M.A.

Naimark

LINEAR
DIFFERENTIAL
OPERATORS
Part 1
Elementary Theory
of
Linear Differential Operators
LINEAR DIFFERENTIAL OP ERATORS
 M. A. Naimark

LINEAR DIFFERENTIAL
OPERATORS

PART I

Elementary Theory o f
Linear Differential Operators

W ITH A D D IT IO N A L M A T E R IA L BY T H E A U TH O R

Translated, by
E. R. DAW SON
Queen’s College, Dundee

English translation edited by

W. N. E V E R IT T
Professor of Mathematics, Queen’s College, Dundee

(T>

F R E D E R IC K U N G A R PU BLISH IN G CO.
NEW YORK
 \

Copyright © 1967 by Frederick Ungar Publishing Co., Inc.

Published by arrangement with Mezhdunarodnaja Kniga, Moscow

Printed in the United States of America

Library of Congress Catalog Card Number: 66-19469

 TRANSLATION EDITOR’S PREFACE
The first, Russian, edition of Linear Differential Operators by N. A.
Naimark appeared in 1952. The English edition of this important book has had
the benefit of advice from, and additional material supplied by, Professor
Naimark, who has also written a preface for this new edition.
Some explanation may be called for in presenting at this time an English
edition of this book. Since the original Russian edition, several presentations
of the theory of linear ordinary differential operators have appeared, notably
those of E. C. Titchmarsh (second edition, 1962), Coddington and Levinson
(1955), K. S. Miller (1964), G. Hellwig (1964), and Dunford and Schwartz
(1964). Some of these authors, notably Dunford and Schwartz, use methods of
functional analysis; others, in particular, Titchmarsh, restrict themselves to
classical analytical methods; none combines these two approaches so effectively
as Naimark does in this book. There is no better introduction to the wide range
of ideas and techniques required for the study of linear differential operators
than that given here.
Part I presents a complete account of the so-called elementary theory of
linear differential operators, in which the methods of functional analysis are
kept in the background whilst use is made of the properties of functions of a
complex variable. There is a detailed investigation of the general, regular,
eigenvalue problem for ordinary differential equations with complex co­
efficients and regular boundary conditions. No comparable account exists else­
where in the literature.
Part II of the book, which is to be published separately, begins with a com­
pact account of the theory of linear operators in separable Hilbert spaces;
these results are then used to present the properties of symmetric operators
derived from ordinary differential expressions with real coefficients. This, in
turn leads to an account of the spectral theory of such operators and the corre­
sponding eigenfunction expansions in square-integrable function spaces. Next
follows a long chapter on the determination of the deficiency indices and the
spectrum of these differential operators. These results, whilst presented in the
terminology of Hilbert space, are obtained by the use of methods of classical
analysis, in particular, the development of certain asymptotic expansions for
systems of differential equations. There is no comparable account of these
expansions, which are largely due to Naimark himself, anywhere else in the
literature. This second part of the book ends with an account of the work of
Russian mathematicians on the inversion of the Sturm-Liouville problem.
VI TRANSLATION EDITOR’S PREFACE

The book has been expertly translated by Mr. E. R. Dawson, who has also
brought the Bibliography up to date.
Mr. Dawson and I wish to express our thanks to several of our colleagues in
Queen’s College, Dundee, for their help in reading the manuscript and in
checking the proofs.
W. N. E v e r i t t .
Queen’s College, Dundee.
July 1966.
 AUTHOR’S PREFACE TO THE ENGLISH
EDITION

This edition incorporates the editorial amendments and additions which were
made in the German edition of the book at the suggestion of the German
editors, Professors H. O. Cordes and F. Riihs; the author takes this occasion to
express his cordial acknowledgment to them.
In addition, a number of new changes and additions have been made in the
English edition; these take into account some of the important advances in the
theory of linear ordinary differential operators which have been made in the
past few years. In this work the author has been assisted by Professor Y. E.
Lyantse, who has also written a Supplement to Part II. This Supplement deals
with the author’s work in the theory of non-selfadjoint singular differential
operators, work which has later been extended by Professor Lyantse’s own
researches. It is the author’s pleasant duty to express his deep gratitude to
Professor Lyantse. The author also expresses his gratitude to Professor W. N.
Everitt for undertaking the task of editing the English translation of the book.

M. A. N aimark.
Moscow, May 1966.
 PREFACE TO THE RUSSIAN EDITION
Many topics in mathematical physics lead to the problem of determining the
eigenvalues and eigenfunctions of differential operators and of expanding an
arbitrary function as a series (or an integral) of eigenfunctions. A problem of
this sort is encountered, for example, whenever the Fourier method is used to
find a solution of a partial differential equation under prescribed initial and
boundary conditions. Consequently a lot of attention has been paid to differ­
ential operators and they are the subject of much current research.
Most of the publishedwork deals with the main problem just mentioned—the
spectral theory of differential operators, i.e., the investigation of the spectra of
differential operators and the expansion of given functions in terms of the
eigenfunctions of these operators. Interest in such problems has been particu­
larly stimulated by the development of quantum mechanics. The spectral
theory of differential operators appears as the basic mathematical method for
investigating many topics in quantum mechanics. In particular, quantum
mechanics demands a detailed study of “singular” differential operators, of
operators, for example, which are defined on some unbounded interval. Such
operators may, in general, have a continuous spectrum as well as a discrete
spectrum; and then an expansion in terms of eigenfunctions takes the form of a
Stieltjes integral.
It should be mentioned that, in recent years, highly important results in this
field have been obtained by Soviet mathematicians. Results such as the proof
of the existence of an expansion in eigenfunctions of a singular, self-adjoint
operator of arbitrary even order, the solution for the inversion of the Sturm-
Liouville problem (see Chapter VIII), the completeness of the system of eigen­
functions and associated functions for a wide class of differential operators
which are not self-adjoint, and many other results, are due, either entirely, or
almost entirely, to the efforts of Soviet mathematicians.
Notwithstanding these fundamental achievements, the problems of spectral
representation of differential operators cannot yet be regarded as exhausted.
Particular attention should be drawn to the problem of determining the multi­
plicity of the spectrum and the deficiency indices of a differential operator in
relation to the behaviour of its coefficients, and to the problem of selecting and
normalizing a minimal system of eigenfunctions when the spectrum is con­
tinuous. In particular, very little work has been done for the case of an operator
generated by a system of differential expressions, that is, for a differential
operator in the space of vector-functions (cf. Chapters V and VII).
IX
X PREFACE TO THE RUSSIAN EDITION

An equally important and interesting problem is that of expanding a given

function in terms of eigenfunctions of a differential operator which is not self-
adjoint. A noteworthy advance with this prbblem was made in a paper by
M. V. Keldysh [47] for the case of a regular differential expression with, in
general, non-regular boundary-conditions, but so far no one has investigated
the case of a singular differential operator which is not self-adjoint.
Among other outstanding problems in the theory of differential operators
we may also mention that of putting on a proper mathematical basis the so-
called “perturbation” theory, which has many applications in quantum
mechanics (see, in this connection, [39e], [56], [61], [96a], [112b], and [27b]).
Despite the importance of the theory of linear differential operators, due
to its applications, and despite the abundance of fundamental results already
achieved, the literature, both Soviet and foreign, is extremely poor in books
devoted to this theory. Thus, in the foreign literature there is just one book in
which a complete and exact account of the fundamental topics in the theory of
second-order differential operators is given; this is [112a] by Titchmarsh,
published in 1946. In the Soviet literature there is one monograph by B. M.
Levitan [63a], published in 1950, in which another, and in fact a very ele­
mentary, presentation of the theory of differential operators is given. No book
is yet available dealing with differential operators of higher order. Moreover,
a whole range of questions arising in the theory of second-order differential
operators are not discussed at all in the books by Titchmarsh and Levitan.
In the present book, the theory of linear, ordinary, differential operators of
arbitrary order is developed. Extensive use is made of the ideas and theorems
of functional analysis, especially those of the theory of linear operators in
Hilbert space. For the reader’s convenience all the necessary theorems in
functional analysis are developed in the text itself, to make the book, as the
author hopes, self-contained and accessible to a wide range of readers-. Many
topics in the theory of linear differential operators can also be discussed without
using functional analysis, as Titchmarsh’s treatment [112a] shows. However,
the author does not consider such a treatment to be appropriate, since it
is only by using the ideas and methods of functional analysis that a deeper
understanding of the theory, and its most general results, can be achieved.
The book is in two parts. In the first part, which may be described as an
elementary theory of differential operators, the use of the methods of functional
analysis is kept to a minimum. This part presents the theory of differential
operators defined on a finite interval, including the case of differential opera­
tors which are not self-adjoint, the theory being developed on the assumption
that the coefficients of the operators are sufficiently smooth, i.e., sufficiently
differentiable. To understand Part I the reader needs only an elementary know­
ledge of the theory of ordinary differential and integral equations and of the
theory of functions.
Part II develops the theory of differential operators, using Hilbert space
 PREFACE TO THE RUSSIAN EDITION XI

methods. Here, in addition to the subjects already mentioned, the reader will
be expected to have some knowledge of the basic results in the theory of
the Lebesgue integral.
Many new results are given in this book, most of them due to the work of
Soviet mathematicians; in particular, the results relating to the deficiency
indices and the spectrum of a differential operator.
The author wishes to express his thanks to M. I. Vishik for reading the
book in manuscript and making a number of valuable suggestions for improv­
ing the text.
M. A. N a i m a r k .
July, 1952.
 CONTENTS

CHAPTER I: FUNDAMENTAL CONCEPTS AND THEOREMS

§1. Definition and Principal Properties of the Linear
Differential Operator 1
§2. Eigenvalues and Eigenfunctions of a Differential Operator 13
§3. Green’s Function for a Linear Differential Operator 27

CHAPTER 11: ASYMPTOTIC BEHAVIOUR OF THE EIGENVAL­

UES AND EIGENFUNCTIONS. EXPANSION OF A PRESCRIBED
FUNCTION IN TERMS OF EIGENFUNCTIONS OF A DIFFEREN­
TIAL OPERATOR
§4. The Asymptotic Behaviour of the Eigenvalues and
Eigenfunctions for Large Values of |A| 42
§5. Expansion in Terms of Eigenfunctions 79

CHAPTER 111: DIFFERENTIAL OPERATORS IN A SPACE OF

VECTOR-FUNCTIONS
§6. Basic Concepts 105
§7.Green’s Function for a Differential Operator 115
§8. Asymptotic Behaviour of the Eigenvalues of a
Differential Operator 118
§9. Expansion in Terms of Eigenfunctions of a
Differential Operator 124

BIBLIOGRAPHY 131

INDEX 143

x iii
 CHAPTER I

FUNDAMENTAL CONCEPTS AND

THEOREMS

§1. Definition and Principal Properties of the Linear Differential Operator

1. General Definition of a Linear Vector-space and of a Linear Operator
A set R of elements x , i s called a linear vector-space if it is constituted as
follows:
1 °. For any two elements x,y £ R there is a sum x + y having the following
properties:
(aj) if x,y e R, then * + y e R ;
(b jx + y = y + x;
(ci) (x + y) + z = x + (y + z );
(dx) there is a “null element” 0 in R such that * + 0 = x for all * e R.
2°. For each element x £ R and for each real or complex number X there is a
product Xx having the following properties:
(a2) if x £ R, then Ax s R;
(b2) X(px) = (Ap)x\
(c2) \.x = x;
(d2)0.x = 0 (where, on the L.H.S., 0 means the number 0, and on the
R.H.S., 0 means the null-element);
(e2) A(x + y) — Xx + Aj';
(ft) (A + P-)* = Ax + fix.
The element ( —l)x is denoted by —x, and by properties (c2), (f2) and (d2)
a: + (—x) = (1 + (—1))* = 0 .x = 0.
The elements x,y,... of the space R are called vectors in R.
In this definition the nature of the elements x,y,... is entirely arbitrary, and
it is equally a matter of indifference how the concepts of the sum x + y and the
product of an element x with a number Aare interpreted in particular cases. It
is required only that these concepts shall satisfy the conditions enumerated
above. Any two operations whatever which satisfy these conditions may be
regarded respectively as addition of vectors, and as multiplication of a vector
by a number; and the set of elements for which these operations are defined
may then be called a linear vector space.
If in the space R multiplication only by real numbers is allowed, then R is
1
2 §1 LINEAR DIFFERENTIAL OPERATOR

called a real space; but if multiplication by any complex number is allowed,

then R is called a complex space. We shall take R to be a complex space unless
the contrary is specifically stated. An expression of the form A ^ -f- A2* 2 + ...
+ Xnxn is called a linear combination of the vectors x1,xr2,...,xrn; a linear com­
bination is said to be trivial if all the numbers A1,A2,...,A„are zero, otherwise it
is non-trivial. The vectors x ltx 2,...,xn are said to be linearly dependent if at least
one non-trivial combination of them vanishes, and to be linearly independent
if no non-trivial combination of them vanishes.
A space R is finite-dimensional, or more precisely, n-dimensional if in it there
are n but not more than n linearly independent vectors. Such a system of n
vectors is called a basis in R. If, on the other hand, there are arbitrarily many
linearly independent vectors in R, then R is infinite-dimensional.
A subset R ' of the finite-dimensional or infinite-dimensional space R is
called a subspace1 of R if every linear combination of elements of R' is itself an
element of R '.
Let 3 be a subset in the linear space R. A function A which assigns to each
element x of 3 some element, x ' — A(x), of R is called an operator in the space
R and 3 is the domain of definition of A. Often we write A x instead of A(x) if
this is not likely to be misunderstood; and we write 3 A instead of 3 when it
is necessary to emphasize that 3 is precisely the domain of definition of the
operator A.
The set of all vectors Ax,x e 3 A, is called the range of values of the operator
A and is denoted by 0tA or A2AA.
If, generally, £ is an arbitrary set, then A £ denotes the set of all vectors
A x,x e £, i.e. the set of all vectors which are obtained by applying the operator
A to all the vectors of the set S’.
An operator A is said to be linear if 3)A is a subspace and if, for arbitrary
vectors x,y e 3 A and for any number A, the relations
/I (A*) = XA(x),

A x + y) = A(x) + A(y)
hold.
Two operators A and B in the space R are regarded as equal if and only if
they are defined on the same domain3 and if A x — B xfor all x e 3 .
An operator A is called an extension of an operator B and we write A => B
or B cc A i£ 3 A => 3 B and if the operators are equal i n 3 B, i.e. if A x = Bx
for all x e 3 B. In such a case we also call the operator B a restriction of the
operator A t o 3 B.
In this book we shall consider only operators which are linear, and therefore,
to avoid constant repetition of the word “linear”, we shall always use the one
word “operator”, but mean by it “ linear operator” .1

1 In the English literature this is usually called a linear manifold.

 § 1 .3 BOUNDARY CONDITIONS 3

2. Linear Differential Expressions

A linear differential expression means an expression of the form
Ky) = p0( x ) / n) + Pi(x)yin~1] + ••• + pn(x)y. (l)
The functions Po(x),Pi(x),...,pn(x) are called the coefficients and the number n
is called the order of the differential expression.
In this first chapter we shall assume that the functions —!~r,/>i(x),Jp2(x),...
Po\x)
p„{x) are continuous on a fixed, closed, finite interval [a,b] \ in some cases we
shall impose further conditions.
We denote by C(n) the set of all functions which have continuous
derivatives up to the nth order inclusive on the interval [a,b].
For every function y e C(n) the differential expression l(y) is well defined,
and represents a function continuous on the interval [a,b\.

3. Boundary Conditions
We denote the values of the function y and its first (n — 1) successive
derivatives at the boundary points a and b of the interval [a,b] by
yay a,---,y(a ~ l)\ y byb’---A n~1)- (2)
Let U{y) be a linear form in the variables (2):
U(y) = a0ya + a ^ ' + •• • + + & y b + Piy'b + ••• + Pn- A n~l)- (3)
If several such forms Uy{y) have been specified, v = l,...,m, and if the con­
ditions
Uy{y) = 0, v = 1,...,m, (4)
are imposed on the functions e C(n), we call these the boundary conditions
which the functions^ must satisfy.
We denote by 2 the set of all functions y e CM which satisfy a specified
system of boundary conditions of the form (4). Clearly is a linear subspace
in C(n) which coincides with C(n) only if the conditions (4) are entirely lacking
or if all their coefficients vanish.
Suppose a certain differential expression l(y) and a particular subspace 3)
defined by conditions of the form (4) are given. To each function y e 9) we let
the function u = l(y) correspond. This relation is a linear operator with as
its domain of definition; we denote it by L. Using the notation previously
introduced, we write
u — Ly.
The operator L is called the differential operator generated by the differential
expression l(y) and the boundary conditions (4).
4 §1 LINEAR DIFFERENTIAL OPERATOR

In this way—and this fact will be very important in the sequel—one and the
same differential expression can generate various differential operators depend­
ing on the way in which the boundary conditions (4) are chosen.
If, in particular, the conditions (4) are absent, we obtain a differential
operator, denoted by Lx, with the domain of definition = C(n). is
obviously an extension of all other operators L which can be generated by the
same differential expression l(y).
For many questions it turns out to be profitable to investigate not merely
the operator Lx which has the widest domain of definition, but also the
operators considered above, which are restrictions of Lv
Certain of the forms U fy) may be expressible as linear combinations of the
remaining forms. The corresponding conditions Uv(y) = 0 are then im­
mediate consequences of the remaining conditions, and therefore, being
superfluous, can be discarded. Hence we may, ab initio, regard the forms
U fy) as being linearly independent. This implies, of course, that the rank of
the matrix formed from the coefficients of these forms is equal to m.
For m = 2n, the equations (4) are equivalent to

The particular differential operator generated by l(y) and these conditions is

denoted by L0.
We can also give the following geometrical interpretation of the boundary
conditions. Let 77 stand for the whole of system (2): 77 = (ya,y'a,■■■,)'i"
K . . . , ^ ) . Clearly we may regard 77 as a vector in the 2/7-dimensional space
R 2„. The conditions (4) separate out from the space R 2n a definite subspace
J i of dimensions (2/7 — m). These conditions imply that the vector ?7 which
corresponds to a function y belongs to the subspace .

4. The Homogeneous Boundary-Value Problem

The problem of determining a function y e C(n) which satisfies the conditions

Ky) = 0 , (5)
U f y ) = 0, v = 1,2,...,/?/. ( 6)

is called the homogeneous boundary-value problem. If L is the operator which is

generated by the differential expression /(_)’) and the boundary conditions (6),
then the homogeneous boundary-value problem amounts to finding, in the
domain of definition of the operator L, a function y for which L vanishes.
Clearly, any homogeneous boundary-value problem always has at least one
solution, viz. the solution y = 0. This is called the trivial solution. A homo­
geneous boundary-value problem can, however, have other “non-trivial”
solutions, i.e. solutions which are not identically zero.
 §1.4 HOMOGENEOUS BOUNDARY-VALUE PROBLEM

We now want to ascertain under what conditions a homogeneous boundary-

value problem has non-trivial solutions.
Let y\,yz,...,y„ be linearly independent solutions of the differential equation
l(y) = 0. Then, as is known from the theory of linear differential equations
(see [108] or [17], for instance), any solution of the equation l(y) = 0, and
therefore the solution of the homogeneous boundary-value problem, too,
must be expressible in the form

)’ = c i ) ’i + c 2.)’2 + ••• + c„_y„,

where c1,c2,...,c„ are certain constants. By substitution in the conditions (6)

we obtain a system of linear, homogeneous equations

+ ci^i(y f) + ••• + cnUi(y„) —

c\Ui(y-d + c2u 2(yd + ••• + cnU2(y„) = 0
(7)

ClUmiyi) + C2^m()ff + ••• + c„Um(y„) =

for the determination of the constants c1,c2,...,c„. Let
matrix
'U fy f) U fy 2) U fy n)
U,(yx) L2( v2) U fy n)
U - ( 8)

Um(y\) Um(y2) Um(yn)

Then there are precisely (n — r) independent solutions of the equations (7) for
the c1,c2,...,c„; and to these correspond (n — r) independent solutions y of the
boundary-value problem.
Hence the following result:
I. If the rank of the matrix U is equal to r, the homogeneous, boundary-value
problem has exactly (n — r) independent solutions.
In particular,
II (a). The homogeneous, boundary-value problem has a non-trivial solution if
and only if the rank r of the matrix U is less than the order n of the
differential expression I.
(b) . For m < n, the homogeneous, boundary-value problem always has a
non-trivial solution.
(c) . For m = n, the homogeneous, boundary-value problem has a non-trivial
solution if and only if the determinant of the matrix U (in this case, a
square matrix) vanishes.
The rank of the matrix U does not depend on the choice of the basic system
6 §1 LINEAR DIFFERENTIAL OPERATOR

y y n. For, the transition from one such system yi,...,y„ to another

is effected by means of a linear transformation
\
n
9i = ^i atjyj, i=
j -1
with a non-zero determinant. Since in this transition the matrix U is multiplied
by the matrix [a[;] (i,j = 1 , the rank of U remains unchanged.
The rank of the matrix U is called the rank of the boundary-value problem.
The definition which we have given of the homogeneous boundary-value
problem may be generalized as follows.
Let U1(y),U2(y),...,Un(y) be linearly independent functionals in C(rt). (As
n
the norm in C(n), we can take |y| = £ max |y k)(*)l-) Let l(y) be a differen-
k = 0 a< x< .b

tial expression of the «th order.

Then by a generalized homogeneous boundary-value problem we mean the
problem of determining a function y e C(n) which satisfies the conditions

/0 0 = 0 , U ,(y )= 0 , V = 1,2,
The results of this section will clearly go over completely to the case of a
generalized boundary-value problem.
Certain special cases of the generalized boundary-value problem are treated
in Tamarkin’s articles [110] and by other writers (see [106] and [45a], Part II,
§5 in which a detailed list of references is given); but the question has not so
far been discussed in the full generality given here.

5. Lagrange’s Formula. The Adjoint Differential Expression

We now assume that the coefficients Pu(x),k = 0 , 1 , 2 ,...,«, of the differential
expression

*00 = Po(x) dfj - + Pi(x) d~ — + ••• + Pn(x)y

have continuous derivatives up to the order (n — k) inclusive on the interval

[a,b]. Further let y and z be two arbitrary functions in C(,,). By k partial
integrations we get

\ ba Pn-kZy(k)d x = [ p n - k Z y ^ - P - ( p n_ kz ) ' y {k- 2) + ...

+ + ( - l ) k \ ba y(pn- k:)(k)dx. (9)

(Here z denotes the complex number conjugate to z, and z = z (a ) denotes the

function whose values are conjugate to those of z(x).)
 §1.5 ADJOINT DIFFERENTIAL EXPRESSION 7
If we put k — n, n — in (9) and add the resulting equations, we
obtain the formula
J a liy)zdx = P(rj, 0 + J* yl*(z)dz, (10)
where
l* (z) = ( - 1 ) " ( M ) (,,) + ( - 0 " - 1(ftz )(" -1) + ( - 1 ) " - 2(/>2z)(" - 2)
+ ••• + P„z, (11)
and is a certain bilinear form in the variables
v = (yay a, •••d’a ~ W b •••A " ~x))
and £ = (za,z'a,...,z<
? - 1\ z b,z'b,...,z{bn- 1)).
The differential expression l*(z) defined by the formula (11) is called the adjoint
differential expression of l(y), and (10) is called Lagrange's formula.
If we apply the preceding considerations to the integral
J a l*(z).ydx,
we obtain an expression of the form

J a
l*{z)ydx = Qi-rjX) + Jf a z.KJ)dx.
Hence the differential expression l(y) is adjoint to /*(z):
l**(y) = Ky).
In other words, the differential expressions l(y) and l*(y) are mutually adjoint.
From the definition (11) of adjoint expressions we see at once that
(/: + /*)* = If + /,*, (12)
and, if Ais any number,
(A/)*=A/*. (13)
A differential expression l{y) is said to be self-adjoint1 if /* = /. It follows
from (12) and (13) that:
III (a). A sum of self-adjoint differential expressions is also a self-adjoint
differential expression.
(b). The product of a self-adjoint differential expression with a real number
is also a self-adjoint differential expression.
We now want to determine the general form of all self-adjoint differential
expressions.
IV. Any self-adjoint differential expression is a sum of differential expressions
of the form
(pyM)(v); k -iO O = W p / ' - * ) ™ + ( ^ (v))(v_1)]>
/8vOO =
where p in each case is a real-valued function on [a,b\.

1 In the E nglish literature such an expression is often said to be formally self-acljoint.

8 §1 LINEAR DIFFERENTIAL OPERATOR

Proof. We consider the integrals

P kXy)zdx and JPa kv-i(.y)zdx,

Ja \
with y ,z e C(n).
By partial integration it is at once clear that l2v(y) and l2v- f y ) are self-adjoint
differential expressions. Consequently a sum of expressions of the form
l2v(y) and 4 v- i 0 ’) is likewise a self-adjoint differential expression.
Conversely, let
Ky) = P o/n) + + ••• + p ny
be a self-adjoint differential expression; we shall prove that l(y) can be
represented as the sum of expressions of the form l2v(y) and 4v-i(>’)-
By definition, l(y) must agree with the adjoint expression
l*(y) = ( - ! ) " ( A>T)(n) + ( - l ) ',_1(A>’)(n“ 1) + ••• + P„y
= ( —l)nPoy(n) + [( —ly ^ n p o + ( ~ l ) n“ 1jp1]>,(''" 1) + •••
Hence, in particular,
Po = ( - 1 YPo- (14)
If n is even and = 2/x, (14) implies p0 = p0, and so p0 takes only real values.
If we subtract from /(>’) the self-adjoint expression
4„G0 = (Poyipl))(ft) = p 0y(n) + (p - i)Poy(n~1) + •••,
we obtain the self-adjoint expression !(y) — l2fl(y) of order n — 1.
Next, if n is odd and = 2 p — \, (14) implies pQ= —p 0, and therefore
Pq = ip, where p takes only real values. By subtracting from l(y) the self-
adjoint expression
4 m- i OO = i[('> /''~ 1))(/i) + ( ^ })(- 1}]
= ( /» j(2'x-1) + ^ / » (2" - 2) + ...
= p0y(n) + $ppoy(n~1) + —,
we obtain the self-adjoint expression l(y) — 4m- i( t) of order (n — 1).
It follows from this argument that if self-adjoint expressions of the form
l2v(y) and l2fl-i(y) are subtracted successively from l(y), then finally a self-
adjoint expression of zero order will be obtained. This must necessarily agree
with lQ(y) = py, where p denotes a real function. This proves assertion IV.
If, in particular, l(y) is a self-adjoint expression with real coefficients, then it
can have no addends of the form l2ft-i(y ). Hence
V. Any self-adjoint differential expression with real coefficients is necessarily
of even order and has the form
Ky) = (Po/'0) ^ + (p1y(tt~1)f>i- 1) + ... + (Pli _!>’')' + Pfly
where p0,Pi,---,Pn are real-valuedfunctions.
 §1.6 ADJOINT OPERATOR 9

6. Adjoint Boundary Conditions. The Adjoint Operator

Let be linearly independent forms in the variables JVVo* -1)>

if m < 2/i, we shall supplement them with other forms
Um+1,...,U 2n to obtain a linearly independent system of 2n forms U1,U2,...,U2n.
Since these forms are linearly independent, the variables
- can be expressed as linear combinations of the forms

We substitute these expressions in the bilinear form P{-qX) which occurred in

Lagrange’s formula (see formula (10)). Then P(rj,£) becomes a linear, homo­
geneous form in the variables Ux,U2,...,U 2n, and its 2n coefficients are them­
selves linear, homogeneous forms, which we denote by V2n,V2n_l,...,V 1, in the
variables z a,z'a,...,z(an~1\ z b,z'b,...,z(bn~l). Lagrange’s formula now takes the
form
J* M z d x = UxV2n + U2V2n. x 4- ... + U2nvx+ j a byl*{z)dx (15)
The forms Vx,v 2y...,v 2n are linearly independent.
To prove this assertion, we first recall that the form P(t),£) arose from the
substitution of the limits x = a and * = b in the integrated terms obtained in
the integration by parts, and so it has the form
P(v, 0 = P M ~ P M ,
and Pa{pX) and Pb(j]X) involve only the values of the functions of y,y',
. . . , / n“ 1),z,z',...,z(n-1) at the points x = a and x = b respectively.
Consequently the matrix of P(-qX) is of the form
—Afl 0 '
( 16)
0 Ab

where A0 and Afcdenote the matrices of Pfj],Q and Pb(rjX) respectively, and 0
the null matrix.
But from formula (9) we see at once that each of the matrices A0 and Ab has
the form
■ .................... (-iy - w
. o

. - p0 ..0 0
_Po 0 ..0 0 _
wherep 0 is to be evaluated at x = a and x = b for A0 and A brespectively. (The
10 §1 LINEAR DIFFERENTIAL OPERATOR

elements above the diagonal in this matrix have not been written down, since
they play no part in the argument.)
The determinants of both these matrices'are non-zero, and from (16) it
follows that the determinant of P{t}X) is also non-zero.
The transition from the variables to the
variables U1,U2,...,U 2n is a linear transformation with non-zero determinant.
Hence the determinant of the matrix of P in the variables
Ux,Ui,...,U 2,n and zfl,z ',...,z ^ -1),z6,zj,...,z£,- 1)
is also non-zero. But this matrix is also the coefficient matrix of the forms

These forms are therefore linearly independent.

The boundary conditions
F1 = 0 ,F 2 = 0,...,F2n_m = 0 (17)
(and all boundary conditions equivalent to them) are said to be adjoint to the
original boundary conditions.
Ui = 0,t/2 = 0,...,£/m = 0. (18)
Boundary conditions are self-adjoint if they are equivalent to their adjoint
boundary conditions.
Let L be the operator generated by the expression l(y) and the boundary
conditions (18). The operator generated by l*(y) and the boundary conditions
(17) will be denoted by L* and called the adjoint operator to L.
It follows from formula (15) and the boundary conditions (17) and (18) that
the equation
L y.zdx = J* y.L*zdx (19)

holds for the operators L and L *, for all y in the domain of definition of L and
all z in the domain of definition of L*.
If we introduce the contracted notation

(y>z) = j l y(x).z(x)dx,
then (19) takes the form
(Ly,z) = (y,L*z). (20)
It follows from the definition of adjoint operators that the operator L is
adjoint to the operator L * :
L** = L.
An operator is self-adjoint if L* = L.
It also follows from the definition of L* that:
 §1.7 ADJOINT BOUND ARY-VALUE PROBLEM 11
An operator L is self-adjoint if and only if it is generated by a self-adjoint
differential expression and self-adjoint boundary conditions.
For a self-adjoint operator L the formula (20) becomes
(Ly,z) = (y,Lz). (21)

7. The Adjoint Boundary-Value Problem

If L* is the operator adjoint to L, the homogeneous boundary-value problem
L*z = 0 (22)
is said to be adjoint to the boundary-value problem
Ly = 0. (23)
In more detail, the boundary-value problem (22) can be written as:
/* 0 0 = 0, (24)
Vv(z) = 0 , v = 1,2,...,2/7 - m. (25)
where /*(z) is the differential expression adjoint to l(y), and (25) are the
boundary conditions adjoint to those of L.
We intend now to obtain a relation between the ranks of adjoint boundary-
value problems.
Let z1,z2,...,z„ be linearly independent solutions of the equation /*(z) = 0,
and let r' be the rank of the matrix
Vl(z1) Vi (* J
V2(Zl) • V2(z„)

• v 2n. m\
Then the adjoint boundary-value problem (24),(25) has exactly (n — r')
independent solutions.
If, on the other hand, y is an arbitrary solution of the boundary-value
problem (23), then the left-hand side of Lagrange’s formula (15) vanishes for
z = z v. Also, U fy) = ... = U Jy) = 0. Hence in this case Lagrange’s
formula simplifies to
Um+i(y)V2n-m(zv) + + U2n(y)Vx(z,) = 0.
Putting v = 1, 2,...,77, we obtain for the forms Um+1{y),...,U2n(y) the system of
equations
Um+l( y )V2n-m(Zl) + ••• + U2n( y ) V f z j ) = 0 , \
Um+l(y)V2n- m(z2) + ... + U2n(y )V fz j) = 0, I

Um+fy ) V 2n-m^ n) + - + U2n{y)V fzn) = 0.

12 §1 LINEAR DIFFERENTIAL OPERATOR

This system has at least (n — r) independent solutions, viz.

Um+i(yj)>--->U2n(yj)> j 1 > 2 — r,

if r is the rank of the boundary-value problem (23), and y x,.. .,yn- r is a system of
linearly independent solutions of (23). Therefore the rank of its matrix is not
greater than 2n — m — (n — r) = n — m + r. Since the matrix of the
system (27) corresponds—up to the arrangement of rows and columns—to the
matrix (26), we have
r' < n — m + r. (28)

We now recall that the two boundary-value problems are mutually adjoint,
and therefore we can interchange their roles. In such an interchange, r and r'
are interchanged, and m goes over into (2n — m); hence r < n — (2n — m)
+ r', i.e.
r < m — n + r\ (29)

Comparing (28) and (29), we have

r' = n — m + r. (30)

VI. The rank r of a boundary-value problem is related to the rank r' of the
adjoint boundary-value problem by the relation

r' = n — m + r.

