334 LINEAR INTEGRAL EQUATIONS.

satisfies the conditions stated above for all values of x, it

was shown that a precisely analogous theorem holds for the
approximation of f(x) by a trigonometric sum of order n or
lower, this result being obtainable as a consequence of the
preceding. It is now shown that decided simplification in
the proof of both theorems may be effected by proving the
second directly (this had been done only for k = 1) and
deducing the first from it.
This method has the further advantage that the numerical
constants involved can be computed more conveniently.
For example, if f(x) satisfies the condition
I/0&2) — /Ol)| S I 32 - XX\
in the closed interval (0,1), it can be approximately represented
in this interval by a polynomial of degree n or lower, with an
error which never exceeds 3/n, for all positive integral values
of n. The same line of investigation leads to results in the
theory of Fourier's series.
13. There is a theorem that the perpendiculars let fall from
the incenters of three out of four lines of given direction upon
the remaining line touch a circle. In Dr. Hodgson's paper a
circle is obtained for any even number of lines, beginning with
four. If we take this circle for any 2n out of 2n + 1 lines,
the 2n + 1 circles touch a line. The question of the reversal
of direction of one or more of 2n lines is then taken up, and
this is followed by the consideration of the configuration of
circles arising from four, five, and six lines.
F. N. COLE,
Secretary.

ON THE FOUNDATIONS OF THE THEORY OF LINEAR

INTEGRAL EQUATIONS.*
BY PROFESSOR E. H. MOORE.

1. The Analogous Systems of Linear Equations.

THE theory of linear integral equations, mathematically
considered, has its taproot in the classical analogies between
* Address of the Vice-President and Chairman of Section A of the Amer-
ican Association for the Advancement of Science, Washington, December
29, 1911.
1912.] LINEAR INTEGRAL EQUATIONS. 335

an algebraic sum, the sum of an infinite series, and a definite

integral.
Consider the linear algebraic equation
(1°) x=ky
for the number y, the coefficient k and the number x being
given. From this single equation 1° we ascend to the algebraic
system
n
(Un 0 ) Xi = S hjVj (i = 1, 2, • • -, n)

of n simultaneous linear equations for the determination of

the set (yi) of n numbers yi, • • •, yn, the matrix (kij) of n2
coefficients &n, • • •, knn and the set (xi) of n numbers
Xi, • • •, xn being given.
To this algebraic system (IIn°) we have by the classic
analogy the two corresponding transcendental systems

(111°) Xi^t.kijyj (i = 1,2, . . . ) ;

(IV°) m = f K(S, t)rj(t)dt (a£s S b).

In 111° the infinite set (yi) is to be found, the infinite set (xi)
and the infinite matrix (&#) of coefficients being given; the
suffixes i, j have the range 1, 2, • • • . In IV° the unknown
function rj and the known function £ are functions of one
variable ranging over the interval a-b of the real number
system, while the known coefficient function or, in Hubert's
terminology, kernel K is a function of two variables ranging
independently over that interval. I t is plain that the theories
of 111°, IV° must involve convergence considerations.
Throughout, the numbers and the functional values of the
functions are real or complex numbers.
You are aware that the study of the algebraic system II n °,
initiated before 1678 by the genial intuition of the philosopher-
mathematician Leibniz, led to the development of the theory
of determinants—a theory which in the nineteenth century
came to permeate all branches of number theory, algebra,
analytic geometry, and pure and applied analysis, exerting
everywhere a profound influence not merely by its usefulness
but perhaps even more by the extreme elegance of its methods
and results.
336 LINEAR INTEGRAL EQUATIONS. [April,

The theory of infinite determinants connected with the

denumerably infinite system 111°, or more exactly with the
equivalent system in which the indices i, j have the integral
values from — oo to + oo, was initiated by G. W. Hill, who
in 1877 made hardy but happily effective use of the deter-
minant of a system of the latter type in his solution of a
differential equation arising in his memorable study of the
motion of the lunar perigee. To supply the requisite con-
vergence proofs, Henri Poincaré in 1886 laid the foundations
of the general theory of infinite determinants, which has since
been developed chiefly by Helge von Koch.
We are led to an analogous determinant, not from the con-
tinuously infinite system IV° of linear equations, but from the
system

(IV) {(*) = v(s) - z f K(S, t)v(t)dt (a^s ^ 6).

*Ja

Here z is a given number, real or complex, and we consider

the regular case, in which the functions involved are con-
tinuous real or complex valued functions of their arguments.
The corresponding systems
(I) x = y - zky,
n
(Lin) Xi = j/i — zJ^kiMj (i = 1, 2, • • -, n),

(III) Xi = yi — zj^kijVj (i = 1, 2, • • •)

are respectively equivalent to the systems 1°, II n °, I I P .

The types IV, IV° are however essentially distinct. For
instance, if we look at IV and IV° as transformations of the
functions y into the functions £, the type IV contains (in
the case of vanishing parameter z or identically vanishing
coefficient function or kernel K) the identical transformation
£ = 7j, while the type IV° does not contain this transformation.
The solution rj of the system or integral equation IV may be
expanded formally as a power series in the parameter z, viz.,

K
v(p) = %(p) + * I (P> Pi)£(Vi)dVi
•J a
nb r*b
+ z2 I I n(p, pi)K(ph P2)%(p2)dpidp2 +
1912.] LINEAR INTEGRAL EQUATIONS. 337

This power series converges near z = 0 uniformly in p, and

accordingly for z sufficiently small has as sum a continuous
function rj which is readily proved to be a solution of IV.
Impelled by the fact that integral equations of type IV
occur very frequently in the linear problems of mathematical
physics, in the late nineties of the last century the Swedish
mathematical physicist Ivar Fredholm undertook the study
of the analytic character of the solution rj as a function of the
parameter z. After earlier notes on the subject, Fredholm
published his fundamental memoir in 1903 in volume 27
of the Acta Mathematica, in one of the two volumes of the Acta
dedicated to the memory of Abel on the occasion of the cen-
tenary of his birth. And this was the more fitting since Abel
first studied special integral equations of the type IV°. Ac-
cordingly Fredholm calls the equation IV° Abel's integral equa-
tion. Mathematicians generally call the equation IV Fred-
holm's integral equation. With Hubert one also designates
the equations IV°, IV as integral equations of the first and
second kind.
Fredholm found that the function r\ is a single-valued
analytic function of the parameter z, having at most polar
singularities in the finite s-plane, and he exhibited it explicitly
as the quotient of two permanently converging power series
in z. The denominator series with coefficients depending only
on the kernel K, viz.,

1—s I K(P, p)dp+~ 1 1 / v / 'irf^i'*',

Ja ^ *' * 2 Ja Ja \K(P2, PI) K(p2, Vm
is Fredholm's determinant of the kernel K with parameter z.
In case z is not a root of this determinant, for every function £
there is a definite solution rj of the equation IV, and the same
is true as to the adjoint equation

(IV) £(*) = v(t) - * f V(S)K(S, t)ds (a <,t ^ b).

