vol. 169 (1919), pp. 769-
770, 950-953.
7. Sur les developpements de Jacobi. Ibid., vol. 172 (1921), pp. 1333-1334; vol. 192
(1931), pp. 915-918.
8. Sur la sommabilite (C, o) de developpements suivant les polynomes d' Hermite. Ibid.,
vol. 192 (1931), pp. 662-663.
9. Nouvelles observations sur le systeme orthogonal de polynomes d' Hermite. Ibid., vol.
192 (1931), pp. 1696-1698.
10. Surles series d'Hermite et de Laguerre. Ibid., vol. 193 (1931), pp. 386-389.
11. Sur la convergence des series d'Hermite. Ibid., vol. 194 (1932), pp. 161-163.
12. Sur les developpements de Laguerre. Ibid., vol. 194 (1932), pp. 1422-1424.
13. Sur la serie de Laguerre. Ibid., vol. 196 (1933), pp. 523-525.
14. Expression approchee du polynome de Laguerre L~o:) (x). Ibid., vol. 196 (1933),
pp. 1079-1080.
15. Sur la determination du saut D(xo) de f(x). Ibid., vol 196 (1933), pp. 464-466.
16. Uber die (C o)-Summierbarkeit der Laplaceschen Reihe fur 1/2 < o < 1. Mathema-
tische Zeitschrift, vol. 14 (1922), pp. 99-109.
17. Analogie entre les series trigonometriques et les series spheriques au point de vue di
leur sommabilite par les rnoyennes arithmetiques. These. Paris, 1923, 65 pp.
18. Sur la sommabilite de la serie ultraspherique a l'interieur de l'intervalle ( -1, + 1)
par la methode des moyennes arithmetiques. Bulletin de la Societe Mathematique
de France, vol. 51 (1923), pp. 244-295.
 408 LIST OF REFERENCES

19. Recherches sur la sommabilite des series ultraspheriques par la methode des moyennes
arithmetiques. Journal de Mathematiques, (9), vol. 3 (1924), pp. 107-187.
20. Recherches sur l'unicite.des series ultraspheriques. Tbid .. (9), vol. 5 (1926), pp.
125-196.
21. Sommation des series et integrales divergentes par les rnoyennes arithmet-iques et
typiques. Memorial des Sciences Mathcmatiques. Vol. 51. Paris, 1931.
22. Recherches sur la sommabilite des series d' II ermite. Annales Scientifiques de I 'Ecole
Normale Superieure, (3), vol. 49 (1932), pp. 137-221.
23. Sur les moyennes arithmetiques des series-noyaux des developpements en series d' II er-
mite et de Laguerre et sur celles de ces series-noyaux derivees termed terme. Journal
of Mathematics and Physics, Massachusetts Institute of Technology, vol. 14
(1935), pp. 37-99.
24. Contrib1ttion d l'etude du saut d'une fonction donnee par son developpement en series
d' Hermite ou de Laguerre. Transactions of the American Mathematical Society,
vol. 38 (1935), pp . 10-47.
KOROUS, J.
1. 0 rozvoji funkci jedne realne promenne v radn II ermiteovych polynomu. Rozpra vy
Geske Akademie, (2), vol. 37 (1928), no. 11, 34 pp.
2. 0 radach Laguerrovych polynomu (with French abstract). Ibid., no. 40, 23 pp.
3. 0 rozvoji funkci jedne realne promenne v radu jist'!)ch ortogonalnich polynomu (with
English -abstract). Ibid., vol. 48 (1938), 12 pp.
4. Uber Reihenentwicklungen nach verallgemeinerten Laguerreschen Polynomen mit
drei Parametern. Vestnik KraJovske Geske Spolecnosti Nauk, Tffda Matemat.-
Pffrodoved, 1937, 26 pp.
5. Ober Entwicklungen der Funktionen einer reellen Verlinderlichen in Reihen einer
gewissen Klasse orthogonaler Polynome im unendlichen Intervalle. Ibid., 1937,
19 pp.
KoscHMIEDER, L.
1. Uber besondere J acobische Polynome. Mathematischc Zeitschrift, vol. 8 (1920),
pp. 123-137.
KowALEWSKI, G.
*1. Einfuhrung in die Determinantentheorie. Leipzig, 1909.
KowALLIK, U.
1. Entwicklung einer willkiirlichen Funktion nach Hermiteschen Orthogonalfunktionen.
Mathematische Zeitschrift, vol. 31 (1930), pp. 498-518.
KRALL, H. L.
1. On derivatives of orthogonal polynomials. Bulletin of the American Mathematical
Society, vol. 42 (1936), pp. 423-428.
2. On higher derivatives of orthogonal polynomials. Ibid., vol. 42 (1936), pp. 867-870.
KRAWTCHOUK, M.
1. Sur une generalisation de8 polynomes d'Hermite. Comptes Rcndus de l'Academie
des Sciences, Paris, vol. 189 (1929), pp. 620-622.
2. Sur la distribution des racines des polynomes orlhogonaux. Ibid., vol. 196 (1933),
pp. 739-741.
KRONECKER, L.
*1. Zur Theorie der Elimination einer Variaheln aus zwei algebraischen Gleichungen.
Monatsberichte der Preussischen Akademie der Wisscnschaften zu Berlin, 1881,
pp. 535-600.
LAGRANGE, J. L.
1. Oeuvres. Vol. 1, pp. 534-539. 1867.
LAGUERRE, E. N.
1. Sur l'integrale f;' :~: -I e -x dx. Bulletin de la Societe Mathematique de France,
vol. 7 (1879), pp. 72-81. Oeuvres. Vol. 1, pp. 428-437.

------------------------------------------------····-··
 LIST OF REFERENCES

2. Sur l'approximation des fonctions circulaires au moyen des fonctions algebriques.

Comptes Rend us de l' .1\.cademie des Sciences, Paris, vol. 90· (1880), pp. 304-307.
Oeuvres. Vol. 1, pp. 104-107.
3. Sur les equations algebriques dont le premier membre satisfait a une equation
lineaire du second ordre. Ibid., vol. 90 (1880), pp. 809-812. Oeuvres. Vol. 1,
pp. 126-132.
LANGER, R. E.
*1. The asymptotic solutions of ordinary linear differential equations of the second order,
with special reference to the Stokes phenomenon. Bulletin of the American
Mathematical Society, vol. 40 (1934), pp. 545-582.
*2. The asymptotic solutions of certain linear ordinary differential equations of the
second order. Transactions of the American Mathematical Society, vol. 36
(1934), pp. 9Q-106.
*3. On the asymptotic solutions of ordinary differential equations, with reference to the
Stokes' phenomenon about a singular point. Ibid., vol. 37 (1935), pp. 397-416.
LAWTON, w.
1. On ihe zeros of certm:n polynomials related to Jacobi and Laguerre polynomials.
Bulletin of the American Mathematical Society, vol. 38, ( 1932 ), pp. 442-448.
LEBESGUE, H.
*1. Sur la divergence et la convergence non-uniforme des series de Fourier. Comptes
Rendus, de I' Academie des Sciences, Paris, vol. 141 (1905), pp. 875-877.
*2. Lerons sur les Series 'Trigonometriques. Paris, 1906.
LEGENDRE, A. M.
1. Exercices de Calcul Integral sur Divers Ordres de Transcendantes. Vol. 2. Paris,
1817.
LERoY, E.
1. Sur les series divergentes et les fonctions definies par un developpement de Taylor.
Annales de la Faculte des Sciences de Toulouse, (2), vol. 2 (1900), pp. 317-430.
L UK.ACS' F.
1. Verschiirfung des ersten Mittelwertsatzes der Integralrechnung fiir rationale Poly-
nome. Mathematisehe Zeitschrift, vol. 2 (1918), pp. 295-305.
2. Uber die Laplacesche Reihe. Ibid., vol. 14 (1922), pp. 250-262.
MARCINKIEWICZ, J.
*1. Quelques remarques sur l'interpolation. Acta Litterarum ac Scientiarum Regiae
Universitatis Hungaricae Francisco-Josephinae, vol. 8 (1937), pp. 127-130.
2 .. Sur la divergence des r.wlynomes d'interpolation. Ibid., vol. 8 (1937), pp. 131-135.
MARKOFF, A.
1. On some applications of algebraic continued fractions (in Russian). Thesis. St.
Petersburg, 1884, 12:1 pp.
2. Demonstration de certaines inegalites de M. Tchebycheff. Mathematische Annalen,
vol. 24 (1884), pp. 172-180.
3. Extrait d'une lettre. .Annales Scientifiques de l'Ecole Normale Superieure, (3),
vol. 2 (1885), p. 183.
4. Sur les racines de certaines equations (second note). Mathematische Annalen,
vol. 27 (1886), pp. 177-182.
5. Differenzenrechnung. Leipzig, 1896.
MEHLER, F. G.
1. Bemerkungen zur Theorie der mechanischen Quadraturen. Journal fi.ir die reine
und angewandte Mathematik, vol. 63 (1864), pp. 152-157.
2. Ueber die Entwicklung einer Function von beliebig vielen Variablen nach Laplace-
schen Functionen hOherer Ordnung. Ibid., vol. (\(\ (1866), pp. 161-176.

------------·-·---······
410 LIST OF REFERENCES

3. Ueber die Verthe£lung der statischen ElektricifiU in einem von zwei K ugelkalotten
begrenzten Korper. Ibid., vol. 68 (1868), pp. 134-150.
4. Ueber die Darstellung einer willkiirlichen Function zweier Variablen durch Cylinder-
functionen. Mathematische Annalen, vol. 5 (1872), pp. 135-140.
5. Notiz uber die D1:richlet'schen lntegralausdriicke fur die Kugelfunction Pn(cos o)
und uber eine analoge Integralform fiir die Cylinderfunction J(x). Ibid., vol. 5
(1872), pp. 141-144.
MEIXNER, J.
1. Orthogonale Polynomsysteme mit einer besonderen Gestalt der erzeugenden Funk-
tion. Journal of the London Mathematical Society, vol. 9 (1934), pp. 6-13.
2. Erzeugende Funktionen der Charlierschen Polynome. Mathematische Zeitschrift,
vol. 44 (1938), pp. 531-535.
MoECKLIN, E.
1. Asymptotische Entwicklungen der Laguerreschen Polynome. Commentarii Mathe-
matici Helvetici, vol. 7 (1934), pp. 24-46.
MUNTZ, c.
1. Uber die Potenzsummation einer Entwicklung nach Hermiteschen Polynomen.
Mathematische Zeitschrift, vol. 31 (1929), pp. 350-355.
MYLLER-LEBEDEFF, v.
1. Die Th.eorie der Integralgleichungen in Anwendung auf einige Reihenentwicklungen.
Mathematische Annalen, vol. 64 (1907), pp. 388-416.
NEUMANN, E. R.
1. Die Entwicklung willkUrlicher Funktionen nach den Hermiteschen und Laguerreschen
Orthogonalfunktionen auf Grund der Theorie der Integralgleichungen. Inaugural-
Dissertation. Breslau, 1912.
2. Beitriige zur Kenntnis der Laguerreschen Polynome. Jahresbericht der Deutschen
Mathematiker-Vereinigung, vol. 30 (1921), pp. 15-35.
0BRECHKOFF, N.
1. Sur la sommation de la serie ultraspherique par la methode des moyennes arithme-
tiques. Rendiconti del Circolo Matematico di Palermo, vol. 59 (1936), pp.
266-287.
2. Formules asymptotiques pour les polynomes de Jacobi et sur les series suivant les
memes polynom1~s. Annuaire de l'Universite de Sofia, Faculte Physico-Mathe-
matique, vol. 1 (1936), pp. 39-133.
PERRON, 0.
1. Uber das injinitar.e Verhallen der Koeffizienlen einer gewissen Potenzreihe. Archiv
der Mathematik und Physik, (3), vol. 22 (1914), pp. 329-340.
2. Uber das Verhalten einer ausgearteten hypergeometrischen Reihe be1: unbegrenztem
W achstum eines Parameters. Journal fi.ir die reine und angewandte Mathe-
matik, vol. 151 (1921), pp. 63-78.
3. Die Lehre von den Kettenbriichen. 2d edition. Leipzig, 1929.
*4. Algebra. Vols. I, II, 2d edition. Berlin, 1932, 1933.
PLANCHEREL, M., and RoTACH, W.
2
1. Sur les valeurs asymptotiques des polynomes d'Herrnite H,.(x) = ('-1)nez / 2
·dn(e-z 12)jdxn. Commentarii Mathemati~i Helvetici, vol. 1 (1929), pp. 227-254.
2