For the particular case m = n, formula (30) gives r' = r. Hence:

VII. If the number of independent boundary conditions is equal to the order of
the differential expression, then the rank of the boundary-value problem is
equal to the rank of the adjoint boundary-value problem.
In particular it follows from this that:
VIII. If the number of boundary conditions is equal to the order of the differential
expression, and if the associated homogeneous boundary-value problem has only
the trivial solution, then the adjoint boundary-value problem has only the
trivial solution.
Additional information on the topics discussed in this section can be
found in chapter 11 of Coddington and Levinson’s book [17].
We have been considering here problems where the boundary conditions
are prescribed at the end-points of an interval. In mathematical physics and
in applications, however, problems are encountered where the boundary con­
ditions are prescribed at n arbitrary points of an interval [a,b]. Chapter 4 of
Sansone’s book [97] may be consulted on these so-called n-point problems.
 §2.1 EIGENVALUES A N D EIGENFUNCTIONS 13

§2. Eigenvalues and Eigenfunctions of a Differential Operator

1. Definition of Eigenvalues and Eigenfunctions

A number Ais called an eigenvalue of an operator L if there exists in the domain

of definition of the operator L a function y ^ 0 such that
Ly = Ay. (1)
The function^ is called the eigenfunction, of the operator L, for the eigehvalue A.
The operator L may be generated by the differential expression /(y) and the
boundary-conditions
Ui(}’) = 0,...,U„(y) = 0. (2)
Since an eigenfunction y must belong to the domain of definition of the
operator L, it must satisfy the conditions (2). Moreover, Ly = /(y), and
therefore (1) is equivalent to
l(y) = Ay. (3)
Hence:
The eigenvalues of an operator L are those values of the parameter Afor which
the homogeneous boundary-value problem
l(y) = Ay, Uf y) = 0, v = l,2,...,m, (4)
has non-trivial solutions', each of these non-trivial solutions is an eigenfunction
belonging to A.
A linear combination of eigenfunctions which belong to one and the same
eigenvalue A is itself an eigenfunction belonging to A. For, if Ly1 = Ayt and
Ly2 = Ay2, then L ( c ^ + c2y2) = Afoyj + c2y2) for any constants and c2.
Since a homogeneous equation (3) can have, for a given A, not more than n
linearly independent solutions, it follows that the aggregate of all eigenfunctions
which belong to one and the same eigenvalue form a finite-dimensional space of
dimension < n. The dimension of this space is simply the number of linearly
independent solutions of the boundary-value problem (4) for the given
eigenvalue A; this number is called the multiplicity of the eigenvalue.
We shall specify conditions for the determination of eigenvalues.
Let

denote a fundamental system of solutions of the differential equation (3)

which satisfy the initial conditions
0 for j ^ v,
j,v = 1,2, ( 6)
1 for j = v,

From general theorems on the solutions of linear differential equations it

14 §2 EIGENVALUES AND EIGENFUNCTIONS

follows that, for each fixed value of x in [a,b\, the functions (5) are integral,
analytic functions of the parameter A.
From the results of §1.4, the boundary-valile problem (4) has a non-trivial
solution if and only if the rank r of the matrix
tfitVi) U M
W =
lU M ... Um(yn)
is less than n:
and if m < n, then r < n; and in this case the boundary-value problem (4) has
a non-trivial solution for any value of A. Hence, if m < n, any value of A is an
eigenvalue.
If in > n, the rank of the matrix ^ will be less than n if and only if all its
minors of order n vanish. But each of these minors is an integral, analytic
function of A. Hence the only possible alternatives are:
1°. All the nth-order minors of the matrix °U vanish identically, and so, by the
previous result, any value of A is an eigenvalue.
2°. At least one nth-order minor of % does not vanish identically. In this case,
only the zeros of these minors can be eigenvalues, and further, a zero of one
particular minor can be an eigenvalue only if it makes all the other, not
identically zero, minors of order n of ^ also vanish.
Now, a non-vanishing integral function has at most denumerably many
zeros (it need not have any at all), and moreover, these zeros have no finite
point of accumulation. Hence: In case 2°, the operator L has at most denumer­
ably many eigenvalues (it may have none at all), and these eigenvalues have no
finite limit-point.
Summarizing, then, we have the following alternatives.
I For any differential operator L, only the following two possibilities can occur:
1°. Every number A is an eigenvalue for L;
2°. The operator has at most denumerably many eigenvalues (in particular
cases, none at all), and these eigenvalues can have no finite limit-point.
The case when m = n is of particular interest, and in the sequel we shall
consider this particular case, if nothing to the contrary is stated.
We put, then,
't'lO 'i) v ,(y S
A(A) = (7)

By the preceding discussion, A(A) is an integral, analytic function of A, the

so-called characteristic determinant of the operator L (or of the boundary-
value problem Ly — 0); and the following theorems hold:
 §2.2 GENERALIZED EIGENVALUE PROBLEMS 15
II. The eigenvalues of the operator L are the zeros of the function A(A). I f A(A)
vanishes identically, then any number A is an eigenvalue of the operator L.
If, however, A(A) is not identically zero, the operator L has at most de-
numerably many eigenvalues, and these eigenvalues can have no finite limit-
point.
If, in particular, the function A(A) has no zeros at all, then the operator L has
no eigenvalues.
An eigenvalue Amay be a multiple zero of A(A).
III. I f A0 is a zero of the characteristic determinant A(A) with multiplicity v, then
the multiplicity of the eigenvalue A0 cannot be greater than v.
Proof. Let r be the rank of the matrix belonging to the determinant A(A0) ; then
the multiplicity of the eigenvalue A0is equal to (n — r). On the other hand, from
a well-known rule for the differentiation of determinants, all the derivatives of
the determinant A(A) up to the (n — r — l)th order inclusive vanish for A = A0.
Since A0 is a zero of multiplicity v, it follows that n — r — 1 < v — 1, and
hence n — r < v.
If, in particular, v = 1, then n — r < 1; on the other hand, n — r > 1,
since A(A0) = 0. Hence n — r = 1, and we have proved the following:
IV. I f A0 is a simple zero of the characteristic determinant A(A), then the
multiplicity of the eigenvalue A0 of the operator L is also unity. An eigenvalue
is called simple if it is a simple zero of the characteristic determinant A(A).

2. Various Generalizations of Eigenvalue Problems

In applications, more general eigenvalue problems occur from time to time.
The most general formulation of the problem is as follows:
The coefficients of a linear differential expression l(y), and perhaps also those
of the boundary conditions Uv(y) = 0,v = 1,2,...,n, may depend on a certain
complex parameter A. The values of A are to be determined fo r which the
boundary-value problem
l{y )= 0, Uv(y) = 0, v = 1,2,...,n (8)
has a non-trivial solution.
These values of the parameter Aare called the eigenvalues of the boundary-
value problem (8), and the corresponding non-trivial solutions y are called its
eigenfunctions.
Using the same considerations as in the preceding subsection, we construct
the characteristic determinant A(A) according to formula (7); the zeros of A(A)
will again be eigenvalues.
If the coefficients of the expression l(y) and of the boundary-conditions are
analytic functions of the parameter Awhich are regular in a certain domain ^
of the A-plane, then the determinant A(A) will also be a regular, analytic function
of Ain the domain < S . There are then the two following possibilities:
16 §2 EIGENVALUES AND EIGENFUNCTIONS

1°. A(A) = 0 in the domain then every number Ae ^ is an eigenvalue.

2°. A(A) ^ 0; there then exist in the domain 3? at most denumerably many
eigenvalues and these have no point of accumulation within
If, in particular, the coefficients of the expression l(y) and the forms UY(y)
are integral, analytic functions of A, then A(A) is also an integral function; and
hence in this case the theorems enunciated at the end of the last subsection
remain valid. The most commonly occurring case is where /(y) = 4(c) + A4(c),
and both 4(c) and 4(c) and the forms UY(y) are independent of A. The
coefficients of the expression l(y) are then linear (and therefore integral)
functions of the parameter A. So in this case, too, the theorems states in sub­
section 2 hold.
We may mention, further, the following important generalization of the
eigenvalue problem defined above:
4(c) + *400 = 0> £400 = 0 , V = 1,2
where 4,4 are self-adjoint differential expressions and Uv{y) = 0(v = l,2,...,n)
are self-adjoint boundary-conditions relative to the differential expressions
4(c) and 4(c)- This problem has been analyzed in detail by E. Kamke (see
[45]).

3. Associated Functions
We now turn our attention to the generalized eigenvalue problem
/(c) = 0, £4,00 = 0, v = 1,2,
for which the coefficients of the differential expression l(y) and of the forms
Uv{y) are integral analytic functions of the parameter A. Let A0 be an eigenvalue
and <p(x) an eigenfunction belonging to A0 for this problem. The functions
<Pl(x),<p2{x),...,cpk(x)
are said to be associated with the eigenfunction <p(x) if all the functions
<Pi(x),<p2(x),...,<pk(x) satisfy the boundary conditions
» 1 gp
L — UV(<P„-P) = 0 , <p0 = <p, fi = 0,1,...,^; v - 1,2,...,?;, (9)
p- o p '
for A = A0, and if for A = A0 the following relations hold:
K<p) = 0 \

( 10)

I(n) + 1. L + - + h w K9) = 0
 §2.3 ASSOCIATED FUNCTIONS 17
dq
Here 1(f) denotes the differential expression whose coefficients are the <7th
derivatives with respect to Aof the corresponding coefficients of the differential
expression 1(f). The number k is called the length of the system of associated
functions. An eigenfunction <f>(x) is said to have multiplicity m if there is a
system of functions associated with <p(x) of length (m — 1) but no system of
length m.
V. I f A0 is a v-fold zero of the function A(A), then the multiplicity of any eigen­
function is not greater than v.
Proof. We assume the contrary; there is a system of functions associated with
(p(x) of length v:

Let >’i(XA) be a solution of the equation l(y) = 0, which, for x = a, satisfies the
same initial conditions as the function

f(x,X) — <p(x) + (A — K)<Pi(x) + — A0)29?2(x) + ... + (A — A0)v<j?v(x).

Then the first (V + 1) terms in the expansion of the function j^foA) in powers
of (A — A0) agree with the corresponding addends in the representation of the
function f(x,X). For, if we put A = A0 in the relations

/O’l) = 0
01)
/ q\a ,X )= f^ (a ,X ), q = 0,!,...,(« — 1)
it follows that the function yi(x,A0) satisfies the same equation and the same
initial conditions for jc = a as the function cp(x) does; hence^(x,A) = <p(x). If
we differentiate the relations (11) with respect to A, and then put A = A0, we get,
for A = A0,
dyi
dX

dy[q)(a,X)
= $ \a ). q = 0 , 1 , — 1),
8X

and comparison with (10) shows that

Repeating the same argument, we arrive at

1 dqy 1(x,X) 1
q = 1,2, ...,v.
q\ 8Xq
Now we construct solutions ^ (x . A),y3(x, A), •• ° f e q u a t i o n l(y) — 0
such that the functions
18 §2 EIGENVALUES AND EIGENFUNCTIONS

are linearly independent for A = A0, and with the help of these functions we
form the determinant A(A).
Since, by hypothesis, the functions <p,<pi,...,<pv satisfy the boundary con­
ditions (9), the power series for U fy i) in powers of (A — A0) must begin with
(A — A0)v+1. The first column of the determinant A(A) would therefore contain
the common factor (A — A0)v+x, i.e. the number A0 would be a zero of A(A) with
at least (v + l)-fold multiplicity, and this contradicts our hypothesis. Hence
the theorem.
Now let be an eigenfunction of maximum multiplicity, <p2an eigenfunction
of maximum multiplicity among the eigenfunctions linearly independent of
<Px, and, in general, <pj an eigenfunction of maximum multiplicity among the
eigenfunctions linearly independent of Clearly, the total number
of such eigenfunctions is equal to the multiplicity p of the eigenvalue A0. We
denote the multiplicity of the eigenfunction <pj by mj, and the system of
associated functions belonging to it by
jl’Tj2>••• j,mj —1•
It follows immediately from the definition of the numbers nij that
m1 > w2 > ... > mp.
The system of functions so obtained
<Pj><Pjl><Pj*»••• -i» j = 1,2, • • •,p,

is called a canonical system of eigenfunctions and their associated functions.

VI. The sum of the multiplicities mx + m2 + ... + mp is equal to the multiplicity
of A0 as a zero of A(A):
i.e. mx + m2 + ... + mp = v.
Proof. Let
4>P+i(x)dp+fx),...,if>n(x)
be any solutions of the equation /(;>) = 0 for A = A0, which together with the
eigenfunctions <Pi,<P2,...,<pp form a fundamental system of solutions of the
equation l(y) = 0 for A = A0. As above, we can construct the solutions
°f the equation l(y) = 0 whose expansions in powers of (A — A0)
have the form
.FiO,A) = <PiM Hr <PfiO)(A - A0) + ... + i(x)(A - Ao)”*'-1
+ '/'i(*)(* ~ Ao)”" + •••, for i = 1,2,...,p,
y t(x,A) = <Af(x) + ... for i = (p + 1),...,//,
and these functions represent a fundamental system of solutions of the
equation l(y) = 0 for A = A0. We form with them the determinant A(A). The
firstp columns of A(A) contain respectively the factors (A — A0)m>,(A — A0)m«,...,
 §2.3 ASSOCIATED FUNCTIONS 19
(A — A0)mp, and so the determinant A(A) contains the factor (A — A0)m>+m»+•••+mp
Therefore
A(A) = (A — A0)m>+m»+ •••+mpA1(A))
and A1(A0) = det[t/,j], i j = 1,2,...,#i, with Uu = U ^ j ) for i = 1,
and j = (p + 1),(p + 2),...,«,
1 d ] 82
but U;j = UM j) + + j j e X 2 u &Pj,mj-2) + •••

1 dmJ
+ n i j ] d \ mJ U i('<Pj) f ° r * = h ' ,n a n d j = h -’p -

To prove that v = mx + m2 + ... + we have only to show that

Ai(A0) 5* 0. Let us assume the contrary, that A1(A0) = 0. Then one column of
this determinant, the column with suffix j for example, is a linear combination
of the columns with suffixes greater than j; let ocj+1)...,oin be the coefficients of
this linear combination.
First, let j > p. Then this implies that
4> = 4>j — “ j+i'Aj+i - ... - <*„</»„
satisfies the boundary conditions t/v(</>) = 0, v = l,...,n, for A = A0. Since in
this case all the are solutions of l(y) = 0, ^ would be an eigen­
function and would have to be representable as a linear combination of the
functions But this contradicts the linear independence of the
system <p1,<p2,...,<pp,ifip+1,...,tfjn.
If, on the other hand,./ < p, then we put
^,(x,A) = (A — A0)"v-m%(x,A) for i < p,
y-,(x,X) = (A — A0)m^i(x,A) for i > p,
and >’(*,A) = ^(x,A) — <xJ+1p j+1(x,X) — ... — ocnyn(x,X).
Then l(y) = 0.
By differentiating this equation my times with respect to A and then putting
A = A0. it follows that the functions
1 8y(x,A) 1 d2y(X>A) 1 d”jy(X>X) _ , .
1! d\ ’ 2\ dX2 ’ " ’’m ! d \ mJ n '

satisfy the equations (10) with k = m}. On the other hand, it also follows that
the boundary conditions (9) are satisfied by the same functions, again with
k = trip and in particular the last equation
W 1 W v/ 1 8mJ~Py\ Q
p? oP'. W \ K ~ p V- dXmj~P)
holds, again because of the vanishing of A^A,,) and the special construction of
20 §2 EIGENVALUES AND EIGENFUNCTIONS

y(;c,A). These functions would therefore form a system of functions (of length
mj) associated with the eigenfunction y(x,A0). But this is impossible, because
y(.x,A0) is a linear combination of the eigenfunctions , and there­
fore, by the definition of these functions, it has a multiplicity of at most mj.
Theorem VI means that a canonical system of eigenfunctions and their
conjugate functions generate a subspace, the dimension of which is equal to the
multiplicity of the zero of A(A) (for it can readily be seen that the eigenfunctions
and associated functions of the canonical system are linearly independent), so
that the associated functions complete the set of eigenfunctions to form a space
of the necessary dimension.
The theory of associated functions is analogous to the theory of elementary
divisors of linear operators in a finite-dimensional space (cf., e.g. [28a]). The
presentation of the theory of associated functions given here was originated
by Keldysh [47].

4. Relation between the Eigenvalues and the Eigenfunctions of Adjoint Operators1

THEOREM 1
I f Ais an eigenvalue of multiplicity p of the operator L, then A is an eigenvalue
of the adjoint operator L* and has the same multiplicity p.
Proof. The operator L is generated by the differential expression l{y) and the
boundary conditions
U fy) = 0, v = 1,2,...,«,
while the operator L* is generated by the differential expression l*(y) and the
conditions
K(y) = 0, v = 1 ,2 ,...,« .
We put Ify ) = l(y) - Ay;
then /* (y) = l*(y) — Ay.
If A is an eigenvalue of multiplicity p of the operator L, then the boundary-
value problem
kiy) = 0, U^y) = 0. v = 1,2,...,/!
has p linearly independent solutions.
By the results of §1.7 (Theorem VII) it follows that the adjoint boundary-
value problem
/* (y )= 0 , Kv(y) = 0, v = 1,2,
also has p linearly independent solutions. This means that Ais an eigenvalue of
the operator L* and has multiplicity p.
Two functions y(x) and z(x) are said to be orthogonal if (y,z) = 0.2
1 We remind the reader that only the case m = // is now being treated.
2W e remind the reader o f the n otation introduced in §1: (y,z) = J* y(x).z(.x)dx.
 §2.5 EIGENVALUES OF SELF-ADJOINT OPERATOR 21
THEOREM 2
Eigenfunctions of the operators L and L* corresponding to the eigenvalues A
and p. respectively are orthogonal if A ^ p.
Proof. Let y be an eigenfunction of the operator L for the eigenvalue A, and z
an eigenfunction of the operator L* for the eigenvalue so that
Ly = A_y
and L*z = p,z.

Hence t Ly,z) = (Ay,z) = X(y,z),

and (y,L*z) = (y,fiz) = fi(y,z).

But from §1.6 (equation (20)), (Ly,z) = (y,L*z); so on subtracting the last
two equations we obtain
(A — fi)(y,z) = 0.

If A ^ p, this implies (y,z) = 0, i.e. the functions y and z are orthogonal.

5. Eigenvalues and Eigenfunctions of a Self-adjoint Operator

THEOREM 3
All eigenvalues of a self-adjoint operator are real.
Proof. For a self-adjoint operator L we have

(Ly,z) = (y,Lz) ;
in particular,
(Ly,y) = (y,Ly).
But (;v,Ly) = (Ly,y); hence (Ly,y) is real.
Now let Abe an eigenvalue of the operator L, and y an eigenfunction belong­
ing to A. The equation Ly = Xy implies that (Ly,y) = A(j,j).
Since ( j ^ ) > 0 and (Ly,y) is real, A = - must also be real.
Ky^y)
Theorems 2 and 3 yield the
COROLLARY

Eigenfunctions of a self-adjoint operator which belong to different eigenvalues

are orthogonal.
Eigenfunctions which belong to one and the same eigenvalue can also be
chosen so as to be mutually orthogonal. To do this, we simply choose an
orthogonal basis in the subspace of eigenfunctions which belong to one and the
same eigenvalue.
Altogether, we obtain in this way a system of pairwise orthogonal eigen­
functions of the self-adjoint operator L.
22 §2 EIGENVALUES A N D EIGENFUNCTIONS

6. Examples of Eigenvalue Problems

1. -/= A y; * 0 )= J < 1 );N / ( 0 ) = / ( l ) . (13)
The general solution of the differential equation is
y = A cosx^/X + B sin x^JX.
By substitution into the boundary conditions, we see at once that the eigen­
values are given by
A„ = (2mr)2, n = 0,1,2,3,...
For n 0, there are, corresponding to the eigenvalue A0, two linearly
independent eigenfunctions
cos (2mrx) and sin (2rmx);
A„ is therefore a double eigenvalue for n A 0.
For n = 0, there is, corresponding to the eigenvalue A0, only one eigen­
function y = 1 (up to a constant factor); A0 is therefore a simple eigenvalue.

2. Flexure of a Rod with one end fixed and one end free.
We consider a rod with one end, x = /, firmly fixed (Fig. 1) and its other end
free, x = 0. At the free end a compressive force P acts along the rod.
It is a familiar fact that for sufficiently small values of P the straight line
shape of the rod is stable, but there is a critical value P0 of the force P such that,
for P > P0, the rod loses its straightness and bends.
We shall investigate this bending by considering an equilibrium position
differing but little from the straight-line position (Fig. 2).
The equation for the bent axis of the rod y = j(x) is then such that
Py = —Ely", (14)

where / is the moment of inertia of the cross-section of the rod and E is the
modulus of elasticity. The two sides of this equation represent the bending
moment due to the applied force and the bending moment in the rod at
distance x from the origin (see [111], for instance).
We restrict ourselves to the simple case of a homogeneous rod of constant
P
cross-section; then the stiffness £ 7 is constant. Using the substitution A = —
El
we obtain the equation
-y"=Xy- (15)

From Fig. 2 it is clear that the function >’(x) satisfies the conditions

* 0 ) = 0, / ( / ) = 0. (16)
 §2.6 TYPICAL EIGENVALUE PROBLEMS 23
^ -------------- 1

>f P

X= /
77777777777/ 77777777777

>r X

Fig. 1.

So, by (15) and (16) we have formulated the eigenvalue problem. By sub­
stituting the general solution
y = A cos x^/A -f B sin x j A
for equation (15) into the boundary conditions (16) we find A = 0,
cos y f\l = 0, and so

These eigenvalues are simple, and to them correspond the eigenfunctions

, s D . 2n — 1
y{x) = B sm ——— ttx.

The critical loads corresponding to the eigenvalues are

Apart from a numerical factor, the eigenfunctions describe the corres­

ponding equilibrium forms of the rod (Fig. 3).
For a rod with varying cross-section or for a non-homogeneous rod, £7 is a
function of x. Writing -J- = />(*),we obtain the following eigenvalue problem:
El
- y " = Pp{x)y, M 0 )= 0 , / ( / ) = 0.
In the general case for an arbitrary function p(x), only approximate methods
are possible for calculating the eigenvalues and eigenfunctions.
24 §2 EIGENVALUES AND EIGENFUNCTIONS

Pi '
y*ix) T

:77777777

3. Flexure of a Rod with one end fixed and one end supported.
In this case there is a horizontal reaction H at the end x = 0 (Fig. 4). So the
bending moment is M = Py — H x = —Ely".
By differentiating this equation twice, we get
{Ely")" = —Py"; (17)
and the function ^(a) must satisfy the boundary conditions
y( 0 ) = 0 , / '( 0 ) = 0 , y{l) = 0, / ( / ) = 0. (18)

Fig. 4.
 §2.6 TYPICAL EIGENVALUE PROBLEMS 25
The first and third of these boundary conditions are obvious; the second means
that the bending moment vanishes at the point a — 0 (since the rod is supported
there); the fourth holds because the end a = / is fixed and therefore the tangent
there is parallel to the x-axis.
So we have the eigenvalue problem (17), (18). Suppose now that E l is
constant; we write PIE l = k 2, and also, for simplicity, take / = 1. Then
equation (17) can be written in the form
= —k-y". (19)
Its general integral, for k ^ 0, is
y = A + Bx + C cos kx + D sin kx. (20)
The boundary conditions (18) give directly that A — C — 0 and that the
eigenvalues are the roots of the equation
tan k = k.
These are the abscissae of the points of intersection of the graphs (Fig. 5)
y =- tan x and y — x.
The eigenvalues so obtained are simple, and the corresponding eigenfunctions
are
y = sin k x — x sin k.

Fig. 5.
26 §2 EIGENVALUES AND EIGENFUNCTIONS

For k = 0, the general integral of the differential equation is

y = A + Bx + Cx2 + Dx3; (21)
\

but the boundary conditions (18) can be satisfied only if A = B = C = D = 0.

Consequently k = 0 is not an eigenvalue of the problem.

4. = - k * y "; / ( 0 ) = /"(0) = X I) = / ( I ) = 0.
For k — 0, the general integral (21) of the differential equation satisfies the
given boundary conditions only if A — B = C — D = 0. If k ^ 0, the
general integral (20) of the differential equations satisfies these boundary con­
ditions again only A = B = C = D = 0. Consequently this problem has
no eigenvalues.

5. y " = - k y ; X0) = X(0) = X(0) = X D = 0.

As above, we find that k = 0 is not an eigenvalue. If, for k ^ 0, we sub­
stitute the general solution (20) of the differential equation into the boundary
conditions, we obtain
sin k — k.
This equation has no real roots other than zero. So the problem considered has
only non-real eigenvalues.

6. Natural Vibrations of a Stretched String.

The equation of free, transverse vibrations of a string has the form
d*u d2u
= a2 ( 22)
dt2
where u = u(x,t) is the displacement of the point of the string with abscissa .r
at time t, and a is a constant.
Solutions of equation (22) which are of the form
y = y(x) sin (aojt + <p) (23)

Fig. 6.
 §3.1 INVERSE OPERATOR 27
represent the natural vibrations of the string. Substitution in (22) shows that
the function .y(x) has to satisfy the differential equation
y" + «>2y - 0. (24)
If we take the length of the string as unity, then the condition that the string is
fixed at its ends (Fig. 6) can be written in the form
X0) = y(l) = 0. (25)
So we have an eigenvalue problem (24), (25).
We substitute the general solution
y — A cos a>x + B sin a>x
of equation (24) into the boundary conditions (25) and obtain A = 0,
sin a> — 0.
Hence
CD = flTT, n = 1,2,3,....
So
= M 2, n = 1,2,3,...
are the eigenvalues. They are simple, and to them belong the eigenfunctions
.y„(x) = Bn sin n7TX-
The eigenvalues and eigenfunctions have a simple interpretation in
mechanics. Up to a constant factor, they are the squares of the natural fre­
quencies of the string and the eigenfunctions yield the modal shapes for the
natural vibrations.

§3. Green's Function fo r a Linear Differential Operator

1. General Definition of the Inverse Operator

An operator B is said to be inverse to an operator A if the domain of definition
SAb of the operator B coincides with the range of values 0tA of the operator A
and if, for all x e A, the relation
B(Ax) = x
holds. The inverse operator to A will be denoted by A ~ x. Hence, for all x e@A,
A '^ A x ) = x (1)
I. I f the operator A has an inverse operator A -1, then the operator A is inverse
to the operator A ~ x: i.e.
( A - 'ff^ A . (2)
For equation (1) shows that the range of values 0tA- x of the operator A -1
28 §3 g r e e n ’s f u n c t i o n

coincides with the domain of definition 3>A of the operator A. Further, if we

apply the operator A to both sides of (1) and put y = Ax, we get
A A ~ xy = y '
for all y e = &>A-i. And this implies precisely that A is inverse to A ~ x.
II. The operator A has an inverse if and only if the equation
Ax - 0 (3)
has only the “ trivial” solution x = 0.
For, suppose that the operator A has an inverse operator A ~ x. If we apply
A ~ x to both sides of (3), we obtain .x = 0. Hence the condition (3) is necessary.
If, on the other hand, we assume that the condition (3) holds, then the
equation
A x — y, (4)
for all y e &A, has a unique solution x e &A; because, for each such y, at least
one solution exists by virtue of the definition of the range 3%A. And the solution
is unique; for if A x' = y were also true, then we should have A(x — x') — 0.
By hypothesis, this is possible only if x — x ' — 0, i.e. if x = x'. To each
element y tS%A we assign the solution x of equation (4). But we see im­
mediately that this correspondence is a linear operator, and moreover it is the
inverse to A.

2. The Problem of Inverting a Differential Operator

Let L be a differential operator, generated from the differential expression

Kr) = M * ) ~ + P i(*) + - +Pn(x)y

and the boundary conditions

£^>00=0, v = 1,2
We assume that the homogeneous boundary-value problem Ly — 0 has
only the trivial solution y = 0; i.e. for an arbitrary fundamental system of
solutions yi,y2,---,yn of the equation l{y) = 0, the rank of the matrix
U1{y1) ... U fy n)
U fy 0 - U,{yn)

_Un{yf) Un{y2) - Un{yn)

is equal to n, and so the determinant
det \U lyj)\ * 0, i j = 1 ,2 ,.. (5)
 § 3 .3 CONSTRUCTION OF GREEN’S FUNCTION 29
By II, L then has an inverse L~ \ whose domain of definition coincides with the
range of values of the operator L. We set ourselves the task of finding an
explicit expression for L ~ l. As we shall see, L r x turns out to be an integral
operator with a continuous kernel; the kernel is called Green s function for the
operator L. We shall give below an explicit expression for the Green’s function.

3. Construction of Green’s Function

Green’s function for an operator L is to be understood to be a function G(x,£)
satisfying the following conditions:
1°. G(x,£) is continuous and has continuous derivatives with respect to x up to
order (n — 2) inclusive fo r all values of x and £ in the interval [a,b\.
2 ° . For any fixed value of £ in the interval (a,b) the function G(x,£) has con­
tinuous derivatives of orders (n — 1) and n with respect to x in each of the
intervals [a,£) and (£,£>]; the (n — 1)th derivative is discontinuous at x = £ with
1 .
a jump of
PoiZ)'
-1 gn-l 1
^ + o , o - ^ G(£ - 0,0 =
dx* Po(0'
3 ° . In each of the intervals [a,£) and {£,b], G(x,£), considered as a function of
x, satisfies the equation 1{G) = 0 and the boundary conditions UfG) = 0,
v = \,2,...,n.
THEOREM 1
I f the boundary-value problem Ly = 0 has only the trivial solution, then the
operator L has one and only one Green's function.
Proof. Let y^y^, ...,ynbe linearly independent solutions of the equation l(y) = 0.
Since the function G in the interval [a,£) also satisfies this equation, we have
G(x,£) = a ^ f x ) + ... + anyn(x) for a < x < £ ,

where a!,a2,...,an are certain functions of £. Similarly,

G(x,£) = ^ ( x ) + ... + bnyn{*) for £ < x < b.
The requirement for continuity of the function G(x,0 and its first {n — 2)
derivatives at x = £ gives the equations

+ ••• + W niO ] ~ + ••• + bnyn{£)\ = 0.

+ - + any'n(0] - (biy'iiO + ••• + bny'n(£)] = 0,

K y i - ' ’® + - + - 1 = <>•
30 §3 g r e e n ’s f u n c t i o n

Moreover, the condition

dn-\ dn~1 1
G({ + 0 ,0 - G{£ - 0 ,0 =
dxn~1 dxn~x MO
is equivalent to
1
[<3i/1" - l)(0 + ... + an^ ”-1)(0] — + ... + 6„^n-1)(0]
Pod)'
Putting
Cy by Qyy V 1,3,
we obtain for the cv the system of equations

W iiO + ••• + c„y„(€) =0,

Ciy'iiZ) + ••• + cny'M) =0,

( 6)
cxy { ~ 2)(0 + ... + cny ("~2\ 0 = 0,

c ^ r 'X Z ) + ••• + cn/ nn_1)(0 = + —-

the determinant of which is the Wronskian determinant for the fundamental

systemyi,--.,y„ for x = 0 and therefore does not vanish.
Hence the system always has a solution and the functions cv are uniquely
determined by the system (6). To determine the functions av and bv we bring in
the boundary conditions; we write the form £/v(y) as
Uv(y) = Uy„(y) + Uvb(y), (7)
where Uva(y) is the sum of all the addends containing y a,y'a, . . . , y and
Uvb(y) is the sum of all the addends containing
Hence
Uy{G) = a-JJyciy^) + ... + anUya(yn) + b-JJyhiy^) + ... + bnUvb(y’n) = 0.
Replacing the coefficients ak by their expressions as ak — bk — ck, we get
b\Uyb(yd + ... + b„Uvb(yn) + (^i — G^vaCFi) + ... + (bn — cn)Uva(yn) = 0.
Hence, by (7),
h U v{y1) + ... + bnUy(yn) = c . U y M + ... + cnUya(yn). (8)
Putting v = 1 , 2 , . . we obtain from (8) a system of equations in the unknowns
b1,b2,...,bn, for which, by (5), the determinant does not vanish. Consequently
the system of equations has a unique solution b1,b2,...,bn. But then the functions
av are also determined uniquely from aY = by — cv. So we have proved the
existence and the uniqueness of the Green’s function.
 §3.4 INVERSION BY GREEN’S FUNCTION 31

4. Inversion of a Differential Operator by means of the Green’s Function

Suppose that the equation Ly = 0 has only the trivial solution y = 0, so that
the inverse L -1 (see subsection 1) and its Green’s function (defined as in
subsection 3) exist. If L -1/ = y, then

Ly = / , (9)

i.e. y is a solution of the equation

Ky) = / , (io)

and also satisfies the boundary conditions

U fy )= 0 , v= l , 2 (11)

We shall show that this solution exists for any function f i x ) which is con­
tinuous in the interval [a,b\, and that it can be determined by means of the
Green’s function. More precisely, the following theorem is valid.

theorem 2

If the equation Ly = 0 has only the trivial solution, then, fo r any function f(x)
which is continuous in the interval [a,b], there exists a solution of the equation
Ly = f ; this solution is expressed by the formula

X*) = J‘ (12)
where G(x,£) denotes the Green's function fo r the operator L.
Proof. Let >»(x) be the function defined by formula (12). We show that X*)
satisfies the conditions (10) and (11), so that it is a solution of the equation
Ly = / .
The function G(x,£) has continuous derivatives up to the (n — 2)-th order
inclusive; so we may differentiate with respect to x under the integral sign in
(12) (n — 2) times. Hence

y(v)(x) = f ^ ^ fiO d i, v = 1,2,...,(« - 2). (13)

Hence the function >>(x) and its derivatives / v)(x) up to the (n — 2)-th order
inclusive are continuous in the interval [a,b\.
fin —I Q
The function------on the other hand, has a discontinuity at x = So in
dxn~ 1
calculating a n d ^ we may not differentiate any more under the integral
32 §3 g r e e n ’s f u n c t io n

sign without preliminary manipulation. For this reason we write formula (13)
for v= n — 2 in the following form:

,(n —
y "w= /“
In each of the intervals (a,x) and ( b,x )the integrand and
respect to x are continuous; we therefore differentiate with respect to x under
the integral sign and with respect to the upper (or lower) limit x, and obtain:
x on- 1
iiOr r g "-2G(x^)
f
J a
- - A m
dx‘ + [ 8xn~
2 £ = x —0
Ax)

6 d»-
+ A m / w m)
dx‘n - 1 t =*+0
dn~ 2G
Since ----- - is continuous at x = £ ,the two integrated terms cancel out, and
oxn~ 2
there remains
(15)

6 a— G (x ,0
i.e. yi n - ‘>W=J (16)
.x -
By differentiating the formula (15) again we find, as above:
x M ix ,® > _1G (x,£) j
/A x ) = f
J a dxn A m + axn I =x-0
/(*)

W ( x ,0 _ r an~1G(x,f)~
+
x axn
Am [ ax"-1
Ax). (17)
f =x+0

From the condition 2° in the definition of Green’s function,

ra— G(x,f)l ra— G(x,0] 1
. ax"-1 L =*-o [ dxn~1 A£=x+Q Po(x)
Hence (17) can be written in the form
_ f 6 P G (x,0 l
y(n\x ) = f (18)
J a w A ()d (+ m
The forms Uv(y) contain merely the function values of ; (x) and its derivatives
up to the order (n —1) inclusive at the points x = and x = hence,
(12), (13), and (16),
Uv(y) = J* Uv( G ) f ( m = 0,
since, by definition of Green’s function, UV(G) = 0. So we have now shown
that the function >>(x) satisfies the boundary conditions (11).
 §3.5 g r e e n ’s f u n c t i o n for a d j o in t 33
We shall now show that it also satisfies the equation l(y) = f.
By substituting the expressions (12), (13), (16) and (18) for the function/(x)
and its derivatives into l(y), we find:

/w = /( o /( m + /o ).
But the integral in the last formula vanishes, because, by hypothesis, the
function G(x,£), regarded as a function of a*, satisfies the equation l{y) = 0 in
each of the intervals [a,£) and (£,£>]. This completes the proof of Theorem 2.
An operator A defined for all continuous functions f{x) by the equation
Af(x) = j ba k { x , m ) d t
is called an integral operator with the kernel k(x,$).
Theorem 2 means that the operator L ~ x is an integral operator with the
kernel G(x,£).