On the other hand, if s is a root of the determinant, it is of

finite multiplicity m, and each of the corresponding homoge-
neous equations
(IVir) y(s) = z f K(S, t)v(t)dt (a ^ s <; 6),
*sa

(WH) V(t) = 2 \ V(S)K(S, t)ds (a^t^b)

%)a
338 LINEAE INTEGKAL EQUATIONS. [April,

has a solution rj not identically vanishing, the number n of

linearly independent solutions rj for one equation being the
same as for the other equation and at most m.
These few results suffice to suggest the close parallelism
between Fredholm's theory of the integral equation IV and the
current theory of the algebraic system I I n .
We have seen, then, that the theories of determinants of
the matrices or kernels of the three types—the finite, the de-
numerably infinite, the continuously infinite—were initiated
by the mathematician-philosopher Leibniz, the mathematical
astronomer Hill, the mathematical physicist Fredholm; and
we appreciate anew the magnitude of the debt owed by pure
mathematics to its most closely related sister sciences—logic,
astronomy, physics—a debt abundantly repaid by the appli-
cations throughout the wide range of the sciences, at least in
the progress of time, of even the most abstract doctrines of
pure mathematics.

2. References to the Literature.

The investigations of von Koch and Fredholm opened the
way for the systematic development, now in rapid progress,
of the analogies and the interrelations between the algebraic
and the two kinds of transcendental theories, and for the im-
mediate application of the new results in various domains of
pure and applied analysis.
Especially noteworthy are the memoirs of David Hilbert and
of Erhard Schmidt. By direct limiting processes Hilbert
obtains from algebraic theorems the corresponding tran-
scendental theorems. Hilbert has thus initiated a theory of
functions of a denumerable infinity of variables, from which
in turn, by the connection between functions of continuous
variables and their Fourier coefficients, one proceeds to the
theory of functions of continuous variables. In particular,
Hilbert has studied real-valued symmetric kernels, obtaining
the transcendental analogues, for example, of the orthogonal
transformation of the algebraic quadratic form to the sum of
squares of linear forms. Geometric analogies of metrical
nature play a considerable rôle in the work of Hilbert, and
perhaps even more in the work of Schmidt, who treats tran-
scendental problems directly, using methods originated by
H. A. Schwarz in the potential theory. Schmidt has also
entered upon the study of non-linear integral equations.
1912.] LINEAR INTEGRAL EQUATIONS. 339

But for details of the extensive literature and present

state of the whole subject I must content myself with referring
to the most recent books:
HEYWOOD ET FRÉCHET: L'équation de Fredholm et ses applications à la
physique mathématique. Hermann, Paris, 1912.
LALESCO: Introduction à la théorie des équations intégrales. Hermann,
Paris, 1912.
and to the reports :
HILBERT: "Wesen und Ziele einer Analysis der unendlichvielen unab-
hângigen Variabeln." Palermo Rendiconti, volume 27, pages 59-74
(1909).
VON KOCH: "Sur les systèmes d'une infinité d'équations linéaires à une
infinité d'inconnues." Compte rendu du\ Congres des Mathématiciens,
tenu à Stockholm 22-25 Septembre, 1909, pages 43-61; Teubner,
Leipzig, 1910.
FREDHOLM: "Les équations intégrales linéaires." Ibid., pages 92-100.
BATEMAN: "Report on the history and present state of the theory of inte-
gral equations." Report to the British Association for the Advance-
ment of Science, Sheffield, 1910, 80 pp.; Burlington House, London,
1911.
HAHN: "Bericht über die Theorie der linearen Integralgleichungen."
Jahresbericht der Deutschen Mathematiker-Vereinigung, volume 20,
pages 69-117 (1911).

3. The Fundamental Problem of Unification. General Analysis.

We are now in position to take up, as the specific subject
of this discourse, the question of foundations of the theory
of linear integral equations. We have seen that the algebraic
theory serves to suggest the corresponding transcendental
theories, or even to determine those theories by suitable use
of limiting processes. But this state of affairs may be recog-
nized as only preliminary to the determination of a general
theory capable of specialization into the various theories.
This is in accordance with a general heuristic principle of sci-
entific procedure, which I have formulated as follows:
The existence of analogies between central features of varions
theories implies the existence of a general theory which underlies
the particular theories and unifies them with respect to those
central features.
For the case of the real-valued symmetric kernel, and in
fact for the more general case of the complex-valued hermitian
kernel, i. e., a kernel K satisfying identically the condition
t h a t K(s, f) and ic(t, s) are conjugate complex numbers, I
340 LINEAR INTEGRAL EQUATIONS. [April,