P6LYA, G.
1. Sur un theoreme de Stieltjes. Comptes Rend us de I' Academic des Sciences, Paris,
vol. 155 (1912), pp. 767-769.
2. Sur un algorithme tou.four.~ convergent pour obtenir les polynornes de rneilleure ap-
proximation de 'rchebychef pour une jonction continue quelconque. Ibid., vol. 157
(1913), pp. 840-843.
 LIST OF REFEIU<;NCES 411

*3. Uber die Nullstellen gewisser ganzer Fllnktionen. Mathematischc Zcitschrift, vol.
2 (1918), pp. 3.52·-383. .
4. Uber die Konvergenz von (Juadraturverfahren. Ibid., vol. 37 (1!J33), pp. 264-286.
P6LYA, G., and SzEnii, G.
1. Aufgaben und Lehrsdlze aus der Analysis. Vols. I, fl. Berlin, 1925.
2. Uber den transjinilen Durchmesser (Kapazitdtskon.~tante) von ebenen und rdumlichen
Punktmengen. .Journal fiir die reinc und angcwandtc Mathematik, vol. 165
~1931), pp. 4-49.
PoPOVICiu, T.
1. Sur la distribution des zeros de certains polynornes rninimisants. Bulletin d<:
l'Acadcmie Roumainc, vol. 16 (1934), pp. 214-217.
2. Sur certains prohleme.~ de maximum de Slieltjes. Bulletin Mathcmatique de Ia
Societe Roumainc des Sciences, vol. 38 (1936), pp. 73-96.
RAu, H.
1. Uber die Lebesgueschen Konstanten der Reihenentw1"cklungen nach Jacobischen
Pol.ynomen. Journal flir die rcine und angcwandtc Mathematik, vol. 161 (1929),
pp. 237-254.
2. Uher eine asyrnptotische Darstellung der .! acoiJischen Polynome durch Besselsche
Funktionen. Mathematische Zeitsehrift, vol. 40 (1936), pp .. 683-692.
RIESZ, F.
*1. Sur certains systemes singuliers d'equations integrales. Annales Scientifiqucs de
l'Ecole N ormalc Supcrieurc, (3), vol. 28 (1911), pp. 33-62.
*2. Uber die Randwerte einer analytischen Funktion. Mathematische Zeitschrift,
vol. 18 (1923), pp. 87-95.
RIEsz, M.
1. Eine trigonometrische .Tnterpolationsformel und einige Ungleichungen fiir Polynome.
Jahresbericht der Deutschen Mathematiker-Vereinigung, vol. 23 (1915), pp.
354-368.
2. Sur le probllm~e des moments. Arkiv for Matematik, Astronomi oeh Fysik, vol. 16
(1921), no. 12, 23 pp.; vol. 16 (1922), no. 19, 21 pp.; vol. 17 (1923), no. 16, 52 pp.
ROTACH, w.
1. Reihenentwicklungen einer willkiirlichen Funktion nach Herrniteschen und Laguerre-
schen Polynomen. Inauguraldissertation, Eidgenossische Technische Hoeh-
sch ule Zurich, 1925.
ScHMIDT, E.
1. Uber die Charlier-Jordansche Entwicklung einer willkiirlichen Funktion nach der
Poissonschen Funktion und ihren A bleitungen. Zeitschrift fur angewandte
Mathematik und Meehanik, vol. 13 (1933), pp. 139-142.
ScHuR, I.
1. Uber die Verteilung der Wurzeln IJei gewissen algebraischen Gleiclwngcn mit ganz-
zahligen Koeffizienten. l\Iathcmatisehc Zcitsrhrift, vol. 1 (1918), pp. 377-402.
2. AjJekt.lose Gleichungen in der Theorie der Laguerreschen und Hcrrniteschen Poly-
nome. Journal fur die reine und angcwandte Mathematik, vol. 165 (1931),
pp. 52-58.
Scuwro, N.
1. The asymptotic forms of the Hermite Q.nd Weber functions. Transactions of the
American Mathematical Society, vol. 37 (1935), pp. 339-362.
SEN, D. N., and RANnACHARIAR, V.
1. Generalized Jacobi polynomials. Bulletin of the American Mathematical Society,
vol. 42 (1936), pp. 901-908.
S!IEHMAN, .J.
1. On the numerators of the convergents of the Stieltjes continued fractions. Trans-
actions of the American Mathematical Society, vol. 35 (1933), pp. 64-87.

_____________________________________ ...
,_
 412 LIST OF REFERENCES

SHIBATA, K.
1. On the distribution of the roots of a polynomial .satisfying a certain differential equa-
tion of the second order. Japanese Journal of Mathematics, vol. 1 (1924), pp.
147-153.
SHOHAT, J.
1. On the· polynomial of the best approximation to a given continuous function. Bulletin
of the American Mathematical Society, vol. 31 (1925), pp. 509-514.
2. On a general formula in the theory of Tchebycheff polynomials and its applications.
Transactions of the American Mathematical Society, vol. 29 (1927), pp. 569-583.
3. On a certain formula of mechanical quadratures with non-equidistant ordinates.
Ibid., vol. 31 (1929), pp. 448-463.
4. On the polynomial and trigonometric approximation of measurable bounded func-
tions on a finite interval. Mathematische Annalen, vol. 102 (1929), pp. 157-175.
5. On interpolation. Annals of Mathematics, (2), vol. 34 (1933), pp. 130-146.
6. TheorieOenerale des Polynomes Orthogonaux de Tchebichef. Memorial des Sciences
Mathematiques.. Vol. 66. Paris, 1934.
7. On mechanical quadratures, in particular, with positive coefficients. Transactions
of the American Mathematical Society, vol. 42 (1937), pp. 461-496.
8. On the convergence properties of Lagrange interpolation based on the zeros of orthogonal
Tche~ycheff polynomials. Annals of Mathematics, (2), vol. 38 (1937), pp. 758-769.
SMIRNOFF, v. J.
1. Sur la theorie des polynomes orthogonaux d une variable complexe. Journal de Ia
Societe Physico-Mathematique de Leningrad, vol. 2 (1928), pp. 155-179.
2. Sur les formules de Cauchy et de Green et quelques problemes qui s'y rattachent.
Bulletin de I' Academie des Sciences de l'.URSS, 1932, pp. 337-372.
SMITH, E. R.
1. Zeros of the Hermitian polynom1'als. American Mathematical Monthly, vol. 43
(1936), pp. 354-:358.
SoNIN(E), N. J.
1. Recherches sur les fonctions cylindriques et le developpement des fonctions continues
en series. Mathematische Annalen, vol. 16 (1880), pp. 1-80.
2. On the precision of the determination of limiting values of definite integrals (in Rus-
sian). Zapiski Akademii Nauk, vol. 69 (1892), pp. 1-30.
SPENCER, v. E.
1. Asymptotic expressions for the zeros of generalized Laguerre polynomials and Weber
functions. Duke Mathematical Journal, vol. 3 (1937), pp. 667-675.
STEKLOFF, w.
1. Sur les expressions asymptotiques de certaines fonctions, definiel5 par les equations
differentielles lineaires du second order, et leurs applications au probleme du
developpement d'une fonction arbitraire en series procedant suivant les-dites !one-
lions. Communications de Ia Societe Mathematique de Kharkow, (2), vol. 10
(1907), pp. 97-200. Remarque complementairc, p. 201.
2. On approximate evaluation of definite integrals by means of formulas of mechanical
quadratures. I. Convergence of formulas of mechanical quadratures (in Russian).
Bulletin de I' Aeademie Imperiale des Sciences, Petrograd, (6), vol. 10 (1916),
pp. 169-186.
STIELTJES, T. J.
1. Quelques recherches sur la theorie des quadratures dites mechaniques. Annales
Scientifiques de !'Ecole Nor male Superieure, (3), vol. 1 (1884), pp. 409-426.
Oeuvres Completes. Vol. 1, pp. 377-394.
2. Note d ['occasion de la reclamation de M. Markoff. Ibid., (3), vol. 2 (1885), pp.
183-184. Oeuvr.es Completes. Vol. 1, pp. 430-431.

------------------------------------------------··- ..
 LIST OF REFERENCES 41 :~

3. Sur certains polyn6mes qui 1•erijient une equation differenHclle Hnea1"re du second
ordre et sur la theor·ie des functions de Lame. Acta Mathematica, vol. 6 (1885),
pp. 321-326. Oeuvres Completes. Vol. 1, pp. 434-439.
4. Sur quelque~> theorbnes d'algebre. Comptes Rendus de I' Academie des Sciences,
Paris, vol. 100 (1885), pp, 439-440, Oew•res Completes. Vol. 1, pp. 440--441.
5. Surles polynumes de Jacobi, Comptes Rcndus de !'Academic des Sciences, Paris,
vol. 100 (1885), pp. (\20-622, Oeuvres Completes, VoL 1, pp. 442-444.
(\, Surles radnes de ['equation X,.= 0. Acta Mathf:'matica, voL 9 (1886), pp. 385-400.
Oeuvres Compteles. Vol. 2, pp. 73-88.
7. Sur la valeur asymptotique de.~ polynornes de Legendre. Comptes Rendus de
I' Academie des Sciences, Paris, vol. 110 (1890), pp, 1026-1027. Oeuvres Com-
pletes. Vol. 2, pp. 234-235.
8. Sur les polyn1'imes de Legendre. Annales de Ia Faculte des Sciences de Toulouse,
vol. 4 (1890), 17 pp. Oeuvres Completes. Vol. 2, pp. 236-252.
9. Sur les racines de la fonction spherique de seconde espece. Annales de Ia Faculte
des Sciences de Toulouse, vol. 4 (1890), 10 pp. Oeuvres Completes. Vol. 2,
pp. 253-262.
10. Recherches sur les fractions continues, Comptes Rend us de I' Academie des
Sciences, Paris, vol. 118 (1894), pp. 1401-1403. Oeuvres Completes. Vol. 2, pp.
398-401.
11. Recherches sur les fractions continues. Ann ales de Ia Faculte des Sciences de
Toulouse, vol, 8 (1894), 122 pp.; vol. 9, 1895, 47 pp. Oeuvres Completes.
Vol. 2, pp. 402-566.
12. Sur certaines inegalites dues d M. P. Tchebychef. Article redige d'apres un manu-
script inedit. Oeuvres Completes. Vol. 2, pp. 586-593.
STONE, M. H.
1. Developments in Hermite polynomials. Annals of Mathematics, (2), vol. 29 (1928),
pp. 1-13.
*2. Linear Transformations ·in Hilbert Space and their Applications to Analysis.
American Mathematical Society Colloquium Publications, vol. 15, 1932.
SzA,sz, 0.
*1. Korldtos hatvanysorokn5l. Mathematikai es Termeszettudomanyi ErtesftO, vol.
43 (1926), pp. 504-520.
SZEGO, G.
1. A Hankel-jete jormdk~r6l. Mathematikai es Termeszettudom:inyi Ertesfto, vol.
3() (1918), pp. 497-53·8.
2. Kin Beitrag zur Theorie der Polynome von Laguerre und Jacobi. Mathematische
Zeitschrift, vol. 1 (1918), pp. 341-35Q.
3. Ober Orthogonalsysteme von Polynomen. Ibid., vol. 4 (1919), pp. 139-151.
4. Beitrage zur Theorie der Toeplitzschen Formen. II. Ibid., vol. 9 (1921), pp.
167-190.
5. Ober orthogonale Polynorne, die zu e1"ner gegebenen Kurve der komplexen Ebene
gehiiren. Ihid., voL 9 (1921), pp. 218-270.
6. 0 ber die Entw·ickelung e·iner analytischen Funktion nach den Polynomen eines
Ortlwyonal~>ystems. Mathematische Annalen, vol. 82, 1921, pp. 188-212.
7. Ober die Ranrlwerte analytischer Funktionen. Ibid., vol. 84 (1921), pp. 232-244.
8. Ober den asyrnptotischen A usdruck von Polynomen, die durch c·ine Orthogonalitats-
eigenschaft definiert sind. Ibid., vol. 86 (1922), pp. 114-139.
9. Uber die Entw·icklung einer willkurlichen Funktion nach den Polynomen eines
Orthogonalsyslems. Mathematischc Zeitschrift, vol. 12 (1921), pp. 61-94.
10. Beitrdge zur Theorie der Laguerreschen Polynome. I. Entwicklungssiitze. Ibid.,
vol. 25 (1926), pp. 87-115.