5. G reen ’s F u n ction for the A djoint O p erator

If L has an inverse, then, by virtue of Theorem VIII, §1.7, the adjoint operator
L* also has an inverse. Consequently L* likewise has a Green’s function, which
we shall denote by H{x,£).
Let us find the relation between the Green’s functions G(x,$) and H(x,$) of
L a n d L*.
Put L " 1/ = (p, and L*~xg = 4>, then / = L<p, g = L*</>, and the equation
(cp,L**p) = (L<p,<p) [see (20), §1.6] can be written in the form
(L~xf,g ) =
Hence
J‘ c (x ,o m ¥ x )d x d ( = f l jlm m x M x ) d x d (
for any continuous functionsf( x ) and g(A). This gives us:
G (x,0 = H & x). (19)
Kernels G and H which are related to one another by the formula (19) are
said to be adjoint to one another.
So we have:
III. The Green's functions of two mutually adjoint differential operators are
mutually adjoint kernels.
In particular, if the operator L is self-adjoint, so that L* — L, then
H (x ,0 = G(x&. The relation (19) then gives the condition
G (x,0 = G{ffxj. (20)
A kernel which satisfies condition (20) is called Hermitian. Hence:
IV. The Green's function of a self-adjoint differential operator is a Hermitian
kernel.
34 §3 g r e e n ’s f u n c t io n

6. B oundary-value P rob lem s involving a P ara m eter and their R ed u ction to an

In tegral E quation x

In many questions we have to investigate boundary-value problems of the form

Ky) = Av + /(* ), (21)

Uv(y)= -0, V = 1 , 2 , ( 2 2 )

where A is a certain parameter, and f(x) denotes a prescribed, continuous

function of x. Using the operator L generated by the differential expression
l(y) and the boundary conditions (22), we can write this boundary-value
problem in the form
L y = \y + f. (23)
For the special case/ = 0, this equation takes the form
Ly = Ay, (24)

and is therefore equivalent to the homogeneous boundary-value problem

l( y ) = \ y ; Uv(j) = 0, v = 1,2, ...,n. (25)

The values of Afor which equation (24) has a non-trivial solution are obviously
the eigenvalues of the operator L (or of the boundary-value problem (25)), and
each non-trivial solution itself is a related eigenfunction.
Suppose that L has an inverse L -1 ; let G(x,£) be the associated Green’s
function. By applying the operator L -1 to both sides of equation (23), we
obtain
y = *L~'y + g (26)
with g = L ~ xf . Equation (26) implies

y(x) = AJ* G(xf)y(£)d£ + g(x). (27)

Hence:
V. I f the differential expression l(y), subject to the boundary conditions
U fy) = 0,v = 1 , 2 has the Green's function G (xf), then the boundary-
value problem
Ky) = Av + f(x ), Uv(y) = 0
is equivalent to the integral equation

;■(->) = a J* + g(x),
where
g(x) = J‘ a (x,m & d (-
 §3.7 g r e e n ’s f u n c t i o n for L — A1 35
In particular, the homogeneous boundary-value problem

Hy) = *y, u„(y) = o, v = i,2,...,/?

w equivalent to the homogeneous integral equation

x*) = a J‘ G ( x , o x m - (28)

Since (j (;c,£) is a continuous kernel, the Fredholm theory (see, for example,
[86a]) can be applied to the related integral equation. Accordingly, the
homogeneous integral equation (28) has only denumerably many eigenvalues
Ai ,A2,A3,..., which have no finite point of accumulation. For all values of A
different from the eigenvalues, the inhomogeneous equation (27) with an
arbitrary continuous function g(x) on the right-hand side has a solution
expressed by the formula

}'(x) = Aj ' r(xf,X)g(0d£, (29)

where T(;c,£,A) denotes the resolvent of the kernel G(x,£) and is a meromorphic
function of A, for arbitrary fixed values of x and £ in the interval [a,6], which can
have poles only at the eigenvalues of the homogeneous integral equation. From
the integral equations we go over to the corresponding boundary-value prob­
lems to obtain the following results:
VI. I f the differential equation l{y) = 0, under the boundary conditions
Uv(y) = 0, v = 1,2,...,/?, has only the trivial solution, then:
(a) the homogeneous boundary-value problem

I ( y ) = * y : U fy) = 0, v = 1,2,...,/?,

has at most denumerably many eigenvalues A1,A2,A3,... with no finite point of

accumulation.
(b) the inhomogeneous boundary-value problem

l{y) = Ay + / : U fy) = 0 , v = 1,2,...,/?,

has a solution fo r all values of A different from the eigenvalues for the corres­
ponding homogeneous problem and fo r any continuous function fix).
We had already obtained proposition Via otherwise, without using integral
equations, in §2.1; and we could have obtained VIb by considerations similar
to those there used.

7. G r een ’s F u n ction for th e O p erator L — XI

As before, let L be an operator generated by the expression l(y) and the con­
ditions U fy) = 0, v = 1,2,...,/?. We want to find an expression for the Green’s
36 §3 g r e e n ’s f u n c t i o n

function of the operator L — A l ; in other words, we want to form the inverse

of the operator L — A l.
We denote by y v = y v(x,A), v = l,2,...,/7,\a system of solutions of the
equation l(y) = Ay which satisfy the initial conditions

We apply the method of variation of the parameters to the equation

l(y) - Ay = f .
and obtain

di (30 a)

and also

d£. (30 b)

Here C[,...,C'n,Ci,...,CI are constants and Wis the Wronskian determinant of

the functions y uy 2,■■■>y„-

Xr-1) •
yUt ~ 2) i;(n—2)
y {r 2) •** sn
W =

yi y% •• y n
and W1,W 2>...,W„ are the cofactors of the elements in the first row of W.
Adding (30 a) and (30 b), and dividing by 2, we get

X*) = E C jv W + J* (31)

where Cu ...,Cn are constants, and the function

y i(x) y?.(x) •• yn(x)

y(r 2\ i ) y(r 2Xt) X"“ 2)|
g (x,0 = ± (32)
2W(i)

yi(0 aXO •• y»(fi

the positive sign being taken if x > £, and the negative sign if x < £.
We next demand that the function y in (31) shall satisfy the boundary con-
 §3.8 ANALYTIC NATURE OF GREEN’S FUNCTION 37
ditions Uv(y) — 0,v — 1 , 2 , i.e. thatitshall be a solution of the boundary-
value problem
l(y) = Ay + / , Uv(y) = 0, „ = 1,2,...,#!.
This requires that

t CjUv(yj) + f/G r)t/V(s)<*r = 0.

i=i
By solving these equations for Cj and substitution in (31), we get

><x) = r G(x,{,\)f(odz (33)

with
G(x,£, A) ( - i )- H(x,$, A), (34)
A(A)
where
tfiO'i) - u m

U2(y i) U2(y2) ••• u f y n)

A(A) = (35)

Un(y i) Un(y2) U„(y„)

and
yi(x) yz(x) - yjix) g (x,0
Ux(yd U fy,) ••• U fy n) UM
H(x,£, A) = U2(yi) UM - U fy n) U~(g) (36)

U M Un{y2) U„(y„) u,(g) 1

If Ais not an eigenvalue of the operator L, then A(A) ^ 0, and the formulae (33)
and (34) are meaningful. Formula (33) shows that G(x,£,A) is the Green’s
function of the operator L — Al. Hence:
VII. The Green's function G(x,£, A) of the operator L — Al is determined by the
formulae (32), (34), (35) and (36).

8. The Analytic Nature of the Green’s Function of the Operator L — XI

The functions A(A) and H(x,g,A), as may be seen immediately from (35) and
(36), are integral, analytic functions of the parameter A. So (34) implies:
VIII. Green's function G(x,£,A) for the operator L — Al is a meromorphic
function of the parameter A, and its poles can only be the eigenvalues of the
operator.
38 §3 g r e e n ’s f u n c t i o n

Let A0 be a simple zero of the function A(A). Then A0can also only be a simple
pole of the function (7(x,f,A), so that
\

G(x,i, A) = — — + G^x&X), (37)

A A0

where (7i(x,£,A) is regular in a neighbourhood of the point A0.

If, then, A0 is not a singular point of the function (/(*,£,A), we have to put
R (x,0 = 0.
By a well-known formula in the theory of residues, we have

H{x,i,X0)
R (x,0 = (-1 )"
AW '

In the expansion of the determinant H(x,£,A) by elements of the first row, the
coefficient of g(x,f) is A(A0) and = 0; hence the function R(x,£) and its first n
derivatives with respect to x are continuous in the square a < x,£ < b.
Finally, it follows from (36) that R(x,£) is a linear combination of the functions
^x(x),...,y„(x); consequently, for a fixed £,R(x,£) satisfies the equation

l(R) - X0(R) = 0.

We also derive from formula (36) that H(x,£,A), and therefore R(x,£) also,
satisfy the boundary conditions.
Therefore, for every fixed £, R(x,£) is an eigenfunction of the operator L for
the eigenvalue A0. Since, however, A0 is a simple zero of the function A(A), there
is only one eigenfunction }>o(*) corresponding to it, up to a factor independent
of x , for the operator L. We must therefore have

R(x,£) = a(£)y0(x). (38)

We now put G*(x,£,A) = (/(£,.*,A); then G*(x:,|,A) is the Green’s function for
the operator L* — Al, provided that the coefficients of the expression l(y)
satisfy the differentiability requirements needed for I* to exist.
From equation (37), on the other hand, we infer that

A A0

So, for a fixed £,R(£,x) is an eigenfunction of the operator L*, corresponding to

the eigenvalue A0. If we denote one of these functions by z0(x), we then have

R (lx ) = b(£)z0(x).
Hence
R (x ,£ )= b (x ).z0(£).
 §3.8 ANALYTIC NATURE OF GREEN’S FUNCTION 39
Comparing this formula with (38), we find

R(x,£) = cy0{x)zfZ). (39)

It remains to determine the constant c.
From (37) and (39), we deduce that

(A — Ao) J* G(.x,£,X)y0(g)dg = cjq(a) J* y^)zj£)dZ;

+ (A -A 0) j ' G M % 0( f)C (40)

The integrand in the last term is a regular function of the parameter A in a

neighbourhood about the point A0; hence we find from (40) that

lim (A — A0) r G{x,{j,X)y0(Od$ = cy0{x) f* y0($)zjf)d£. (41)

JWA. Ja J0
On the other hand,
(L Al)_y0 — (A0A)_}’0,
whence
(L - Al)-Vo = — -^7o-

Since G(x,£,A) is the Green’s function for the operator L — Al, the last equation
can be written in the form

f G (x ,(S > y jm = j - J — *,(*).

Ja A

By substitution in (41) we obtain

■y0(x) = cy0(x) J* y0{Qz0{£)d{,
so that c is given by
= _______ 1
J* yo(QzJf)d£
and therefore
r (x,o = - .
j. y jjfr jm
It follows from (37) that:
IX. For every simple zero A0 of the function A(A),
yo(x)z0(j) + G'1(x,^,A), (42)
G(x,£, A) = -
(A - A0) J* y 0{0zjf)d£

where G f x ^ f ) is regular in a neighbourhood of the point A0.

40 §3 g r e e n ’s f u n c t io n

The formula becomes particularly simple if the functions aQd zo(x ) are
normalized so that
= i.
It then becomes
G(x,i,\) = - + Gl(x,£,A). (43)
A A0

9. The Case of a Multiple Pole of the Green’s Function

The above results may be generalized to the case of a multiple zero A0 of the
function A(A). In general, A0 is then also a multiple pole of the Green’s function
G(x,^,A). We shall again assume that G(x,^,A) is the Green’s function for the
boundary-value problem
Uv(y) = o, V = 1,2, (44)
where the coefficients of the differential expression l(y) and of the forms t/v(j)
are integral, analytic functions of the parameter A; the boundary-value problem
(L — Al)y = / is obviously a special case of this boundary-value problem. We
observe that all the results of sections 7 and 8 carry over completely to the
general case, >'i(x,A),>'2(x,A),... ,^n(x,A) now being understood to be a funda­
mental system of solutions of the equation l(y) = 0, satisfying the same initial
conditions as before, viz.
yk (<*,*) (o for k

Let, then, A0 now be a pole of the v-th order for the Green’s function G(jc,£,A)
for the boundary-value problem (44); then

Gfx,£,A) = R(x,Q Rv-i(x,Q

+ C/^Xj^A), (45)
(A - A0)v (A - K Y - 1 A — A0
G^ a-j^A) being a function regular in a neighbourhood of the point A0.
For a fixed £, the function G(x,£,X) satisfies the equation 1(G) = 0 in each of
the intervals [a,£) and (£,&]. By substituting for the function G its expression as
in (45), we obtain the relations, for A = A0:

m = o,

=o.
(46)

/<* - ,) + 1i i KR'-*>+ • + (7 = 1 )! KR) = a

 §3.9 g r e e n ’s f u n c t i o n w it h m u l t ip l e pole 41
Moreover, it follows from the formulae (34) and (36) that, with a fixed £ and
for A = A0, the functions i?(x,^),/?1(x-,^),...,i?v_1(x,|) satisfy the boundary
conditions
A 1 dP
I ~1W p U^ R p- p) = 0 , R0 = R , ^ = 0,1,...,(v - 1) ,q = 1 , 2 ( 4 7 )
p = Qp i CAF

By making use of the relations (46) and (47) as well as the definition of the
eigenfunctions and the associated functions (§2.3), we can satisfy ourselves of
the truth of the following theorem, which we merely state without proof (see
[45a]):
X. The principal part of the Green's function in a neighbourhood of the zero
A0 of the function A(A) has the form

y , zJF)yn(x) + zjiiQyA*) ,
A 1(A -V "i + (A - A 0)"J-1 ~
+ z f f i y j w - iO) + 2(x) + ••• + Zj>r i ( W | (48)
A — A0 I
where
yj>yji>•••>yj.mj - 1» j = 1 ,2 ,—,p
is an arbitrary, canonical system of eigenfunctions and associated functions of
the boundary-value problem concerned, while
j 1>2,...,/?
is an appropriately normalized, canonical system of eigenfunctions and
associated functions for the adjoint boundary-value problem.
 CHAPTER II

ASYMPTOTIC BEHAVIOUR OF THE

EIGENVALUES AND EIGENFUNCTIONS.
EXPANSION OF A PRESCRIBED FUNCTION
IN TERMS OF EIGENFUNCTIONS OF A
DIFFERENTIAL OPERATOR

§4. The Asymptotic Behaviour of the Eigenvalues and Eigenfunctions for

Large Values of |A|

1. S ta tem en t o f th e P rob lem

For large values of |A|, approximation formulae and, indeed, asymptotic

formulae can be given for the eigenvalues and eigenfunctions of a differential
operator. Such formulae are not only of interest in themselves, but they also
find application at a decisive point in the proof of certain theorems in the theory
of differential operators, particularly in expansion theorems.
It turns out that the behaviour of the eigenvalues Av and the eigenfunctions
for an arbitrary differential operator as |AV| increases can, to a first approxima­
tion, be characterized by the eigenvalues and eigenfunctions of the operator
which is generated by the same boundary conditions but the simplest possible
(Jny
differential expression of the nth order, l(y) = ~ .
We shall first investigate the asymptotic behaviour of solutions of l(y) = Ay
for large |A|, where l(y) is an arbitrary differential expression. Then by sub­
stituting the asymptotic approximation expressions so obtained into the
equation A = 0 (cf. §2.1), we shall obtain the corresponding asymptotic
estimates for the eigenvalues.
First we put A = —p"; the equation l(y) — Aj> then takes the form

Ky) + Pny = 0, (1)

or, in more detail,

dnv dn~
— + P iW + ••• + pn(x)y + pny = 0. (2)

[For simplicity, we have at first taken p 0(x) = 1; see §4.11(b) for the case
42
 §4.1 THE PROBLEM 43
Po(x) # 1]. Without loss of generality, we may assume that /^(x) = 0. For if
Pi(x) & 0, then by using the substitution

y = y -exp^- i J
we find that equation (2) takes the form

dnv dn~2v
^ + AW + ••• + P„(x)y + Pny = 0,

wherep fx ),...,p n(x) are likewise continuous with respect to .x in the interval
[a,b], and p has not changed.
Also without loss of generality we can replace the interval [a,b] by [0,1], for
we can return immediately to the general case by making the substitution

x = a + (b — a)t.

2. The Domains S and T

We divide the complex p-plane into 2n sectors S k, k = 0,1, ,(2n - 1),
defined by
kir {k + 1)tt
< arg p (3)
n n
(Fig. 7).
As we shall see later, the asymptotic formulae for the solution y of equa­
tion (2) depend essentially on which sector Sk the point p lies in.
We denote by

the different nth roots of —1 arranged in an order in each case to suit later
requirements.
Later we shall need to use the following essential properties of the sectors Sk.
I. For each of the sectors S k the numbers can be ordered in such a
way that, fo r all p e S k, the inequalities

&(p<»i) <^(/3Cl>2) < ... (4)

hold, where 0t{z) means the real part of z.

Proof. It is sufficient to prove the assertion for two adjacent sectors S k. For,
multiplication by any particular nth root of unity, say

transforms each sector S k into the sector S k- obtained from it by a rotation

44 §4 ASYMPTOTIC BEHAVIOUR

Fig. 7. Fig. 8.

through the angle — , 0 < k < In — 1, and we have k ' = k (mod2) ; and the
n
numbers

differ from the numbers (o1,w2,..., oj„ only in their order.

The inequalities (4) can now be written in the form
^ (p e.e -1^ ) < ^ (p e .e -1w2) ^ ... < ^ (p e .e -1 a>„),
and by what has just been said, they provide the corresponding inequalities
for the sector Sk>. Since 1is arbitrary, they give the corresponding inequalities
for all S k>with k ' — k = 0 (mod 2), provided only that they hold for the
sector S k. Hence it suffices to prove the assertion for two adjacent sectors, say
Sin—i ^nd So.
Suppose n is odd: n — Ip. — 1. Let, then, the order of the {a)*} be chosen so
that
.2-77 277 477
arg <d1 - 77, arg o>2 — it H-----, arg co3 = 77------ , arg o>4 = 77 ----- ,
n n n
4tt 77
arg o>5 = 77------ ,..., arg wn = -.
n n
If we multiply any number p e 52n_1 in turn by cu1,to2,...,con, then the points
p1,p2,--;Pn s0 obtained will be situated as shown in Fig. 8. For the domain
*^2n-i> therefore, the inequalities (4) are geometrically obvious.
Similarly, we can show the validity of the assertion I for the domain 50 and
also for the case when n is even.
Later we shall find it expedient to consider more general domains, namely
those which are obtained from the sectors S k by a translation p p — c,
 §4.3 REDUCTION TO INTEGRO-DIFFERENTIAL EQUATION 45
where c is a fixed complex number. These new sectors with their vertices at the
point p = —c will correspondingly be denoted by Tk, k — 0,1,2,...,2n — 1.
Taking account of the way in which the Tk are produced from the S k by
translation, we see that, for p e Tk>the inequalities

&((p + c)a>x) < ^2((p + c)a>2) < ... < 8$((p + c)a>„) (5)
hold, for a suitable ordering of the numbers couu>2,...,u>n.
In the sequel we shall let p vary in a fixed domain Tk and so we shall write
simply S and T instead of Sk and Tk. The order of the numbers a ^ a ^ ,...,^
will be such that for p e T the inequalities (5) are valid.

3. R ed u ction o f th e E quation l ( y ) + pny = 0 to an Integro-differential E quation

The derivation of the subsequent asymptotic approximation formulae depends

on the fact that the equation
liy) + Pny = 0- ( 6)
can be reduced to a certain equivalent integro-differential equation. To this
end we put
m{y) = - p 2/ n- 2> - ... —pny, (7)
and we can then write equation (6) in the form
y (n) + pny = m(y). (8)
The homogeneous, linear, differential equation y(n) -f pny = 0 has, for p ^ 0,
the fundamental system
eP<oix yep(0nX.

If we regard (8) as an associated inhomogeneous equation with the inhomo­

geneous term m{y), then the method of variation of the parameters gives the
general solution y in the form
+ ... + (o„e‘,P<On(X~Z)
;; = c1ep“lX + ... + cnepu>nX + -m Ay)dt (9)
npn- 1
where m^iy) denotes the value of m{y) for x =
The equations (8) and (9) are equivalent in the sense that any solution of (9)
with arbitrary constants cx,...,cn is also a solution of (8), while conversely for
each solution of (8) there exist constants cx,...,c„ with which (9) is satisfied. In
this, the constants cls...,cn may depend on p. We keep k fixed at one of the
values k = 1,2,...,n, and put
c'j = Cj fory = 1,2,...,/:
( 10)
c'.=c.+ i (OTj = ( k + l ) , . . .
a=k+ l J 0 ftp 4
46 §4 ASYMPTOTIC BEHAVIOUR

Equation (9) then takes the form

y = c[e/>0>lX + ... + c'nepmnX + —~ f K1(x,£,p)m£y)dg

np Jo

------—I f K fxf,p )m fy)d £ , (11)

np Jx
where
Ki(x,g,p)= Z “>xepac‘(x~i)> K2( x ,t P) = t (12)
a=1 a « fc+ 1

Equation (11) is the integro-differential equation referred to.

4. A L em m a concerning a S y stem o f In tegral E q u ation s

LEMMA 1

I f the system of integral equations

y ix ) =fi(x) + £ J* A ij(x ,i,\)y j(0 ^ , i = 1,2,.../, (13)

satisfies the following hypotheses:

1. the functions f f x ) are continuous in the internal [a,b\\
2. for any fixed £, the functions A {j(x,£, A) are continuous in a < x < £ and
£ < x < b;
3. for any fixed x and £ (a < x,£ < b), the A ifix J ,A) are regular analytic
functions of the parameter X, defined in a region of the complex X-plane
extending to infinity;
4. there exist constants R and C such that

fo r |A| > R and a < x,£ < b:

then, fo r a sufficiently large R0, there exists, for |A| > R0, one and only one
solution
y&x)=yfcx,X), i=\,...,r,
such that the functions ^,(x,A) are regular analytic functions of Afor |A| > R,
and such that
7 i(^ A) =/<(*) + O ^ for A oo.1 (14)

1 The expression 0 ( 1 /Ak) denotes, here and elsew here, a function o f the form /(*,A)/Ak,
where |/(*,A)| for a < x < b and for sufficiently large |A| alw ays rem ains less than a
constant.
 §4.4 SYSTEM OF INTEGRAL EQUATIONS 47
Proof. If the system (13) has a solution y f x f i = 1 , 2 t hen using the
method of successive approximations we obtain the equations

> ',(* )= /,•(* )+ z f

y—1 J a

j»jl~I*' J

= u x ) + i r A ij(x ,z ,\)fj{ m + . . .
j-1j

+ . Z . L ••• L A ij>(*>^1>A)■■■AJn_ -1,L,X)fjSU d£l■■■<!£„

Ji» • • >iJn * 1

+ . Z J0••• jn+,(£»>£.+!.%;„+,(£,+l)^l•••#„+1-
We put 5 = max |j,•(*)!> a < x < b , i = 1,2
Then, for |A| > R, the terms in the last row of this equation chain add up to
an amount not greater than

jXjir. [(» - ° ) C r r 'B ,

and therefore tend to zero as n -> oo, provided |A| > R 0, where

R 0 = max (f?,(6 — a)Cr}.

If this condition is satisfied, the solution is expressible in the form of an

infinite series, viz.

y fx ) = y fx , A)

= /,m + i J‘

+ i f‘ JI + ••• •
a j> = iJ J

Conversely, it can easily be seen that this series converges uniformly for every
case with |A| > R 0,a < x < b , and represents a solution of the system (13).
Hence follow all the assertions of the lemma, and in particular the formula (14)
and the regularity of the functions y fx . A) for |A| > R0.
48 § 4 ASYMPTOTIC BEHAVIOUR

5. Asymptotic Estimates for Solutions of the Equation l ( y ) + = 0

First we need estimates for the kernels X and K2 in (11) and for their
derivatives.

lem m a 2

There exists a constant C such that for all p e T the following inequalities hold:

-f-v *i(x,£,A) < C l p Y k l e ^ - ^ l for 0 < £ < x < 1, (15 a)

ox

^ - v K2(x,£,A) < c \p \\n - h)\er**-*>\ for 0 < * < £ < 1, (15 b)

V = 0,1,2,3,....

Proof. We choose a constant C such that, for all j,k = 1,...,n and all x,£ in the
interval [a,b],
|ec(^-<ok)(x-0| < C ; (16)

this is possible, since the left-hand side of (16) is a continuous function of the
variables x and £. It follows from the inequalities (5) that, for p e T and for
a. < k,
@(po>a) < + 0 + c)(a>k — a>J),
which, for 0 < £ < x < 1 , implies
^ |e[pco«+(p+c)(<ok-<Oo>)](*-?)| ^ £ |eP<ok(x-()^

Therefore (see (12))

k
dv
U x,£ ,p ) I < C k \p\v\ep(°k<-x - ()\
dx' a =1
which is the inequality (15a).
Inequality (15 b) may be proved similarly.
THEOREM 1

If the functions p2,... ,pn are continuous1 in the interval [0,1], then the equation

yin) + P iy {n~i) + ••• + p ny + pny = o

has, for each region T of the complex p-plane, n linearly independent solutions
y x,...,yn, which are regular for p s T and for sufficiently large |p|, and which.

1 In fact, as will be seen from the p ro o f given, the con clu sion o f the theorem rem ains true
if p 2, . . . ,pn are arbitrary functions summable in the interval [a,b].
 §4.5 ASYMPTOTIC ESTIMATES 49
with their derivatives, can be expressed in the form

= + O ^ Y |,

dx'
Proof. Suppose that the equation (6) has a solution y k for which
c' = 0 for v ^ k and c'k = 1.
Then

T* = + — I* Kl{x , l p)mi{yk)d! - - L - I ' U x , { , p)mi(y k)d t (18 a)

ftp J0 ftp J x

The results of differentiating this equation (n — 1) times, together with the

original equation, give the system of equations

d vy k = f a i l e d * + 1 r dvK1(x,i,p)
d xv nPV- 1 Jo dxv ”h(yk)d£

1 f 1 d 'K fx,{,P)
V—1 J* v = 0,1,2, ,(« — 1); (18 b)
V*
np ex'1
for, the result of differentiating with respect to the upper and lower limit x is
equal to zero when the sum of the two expressions is formed; this can be seen
by differentiating the integral in (9) with respect to the upper limit, and using
the known result + ... + aSn = 0 for v = 1,2,....
If in (18) we put
(19)

we obtain the system of equations for the functions z kv = z kv(x,P):

+ - + ■■■ «((•/>)] d{
p p J

+ - f L ( f ) z „ n_*({,,>)
np J x dx I

+ - p f O z k ,n - z { L p ) + ••• + ~ WZ 2 />n(£)**,o(&P)l & ( 20)

P P i
50 §4 ASYMPTOTIC BEHAVIOUR

If we now put
_ e-P<°k(x-$)p-v-*+\
n

_j_ i. e-P<Ok(x-S)p-v-«+z
n

k — l>2,...,/z, v 0,1,2,.. .,(/t 1), oz2,3,

then the system (20) can be reduced to the form

z kv(x,p) = + - Z f Kkva(x,tp )zka($,p)dt (21)

P a=2 J 0

For a fixed k, then equations obtained from (21) by puttingv = 0,l,...,(n — 1)

form a system of integral equations for the functions zkv(x.p), v = 0,1,...,«. It
follows from Lemma 2 that all the functions Kkvx(x,£,p) are bounded for
0 < x, f < l , p e T , and sufficiently large |p|. Hence all the assumptions in
Lemma 1 of section 4 are satisfied. By this lemma, the system (21) has one and
only one solution zkv = z kv(x,p), which is analytic in p and which satisfies

(22)

From this and from (19), the relations (17) follow immediately, and hence
also the linear independence of the functions .y^XjA), k = 1,2,...,«.
It remains to prove that a solution y*0>i:,A) of equation (6) exists, satisfying
equation (18 a); to do this, it is only necessary to show that for arbitrarily
chosen constants c'v (i.e. for numbers independent of p) there is a solution of
equation (6) which for these c' satisfies equation (11).
The equations (10) represent a linear transformation of the Cj into the cj (we
remark that y and therefore also the expression m ^y) on the right-hand side of
(10) depend linearly on c1,...,cn). Clearly it is sufficient to show that the deter­
minant of the transformation (10) is different from zero for sufficiently large
|p|, p e T\ we can then solve equations (10) with arbitrarily prescribed c'j for Cj.
The solution y of equation (6), or, what is the same thing, of equation (9) for
these values of Cj will then be the required solution.
If now the determinant of the transformation (10) vanishes for some p e T,
then the equations (10) with c[ = c% — ... = c'n = 0 for this p have a non­
trivial solution cj. The corresponding function y will consequently be a
nontrivial solution of the equation

y = — f Kiix,^p)m ^y)d^------— f K2(xJ,p)mt(y)df, (23)

np Jo np Jx

which is obtained from (11) on putting c’x = c'2 = ... = c' = 0.

 §4.5 ASYMPTOTIC ESTIMATES 51
We shall show that, for sufficiently large |/>|, this is impossible.
By differentiating (23) (n — 1) times and using the substitutions

0 ( f V » r ■, V = 0 ,1 ,...,( „ - 1),(24)

we obtain for the functions zv the system of equations

= (25)

“ ^ ri + - + P .d S M £ j> ) } d l+

_L_ + ... +

Let
m{p) = max |zv(x,/>)|, v = 0,1,...,(« - 1).

By using Lemma 2 to estimate the right-hand side of (25), we obtain the

inequality

+ ^ c (n - k ) £

Since the left-hand side attains its maximum m(p), it follows that

* » < ^ £ {w + - + ^

where Cx is a certain constant.

For large |/>| this inequality is possible only if m(p) = 0; hence certainly we
have zv(x,p) = 0. It follows from this immediately by (24), for v = 0, that
y = 0. This proves the theorem completely.
Remark. The equation
U(n) + PnU = 0

has, for p ^ 0, the fundamental system of solutions

Uk = ep<0kX, k = l,2,...,n.

From formula (17) we deduce that the solutions y k in Theorem 1 can be

approximated by these functions Uk for large values of |/>|.
52 §4 ASYMPTOTIC BEHAVIOUR

In other words, for large values of |/>|, only the first and last terms in the
equation
/ ° +/>2/ n- a) + ... ^ p „ y + pny = 0
play an essential part.
In the case of a second-order equation of the form y" + p(x)y — p2y = 0,
there are four of the domains S (fig. 9); by the theorem just proved there exist
in each of them linearly independent solutions y 1 and y 2 which can be
represented asymptotically in the form

, 1 = ^ [ i + o (i)\

For real and positive p it is often convenient to replace these solutions by the
linear combinations of them

& ± * -c o ,fr * ) + o(i).

^ = s in W + o (i).

The asymptotic behaviour of solutions of differential equations for large

values of the parameter has been investigated in many works. Additional
information and references can be found in the books [17], [97], [95b].

S! S0

S3

Fig. 9.
 §4.6 REFINED ASYMPTOTIC ESTIMATES 53

6. R efin em en t o f th e A sym p totic F orm ulae

If the coefficients p 2,...,pn of the differential expression are not only themselves
continuous but also have continuous derivatives up to a certain order in, then
the asymptotic formulae already obtained can be made more precise.
Suppose, for example, that the coefficient p 2(i) has a continuous first
derivative p'2({).
We substitute in the right-hand side of (18 a), instead of the functions and
their derivatives, the expressions from (17). Then using the estimates (15 a)
and (15 b) for v = 0, we obtain:

yk = - - \xK ^ Pv ^ p M n
np J o

(26)

But by (12)

j; u x ^ p v ^ p i m = \ ] e - ^ - ^ x-v p 2{ m

and by integrating by parts and using the inequalities (5) we can easily convince
ourselves that the first addend in the brackets is an .expression of the form
O . Hence

_ eP
= UakX
ep<kx [wk J % 2( 0 # + o ( - ) ,

and similarly

Substituting these expressions in (26), we get

(27)

where
(28)
54 §4 ASYMPTOTIC BEHAVIOUR

Similarly, by substituting the expressions from (17) into the right-hand side of
(18 b), we obtain:

^ = Pvw l ^ x ^l + + ° (-)j, v = 0,1..... (n - 1); (29)

in this case it can readily be checked that the first two terms in the brackets in
formula (29) result from the fact that we differentiate the expression

in (27) v times and, after taking out the factor pvwvkept0kX, discard all terms of
order o ( t ) .
If p 2 and p 3 have continuous derivatives up to the second and first order
respectively, then by substituting the expressions (29) in the right-hand sides
of (18 a) and (18 b), we derive in a similar way to the above

dy* = Pvcoiepmkx\ i + ^ (30)

dxv L p p2
Repetition of the argument leads to the following
Corollary. I f the functions Pz,Pz,-..,p„ have continuous derivatives up to the
orders m,(m — 1),(m — 2),... respectively in the interval [0,1], then the follow­
ing asymptotic formulae hold fo r the solutions yi,yz,...,yn constructed in
Theorem 1:

yk = e ^ l + + ... + + o ( — j^) , (31a)

^ = p x ^ ^ i + +... + + o ( — i) (31b)

k l,2,...,n, v = \,2 ,...,n 1,

where y kvi(x),--,ykVm(x) are continuous functions in the interval [0,1].