took up six years ago this problem of unification for the

Hubert theory as presented by Schmidt. This was the theme
of my series of lectures: " On the theory of bilinear functional
operations," at the colloquium of the American Mathematical
Society, held in September, 1906, in New Haven, under the
auspices of Yale University.
Subsequent study led to the recognition that the general
theory of linear integral equations is merely a division in
the theory of a certain form of general analysis, an introduction
to which, instead of the colloquium lectures, I published,*
as pages 1-150, in the volume The New Haven Mathematical
Colloquium, etc., Yale University Press, New Haven, 1910.
This morning I wish to establish, in the sense of general
analysis, an adequate and satisfactorily simple foundation
for the general theory of linear integral equations, embracing
by specialization, as we shall see, together with an interesting
variety of new theories, the algebraic and both types of tran-
scendental theories, now current, at least in so far as regular
kernels are concerned.
As to the system I I I , I call the matrix-kernel (&#) regular
in case there is a set (ki) of numbers of finite norm 2»- | hi | 2 ,
such that for every i and j \ lctj | ^ | hihj |. This regular kernel
satisfies the latest condition found t by von Koch as sufficient
that the infinite determinant and all its minors converge abso-
lutely. Then, if the sets (xi), (yî) are of finite norm, the sys-
tem I I I may be treated either by the method of infinite
determinants or by Hubert's theory of functions of infinitely
many variables.
I shall indicate, first, the terminology or basis of the founda-
tion of the general theory, and then two sets of postulates,
the former effective for the validation of the general Fredholm
theory for the general kernel, and the latter effective for the
validation of the general Hilbert-Schmidt theory of the real
symmetric or the more general hermitian kernel. And, in
advance, I notice that for the latter theory we need more
postulates than for the former theory. This is in accordance
* Cf. also my paper, " O n a form of General Analysis, with application
to linear differential and integral equations," read before the Section
on Analysis of the Rome Congress of 1908, Atti, etc., vol. 2 (1909), pp.
98-114.
t Loc. cit., pp. 49, 50; Palermo Rendiconti, vol. 28 (1909), pp. 257, 263.
The condition is that the kernel (fan) have the form ha = UijVi/vj (i,j)y
where vi =H 0 (i) and the series S»-,-w»,-2, XiUu converge absolutely.
1912.] LINEAR INTEGRAL EQUATIONS. 341

with the nature of things logical: we must pay for the elaboration
of theory by the imposition of additional postulates and the cor-
responding restriction of scope of application»

4. Fredholm's Equation in General Analysis. The Basis S x .

In order to bring the equation-systems I, II, I I I into no-
tational conformity with the integral equation IV, we regard,
for instance, in I I I the set (xi) of numbers Xi (i = 1, 2, 3, • • •)
as a function x or £ of the argument i oi s with the range
i = s = 1, 2, 3, • • • . Then I, II W , I I I , IV are special cases
of the general equation

(G) £ = if - zJitri,
with the meaning
(Ö) it(s) = rj(s)-zJtK(st)r,(t) (s),
which we designate as Fredholm's equation in general anal-
ysis. The kernel K, the parameter z, and the function £ being
given, the function TJ is to be determined as a solution of the
equation G.
The understanding is that (1) £ and rj are functions of an
argument p or s or t having a certain range ^3; (2) /c or ic(st)
is a function of two arguments ranging independently over ^5 ;
(3) J or Jt is a functional operation turning a product KTJ
or K(st)rj(t) into a function of the argument 5; and (4) the
equation G holds for every value of s on the range ^3.
For the general theory this range ^5 is simply a class of
elements p. These elements p are of any nature whatever,
e. g., numbers, sets of numbers, functions, points, curves; and
they are not necessarily all of the same nature. Thus, the
range ^5 is a general class of general elements. This " general "
is the true general, in the sense of arbitrarily special, that is,
capable of arbitrary specification—without the exclusion of
exceptional or singular cases.
Thus, for the general theory of the equation G the range $P
enters without the imposition of restrictive properties or
features, and it is this presence in the theory of a general class
which constitutes the theory a doctrine of the form of general
analysis which we are developing.
For the respective instances I I n , I I I , IV, the range ^3 is
342 LINEAR INTEGRAL EQUATIONS. [April,

finite, denumerably infinite, continuously infinite, consisting of

the respective elements
p = 1, 2, • • *, n; p = 1, 2, 3, • • • ; a ^p £b.
We denote these ranges by the notations
spn». çpin. çpiv#

The respective functional operations J are

*=1 ?=1 Ja

In the instances I I I and IV the functions £, rj, K are necessarily

subject to certain conditions of convergence or of continuity.
The conditioning properties are defined in terms of features
possessed by the special classes $p m , ^3 IV ; we are able to speak
in ^ 3 m of p tending to oo, and in ^3 IV of the difference p\ — p2
of two elements p. Similarly, in the general theory the func-
tions must possess certain properties, which must however be
postulated and not explicitly defined, since we attribute to the
general range no features available for use in the definitions.
Now, instead of postulating properties of the functions, it is
technically more convenient to postulate classes of functions
to which the functions shall belong, viz., the classes of func-
tions are the classes of all functions possessing the respective
properties.
Accordingly, the form of the general equation G leads to
the following first basis:
Zi = (31; $; 3»; ft; J )
for the construction of a general theory of the linear equation
G. Here 31 denotes the class of all real or the class of all com-
plex numbers a; ^3 denotes a general class of general elements
p or s or t or u or v or w; 3)t denotes a class of single-valued
functions JJL on ^ to 31, that is, for every function ^ and argu-
ment p, fjt,(p) denotes a definite number a of the class 3t; ft
denotes a class of functions K on ^3^3 to 3Ï, that is, for every
function K and (ordered) pair (st) of arguments of ^3, n(st)
denotes a definite number a of the class 31; and J or Jt denotes a
definite functional operation on ft9Jl or fts*9ft* to 3)i or 3JÎS,
that is, for every function K of ft and rj of 3JÎ, JKTJ or Jt K(st)rj(t)
denotes a definite function of 3)1 or 3)l5.
1912.] LINEAR INTEGRAL EQUATIONS. 343

For the basis Si the 'problem of foundations of a general

theory of the equation G is then: to specify properties of the con-
stituents of the basis Si sufficient to validate the desired theory.
The range ^5 is to remain general, and the properties specified
are to be of general reference, that is, defined with respect to
the general range ^5.
I t is convenient here to refer to the important memoirs of
S. Pincherle:
"Sulle equazioni funzionali lineari". Rendiconti delta R. Accademia dei
Lincei, ser. 5, vol. 14 (1905), pp. 366-374;
"Sulle equazioni funzionali lineari". Bologna Memorie, ser. 6, vol.3 (1906),
pp.143-171;
"Appunti di calcolo funzionali". Bologna Memorie, ser. 6, vol. 8 (1911),
pp. 1-38.
In these memoirs, from the standpoint of the general theory
of linear distributive functional operations, Pincherle inves-
tigates the problem of foundations for a theory of the equation
(G') ï=v-zJ'v,
which includes the equation G, with the basis
2 ' = (21; <$; Wt; J ' ) ,
where J' is a functional operation on -BÎ to 9JÏ.