__ _____________________________
, ____
, ,,_,_,'
 414 LIST OF REFERENCES

11. Remerkungen zu einer Arbeil I'On Herrn Fejer uber die Legendresrhen Polynome.
Ibid., vol. 25 (1926), pp. 172-187.
12. Ein Beitrag zur Theorie der Thetafunktionen. Sitzungsberichte der Preussisehen
Akademie der \Yissenschaften, physikalish-mathematische Klasse, 1926, pp.
242-252.
13. J<oeffizienlenabschdt;oungen bei ebenen und rtiumlichen harnwrtilichen Enlwicklungen.
:\fathematische Annalen, vol. 96 (1927), pp. 601-632.
1--1. Ober gewisse Interpolationspolynonie, die zu den Jacobischen und Lagucrreschen
Abszissen gehorcn. l\Iathrmatische Zeitschrift, vol. 35 (1932), pp. 579-602.
15. Ober einige asymplotische Enl!cicklungen der Legenrlreschen Funktioncn. Proceed-
ings of the London :\Iathrmatiral Society, (2), vol. 36 (1932), pp. 427-450. ·
16. Ober eine von Herrn S. Bernstein herruhrende Abschi.itzung der Legendreschen P(!ly-
nome. :\Iathematische Annalen, vol. 108 (1933), pp. 360-369.
17. Asymplolische Entwicklungen der Jacobischen Polynome. Schriften der Konigs-
berger Gelehrten Gesellschaft, naturwissenschaftliche Klasse, vol. 10 (193:3),
pp. 35-112.
18. Bemerktmgen zu e·incm Satz von E. Schmidt ii.ber algebraische Gleichungen. Sitzungs-
berichte der Prcussisrhen Akademie der \Vissensrhaften, physikalish-mathema-
tisehe Klassf:', 19a4, pp. 3-15.
19. Ober gew1"sse orthogonale J>olynome, die zu einer oszillierenden Belegungsfunktion
gehoren. Math(•matische Annalen, vol. 110 (1934), pp. 501-513.
20. Inequalities for the zeros of Legendre polynomials and related functions. Trans-
artions of the American !\!athematical Society, vol. 39 (1936), pp. 1-17.
21. An 1'ntegral equation for the square of a Laguerre polynomial. .Journal of the London
l\Iathematical Society, vol. 12 (1937), pp. 162-163.
TAMARKIN, J.D.
1. On the Theory of Polynomials of Approximation. Lecture delivered at Brown
University, 1935-1936.
Tcn.'.KALOFF, L.
1. Sur la structure des ensembles lineail'es dejinis par nne certaine propriete minimale.
Acta :\fathematica, vol. 63 (1 934), pp. 77 -·97.
TCHEBICHEF, P. L.
1. Sur les fractions continues. Journal de :\Iath{~matiqucs, (2), vol. 3 (18.58), pp.
289-323. Oeuvres. Vol. 1, pp. 201-230.
2. Sur l'1'nterpolat-ion par la methode des rnoindres carres. ?liemoires de I' Acad<~mic
Imperiale des Sciences de St. PHersbourg, (7), vol. 1 (1859), pp. 1-24. Oeuvres.
Vol. 1, pp. 471-498.
3. Sur le developpement des functions d une seule variable. Bulletin de I' Academic
Imperiale des Sciences de St. Petcrsbourg, vol. 1 (1859), pp. 193-200. Oeuvres.
Vol. 1, pp. 499-508.
4. Sur l'interpolat1'on. Zapiski Akadcmii Kauk, vol. 4, Supplement no. 5, 186~.
Oeuvres. Vol. 1, pp. 539-560.
5. Surles fonctions analogues d celles de Legendre. Zapiski Akadcmii :\auk, vol. 16
(1870), pp. 131-140. Oeuvres. Vol. 2, pp. 59-68.
6. Sur les valeurs z.im.1'tes des 1'ntegrales. .Journal de Mathcmatiqucs, (2), vol. 19
(1874), pp. 157-lHO. Oeuvres. Vol. 2, pp. 181-185.
7. Sur le rapport de rleux httegrales etenrlues aux memes valeurs de la van·able. Za-
piski Akn.clemii :\auk, vol. 44, Supplement no. 2, 1883. Oeuvrefl. Vol. 2,
pp. 375~402.
8. Sur la rcpresr:ntation des valeurs linrites des 1'ntegrales par des res1'dus 1'ntegraux.
Ibid., vol. 51, Supplement no. 4, 1885. Oeuvres. Vol. 2, pp. 419-440. Acta
:\Iatlwmatica, vol. 9 (1886) pp. 35-56.

------·---------·--------------·----------·--·
 LIST OF REFERENCES 415
TITCHMARSH, E. c.
*1. The Theory of FuncHons. Oxford, 1932.
TRICOMI, F.
1. Trasformazione di Laplace e polinomi di Laguerre. I. lnversione della tras-
formazione. Rendiconti della R. Accademia dei Lincei, (6), vol. 21 (1935), pp.
232-239.
UsPENSKY, J. V.
1. On the development of arbitrary functions in series of Hermite's and Laguerre's
polynomials. Annals of Mathematics, (2), vol. 28 (1927), pp. 593-619.
vAN VEEN, s. c.
1. Asymptotische Entw?:cklung und Nullstellenabschdtzung der Hermiteschen Funk-
tionen. Proceedings, Koninklijke Akademie van Wetenschappen te Amsterdam,
vol. 34 (1931), pp. 257-267.
2. Asymptotis~he Entwicklung und Nullstellenabschdtzung der Hermiteschen Funk-
tionen. Mathemat.ische Annalen, vol. 105 (1931), pp. 40R-436.
3. Zusatz zum vorangehenden Berichte (of W. Hahn). Jahresbericht der Deutschen
Mathematiker-Vereinigung, vol. 44 (1934), pp. 236-238.
VITALI, G., and SANSONE, G.
1. Moderna Teoria delle Funzioni di Variabile Reale. II. Sviluppi in Serie di Fun-
zioni Ortogonali. Bologna, 1935.
WALSH, J. L.
1. Interpolation and A11proximation by Rational Functions in the Complex Domain.
American Mathematical Society Colloquium Publications, vol. 20, 1935.
WANGERIN, A.
1. Theorie der Kugelfunktionen und der verwandten Funktionen, insbesondere der
Lameschen und Besselschen (Theorie spezieller, durch lineare Differentialglei-
chungen definierter Funktionen). Encyklopadie der Mathematischen Wissen-
schaften, vol. II.1.2, pp. 695-759.
WATSON, G. N.
1. The harmonic functions associated with the parabolic cylinder. Proceedings of the
London Mathematical Society, (2), vol. 8 (1910), pp. 393-421; (2), vol. 17 (1918),
pp. 116-148.
2. Approximate formulae for Legendre functions. Messenger of Mathematics, vol.
47 (1918), pp. 151-160.
3. A Treatise on the Theory of Bessel Functions. Cambridge, 1922.
4. Noles on generating functions of polynomials: (1) Laguerre polynomials. Journal
of the London Mathematical Society, vol. 8 (1933), pp. 189-192.
5. Noles on generating functions of polynomials: (2) Hermite polynomials. Ibid.,
vol. 8 (1933), pp. 194-199.
6. Ober eine Reihe a us verallgemeinerlen Laguerre' schen Polynomen. Sitzungsberichte
der mathematisch--naturwisscnschaftlichen Klasse der Akademie Wien, Ila,
vol. 147 (1938), pp. 151-159.
WEYL, H.
1. Singulilre lnlegralgleichungen mil besonderer Beriicksichligung des Fourierschen
lnlegraltheorems. Inauguraldissertation, Gottingen, 1908. Mathematische An-
nalen, vol. 66 (190!)), pp. 273-324.
WHITTAKER, E. T., AND WATSON, G. N.
1. A Course of Modern Analysis. 4th edition. Cambridge, 1935.
WIGERT, S.
1. Contributions a la theorie des polynomes d' Abel-Laguerre. Arkiv for Matematik,
Astronomi och Fysik, vol. 15 (1921), no. 25, 22 pp.
2. Surles polynomes orlhogonaux et ['approximation des fonclions continues. Ibid.,
vol. 17 (1923), no. 18, 15 pp.
 416 FURTHER REFERENCES

WIMAN, A.
1. Ober eine asymptotische Eigenschaft der Ableitungen der ganzen Funktionen von den
Geschlechtern 1 und 2 mit einer endlichen Anzahl von Nullslellen. Mathematische
Annalen, vol. 104 (1931), pp. 169-181.
WINSTON, C.
1. On mechanical quadratures formulae involving the classical orthogonal polynomials.
Annals of Mathematics, (2), vol. 35 (1934), pp. 658-677.
WRIGHT, E. M.
1. The coefficients of <t certain power series. Journal of the London Mathematical
Society, vol. 7 (Hl32), pp. 256-262.
YouNG, W. H.
1. On the connexion bdween Legendre series and Fourier Series. Proceedings of the
London Mathematical Society, (2), vol. 18 (1919), pp. 141-162.
ZERNIKE, F.
1. Eine asymptotische Entwicklu.ng .fiir die grosste Nullstelle der Hermiteschen Polynome.
Koninklijke Akademie van Wetenschappen te Amsterdam, Proceedings, vol. 34
(1931), pp. 673-680.
ZYGMUND, A.
1. Sur la theorie riemannienne de ~ertains systemes orthogonaux, II. Prace Matema-
tyczno-Fizyczne, vol. 39 (1932), pp. 73-117.
2. Trigonometrical Sel"ies. Warszawa-Lw6w, 1935.