Remark 1. The formula (31 b) follow from (31 a) by termwise differentiation,
taking the factor pvavkepmkX outside the brackets, and collecting all those terms in
the brackets which contain factors 1jpk with k > m into an expression of the
form o ( — ).
We can easily satisfy ourselves that this is so, by considering that the term
in (31 a) is an expression of the form
 §4.7 NORMALIZED BOUNDARY CONDITIONS 55

where <p(£,p) denotes a certain bounded function; from the Lemma in section 5
it then follows that the vth derivative of this expression with respect to x
(v = l,2,...,(n — 1)) is of the form O
The functions y kvj{x) can therefore be expressed in terms of the functions
.W * )-
Remark 2. We can determine the functions y k^{x), up to additive constants,
even more simply by substituting the expressions (31 a) and (31 b) into the
equation l(y) + pny = 0, and, after removing a factor pnepa>kX, comparing
terms with the same powers of p, from 1/pto\/pminclusive. In this way we obtain
a system of recursive differential equations of the first order, from which we can
successively determine the functions y k0ft(x), up to additive constants.
In fact, after the substitutions mentioned, and the removal of the factor, we
obtain the result

^ + 4 r + . . . + — + o ( — ) = o, (32)
p p p \p /
where A Jpvdenotes the sum of all terms of the equation which contain l/pv but
not l/pv+1.
Obviously, equation (32) can hold only if

^ = 0, A s = 0,..., Am = 0. (33)
If we actually calculate the expressions for A k,Az,...,A m, we obtain from the
equations (33) the following expressions for the functions y k0v:

>>*oi = ai — nojk
— jf o (34 a)

y k = «, - i ’f - - 2,3,4,....m. (34b)
ft a—0 0=0 J0
The constants av (v = 1,2,3,...) could already be found from the equations
(18). Thus we saw earlier that = 0, for example. In the same way the
remaining constants av can be determined.7

7 . N o rm a liz a tio n o f th e B oundary C ond ition s

We wish to investigate the different systems Uv{y),v = 1,2,.. ,n, of linear forms
which define a given differential operator. If yik) or j4ft) appear explicitly in the
form U(y) but y ^ and / xv) do not, for any v > k, then we say that the form
56 §4 ASYMPTOTIC BEHAVIOUR

U(y) has order k .1 We consider the forms Uv(y) of order (n — 1), if there are
any. By replacing them, if necessary, by equivalent linear combinations, we can
arrange that the maximum number of forms of order (n — 1) is < 2. The
remaining forms have orders <(« — 2); we apply the same process to the
forms of order (n — 2) and reduce their number to a minimum; and so on.
The operations described are referred to as the normalization of the boundary
conditions, and the boundary conditions finally resulting are said to be
normalized. From the way in which they are constructed it follows that the
normalized boundary conditions must have the form

U fy) = Uv0(y) + l \ f y ) = 0, (35)

where
(36)
j =o
fcv-l
u vl(y) = iM *v) + I M * (37)
i=n

n — 1 > k x > k 2 > ... > kn > 0, k v+2 < K ,

and for each value of the suffix v at least one of the numbers av,/?v is non-zero.

8. R egular Boundary C onditions

We consider a fixed domain Sk; as before, we number so that, for

p e S k,
&(pwi) < ^?(po>2) < ... *C&(pcun).

For our later purposes it is convenient to particularize a class of boundary

conditions which we shall refer to as regular. This class is defined differently for
odd and even values of n, as follows:
(a) n is odd: n =2f* — 1.
The normalized boundary conditions (35) are said to be regular if the numbers
60 and 0k defined by the identity
kt
(a l + i

^2^1* a 2a ,^>- 1 (a2 + ^ 2 ^ +1

00 “1“ @1S “

« X " •• • « X B- i (“» +

are both different from zero.

1 correspond to y a,yi, o f §1.3.

 §4.8 REGULAR BOUNDARY CONDITIONS 57
(b) n is even: n = 2/x.
The normalized boundary conditions (35) are said to be regular if the
numbers 0_1 and 9k defined by the identity
0 -i
~ + 0Q + 0lS =

(<*i + ^ i ) ^ ‘ (<*! + - & ) k ' + . 2 •

•• “2 ^ - 1 ( a 2 + ^ 2) aj^ (<*2 + ^ 2) R ' + > +2 *

(3 9 )
1

,o>f" . ( a n + s /3 „ )a > * n
( a"+ X )“ j+ i £X "+2 " ■

are different from zero.

This definition of regularity is independent of the choice of the region S for
which the numbers a>k were arranged in order. In order to show this, we first go
back to section 2 and recall that there we were able to obtain the different
ordered-arrangements of the <x>k belonging to individual domains S k by
multiplying by an nth root of unity the values of the <ok in the order in which
they occurred for one or other of two basic, adjacent domains, say the domains
S0 and Szn-i. If now all the a>k in equation (38) or (39) are multiplied by the
same factor with modulus equal to 1, then the numbers 9 will change by a
common factor with modulus equal to 1. In particular, the ratios 90: 9k or
90 :9k : 9_v and so also the roots of the equations 9 ^ — 90 = 0 or 9a£2 +
+ 0 -i = 0 , remain invariant; we shall apply this result later.
So it now remains to find out how the numbers 9 change when we change
from an appropriate domain Sk with k even to a suitable chosen domain Sk-
with k ' odd. We shall, in fact, find that this change, too, can be achieved for
each separate number 9 by multiplication by a suitably chosen number with
modulus equal to 1 but that now, in general, all the numbers 9 will not be
multiplied by the same factor.
We discuss this first for the case when n is odd:
If n — 2p. — 1, the numbers <pk — e,n+2kiuin} k = 0 ,± 1 ,± 2 ,...,+ (/! — 1)
obviously represent all the distinct nth roots of —1. We can correlate any
particular order of the a>k to the tpk simply by specifying the row of suffixes. For
example, the row of suffixes in the proper order for the inequalities (4) to hold
in the domain runs:
(0, + 1, —1,+2, 2,...,(/* 1), (p 1)),
while that for the domain S0 is
( 0 ,-l,+ l,-2 ,+ 2 ,...,(- /x - !),(/* - 1)).
58 §4 ASYMPTOTIC BEHAVIOUR

Now, we have
aiO)
iwi ot*a)
l wn Atf*
^M+l A a>

«X " a„w£n M+ l _a>*"

AW rt
and the determinant for 01 has the same form except that in the ^th column the
oik are replaced by j8k.
Let us suppose that 60 and 0Xhave been formed for the domain and let
the corresponding numbers for the domain So be denoted by B0 and Bk. It is
clear that B0,Bx are obtained from 9{i,6l by interchanging the numbers o>k in the
second and third columns, in the fourth and fifth columns, etc. For n odd, this
reduces to an actual interchange of columns in the determinant for just one of
the numbers 0O or 6V This will become clear by considering the following
arrays:
(ax) iu, even, n = 4v — 1.
0O a a a ... a a a
(0, 1,-1,...,(- - l),-(v - lV,-v,...,(2v - l),-(2v - 1))
6± a oc a ... oc a j8... /?/J

(a2) p odd, p = 2v + 1, n = 4v + 1.
60 a oc oc ... oc oc jS /? ... jS
(0, 1, —l,...,v,—v,(v 4- 1),—(v 4- l),...,2v,—2v) .
6k a a a ... a j8 /? /J ... fi j3

These arrays merely show, for each column of the determinants for 60 and
0l5 the ^-suffix for the a>k occurring in that column, and the letters a. or /J above
and below this index shows whether one of the a j,...,^ or one of the
goes with it. For the transition from S2n - i t0 S0, the index row (0,1, —1,...) is
simply replaced by the row (0, —1,1,...) and the (a,/^-distribution remains
unaltered.
This can be accomplished in case (ax) for 0X, and in case (a2) for 0O, by a
simple interchange of columns, and it follows that
B1 = —0X for n = 4v — 1,

B0 — 0q for n : 4v 4- 1.

To calculate the value of the other 9 in each of these cases, we proceed thus:
 §4.8 REGULAR BOUNDARY CONDITIONS 59
A multiplication by e = e2nl^n transforms the system wl,...ywninto itself; only
the sequential order is permuted—and this simply by adding / to each number
in the index row.
From the suffix row:
.......................... a a jS
(0,1, — 1 ,...,0 — 1),—(v — l),v,—v,(v + 1),...,—(2v — 1))
the row:
....................................... a a j8 ...
( —1,0,—2,1,—3,...,(v - 2),-v,(v - l) ,- (v + l),v,...,(2v - 2),2v - 1)
can be obtained (with / = —1).
We see that this last row can be brought into the form
............................................. . a a /?
( 0 , - 1 , + 1 , - 2 , + 2 , . . . . —O' — 1),(*^ — \ ) , — v,v,— (v + !),(»/ + 1),...,

~{2v - 1),(2„ - 1))

by interchanging the members of its first (2v — 1) pairs.
Hence it follows that
0O = —e2ni{kl+ - +k’')ln 60 for n = 4v - I.
Similarly it can be shown that
8j_ = e~2*«k«+-+*-)/" ex for n = 4v + 1.
These results imply that, for odd n, the definition of regularity is independent
of the-choice of the zone Sk. Moreover, it has been shown that:
The root of the equation 6 — 0O= 0 remains invariant for the transition
from S k to S k>if k = fc'(mod 2), but for the transition from S2l+1 to S2l it is
multiplied by a factor e±2n*k,+ where the plus sign is to be taken if
tt = 4v -f- 1» and the minus sign if n = 4v — 1. If f (2) are the roots for Sk
with even and odd k, then ? 2) = e±2ni(k' +
For even n, n = 2/x, we can attain our goal rather more conveniently. The
numbers 4>k = ein+knt/n, k — ± l,± 3 ,d b 5 ,...,± (« — 1), now denote all the nth
roots of —1, and so we can again characterize any special order of the numbers
a>!,...,<o„, and consequently any particular domain Sk, by giving the corres­
ponding row of suffixes for its <pk. For the domains So, *Si respectively we
obtain in this way the suffix rows:
( 1 ,-1 , 3 ,- 3 , , (n — 1),—(« — 1))
(-1 , 1 ,-3 , 3, , - ( n - 1), (« - 1))
( - 1 ,- 3 , 1 ,-5 , 3 ,- 7 , 5, ...,—(« — 1), (« -3 ), (n - 1)).
60 §4 ASYMPTOTIC BEHAVIOUR

We again distinguish two cases:

Oh) ix = 2v, n == 4v.

The row of suffixes for Sk can be obtained from that for S0 by interchanging
the second and third columns, the fourth and fifth, ..., the (n — 2)th and
(n - l)th.
If the corresponding columns are interchanged in the determinant (39)
formed for the domain S0, and if s is replaced by 1/s, then clearly the determinant
(39) for the domain is obtained.
Since the number of column interchanges carried out is odd, we have
Dk(s) = —D0(\/s), where Dk(s) denotes the determinant (39) for the domains*.

(b2) fx = 2v -f- 1, n = 4v -f- 2.

In this case we obtain in exactly the same way

D&1-i(s) — D0 (1/s).

From the equations obtained in cases (bx) and (ba) it follows, first of all, that
the definition of regularity is independent of the special choice of the domain

We further notice that:

The roots of the quadratic equation Dk(s) = 0 for the regular boundary
conditions remain invariant when k is changed to k ' with (k' — k) even; but
they go over into their reciprocals l / £ ', l/£" if (k ' — k) is odd.
Finally we remark th at:
If the numbers a1,...,an,j81,...,j8n are all real, then the numbers 0 go over into
their complex conjugates for the transition from to S0. In this case, then,
the independence of the definition of regularity from choice of S k follows
immediately.
We shall now discuss a few examples of regular boundary conditions.

(a) Conditions of Sturm Type for even n (n — 2/x).

This is the name given to boundary conditions of the form:

(40)
U j . i y ) - / ^ + X f jvy ^ - 0,

with/j — 1 > k k > k 2 > ... > k^ > 0; n — 1 > k[ > k'2 > ... > k^ > 0.
 §4.8 REGULAR BOUNDARY CONDITIONS 61
Here half the boundary conditions contain only the function values of y and
its derivatives at the point a = 0, and the other half of them only the values at
the point a = 1.
In this case.
.. •. a>‘*
fi -1. “V* « i‘+i 0 ... 0

1 0 ... t 0
------ h @o + ®is — i
5
0 0 sa>*; a>lfi +2 ... *>?
s M+1

0 0 I ... CO?
J " +1

(The ± sign occurs because the rows are not in the same order as in formula
(39)). Consequently

0O= 0
QJ^1 " J'-l <»? <»?.> toK

0i = ± (41)
to. « i“. i " I 1 “ J"-2
A

0 -i = ± (42)
to. OJ„

Hence conditions of Sturm Type are regular provided that all the determinants
in the formulae (41) and (42) do not vanish.
If, for example,
k 2 — k-± 1, k$ k-± 2,..., k^ ki /2 “b 1,
k'2 = k[ — 1, k'z= k{ — 2,..., & '= & !—/* + 1,
these determinants are different from zero and so the conditions are regular.

(b) Conditions of Periodic Type.

This is the name given to conditions of the form :
Uv(y) = ^ov) — Tiv) = v= 0 ,1 ,2 — ,(« - - 1).
62 §4 ASYMPTOTIC BEHAVIOUR

Conditions of periodic type are regular.

For, if n is even (n = 2/a), then

1 ... 1 (1 s ) | i 1 ... 1

+ 0O+ 0i*y — i
a>i ... (i — (
H) k + i W/l + 2 ... a>„

n—1 . ,n -l
a, ; - 1 ... k ; l ^ + 2 ... <

= ± c ( i -* )(i--)>

where C is the Vandermonde determinant for the numbers a ^ ,...,^ , and

obviously does not vanish.
Hence
0O= ± 2C ; 0j = 0_i = ± C 5* 0j
i.e. the conditions are regular,
If h is odd (n = 2/u. — 1), then

1 .. l (1 - s ) 1 ... l
a>x (1 - ^ K ^ /l + l ... a>„
0q+ 0i*s — i ± c o - j).

,OJx,n-l , ,n-l ... a,:-1

.. « £ ii (1 “V+l
We deduce that in this case, too, the conditions are regular.
(c) Boundary Conditions for n = 2.
The most general boundary conditions for n = 2 have the form

aiTo + bxy[ + a^o + b0yx = 0,

ciTo + di y[ + coJo + d0y x = 0.
We distinguish the following cases:
1. axdx — b^Cx t- 0.
By solving (43) for y'0 and y[, we can bring the conditions to the form

J o “H a llJ o “b 0<12j l — 0 ,

J l "H a 2 lJ o "f" a 22 J l = O'

 §4.9 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES 63
Then
W, to2
— ■+ T- Qxs = 1 = — ■*;
s —ai2
s
e0= o, el = - 1 ,^ = 1,
1. e. the conditions are regular.
2. Q\dx b1c1 = 0, |ax| + |Z>x| > 0.
In this case we can transform the conditions (43) so:

a iTo + bjyi + a0y0 + b^yx = 0,

W o + d0yi = 0,

0*i + sb1)to1 + - b ^ w1
—- + “H ^i*y =
s
c0 + sd0 Co + -do
s

= <t>i(bxC0 + axd0) ^ + 2(a1,c0 + b id j)^;

the conditions are therefore regular if i xc0 + axd0 ^ 0.

3. ax = bx = cx = d-L = 0.
Since the formulae (43) must be independent, in this case a0d0 — b0c0 ^ 0;
hence the conditions (43) are equivalent to y 0 = 0, yx = 0. Hence

1 1
1
— - + ^0 + @1S — 1 — ■*>
s s - s
s
i.e. the conditions are regular.
Hence the conditions (43) fo r n = 2 are regular in just these cases:
1. axdx bxcx 5^ 0 j
2. aldl — bxci = 0, |ox| + |&x| > 0, V o + V o * 0;
3. = bi = Ci = di = 0, o0^/0 b0c0 ^ 0.

9. Asymptotic Behaviour of the Eigenvalues

The results of section 5 enable us to prove the existence of infinitely many
eigenvalues for differential operators with regular boundary conditions and
64 § 4 ASYMPTOTIC BEHAVIOUR

to give asymptotic approximation formulae for the behaviour of the sequence

of eigenvalues at infinity. It turns out that the principal terms in these approxi­
mations are already completely determined by the numbers 0O,0! or 0o,61,6_l,
which are calculated from the prescribed boundary conditions by means of
formulae (38) or (39). The further nature of the differential expression and
boundary conditions does not affect them at all.
We assume below (and for brevity we do not mention this again in the
hypothesis for the following theorem) that the coefficients of the differential
expression considered are continuous in the interval [0,1]. It should, however,
be expressly mentioned that all the results of this section would still hold if we
required only that these coefficients be summable in the interval [0,1].
theorem 2 .

A differential operator of the nth order which is generated by an expression of

the reducedform described in section 1 and by regular boundary conditions has
precisely denumerably many1 eigenvalues, whose behaviour at infinity is
specified by the following formulae:
For odd n such that n = Av — 1, the following two sequences are obtained: 2

A ; = ( - 2 W ) f - 2 J ^ 1,+ 0 (l)J , (4 4 a)

^ = (M '+W + 0 (f )}' (44b>

k = N ,N + 1,...;

and for odd n such that n = 4v + 1, the two sequences:

K = ( 2 f a m ,) " ( l + ' ^ L + ( 4 4 a ')

a: = ( - 2 « 4 - ^ + 0 ( 1 ) } . (44b-)

k = N ,N + l,...;

here | (l) and £(2) denote the two solutions defined earlier of the equation
0 ^ + 0o = 0 for S k with k odd and k even respectively; and N is a chosen,
sufficiently large, positive integer.

1 In view o f the am biguous usage o f the w ord “ denum erable” in English m athem atical
texts, w e have follow ed the au th or’s term inology.
2 Here ln0f is any fixed branch o f the natural logarithm .
 §4.9 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES 65
For even n, n = 2p, and 6$ — 4 6 ^ # ^ 0, the following sequences are
obtained:

K = ( l)"(2/crr)2,l<| 1 T ~ f + < ?(-)} , (45 a)

A;' = ( - l ) ^ ( 2 ^ f j l T (45 b)

where £' and £" are the roots of the equation

+ 0o£ + ^ -i = 0 (4 6 )

for the domain S0, i.e. are roots of the equation D ff) = 0, and where the
upper or lower sign is to be taken according as n = 4v or n = 4v + 2.
For even n, n = 2p, and 0~o — 4010_1 — 0, the following sequences are
obtained:

K = (-l)'(2fo7)2«(l T- (47 a)

A,' = ( - l ) « ( ^ ' j l T ^ + o L U ), (47 b)

k = N , N + 1,...,
where £ is the double root, occurring in this case, of the equation (46) (relative
to the 6 for the domain S0). The signs are to be chosen in the same way as for
the equations (45).
In the first three cases, all eigenvalues of sufficiently large modulus are simple;
but in the last case, all eigenvalues of sufficiently large modulus can be either
simple or double.
Proof. First let n be odd {n = 2p — 1). We consider a fixed domain T.
Let the numbers wk be numbered in the order such that, for p e T,
£%((p + c)o>1) < <%((p + c)a>^ < ... < &((p + c)a>„). (48)
We put
pk = (p + k = \, 2,...,n. (49)
The points Pi,p2,---,pn *ie on a circle of radius |p + c\ and divide the circum-
2^ 2,77
ference into n equal parts, each corresponding to an angle of — = -------
n 2pi — 1
In the closed, right half-plane there lie certainly at most p points, since more
than /x points (Fig. 10) would lead to the contradiction
2tt
7r > P > V.
2p — 1

Hence, and by the inequalities (48), it follows that at least the first (p — 1)
66 §4 ASYMPTOTIC BEHAVIOUR

points Pi,P2,---,Pf,-i must lie in the open, left half-plane. Similarly, the last
Gu — 1) points pfl+1,pll+2 >--->Pn must lie in the open, right half-plane. In other
words, we have:

&(Pi) < 0, ^(pjs) < 0 ,...,^ (p M_i) < 0, (50 a)

1) > 0, 5m+2) > 0,...M p n) > 0. (50 b)

If p oo and remains in the domain T, then the left-hand sides of (50 a) and
(50 b) tend to —co and + oo respectively.
will only then not tend to —oo if the angular distance between
and the positive or negative imaginary axis tends to zero. But this would
imply that, for sufficiently large |/>|, the (p -f 1) points lie on the
arc
7r ^ ^■7T ,
— 2 — e < ar§ P < 2 +

implying that, for arbitrarily small e > 0,

2irp
tt + 2e >
2p — 1
which is impossible.
From the assertion just proved it follows that, if p -»• oo, p e T, then
(a) eiij tends exponentially to zero if j < p
(b) e^J tends exponentially to infinity if j > p.
 §4.9 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES 67
By Theorem 1 in section 5 there exists in the domain T a system of n in­
dependent solutions of the equation l(y) + pny = 0 suchthat

With these solutions we form the determinant A(A) = det[t/vQ>y)]; as was

proved in §2.1, the eigenvalues are the zeros of A(A).
So we shall now concern ourselves with this determinant A(A).
For brevity, we introduce the notation [a] = a + O
By substituting the expressions (51) into the normalized forms Uv(y), we
obtain the equations

U.oO’j) = (pfj;)*')*, + <">(-) j = 0<u;)*’K].

tf.iOj) = + <?(£)) =

and hence

U¥(yj) = Uy0(yj) + Uvl(yj) = ( p o j j ) ^ ] + e ^ m } -

If j < /ii, the function

£pcaj __ e -cmjePj

decreases exponentially as p oo, p g T; hence

Uv(yj) = ( p w / l a j for j < /i. (52 a)

Similarly we find
Uv(yj) = (p c P jf^ J [ft] for j > fi. (52 b)
Finally, we have
Uv(y„) = ( p - / 1 K ] + ep^ m ) . (52 c)

We substitute all these expressions in the equation

A = det [Uy(yj)] = 0 (53)

and divide out the common factors p \ p \ . . . , p kn of the rows and also the
68 §4 ASYMPTOTIC BEHAVIOUR

common factors epca‘i+,,ep0,"+a,...,ep‘0n of the last p. columns of the determinant

D(A). The equation can then be written in the form
A0 = 0 , '
where

.. k K ‘- i [ f t K v i ••• k R 1
k k - • ( W i a K vi .. [&R**
■ (54)

[ « > :- ••• k K " - i ( W + [ A K " " K - ia K '. i •

From the definition of the numbers 0Oand 01(see (38) in section 8), we therefore
have
a0= m +
For a root p of the equation A0 = 0 with a sufficiently large absolute value, we
have
eP&n — _ R ]
P J’

.e. ep®*‘ — —
0o ~j- O
C)_- s i ' + »a
-ll'+ o f!)), <!!)
e1 + o { 1-
\p
since, by the regularity of the boundary conditions, 60 0 and 61 # 0. Hence

p — — |ln0£ “I” 2kiri + k —0 ,il> i2 ,— (56)

We now show that zeros of the function A(A) which are given by formula
(56) do actually exist.
We put
Pk = — (2tori+lnof); (57)

then equation (56) takes the form

P k — 0 ,± l ,i 2 ,.... (58)
~ p f t + 0W ’
The numbers pk obviously lie on a parallel to the bisector of the domain T,
so that for sufficiently large k there is only one choice possible for the sign of the
integer k if the point pk is to lie in T. Closer investigation shows that:
(a2) n = 4v — 1.
 §4.9 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES
« 69
For a T with even subscript, k must be chosen to be positive; and for a T
with odd subscript, k must be negative,
(ajj) n = 4v + 1.
For a T with even subscript, k must be chosen to be negative; and for a T
with odd subscript, k must be positive.
About each point pk we now describe a circle I \ with the same radius r in
each case. By what we have just said, for a sufficiently large |fc| these circles will
lie entirely within the domain T. The equation (53) is equivalent to equation
(55), and since £ = ePk“M, the latter can be written in the form

e<Mp-p><) — l _ = o. (59)

Outside the circles T k the modulus of the function

j = e <*f*(P~Pk) _ _ l — £<0,i(P~Po) i

is always greater than a positive constant.

For, if we introduce a new variable £ by putting

- Po) = Z,
then
f = £ — 1,
and the circles Tk transform into circles Tk with the same radius r about the
points £ = Ikrri. Since/ = f(£) is a periodic function with period 2ni, we have
only to show that its modulus is bounded from below by a positive constant in
the domain D bounded by the lines «/£ = ±77 and the circle Tq (Fig. 11). In
this domain the fu n ctio n /© never vanishes, however, and for sufficiently
large \@£\, say \&£\ > N, |/(£)| remains greater than a positive constant,
because
lim |/(0 | = co
S li-k + CO
lim |/ © | = 1.
- 00

Fig. 11.
70 §4 ASYMPTOTIC BEHAVIOUR

This proves our assertion. We conclude from it that, for sufficiently large p
the function A has no zeros outside the circles Tk.
Let m be the minimum of the function |<s"m(p“Pi<) — 1| on I \ . Since
p — p. = re'9 on I \ , m does not depend on k. For sufficiently large p,
\0 ( l/ P)\ < m on I \.
Hence, by Rouche’s theorem (see, e.g. [91]), it follows that, within I \ ,
the equation (59) has the same number of zeros as the equation
e<*,Ap-pk) __ 1= 0, i.e. just one zero, which we denote by p'k.
By (56),
^ = ~ - { u wi + l n o f + 0 ^ } ;

but by the same formula,

We have therefore
pk = — 12^77-/ -f- ln0£ j J,

If we apply the same arguments to a domain T, first with even subscript, then
with odd subscript, and raise the result to the oth power, then, taking account
of the proper sign of the number k, we obtain for each case the desired sequence
of eigenvalues (44).
That these eigenvalues are simple for sufficiently large k rests on the fact
shown above that they are simple zeros of the determinant A(A).
Now let n = 2ja, i.e. n is even. We again consider a fixed domain T for which
the inequalities (48) hold. By the same considerations as were used for n odd,
we conclude that
m{pk) < 0, $ (p 2) < 0 < 0, (61)
^0W > 0, ^ % +3) > 0,...,<%„) > 0, (62)
and the left-hand sides of (61) and (62) tend exponentially to —oo and +oo
respectively, as p -*■ oo while remaining in the prescribed domain T.
Hence, we conclude, exactly as in the case of n odd, that in the domain T
V jy j) = (?«’/)*•[“,] for _/ < / i —1,1
(63)
U,(yj) = for j > ii + 2,j
and moreover
W = W kv{ k ] + epco^ v]}, (64 a)
t f v ( W = 0>%+i)k1 k ] + (64 b)
 §4.9 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES 71
When n = 2n, if u>k is a root of the equation wn + 1 = 0, then so is —a>k.
Hence, and by the inequalities (48), it follows that

<0, ^ ( p M+1) > 0 ,

and

Hence (64 b) can be re-written as

1) = 0 > < W v( k ] + (64 b')

By substituting the expressions (63), (64) for Uv(j >j) into the equation A = 0
and removing the common factors, we obtain an equation of the form
(65)
II
oo
where A0 = o
• [«i"-*MJ “>;'{[«■] + e ~ 'lA S o>{V i{[«J+«-'"lA ]} [/V jV J- ••[ft™!!']
• [«**{•-J ..[ M S
• • ♦
«•:“{ [« .)+ e ' - M - h. t w+ m ;*j ••• [ M " l
(66)
By the definition of the numbers 0oA>0-i (see (39), section 8),

A0 = [0O] +
and therefore
ep*vAo = + [d0]ep^ + [#_ j = + e .k + (67)

because, by virtue of the relation ^ ( p M

) < 0, we clearly have
|ePo>„| =

i.e. the function ep<0» is bounded in the domain T.

We consider the quadratic equation

0i£2 +0o£ + 0 - i = O ; (68)

let i" be its roots, so that

«rf, + 0o£ + 0 -i = *i( t - n c t - n

Then (67) can be rewritten as
- f ' ) (e '” ' - (') + o(jj. (69)
72 §4 ASYMPTOTIC BEHAVIOUR

The equations
ep<0 » - f ' = 0 and eT* - f ' = 0
have respectively the roots v

—(ln0 £' + 2kiri), — 0no£" + 2km), k = 0 ,± l,:t2 ,...,

but we are concerned only with those which lie within the domain T.

Fig. 12.

Now the roots of both these types obviously lie on a parallel to one or other
of the boundary lines of the domain S. Consider the domain S0, for example.
Then we have
= —ie~tK,n for n — 4v,
a»M= i for n = 4v + 2,

and the above roots are given by

ieinln ln0 £' — 2kir, iein/n ln0 f" - 2kn for n = 4v,
and by
—i ln0 r + 2kn, —i ln0 f " + 2k-rr for n = 4v + 2.
Hence it follows that, for n = 4v all roots with k < 0, and for n = 4v + 2
all roots with k > 0, are situated inside T0 and at a positive distance from its
boundaries, if T0 is suitably chosen, i.e. if c is suitably chosen.
In the first of these cases, we replace k by —k and so arrive at the two
sequences
Pk = — Ono =F 2km), Pl = ~ (ln0 ? ± 2km), (70)
% “V
which, for k = 1,2,... lie inside T0, provided that the upper sign is chosen for
 § 4 .9 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES 73

n = 4v and the lower sign if « = 4v -(- 2. Using p', p" and (69), the equation
A0 = 0 can be put into the form
_ i] [e^ ( P - p l ) _ i] + = 0. (7 1 )

About each of the points pi, pi, k = 1,2,..., we describe a circle r* or r£ with
the same radius r. If we choose r to be sufficiently small, all these circles will lie
entirely within T0. If we apply Rouche’s theorem again, as we did for n odd,
then it follows that for sufficiently large \p\, p e T0, the equation A0 = 0 can
have zeros only within the circles r* and r*, and that it has the same number of
zeros, in fact, as the equation
_ i] [g<M p-pJ) - i] = o (7 2 )
has.
Now suppose that — 46k d_k ^ 0 ; then £' ^ £". Consequently the
numbers p'k and pi are pairwise distinct. Hence, by choice of a sufficiently small
common radius r it can be ensured that the circles T* and T* do not touch or
intersect each other.
In each of these circles the equation (72), and therefore also the equation
A0 = 0 for sufficiently large k , have exactly one zero, which we denote by
Pk or Pfc.
In the circle T* the modulus of the factor ea,l{p~p") — 1 remains greater than
a certain positive constant; hence by (71) it follows that in the circle 1^ the
equation A = 0 is equivalent to the equation

e®M<p-p;> _ i = O ^ -j.

Hence

A + °(^)-
and similarly
A = K + o (^ )-
As for n odd, this implies that
2kvi lnor
Pl = zF 1 =F + O >
2ktti k
and similarly for pi.
On raising this to the «th power, we obtain the formulae (45).
It is important to notice here that consideration of a domain Tk with k odd
would yield no new eigenvalues. For if we consider, for n = 4v, instead of the
domain T0 a suitable domain, for example, Tlt then f and £' go over into 1 /f
and l/r> and so the logarithms ln0f and ln0f ' become —ln0£' and —ln0f '
respectively, apart from a possible addend Ik'-rri. On the other hand, is also
74 §4 ASYMPTOTIC BEHAVIOUR

to be replaced by —wM, as may easily be seen from the considerations of

section 8. Hence, in order that the numbers — (ln0 £' + 2/cm) shall lie in Tlt
v%
it is now necessary that k be positive, and this implies that in formula (70) for
n — 4v the plus sign now stands.
Hence the pk and also the p"k remain unchanged. Since the equation (71) also
remains unchanged (up to multiplication by a constant factor), the samep* and
p'k result. The same thing holds for n = Av + 2 if we go over from the domain
T0 to T2n_v Hence, for large |A|, the eigenvalues (45) are the only eigenvalues
of the operator L.
Now suppose that 0\ — 4010_1 = 0. Then we have £' = and so p'k = p'k,
and the circles T* and F'k coincide for each k. The equation A0 = 0 therefore
has, for sufficiently large |/>|, precisely two zeros in each of these circles
Tk = T* = T'k, though these may possibly coincide to form a double root.
Let pk be either of these zeros. The equation (71) then takes the form

It follows that

and hence

The last part of the argument as in the case 9%—■49x9k ^ 0 enables us to

derive the formulae (47).
Remark. If the coefficients P z { x ) , . . . , p n( x ) have continuous derivatives up to a
definite order, then more precise asymptotic formulae can be derived, involving
higher powers of 1/p (see, for example, [54] and [110]).

10. Asymptotic Behaviour of the Eigenfunctions

We shall now use the results of sections 5 and 9 to elucidate the asymptotic
behaviour of the eigenfunctions for eigenvalues with very large moduli.
We again assume that L is a differential operator in the interval [0,1] and is
generated by a differential expression with continuous coefficients and by
regular boundary conditions.
Let yi,y2,---,yn be linearly independent solutions of the equation
Ky) + Pny = 0 which satisfy the relations (17) in a certain domain T (cf.
Theorem 1, section 5). An eigenfunction which belongs to a prescribed eigen­
value A = —pn withp e T must be expressible as a linear combination of the
functions yi,y2,. -.,yn’.
y = ciLi + c2y 2 + ... + c„y„.
 §4.10 ASYMPTOTIC BEHAVIOUR OF EIGENFUNCTIONS 75
where the coefficients Cj are non-trivial solutions of the system of homogeneous
equations
Uv(yi)ci + Uv(y^)c2 + ... + Uv(yn)cn = 0 , v = \,2,...,n.

For simplicity, we consider only a simple eigenvalue A, for which the rank of
the determinant A = det [i7v(yk)], v,k = is equal to (n — 1).
Then
yi r2 y«
U £y i) U2(y 2) U M

Un(y\) U M V M
is an eigenfunction belonging to the eigenvalue Aif we assume that not all the
minors of the elements of the first row of the determinant A vanish. (If the
latter were the case, we should have to arrange the y x,y2,...,y„ in a row, of
which not all the minors vanish).
We examine the cases for n even and n odd separately.