5. Certain Definitions. The Closure Property C\. Relative

Uniformity of Convergence.
We do not however retain the basis Si. With the purpose
of obtaining finally a general theory characterized by its
simplicity and by its possession of a number of important
closure properties, we set up other bases S 2 , • • -, SO. When-
ever a general theory, as a matter of fact, includes, as a special
instance, a theory analogous to but not a priori one of its
instances, we speak of a closure property of the general
theory. The terminology is adapted from that in use in the
theory of point sets.
That the general theory include, as special instances, the
current theories of the equations I, I I n , I I I , IV with regular
kernels is the closure property Ci, fundamental to the whole
inquiry.
You are familiar with the notion, uniformity of convergence
of a sequence of functions over a range of values of the argu-
ment of the functions, and you appreciate the fundamental
344 LINEAR INTEGRAL EQUATIONS. [April,

rôle played by the notion in the development of analysis

during the last half century. A sequence {jun} of functions
/in (n = 1, 2, 3, • • •) of the argument p on the range $P con-
verges over the range 9$ uniformly to a function 6 as limit
function, in notation
L ixn = e CP),
n
in case for every positive number e and index n greater than a
number ne, dependent upon e alone, the difference /*n(p) — 0(p)
is, for every value of p on the range $P, in absolute value at
most e:
I A * n ( p ) - 0(p}\ <> e.
For investigations in general analysis we need a more gen-
eral notion, the notion of relative uniformity of convergence, or,
uniformity of convergence relative to a scale function. If the
function a (defined on the same range ^3) is the scale function,
this relative uniformity, in notation
LfJLn= 6 («P; or),
n

has the same definition except that the final inequality is

replaced by
I/*»(p) - »(p) I Se |<r(p)|.
We speak also of relative uniformity as to a class © of scale
functions; the notation
£ / * » = e cp; © )
n
means that for some function a of the class S we have the
relative uniformity as to the scale function v.
One observes that the classical uniformity is that instance of
relative uniformity in which the scale function is identically 1.
In illustration of this notion of relative uniformity, consider
a sequence {/*n} of real-valued nowhere negative functions ixn
of the real variable p on the infinite interval ^ = 1 — oc
of the number system, and let the functions fin be individ-
ually integrable from 1 to oo. If relative to a scale function
a of the same kind the sequence {/*»} converges uniformly on
^P to a limit function 6, then 6 also is integrable from 1 to oo,
and the integral of 6 is the limit as to n of the integral of fin.
But the limit function 6 is no longer necessarily integrable
1912.] LINEAR INTEGRAL EQUATIONS. 345

from 1 to oc, if the convergence is merely uniform in the class-

ical sense (a = 1), as appears from the following example:

/*n(p) = 1/p (1 Û V ^ n)> n/P2 (P ^ n)>

»(p) = 1/P (1 ^ p ) .
To facilitate the exposition of the sequel we need certain
additional definitions.
Consider a class 9ft of functions ^ on ^3 to 91, that is, on
the range ^3 with functional values belonging to the class 21.
The class Wlz, the linear extension of the class 9ft, is the class
of all functions fxL of the form
MX = aim + «2M2 + • • • + ClnVn,
viz., the class of all linear combinations of a finite number of
functions belonging to the class 9ft with numerical coefficients
belonging to the class 2Ï. Further, © being a class of functions
a on ^3 to 31, the class 3)1®, the class 9ft extended as to the class ©,
is the class of all functions 6 of the form
0 = Lfxn (?P; cr),
n
a form which has been defined above. The class 9ft is con-
tained in the class 9ftz and for every S in the class 9ft@. If
the classes 9ft and 9ft/, are identical, the class 9ft is said to be
linear, to have the property L. If the classes 9ft and 9ft^
are identical, the class 9ft is said to be closed, to have the
property C; otherwise expressed, the class 9ft is closed in
case every function 0 of the form

n
belongs to the class 9ft. The class 9ft*, the ^-extension of the
class 9ft, is the class (9ft/,) ^ that is, the extension as to 9ft of
the linear extension of 9ft. The class 9ft2 is the class of all
functions of the form /XIJU2 or fii(p)mip)> that is, of all products
of pairs of functions of the class 9ft, the arguments of the two
functions being the same.
In illustration of these definitions, if 9ft is the class 9ftIV of
all continuous functions on the finite linear interval 5)3IV, a~b,
we have
m = mL = mm=m* = 9ft2.
Further, if 9ft is the class 9ftni2 of all functions /* on $
346 LINEAR INTEGRAL EQUATIONS. [April,

(p = 1, 2, 3, • • •) such that the series SPM(P)2 converges ab-

solutely, we have
m = mL = mm = m*; aft2 = an1111,
where 9Jlni1 denotes the class of all functions \x on ^ such that
the series S pi u(p) converges absolutely. Thus the classes
•Di111* and 3JtIV occurring in the regular cases of equations I I I
and IV are linear closed classes of functions.
Consider further two general ranges 5)3', %" conceptually but
not necessarily actually distinct. The product range %'%"
is the class of all composite elements (p', p") or p'p", the first
constituent p ' being an element p ' of the class ^3', and the
second constituent p " being an element p" of the class ty".
For example, if $P' is the linear interval a' - 6 ' and ^3" is the
linear interval a"-b", the product $|3'^3" is, to speak geo-
metrically, the rectangle a' ^ p' ^ b', a" £ p" ^ b". The
product class WW of two classes W, W of functions on the
respective ranges ^3', ^3" consists of all products M V or
/*' (p') P" (p") of a function // of the class W on ^ r and a function
M" of the class 3ÏI" on $ " . The class (WW%, the ^-com-
posite of two classes W, 3K" on the respective ranges ^3', ^3"
is, as indicated by the notation, the -^-extension of the product
class WW, viz., the extension as to the product class WW
of the linear extension of the product class WW. The
classes 9tt'9Jt" and (WW) * are classes of functions on the
product range ^3'^3", and accordingly, if the ranges $P', 5)5"
are identical, $P' = 5)3" = 9$, the arguments of the functions
of those classes are variables (p', p" or pi, p 2 ) ranging inde-
pendently over ^3.
The suitability of these notions for use in a general theory
of linear integral equations is indicated by the fact that for
the regular case of the equation I I I or IV the functions £
and rj belong to a class 33Î (SDÎ1112 or WY) whose -^-composite
with itself is the class ft to which the kernels belong, viz.,
ft = (3B3R)*.