FURTHER REFERENCES
AcHIESER, N. I. (See List of Heferences above.)
3. Theory of Approximation. New York, 1956.
4. The Classical Moment Problem. Oliver and Boyd, Edinburgh and London, 1965.
ASKEY, R.
1. Orthogonal expansions with positive coefficients. Proceedings of the American Mathe-
matical Society, vol. 16 (1965), pp. 1191-1194.
2. Jacobi polynomial expansions with positive coefficients and imbeddings of projective
spaces. Bulletin of the American Mathematical Society, vol. 74 (1968), pp. 301-304.
3. A' transplantation theorem for Jacobi series. Illinois Journal of Mathematics, vol. 13
(1969)' pp. 583-590.
4. Linearization of the product of orthogonal polynomials, in Problems in analysis. Edited
by R. Gunning, Princeton University Press, Princeton, N. J., 1970, pp. 223-228.
5. Positivity of the Cotes numbers for some Jacobi abscissas. Numerische Mathematik,
vol. 19 (1972), pp. 46-48.
6. Mean convergence of orthogonal series and Lagrange interpolation. Acta Mathematica
Academiae Scientiarum Hungaricae, vol. 23 (1972), pp. 71-85.
7. Grllnbaum 's inequality for Bessel functions. Journal of Mathematical Analysis and
Applications, vol. 41 (1973), pp. 122-124.
8. Summability of Jacobi series. Transactions of the American Mathematical Society,
vol. 179 (1973), pp. 71-84.
9. Certain rational functions whose power series have positive coefficients. II. SIAM
Journal on Mathematical Analysis, vol. 5 (1974), pp. 53-57.
10. Jacobi polynomials. I. New proofs of Koornwinder's Laplace type integral representation
and Bateman's bilinear sum. SIAM Journal on Mathematical Analysis, vol. 5 (1974),
pp. 119-124.
ASKEY, R., AND FITCH, J.
1. Positivity of the Cotes numbers for some ultraspherical abscissas. SIAM Journal on
Numerical Analysis, vol. 5 (1968), pp. 199-201.
2. Integral representations for Jacobi polynomials and some applications. Journal of
Mathematical Analysis and Applications, vol. 26 (1969), pp. 411-437.

----·--------·----------------------·------------·-- ... -
 FURTHER REFERENCES 417
ASKEY, R., AND GASPER, G.
1. Linearization of the product of Jacobi polynomials. III. Canadian Journal of Mathe-
matics, vol. 23 (1971), pp. 332-338.
2. Jacobi polynomial expansions of ,Jacobi polynomials with non-negative coefficients.
Proceedings of the Cambridge Philosophical Society, vol. 70 (1971), pp. 243-255.
3. Certain rational functions whose power series have positive coefficients. American
Mathematical Monthly, vol. 79 (1972), pp. 327-341.
4. Positive Jacobi polynomial sums. II. American Journal of Mathematics.
AsKEY, R., AND PoLLARD, H.
1. Some absolutely monotonic and completely monotonic functions. SIAM Journal on
Mathematical Analysis, vol. 5 (1974), pp. 58-63.
ASKEY, R., AND STEINIG, J.
1. Some positive trigonometric sums. Transactions of the American Mathematical
Society, vol. 187 (1974), pp. 295-307.
2. A monotonic trigonometric sum. American Journal of Mathematics.
ASKEY, R., AND WAINGER, s.
1. Mean convergence of expansions in Laguerre and Hermite series. American Journal
of Mathematics, vol. 87 (1965), pp. 695-708.
2. A transplantation theorem between ultraspherical series. Illinois Journal of Mathematics,
vol. 10 (1966), pp. 322-344.
3. A transplantation theorem for ultraspherical coefficients. Pacific Journal of Mathe-
matics, vol. 16 (1966), pp. 393-405.
4. A convolution structure for Jacobi series. American Journal of Mathematics, vol. 91
(1969), pp. 463-485.
BAILEY, w. N.
1. The generating function of Jacobi polynomials. Journal of the London Mathematical
Society, vol. 13 (1938), pp. 8-12.
2. On the product of two Laguerre polynomials. Quarterly Journal of Mathematics,
vol. 10 (1939), pp. 60-66.
BALAZS, J., AND TURAN, P.
1. Notes on Interpolation. II. (Explicit formulae). Acta Mathematica Academiae
Scientiarum Hungaricae, vol. 8 (1957), pp. 201-215.
2. Notes on Interpolation. III. (Convergence). Ibid., vol. 9 (1958), pp. 195-214.
BATEMAN, H.
1. A generalization of the Legendre polynomial. Proceedings of the London Mathematical
Society, ser. 2, vol. 3 (1905), pp. 111-123.
2. The solution of linear differential equations by means of definite integrals. Transactions
of the Cambridge PhHosophical Society, vol. 21 (1909), pp. 171-196.
3. Partial Differential Equations. Cambridge University Press, Cambridge, 1932.
BATEMAN MANUSCRIPT PRoJECT (Director: A. ERDELYI).
1. HigherTronscendental Functions.Vols. 1, 2, 3. New York-Toronto-London, 1953, 1955.
BoCHNER, S. (See List of References above.)
2. Positive zonal functions on spheres. Proceedings of the National Academy of Sciences,
vol. 40 (1954), pp. 1141-1147.
BONAMI, A., AND CLERC, J.-L.
1. Sommes de Cesaro et multiplicateurs des developpements en harmoniques spheriques.
Transactions of the American Mathematical Society, vol. 183 (1973), pp. 223-263.
BuTLEWSKI, Z.
1. Sur les integrates d'une equation differentielle du second ordre. Mathematica (Cluj),
vol. 12 (1936), pp. 36-48.
CARLE SON' L.
1. On convergence and growth of partial sums of Fourier series. Acta Mathematica,
vol. 116 (1966), pp. 135-157.
COIFMAN, R., AND WEISS, G.
1. Analyse harmonique non-commutative sur certaines espaces homogenes. Lecture Notes
in Mathematics, vol. 242, Springer-Verlag, Berlin and New York, 1971.

·------·-------------------------------·--------------·-·-- .. ·-·· ...

 418 FURTHER REFERENCES

CSORDAS, G., AND WILLIAMSON, J.

1. On polynomials satisfying a Turan type inequality. Proceedings of the American
Mathematical Society, vol. 43 (1974), pp. 367-372.
DAVIS, J.,AND HIRSCHMAN, I. 1., JR.
1. Toeplitz forms and ultraspherical polynomials. Pacific Journal of Mathematics,
vol. 18 (1966), pp. 73-95.
DAVIS, J., AND RABINOWITZ, P.
1. (a) Some geometrical theorems for abscissas and weights of Gauss type. Journal of
Mathematical Analysis and Applications, vol. 2 (1961), pp. 428-437. (b) Erratum.
Ibid., vol. 3 (1961), p. 619.
DE BRUIJN, N. G.
1. Uncertainty principles in Fourier analysis, Inequalities, Edited by 0. Shisha, Academic
Press, New York, 1967, pp. 57-71.
DoETSCH, G. (See List of References above.)
3. Handbuch der Laplace- Transformationen. Vol. 2: Anwendungen der Laplace- Trans-
formationen, 1. Abteilung. Basel-Stuttgart, 1955.
EAGLESON, G.
1. A characterization theorem for positive definite sequences on the Krawtchouk polynomials.
Australian Journal of Statistics, vol. 11 (1969), pp. 29-38.
EGERVARY, E., AND TURAN, P.
1. Notes on interpolation. V. (On the stability of interpolation). Acta Mathematica
Academiae Scientiarum Hungaricae, vol. 9 (1958), pp. 259-267.
2. Notes on interpolation. VI. (On the stability of the interpolation on an infinite interval).
Acta Mathematica Academiae Scientiarum Hungaricae, vol. 10 (1959), pp. 55-62.
ERDELYI, A. (See List of References above.)
3. Asymptotic forms for Laguerre polynomials. Journal of the Indian Mathematical
Society, Golden Jubilee Volume, vol. 24 (1960), pp. 235-250.
ERDELYI, A., AND SWANSON, c. A.
1. Asymptotic forms of Whittaker's confluent hypergeometric functions. Memoirs of the
American Mathematical Society, No. 25, 1957.
ERDOS, P.
1. On the distribution ol normal point groups. Proceedings of the National Academy of
Sciences, vol. 26 (1940), pp. 294-297.
2. On divergence properties of the Lagrange interpolation parabolas. Annals of Mathe-
matics, vol. 42 (1941), pp. 309-315. . ·
3. Problems and results on the theory of interpolation. II. Acta Mathematica Academiae
Scientiarum Hung:aricae, vol. 12 (1961), pp. 235-244.
ERDOS, P., AND GRONWALD, G.
1. Ober einen Faber'schen Satz. Annals of Mathematics, vol. 39 (1938), pp. 257-261.
ERDOS P., AND LENGYEL, B. A.
1. On fundamental functions of La.grangean interpolation. Bulletin of the American
Mathematical Society, vol. 44 (1938), pp. 828-834.
ERDOS, P., AND TURAN, P. (See List of References above.)
3. On Interpolation. III. Interpolatory theory of polynomials. Annals of Mathematics,
vol. 41 (1940), pp. 510-553.
4. An extremal problem in the theory of interpolation. Acta Mathematica Academiae
Scientiarum Hung:aricae, vol. 12 (1961), pp. 221-234.
FELDHEIM E.(See List of References above.)
5. On the positivity of certain sums of ultraspherical polynomials. (Translated and edited
by G. Szego.) Journal d'Analyse Mathematique, voh 11 (1963), pp. 275-284.
FREUD, G.
1. Az Hermite-Fejer-{ele interpolacios eljaras konvergenciajar6l. Magyar Tudomimyos
Akademia IH. osztalyimak kozlemenyei, vol. 5 (1955), pp. 29-47.
2. Ortogonalis polin6mokr6l. Dissertation. Budapest, 1956.
3. Ober die Asymptotik orthogonaler Polynome. Academie Serbe des Sciences, vol. 11
(1957)' pp. 19-32.
4. Orthogonale Polynome. Basel, 1969. English translation, New York, 1971.