(a) Let n be odd (n — 2/x — 1)

In (73) we substitute for yi and Uy{y}) their representations (17) and (52) and
divide through by the inessential factors pk\p k\...,p kn,epa>l*+ ep0>n. Then an
eigenfunction y 0 is obtained, perhaps not normalized, for which the following
formula holds:
epa‘x[l] . . ep(0»-'x[\] +
ep(0»x[l] eP<on{x- 1)JJJ
ePC0M 1(x-l)|-jj

(M + R ’) [&R\i •• [&K‘
• k R ”-i (K ] + ^*[&R") [A.R+1 • [ftK-
(74)
Now for sufficiently large |A| the number p must correspond to one of the
Pk or Pk lying in T\ we therefore obtain for a domain T with an odd subscript

P = p'k = -L | t 2 kiri + ln0 f(1) + o { ^ j ,

and for a domain T with even subscript

± 2 kni + ln0 + o (-)

CO
for a suitable positive k ; and in these formulae the upper or lower sign is to be
76 §4 ASYMPTOTIC BEHAVIOUR

taken according as n = 4v — 1 or n = 4v + 1. We are therefore led to define

rf> = -1- + In, p ? = - (± 2 fori + In,

COH (Otl
M
in each case; and we then have

= + °(1 )> pi = p 'P + o i ) \ .

Then obviously
£ po>nX _ £ pka>Mx _ e "»iPk ^ [1 ] ^

since

<,oG) = 1 + °(r) + (°(r)) + - - 1 + °(1) = [1].

In particular,
_ e±2ftni +ln,^|j _

By substituting these expressions in (74) we obtain the following representa­

tion for each of the eigenfunctions ^ J), y£2) belonging to the eigenvalues
K,K °f formula (44)

e"^)jc[l] . e<0„pMx[ JJ + ...

KKa • W^-i te + tfK 1 t&Kvi - [&R2

kK" ■■■KK-1■1 k + SHnS^’p

f&K" [^K"+
L“ n ~ lHniw p + i
l - [AK-
••• L^nJ^n
(75)
k = N , ( N + 1),..., a = l,2.
If we restrict ourselves to a closed sub-interval a ' < x < x" inside the
interval [0,1], formula (75) can be simplified considerably. For, if we take
outside the determinant, then the /nth element of the first row is [1], and
the other elements of this row become:
for v < fjL and for p > ^
and these expressions tend exponentially to zero when k -> oo.
Hence
a2wti‘ 1 1 fa * ..

j4ff)= ( —i f * . (76)
«XM fa ^ l •• M n
It must be noted here that, in general, the numerical order of the numbers
£t>i»...>a,n wfil t>e different for the sequences y (P and y f \
 §4.10 ASYMPTOTIC BEHAVIOUR OF EIGENFUNCTIONS 77
(b) Let n be even (n = 2p).
We repeat the above argument and obtain two sequences of eigenfunctions
corresponding to the eigenvalues X' and A". To the eigenvalue A£ corresponds
the eigenfunction

=
eu>'p'«x[l) .

KM* .. • K K ‘-1 [a2 + r/3,]cp^ |a 2 + 1 /32j a ^ +1 [ ^ K v2 ' ...[]82K ‘

K ]^ i" ••• [a„]^n- i K + j^an + 1 /3„ja>*"+1 [/3„]a^

(77)
= 1,2,3,... .

For any closed sub-interval of the interval [0,1] this formula simplifies to
y^ =
/
a2wl ••• “2 ^ -1 ( tt2 + ~ Pt'j <•>£*+ 1 / V £ a+ 2 • • •

+ % .
••• “X " - l + ^ n) wp +l ^ +2 ••• fa n "

«2a'l2 a2a'#j2- 1 (a2 "F £ h>2)C°^ PiWp*+2 ••• 2^n

+ 0 7 (78)
’( 1 )
« X " ••• <vV -i (<*„ + I '& K " M n+2 &

The formula for the eigenfunction y ki corresponding to the eigenvalue X'k is

obtained by replacing pk and £' by pk and £".
For n = 2, the formulae are particularly simple. If, for example, the domain
Tis the first quadrant of thep-plane, then we have co„ = i and wfl+1 = —i, and
formula (77) then has the form

y kl = ( - i ) k‘eip'kX « 2 + - ^ 2 + C>(- (O^-'Pi^aa+^jSj+ 0

The results presented in this section are derived essentially from Birkhoff’s
work [5]; the method of proving Theorem 1 stems from Stone [109a].
78 §4 ASYMPTOTIC BEHAVIOUR

Theorem 1 has been proved by Birkhoff under more general conditions; viz.
the coefficients p k, k = 2,3,...,« can be analytic functions of the parameter p
which are representable in the neighbourhood of infinity by
.00

Pk(x,p) = £ akv(x)p~v
v=0
(and this representation may even be taken to hold in the sense of asymptotic
convergence). The domains 5 and T also become more general, and instead of
formula (17) the following formula is obtained:

yk = ["*(*) +

where co^x),...,^.*) are the roots of the equation

o>n + tfn_lio(x)cy"-1 + ••• + a0,o(*) = 0,
and uk(x) are certain functions of x which are defined by formal substitution
for y k in the differential equation.

11. Various Generalizations of the Asymptotic Estimates

(a) Arbitrary Interval.
An operator L defined in an arbitrary interval [a,b],a < b, can always be
brought, by the transformation x = a + t(b — a), into a form for which the
above asymptotic formulae are applicable, provided only that the differential
expression for L satisfies the conditions required in section 1. Correspondingly,
in the formulae (17) and (75) to (78) x is to be replaced by (x — a)/(b — a),
and in the formulae (44) to (47) for the eigenvalues the number tt must be
replaced by 77/(6 — a).
(b) . The Boundary-value Problem L{y) — Xpy.
Let L be an arbitrary differential operator for the interval [a,b] and in whose
differential expression l(y) the coefficientp 0(x) is = 1. Suppose that the function
p(x) is real and continuous and has the same sign throughout [a,b]; without loss
of generality, we may takep(x) > 0, for if this is not so initially, we have only
to replace p and Aby —p and —A. We introduce a new independent variable t
by the substitution

' = 1 £ 7 ^ (0 (79)
where
b ____
y/p(0di. (80)
a
AVith the variable t, the boundary-value problem L(y) = Xpy can be written
in the form Lxy = Aj, where Lx is the operator which is obtained from L by the
 §5.1 FOURIER METHOD 79
change of variable to t and division by p(x). For its differential expression the
coefficient of the highest derivative is again equal to 1.
Then by a change of the dependent variable to y, where

we obtain finally from L x an operator L2 for which all the assumptions of

Theorem 2 are satisfied and accordingly the previous asymptotic estimates
(17), (44)-(47), (75)-(78) may legitimately be applied, provided that the con­
dition for regularity of the boundary-conditions, and the incidental assumption
that the eigenvalues are simple, are satisfied. If we now transform these
formulae back to the old variables, we obtain the corresponding formulae for
the boundary-value problem L{y) = Apy. In particular, the formulae for the
eigenvalues are obtained from (44) to (47) by replacing ir by d*}-
Obviously we have also arrived at asymptotic approximations for any
eigenvalue problem L(y) = Ay of the type considered in Chapter 1, for which
the condition p0(x) = 1 is unnecessary.
The important assumption, of course, here is that p0(x) is real in [a,b\, does
not vanish, and does not change sign.
Considerable complications arise in the case where the function p(x)
vanishes at some points of the interval [a,b], or, what comes to the same thing,
if the coefficient p0(x) in an eigenvalue problem L{y) = Ay vanishes at certain
points of (a,b). This case and some of its various generalizations were the
subject of investigations by R. Langer and other authors (see the paper by
R. Langer [57e], which deals with the case n = 2).
The most general results in this direction have been obtained by V. S.
Pugachev [92].
(c). J. D. Tamarkin [110] obtains asymptotic expressions for the eigenvalues
of the generalized boundary-value problem
/ n) + />i(jc,A)j>(""1) + ... +/>„(*,A)j = 0; Uv(y) = 0 , v = l,2,...,n,
where p k(x,\) is a polynomial of the £th degree in A, and the coefficients of the
forms Uv(y) are polynomials of the nth degree in A.
Even more general results (for differential operators with partial derivatives)
have been obtained by M. V. Keldysh [47].

§5. Expansion in Terms of Eigenfunctions

1. The Basis of the Fourier Method

The solution of partial differential equations by Fourier’s method leads to a
most important problem: the expansion of a given function in terms of the
eigenfunctions of differential operators.
80 §5 EXPANSION IN EIGENFUNCTIONS

Suppose, for example, that a solution of the equation

e2u dnu , , . dn~'u , .Xt v „ . v . .
--- = P-\{x) “—^ “h Pn\X)M> Q ^ x ^ b) (1)
dt 2 ex" ™ exn-1 jPnW
is to be found which shall satisfy the initial conditions

[4=o = /(* ); ^ J (_o = <p(x); ( 2)

and the boundary conditions

(3)
I “- ( S t y p ' - i & L r 0- j = h 2 ....-•
We shall seek a solution of equation (1) which satisfies the boundary con­
ditions (3) and has the form
u = j(x) (A cos pt + B sin pt). (4)
By substitution in (1) and (3), we find that the function >»(*) must satisfy the
differential equation

Ky) = ^ + Pi(x) — y + ••• + p„(x)y = - p 2y (5)

and the boundary conditions

Uj(y) = " f + "l = 0. (6)

v= 0 v=0

If now y ^ 0, then y is an eigenfunction of the boundary-value problem (5),

(6) for the eigenvalue —p 2. Let
- p l - p l - p l - ■■
be all the eigenvalues of this problem and
Pi(x), y*(x), y3(x \...
the corresponding eigenfunctions, where each eigenvalue is repeated as many
times as it has linearly independent eigenfunctions belonging to it. Then the
infinite series
00
w = Z y»(x) (An cos p„t + B„ sin p„t)
u=1

satisfies, at least formally, the equation (1) and the boundary conditions (3).
It still has to satisfy the initial conditions. Substitution into the first of the
initial conditions yields
f(x) = Z An}’n(x). (7)
n =1
 §5.2 SELF-ADJOINT OPERATOR 81
This equation represents a series expansion of the prescribed function fix ) in
terms of eigenfunctions of the boundary-value problem.
The question as to the precise basis of the Fourier method thus leads directly
to the following problem: under what conditions may a prescribedfunction fix )
be expanded as a series of eigenfunctions of a proposed boundary-value problem ?
The solution of this problem is easiest in the case of a self-adjoint boundary-
value problem, i.e. if the expression l(y) and the boundary conditions
Uj(y) = 0> j = 1 > 2 , . . . , « , generate a self-adjoint differential operator.

2. The Case of a Self-Adjoint Operator

Let L be a self-adjoint operator which is developed from a differential ex­

pression l(y) and the boundary conditions Uj(y) = 0, j = 1,2,...,n. Without
loss of generality we can assume that Ly = 0 only for y = 0, i.e. that the
boundary-value problem

/C f)= 0 , U fy) = 0, j = 1 , 2 , . ( 8 )

has only the trivial solution y = 0. For, otherwise, we have only to replace
l(y) by the expression l(y) — cy, where c is a real number different from all the
eigenvalues of the operator L. And we know that such a number exists because
a self-adjoint operator can have only denumerably many eigenvalues.
If the boundary-value problem (8) has only the trivial solution, then the
operator L has a Green’s function (§3, sections 1-6), and this is a Hermitian
kernel.
We consider an arbitrary function f(x ) in the domain of definition of the
operator L : this means that the function fix ) has continuous derivatives up to
the nth order and that it satisfies the boundary conditions (6). If we put

L f = h,
then
fix ) = J* G(xf)h(f)di,

i.e. the function f(x ) is “sourcewise” representable1 by the continuous kernel

G(x,0-
By reason of the Hilbert-Schmidt expansion theorem in the theory of
integral equations (see [86a], for example), the function f(x) can be expanded
in a uniformly convergent series of eigenfunctions of the kernel G(x,£). But
we have seen that the kernel G(x,£) and the operator L have one and the same
set of eigenfunctions. This proves the following theorem.

1 T h is term in ology is com m on ly used by R ussian authors where a function is expressed

u sing a G reen ’s fun ction as kernel.
82 §5 EXPANSION IN EIGENFUNCTIONS

THEOREM 1

Any function in the domain of definition of a self-adjoint differential operator

can be expanded in a uniformly convergent, generalized Fourier series in terms
of eigenfunctions of this operator.
We remind ourselves that the domain of definition of a differential operator
of the «th order consists of all functions which have continuous derivatives up
to the «th order inclusive and which satisfy the boundary conditions belonging
to the operator. The above assertion about expansion in eigenfunctions
therefore applies to all functions having these properties.
Since the domain of definition of the operator L is dense in L 2(a,b), we
deduce from Theorem 1 the immediate
COROLLARY

The eigenfunctions of a self-adjoint differential operator form a complete

system in L\a,b).
If the eigenfunctions .^(x), are chosen so that they form an ortho­
normal system, then the coefficients A„ in (7) are determined by the formulae

A„ = (f,y „) = j l f(x)yjx)d x, n = 1,2,3,....

These coefficients A„ satisfy ParsevaVs relation

r \f(.x)\2dx = f \A„\2.
J 71=1

Theorem 1 remains true in the case of the general boundary-value problem

l(y) - Ap(x)y = 0, U fy) = 0, v = 1,2

where l(y) is self-adjoint and the boundary conditions are self-adjoint, pro­
vided that the function p(x) is continuous and positive in the interval [a,b].
For, in this case, the eigenvalue problem L(y) = Apy is equivalent to

P(x) = A G(x,Op(£)y($)d£,

and the integral operator occurring here has the kernel G fx ,i) = G(x,£)p($)
and is a Hermitian operator in the space & of the functions/(x) with the scalar
product
(fvfz) = J* /i (x)fJx)P(x)dx, (9)

i.e. G f x f ) is a so-called symmetrizable kernel.

So the Fourier coefficients A„ are now given by the formulae

An = (f , y n) = J 0/ ( X)>,7iWp(^yA,
 §5.3 REGULAR BOUNDARY CONDITIONS 83
where the form a complete system of eigenfunctions of the
operator —L, orthonormal in the sense of the scalar product (9).
Kamke [45] weakened the condition p{x) > 0, replacing it by p(x) Owith
another restriction. He also generalized Theorem 1 to the case of a self-adjoint
boundary-value problem

kiy) + A/20 ) = 0 for any real A, Uv(y) = 0, v = 1,2

under certain additional assumptions relative to the differential expression

l2(y) and the boundary conditions.

3. Expansion in terms of Functions of a Differential Operator with Regular

Boundary Conditions

The considerations in §5.2 do not attain the objective if the operator L

is not self-adjoint. So we shall now adopt another method which depends on
the analytic properties of the operator L — A1 and on the asymptotic formulae
obtained in §4. We shall assume that the differential expression from which L
is generated has the special form described in §4.1 and that the boundary con­
ditions are regular in the sense of the definition laid down in §4; we shall
further assume1, as in §5.2, that A = 0 is not an eigenvalue of the operator,
and therefore L has a Green’s function G(x,f).
In the complex A-plane we consider a sequence of circles Tk (k = 1,2,...)
with the origin as their common centre, and having the following properties:
1°. The radius R k of the circle T* increases without limit as k -> oo.
2°. There is a positive number S such that all the eigenvalues of the operator lie
at a distance from each of the circles T*.
By virtue of the asymptotic properties of the eigenvalues proved in §4.9 such
circles T* do exist. Let G(x,j,A) be the Green’s function of the operator L — Al,
and in particular, let G(x,j,0) = G(x,s) be the Green’s function for the
operator L.
We apply the theorem on residues to the integral

r 1 X G(x,s,X)dX
Ik ~~ 2ni J r k A
and obtain
It = 6 H + E 00)
V= 1

1 It fo llo w s from the argum ent at the beginning o f §5.2, and from the asym ptotic
form ulae for the eigen values that this assum ption does n ot restrict the generality.
84 §5 EXPANSION IN EIGENFUNCTIONS

where H fx,s) is the residue of the function G(x,s,A) at the pole Av and mk is the
number of such poles in the circle I \. \
We shall prove that
lim Tfc = 0 (11)
h—>oo

uniformly with respect to x and s in the interval [a,b]. The equation (10) will
then lead to the series expansion

G(x,s) = — EV=1 Hv(x,s)

K

with uniform convergence for all x and s in [a,b\.

In fact, it follows from the asymptotic formulae for the eigenfunctions that
the circles Tfccan be chosen so that mk+l = mk+2; on the other hand, it is easily
seen that, in the square a < x,s < b,

tends uniformly to zero as k -> oo.

A*

The proof of the relation (11) is based on

LEMMA 1

On the circumference Tk the function G(x,s, A) satisfies the inequality

|G (x,s,A )| < ( 1 2 )

where M is a certain constant.

Proof. We put A = —p". Then, with a suitable choice of arg p the circumference
Tk transforms into the arc y k: 0 < arg p < 2n/n, |p| = ^/|A|; y k varies entirely
within the two neighbouring domains S0 and Sk of the complex p-plane.
In order to prove the inequality (12), we make use of formulae (32)-(34) of
§3.7; it is convenient to deal with the cases n even and n odd separately.

(a) Let n be odd; n = 2p — 1.

Let the numbers <0 ^ ( 02, . . . , w n be numbered in the order such that, for p e 5"0,

t%(po>1) < < ... < b^(pajn).

Then, for p e S0, we have

^(poii) < 0 , . . . , 0 ? ( p a > ft_ 1) < 0 ,

^ ( P <°n + 1) > 0 , . . . , & ( p a ) n) > 0 .
 §5.3 REGULAR BOUNDARY CONDITIONS 85
We denote the two parts of the arc yk which lie in S0 and for which
<0 or > 0
by y'k and y'k respectively (see Fig. 13).
We evaluate the function G(x,0A) on the arc yk, making use of the formulae
(32)-(36) of §3.7.

We denote by Wv the cofactor of the element y (" x). in the determinant

y[n- l}(0 y(r 1}(0 ... /r 'X o

y[n- 2)(0 y(r \ 0 - y(r 2KO

y^O y a(0 yn(0

and put
WM) (14)
z v( 0 =
no’
Then formula (32) of §3.7 can be written as

g(x,0 = ± \ £ y v(x)zv(0- (15)

v—1

By Theorem 1 of §4.5, we have, for p e S0,

y (?X0 = j = 1 , 2 v = 0,1,2,...,(» - l).
We substitute these expressions into (14) and cancel the factors
86 §5 EXPANSION IN EIGENFUNCTIONS

P,P2,...,/>"-2, pP<ȣ
C CP<»%z>, • • • eP<°nS jn
5p J^g n u m e ra to r a n d d e n o m in a to r, so
o b ta in in g \
1 ft
z v( 0 = e - ^ (16)
p - 1 ft
w here
n - 1 n - 1 . 7*- 1
w1 oj2
n - 2 n - 2 n - 2
0>i CO 2

P= 5

11 1 ... 1

and j3vis the cofactor of the element w""1 in this determinant. Hence

» . jiv fa fory = 0 , 1 , 2 , — 2),

(17)
v-i v P 1 for j = (n — 1)..

The system (17) has a unique solution, for otherwise it would be satisfied by

f t ___
J - T
since to" = —1. Consequently (16) takes the form

2V(0 = —
«/> [ - « v], v = 1.2,...,(« - 1). (18)

From formula (15) and formula (32) of §3 it follows that

U ,(s) = - i i V J y M 0 + i i » .* > /{ ) ■ (19)

J=1 J=1
We consider the function G(x,£,X) for x > f ; then the sign in (15) must be + .
We multiply the first/u. columns of the determinant //(x,£, A) in the numerator of
formula (34) of §3.7 respectively by fa iQ , iz a(£),—,iz„(£), and also the
+ l)th to the/ith columns by - i z „ +1(f),— J-r„(f) respectively,
and then add all to the last column. The elements of the last column then have
the form
P
j =i
Z
yj(x)zj(0 in the first row,
and

Z
J~1
U*i(yj)zj(Q —
j-tt +1
Z
Uv0{yj)zj(i), v= l,2,...,n,
in the 2nd to (n + l)th rows. By formulae (17) and (52) of §4 and formula (18)
 §5.3 REGULAR BOUNDARY CONDITIONS 87
above, these elements can be written in the following form, with the introduc­
tion of P v,v = 0,

A,r V“ 1 +l J
Moreover, we have the asymptotic formulae proved in §4.5 and 4.9:

y v = epo>vX[l], v = 1,2,
^ tvk ^ v] for_/ = 1,2,...,(// - 1),
Uv(yj) = I />*V(K ^ * V] + ep^ [ ^ ] ) for j = p,
{pKep^ u >)v] f o ry = 0* + 1), 0* + 2 ) ,...,

A ( A )v== in / vy=/i
f l+1 +

We substitute these expressions into formula (34) of §3 in which the last column
of the determinant H (x,i,A) has been changed in the above way, and distribute
the factors of the denominator A(A) as follows. We divide the (v + l)th row by
pkv for v = 1,...,«, the /u.th column by [0O] + ep“"[0i], and they'th column by
ep0>J for j = (p + 1), (p + 2 n. The formula then takes the following form

e*"**[l] ePtan(X- i ) ^ Po
epm'x[1]
[0O] + ep^ [ e x]
^■( k j +
(-1 )" [«i«M ifiitf] Pi
(?(*,£, A) -1 m + ep^[0,]

^ - q ° u + ep^ m
DW "] Pn
[0O] + ep^ m

By the conditions (13), all the exponents in this determinant have a real part
less than or equal to zero; further, by the results of §4.9 (p. 69), the denominator
[0O] _|_ epo,"[01] is bounded, in modulus, from below by one and the same
number on all the arcs y*. Hence all the elements of the determinant, and
therefore the determinant itself, are bounded on these arcs.
Hence on the arcs y* the inequality

|G(*,f,A)l < £ p j (20)

holds, M being a certain constant.

88 §5 EXPANSION IN EIGENFUNCTIONS

We now show that the same inequality holds on the arcs y'k. To do this we
need only to multiply the elements of the first (p. — 1) columns of the
determinant H(x,$,A) by iz i(0 ,iz 2(O> •••> - i(0> and those of the pth,
0* + l)th,...,«th columns by —iz„(f)» —i^ + ^ O ,...,—iz„(0, and add all to
the last column. Then by repeating the previous argument, we find that the
inequality (20) holds also on the arcs y ' f
Hence the inequality (20) has been shown to hold on the part y k + y'k °f the
arc y k lying in S 0. But the preceding arguments are clearly applicable to any
domain S^. Hence it follows that (20) holds on the part of the arc y k lying in
Si and indeed on the whole arc y k . Changing from p to A, we obtain the
inequality (12), and the lemma is proved for n odd.
(b). Let n be even; n = 2p.
This case differs from the preceding one only in that, in two columns, the
pth and (ju. + l)th, the numerators
ep^[l] - [f] and epm»[\] - [£"]
appear. These are bounded from below on the arcs y k, and therefore the
inequality (12) holds on the arcs y'k . Similarly, it may be shown that it holds on
the arcs y ' f
By applying the lemma just proved, we obtain for the integral Ik the estimate

from which follows immediately the uniform convergence I k 0 as k oo.

The considerations on p. 84 therefore lead to the following theorem.
theorem 2

The Green’s function for a differential operator L generated by regular boundary-

conditions can be expanded in a uniformly convergent series

G M = - V E=1 A (21)
H f x f ) being the residue of the function G(x,£,\) at the pole Av.
If all eigenvalues of the operator L are simple zeros of the function A(A), and
if the coefficients of the corresponding differential expression I satisfy the
differentiability conditions necessary for the existence of the adjoint expression,
then (see §3.8)
= y,(x)z,(0.
where >>v(x),zv(x) are eigenfunctions of the operators L and L* corresponding
to the eigenvalues Av and Av and are normalized so that1

1 W e shall assum e these norm alization con d ition s to be satisfied in T heorem s 3 and 4,
which follow .
 §5.3 REGULAR BOUNDARY CONDITIONS 89
If we apply Theorem 2 to this case, we obtain

THEOREM 3
I f all eigenvalues of an operator L of the form considered in §4.1, which is
generated by regular boundary conditions and for which the adjoint differential
operator exists, are simple zeros of the function A, then its Green's function can
be expanded in a uniformly convergent series

(22)

From this theorem we easily obtain one on the representation of a given

function.

theorem 4
Let L be an operator generated by regular boundary conditions with a differential
expression of the form considered in §4.1 and for which the adjoint differential
operator exists. Let all eigenvalues of L be simple zeros of the function A.
Then any function f{x) in the domain of definition of the operator L can be
expanded in a uniformly convergent series of the eigenfunctions:
oo
/(*) = Z “vl'vW.
*, = /:
where j\,(x), z fx ) are the eigenfunctions corresponding to the eigenvalues
Av, 3Vo f the operators L, L*.
Proof. We put L f = h; then

/ w = j;

In this we substitute for G(x,£) its expansion (22). Since the series is uniformly
convergent, we may integrate termwise. Hence
CO
fix) =Z
where

= (/>v)-
90 §5 EXPANSION IN EIGENFUNCTIONS

The asymptotic formulae obtained in §4 also enable the convergence to be

investigated of an expansion in terms of eigenfunctions of a given function
which does not belong to the range of definition of the operator L. The results
found are similar to those for ordinary Fourier series; see the papers by
Birkhoff [5], Tamarkin [110], and Stone [109a].
Remark. It follows from the Corollary to Theorem 1 in §5.2 that every func­
tion / belonging to L 2(a,b), and not merely functions f belonging to the
domain of the operator, can be expanded as a series in the eigenfunctions of a
self-adjoint operator L and this series is convergent in the quadratic mean.
Since the eigenfunctions of a self-adjoint operator are orthogonal to one
another, the expansion also converges unconditionally.
The question arises whether a similar proposition holds for a non-self-
adjoint operator L. G. M. Keselman [49] and V. P. Mikhailov [75] have given
an affirmative answer to this question on the assumption that the operator L
is generated by regular boundary conditions, and that, for an operator of
even order, 6\ ■£ d-idv Their results contain sufficient conditions to ensure
that the eigenfunctions of the operator L should form a Riesz basis in the
space L2(a,b) — (see N. B. Bari [3]). We say that a complete system

C L\a,b)

is a Riesz basis if for all f e L2(a,b)

Z I \ \ A xyPn(x)dx < 00,

n=l 1 Ja
and if, for any sequence of numbers cly c2, . . . such that

Z kl2< °°>
71= 1

there exists a function / e L2(a,b) such that

J* fi.x)<p„(x)dx = cn , n = 1 ,2 ,3 ,... .

If is u Riesz basis, then the following lower and upper estimates

hold, instead of Parseval’s relation:

m J* |f(x) 12dx < | |V(-v)«pn(.v)^Y |2 < A /J * |/(.v) \2dx,

where the numbers m and M, M > m > 0, are independent of/.

We remark that operators whose eigenfunctions and associated functions
form a Riesz basis are spectral operators: see N. Dunford’s survey [21],
 §5.4 SEPARABLE BOUNDARY CONDITIONS 91

4. Expansion in terms of Eigenfunctions when the Boundary Conditions are

Separable

The boundary conditions are said to be separable if each separate U fy ) con­

tains the values of the function and its derivatives at only one of the two
boundary points of the interval. Separable boundary conditions therefore have
the form

u j(y) = Z <w i'° = 0 forj < m,)I

(23)
Uj(y) = Z Pjk/P = Ofory > m.J
k= 0 '
For m = 0 or /// = n, they are equivalent to the Cauchy initial conditions
= 0 or y[k) = 0, k = 0,1,...,(» — 1).
It may readily be seen that, if 0 < m < n, the separable boundary conditions
are not regular; the only exception is the case where n is even and m = \n. In
this case the boundary conditions are of the Sturm type and are therefore,
under certain conditions, regular (see §4.8).
In the case of separable boundary conditions, the concept of the weight of
a matrix is useful.1 Let A be a matrix of rank m\

a10 ••• “l.n —1

A =

_ am0 a m ,n-l_

then at least one of the determinants

“i; m
(24)

does not vanish. We call the sum j\ + ... -f j m the weight of the determinant
(24), and the weight of the matrix A is the greatest weight of its non-vanishing
determinants (24).
lem m a 2
Among the determinants (24) of the matrix A there is just one non-vanishing
determinant whose weight is equal to the weight of the matrix A.

1 T his con cep t and also the p ro o f given later in this section are due to M . V . K eldysh [47].
92 §5 EXPANSION IN EIGENFUNCTIONS

Proof. We prove this assertion by induction. For m = 1, the assertion is

clearly true. v
Suppose it holds when the number of rows in the matrix A is less than m ; we
shall show that it then holds for a matrix of m rows.
Let the number p be chosen so that the (p + l)th column contains at least
one non-zero element, while all the later columns contain only zeros, i.e.

«10 «1p o 0
A =
La mO ~mp 0 ... 0

Clearly, p + 1 > m.
We consider the determinant

a l tp-m + l a lp

a m,p —m + 1 ••• a mp

if it is not equal to zero, the proposition obviously holds.

Suppose, however, that it vanishes, and let q be the rank of its matrix, so that
q < m; then (m — q) of its rows are linear combinations of its remaining rows.
Without loss of generality we may assume that these are its last (m — q) rows,
since an interchange of rows in the matrix A merely changes the signs of the
determinants (24). We add to the last (m — q) rows of matrix A certain linear
combinations of its other rows (this does not alter the determinants (24)) to
bring A into the form

a 10 •• «]i p - m a l,p-m + l •• a lp 0 . .. 0

/
a q0 a q ,p - m ^q .p —m + l •• <p 0 .... 0
/
a q + 1,0 + l,p —m 0 .. 0 0 0

«m0 m 0 0 0 0

The rank of the matrix

a Up- m+l a i.p
a; =
L °W - m+1 qp _J
 §5.4 SEPARABLE BOUNDARY CONDITIONS 93
is obviously equal to q, and the rank of the matrix
a.9 + 1,0 /
a4+l,p—m
a; =

mO ocm,p —m

is equal to (w — q). By the hypothesis of the induction, the matrix A[ has just
one non-zero determinant of the maximum weight; let this be
a 1Jm 9+1 XUm

ry rv .
tlijm-q +i *•*
Similarly, the matrix A'2 has just one non-zero determinant of maximum
weight, say
“9+lJi •• Jm-q

t
0im,jl tjm—q
Then the determ inant
t / <*-,t v - q+i /
“iJi Jm-q ••• “Urn

&q>j•i ^aJm-q OLQ

*tJm
• - q+i a9.Jm
D =
cV W i ••• ^ Jm- q 0 ... 0
• • • ... .
/ 0 0
0Cm,j1 m>Jm- q
is obviously non-zero, and all other non-vanishing determinants of the matrix
A ' have a smaller weight. For, any such determinant A has at least one sub­
determinant Aq formed out of the first q rows of the matrix A and which,
together with its complement A^, is non-zero. The weight of the determinant
A is equal to the sum of the weights of \ and A^.
But by definition of Dq and Dq, we have
weight of \ < weight of Dq, (25)
weight of A^ < weight of Dq, (26)
and the equality signs are possible only if A? coincides with Dq or A^ with Dq.
Hence
weight of A < weight of D,
94 §5 EXPANSION IN EIGENFUNCTIONS
and the equality sign appears only if it occurs in each of (25) and (26). This is
clearly only possible if A coincides with D.
Hence Lemma 2.
We now consider the matrices
“10 fim+1,0 ^w+l.n-1
A = and B -

_am0 awi,n—1_ _PnO ‘• fin,n-1

formed from the coefficients of the boundary conditions (23).
Since these boundary conditions are linearly independent, the rank of each
matrix A and B is equal to the number of its rows; hence we may speak of the
weights of these matrices. In future, we shall denote the weight of the matrix A
by h and that of the matrix B by /. Further, we shall put a = 0 and b = 1; as
before, this causes no loss of generality.
lemma 3 (M. V. Keldysh)
An operator L which is generated by the differential expression l(y) = _j/n)
and the separable boundary conditions

Z £W o c) = 0 for j < m,
k= 0

Z Pjk/P = 0 for j > m,

k=0
which are not equivalent to the Cauchy initial conditions has infinitely many
eigenvalues Av.
For these eigenvalues the following asymptotic formulae hold:

K = Pi, (27)
where
V + - - h- ± I

' ’, = ” 2 2k, " + 0 ( ') ’ (2 8 a)

sin ---- ' 7
n
if n —m = 2k and i f v is sufficiently great, and

v + ri- - lL ± l

p’ = sin ^ +!—1—
> + ° (N‘)7 (28 b)
n
if n — m = 2k + 1 and ifv is sufficiently great.
 §5.4 SEPARABLE BOUNDARY CONDITIONS 95

Proof. We put A = Pn; the equation / n) = pny has the general integral

y = iZ C ^ t
-1
where a>j,j — 1,2,...,//, denote the different /ith roots of unity. By substituting
this expression in the boundary conditions (23) we obtain a system of homo­
geneous equations

Z cq Z = 0, 7 = 1,2,...,1/1
g-1 k-0

Z Z Pjk<*kqPke<0qp = 0, y = (w + 1), (m + 2),...,n

9= 1 fc= 0

for the unknowns c1,c2,...,cn. Hence the eigenvalues for the operator L in
question are determined by the equation

" £ ... £ <Xik<°kp k

k- 0 fc= 0

E a m k ™ lp k ...
k =0 k= 0
D(P) = = 0 (29)
i ... " i p .+ ijy y < f
k- 0 k-0

"l ...
k-0 k-0

Clearly, we need discuss only the case 1 < m < \n, since the case \n < m
< n — l can be reduced to the first case by the substitution = 1 — x.
We put

A{j = E aik(tijP > i ~ l,2,...,m, j 1,2,

k=0 (30)
B u = E / = (m + 1),. .,«; j = 1,2,.
k-0
By Laplace’s expansion theorem,
D
^ •*• ^1/m +i *•• Bm+i,jn
A(/>) = E ± D
.e<®/m+,+- +w/> , (31)
>4 .
•• ^nijm •• Bnj„
96 §5 EXPANSION IN EIGENFUNCTIONS

the summation being over all possible permutations +

of the numbers 1 , 2 satisfying j\ < j 2 < ... < j n and j m+1 < ... < j n. In
other words, D(p) has the form v

Dip) = t Pki p ) ^ p, (32)

*=i
where Pk(p) is a polynomial in p of degree < n{n — 1), and all the exponents
p.k are different.
We are concerned with those terms of this sum in which the real part of the
expression coJm+i + ... + <*>j„ takes the greatest value. We shall shortly see
that for each such term the polynomial Pk(p) associated with the term consists
of just a single product of two determinants from (31). Hence it suffices to
show in each case that neither of the determinants

.. Au D
A lji •^ro+ljm +i Bm+ljn
♦ * 9 • • (32 a)
d mjl A .
** ^ntjm Bn .• +1
njm ■■ K , .
in this product vanishes identically. We shall prove this assertion for the first
of the determinants; the proof for the second of the determinants is similar.
We have:
.
A lj\ • •• A ljm “i • •• « 1 hm “ 5l -
U )hi m
J1

p hl + ... + hm
= I ± (33)
d mj l •••
A .
^ tn jm “ m/i, • ** ^ m h m . Jm
••

But if &(a>Jrn+i + ... + iOj-n) takes the greatest value, then the points
o>Jm+i,ojJm+i,...,ajjn on the unit circle are situated as far to the right as they can
be in any of their possible arrangements, and consequently, when numbered
appropriately, they lie within an angular distance of 2tt//i from one another
(Fig. 14). If the numbers a>1,a>8,...,ct>,l are numbered cyclically in a suitable way,
then
^jh ^ ^jiy ^ ~ 1?2,...,!?,
where £ = e2mln. Introducing the notation t k — £*, we have
.. 1 1 l
ji

••• £/>,» U)y + - + h m 0) (34)

• =

ym - 1 ym - 1
.. a>)m
Jm *** W»m

since the numbers £v ihi,...£ hm are all different.