6. The Bases S 2 , S 3 , S 4 .
With the aid of the notions and notations now at hand we
are able to proceed rapidly towards our goal. We recall that
the basis Si
Si = (21; $ ; m; ft; J ) ,
1912.] LINEAK INTEGRAL EQUATIONS. 347

was dictated by the mere form of the general equation 6?

(G) S(8) = l,(8)-zJtK(st)r,(t) (S).
By the consideration that in the typical instances I, I I n , III>
IV the kernel n(st) for every s as a function of the argument t
belongs to the class 9JÎ* to which the function rj(t) belongs, we
are led to a basis S 2 , simpler than Si, in the form
S 2 = (21; $; 3R; 9Î ï 3ÏÏ2; ft; J ) .
Here the class -Ji is defined as the class 5DÎ2 of all products jui/z2
of pairs of functions of the class 3JI, the arguments being the
same for the two functions; and J is a functional operation
on ît to 21, so that for every function v of 9Î, Jv denotes a number
a of the class 2Ï.
This system S 2 was basal for my lectures at the New Haven
Colloquium of 1906. The development of the theory of the
general equation G on the basis S 2 requires numerous postu-
lates. We must, for instance, arrange to extend the scope of
the functional operation J from the class 9Î to its linear
and -^-extensions, 3ÎL and $1*, in such wise that from the equa^
tions
vL = aivi + • • • + anvn) 0 = LvLn OP; v)
n
we have the conclusions
JvL = CliJvi + • • • + CLnJVn', JO = LJvLn*
n
Accordingly, if we define the class 9t as the class 9W*, the -^-ex-
tension of the class 9tt, and take J as a functional operation on
this class 9Î to 2Ï, we have a simpler basis S 3
S 3 = (SI; ^;m;m^ 2tt*; 51 ; J).
A theory based on S 3 requires postulates involving the kernel
class ft. These postulates are avoided if we define the class
ft as the -^-composite of the class Wl with itself. Thus we
obtain the still simpler basis S 4
S 4 = (31; $; 3»; 31 s 3R*; ft » (2K2R)*; J ) .
Here J is still a functional operation on -K to 21. For the
regular cases of the typical instances I, I I n , I I I , IV the defi-
nition just suggested of the kernel class ft is appropriate.
348 LINEAB INTEGKAL EQUATIONS. [April,

As to irregular cases, it is clear that a greater variety could be

treated on the basis 2 2 or S 3 than on the basis S4.

7. The Closure Property C2. The Basis S5.

In response to the desire that our theory of the general
equation G
(G) m = rj(s)~zJtK(stMt) (s)
shall possess a certain closure property C2, in addition to the
fundamental property C\ of having as special instances the
current theories for the regular cases of the equations (I, Hn>
III, IV), we are led to the basis 2 5 ,
Under the postulates to be imposed on the class 5DÎ, the kernel
a{sf) of the class 51 = (-ïïï-iDÎ)* is, as in the typical instances,
for every s as a function of t of the class 9ft* to which also rj(t)
belongs: thus, the operation J enters the equation G in the
form
Ja/3 or JPa(p)p(p),
where a and j8 are functions of the class 3)1. Now for the
instance IIn the expression
n
JIIna/3 or Z)a(p)/3(p)
P=i

is Grassmann's inner product of the n-dimensional vectors

a = (a(l), • • -, a(n)); 0 = (0(1), • • -, j8(n)),
and accordingly we term the expression JPa(p)P(p) or Ja&
the inner product of the functions a, /3. In this we follow usage
already established for the instances III, IV.
For the basis S4 it is convenient temporarily to denote the
operation J by J 4 and the equation 0 by G\ J 4 is then a
functional operation on 9? = 9tt*2 to 21. The first question
that suggests itself is the following: Is it possible to secure
a basis S 5 for the equation 6?5
(G5) { = v - ZJ'KV

of such a nature that the corresponding inner product

J5a$
1912.] LINEAR INTEGRAL EQUATIONS. 349

shall have as one of its instances the generalized inner product

J\JAua(f)<a(tu)P(u),
where œ is a function of the class $? To that end we set

and assign to G5 the meaning

6
(G ) m = v(s) - zJLAstMu) (*)•
Further, since the operand a(f)(3(u) belongs to the product
class 3J13JÏ, just as we were led to replace J 2 on 9J12 by J 3 and
J 4 on 31 E= 9Jl|, we stipulate that J 5 shall be a functional
operation on the class (2J19JÎ)*, that is, on the kernel class ^.
Thus we have reached the basis
2 5 = (»; ?P; 3tt; « s (3JÏ3ÏÏ)*; J),
where J is a functional operation on £ to 31.
This basis S5 is, by the omission of the class 9? = 9Ji|,
simpler than the basis S4. Further, the equation G4 is an
instance of the equation G5, viz., for the operation J 5 with

for every function cp of $. This stipulation is legitimate

since for every function (p(tu) of &tu = (3#$WW)* the re-
duced function <p(tt) belongs to %lt = 9#«*. Accordingly,
the general theory of the equation G5 based on S5 yields, as a
special instance, a theory of the equation G4 based on S4.
Furthermore, in accordance with the derivation of the basis
S 5, the general theory of the equation G5 contains as an
instance a theory of the equation
(G4") *(*) = v(s) - zJ\ruK{st)^{tu)n{u), (,),
5
where w is a function of $. The operation J for this instance
is the operation
Juu)<p(fu) = JétJi<p(tu)œ(tu)
for every function <p of ^. Under the postulates to be laid
on the basis S4 the double operation J\Ji is applicable to
the product <p(tu)a>(tu) of two functions of the class ^.
It is to be noted, however, that the geometric analogy, let
us say, between the sphere and the ellipsoid had as primary
function not that of enabling us to treat the equation G4<rt
350 LINEAR INTEGRAL EQUATIONS. [April,