--------------·----------------·-----------·-··-··--····-----
 FURTHER REFERENCES 419

GASPER, G.
1. Linearization of the product of Jacobi polynomials. I. Canadian Journal of Mathe-
matics, vol. 22 (1970), pp. 171-175.
2. Linearization of the product of Jacobi polynomials. II. Canadian Journal of Mathe-
matics, vol. 22 (1970), pp. 582-593.
3. Positivity and the convolution structure for Jacobi series. Annals of Mathematics,
(2), vol. 93 (1971), pp. 112-118.
4. Banach algebras for Jacobi series and positivity of a kernel. Annals of Mathematics,
(2), vol. 95 (1972), pp. 261-280.
5. An inequality of Turan type for Jacobi polynomials. Proceedings of the American
Mathematical Society, vol. 32 (1972), pp. 435-439.
6. Nonnegativity of a discrete Poisson kernel for the Hahn polynomials. Journal of
Mathematical Analysis and Applications, vol. 42 (1973), pp. 438-451.
7. Projection formulas for orthogonal polynomials of a discrete variable. Journal of
Mathematical Analysis and Applications, vol. 45 (1974), pp. 176-198.
GATTESCHI, L.
1. Approssimazione asintot£ca degli zeri dei polinomi ultrosferici. Rendiconti di Matematica
e delle sue Applicazioni, (5), vol. 8 (1949), pp. 399-411.
2. Limitazione degli errori nelle formule asintotiche per le funzioni speciali. Rendiconti
del Seminario Matematico dell'Universita e del Politecnico di Torino, vol. 16
(1956-1957)' pp. 83-94.
GINIBRE, R.
1. General formulation of Griffiths' inequalities. Communications on Mathematical
Physics, vol. 16 (1970), pp. 310-328.
GRENANDER, u., AND SZEGO, G.
1. Toeplitz Forms and thelr Applications. Berkeley-Los Anageles, 1958.
GRONBAUM, F. A.
1. A property of Legendre polynomials. Proceedings of the National Academy of Sciences,
vol. 67 (1970), pp. 91)9-960.
2. A new kind of inequality for Bessel functions. Journal of Mathematical Analysis and
Applications, vol. 41 (1973), pp. 115-121.
GRONWALD, G. (See List of References above.)
2. On a convergence theorem for the Lagrange interpolation polynomials. Bulletin of the
American Mathemati,cal Society, vol. 47 (1941), pp. 271-275.
3. On the theory of Interpolation. Acta Mathematica, vol. 75 (1943), pp. 219-245.
HAHN, W. (See List of References above.)
5. Vber Orthogonalpolynome, die q-Differenzengleichungen genugen. Mathematische
N achrichten, vol. 2 (1949), pp. 4-34.
HARTMAN, P., AND WINTNER, A.
1. On non-conservative linear oscillators of low frequency. American Journal of Mathe-
matics, vol. 70 (1948), pp. 529-539.
HIRSCHMAN, I. 1., JR.
1. Variation diminishing transformations and orthogonal polynomials. Journal d'Analyse
Mathematique, vol. ~9 (1961), pp. 177-193.
2. The strong Szego limit theorem for Toeplitz determinants. American Journal of
Mathematics, vol. 88 (1966), pp. 73-95.
HORTON, R.
1. Expansions using ortlwgonal polynomials. Ph.D. Thesis, University of Wisconsin,
Madison, 1973, 60 pp.
Hso,H.
1. Certain integrals and infinite series involving ultrospherical polynomials and Bessel
functions. Duke Mathematical Journal, vol. 4 (1938), pp. 374-383.
HuA, L. K.
1. Harmonic Analysis of Functions of Several Complex Variables in the Classical Domains,
Translations of Mathematical Monographs, vol. 6 (1963), American Mathematical
Society, Providence, R. I.

---------------------------------------·-·--
 420 FURTHER REFERENCES

HUNT,R.
1. On the convergence ol Fourier series, Orthogonal Expansions and their Continuous
Analogues, Edited by D. Haimo, Southern Illinois University Press, Carbondale,
Illinois, 1968, pp. 235-255.
KARLIN, S., AND McGREGOR, J. L.
1. The differential equations of birth-and-death processes, and the Stieltjes moment problem.
Transactions of th<e American Mathematical Society, vol. 85 (1957), pp. 489-546.
2. The Hahn polynomials, formulas and an application. Scripta Mathematica, vol. 26
(1961)' pp. 33-46.
3. Classical diffusion processes and total positivity. Journal of Mathematical Analysis
and Applications, vol. 1 (1960), pp. 163-183.
KARLIN, s., AND STUDDEN, w. J.
1. Tchebycheff Systems w.ith Applications in Analysis and Statistics. Interscience Publishers,
New York, 1966.
KARLIN, s., AND SZEGO, G.
1. On certain determinants whose elements are orthogonal polynomials. Journal d'Analyse
Mathematique, vol. 8 (1960), pp. 1-157.
KooRNWINDER, T.
1. The addition formula for Jacobi polynomials. I. Summary of results. Indagationes
Mathematicae, vol. 34 (1972), pp. 188-191.
2. The addition formula for Jacobi polynomials and spherical harmonics. SIAM Journal
on Applied Mathematics, vol. 25 (1973), pp. 236-246.
3. Jacobi polynomials. H. An analytic proof of the product formula. SIAM Journal on
Mathematical Analysis, vol. 5 (1974), pp. 125-137.
4. Jacobi polynomials. Ill. An analytic proof of the addition formula. SIAM Journal
on Mathematical Analysis, vol. 6 -(1975), pp. 533-543.
LoCHER, F.
1. Norm bounds of quadrature processes. SIAM Journal on Numerical Analysis, vol.
10 (1973), pp. 553-558.
LoRCH, L.
1. The Lebesgue constants for Jacobi series. I. Proceedings of the American Mathematical
Society, vol. 10 (1959), pp. 756-761.
2. The Lebesgue constants for Jacobi series, II. American Journal of Mathematics, vol.
81 (1959), pp. 875-888.
· 3. Comparison of two formulations of Son in's theorem and of their respective applications
to Bessel functions. Studia Scientiarum Mathematicarum Hungarica, vol. 1
(1966), pp. 141-145'.
LoRCH, L., MULDOON, M. E., AND SZEGO, P.
1. Higher monotonicity properties of certain Sturm-Liouville functions. Ill. Canadian
Journal of Mathematics, vol. 22 (1970), pp. 1238-1265.
2. Higher monotonicity properties of certain Sturm-Liouville functions. IV. Canadian
Journal of Mathematics, vol. 24 (1972), pp. 349-368.
LoRCH, L., AND SZEGO, P.
1. Higher monotonicity pmperties of certain Sturm-Liouville functions. Acta Mathematica,
vol. 109 (1963), pp. 55-73.
2. Higher monotonicity properties of certain Sturm-Liouville functions. II. Bulletin de
I' Academie Polonaise des Sciences. Serie des Sciences Mathematiques, Astro-
nomiques et Physiques, vol. 11 (1963), pp. 455-457.
MAGNUS, W., AND 0BERHETTINGER, F.
1. Formeln und Lehrsiit:ze fur die speziellen Funktionen der mathematischen Physik.
Berlin, 1948.
MAKAI, E.
1. Ober die Nullstellen von Funktionen, die Losungen Sturm-Liouville'scher Differential-
gleichungen sind. Commentarii Mathematici Helvetici, vol. 16 (1944), pp. 153-199.
2. On a monotonic property of certain Sturm-Liouville functions. Acta Mathematica
Academiae Scientiarum Hungaricae, vol. 3 (1952), pp. 165-172.

__________________________________ .... _.
 FURTHER REFERENCES 421

3. On sy.~tems of polynomicils orthogonal in two intervals. Publicationes Mathematicae,

vol. 2 (1952), pp. 22~!-228.
MAKAI, E., ANDTURAN, P.
1. Hermite expansions and distribution of zeros of polynomial.~. Magyar Tud. Akademia
Mat. Kutat6 Int. Kozll. vol. 8 (1963), pp. 157-163.
MUCKENHOUPT, B.
1. Asymptotic forms for Laguerre polynomial.~. Proceedings of the American Mathematical
Society, vol. 24 (1970), pp. 288-292.
2. Mean convergence of Hermite and Laguerre series. I. Transactions of the American
Mathematical Society, vol. 147 (1970), pp. 419-432.
3. Mean convergence of Hermite and Laguerre series. II. Transactions of the American
Mathematical Society, vol. 147 (1970), pp. 433-460.
4. Equiconvergence and almost everywhere convergence of Hermite and Laguerre series.
SIAM .Journal on Mathematical Analysis, vol. 1 (1970), pp. 295-321.
5. Mean convergence of Jacobi series. Proceedings of the American Mathematical
Society, vol. 23 (1969), pp. 306-310.
MucKENHOUPT, B., AND STEIN, E.
1. Classical expansions and their relation to conjugate harmonic functions. Transactions
of the American Mathematical Society, vol. 118 (1965), pp. 17-92.
NEWMAN, J., AND RUDIN, w.
1. Mean convergence of orthogonal series. Proceedings of the American Mathematical
Society, vol. 3 (1952), pp. 219-222.
NoviKOFF, A.
1. On a special system of orthogonal polynomials. Dissertation, Stanford University, 1954.
OLVER, F. w. J.
1. A paradox in asymptotics. SIAM Journal on Mathematical Analysis, vol. 1 (1970),
pp. 533-534.
PEETRE, J.
1. The Weyl transform and Laguerre polynomials. Le Matematiche, vol. 27 (1972),
pp. 301-323.
POLLACZEK, F.
1. Sur une generalisation des polynomes de Legendre. Comptes' Rend us de I' Academie
des Sciences, Paris, vol. 228 (1949), pp. 1363-1365.
2. Systemes de polynomes biorthogonaux qui generalisent les polynomes ultraspheriques.
Ibid., vol. 228 (1949), pp. 1998-2000.
3. Sur une famille de polynomes orthogonaux qui contient les polynomes d 'Hermite et de
Laguerre comme cas limites. Ibid., vol. 230 (1950), pp. 1563-1565.
4. Sur une generalisation des polynomes de Jacobi. Memorial des Sciences Mathematiques,
vol. 131 (1956).
POLLARD, H.
1. The mean convergence of orthogonal series. I. Transactions of the American Mathematical
Society, vol. 62 (1947), pp. 387-403.
2. The mean convergence of orthogonal series. II. Transactions of the American Mathe-
matical Society, vol. fl3 (1948), pp. 355-367.
3. The mean convergence of orthogonal series. III. Duke Mathematical Journal, vol.
16 (1949), pp. 189-191.
RoosENRAAD, C. T.
1. Inequalities with orthogonal polynomials. Ph.D. Thesis, University of Wisconsin,
Madison, 1969, 52 pp.
SAPIRO, R. L.
1. The special functions connected with the representations of the group SU(n) of class
one with respect to SU(n - 1) (n ~ 3). Izvestija Vyssih Ucebnyh Zavedenii,
Matematika, vol. 71 1(1968), pp. 97-107, (Russian).
SARMANOV' I. 0.
1. A generalized symmetric gamma correlation. (Russian; translation in) Soviet Mathe-
matics, Doklady, vol. 9 (1968), pp. 547-550.