 §5.4 SEPARABLE BOUNDARY CONDITIONS 97

Hence we have only now to show that among the various non-zero deter­
minants
“i*. ai *m
(35)
a mh, a mh,„

there is just one for which the exponent /ix + ... +/z,„ of p takes the greatest
value. But this exponent is equal to the weight of the determinant (35); hence
our assertion coincides with the enunciation of Lemma 2. Hence the first, and
similarly the second, determinant does not vanish. On the other hand, the
values of +l + ... + a>Jn for the separate combinations with greatest real
part are obviously different. Hence in the corresponding polynomials Pk only
one single product of determinants appears, and it has been proved that
the terms which have the greatest value of +i + ... + atjn) cannotvanish
identically. For {n — m) even, there are just two such combinations (cf.
Fig. 15); for (n — m) odd, just one. In the latter case, we notice that the two
combinations having a not smaller real part of + ... + u>Jn also have
non-vanishing coefficients in formula (32), since these too consist of only a
single product of determinants.
We examine the cases (n — m) even and (n — m) odd separately.
1. (n — m) even, n — m = 2 k .
By numbering the numbers ojj suitably we obtain two sums
u t_ K + co_K+1 + ... + ^-K +l + W-K+2 + ” • +
98 §5 EXPANSION IN EIGENFUNCTIONS

with the greatest real part (see Fig. 15, drawn for k — 2). We put arg p = <p
7T 7T
and consider the function D (p ) in the sector S : ------ < <p < - . In this sector
n n
at least one of the exponents
( °> -k + "-k+i + ••• + w k - i )p> (° > -k +i + a , -K+2 + •••+

has a real part greater than the real part of all other exponents in the expression
for D ( p ) . Let h be the weight of the matrix
*10 a l,n - 1

A -
xm0 a m ,n - 1J

and / the weight of the matrix

Pm +1,0 •• Pm + l,n —l

B -
_AiO Pn,n- 1

Then by the above investigation, in the sector S' we have

D(p) - +
e <oK + i + £ 0 _ K + a + ... +coK) p p ' p h + 1
1+ 0

Now a.' and j8' clearly differ only by a factor of (—i)n“ +»/«.

fS' = l)"“ V *I(fc+,)/",

because we obtain the coefficient jS' from the coefficient a' by replacing the
orderh , j z,...,jn in the determinants (32 a) by the order Hence the
 §5.4 SEPARABLE BOUNDARY CONDITIONS 99

sign of the product of determinants in formula (31) changes by ( - 1 ) " '1,

while each of the numbers <ajk is multiplied by £ = eui,n, so that, for example,
a in formula (34) is multiplied by e2Kih,n. Hence we have

D(p) = 0c'ph+le(.a>-K+co-K+l +...+(oK. l)pl

_j_ ( _+‘)lne (o K - <oK)p

We put a)K = a + ir; then oj_k coK = a — fr, where

. 2ktt
t = sin — . (36)
n
Hence we have
D ip) = oc'ph +

+ +C>(p) }’ ^
so that the equation Dip) = 0 is equivalent to the equation

(-1 )"-1 + 0\^j + e2i^ +> +oi = o.

Hence
e2,Op^(i,+0 ] = ( _ 1)™+ o Q ;

and so for the required eigenvalues we have

7T
2 i I Tp -\— {h + /) = (2v + n)ni + 0 ( - J , v = 0,±1,±2,....
n
Therefore
. n h-\-l
v h---------------
2 n n /l
p = 7T-------------------------------1- O
T \p

and so
n h+ l
v H---------------
2 #1,^/1
p ~ TT ---------------- + O
T \ V

On the other hand, by applying Rouche’s theorem, it may easily be shown

that (cf. the detailed investigation in §4.8 for the case of regular boundary
conditions), for sufficiently large |v|, there is in the neighbourhood of the point
100 §5 EXPANSION IN EIGENFUNCTIONS

just one zero pv of the determinant D(p), and

i.e. by (36),

2. (n — Jii) odd; n — m = 2 k + 1.

In the sector S: — < < P < 0 the expressions

n

(a>_K+ a»_K+1 + ... + wK)p and (o>_K+1 + oj_k+2 + ... + 0JK+1)p

have a real part greater than any other expression

(°Vm+i + ••• + w; > -

By considerations similar to those of case 1 we find

D{p) = c (»-K+®-K+i +-+®-)pa ^ i +

_|_ ( _ l ) n- 1e(“ - K+1+" - K+*+-" +<°K+1)pa'e?5iT±^ 1 + o

Hence it follows that D(p) has infinitely many zeros pv in the sector S, where

. n h-\-l
V + -----------
_ in 2 n
pv = 7TC n
. 2k -j-1
s in --------77

This concludes the proof of Lemma 3.

We now consider briefly the question of expansion in terms of eigen­
functions and associated functions.
An expansion in terms of the eigenfunctions of a differential operator does
not, in the general case, have the usual orthogonal properties. However, as
we proved in §5.3, this expansion does have properties similar to those of an
ordinary Fourier series expansion if the operator is defined by regular bound­
ary conditions. If the boundary conditions are separable, the situation is
more complicated.
 §5.4 SEPARABLE BOUNDARY CONDITIONS 101
W. Ward, [117], investigated the eigenfunction expansion for the boundary-
value problem

y"> + Ay = 0, y(0) = . . . = / " " 2>(0) = y( 1) = 0. (38)

He proved that an expansion in a uniformly convergent series in the eigen­

functions is possible only for functions which satisfy certain analyticity condi­
tions, and consequently such an expansion is far from being always possible
for all functions in the domain of definition of the corresponding operator L.
On these questions of eigenfunction expansions there is quite an extensive
literature (see [17], [95], [122], [50a, b], [43], [44], [102], [116]).
Here we shall only go into some results due to A. P. Khromov [50c] and
W. Eberhard [24], who make use of Lemma 3 (M. V. Keldysh) in their
investigations.
A. P. Khromov considered the operator L corresponding to the differential
expression

/[}'] = >’(n) + P2( * ) / n_2) + • • • + p„(x)y, Pv e C ^ O , ! ) (39)

and general separable boundary conditions. It is not difficult to see that in

this case the assertion of Lemma 3 remains true. In what follows we need the
concept, introduced by M. K. Fage [26], of an /-analyticalfunction. We mean
by this a function / to which the differential operator / can be applied arbi­
trarily often and which satisfies inequalities of the form

p = 0,1,2,..., q = 0 ,! ,...,( » - 1),

(with the exclusion, possibly, of p = 0 and q — 0) in an arbitrary interval

[a,/3] c: (0,1); here C is a constant which depends only on / ,/, a and ft. It
turns out that the properties of an /-analytical function are in many respects
similar to those of an ordinary analytical function.
Let {<pk(x)}f=i be the sequence of all the eigenfunctions and associated
functions of the operator L under consideration, numbered in order of increas­
ing absolute values of the eigenvalues. A. P. Khromov has established the
following proposition.

Theorem 5.
00

I f the series £ ak<pk{x) converges uniformly on some interval [x0>*i] <= (0,1),
then:
102 §5 EXPANSION IN EIGENFUNCTIONS

(a) the several series

P = 0,1,2,..., ? = 0,1...... (n -1 )

converge absolutely and uniformly on every interval [0,a] c: (0,xx).

(b) the sum f i x ) o f the series
00
Z
k=1
a k<Pk(x)

w l-analytical function;
(c) functions
fix ), /[/(*)L /2[/(*)L---
satisfy the boundary conditions at x = 0.
Conversely, if a function fix ) which is summable on the interval (0,1) is an
l-analytical function in the interval [0,u), 0 < a < 1, and if the functions
fix), /[/(*)], /2[/(x)],... satisfy the boundary conditions at x = 0, then fix )
can be expanded in the system {9>*(x)}”=1 as a series which converges uniformly
in every interval [0,6] c [0,a) O [0,^?), where the number R is defined by
d_ k—nlkfri] 1Ik
— = lim lWn\ f i x )]
R £-*oo h . fix. x =0

[x] being the integral part o f the number x.

W. Eberhard’s results ([28]-[31]) have much in common with those of
Khromov. Eberhard considered an operator L corresponding to the differ­
ential expression (39) with coefficients p„(x) which are holomorphic in the
circle |x| < 1 in the complex x-plane, and to general separable boundary-
CO
conditions. He proved that, if the series Z afc<Pfc(A) converges uniformly on
k~l
the interval [x1?x2], then this series converges uniformly in every closed sub-
domain in an open polygon (and consequently that the sum of the series is
regular in this polygon) whose vertices are at the points
COS iKTTfi e{2
n.
COS 7rjn
in the case where n — m — 2,
and at the points
COS (2k + 1f i r i n ' e2pi„/n
xr v 1
cos n/n
in the case where n — m = 2k + 1;
here, as usual, m is the number of boundary conditions at x = 0. Eberhard
has also obtained certain necessary conditions, in terms of ordinary analy-
 §5.5 g r e e n ’s f u n c t i o n w it h m u l t ip l e pole 103
ticity, for the function/(.v) to admit a uniformly convergent expansion as a
series in the system {9>*(x)}“=1; and here certain additional restrictions, as well
as holomorphism, are imposed on the coefficients ofthe differential expression
(39).

5. The Case of Multiple Poles of the Green’s Function; m-fold Completeness;

Keldysh’s Theorem

We again consider, as we did earlier, the generalized eigenvalue problem

Ky) = f , (60)
U * (y )= 0 , v = 1 , 2 , ( 6 1 )
let A^A^Aa,... be all the eigenvalues of this problem. They may in general be
multiple zeros of the characteristic determinant A(A) and consequently
multiple poles of the Green’s function. Let Abe any one of these eigenvalues,
and let the eigenfunction and its associated functions which occur in the
canonical system belonging to Abe

We construct m systems of functions

yk + V = 0,l,2,...,(m — 1).
+ 4 f=0
Each of these systems
y v,o)jyv,i)j __ (■v,q- 1)

will be called a derived system from the original system y,yi,---,yq-!.

The system of all eigenfunctions and associated functions of the boundary-
value problem (60), (61) is said to be m-fold complete if every system of m
functions
(AW.AW,
each of which has continuous derivatives up to order n inclusive and satisfies
the boundary conditions (61), can be represented as the limits of m uniformly
convergent sequences of finite linear combinations

fy(x) = Hm I C $y(r k)
N —*oc> i,k

whose coefficients are independent of v.

The considerations in §5.3, §5.4, together with certain refined results in
the theory of analytic functions, and a formula for the principal part of the
Green’s function (see §3.9), enable the following general theorem to be derived,
which we merely state here without proof.
104 §5 EXPANSION IN EIGENFUNCTIONS

theorem 6. (M. V. Keldysh, [47])

Let
Ky) = +a(*,A )/,-1) + >.. + [pJLx,A) + n y ,
Jcfl'l
where p k(x,A) is a polynomial in A of degree < — ; and suppose the boundary
n
conditions may take the form

ZWfl*"1)=0> / = 1,2,...,/?; 0 <p<n,

1

Z = °» j = (p + l).(p + 2),-••,«,
&=1
or
k = 0 , 1 , 2 , — 1).
Then the system of eigenfunctions and associated functions fo r this boundary-
value problem is m-fold complete.
In the article by V. N. Vizitei and A. S. Markus [114] it is proved that
Theorem 6 remains valid also in the case of regular boundary conditions,
where, when n is even, it is required that 6% ^ Further, it remains true
in the case of general self-adjoint boundary-conditions; the conditions of
periodic type, which feature in Theorem 6, are a particular case of these last.
An expanded exposition of M. V. Keldysh’s results (in particular, those
on which Theorem 6 is based) is given in the monograph by I. Ts. Gokhberg
and M. G. Krein [33]. In this monograph have been collected the most
recent results in the spectral theory of completely continuous non-selfadjoint
operators. (We remark that the resolvent, which we have discussed above, of
the operator L is a completely continuous operator.)
In addition to the problem of eigenfunction expansions, a problem of great
interest is that of determining a differential operator from the specification
of its spectral characteristics, e.g., from the eigenvalues and norms of the
eigenfunctions (with known initial values). Particular aspects of this problem
are considered in V. M. Levitan’s book [63f] and in the articles by Z. L.
Leibenson [59a, b].
Surveys of the general state of the theory of linear non-selfadjoint opera­
tors, both differential and those specified abstractly, are given in the papers
[21], [81i], [48], [73].
The articles [52], [72a], [100], [23b], [59a] may also be consulted on the
subjects discussed in this chapter.
 CHAPTER III

DIFFERENTIAL OPERATORS IN A SPACE

OF VECTOR-FUNCTIONS

§6. Basic Concepts

1. Linear Differential Expressions in a Space of Vector-Functions

Let Rm denote an ///-dimensional complex vector space; i.e. Rm consists of all
vectors
y = O 'i.jv
where each y r is a complex number.
Functions y — of the real independent variable x, whose values are not
numbers but vectors in Rm are called vector-functions. A vector-function is
therefore simply a system of /// complex-valued functions
>’(*) = 0t(*),T2(*)»... ,ym(x));
each of the scalar-functions jvU) is called a component of the vector-function
}<x).
The function X*) is said to be continuous at the point x0 if all its components
are continuous at x 0.
Similarly, a function ;»(x) is said to be differentiable if each of its com­
ponents is differentiable, and by definition
/(* ) =
Derivatives of higher order are defined in a similar way.
It may easily be seen th a t:
O' + *)' = y ' + z', (Ay)' = A> + A /,
(y,zY = (y',z) + O',*')
where
m ___
(y,z) = Z yk(x)zk(x).
k= 1
In addition to vector-functions we shall also be concerned with operator-
functions. The values of operator-functions are linear1 operators in Rm. Such

1 T h e reader is rem inded that the author’s definition o f linearity, on p. 2, includes

h om ogen eity.

105
106 §6 BASIC CONCEPTS

operators can be represented by means of square matrices A(x) = [<3jn(x)] of

order m, whose elements are scalar functions. We shall not, from now on, make
any distinction between operators and the matrices which represent them.
Essentially, operator-functions are also vector-functions, since the aggregate
of all linear operators is a vector-space of dimension m2. Consequently an
operator-function A(x) will be said to be continuous at the point ,x0 if all its
functions ajk(x) are continuous at x0, and to be differentiable at x0 if all the
functions aJk(x) are differentiable at x0. So, by definition, A'(x) is the matrix
whose elements are a'jk(x). We see without more ado that the following rules
hold:
(A + B)' = A '- h B', (AA)' = A'A + \ A ',

(AB)' = A 'B + A B \ (Ay)' = A 'y + Ay'.

We denote by C(n) the aggregate of all vector-functions y(x) which have

continuous derivatives up to the nth order inclusive in a fixed interval [a,b}.
Let P0(x),Pi(x),...,Pn(x) be operator-functions which are continuous in [a,b\,
and suppose det P0(x) ^ 0 in [a,b\. An expression of the form1

l(y) = Po(x)y(n) + P1(x)y(n~1) + ... + Pn(x)y (1)

is called a linear differential expression in the space of vector-functions.

It is defined for an arbitrary vector-function X*) belonging to C(n), and
itself represents a vector-function, continuous in the interval [a,b\. The theory
developed below for such differential expressions and the differential operators
corresponding to them is formally similar to the theory already presented for
scalar functions.
We remark that, essentially, l(y) is a system of m differential expressions of
the «th order which depend on m scalar functions For
n = 1, we consider, for example, the expression

Ky) = / + P(x)y,

and resolve the vector-function z = l(y) into its components by putting

K y )= (il(y )M y ),-,L (y )Y
Then we have
m
lj(y) = y ) + Yj p jk(x)yk, j = 1,2,...,m,
*=i
where Pjk(x) are the elements of the matrix P(x).

1 W e rem ind the reader that, e.g., the expression A ( x ) y {n~ l) denotes the result o f applying
the operator A (x ) to the vector >,ln~ 1:(a')-
 §6.3 HOMOGENEOUS OPERATOR EQUATION 107

2. Boundary Conditions
We denote by
yay 0>■•■y on"1}; y b>y'b>■••,y (2 "1}
the value of the vector-function and its first (n — 1) derivatives at the points a
and b respectively, so that ^ are vectors in the space Rm. We put
u (y) = A y a + Ay'a + . . . +
+ B0y„ + Bxy'b + ••• + (2)
where A 0,...,A n^ x,B{),...,B„^x are fixed linear operators in the space Rm. If
several such forms are given, U^y^Uziy),- ■-,Vq{y), then equations of the form
Ux(y) = 0, U2(y) = 0,...,Uq(y) = 0 (3)
are called boundary conditions.
We denote by Q) the aggregate of all functions y e C(n) which satisfy the
conditions (3). Let L be an operator, with S> as its domain of definition, which
is defined by the equation
Ly = /(>’);
then L is called the differential operator which is generated by the differential
expression l(y) and the boundary conditions (3).
We put
Uv{y) — Ayi0y a + ... + + 7?v.Oy b+ ... + T ^ n -l/fr" x)>

V = 1,2,...,?.
We will assume in the definition of that the forms Uv(y) are linearly in­
dependent ; this implies that the rank of the matrix formed from all the elements
of the matrices [Avj], [5VJ], viz.
A w ,.

A 20^ • '

Aq 0 ) • «’ ')Aq n—\>BqQ>. <

is equal to m q; for each form Uv(y) has m components.
From now on we shall mainly be concerned with the case q = n.

3. The Homogeneous Operator Equation

We want now to examine the following homogeneous equation in an unknown
operator function Y(x):
l(Y) = P0(x) Y (n) + Pi(x) Y ^ - v + ... + Pn(x) Y = 0. (4)
108 §6 BASIC CONCEPTS

As before, it is assumed that P0(x),...,Pn(x) are continuous and that det P0 ^ 0

in the fixed interval [a,b].
We mention some of the most elementary properties of solutions of this
equation; they resemble the properties of solutions of ordinary, homogeneous,
nth-order equations.
Solutions Yx, 72,..., Y„ of (4) are said to be linearly independent if the
equation
YXCX + 72C2 + ... + Y„C„ = 0,
where Ca,C2,...,Cn are constant operators, holds only for Cx = C2 = ...
= C„ = 0; otherwise they are linearly dependent.
A system of n solutions Yx, 72,..., 7„ is linearly independent if and only if the
determinant of the «2 x n2 matrix

Yx Y2 Y„
r; y; rn

y(n-l) 7 (2" - X) .. yA n(n-1)

formed from the matrices Y (f is non-zero.

I f the solutions Yx, 72,..., 7„ are linearly independent, then any other solution
Y has the form
Y = YXCX+ Y2C2 + ... + YnCn. (5)
Consequently, formula (5) defines the general integral of equation (4).
If 7 is a solution of equation (4), and if c is a fixed, constant vector, then we
get a solution of the equation l(y) = 0 on putting j(x) = 7(x)c. To see this, we
need only apply both sides of equation (4) to the vector c.
Hence, and from formula (5), we can at once infer:
Any solution of the equation l(y) = 0 has the form

y — Yxcx + 7 2c 2 + ... + Yncn, (6 )

where cx,c2,...,c„ are arbitrary constant vectors in Rm.

4. The Homogeneous Boundary-Value Problem

The problem of determining a vector-function y which shall satisfy the con­
ditions
Ky) = 0, (7)
Uv(y) = 0, v= 1 , 2 (8)
is called the homogeneous boundary-value problem.
 §6.5 ADJOINT DIFFERENTIAL EXPRESSION 109
We consider the n2 x n2 matrix

v m u t( n ... u t( r y
u rn u jir a ... U2(r„)
(9)

u .m um u,m J
The results in §6.3 imply:
I. A homogeneous boundary-value problem (7), (8) has a non-trivial solution if
and only if the determinant of the matrix U vanishes.

5. Lagrange’s Formula; the Adjoint Differential Expression

In addition to the assumptions made in §6.1, we now further require that,
for k = 0,1,2,...,«, the coefficient matrices Pk(x) shall each be continuously
differentiable (n — k ) times. We denote the scalar product of the vectors
y, z e Rm by (y,z). Integrating by parts, we obtain

J* (l(y),z)dx = P(v ,0 + (y,l*(z))dx, (10)

where P(r),Q is a bilinear form in

v = ( y ay a,-~,y(a ' 1}jw'J,•••, / b ' 1})

and
£ = (za,z'a,... ,z<" ” ...,z(b~ V)>
and where1
l*(z) = ( - 1 )n(P *zfn) + ( - 1 ) " ' \P ? z)(n- 1) + ... + P*z. (11)
The differential expression l*(z) is said to be adjoint to l(z). Formula (10) will
be referred to as Lagrange's formula. A differential expression l(y) is said to be
self-adjoint if l*(y) = l(y).
Repeating the argument in §1.5, the following general representation for a
self-adjoint differential expression is obtained:
II. Any self-adjoint differential expression is a sum o f self-adjoint differential
expressions o f the form

h.(y) = 4 .- i« =
where P = P(x) denotes an operator function all o f whose values are Hermitian
matrices in R„.

i I f A = [apq] is a square m atrix o f the «th order, w e denote by A* the adjoint matrix, i.e.
the m atrix B — [bpq] for w hich bpq - aqp. A m atrix is called Hermitian if A* = A.
110 §6 BASIC CONCEPTS

6. Adjoint Boundary Conditions; Adjoint Operators

We supplement any given set of linearly independent forms Uly...,Un to form a
complete system of linearly independent forms U1,U2,---,U2n. Repeating the
argument of §1.6, we can transform Lagrange’s formula to

(Ky\z)dx = (UlyV2n) + (U2yV2n. x) + ... + (U2nyVx) + j ba (y,l*(z))dx, (12)

where VlyV2,...,V 2n are linearly independent forms in the variables

The boundary conditions

Vv = 0, v = \ , 2 (13)
(or any conditions equivalent to them) are said to be adjoint to the boundary
conditions
Uv = 0, v = l,2,...,n. (14)
The operator generated by the differential expression l*(y) and the boundary
conditions (13) is said to be adjoint to the operator L generated by the
differential expression l(y) and the boundary conditions (14). It will be denoted
by L*.
It follows from formula (12) that, for the operators L and L*, the equation

(Ly,z)dx = J* (y,L*z)dx (15)

holds.
Conversely, if a differential operator L x satisfies the condition

(Ly,z)dx = (y,L1z)dx

for all functions y,z in the domains of definition of L and Lx respectively, then
Lx = L*.
For vector-functions y(*) we define the scalar product (y, z) by

0\2> = J* (y(x),z(x))dx.

Then formula (15) can also be written as

<Ly,z> = <y,L*z>. (16)
An operator L is self-adjoint if L* = L. In other words, an operator L is
self-adjoint if it is generated by a self-adjoint differential expression and self-
adjoint boundary conditions. For a self-adjoint operator L, equation (16)
takes the form
<Ly,z> = <y,Lz>.
 §6.7 EIGENVALUES AND EIGENFUNCTIONS Ill
The homogeneous boundary-value problem L*z = 0 is said to be adjoint to
the homogeneous boundary-value Ly = 0. By an argument essentially the
same as that in §1.7 we obtain:
The rank o f a homogeneous boundary-value problem is the same as that o f the
adjoint homogeneous boundary-value problem.
In particular,
A homogeneous boundary-value problem has a non-trivial solution if and only if
the adjoint homogeneous boundary-value problem has a non-trivial solution.

1. Eigenvalues and Eigenfunctions of a Differential Operator

A number Ais called an eigenvalue of an operator L if there is in the domain of
definition of L a vector-function y & 0 such that
Ly = A_y.
All the basic theorems in §2.1,2.4,2.5 remain true for operators in the space
of vector-functions. In particular, the eigenvalues are the zeros of the
characteristic determinant
u m V f Y n)
A(A) = (17)

U fY i) u„m
where Yx, Yz,...,Y n are linearly independent solutions of the operator equation
1 ( Y ) - \ Y = 0.
These solutions can be chosen to be integral, analytic, operator functions of
the parameter A; hence A(A) also is an integral, analytic function, and Theorem
I of §2.1 carries over to the present case.
The argument in §2.2 also applies here, so far as the generalized eigenvalue
problem is concerned.
We may observe that the ordinary eigenvalue problem in the space of vector-
functions is equivalent to a certain generalized eigenvalue problem for scalar
functions. To illustrate this, we will examine the eigenvalue problem

+ A(x)y — 0 < x < 1, (18)

y( 0) = J<1), (19)
where y is a vector-function in the 2-dimensional space R z = y(x) = {.Vi(*)>
y fx )}, and A(x) is the operator defined in this space by the matrix

’0 1'
A(x) =
p(x) 0
112 §6 BASIC CONCEPTS

Consequently
A x )y = W *)» />(*)Ti(*)}>
and written in terms of coordinates the prbblem (18), (19) takes the form

^ + y & ) = \y t(x), \
( 20)
— + p (x)y1(x) = A^2(x), j

*(0) = >>i(l); J 2(0) = TzO)- ( 21)

By differentiating the first equation in (20) to obtain

<pyi . dy* _ ■>dyx

dx2 dx dx ’

and then using this equation with (20) to eliminate y 2 and — , we obtain for y x
the differential equation

2d h ^ 2 A ^ + [ A * - PW K = 0 . (22)
dx2

By the first equation in (20) again, the boundary conditions may be written in
the form:
Ti(0) = TiO),
(23)
[ £ - 4 - . - [ £ - 4
The original problem therefore reduces to the following generalized eigen'
value problem:

cfyi
- 2A ~ + [A2 - pix)]}’! = 0,
dx2

*(0) = * ( 0 ; [ ~ ‘ - Aj.,] o= - V .] ^ .

This remark often enables a generalized eigenvalue problem in the space of

scalar or vector-functions to be reduced to a certain ordinary eigenvalue
problem in the space of vector-functions.
In connection with the generalized eigenvalue problem in the space of
vector-functions, see also [95a,b, 122].
 § 6 .8 FIRST-ORDER DIFFERENTIAL OPERATOR 113

8. Differential Operators of the First Order

The eigenvalues and eigenfunctions are determined most easily in the case of
a differential operator L of the first order. The operator is generated by the
differential expression
A» = y' + P(x)y (24)
and the boundary conditions
Aya + Byb = 0. (25)
We want to find the eigenvalues and eigenfunctions of the operator L. To do
this, we first determine the solution of the equation
v' + P{x)y = Av. (26)
Putting y — eXxz in (26), we see that z has to satisfy the equation ^
z' + P(x)z = 0. (27)
To each vector c e Rm we assign that solution z(x) of (27) which satisfies the
initial condition [z]x=a = c. Since the solution is unique, the correspondence
c ->■ z(x)
is one-to-one. Moreover, it is obviously linear: cl ~>zl(x) and c2 z2(x)
imply Acx + /ac2 -> Azx(x) + pz2(x) for arbitrary numbers A,/a. Hence for each
fixed x, this correspondence defines a linear operator Q(x):
z(x) = Q(x)c;
and, by the definition of Q(x), Q(a)c — c.
Hence the general solution of equation (26) is of the form
y = eXxQ(x)c, (28)
where c is an arbitrary vector in Rm.
Substituting this expression in the boundary conditions (25) and removing
a factor eXa, we obtain
[A + e*b- a)BQ(b)]c = 0. (29)
Next we consider the generalized eigenvalue problem
[A + yBQ(b)\c = 0 (30)
for the operators A and BQ(b) in the finite-dimensional space Rm. Let
be all the eigenvalues, and cx,c2,...,cr the corresponding linearly
independent eigenvectors for this problem. Equation (29) shows that eHb~a) is
an eigenvalue and c the corresponding eigenvector of problem (30).
114 §6 BASIC CONCEPTS

Hence all the eigenvalues of the operator L are of the form

A = r------lnju.- + 2km, j — 1,2,...,r:fc = 0 ,;tl, ± 2 , ± 3 ,...;

b —a J
and the corresponding eigenfunctions are

J^exp^ ^ In fj-j + 2kvi^jx Q{x)cj.

Many authors (see, e.g. [6]) have investigated the more general eigenvalue-
problem
— + P(x)y = AQ(x)y, (31)

Aya + Byb = 0, (32)

$ith operator functions P(x),Q(x) continuous in [a,b].
The corresponding results are formulated most simply for the case where,
for each a in the interval [a,b], all the roots of the characteristic equation of the
operator Q(x) are different, say,/^,/^,...,/^. In this case, they can be numbered
so that for each x the functions Pi(x),fj.2(x),...,fj.m(x) are continuous. Hence
there exists a non-singular matrix S (i.e. one for which det S(x) ^ 0) such that
S ~ 1(x)Q(x)S(x) = M(x),
where M(x) is a diagonal matrix having ^fx),y.fx),...,jj.m(x) as its diagonal
elements.
We assume that Q(x) is a differentiable function in the interval [a,b]. Then
S(x) and M(x) are also differentiable. Using the substitution y = S(x)p, we
can reduce equation (31) to the form

— + P(x)p - AM(x)p,
where
P(x) = S~ 1(x)S'(x) + S~'(x)P(x)S(x).
Henceforward, therefore, Q(x) may be taken to be a diagonal matrix. It
turns out that for large |A| the solution of (31) behaves approximately like that
of
dy
Tx=
More precisely: there is a fundamental matrix Y(x) of solutions of equation
(31) for which
Y(x) = Y fx ) + O

where T0W IS a definite matrix independent of A.

 § 7 .1 INVERSION OF DIFFERENTIAL OPERATOR 115
This result enables asymptotic formulae for the eigenvalues and eigen­
functions to be obtained, and also a theorem to be derived on expansions in
terms of eigenfunctions of a boundary-value problem, exactly similar to the
expansions in §§4 and 5. For more details, see [6].

§7. Green's Function for a Differential Operator

1. Inversion of a Differential Operator

Let L be a differential operator for which the boundary-value problem Ly = 0
has merely the trivial solution. Then L has an inverse operator L ~ x. We shall
show that L ~ 1 is an integral operator, the kernel of which is represented by an
operator function in two variables. This function is called the Green's function
for the operator.
The precise definition of Green’s function for the present case is similar to
that given in §3.3 and now runs:
An operator function G(x,£) in the variables x,£ is called the Green'sfunction
for the operator L if it satisfies the conditions:
1°. Each value of the function G(x,£) is a linear operator in the space Rm.
2°. The function G(x,£) is continuous and has continuous derivatives with respect
to x up to the (n — 2)th order inclusive in the square a < x,£ < b.
3°. For each fixed value of £ in the interval (a,b), the function G(x,£) has con­
tinuous derivatives with respect to x of the (n — 1 )th and nth orders respectively
in each of the intervals [a,£) and (06]. The derivative of the (n — 1)th order
makes a jump in value, [P0(^)]“ \ at x = £: i.e.

G(£ + 0 ,0 - ^ G(£ - 0 ,0 = [PoiOV1.

4°. For a fixed £ in the interval (a,b), the function G(x,£), regarded as a function
of x, satisfies the boundary conditions UV(G) = 0,v = 1,2,...,/?, and, in each of
the intervals [a,£) and (06], the operator equation 1(G) = 0, l(y) and U fy ) = 0
being the differential expression and boundary conditions which generate the
operator L.
Repeating the arguments of §3.3-3.5, we obtain the following results:

THEOREM 1
I f the boundary-value problem Ly = 0 has only the trivial solution, then the
operator L has just one Green's function.

THEOREM 2
I f the boundary-value problem Ly = 0 has only the trivial solution, then the
116 §7 green’s function

equation Ly = / , where f is any continuous vector function f(x), has just one
solution, which is given by the formula1

y(x) = J* G(x,myt- (1)

Theorem 2 means that L ~ l is an integral operator with the kernel G(x,£).
The kernel G*(x,€) = [(?(£,x)]* is called the adjoint of the kernel G (x f ). A
kernel is said to be Hermitian if it is the same as its adjoint: G(xJ) = [(/(f,x)].*
Theorems III and IV of §3.5 and all the theorems in §3.6 remain valid.

2. Green’s Function for the Operator L - X I.2

We seek an expression for the Green’s function for the operator L — Al. Let
Yv = Y fx , A) be a fundamental system of solutions of the equation /(Y) — AY.
We choose this system so that the Tv(x,A) are integral, analytic, matrix-
functions of the parameter A(it is always possible to do this). We write | W\ for
the determinant of the matrix
1)
5h
1

rH

y(n -
1

J n
y(n- 2) ... y ( n - 2)
j. n
■

Y1
^

and denote by Vv, v — 1,2,..., n, the mth-order matrices consisting of the

cofactors of the elements Yv in the determinant \ W\. Further, we let Wv,
v = l,2,...,n denote the transposed matrices of Vv.
We put

^V = IH/l ^V» (3)

C i Z r , ( x ) Z jo for(<x,
(4)
-} £ for ( > x,

” U-jf Yj) .. ■ u f Y ny • wu~

u = * , u~i =
_ u jr ,) .. ■ u„(Yny JKi •• ■ w nn_
Here the W lv are m X m matrices.