based on 2 4 as a special instance of the equation 0 5 based on

S5; at least for the Fredholm theory, the equation G4(ü may
be treated as the equation 6?4 with the kernel K(st) replaced
by J*K(sf)a>(tu), since under the postulates to be imposed on
the basis 2 4 this is a function of the kernel class ^ and the
operations J*, J* are commutative. Its primary function was
rather to lead us from the basis 2 4 to a basis S5, possessing
in common with 2 4 the closure property C\, and possessing
furthermore the closure property C2—that for the basis S 5 a
similar use of the geometric analogy leaves us on the basis 2 5 .
In fact, if we seek a basis 2 for the equation G of such a
nature that the corresponding inner product
Ja/3
shall have as one of its instances the generalized inner product

where co is a function of the class $, we are led to set

Jap s J(tw)a(t)p(w);
thus the new operation J is still an operation J on 3JÎ9JÏ or
preferably on (9J19JI)* = &, that is, an operation of precisely
the same type as J 5 .
As in the preceding remarks, we notice that the equation
(G5w) m = v(s)-zJttn)JlwA^Muv)v(w) (s)
may be treated, either as the equation G6 with the kernel
K(sf) replaced by J(tu)ic(st)ü)(uv), or as the equation Gb with
the operation J6{tu) replaced by a new operation J{tu) defined by
the equation
J[5tu)<p(tu) = J5itw)J5{vu)<p(tu)ü)(wv)
for every function <p of ^.
8. Resume.
To summarize our course to this point: We have as data
the four analogous equations (I, II n , III, IV) with their four
analogous theories. The theories are of two stages: for brev-
ity, F, the Fredholm theory of the general kernel, and H,
the Hilbert-Schmidt theory of the real symmetric or of the
more general hermitian kernel.
1912.] LINEAR INTEGRAL EQUATIONS. 351

Guided by the heuristic principle of unification by abstrac-

tion, we formulate the general equation
(G) £ = rj - ZJKTJ,
embracing as instances the four typical equations; and we
seek the foundations, viz., the terminology or basis, and the
postulates, of a general theory of the equation G which shall
embrace as instances the four theories of at least the regular
cases of the respective equations. This is the fundamental
closure property Ci.
The form of the equation G dictates the basis Si, which by
consideration of the typical equations and of the obvious
nature of their theories we simplify to S 2 , S3, S 4 in succession.
The metric-geometric analogy of the sphere and the ellipsoid
leads on to the present basis
S 5 = ( » ; $; m; i t s (3JÎ9JÏ)*; J on 51 to 21),
with the general equation 6? interpreted as meaning
(G) £0) = i?W — zJ{tu)K(st)ri(u) (s).
And for this basis and its theory there is the closure property
C2.
The operation J is a functional operation on & to 3Ï, that is,
if <p is a function of the class $,
J<P = J (tu) (P (tu)
denotes a definite number a of the class 31. For purposes of
application J or J(tU) is often definable as a double operation,
in the form J(tU) = J/Ju"* For purposes of the general
theory of the basis S 5 , however, J(tU) is an indivisible operation.
Of course J(tu)<p(tu) is equal, e. g., to J(tw)<p(tw) or J(VU)<p(vu)
or J(vw)<p(vw), but it is in general not equal to J(ut)<p{tu).

9. The Definitive Basis Se.

The basis S 5 is in effect definitive for the general Hilbert-
Schmidt theory H. Suitable postulates will be given for
both bases S 4 , S 5 . However, for the general Fredholm theory
F the metric-geometric analogy, by leading us from the oper-
ation Jt on %t over to the operation J(tu) on §ttuy enables us
to proceed to a still more general basis S 6 .
352 LINEAR INTEGRAI/ EQUATIONS. [April,

In fact, for the theory H we have the respective definitions

K(ts) = K(st) and n(ts) = ic(sf),
of the symmetry of the real-valued kernel K and the hermitian
character of the complex-valued kernel K. These definitions
require the arguments s and t to have the same range ^3.
There is, however, no such necessity in the theory F>
Thus we are led to the following definitive basis for the
theory F, viz.,

^«= \%' & û; à = mm*' Jon

* )'
Here 5|3, $ are two conceptually but not necessarily actually
distinct classes of elements p, p respectively; 3ft, 9JÏ are two
classes of functions /I on $ , jx on 9|3 respectively; $ and È are
the * -composites of the two classes 9K, 9JI in the respective
orders 2JI, 2JI; m, Wl—that is, the functions K of $ bear their
arguments p, p in the order (pp), while the functions /c of È
bear their arguments in the reverse order (pp) ; thus there is a
correspondence of the functions K, îc of such a nature that for
two corresponding functions K, îc we have n{pp) = ^(pp)
for every pair of arguments p of %, p of ^j3; and functions K, K
occurring together are understood to be in this sense corre-
sponding functions, each the transpose of the other. Finally,
J is a functional operation on & to 91, so that for every function
K of È the expression
JK S J(tuyl<(tu),
denotes a definite number a of 9Ï. Here, in order to make
the formulas based on S 6 readily comparable with those based
on 2 5, we agree that the elements
p, s, u, ID ;
p, i, v, r,
are generic elements of the classes ^ ; $ respectively.
Based on S 6 we have the general pair of adjoint equations
(G) ? == V - zJicrj;
((?) | = v - zJ^/c,
1912.] LINEAR INTEGRAL EQUATIONS. 353

with the meanings

(G) % (s) = rj (s) - zJ(tu) K (st) rj (u) ($) ;
«?) l(t) = ij(t) - zJ(r8)v(r)K(st) (0.
Introducing the functional operation J, the transpose of «7,
by the equation: JK = JK, for every function K of $, so that
J is on £ to 21, we have the adjoint equation G6 in the form
(G) 1 = 5 - J**,
viz., in the form G for the transpose basis 26> which is the
basis SO with the interchange of rôles of ^3, $ ; 3ft, 501; SÎ, $;
J, J.
The basis 2 5 is secured from the basis 2 6 by the supposition
$ = $ E= ^5; 9ft = 3ft m 9ft.
Thus the theory for the basis S 6 has the closure property C\.
It has moreover the closure property C^ on the understanding
that the function co or u(uv) is any function of the class $;
that is, the functional operation J'
JL)<P(tu) s J6(tw)Jlu)<p(tu)œ(wv)
for every function #> of the class &, is a functional operation
of the type J 6 on IE to 2t. Further, as on the basis S5, the
equation

may be treated, either as the equation Ge with the kernel

n(st) replaced by J^tu)K(st)o)(uv), or as the equation GQ with
the operation J 6 replaced by the operation J' defined above.
Setting