------·-------------------------------·-------------- . -·------
 422 FURTHER REFERENCES

SCHMEISSER, G.
1. Optimale Schranken zu einem Satz iiber Nullstellen Hermitescher Trinome. ,Journal
fUr die reine und angcwandte Mathematik, vol. 246 (1971), pp. 147-160.
ScHOENBERG, 1., AND SzEGO, G.
1. An extremum problem for pol:vnomials. Compositio Mathematica, vol. 14 (1960),
pp. 260-268.
SEIDEL, W., AND SzAsz, 0.
1. On positive harmonic functions and ultraspherical polynomials. ,Journal of the London
Mathematical Society, vol. 26 (1951), pp. 36-41.
SHOHAT, J. A., HILLE, E., AND WALSH, J. L.
1. A Bibliography on Orthogonal Polynomials. Washington, 1940.
SuRANYI, J., AND TuRAN, P.
1. Notes on interpolation. I. (On some interpolatorical properties of the ultraspherical
polynomials). Acta Mathematica Academiae Scientiarum Hungaricae, vol. 6
(1955), pp. 67-79 .
SzAsz, 0. (See List of References above.)
2. On the relative extrema of ultraspherical polynomials. Bollettino della U nione Matematica
Italiana, series 7, vol. 5 (1950), pp. 125-127.
3. On the relative extrema of the Hermite orthogonal functions . •Journal of the Indian
Mathematical Society, vol. 25 (1951), pp. 129-134.
SzEGO, G. (See List of References above.)
22. On an inequality of P. Turan concerning Legendre polynomial.~. Bulletin of the
American Mathematical Society, vol. 54 (1948), pp. 401-405.
23. On the relative extrema of Legendre polynomials. Bollettino della Unione Matematica
Italian a, (3), vol. 5 (1950), pp. 120-121.
24. On certain special sets of orthogonal polynomials. Proceedings of the American
Mathematical Society, vol. 1 (1950), pp. 731-737.
25. Ultrasphaerikus pol.inomok osszeg~rol. Matematikai es Fizikai Lapok, vol. 45
(1938)' pp. 36-38.
26. Uber gewisse Potenzreihen mit Lauter positiven Koeffizienten. Mathematische
Zeitschrift, vol. 37 (1933), pp. 67 4-688.
27. On some Hermitian forms associated with two given curves of the complex plane.
Transactions of the American Mathematical Society, vol. 40 (1936), pp. 450-461.
SzEGO, G., AND TuRAN, P.
1. On the monotone convergence of certain Riemann sums. Publicationes Mathematicae
(Debrecen), vol. 8 (1961), pp. 326-335.
THORNE, R. c.
1. The asymptotic expansion of Legendre functions of large degree and order. Technical
Report, Office of Naval Research, California Institute of Technology, 1956.
TRICOMI, F. G. (See List of References above.)
2. Sul comportamento asintotico dell 'n-esimo polinomio di Laguerre nell 'intorno dell 'ascissa
4n. Commentarii Mathematici Helvetici, vol. 22 (1949), pp. 150-167.
3. Sul comportamento asintotico dei polinomi di Laguerre. Annali di Matematica, ( 4),
vol. 28 (1949), pp. 263-289.
4. Sugli zeri dei polinomi sferici ed ultrasferici. Ibid., (4), vol. 31 (1950), pp. 93-97.
5. Vorlesungen tlber Orthogonalreihen. Berlin-Gottingen-Heidelberg, 1955.
TuRAN,P. •
1. On the zeros of the polynomials of Legendre. Casopis pro Pestovimi Matematiky a
Fysiky, vol. 75 (1950), pp. 113-122.
2. On Descartes-Herriot 's rule. Bulletin of the American Mathematical Society, vol.
55 (1949), pp. 79'7-800.
3. Remark on a theorem of Erhard Schmidt. Mathematica, vol. 2 (1960), pp. 373-378.
VIETORIS, L.
1. Ober das Vorzeichen gewisser trigonometrischer Summen. Sitzungsberichte der
mathematisch-naturwissenschaftlichen Klasse der Akademie der Wissenschaften
in Wien, vol. 167 (1958), pp. 125-135.

·-------------------------------·--·---·-·-·
 FURTHER REFERENCES 423
VITALI, G., AND SA:NSONE, G.
See List of References above. Third edition. Bologna, 1952.
WEBSTER, M. s.
1. A convergence theorem for certain Lagrange interpolation polynomials. Bulletin of the
American Mathematical Society, vol. 49 (1943), pp. 114-119.
WmoM, H., AND WILF, H.
1. Small eigenvalues of large Hankel matrices. Proceedings of the American Mathematical
Society, vol. 17 (1966), pp. 338-344.
ZYGMUND, A.
See List of References above. Second edition. New York, 1952.

_________ ______________________ _____________

, , , ,_,, ......... .
 INDEX
The numbers refer to pages

Abel, v, 100, 401. - - - - -of Mehler-Beine type, 193,194,

- 's continuity theorem, 92. 217.
- 's transformation, 2, 91, 135. - - - - -of Plancherel-Rotach type, 200,
- summability, 245, 273. 201, 227 ff.
See Inequality. See Fejer's formula, Perron's formula.
Achieser, 5, 37, 42, 401, 416. - - - Legendre functions of second kind,
Adamoff, 203, 248, 249, 250, 401. 198, 212, 222 ff.
Addition theorems, 58, 97 ff. - - - -polynomials, 191, 193, 194 ff., 243.
Additive operation, 11, 12. Asymptotic formula of Legendre polynomials
Airy's function, 18, 19, 131, 2:ol, 234, 239, ofHilb's type, 195, 202, 212 ff.
243, 377' 382. See Laplace formula, Laplace-Heine
See Zeros. formula.
Antipole condition, 246, 249, 26!5. - - -polynomials orthogonal on a curve,
Approximation, 5 ff. 371 ff.
Askey, xi, 98, 99, 110, 273, :3:48, 363, 391, - - - - - - the unit circle, 297 ff.
392,416. - - - ultraspherical polynomials, 196. 197,
Askey-Fitch,96,363,416. 206, 208, 209, 212.
Askey-Gasper, 97, 99, 110, 190,392,417.
Askey-Pollard, 273, 417.
Askey-Steinig, 97, 158, 363, 417. Bailey,273,392,417.
Askey-VVainger,99, 190,273,417. Balazs-Turim, 347,417.
Associated function, Legendre, 84. Banach, 13, 401.
Asymptotic formula of Bessel functions, Bateman, 96, 98, 99,273,417.
15, 16. Bateman Project, ix, 95, 110, 243, 393, 394,
- - -classical polynomials, HH ff. 417.
- - - general orthogonal polynomials, v, Bernstein, S., vi, 9, 31, 42, 159, 165, 168 ff.,
vi, 296 ff. 296,299,300,303,314,315,330,401.
- - -Hermite polynomials, ll32, 191, 194, - 's theorem on trigonometric polynomials,
198 ff., 218 ff., 235, 242. 5, 280,304.
- - -Jacobi functions of second kind, See Polynomials.
201, 225. Bessel functions, v, 14 ff., 95, 96, 102 ff.,
- - - -polynomials, 58, 167, 168, 191 ff., 126 ff., 140, 166, 167, 191, 193, 202, 203,
201,202,212,225. 212 ff., 225, 243, 272, 353, 362, 363, 380,
- - - - -of Hilb's type, 169, 197, 202, 384,387,392,400.
214. See Asymptotic formula, Zeros.
- - - - -of Mehler-Beine type, 167, - 's inequality, 25, 38, 289, 316, 322, 333, 367.
192 ff., 202. Blumenthal, 268, 310, 401.
See Darboux's formula. Bochner, 108,273,401,417.
- - -kernel polynomials, 369 ff. Bonami-Clerc, 273,417.
- - -Laguerre polynomials, ll32, 191, 193, Bottema, 132, 401.
194, 198 ff., 236, 237, 240 ff. Bounded variation, functions of, 12.
- - - - -of Hilb's type, 177, 199, 203, Brauer, A., 131, 401.
216 ff., 219 ff., 384. Bruns, 122, 125, 136, 138, 401.

425

------------------------------------------··--------------
 426 INDEX

Buell, 124 ff., 401. Davis-Hirschman, 99,418.

Butlewski, 166, 417. Davis-Rabinowitz, 158, 418.
De Bruijn, 110, 418.
Capacity, see Transfinite diameter. de la Vallee- Poisson summability, 273.
Caratheodory, 364, 402. Deviation, quadratic, 38, 41, 288.
See Osgood-Caratheodory. -, Tchebichef, 41, 368.
Carleman, 366,402. Differences,34,35, 136.
Carleson, 273, 417. Differential equation, v, 16 ff., 37, 152, 159,
Cauchy-Hadamard formula, 248, 253, 312. 164, 166, 210.
Cauchy's principle value, 2'78. - of Bessel functions, 15.
-theorem, 69, 106, 221, 275, 289, 324, 370, - - - Hermite polynomials, 106, 176, 380.
372. - - - Jacobi'(hypergeometric) polynomials,
Centroid, 189. iv, 60 ff., 117, 141.
Cesaro summability, vi, 244, 246 ff., 258. - - -Laguerre polynomials, 100, 117, 176.
- -(means) of Fourier series, 12, 14, 246, - - - ultraspherical polynomials, 80, 81.
250. Dirichlet, 85, 402.
- - -Hermite series, 250, 251. Dirichlet- Mehler, see Integral representation.
- - -Jacobi series, vi, 24{), 248, 249, 256 ff. Dirichlet's integral, 12, 14.
Cesaro summability of Laguerre series, 247, Discriminant of classical polynomials, 143 ff.
248, 250, 251, 271 ff. Distribution, 8, 9, 26, 29, 38, 40, 57, 181, 274,
-Legendre series, 14, 249. 330, 351, 378.
- - - power series, 325. - ofStieltjes type, 9, 33 ff., 38.
- - - ultraspherical series, 248. Doetsch, 34, 35, 380, 402, 418.
Characteristic values, 49, 187. Du Bois, Reymond, 14, 402.
Charlier, see Polynomials.
Christoffel, iii, 42, 43, 47,402.
- (-Darboux) formula, 42 fl'., 318,326. Eagleson, 37, 418.
-formula of, 29 ff., 397. Egerviuy-Turim, 348, 418.
-numbers, 46 ff., 114, 115, 187, 351 ff., 378. Ellipse, 8, 21, 251 ff.
Clerc, see Bonomi-Clerc. -of convergence, 245, 248, 252, 311, 312.
Closure, see Orthogonal polynomials. Elliptic functions, v, 60.
Coifman-G. Weiss, 273, 417 . Equiconvergence, 31, 244, 246, 247, 249,
Completely monotonic sequence, 136 ff., 155. 313 ff.
Confluent hypergeometric function, 104. Erdelyi, ix, 104, 243, 273, 380, 402, 418.
Conformal mapping, 21, HiO, 364, 365, 372. Erdelyi-Swanson, 243, 418.
Conjugate function, 279. Erdos, vii, 347, 348, 418.
-points (interpolation), 33!!, 347. Erdos-Feldheim, 334, 402.
Continued fractions, v, 54 ff. Erdos-Grunwald, 347, 418.
Continuous operation, 12. Erdos-Lengyel, 347, 418.
Convergence in mean, 330. Erdos-Turan, 113, 114, 115, 332, 333, 335,
- - - , generalized, 330. 347,348,402,418.
-of interpolation, 330. Euler, 72, 402.
- , quadrature, 330, 333, 350 ff. - 's constant, 15.
Convergent (continued fraction), 55 ff. - 's integral of the first kind, 14.
Cotes numbers, 12, 349 ff., 3~33. - - - -second kind, see Gamma function.
Courant-Hilbert, see Hilbert-Courant. Expansion, v, vi, 58.
Cramer, 251, 402. - , finite cosine, of Legendre polynomials,
Csordas-Williamson, 388, 418. 90 ff.
- , infinite sine, of Legendre polynomials,
Darboux, 42, 43, 195, 196, 211, 402. 90 ff.
Darboux's formula for Jacobi polynomials, -in power series, 313, 314.
168, 196, 225, 236, 237, 248, 253, 337, 353. - in series of classical polynomials, 244 ff.,
-method, 202, 206 ff., 265, !269, 395. 251 ff.