1 F or a fixed x and P(x,Of(0 m eans the result o f applying the operator P(x,£) to the
vector / (£).
2 In this section 7.2, the author assum es that A is not an eigenvalue.
 §7.3 ANALYTIC NATURE OF GREEN’S FUNCTION 117
I. The Green s function G(xf,X) fo r the operator L — A1 is given by the formula
tt
G(x,£,X) = g(x,£,\) - £ Y fx W j.U fg ). (5)
J,v =l

The proof of formula (5) is similar to the proof of formula (34) in §3.7. If
we apply the method of variation of the parameters to the equation
Ky) — V = / , we obtain for its solution y the formula

y = Z Yv(x)cv + P g(x,£,X)M)d£, ( 6)
v-1 J0

the analogue of formula (31) in §3.7; the cv here, of course, are constant
vectors. By substituting this expression for y into the boundary conditions
Uv(y) = 0, v = 1,2, we find these vectors c„; then by substituting the
expressions for cv into (6), we arrive at formula (5).
If we write A for the determinant of the matrix U, then

(7)

where Vjv is the transpose of that wth-order matrix consisting of the cofactor
of the element U f Yj) in the determinant A. Formula (5) can then be rewritten
as

G ( x J , A ) = g ( x , ^ ) - ~ t Y/x)VJVUv(g). ( 8)
A 1

3 . T h e A n a ly tic N a tu re o f the G reen ’s F unction for the O p erator L — XI

The functions T^jc) and consequently also g(;c,£,A), Vuv, as well as Uv(g) and
A are integral, analytic functions of the parameter A. Hence it follows from (8)
that:
II. The Green's function fo r the operator L is a meromorphic matrix-function of
the parameter A, and only eigenvalues of the operator L can be poles of this
function.
We want to look at the case when A0 is a simple zero of the function A in more
detail.
We assume that the adjoint operator exists as a differential operator of the
type under consideration here, and let z(x) be one of its eigenfunctions corres­
ponding to the eigenvalue 10; we denote by y(x)z*(£) the matrix composed of
the elements >>k(;c)z7(f). Using the result in §7.2, and repeating the argument
in §3.8, we can deduce:
118 §8 ASYMPTOTIC BEHAVIOUR
III. At a simple zero A0 of the function A(A),
A.x)z*(f)
G(x,Z, A) = (9)
(A — A°) JI ( y z)dx
where G f x f f ) is regular in the neighbourhood of the point A0. If, in particular,
fb
Ja(y,z)dx = 1,
then
y(x)z*(f)
G{x,£, A) = + Gfx,^,^). ( 10)
A — Aft
For differential operators of the first order in the space of vector functions,
the Green’s function was the subject of investigation in papers by Bunitski
[13], Bocher [11], Birkhoff and Langer [6], and Bliss [8].

§8. Asymptotic Behaviour of the Eigenvalues of a Differential Operator

1. Statement of the Problem

In §4 we examined the asymptotic behaviour of eigenvalues and eigenfunctions
of a differential operator in the space of scalar functions. The results obtained
enabled us to derive in §5 some general theorems on expansions in terms of
eigenfunctions. In this section we shall be concerned with the analogous
problem, the asymptotic behaviour of the eigenvalues and eigenfunctions of
a differential operator, in the space of vector functions.
To solve this problem, we first investigate the asymptotic behaviour of
solutions of the matrix equation
l ( Y ) + PnY = 0 (1)
for large values of |/>|. The decomposition of the complex p-plane into the
domains S k or Tk (see §4.2) plays an important part here.
For simplicity of exposition we discuss the case where P0(x) = 1, and there­
fore take as our differential expression
l(Y ) = Y (n) + + ... + PnY. (2)
(The possible generalizations to the case P0(x) ^ 1 still remain open for con­
sideration). Further, without loss of generality, we may assume that Pk(x) = 0.
For, otherwise, we can apply to equation (1) the substitution Y = U f, where
U is a solution of the matrix-equation

U' + - p 1t/ = o
n
satisfying the condition
det U ± 0.
 §8.2 SOLUTION OF /( Y) -j- pnY = 0 119
We then obtain for ? an equation of the form
l{f) + P* ? = 0
in which the coefficient of vanishes.
Finally, we can assume a = 0 and b = 1, and so we are considering the
interval [0,1]. (See §4.1).

2. Asymptotic Behaviour of the Solution of the Matrix Equation l ( Y ) + p ”Y = 0

for large |p|
Let then
l(Y) = Y in) + P2Y (n~-) + ... + PnY. (3)
We consider the matrix-equation
/(F) + pnY — 0. (4)
By applying to this equation essentially the same considerations as in §4.5 we
obtain the following theorem.
THEOREM 1

I f the matrix-functions P2,...,Pn are continuous in the interval [0,l],1 then the
matrix-equation
F (n) + P2Y (n~2) + ... + P „ Y + pnY = 0
has in each domain T of the complex p-plane n linearly independent solutions
F1; y2,..., Yn which are analytic in p e T and which for sufficiently large |p|,
p e T , satisfy the relations2

Here o Q is a matrix of the form with A(x,p) uniformly bounded:

|A u M l < M for |p| > R, 0 < x < 1,

where M and R are independent of x.

1 S ee the fo o tn o te , p. 48.
2 H ere 1 stand s for the /ith-order unit m atrix.
120 §8 ASYMPTOTIC BEHAVIOUR

If the functions Pk(x) are not only continuous but also sufficiently often
differentiable, then the asymptotic formulae (5) can be made more precise, and
formulae of the form x

Y k( x ) = e po>kX^ l + ~ ~ + + ... + + O

can be obtained, with similar formulae for the derivatives Y (k \v = 1,2,...,«—1.

(See §4.6 for the details).

3. Normalization of the Boundary Conditions

A given differential operator is characterised by the boundary conditions
Uv(y) = 0, v = 1,2,...,n.
The number k is called the order of a form U(y) if U(y) contains at least one
of the vectors yjjfc) and y ^ but does not contain the vectors or j^v) for v > k.
We consider forms U(y) of order (n — 1), if there are any; they have the
form
Uv(y) = + ••• •

The rectangular matrix Mv,n-i»^v,n-i] has m rows and 2m columns. The

maximum number of linearly independent rows of 2m elements is, however,
2m; if, then, we replace the rows in the forms of order (n — 1) by linear com­
binations of these rows (if this process is necessary), we can arrange that not
more than two forms of order (n — 1) occur.
Continuing in the same way with the remaining forms, we can, after a finite
number of such steps, reduce the boundary conditions to the form

Uv(y) = u j y ) + Uvl(y) = 0
where

U J y ) = A vy ^ Avjy(0J\
y=o

U Ay) = Bvy f ^ Bvjy(A

J =0

n — 1 > ki > k2 > ... > k n > 0, k v+2 > k v,

and where, for each v, v = 1,2,... ,n, at least one of the matrices A V,BVis different
from the zero-matrix.
The operations just described are referred to as the normalization of the
boundary conditions, and the finally resulting boundary conditions of the
form (7) are called normalized boundary conditions.
 §8.4 REGULAR BOUNDARY CONDITIONS 121

4. Regular Boundary Conditions

We consider a fixed domain S k, and number a>lfw2,...,a>n so that, for p e S,

&(puji) < &(pu>2) < ... < @(pa>n).
The asymptotic formulae which follow are going to be derived for a
particular class of boundary conditions which we shall call regular. The
definition of regular boundary conditions depends on whether rt is even or odd.
(a). Suppose n is odd; n = 2p. — 1.
The normalized boundary conditions (6) are said to be regular if both the
numbers 60 and 6m defined by the equation
@0 + @ls + ■•• + ~
(A , + s B j a,*1 ... B ^
A 2u>i z ^ 2 % -l (A 2 + sBf)ofij ... b 2, ^

A„co f- A„a>lh (A„ + sBn)ojk" B A % i - B nw

do not vanish.
(b). Suppose n is even; n = 2p.
The normalized boundary conditions (6) are said to be regular if both the
numbers d_m and 0m defined by the equation
e . ms - m + e_m, lS- m+1 + ... + e j * =

- A lU>ku ( A . + s B J B 1u>kl+2 ... Bi<41

A2a>k‘ ... A 2a>k,_! (A2 + sB2)a>k> ^A2 + - B ^(uk\ l B2a>k\ 2 ... B * Z *

(8 b)
•

A„a>k" ••• A„cuk-_1 (An + s B n)oflr ( a „ + ••• Bna}kn

1
do not vanish.
We can see at once that regularity so defined does not depend on the particu­
lar domain S k selected. For, the numbers d0 and 6m (or 6_m and 6m) are
determinants, which are obtained from those used in §4 to define the numbers
6, merely by replacing ak and by the matrices A k and Bk. Hence precisely the
same laws hold for their transformation due to a change in domain Sk ; in
particular, for odd n, they are multiplied by a factor of modulus 1. For n even,
the whole of the calculation carried out in §4.8 can be analogously repeated.
122 §8 ASYMPTOTIC BEHAVIOUR

It follows, therefore, that here too the roots of the equation

<L„S-" + P.„,, ^ ... + #„/• = 0
transform into their reciprocals when the domain is changed from S0 to Si or
5 2„_!. For odd n, there are two classes of equations
0O “F ®is ~F • •• + @ms>n =
with the same roots in each case in each class. For S k with k odd, the roots are
f°r $k with k even, the roots are

5. Asymptotic Behaviour of the Eigenvalues

In questions concerning the asymptotic behaviour of the eigenvalues, the
equation
0O -F @is + ••• + = 0 (9 a)
for odd n, and the equation
e . ms - m + e_m+lS- m+1 + ... + emSm = o (9 b)

for even n, both play an important part; the numbers 6j here are the numbers
defined in the previous subsection 4.
If the boundary conditions are regular, these equations are respectively of
order m and 2m, and all their roots are different from zero.
In the following theorem we assume (though for brevity we do not mention
this again in the hypothesis) that the coefficients of the differential expression
considered are continuous matrix-functions in the interval [0,1]. It should,
however, be particularly mentioned that all the results of this subsection
would still hold if we required only that these coefficients be arbitrary, sum-
mable, matrix-functions in the interval [0,1].
theorem 2

Let L be a differential operator of the nth order, defined in the interval [0,1],
whose differential expression contains no derivative of the (n — 1)th order, and
whose boundary conditions are regular.
(1). If n is odd, then to each simple root corresponds a sequence A['}, and to
each simple root tfp corresponds a sequence Ajfj, of eigenvalues of the operator
L, and

A ® -O F 2W )P [1T ^ + 0 ^
k = N,(N + 1),..., (10 a)
»ff-<±2W r [ l ± " £ f + o ( f
where the upper or lower signs hold according as n = 4v + 1 or n = 4v — 1.
 §8.5 ASYMPTOTIC BEHAVIOUR OF EIGENVALUES 123
To an r-fold root or £(p , on the other hand, correspond in general r
sequences of eigenvalues X%] or Ag> which fo r large values of k satisfy the
following formulae:

n ln0 £ f
= (T 2 ^ 0 "[l T J- + 0
2kni 7 —7o>(7o + Ov>Oo Jr r — 1)>
n ln0 £ f k = N { N + 1),.... (11a)
Aff = ( ± 2 k n i) ^ l ±
2kni + o

Here it may be taken that = £{£ +1 = ... = | ^ + r l .

The same rule fo r upper or lower signs holds as for formulae (10a). These
eigenvalues X(f ) ,j = y 0,(y0 + 1),...,O'o + r — 1), may coincide fo r certain j and
k to give multiple eigenvalues.
(2). If n is even, then to each simple root £ of the equation (9 b) for the domain S0
corresponds a sequence Xk of eigenvalues of the operator L, and

Xk = { I k n if^ l T + o (i) j, k = N,(N + 1),..., (10 b)

where the upper or lower signs hold according as n = 4 v or n = 4v + 2.

To each multiple zero £ of equation (9 b), with multiplicity r, correspond r
sequences of eigenvalues Xkj of the operator L, and
n ln0 £
Xkj = (2 k m f 1T + 0 _ i ' , y = l,2,...,r ;k = N ,(N + l),..., (lib )
2kni k 1+1/r
and again the sign is — or + according as n = 4v or n = 4v + 2.
For sufficiently large |A|, the operator L has no other eigenvalues and the
multiplicity of any particular eigenvalue X is equal to the number of eigenvalues
from the above sequences which happen to coincide there, i.e. the multiplicity
is not greater than the multiplicity of the corresponding root £ of equation (9).
Proof. First let n be odd, say n = 2p — 1.
We consider a fixed domain T; let the numbers oi1,o)2,...,a)n be numbered in
sequence so that, for p s T ,

&((p + c M ) < &((p + c)co2) < ... < &((p + c)<Dn).

By applying formula (5) of subsection 2, and repeating the argument in the
proof of Theorem 2 in §4.9, we obtain
UfYj) = (PWj)kv[Av] for j < p,
U v( Yj ) = ( p c o j f f i e ^ W for j > p, ( 12)

Uv( Y,) = ( ^ { [ / g + e ^ [ B v]}. ,

124 §9 EXPANSION IN EIGENFUNCTIONS

We substitute all these expressions into the equation A = 0; it follows from

the definition (8 a) of the numbers 6k that, after removing the factors
pk\pk,,...,pkn,epoi>
l+ e1’™*, this equation talfes the form

K l + [0iK"P + + - + [ W ^ p = 0.
Hence, and by arguments similar to those used in proving Theorem 2 of §4.9,
follow all the assertions of the Theorem, for n odd.
The case of n even is treated in similar fashion.
By using Theorems 1 and 2 we can obtain asymptotic formulae for the
eigenfunctions, but we shall not go into this question in detail.

§9. Expansion in Terms of Eigenfunctions of a Differential Operator

1. The Case of a Self-Adjoint Differential Operator

Let L be a self-adjoint differential operator in the space of vector-functions. By
repeating the argument of §5.2, we obtain the following theorem.
THEOREM 1

Any vector-function in the domain of definition of a self-adjoint operator L can

be expanded in a uniformly convergent, generalized Fourier series in terms of
the eigenfunctions of L.
If is an orthonormal system of eigenfunctions of the operator
L, so that all eigenfunctions of L for arbitrary eigenvalues can be expressed as
linear combinations of these vectors, then the expansion referred to in Theorem
1 has the form
fix ) = £ cnj>n(x) (1)
n-1
with
Cn = J l (f(x^y»(x))dx- (2)

We recall once more that f(x ) and yn(x) are vector-functions; if, then, we put

fix) = {A(x)Mx), •••J mix)},

y*ix) = n = 1,2,3,...,

then the representation (1) is equivalent to the system of expansions in

uniformly convergent series
oo
fjix ) = Z CnynJ(x), j = l,2,...,m,
n= 1

the same coefficients cn occurring in each of the series.

 §9.2 REGULAR BOUNDARY CONDITIONS 125
It follows from Theorem 1 that the functions form a
complete orthonormal system in the space I}m(a,b) of all measurable vector
functions f(x ) which satisfy the condition

J* \Kx)\*dx < + oo.

[|/(*)l2 means here the scalar product of the vector f(x) with itself.]
It follows that

(3)
with
J* (f(x),yn(x))dx.

2. Expansion in terms of the Eigenfunctions of a Differential Operator Generated

by Regular Boundary Conditions
The considerations in §9.1 are not applicable if L is not a self-adjoint operator.
So we now make use of a different method which depends on the analytic
properties of the Green’s function for the operator L — XI and on the
asymptotic formula obtained in §8. Here we can assume that the boundary
conditions which generate L are regular; further, without loss of generality,
we may assume that the boundary-value problem Ly = 0 has only the trivial
solution. Then L has a Green’s function G(x,f). As in §5.3, we consider a
sequence of circles I \ , k — 1,2,..., with centre at the origin, and having the
following properties:
1°. The radius Rk of the circle Tk increases without limit as k oo.
2°. There is a positive number S such that all eigenvalues of the operator L
have a distance >S from each of the circles Tk.
By virtue of the asymptotic properties of these eigenvalues proved in §8.5,
such circles do exist.
Let (J(x,£,A) be the Green’s function for the operator L — Al, and, in
particular, let G(xfi,0) = G(x,£) be the Green’s function for the operator L.
We consider the integral

(4)

LEMMA

On the circles Tk the function G(x,g, A) satisfies the inequality

IM *,£A )I (5)

where M is a fixed constant, uniformly fo r k = 1,2,3,... and a < x,f < b.

126 §9 EXPANSION IN EIGENFUNCTIONS

Proof. If we put A = — then, for a suitable choice of arg p, the circle I \ is

transformed into a circular arc yk subtending an angle 2ir/n at the centre, and
yk goes through two neighbouring domains Sk,S2 of the complex />-plane. We
want to estimate the value of the function G(x,£,A) on the arc yk.
Let the numbers ai1,a)2>...,cDn be numbered in sequence so that, for p e Sk,

& ( pwi ) < 3$(paj2) < ... < {%(pwn).

We use the formula for G(x,£,X) from §7.2, and substitute therein the asymptotic
expressions for Yj(x,A) found in §8.2. We get

z ,( (6 )
ftp
and so, by (4) in §7.2,

*w ) = t (i (7)

where the — sign is to be taken if £ < x, and the + sign if £ > x.

For the rest of the proof it is best to deal with n odd and n even separately.
First let n be odd; n = 2p — 1. Then, for p e S1} we have

&(pu>i) < 0 ,..., 1) < 0 ,

&(p°>n+1) > 0 ,..., &(pw„) > 0.

We denote by y'k that part of the arcyfc which lies in Sk and on which ^(pw f}
> 0, and by yk the part of yk which lies in Sk and on which 2%{pwf) < 0. We
want to estimate the value of the function G(x,£,X) on the arc yk. By using
formula (12) of §8.5, it can easily be shown that, on yk,

Y jix W j, = o ( - j-) . (9)

To obtain an estimate of G(x,£,A) ony*, we first transform the expression for

G(x,£,A) in formula (5) of §7.2. By virtue of formula (4) in §7.2, for g(x,£) we
have the equations

*<o,f) =-i a=l

i r,(0)z.(f). s(i.f) =i ai=l y.<i)z.(0;
consequently
n
V*(g) = i I [tfv iW - Uv0(Y a)]Za(£).
a=1
 §9.2 REGULAR BOUNDARY CONDITIONS 127
We write this expression in the form

- i £ I - u , ( y .) + lu ^ Y jjz x t) + i £ [ v ,( r a) - 2 t / , 1(K j]z .(o

a=1 a=fi +1

= - i t u v( r a) z a(f) + t u vl(Y a) z a(o

ol = 1 a= 1

+ i £ u ,( r .) z ,( () - £ u j . r j z j . 0 . (io)
a= /i + 1 a -f 1

By (6), we have, however, for a < fx.

U A Y .)Z ,(0 - -

hence, and by (8), it follows that

Uvl(Y ,) Z .m = ° ( - i r i r 1).
Similarly, for a > n, we get

U ^ Y ,)Z .(() = 0 ( ^ v- ,)•

So (10) can be written in the form

% ) = -! I U £Ya)ZJ® + \ t UAYa) Z ^ ) + o ( n\ —\ . (11)

a= 1 a +1 \P/
If this representation of Uv(g) is now substituted into formula (5), §7.2, for
CQc,|,A), we obtain

<7(*,!,A) = g(x,£. A) + i t Yj(x)Zj(0 - i t YjM ZjiO + o f-jJn V (12)

7-1 7 = M +1 V /

since, by definition of the inverse matrix,

£ wp u , m = S/.i,
V»1

and further, by (9),

Combining (12) with (4) from §7.2, we arrive at the formulae

£ r ,( x ) z /f ) + o ( 4 n ) for ( < x,
j-i \p !
G(x,£,\) = (13)
- £ r/* ) Z /{ ) + for { > x.
^ j - H+1 \p ]
128 §9 EXPANSION IN EIGENFUNCTIONS

By (8) and the asymptotic formulae for Yj(x) and Zj(g), it follows that on the
arc y'k for £ < x,

|/ M f ) = - -in - o ( - r ,) ,

and for £ > x,

£ l '/ x j z / f ) = - - L , i w 1^ - ° = o ( — i ).
j-H +1 nP j-fi +l \r /
In conjunction with (13) this gives the desired relation

G(x,£,X) = o ( — ^
on the arc y'k.
Similarly we can verify this relation on the arc y'k, and on that part of the
arc yk which lies in the domain S2.
This proves the lemma for the case of n odd.
For n even, the proof runs on exactly similar lines.
Using these Lemmata, the following theorems may be deduced exactly as
in §5.3:
THEOREM 2
The Green's function G(x,£) of a differential operator which is generated by
regular boundary conditions can be represented as a series

G{x,£) = -
„ -i K
which, at least for a certain definite order of its terms, converges uniformly.
Here H v(x,g) denotes the residue of the function G(x,^,A) at the pole Av.
THEOREM 3
I f all the eigenvalues of an operator L generated by regular boundary conditions
are simple zeros of the function A(A), then its Green's function G(x,£) can be
represented as a series
G(x,g) =
v-1 \

which, at least for a certain definite order of its terms, converges uniformly.
Here >^v(x) and z f x ) are the eigenfunctions, of the operators L and L*
respectively, associated with the eigenvalues Av and I v; and those eigenfunctions
may both be so normalized that

(yy{x),zfx))dx = 1.
 §9.2 REGULAR BOUNDARY CONDITIONS 129
THEOREM 4
Let L be an operator which is generated by regular conditions and of which all
the eigenvalues are simple zeros of the function A. Then any function f(x) in the
domain of definition of the operator L can be expanded as a series in eigen­
functions of L:

V= 1

The series converges uniformly, at least for one particular order of its terms.
Here y v(x) and zv(„x) are eigenfunctions of L and L*,for the eigenvalues Av and
Av respectively, these eigenfunctions may be normalized so that

/ ! O’v W ^ v W )^ = l-
For simplicity we have restricted ourselves to the ordinary eigenvalue-
problem and have not discussed the generalized problem Ly = AMy at all, nor
the problem Ly = A5(x)y in particular. For the latter problem, if certain
restrictions are imposed on B(x), it can be assumed that B(x) is a diagonal
matrix which h a sp1(x),p2(x),...,iJin(x) as its diagonal elements, th e pv(x)being
continuous and all different (see §6.8). We can obtain, for the solution of the
corresponding differential equation and for the eigenvalues, certain asymptotic
formula in which

takes over the role played by exp {p^j(x — a)} previously. These formulae can
be used to obtain expansions in terms of eigenfunctions. It may be mentioned
that no one has investigated up to now either this case or the general case
Ly = AMy in any detail (except for the special cases n = 1 or m = 1).
 BIBLIOGRAPHY
1. N. I. and G l a z m a n , T. M. Theory o f linear operators in Hilbert space,
A k h ie z e r ,
Eng. trans., Ungar, N.Y., 1961.
2. A m b a r t s u m y a n , V . A. ‘On a question in eigenvalue theory,’ Z. f. Physik, 53,
690-5 (1929). (Ger.)
3. B a r i, H. K. ‘Bi-orthogonal systems and bases in Hilbert spaces,’ Uchen. Zap.
M.G.U. 148, 4 (1951), 68-107. (Russ.)
4 . B e r e z a n s k ii , Y u. M.
(a) ‘Expansion in eigenfunctions of partial difference and partial differential
equations,’ Trudy 3 Vse-Soyuz. matem. S ’ezda, 2, 36-7 (1956), (Russ.)
(b) Expansion in eigenfunctions o f self-adjoint operators, 2nd ed., Izd. “Naukova
dumka” , 1966. (Russ.)
5. B i r k h o f f , G. D.
(a) ‘On the asymptotic character of the solutions of certain linear differential
equations containing a parameter,’ Trans. Amer. Math. Soc., 9, 219-23 (1908).
(b) ‘Boundary-value problems and expansion problems of ordinary linear differ­
ential equations,’ Trans. Amer. Math. Soc., 9, 373-95 (1908).
(c) ‘Note on the expansion problems of ordinary linear differential equations,’
Rendiconti Palermo, 36, 115-26 (1913).
(d) ‘Quantum mechanics and asymptotic series,’ Bull. Amer. Math. Soc., 39,
681-700 (1933).
6. B i r k h o f f , G. D. and L a n g e r , R. E. ‘The boundary problems and developments
associated with a system of ordinary linear differential equations of the first
order,’ Proc. Amer. Acad. Arts & Sci., 58, 51-128 (1923).
7. B ir m a n , M. S h . ‘On the spectrum of singular boundary-value problems,’ Mat.
Sb., 55(97) (1961), 125-74. (Russ.)
8. B l i s s , G. U. ‘A boundary-value problem for a system of ordinary linear differ­
ential equations of the first order,’ Trans. Amer. Math. Soc., 28, 561-84 (1926).
9. B l o k h , A . S h . ‘On th e d e te r m in a tio n o f a d ifferen tia l e q u a tio n b y its sp ectra l
m a tr ix f u n c t io n ,’ Doklad. Akad. Nauk, SSSR, 92, 209-12 (1953).
10. B l o k h i n t s e v , D. I. Foundations o f Quantum Mechanics, GTTI, 1950. (Russ.).
Ger. trans., Deutschen Verl. d. Wiss., Berlin, 1953.
11. B o c f i e r , M. Lessons on Sturm’s methods, Paris, 1917. (French)
12. B o r g , G.
(a) ‘An inversion of the Sturm-Liouville eigenvalue problem. Determination of
the differential equation from the eigenvalues,’ Acta Math., 78, 1-96 (1946).
(Ger.)
(b) ‘Inverse problems in the theory of characteristic values of differential sys­
tems,’ C.R.10 Congr£s Math. Scandinaves, 1946, Copenhagen 1947, 172-80.
(c) ‘On the point spectra of y" + (f- q (x j)y = 0,’ Amer. J. Math., 12>, 122-6
(1951).
13. B u n i t s k i i , E.
(a) ‘On Green’s function for linear ordinary differential equations,’ J. de Math.,
5(6), 65-125 (1909). (French)
131
132 BIBLIOGRAPHY

(b) ‘On the theory of Green’s function for ordinary linear differential equations,’
Odessa 1913. (Russ.)
14. C e s a r i, L. ‘Asymptotic behaviour and stability problems in ordinary differ­
ential equations, 1959.’ Springer, Berlin/Heldelberg/N.Y.
15. C h u d o v , L. A. ‘The inverse Sturm-Liouville problem,’ Mat. Sb., 25(67), 451-6
(1949). (Russ.)
16. C o d d i n g t o n , E. A. ‘The spectral representation of ordinary self-adjoint differ­
ential operators,’ Annals o f Math., 60, 191-211, (1954).
17. C o d d i n g t o n , E. A. and L e v in s o n , N. Theory o f ordinary differential equations,
McGraw-Hill, N.Y., 1955.
18. C o l l a t z , L. Eigenvalue problems with engineering applications, Leipzig, 1949
(reprint 1962). (Ger.)
19. C o u r a n t , R. and H i l b e r t , D. Methods o f mathematical physics, Interscience
Publishers, N.Y., 1953.
20 . D e l a V a l l £ e P o u s s in , Course of infinitesimal analysis, Paris, 1937 (Eng.
trans., Dover Books, N.Y., 1946). (French)
21. D u n f o r d , N. ‘A survey of the theory of spectral operators,’ Bull. Amer. Math.
Soc., 64, 217-74 (1958).
22. D u n f o r d , N. and S c h w a r t z , J. T. Linear operators; Part /, general theory;
Part II, spectral theory, Interscience Publishers, N.Y., 1964.
23. D z h a v a d o v , M. C.
(a) ‘On the completeness of a certain part of the eigenfunctions of a non-self-
adjoint differential operator,’ Doklad. Akad. Nauk SSSR, 195, 4, 723-5 (1964).
(Russ.)
(b) ‘On the n-fold completeness of half the eigenfunctions and associated func­
tions of an ordinary, 2m-th order, differential operator,’ Doklad. Akad. Nauk
SSSR , 160, 4, 754-7 (1965). (Russ.)
24. E b e r h a r d , W .
(a) ‘The asymptotic behaviour of the Green’s function of irregular eigenvalue
problems with separable boundary-conditions, Math. Zeit., 86, 45-48 (1964).
(Ger.)
(b) ‘Eigenfunction expansions of irregular eigenvalue problems with separable
boundary-conditions,’ Math. Zeit., 86, 205-17 (1964). (Ger.)
(c) ‘Eigenfunction expansions of irregular eigenvalue problems with separable
boundary-conditions, II,’ Math. Zeit., 90, 126-32 (1965). (Ger.)
(d) ‘Linear eigenvalue problems of arbitrary order with separable boundary
conditions,’ Ber. Kernforschungsanlage Jiilich (1964), 176, 925. (Ger.)
25. E l l i o t , J o a n n e . ‘Eigenfunction expansions associated with singular differential
operators,’ Trans. Amer. Math. Soc., 78, 406-25 (1955).
26. F a g e , M. K. ‘Operator-analytical functions,’ Trudy Mosk. Mat. Obshch., 7,
227-268 (1958).
27. F r i e d r i c h s , K. O.
(a) ‘Criteria for the discrete character of the spectra of ordinary differential
operators,’ Studies and essays presented to R. Courant on his 60th birthday,
Jan. 8, 1948, 145-60.
(b) ‘On the perturbation of continuous spectra,’ Comm, on Appl. Maths., 1, 361-
406 (1948).
28. G e l f a n d , I. M.
(a) Lectures on linear algebra, GTTI 1950. (Russ.)
(b) ‘Eigenfunction expansions for an equation with periodic coefficients,’
Doklad. Akad. Nauk SSSR, 73, 1117-20 (1950). (Russ.)
 BIBLIOGRAPHY 133
(c) ‘Remarks on N. K. Bari’s article’ (Ref. 3 above). Uchen. Zap. M.G.U., No.
148, 4, 224-225 (1951). (Russ.)
29. Gelfand, I. M. and Kostyuchenko, A. G. ‘Eigenfunction expansions of differ­
ential and other operators, Doklad. Akad. Nauk SSSR, 103, 349-52 (1955).
(Russ.)
30. Gelfand, I. M. and Levitan, B. M.
(a) ‘On the determination of a differential equation from its spectral function,’
Izv, Akad. Nauk SSSR, ser. math., 15, 309-60 (1951). (Russ.)
(b) ‘On a simple identity for eigenvalues of a 2nd-order differential operator,’
Doklad. Akad. Nauk. SSSR, 88, 593-6 (1953). (Russ.)
31. Gelfand, I. M. and Shapiro, S. Ya. ‘Representations of the rotation group
in 3-dimensional space and their applications,’ Uspekhi Mat. Nauk, 7, 1(47),
3-117 (1952). (Russ.)
32. Glazman, I. M.
(a) ‘On the deficiency indices of differential operators,’ Doklad. Akad. Nauk
SSSR, 64, 151-4 (1949). (Russ.)
(b) ‘On the theory of singular differential operators,’ Uspekhi Mat. Nauk, 5,6(40),
102-35 (1950). (Russ.)
(c) ‘On the spectrum of linear differential operators,’ Doklad. Akad. Nauk SSSR,
80, 153-6 (1951). (Russ.)
(d) Direct methods for the qualitative spectral analysis o f singular differential
operators, Moscow, Fizmatgiz, 1963. (Russ.)
33. Gokhberg, I. Ts. and Krein, M. G. Introduction to the theory o f linear non-
selfadjoint operators in Hilbert space, Izd. “ Nauka” , Moscow 1965. (Russ.)
34. Goursat, E. Course o f mathematical analysis, Paris, 1942 (Eng. trans., Dover
Books N.Y.). (French)
35. Graff, A. A. ‘On the theory of linear differential operators, Part I, Mat. Sb.
18(60), 305-28 (1946); Part II, Mat. Sb. 21(63), 143-59 (1947). (Russ.)
36. Halperin, I. ‘Closures and adjoints of linear differential operators,’ Annals o f
Math., 38, 880-919 (1937).
37. Hartmann, P.
(a) ‘The L2-solutions of linear differential equations of second order,’ Duke
Math. J., 14, 323-6 (1947).
(b) ‘On differential equations with non-oscillatory eigenfunctions,’ Duke Math.
J., 15, 697-709 (1948).
(c) ‘On a theorem of Milloux,’ Amer. J. o f Math., 70, 295-308 (1948).
(d) ‘Unrestricted solution fields of almost-separable differential equations,’
Trans. Amer. Math. Soc., 63, 560-80 (1948).
(e) ‘On the spectra of slightly disturbed linear oscillators,’ Amer. J. o f Math., 71
71-9 (1949).
(f) ‘On the linear logarithmico-exponential differential equations of the second
order,’ Amer. J. o f Math., 71, 764-79 (1949).
(g) ‘A characterization of the spectra of /-dimensional wave equations,’ Amer.
J. o f Math., 71, 915-20 (1949).
(h) ‘The number of L2-solutions of x" + q(t)x = 0,’ Amer. J. o f Math., 73, 635-
45 (1951).
(i) ‘On bounded Green’s kernels for second-order linear ordinary differential
equations,’ Amer. J. o f Math., 73, 646-56 (1951).
(j) ‘On the eigenvalues of differential equations,’ Amer. J. o f Math., 73, 657-62
(1951).
 134 BIBLIOGRAPHY

(k) ‘Some examples in the theory of singular boundary-value problems,’ Amer.