S
\h • • •' tn) '^ ^ ' ^ ^ ^ 1 ' ' ' *' n ^
we define the Fredholm determinant FK(&) of the kernel K
and the parameter z, for the general theory of the adjoint
equations G9 G based on S6, as follows:

^(S) s h~krJ(hsd • • • « ^ K U • ••,<*;•

354 LINEAR INTEGRAL EQUATIONS. [April,

Here </(*l8l) . . . (tksk) denotes the &-fold operation

J(hSl)J(t2S2) • • • J(tk8k)-
The M i minor (h = 1,2, •••) has the definition
p f su •••, shf \

~( 1}
k h\ J«>»•>»>• ~«»*»**\tu ..., tv»)'
The initial terms (jk = 0) of the determinant and of the M i
minor are respectively
H 1 S h
i; (- I) K(* ' '"' \

Under the postulates to be specified below, Hadamard's

theorem on determinants may be utilized to show that these
series are permanently convergent power series in z, as to
the M i minor for all values of the arguments sh • • •, th.
Further, on every finite circle in the s-plane the fab. minor
series converges uniformly over the composite range ^$i • • • $/»
relative to the class
ft* . . . * s (9W* • • • afc>;
and its terms belong to the class ft* • • • th; and, as this class is
closed, the M i minor for every s as a function of the arguments
si, • —,th belongs to the class ft* • • • th.
The general theory of the adjoint equations G, G proceeds
along the usual lines. Thus, if the parameter z is not a
root of the determinant FK{z), the kernel K has the reciprocal
kernel X

V
X(rf)r- ' '
FA*) '
belonging to the class ^ and satisfying the equations
n(st) + \(st) = zJ(VW)K(sv)\(wt) — zJ(VW)\(sv)i<(wt) (st) ;
and the equations O, G have the solutions
^ = 1 - zJ\% v = % - zJ%\.
1912.] LINEAR INTEGRAL EQUATIONS. 355

10. Adjundional Composition. The Closure Property C$*

In the algebraic domain we proceed from the single equa-
tion I
(I) x = y - zky
to the system I I n of n simultaneous equations
n
(II n ) Xi = yi — zShjyj (i = 1, • • -, n).

Similarly, on the basis S 6 , we proceed from the single equa-

tion G
(G) Î O ) = rj(s) — zJ\tu)ic(st)rj(u) 0)
to the system Gn of n simultaneous equations
n
(Gn) ?• (*) = rji (*) — z Z ) J(tu)Kij (st) rjj (u) (si).
3=1

Here we have given the parameter z, the n2 functions K# of

the kernel class ft, and the n functions & of the class 9)i; and
we are to determine the n functions rji of the class 9ft. For
the instance IV on the linear interval a-b Fredholm showed
how to reduce the system IV n to a single equation IV on the
linear interval a-bn (bn = a + n(b — a)). A similar pro-
cedure is effective to reduce the system Gn on the basis 2 6
to a single equation G on an enlarged basis. The procedure,
however, is capable of further generalization, and, as thus
generalized, of an important application to the theory of mixed
linear equations.
To this end, consider a system of n bases 26* (i = 1, • • •, n),
viz.,
/ t}* W ft* s (WW) ~ \
Ze
"" v * sp*' W' a* s (aR*a»o' /
The class 21 is the same in the n bases 2*; otherwise the bases
are conceptually but not necessarily actually distinct; it is
however convenient to suppose that no two of the n classes
^ have common elements and that no two of the n classes $*
have common elements; this state of affairs, being always
securable by transformation, involves no restriction of
generality.
356 LINEAR INTEGRAL EQUATIONS. [April,

As a generalization of the system Gn on the basis S6 we

have on the system of n bases S6* the system Gn1 ' * *n
n
(Gn1 * ' ' ») ?(*9 = W) " 2 £ Ji^isHWiu*) (8%
4=1

of n simultaneous equations. Here we have given the param-

eter z, the n functions |* of the respective classes 9JI', and
the 7i2 functions K*' of the respective classes &ij = (9J^9JîO*>
so that $** = £*; and we are to determine the n functions y*
of the respective classes 3Jt\ The n equations (i = 1, • • -, n)
of the system (?„* ' * *n are to hold for every value of the
respective arguments si of the class ^\
We impose on each of the n bases Se* the postulates to be
specified below. Then the adjunctional composite Se1 * ' *n of
the n bases S6* is a basis S6 satisfying the same postulates,
and the system Gn1 * * ' n of n simultaneous equations is equiva-
lent to a single equation 6? on the composite basis S6. This
is the closure property Cz of the theory of the linear equation
G on the basis S6.
For the adjunctional composite S6X * * 'n of the n bases Se*
the class 3Ï is the common class 2Ï of the component bases Se*.
The ranges $, $ are the adjunctional composites or logical sums
respectively of the ranges $*, ^5*; the range ^3* is the ith com-
ponent of the range Ç, and the range $* is the ith component
of the range $. A function 0 on the range % determines n
component functions di on the respective component ranges
^3*; and, conversely, any n functions 0* on the respective ranges
5p* are the n components of a definite function 0 on the com-
posite range ^ ; this function 0 is the adjunctional composite
of the n functions 0\ The functions /x on the range % obtained
thus by adjunctional composition of n functions /x* of the
respective classes 9JÎ* on the ranges %l constitute the ad-
junctional composite 9JÏ of the n classes Tt\ The classes 3JÏ,
9JÏ of functions ju, jx on the composite ranges Sp, % of the com-
posite basis Se 1 ""* are the adjunctional composites, in this
sense, respectively of the classes 2ttl', 3JI* of functions jut*, /I*
on the component ranges $*, $ \ The product ranges %%, %%
are the adjunctional composites respectively of the n2 product
1912.] LINEAR INTEGRAL EQUATIONS. 357

ranges f^-7', $ * ^ ' . Then, under the postulates, the classes

? f (ftp)*> & = ( ™ ) * of functions *, Ï on the ranges
$P$, $^3 are the adjunctional composites respectively of the
classes ft* = (2R*t»0* & * s (%"#") * of functions Kl\\ij on the
component product ranges $P*$', %iÇ$j. The functional
operation J on the class ^ of the composite basis 26* ' " ' n is
the afunctional composite of the w functional operations J*
on the respective classes $1* or £** of the component bases
2*, viz., if the function K is the adjunctional composite of the
functions Hcij,
n