-----·-----------------------------·--·
 INDEX 427

- - - -general orthogonal polynomials, fundamental polynomials of Hermite inter-

313 ff. polation, 330 ff.
- - - -Hermite polynomials, 244, 245, - - -Lagrange interpolation, 12, 47, 329 ff.
247,251,253,269ff.
- - - -Jacobi polynomials, 244 ff. Galbrun, 251, 404.
- - - -Laguerre polynomials, 244 ff., Gamma-function, 14, 75.
249 ff., 253, 266 ff., 313. Gasper, 37, 98, 99, 190, 273, 419.
- - - -Legendre polynomials, 248, 249. See Askey-Gasper.
- - - - Tchebichef polynomials, 248, 314. Gatteschi, 242, 419.
- - - - ultraspherical polynomials, 248, Gauss, iii, 33, 47, 49, 62, 404.
249. See Mean-value theorem, Mechanical
See Fourier series, Orthogonal poly- quadrature.
nomials. Gauss-Weierstrass summability, 273.
Gegenbauer, 80, 94, 98, 99, 404.
Faber, 299,330,368, 369,372, 402. "Gegenbeispiel", 250, 272.
See Polynomials. Generating function, 35, 36.
F atou, 275, 403. - - of Hermite polynomials, 106, 380.
- 's theorem, 274. - -Jacobi polynomials, 69 ff., 82, 83, 95.
F avard, 43, 403. - - -Laguerre polynomials, 101, 202, 379.
Fejer, vii, 14, 45, 58, 89, 91, 96, 134, 136 ff., - - -Legendre polynomials, 90, 206, 207,
157,165, 172, 174, 175, 179, ll98, 202, 249, 383.
264, 330 ff., 336, 348, 350, 386, 403. - - - Pollaczek polynomials, 393.
See Stekloff-F ejer. - - - ultraspherical polynomials, 82, 83.
- 's asymptotic formula for Laguerre poly- Geometric mean, 275, 296, 385.
nomials, 198, 202, 203, 237, 240, 269. Geronimus, 42, 189, 404.
Fejer's generalization of Legendre poly- Gibbs' phenomenon, 251.
nomials, 135 ff., 17 4 ff., 206. Ginibre, 392, 419.
- 's second generalization of Legendre poly- Gottlieb, 37, 405.
nomials, 137, 138, 155. Grenander-Szego, 274, 287, 419.
- 's integral, 12. Gronwall, 165, 249, 405.
- 's representation of positive trigonometric Grunbaum, 392, 419.
polynomials, 3, 4, 274. Grunwald, vii, 330, 347, 405, 419.
- - - - - - , generalization of, 275 ff. See Erdos-Grunwald.
F ejer-Szego, 175, 404. Grunwald-Turim, 346,405.
Fekete, 369, 404.
Feldheim, 95, 96, 97, 334, 335, :386, 404, 418. Haar, 14, 248, 405.
See Erdos- F eldheim. Hahn, W., 33, 107, 111, 112, 129, 132, 151,
Fitch, see Askey- Fitch. 405,419.
Fourier, v. Hamburger, 57, 110, 405.
-coefficient (constant), 24, 287 . Hankel, see Quadratic form.
- 's inversion formula, 387. Hardy, 102,380,405.
-series, 12, 14, 24, 25, 28, 38, 39, 244, 246, Hardy-Littlewood-P6lya, 2, 405.
253,274,289,311,314,323,347,367. Hartman-Wintner, 20,419.
See Cesaro means. Hausdorff, 136,155,405.
Freud, 309, 418. Heine, iii, 27, 37, 47, 54, 80, 90 ff., 151 ff.,
Friedrichs, 404. 192, 194, 251, 405.
Fujiwara, 145, 404. Heine-Stieltjes theorem, iv, 151 ff.
Functions of second kind, Jacobi's, 73 ff., 251. Belly, 13, 405.
- - - -, Legendre's, 74, 78, 88, 89, 92, - 's theorem, 13, 14, 258, 272, 330, 339, 350.
379,383. Hermite, v, 156, 405, 406.
See Asymptotic formula, Integral -interpolation, 330 ff., 340 ff., 347,348.
representation, Zeros. Hermite-Stieltjes, 89, 156, 157, 172, 406.

---------------------------------------·-----··--·-··
 428 INDEX

Hermitian form, 287. -on Hermite abscissas, 340.

Hilb, 195, 249, 406. - -Jacobi abscissas, 335 ff.
Hilbert, 142, 145, 406. - -Laguerre abscissas, 340, 344 ff.
Hilbert-Courant, 58, 59, 100, 106, 109, 406. - - Legendre abscissas, 386.
Hildebrandt, 8, 10, 406. - - Tchebichef abscissas, 330, 334, 335, 386.
Hille, 102, 105, 126, 131, 242, 251, 406. - ultraspherical abscissas, 386.
See Shohat-Hille-Walsh. See Conjugate points, Fundamental
Hirschman, 99, 273, 419. polynomials, Hermite interpolation,
$ee Davis-Hirschman. Lagrange polynomials.
Hobson,iv,58,84,92,382,406.
Hollo, 335, 406. Jackson, vii, 6, 7, 42, 331, 406.
Horton, 99, 273, 419. Jacob,251,406.
!'Hospital, rule of, 307. Jacobi, iii, 49, 58, 69, 86, 406, 407.
Howell, 392. See Functions of second kind, Me-
Hsi.i, 390, 419. chanical quadrature, Polynomials,
Hua, 99,419. Series.
Hunt, 273,420. Jensen's theorem, 301.
Hurwitz, theorem of, 21, 150, 193, 239, 371. Joo, 391.
Hypergeometric function, iii, 62 ff., 83, 84, Jordan, C., 58, 407.
96. - arc, 8, 364.
See Differential equation. -curve, 8, 21, 364, 365.
- , Ch., 34, 407.
Inequalities, 159 ff. Jordan-Pochhammer integral, 75.
Inequality, Abel's, 2, 174, 205. Julia, 369, 407.
-, Cauchy's, 2, 39, 120, 1160, 183, 291, 302,
306,321,370. Kaczmarz-Steinhaus, 1, 10, 407.
- for the arithmetic and geometric mean, 2, Kanter, vii.
300. Karlin- McGregor, 37, 273, 389, 420.
-, Schwarz's, 2, 9, 110, 1.36, 162, 268, 276, Karlin-Studden, 5, 420.
317,376. Karlin-Szego, 190, 388, 420.
-, Tur{m's, 190,388. Keldysch-Lawrentieff, 368, 407.
See Bessel's inequality. Kernel polynomials, 39, 40, 44, 71, 101, 180,
Integral, Lebesgue, vii, 9, 26, 159, 246 ff., 183, 249, 290, 322, 323, 333, 351, 368, 377.
274, 275, 287, 291, 364, 384. See Asymptotic formula, Zeros.
- , Riemann, 9, 10, 280, 281, 282, 298, 310, Klein, 145,407.
333, 335, 351, 361. Kogbetliantz, 102, 168, 172, 203, 240, 248,
-, Riemann-Stieltjes, 8, 9, 11, 50, 333, 350, 249,251,256,273,380,407.
351. Koornwinder, 98, 99, 420.
-, Stieltjes-Lebesgue, v, 1, 8, 10, 38, 187. Korous, 131, 13~, 162, 167, 212, 251, 303, 408.
-equations, v, 218, 251. Koschmieder, 60, 408.
- representation of Legendre functions of Kowalewski, 24, 408.
second kind, 88 ff. Kowallik, 251, 408.
-polynomials, 85 ~r. Krall, 107, 408.
- -, Dirichlet-Mehler, 85 ff., 96. Krawtchouk, 113, 408.
- - , Laplace (first), 86, 176. See Polynomials.
- - , Laplace (second), 86 ff. Kronecker, 408.
- - , Stieltjes, 87 fl'.
- - - ultraspherical polynomials, Dirichlet- Lagrange, 100, 408.
Mehler, 89. -polynomials, 329 ff., 347.
- - - - - , Stieltjes, 89 fl'. -series, 70.
Interpolation, v, ix, 12, 14, 58, 329 ff., 347. See Interpolation.
-, Lagrange, 12, 47, 180, 329 ff., 348, 349, Laguerre, v. 100, 117, 131, 408, 409.
389. See Polynomials.

------------·-----------·-----------·-··
 INDEX 429

Lame function, 151. - -for classical abscissas, 352 ff.

Langer, 204, 210, 409. - - -Jacobi abscissas, 355 ff., 378, 379.
Laplace, iv, 86. - -,Gauss-Jacobi, 47 ff., 111, 348 ff.
Laplace's formula for Legendre polynomials, Mehler, iii, 47, 49, 85, 192, 380, 394, 409, 410.
194, 198, 201, 202, 204 ff., 211 fi., 224. Mehler-Beine, see Asymptotic formula.
- - - - -, Darboux's generalization, 195, Meixner, 34, 35, 410.
201, 206 ff., 211. Method of Liouville-Stekloff, see Liouville-
- - - - -, Stieltjes' generalization, 195, Stekloff.
202, 209 ff. - - steepest descent, 202, 203, 221 ff.
Laplace-Heine formula, 194, 204 ff., 208, 243. Modulus of continuity, 6, 7, 335, 336, 340,
- - , generalization, 194. 346.
Laplace series, 96, 249. Moecklin, 194, 203, 410.
-transform, 379. Moment problem ofStieltjes, iii, iv, v, 40.
See Integral representation. Muckenhoupt, 190,243,273,421.
Laurent expansion, 252. Muckenhoupt-Stein, 273, 421.
Lawton, 151, 409. Muldoon, see Lorch-Muldoon-P. Szego.
Lebesgue, 14, 15,409. MUntz, 251, 410.
See Integral. Myller-Lebedeff, 251;410.
-constant, 13, 258, 330, 336, 388, 339, 350.
Legendre,v, 70,409. N eumarin, E. R., 130, 251, 380, 410.
See Associated function, J?olynomials. Neumann, F., 248.
Lengyel, 113. Neumann, J. von, 108.
See Erdos-Lengyel. Newman-Rudin, 273, 421.
LeRoy, 103, 409. Newton's formula, 259.
Level curve, 8. Normalization, 28, 58, 160.
Leibniz, rule of, 67, 68, 101. Norm function, see Weight function.
Limited operation, 12. Norm of operation, 12.
Linear operation, see Operation. N ovikoff, 394, 396, 421.
Liouville-Stekloff, method of, 202 ff., 210 ff.,
299. Obrechkoff, 198, 248, 249, 410.
Lipschitz, 383. Olver, 243, 421.
- condition, 6, 162, 163, 186. Operation, linear functional, 11 ff.
Lipschitz-Dini condition, 279, 29'7, 324. Orthogonal polynomials and continued
Littlewood, see Hardy-Littlewood-P6lya. fractions, 54 ff.
Locher, 363, 420. - -,asymptotic formula of, v, ix, 4, 296 ff.
Lorch, xi, 190, 249, 420. - -, Christoffel-Darboux formula for, 42 ff.
Lorch-Muldoon-P. Szego, 158,420. - -,classical, 29, 49.
Lorch-P.Szego,158,420. - - , closure, 40, 108 ff.
Lukacs, 4, 178 ff., 249, 409. - - , definition, ix, 23 ff.
- -,expansion in series of, v, ix, 28, 38, 39,
Magnus-Oberhettinger, 420. 41, 311, 312, 313 ff.
Makai, 20, 148, 157, 190, 420, 421. - , extremum properties of, 28, 38 ff.
Makai-Turan, 158,421. - , general properties of, 38 ff.
Marcinkiewicz, 273, 330, 347, 409. - , highest coefficient of, 28.
Markoff, A., v, 33, 37, 50, 57, 115, 116, 121, -, recurrence formula for, 42 ff., 55, 391.
122,139,259,378,409. - , representation of, 27.
McGregor, see Karlin-McGregor. -, zeros of, v, ix, 44 ff., 121 ff., 188, 189.
Mean approximation, 10, 11. Orthogonality, Orthogonalization, v, vi, 8,
Mean-Value theorem, 383. 23 ff., 44.
- - - ofdauss, 275,310,311. Orthonormal set, 23, 25, 68.
- - -,second, 2, 202,357,363. Osgood-Caratheodory's theorem, 364.
Mechanical quadrature, v, ix, 12, 14, 58, 187,
329,348 ff. Parabola of convergence, 253.
430 INDEX