J. o f Math., 74, 107-25 (1952).
(l) ‘On non-oscillatory linear differential equations of second order,’ Amer. J. o f
Math., 74, 389-400 (1952).
(m) ‘On boundary-value problems for systems of ordinary non-linear, second-
order differential equations,’ Trans. Amer. Math. Soc., 3, 96, 493-509 (1960).
(n) Ordinary differential equations, Wiley, N.Y., 1964.
38. Hartman, P. and Putnam, C. R.
(a) ‘The least cluster point of the spectrum of boundary problems,’ Amer. J. o f
Math., 70, 849-55 (1948).
(b) ‘The gaps in the essential spectra of wave equations,’ Amer. J. o f Math., 72,
849-62 (1950).
39. Hartman, P. and Wintner, A.
(a) ‘An oscillation theorem for continuous spectrum,’ Proc. Nat. Acad. Sci., 33,
376-9 (1947).
(b) ‘The asymptotic arcus variation of solutions of real linear differential equa­
tions of second order,’ Amer. J. o f Math., 70, 1-10 (1948).
(c) ‘Criteria of non-degeneracy for the wave equations,’ Amer. J. o f Math., 70,
295-308 (1948).
(d) ‘On the orientation of unilateral spectra,’ Amer. J. o f Math., 70, 309-16
(1948) .
(e) ‘On the asymptotic problems of the zeros in wave mechanics,’ Amer. J. o f
Math., 70, 461-80 (1948).
(f) ‘On non-conservative linear oscillators of low frequency,’ Amer. J. o f Math.,
70, 529-39 (1948).
(g) ‘A criterion for the non-degeneracy of the wave equation,’ Amer. J. o f
Math., 71, 206-13 (1949).
(h) ‘On the location of spectra of wave equations,’ Amer. J. o f Math., 71, 214-17
(1949) .
(i) ‘Oscillatory and non-oscillatory linear differential equations,’ Amer. J. o f
Math., 71, 627-49 (1949).
(j) ‘A separation theorem for continuous spectra,’ Amer. J. o f Math., 71, 650-62
(1949).
(k) ‘On the essential spectra of singular eigenvalue problems,’ Amer. J. o f Math.,
72, 545-52 (1950).
(l) ‘On perturbations of the continuous spectrum of the harmonic oscillator,’
Amer. J. o f Math., 79-85 (1952).
40. Hausdorff, F. Set theory, W. de Gruyter, Berlin, 1927. (Eng. trans. J. R.
Aumann, N.Y., 1957.)
41. Hellwig, G. Differential operators o f mathematical physics. Springer 1964
Berlin/Heidelberg/N.Y. (Ger.)
42. Heywood, P. ‘On the asymptotic distribution of eigenvalues,’ Proc. Lond. Math.
Soc., 4, 456-70 (1954).
43. Hopkins, J. W. ‘Some convergent developments associated with irregular
boundary conditions,’ Trans. Amer. Math. Soc., 20, 245-59 (1919).
44. Jackson, D.
(a) ‘Expansion problems with irregular boundary conditions,’ Proc. Amer. Acad.,
51, 383-413 (1916).
(b) ‘Algebraic properties of self-adjoint systems,’ Trans. Amer. Math. Soc., 17,
418-424 (1916).
 BIBLIOGRAPHY 135
45. K a m k e , E.
(a) Differential equations. Methods o f solution and solutions; Vol. I, ordinary
differential equations. 4th ed., Akad. Verlag, Geest & Portig, Leipzig 1951.
(Ger.)
(b) ‘On the expansion theorem for polar eigenvalue problems.’ Math. Zeit., 45,
706-18 (1939). (Ger.)
(c) ‘On definite self-adjoint eigenvalue problems for ordinary differential equa­
tions; I,’ Math. Zeit., 45, 759-87 (1939); II, Math. Zeit., 46, 231-50 (1940);
III, Math. Zeit., 46, 251-86 (1940). (Ger.)
46. K a t s , I.
(a) ‘On Hilbert spaces generated by monotonic matrix-functions,’ Zapiski
Nauchno-Issled. Inst. Mat. i Mekh., Kharkov. Mat. Obshch., 22, 95-114 (1950).
(Russ.)
(b) ‘Two general theorems on the asymptotic behaviour of the spectral functions
of 2nd-order differential systems,’ Izv. Akad. Nauk SSSR, ser. Math., 26 (1962),
53-78. (Russ.)
(c) ‘Behaviour of spectral functions for differential systems with boundary con­
ditions at a singular end-point,’ Doklad. Akad. Nauk SSSR, 1,157, 34-7 (1964).
(Russ.)
(d) ‘The existence of spectral functions of generalized differential systems of the
second order with boundary conditions at a singular end-point,’ Mat. Sb., 2,
28(110), 174-227, (1965). (Russ.)
4 7 . K e l d y s h , M . V . ‘On eigenvalues and eigenfunctions of some classes of non-
selfadjoint equations,’ Doklad. Akad. Nauk SSSR, 77, 11-14 (1951). (Russ.)
48. K e l d y s h , M . V. and L id s k ii, V. B. ‘Topics in the spectral theory of non-
selfadjoint operators’, Trudy Vse Soyuz. Mat. S'ezda, 1, 1963, 101-20.
49. K e s e lm a n , G. M. ‘On the absolute convergence of expansions in eigenfunctions
of certain differential operators,’ Izv. Vuz SSSR, Mat., 2, 1964, 82-93. (Russ.)
50. K h r o m o v , A. P.
(a) ‘Eigenfunction expansions for ordinary differential operators in a finite
interval,’ Doklad. Akad. Nauk SSSR, 146, 1294-1297, (1962). (Russ.)
(b) ‘Eigenfunction expansions for ordinary differential operators in a finite inter­
val,’ Trudy 2 Nauch. Konf. Mat. Kafedr Pedinst. Povolzhya, Kuibyshev, 1962,
109-113. (Russ.)
(c) ‘Eigenfunction expansions for ordinary differential operators with irregular
separable boundary-conditions,’ Doklad. Akad. Nauk SSSR, 152, 6, 1324-
1326(1963). (Russ.)
(d) ‘Eigenfunction expansions for ordinary linear differential operators with non­
regular separable boundary conditions,’ Mat. Sb. 3, 70(112) (1966), 310-29.
(Russ.)
51. K o d a i r a , K.
(a) ‘Eigenvalue problems for ordinary differential equations of the second order
and Heisenberg’s theory of S-matrices,’ Anier. J. o f Math., 71, 921-45 (1949).
(b) ‘On ordinary differential equations of any even order and the corresponding
eigenfunction expansions,’ Amer. J. o f Math., 72, 502-44 (1950).
52. K r a m e r , H. P. ‘Perturbations of differential operators,’ Pacific J. o f Math., 1,
1405-1443 (1957).
53. K r e i n , M. G.
(a) ‘On a general method of resolution of positive-definite kernels into elementary
products,’ Doklad. Akad. Nauk. SSSR, 53, 3-6 (1946). (Russ.)
136 BIBLIOGRAPHY

(b) ‘On Hermitian operators with direction functionals,’ Zbirnik Prats Inst.
Mat. Akad. Nauk URSR, 10, 83-105 (1948). (Ukr.)
(c) ‘Theory of self-adjoint extensions of semi-bounded Hermitian operators and
its applications: I,’ Mat. Sb. 20(62), 431-95 (1947); II, Mat. Sb., 21(63),
365-404 (1947). (Russ.)
(d) ‘Foundations of the theory of representations of Hermitian operators with
deficiency-indices (m, m ),’ Ukr. Mat. Zh., 2, 3-66 (1949). (Russ.)
(e) ‘On the Sturm-Liouville boundary-value problem in the interval [0,co) and
on one class of integral equations,’ Doklad. Akad. Nauk SSSR, 73, 1125-8
(1950). (Russ.)
(f) ‘On a one-dimensional singular boundary-value problem of even order in the
interval [0,co),’ Doklad. Akad. Nauk SSSR , 74, 9-12 (1950). (Russ.)
(g) ‘Solution of the inverse Sturm-Liouville problem,’ Doklad. Akad. Nauk SSSR,
76, 315-18(1951). (Russ.)
(h) ‘Determination of the density of an inhomogeneous symmetrical string from
its frequency spectrum,’ Doklad. Akad. Nauk. SSSR, 76, 315-18 (1951) (sic).
(Russ.)
(i) ‘On the inverse problem for an inhomogeneous string,’ Doklad. Akad. Nauk
SSSR,' 82, 669-72 (1952). (Russ.)
(j) ‘On a generalization of Stieltje’s investigations,’ Doklad. Akad. Nauk SSSR,
87, 881-4(1952). (Russ.)
(k) ‘Analogues of the Chebyshev-Markov inequalities in a one-dimensional
boundary-value problem,’ Doklad. Akad. Nauk SSSR, 87, 5-8 (1953). (Russ.)
(l) ‘On the transition function of a one-dimensional boundary-value problem of
the second order,’ Doklad. Akad. Nauk SSSR, 88, 405-8 (1953). (Russ.)
(m) ‘On some cases of effective determination of the density of an inhomogeneous
string from its spectral function,’ Doklad. Akad. Nauk SSSR, 93, 617-20
(1953). (Russ.)
(n) ‘On inversion problems in the theory of filters and the A-zone of stability,’
Doklad. Akad. Nauk SSSR, 93, 767-70 (1953). (Russ.)
(o) ‘On a method of effective solution of the inverse of a boundary-value prob­
lem,’ Doklad. Akad. Nauk SSSR, 94, 987-90 (1953). (Russ.)
54. K r e in , M. G. and K r a s n o s e l s k i i , M.A. ‘Fundamental theorems on the exten­
sion of Hermitian operators,’ Uspekhi Mat. Nauk, 2 , 3 (1 9 ), 6 0 - 1 1 7 (1 9 4 7 ).
55. Krein, M. G., Krasnolselskii, M. A. and Milman, D. P. ‘On the deficiency
indices of operators in Banach space and on some geometrical questions,’ Sb.
Trud. Inst. Mat. Akad. Nauk, Ukr. SSR, 11, 97-112 (1948). (Russ.)
56. Landau, L. D. and Lifschitz, V. I. Quantum Mechanics, GTTI, 1948. (Eng.
trns, Pergamon Press.)
57. Langer, R.
(a) ‘On the asymptotic solutions of ordinary differential equations, with an
application to the Bessel functions of large order,’ Trans. Amer. Math. Soc., 33,
23-64(1931).
(b) ‘On the asymptotic solutions of differential equations, with an application to
the Bessel functions of large complex order,’ Trans. Amer. Math. Soc., 34, 447-
80 (1932).
(c) ‘The asymptotic solutions of certain linear ordinary differential equations of
the second order,’ Trans. Amer. Math. Soc., 36, 90-106 (1934).
(d) ‘The asymptotic solutions of ordinary linear differential equations of the
second order, with special reference to the Stokes phenomenon,’ Bull. Amer.
Math. Soc., 40, 545-82 (1934).
 BIBLIOGRAPHY 137
(e) On the asymptotic solutions of ordinary differential equations with reference
to the Stokes phenomenon about a singular point,’ Trans. Amer. Math. Soc.,
37, 397-416 (1935).
(f) ‘On the connection formulae and the solutions of the wave equation,’ Phys.
Rev., 51, 669-76 (1937).
(g) The boundary problem of an ordinary linear differential system in the com­
plex domain, Trans. Amer. Math. Soc., 46, 151-190 (1939), correction p. 467.
(h) ‘A theory for ordinary differential boundary problems of the second order
and of the highly irregular type,’ Trans. Amer. Math. Soc., 58, 292-361 (1943).
(i) ‘The asymptotic solutions of ordinary linear differential equations of the
second order, with special reference to a turning point,’ Trans. Amer. Math. Soc.,
67, 461-90(1949).
(j) ‘On the wave equation with small quantum numbers,’ Phys. Rev., 72, 1573-8
(1949).
58. L e a v i t t , G. W . ‘On systems of linear differential equations,’ Amer. J. o f Math.,
73, 690-6 (1951).
59. L e ib e n z o n , Z. L.
(a) ‘The uniqueness of solution of the inverse problem for ordinary differential
operators of order n > 2, and transformations of such operators,’ Doklad.
Akad. Nauk SSSR, 3, 142, 534-7 (1962). (Russ.)
(b) ‘The connection between the inverse problem and the completeness of the
eigenfunctions,’ Doklad. Akad. Nauk SSSR, 3, 145, 519-22 (1962). (Russ.)
60. L e ig h t o n , W .
(a) ‘On self-adjoint differential equations of second order,’ Proc. Nat. Acad, o f
Sci., 35, 656-7 (1949).
(b) ‘The detection of oscillation of solutions of a 2nd-order linear differential
equation,’ Duke Math. J., 17, 57-62 (1950).
61. L e v in , B. Ya. ‘Fourier and Laplace transformations by means of solutions of
2nd-order differential equations,’ Doklad. Akad. Nauk SSSR, 106, 187-90
(1956). (Russ.)
6 2 . L e v in s o n , N .
(a) ‘Criteria for the limit-point case for 2nd-order linear differential operators,’
Chasopis Pchst. Math. Fys., 74, 17-20 (1949).
(b) ‘The inverse Sturm-Liouville problem,’ Math. Tidsskr. B., 25-30 (1949).
63. L e v i t a n , B. M .
(a) Expansion in eigenfunctions for 2nd-order differential equations, GTTI, 1950.
(Russ.)
(b) ‘On the theorem of expansions in eigenfunctions for 2nd-order differential
equations,’ Doklad. Akad. Nauk SSSR, 71, 605-8 (1950). (Russ.)
(c) ‘Proof of a theorem on expansions in eigenfunctions for selfadjoint differ­
ential equations,’ Doklad. Akad. Nauk SSSR, 73, 651-4 (1950). (Russ.)
(d) ‘On a uniqueness theorem,’ Doklad. Akad. Nauk SSSR,16,485-8 (1951). (Russ.)
(e) ‘On the asymptotic behaviour of the spectral function and an expansion in
eigenfunctions for a 2nd-order differential equation,’ Izv. Akad. Nauk SSSR,
ser. Math., 19, 33-58 (1955). (Russ.)
(f) Operators o f a generalized displacement and some applications o f them, Moscow
1962.
64. Lidskii, V. B.
(a) ‘On the number of quadratically integrable solutions of the system of differ­
ential equations y" + P{t)y — Av,' Doklad. Akad. Nauk SSSR, 95, 217—20
(1954). (Russ.)
 138 BIBLIOGRAPHY

(b) ‘On the completeness of a system of eigenfunctions and associated functions

for a non-selfadjoint differential operator,’ Doklad. Akad. Nauk SSSR, 2, 110,
172-5(1956). (Russ.)
(c) ‘On the completeness of a system of eigenfunctions and associated elements
for a completely continuous operator,’ Doklad. Akad. Nauk, 2, 115, 234-6,
(1957). (Russ.)
(d) ‘Theorem on the completeness of a system of eigenfunctions and associated
elements for an operator with a discrete spectrum,’ Doklad. Akad. Nauk SSSR,
6, 119, 1088-91 (1958). (Russ.)
65. L i f s c h i t z , I. M. ‘On the theory of regular perturbations,’ Doklad. Akad. Nauk.
SSSR, 48, 83-6 (1945). (Russ.)
66. L i f s c h i t z , M. S. On the spectral representation of linear non-selfadjoint
operators,’ Mat. Sb., 34(76), 145-99 (1954). (Russ.)
67. L in d P o t t e r , R u t h . ‘On self-adjoint 2nd-order differential equations,’ Pacific
J. Math., 3, 467-91 (1953).
68. L y a n t s e , V. E. ‘On a differential equation with spectral singularities,’ Mat. Sb.,
4, 64(106), 520-61 (1964), and 1, 65(107), 47-103 (1965). (Russ.)
69. L y u s t e r n i k , L. A. and S o b o le v , V. I. Elements o f functional analysis, GTTI,
1951. (Eng. trans., Ungar, N.Y., 1961).
70. M a l t s e v , A. I. Foundations o f linear algebra, GTTI, 1948. (Russ.)
71. M a r c h e n k o , V. A.
(a) ‘Some questions in the theory of a 2nd-order differential operator,’ Doklad.
Akad. Nauk SSSR, 72, 457-60 (1950). (Russ.)
(b) ‘Some questions in the theory of /-dimensional linear differential equations
of the second order,’ Trudy Moskov. Mat. Obshch., 2, 3-83 (1953). (Russ.)
72. M a r k u s , A. S.
(a) ‘On an expansion in root vectors of a weakly perturbed self-adjoint operator,’
Doklad. Akad. Nauk SSSR, 3, 142, 538^11 (1962). (Russ.)
(b) ‘Some criteria for completeness of root vectors of a linear operator and on
the summability of series in this system,’ Doklad. Akad. Nauk SSSR, 4, 155,
753-6 (1964). (Russ.)
73. Mathematics in the USSR during the last 40 years, J, 763-72, Moscow 1959.
(Russ.)
74. M a u t n e r , F. I. ‘On eigenfunction expansions,’ Proc. Nat. Acad. Sci., USA, 39,
49-53 (1953).
75. M i k h a i l o v , V. P. ‘On Riesz bases in L 2 (0, 1), Doklad. Akad. Nauk SSSR, 25,
144, 981-84 (1962). (Russ.)
76. M i l l e r , K. Linear Differential equations, Routledge & Kegan Paul Ltd., Lon.
1964.
77 . M i l n e , W . E . ‘O n th e d eg re e o f c o n v e r g e n c e o f e x p a n s io n s in a n in fin ite in te r ­
v a l,’ Trans. Amer. Math. Soc., 3 1 , 9 0 6 - 1 8 (1 9 2 9 ).
78. M o l c h a n o v , M. A. ‘On conditions for discreteness of the spectrum of a self-
adjoint differential equation of the second order,’ Trudy Moskov. Mat. Obshch.,
2, 169-200(1953). (Russ.)
79. M o s e r , J u r g e n .
(a) ‘Perturbation theory of the continuous spectrum for ordinary 2nd-order
differential equations,’ Math. Ann., 125, 366-93 (1953). (Ger.)
(b) ‘Singular perturbations of eigenvalue problems for linear differential equa­
tions of even order,’ Commun. Pure & Appl. Math., 8, 251-78 (1955).
80. N a g y , B e l a de S z . ‘Perturbations of selfadjoint transformations in Hilbert
space,’ Comment. Math. Helv., 19, 347-66 (1947). (French)
 BIBLIOGRAPHY 139
81. Naimark, M. A.
(a) ‘On self-adjoint extensions of the second sort of a symmetric operator,’
Izv. Akad. Nauk SSSR, ser. Mat., 4, 53-104 (1940). (Russ.)
(b) ‘Spectral functions of a symmetric operator,’ Izv. Akad. Nauk SSSR., ser.
Mat., 1, 277-318 (1940). (Russ.)
(c) ‘On spectral functions of a symmetric operator,’ Izv. Akad. Nauk SSSR, ser.
Mat., 7, 285-96 (1943). (Russ.)
(d) ‘On the deficiency indices of linear differential operators,’ Doklad. Akad. Nauk
SSSR., 82, 517-20 (1952). (Russ.)
(e) ‘On the spectrum of singular non-selfadjoint differential operators of the
2nd-order,’ Doklad. Akad. Nauk SSSR, 85, 41-4 (1952). (Russ.)
(f) ‘On expansions in eigenfunctions for non-selfadjoint singular differential
operators of the 2nd order,’ Doklad. Akad. Nauk SSSR 89, 213-16 (1953).
(Russ.)
(g) ‘Investigation of the spectrum and expansion in eigenfunctions of a non-
selfadjoint differential operator of the 2nd order on the half-axis,’ Trudy
Moskov. Mat. Obshch., 3, 181-270 (1954). (Russ.)
(h) ‘On certain criteria for the completeness of a system of eigenvectors and
associated vectors of a linear operator in Hilbert space,’ Doklad. Akad. Nauk
SSSR , 98, 727-30 (1954) (Russ.)
(i) ‘Spectral analysis of non-selfadjoint operators,’ Uspekhi Mat. Nauk, II,
6(72), 183-202 (1956). (Russ.)
82. Natanson, 1. P. Theory offunctions o f a real variable, GTTI, 1950. (Eng. trans.,
Ungar, N.Y., 1955.)
83. Newell, H. E. ‘The asymptotic forms of the solution of an ordinary linear
matric differential equation in the complex domain,’ Duke Math. J., 9, 245-58
(1942); 10, 705-9 (1943).
84. Orlov, S. A. ‘On the deficiency indices of linear differential operators,’ Doklad.
Akad. Nauk SSSR. 92, 483-6 (1953). (Russ.)
85. Perron, O. ‘On linear differential equations in which the independent variable
is real,’ J. reine angew. Math., 142, 254-70 (1913); 143, 25-50 (1913). (Ger.)
86. Petrovsky, G. I.
(a) Lectures on the theory o f ordinary differential equations, GTTI 1949. (Eng.
trans., Interscience, N.Y., 1954.)
(b) Lectures on the theory o f integral equations, GTTI 1951. (Eng. trans., Gray-
lock Press, N.Y., 1957.)
87. Pleijel, A. ‘On the problem of the asymptotic distribution of proper values in
the theory of ordinary differential equations,’ Rung. Fysiografiska Sallskapets
i Lund Forhandlingaar, 17 (1947).
88. Plesner, A. I. ‘Spectral theory of linear operators, I,’ Uspekhi Mat. Nauk, 9,
3-125 (1941). (Russ.)
89. Plesner, A. I. and Rokhlin, V. A. ‘Spectral theory of linear operators, II,’
Uspekhi Mat. Nauk, 1 (11), 71-191 (1946). (Russ.)
90. Povsner, A. Ya.
(a) ‘On differential operators of Sturm-Liouville type on the half-axis,’ Mat. Sb.,
23(65), 3-52 (1948). (Russ.) . . xr
(b) ‘On the method of direction-functionals of M. G. Krein,’ Zaptski Nauch.
Issled. Inst it. Mat. i Mekh., Kharkov. Mat. Obshch., ser. 4, 20, 43-52 (1950).
(Russ.)
91. P r i v a l o v , I. I. Introduction to function theory, GTTI 1950. (Russ.)
140 BIBLIOGRAPHY

92. Pugachev, V. S. ‘On asymptotic representations of the integrals of systems of

ordinary differential equations containing a parameter,’ Mat. Sb., 15(57), 13-54
(1954). (Russ.)
93. Putnam, C. R. n
(a) ‘On the spectra of certain boundary value problems,’ Amer. J. Math., 71,
109-11 (1940).
(b) ‘The cluster spectra of bounded potentials,’ Amer. J. Math., 71, 612-20 (1949).
(c) ‘On isolated eigenfunctions associated with bounded potentials,’ Amer. J.
Math., 72, 135-47 (1950).
(d) ‘The spectra of quantum-mechanical operators,’ Amer. J. Math., 74, 377-87
(1952).
(e) ‘Note on a limit-point criterion,’ J. London Math. Soc., 29, 420-6 (1954).
(f) ‘A sufficient condition for an infinite discrete spectrum,’ Q. Appl. Math., 11,
484-7 (1954).
(g) ‘On the continuous spectra of singular boundary value problems,’ Canad. J.
Math., 6, 420-6 (1954).
94. Rapoport, I. M.
(a) ‘On the asymptotic behaviour of solutions of linear differential equations,’
Doklad. Akad. Nank SSSR, 78, 1097-1100 (1951). (Russ.)
(b) ‘On a singular boundary-value problem for ordinary differential equations,’
Doklad. Akad. Nauk SSSR, 79, 21-4 (1951). (Russ.)
(c) On certain asymptotic methods in the theory o f differential equations, Kiev, 1954.
(Russ.)
95. Rasulov, M. L.
(a) ‘Residue method for solution of mixed problems for differential equations
and a formula for the expansion of an arbitrary vector-function in fundamental
functions for a boundary problem with a parameter,’ Mat. Sb., 3, 48(90),
(1959). (Russ.)
(b) The contour integration method, Moscow, 1964. (Russ.)
96. Rellich, F.
(a) ‘Perturbation theory of spectral resolution, I,’ Math. Anna!. 113, 600-19;
II, 113, 677-85 (1936); III, 116, 555-70 (1939); IV, 117, 356-82 (1940); V, 118,
462-84 (1942). (Ger.)
(b) ‘Semi-bounded ordinary differential operators of the 2nd order,’ Math. Anna!.,
122, 343-68 (1951). (Ger.)
97. Sansone, G. Differential equations in the real field, Vols. I, II, 2nd ed., Nicola
Zanichelli, Bologna, 1948, 1949. (Ital.)
98. Schafke, F. W. ‘On the parametric dependence of the initial value problem
for ordinary linear differential equations,’ Math. Nachr., 3, 20-39 (1949). (Ger.)
99. Schubert, H. ‘On the eigenfunction expansion of admissible functions in
definite self-adjoint boundary-value problems,’ S.-B. Heidelberg Akad. Wiss.
Math. Nat. Kl., 8 (1948). (Ger.)
100. Schwartz, J. ‘Perturbations of spectral operators and applications, I.
Bounded perturbations,’ Pacific J. Math., 4, 415-58 (1954).
101. Sears, D. B.
(a) ‘Note on the uniqueness of the Green's function associated with certain
differential equations,’ Canad. J. Math., 2, 314-25 (1950).
(b) ‘Some properties of a differential equation (II),’ J. London Math. Soc., 29,
354-60 (1954).
102. Seifert, G. ‘A 3rd-order irregular boundary-value problem and associated
series,’ Pacific J. Math., 2, 395-A06 (1952).
 BIBLIOGRAPHY 141
103. Shin, D.
(a) ‘On the solutions in L2(0,co) of the self-adjoint differential equation u(n) =l(u),
7(/)= 0,’ Doklad. Akad. Nauk SSSR, 18, 519-22 (1938). (Russ.)
(b) ‘On quasi-differential operators in Hilbert space,’ Doklad. Akad. Nauk
SSSR , 18, 523-6 (1938). (Russ.)
(c) ‘On the solutions of a linear quasi-differential equation of the nth order,’
Mai. Sb., 7(49), 479-532 (1940). (Russ.)
(d) ‘On quasi-differential operators in Hilbert space,’ Mat. Sb., 13(55), 39-70
(1943). (Russ.)
104. Shnol, E. ‘The behaviour of eigenfunctions and the spectrum of Sturm-
Liouville operators,’ Uspekhi Mat. Nauk, 9, 113-32 (1954). (Russ.)
105. Smirnov, V. A. Course o f Higher Mathematics, Vols. IV & V, GTTI, 1947.
(Eng. trans, Pergamon Press.)
106. Smogorzhevskh, A. ‘Green’s functions for linear differential systems in a
1-dimensional domain,’ Mat. Sb., 7(49), 179-96 (1940). (French)
107. Steklov, V. A. ‘The problem of the cooling of an inhomogeneous rigid rod,’
Kharkov. Mat. Obshch., 1896. (Russ.)
108. Stepanov, V. V. A course in differential equations, GTTI 1950. (Russ.)
109. Stone, M. H.
(a) ‘A comparison of the series of Fourier and Birkhoff,’ Trans. Amer. Math.
Soc., 28, 695-761 (1926).
(b) ‘Certain integrals analogous to Fourier integrals,’ Math. Zeit., 28, 654-76
(1928).
110. Tamarkin, Ya. D.
(a) On certain general problems o f the theory o f ordinary linear differential equa­
tions, Petrograd 1917. (Russ.)
(b) ‘Some points in the theory of differential equations,’ Rend, di Palermo, 34,
354-82 (1912). (French)
(c) ‘Some general problems of the theory of ordinary linear differential equations
and expansion of an arbitrary function in series of fundamental functions,’
Math. Zeit. 27, 1-54 (1927).
111. Timoshenko, S. P. Theory o f elasticity, GTTI 1937. (Eng. trans. McGraw Hill.)
112. Titchmarsh, E. C.
(a) Eigenfunction expansions associated with 2nd-order differential equations,
Oxford 1946.
(b) ‘Some theorems in perturbation theory,’ Proc. Roy. Soc. London, ser. A, 200,
34-46 (1949).
(c) Eigenfunction expansions associated with 2nd-order differential equations,
Oxford U.P., Part I (2nd ed.) 1962, Part II 1958.
113. Vilenkin, H. Ya. ‘On the theory of orthogonal kernels,’ Uspekhi Mat. Nauk,
7, 3(49), 63-82 (1952). (Russ.)
114. Vizitei, V. N. and Markus, A. S. ‘On the convergence of multiple expansions
in the system of eigenvectors and associated vectors of a pencil of operators,’
Mat. Sb., 66(108), 2, 287-320 (1965). (Russ.)
115. W allach, S.
(a) ‘On the location of spectra of differential equations,’ Amer. J. Math., 70,
4, 833-41 (1948).
(b) ‘The spectra of periodic potentials,’ Amer. J. Math., 76, 842-8 (1948).
(c) ‘The stability of differential equations with periodic coefficients,’ Proc. Nat.
Acad. Sci., USA, 34, 203-4 (1948).
142 BIBLIOGRAPHY

116. W a r d , L. E.
(a) ‘Some 3rd-order irregular boundary-value problems,’ Trans. Amer. Math.
Soc., 29, 716-45 (1927)
(b) ‘A 3rd-order irregular boundary-value problem and associated series,’
Trans. Amer. Math. Soc., 34, 417-34 (1932).
117. W a r d , W. ‘An irregular boundary value and expansion problem,’ Annals o f
Math., 26, 21-36 (1925).
118. W e y l , H. ‘On ordinary differential equations with singularities and the
associated expansions of arbitrary functions,’ Math. Annal., 68, 222-69 (1910).
(Ger.)
119. W in d a u , W. ‘On linear differential equations of the 4th order with singulari­
ties, and the related representations of arbitrary functions,’ Math. Annal., 83,
256-79 (1921). (Ger.)
120. WlNTNER, A.
(a) ‘Small perturbations,’ Amer. J. Math., 67, 417-30 (1945).
(b) ‘Asymptotic integration constants,’ Amer. J. Math., 68, 553-9 (1946).
(c) ‘On the normalization of characteristic differentials in continuous spectra,’
Phys. Rev., 72, 516-17 (1947).
(d) ‘The adiabatic linear oscillator,’ Amer. J. Math., 68, 385-97 (1946).
(e) L2-connections between the potential and kinetic energies of linear systems,’
Amer. J. Math., 69, 5-13 (1947).
(f) ‘Asymptotic integrations of the adiabatic oscillator,’ Amer. J. Math., 69,
251-72(1947).
(g) ‘On the location of continuous spectra,’ Amer. J. Math., 70, 22-30 (1948).
(h) ‘A norm criterion for non-oscillatory differential equations,’ Q. o f Appl.
Maths., 6, 183-5 (1948).
(i) ‘A criterion of oscillatory stability,’ Q. o f Appl. Maths., 7, 115-17 (1949).
(j) ‘Asymptotic integrations of the adiabatic oscillator in its hyperbolic range,’
Duke Math. J., 15, 56-67 (1948).
(k) ‘A priori Laplace transformations of linear differential equations,’ Amer. J.
Maths, 71, 587-94 (1949).
(l) ‘On almost free linear motions,’ Amer. J. Math., 71, 595-603 (1949).
(m) ‘On the smallness of isolated eigenfunctions,’ Amer. J. Math., 71, 603-11
(1949).
121. W o l f s o n , K. G. ‘On the spectrum of a boundary-value problem with 2 singu­
lar end-points,’ Amer. J. Math., 72, 713-22 (1950).
122. Z h d a n o v i c h , B. F. ‘Non-selfadjoint differential operators with non-regular
boundary conditions in a space of vector-functions,’ Trudy 5th Vsesoyuz. Conf
on functional analysis, Baku 1961, 58-64. (Russ.)
123. Z i a t d i n o v , F. Z .
(a) ‘On linear differential operators of the 2nd order in a Hilbert space of vector-
functions with values in an abstract Hilbert space,’ Izv. Vuz. 4(17), 89-100 (1960).
(b) ‘On the spectrum of a linear differential operator of 2nd order in a Hilbert
space of vector-functions with values in an abstract Hilbert space,’ Izv. Vuz.
6(43), 46-59 (1964); 2(45), 69-76 (1965). (Russ.)
 INDEX

A djoin t, boundary con d ition s, 9, 110 D ifferential operator, 3, 107

boundary-value problem , 11 inversion of, 31, 115
differential expression, 6, 109 o f first order, 113
kernel, 116 D im en sion o f a vector space, 2
m atrix, 109 D o m a in o f definition, 2
operators, 10
eigenvalues and eigenfunctions of, 20 E igenfunctions, 13 et seq., I l l
G reen ’s function for, 33 asym ptotic behaviour, 74
A ssociated functions, 16 expansion in, 79 et seq., 91, 125, 129
A sym p totic behaviour o f eigenvalues and m ultiplicity of, 17
eigen fu nctions, 42 et seq., 118 o f adjoint operators, 20
o f self-adjoint operators, 21, 124
Basis, 2 Eigenvalues, 13 et seq., I l l
R iesz, 90 asym ptotic behaviour, 63, 118, 122
Bending o f a rod, 2 2 -6 m ultiplicity, 13
Boundary con d itions, 3, 107 Eigenvalue problem , generalized, 15
adjoint, 9, 110 exam ples, 22
h om ogen eou s, 4 Eigenvalues
norm alized, 56, 120 o f adjoint operators, 20
periodic type, 61 o f a self-adjoint operator, 21
regular, 56, 83, 121, 125 Expansion in eigenfunctions, 79 et seq., 91,
self-adjoint, 10 125, 129
separable, 91 Expansion o f G reen ’s function, 88, 128
Sturm type, 60 Extension o f an operator, 2
B oundary-value problem
adjoint, 11, 111 Fourier m ethod, 79
h om ogen eou s, 4, 108
h om ogen eou s, generalized, 6 Generalized eigenvalue problem , 15
in volvin g a param eter, 34 G reen’s function, 29 et seq., 88, 115, 128
rank of, 6, 14, 111 analytic nature of, 37, 117
reduction to integral eq u ation , 34 construction of, 29
self-adjoint, 81 for adjoint operator, 33
for the operator L— Al, 35, 115
C an on ical system o f eigenfunctions and with m ultiple pole, 40, 103
associated functions, 18, 104
C haracteristic determ inant, 14 H erm itian m atrix, 108
C om pleteness, m -fold , 103 H om ogen eou s
boundary-value problem , 4, 108
D erived system , 103 operator equation, 107
D ifferential expression, 3,
adjoint, 6, 109 Integro-differential equation, 45
linear, 3, 106 Integral equation,
self-adjoint, 7, 109 h om ogeneous, 35

143
144 INDEX

Integral equations, Operator, 2

system of, 46 differential, 3, 107
Integral operator, 33 adjoint, 10, 110
Inverse operator, 27 extension of, 2
Inversion form ula for differential operator, integral, 33
31, 115 inverse, 27
linear, 2
restriction of, 2
K eldysh’s theorem s, 94, 104 self-adjoint, 10, 81, 110, 124
K ernels, 33 O perator-functions, 105
adjoint, 33, 116 Operators, equality o f, 2
H erm itian, 33, 116 Order o f a form , 5 6 -7 , 120
sym m etrizable, 82 O rthogonal functions, 20

R ange o f values, 2
L agrange’s form ula, 6, 109 R ank o f a boundary-value problem , 6, 111
/-analytical function, 101 R estriction o f an operator, 2
Length o f system o f associated functions, 17
Linear Scalar product, 109
com b ination , 1 Self-adjoint
differential expression, 3 boundary con d ition s, 10
in space o f vector functions, 106 differential expression, 7, 109
m anifold, 2 operator, 10, 21, 110, 124
operator, 2 Sim ple eigenvalue, 15
vector space, 1 “ Sourcew ise” representable fun ction , 81
Linearly independent, 2, 108
Subspace, 2

V ector, 1
M atrix
space, linear, 1
adjoint, 109
finite-dim ensional, 2
H erm itian, 109
M ultiplicity infinite-dim ensional, 2
function, 105
o f an eigenfunction, 17
o f an eigenvalue, 13
W eight
o f a determ inant, 91
N atural vibrations o f stretched string, 26 m atrix, 91
N orm , 6 W ronskian, 30