Parseval's formula, 40, 289, 368. -,Legendre, vi, 29, 30, 33, 34, 48, 58, 63, 70,
Peano, 1. 85 ff., 95, 96, 136 ff., 162, 164 ff., 167' 172,
Peetre, 110, 421. 189, 346, 348, 379, 382 ff., 392, 393.
Pencil, 49. See Asymptotic formula, Expansion,
Perron,45,54, 144,202,410. Generating function, Integral repre-
- 's formula for Laguerre polynomials, 198, sentation, Interpolation, Zeros.
199, 202, 203, 220, 221, :~25 ff. - , Krawtchouk, 35 ff.
P-function, contiguous Riemann, 71. -, ofS. Bernstein and Szego, 31 ff.
Plancherel, 250. -, Poisson-Charlier, 34, 35, 37, 377, 389, 390.
Plancherel-Rotach, 132, 201, 203, 410. -, Pollaczek, 37, 393 ff.
Pochhammer-Bames, notation of, 103. - orthogonal on a curve, vii, 364 ff.
Poincare, 310. See Asymptotic formula, Zeros.
Poisson's integral, 276, 29~!. - - - -segment, see Orthogonal poly-
See Polynomials. nomials.
Pollaczek, ix, 37, 393, 421. Polynomials orthogonal on the unit circle, ix,
Pollard, 273, 421. 287 ff., 384.
. See Askey-Pollard. -, Stieltjes-Wigert, 33.
Polya, vii, 42, 53, 117, 134, 154, 166, 350, -, Tchebichef (of the first and second kind),
388, 410,411. 3, 26, 29, 30, 60, 63, 112, 136 ff., 162, 347,
See Hardy-Littlewood-Polya. 348, 387, 391.
Polya-Szego, vii, 5, 11, 21, 24, 34, 40, 70, 87, See Expansion, Interpolation, Zeros.
105, 117, 134, 135, 175, 178, 179, 184, 212, - , ultraspherical, vi, 29, 58 ff., 80 ff., 93 ff.,
310,311,411. 107,135 ff., 167, 170 ff., 379,387,390,394.
Polynomials associated with a curve See Asymptotic formula, Differential
(Tchebichef polynomials), 368, 385. equation, Expansion, Generating
-,classical, v, 29, 159. function, Integral representation,
See Asymptotic formula, Expansion, Interpolation, Recurrence formula,
Mechanical quadrature. Rodrigues' formula, Zeros.
-,Faber, 372,374. Popoviciu, 46, 140, 142, 411.
-, Fejer's generalization of Legendre, see Principle of argument, 21, 157.
Fejer. Probability, calculus of, 35.
-, Gegenbauer, see Polynomials, ultra-
spherical.
-, Hermite, vi. ix, 29, 3{i, 37, 105 ff., 110, Quadratic form, 24, 123, 187, 366.
111, 176 ff., 190, 331, 388 ff., 392. - -of Hankel (recurrent) type, 27, 309.
See Asymptotic formula, Differential Quantum mechanics, iii.
equation, Expansion, Generating
function, Interpolation, Recurrence Rabinowitz, see Davis-Rabinowitz.
formula, Rodriqwes' formula, Zeros. Rau, 197, 214, 249, 411.
-,Jacobi, vi, ix, 3, 29, 58 III., 94 ff., 103, 105, Recurrence formula, general, 42 ff.
107, 161, 167 ff., 172 ff., 179 ff., 243, 249, - -of Hermite polynomials, 106.
295,348,383,385,399. - - -Jacobi functions of second kind, 78 ff.,
See Asymptotic formula, Differential 379.
equation, Expansion, Generating - - polynomials, 71 ff.
function, Interpollation, Mechanical - - -Laguerre polynomials, 101.
quadrature, Recurrence formula, - - -polynomials orthogonal on the unit
Rodrigues' formula, Zeros. circle, 293.
-, Laguerre, vi, ix, 29, 35, 100 ff., 110, 111, - - - ultraspherical polynomials, 81, 82.
164, 176 ff., 184, 185, 190, 243, 379, 380, Riemann, see Integral, P-function.
382,387 ff., 391,393,395. - 's lemma, 254,267,319.
See Asymptotic formula, Differential - 's theory of trigonometric series, 248.
equation, Expansion, Generating Riesz, F., 12, 275, 411.
function, Interpollation, Recurrence Riesz, M., 5, 57,304, 411.
formula, Rodrigues' formula. Robin's constant. see Transfinite diameter.

_______________________________ ............. .
 ~--I

432 INDEX

Turan, vii, ix, 158, 190, 388, 389, 422. Williamson, see Csordas-Williamson.
See Balazs-Turan, Egervary-Turan, Wiman, 131, 416.
Erdos-Turan, Grunwald-Turan, Winston, 131, 351, 416.
Makai-Turan, Suranyi-Turan, Wintner, see Hartman-Wintner.
Szego-Turan. Wright, 203,384, 416.

Uniqueness theorems, 248 Young, W. H., 10,248,414.

Uspensky, 53, 107, 203, 219, 220, 251, 415.
Zernike, 132, 416.
VanVeen, 132, 204, 415. Zero-function, 9, 40, 45.
Vector, 9. Zeros, distribution of, v, 310.
-space, 10. -, electrostatical interpretation of, 140, 382.
Vietoris, 363, 422. -of Airy's function, 18, 19,377, 382.
Vitali-Sansone, 415, 423. - - analytic functions, 21.
Vitali's theorem, 57. - -Bessel functions, 126 ff., 140, 192, 193,
Volterra equation, 211. 381.
- -Hermite polynomials, 117 ff., 123,
Wainger, see Askey-Wainger. 127 ff., 141 ff., 240, 353.
Walsh, 7, 366, 415. - -Jacobi polynomials, vi, 116 ff., 140 ff.,
See Sholat-Hille-Walsh. 144 ff., 192, 193, 237 ff., 250 ff., 379, 381.
Wangerin, 84, 415. - -kernel polynomials, 369, 377.
Watson, 18, 20, 96, 102, 104, 107, 159, 162, - -Laguerre polynomials, 116 ff., 122 ff.,
192, 193, 202, 203, 222, 251, 253, 363, 380, 127 ff., 141 ff., 150 ff., 237 ff., 353, 381, 382.
384,415. - - Legendre functions of second kind,
See Whittaker-Watson. 155 ff.
Webster, 347,423. - -polynomials, 111, 122, 125, 353.
Weierstrass, theorem of, fi ff., 10, 110. -numerators of continued fractions, 57.
Weight function, 9, 37, 68, 159, 160, 162 ff., -polynomials orthogonal on a curve, 369.
185, 188, 287, 294, 296, 297, 299, 309, 312, - - - - the unit circle, 292, 384.
314,316,347,380,385. -solutions of differential equations, see
Weiss, G., see C oifman- G. Weiss. Sturm's theorem.
W eyl, 110, 251, 310, 415. - - Tchebichef polynomials, 330, 351.
Whittaker-Watson, 14 ff., 58, 63, 65, 66, 71, - - ultraspherical polynomials, 119, 121 ff.,
75,84,86,88,93,103,2:48,415. 138, 352, 381.
Widom-Wilf, 243, 423. See Orthogonal polynomials.
Wigert, 33, 102, 251, 413. Zygmund, 248, 253, 254, 274, 276, 279, 281,
See Polynomials. 359,416,423.
Wilf, see Widom-Wilf.
 INDEX 431

Rodrigues' formula for Hermite polynomials, -polynomial, 329, 339 ff.

106. - - , generalized, 331.
- - -Jacobi polynomials, 67 ff., 73, 74, 94, Stieltjes, v, 33, 37, 46, 50, 51, 53, 54, 58, 87,
99,117. 89, 90, 92, 93, 121 ff., 136, 139, 140, 142,
- - -Laguerre polynomials, 101, 117, 388. 145, 151 ff., 156, 165, 172, 174, 175, 193,
- - - ultraspherical polynomials, 81. 195,211,382,412,413.
Roever, vii. See Heine-Stieltjes, Hermite-Stieltjes,
Rogosinski, 34 7. Integral, Laplace, Moments, Poly-
Rolle's theorem, 51, 53, 117, 378. nomials.
Roosenrad, 110, 421. Stirling's formula, 227.
Ross, vii. - series, 212.
Rotach, 203, 249, 250,411. Stone,10,23,251,413.
See Plancherel-Rotach. Strip of convergence, 253.
Rouche's theorem, 21, 149. Studden, see Karlin-Studden.
Rudin, see Newman-Rudin. Sturm-Liouville type, 210.
Runge, 7. Sturm's theorem on differential equations,
Runge-Walsh, theorem of, 7. 19 ff.
- -(method)- zeros, vi, 45, 111, 121,
Saddle-point, 219, 231. 124 ff., 139.
Sapiro, 98, 421. Summability, see Abel, Cesaro.
Sarmonov, 392,421. Suranyi-Turan, 347, 422.
Scalar product, 8, 9, 10, 25,365,3:67,378. Szasz,5,190,413,422.
Schmeiser, 158, 422. See Seidel-Szasz.
Schmidt, E., 34, 35,411. Szego, G., 19, 27, 31, 33, 37, 45, 59, 87, 93, 94,
Schoenberg, 273. 96, 110, 125, 127, 128, 135, 137, 140, 158,
Schoenberg-Szego, 190, 422. 163, 164, 167 ff., 189, 190, 197, 198, 203,
Schur, 1., vi, 140, 142, 411. 206, 214, 243, 248, 249, 251, 274 ff., 287,
Schwid, 204, 416. 289, 294 ff., 299, 300, 309, 310, 312, 313,
Seidel-Szasz, 94, 422. 340, 346, 355, 360, 361, 366, 369, 371, 377,
Sen-Rangachariar, 145,411. 384,388,392,393,394,413,414,422.
Series, Jacobi, vi, 273. See Fejer-Szego, Grenander-Szego,
- , Laguerre, 240. Karlin-Szego, Polya-Szego, Poly-
-,Legendre, vi, 42. nomials, Schoenberg-Szego.
See Expansion, Orthogonal poly- Szego, P., xi.
nomials. See Lorch-P. Szego, Lorch-Muldoon-
Sherman, 57, 411. P. Szego.
Shibata, 145, 412. Szego-Turan, 157, 158,422.
Shohat, v, vii, 42, 48, 159, 163, 189, 309, 333,
336,340,345,347,350,385,386,412.
Shohat-Hille-Walsh, 422. Tamarkin, vii, ix, 42, 414.
Singular integral, 13. Tchakaloff, 189, 383, 414.
Smirnoff, 275, 27:6, 289, 368, 369, 412. Tchebichef, v, 33, 47, 50, 54, 70, 100, 186,
Smith, 412. 188, 189, 414.
Sonin, 96, 100, 102, 104, 159, 16:6, 169, 176, See Deviation, Polynomials.
189,379, 412. Thorne, 243, 422.
Spencer, 132, 412. Titchmarsh, 37, 57,301,310,311,415.
Statistics, mathematical, v. Toeplitz matrix, 287, 399.
Stein, see Muckenhoupt-Stein. -,operator, 99.
Steinig, see Askey-Steinig. Total variation, 12, 34, 35.
Stekloff, 9, 210, 350, 412. Transfinite diameter, 364, 369.
See Liouville-Stekloff. Tricomi, ix, 242, 243, 415, 422.
Stekloff-Fejer, theorem of, 350 ff., 360. Trigonometric polynomials, 3, 5 ff., 11.
Step function, 34, 35, 37, 164. - representation, 90 ff.

·-----------------------------·-··--""""