Now we see readily that the linear equation

(G) 1 — y — ZJKTJ

for the composite basis 2 6 * * ' ' n , and the system

n
(Gn1'"") f = y - 2 E W ? ( i = 1, - , n )

of n simultaneous equations for the system of component

bases 26*, are precisely equivalent. The functions £, ??, K
are the adjunctional composites respectively of the functions
V, V\ K\
Accordingly, the theory of the equation G based on 26
covers the theory of the system Gn1'"n based on the 26*.
Thus, if the parameter z is not a root of the Fredholm de-
terminant FK(z), the kernel K, the composite of the functions
nij, has a reciprocal kernel X, the composite of certain functions
\ij, and, corresponding to the equations
K + X = ZJK\ = ZJ\K,
we have the system of equations
n n
Kij + \iJ = Z ] T JkKikXkj = z ] T Jk\MKkJ (ij)f

and for the system Gn1 ' '% n we have the solution
n
rf = £*' - 3 2 JjXijf' (i = 1, • • -, ri).
7=1

For the adjoint system of equations

358 LINEAR INTEGRAL EQUATIONS. [April,

{Gn1 •••") & = tf - zZ JWj 0' = 1, • • •, n)

we have, with the same functions \ij, the solution
n

V = £ - * E J'ÏW (j = 1, • • •, n).
T 7 ^ Mixed Linear Equation.—Consider a basis S 6 with w
functional operations J i , • • •, Jn (instead of merely one) on
the class ^, and the corresponding mixed linear equation
n

Here the function £ of the class 3JÎ and n functions KJ of the

kernel class $ are given, and the function rj of the class 9ft
is to be found.
We may treat this mixed basis as a system of n bases S6*,
identical except that the functional operations J1 of the bases
2* are the respective operations Ji of the mixed basis SO.
Then the mixed linear equation n times repeated constitutes
a particular case of the system Gn1 " 'n of n simultaneous
equations on the system of bases Se*. Accordingly, if the
parameter z is not a root of the Fredholm determinant, the
system of kernel functions KJ has, with respect to the system
of functional operations JJf a reciprocal system of kernel
functions Xy; and we have the equations
n n
*3 + X; = Z 2 Jk Kkkj = Z ] £ Jk^kKj (j),
k=l ?c=L

and for the mixed equation the solution

n

3=1

11. ^-Composition. The Closure Property C±.

We have seen that the theory of the general linear equation
G based on S 6 has (closure property Ci) as instances the theories
of the equations (I, I I n , III, IV), and furthermore (closure
property C3) as instances the theories of the systems Gn,
Gn1 ' * 'n of linear equations based respectively on S 6 and on
a system of bases S6Z. Now the step from G on S 6 to Gn
on S 6 was analogous to and has as instance the step from I
to I I n . The question arises whether the general theory has
as instances the theories of equations arising from G by steps
1912.] LINEAR INTEGRAL EQUATIONS. 359

analogous to the steps from I to I I I , IV, and even to the general

equation 0 based on 2 6 . Under the postulates to be speci-
fied, this question is to be answered in the affirmative, and this
is the closure property (74 of the general theory.
Consider again a system of n bases 2 6 * having the same class
3Ï of all real or of all complex numbers, and otherwise conceptu-
ally, but not necessarily actually, distinct. From this system
of bases 2* we determine the ^-composite basis 2 1 * * *w* or 2*
as follows. The class 21 of 2* is the common class 31 of the
constituent bases 2 \ The ranges ty, $|3 of 2* are the product
classes respectively of the ranges $*, 9|3* of the bases 2 \ The
classes 9R, 5DÎ of functions on f$, Sp are the -^-composites re-
spectively of the classes 2tt*", SOI' on 5p', $*' of the bases 2*.
Then the product ranges $ $ , $ ^ of 2* are the product classes
respectively of the product ranges $*$*, $*$* of the bases 2*;
and the classes, $ = (3ÏÏ3JI)*, ^ = (9K$!tt)* of functions K, K
on ^3^3, $ $ are the -^-composites respectively of the classes
fl* s ( $ » % , «* s (3tt*2R% of functions K\ Ï* on $ ' $ ' , $ < $ *
of the bases 2 \ The functional operation J on the class K has
the definition

for every function Àc of the class ^ .

This -^-composite basis 2* satisfies the postulates laid
on the bases 2 6 , and accordingly the general theory of the
equation G based on 2 6 has as instance the theory of the equa-
tion G for the basis 2*, that is, of the equation

The -^-composition of the bases I I n and 2 6 leads to the

equation
n
1 (is) = v (is) —zJ2 J (tu) K (isjt)y (ju) (is),

which is, notation apart, the system Gn on the basis 2 6 . In

fact, the -^-composite of I I n and 2 6 is identical with the ad-
junctional composite of n systems identical with 2 6 . But
the general adjunctional composition of n systems possibly
distinct is not an instance of the -^-composition here defined •
360 LINEAR INTEGRAL EQUATIONS. [April,

12. Additional Definitions.

We are to specify postulates on the bases S 4 , S5, 2 6 enabling
us to secure general theories F, H. To that end we have need
of several additional definitions.
Consider a class 9JÎ of functions /* on the range 5)3 to 31.
We have already defined the properties linearity (L), closure
(C), and now define two dominance properties (D, Do) and
a reality property (R).
The function a is dominated by the function 0 in case for
every argument p \a(p) | ^ | p(p) |. The class 9JI has the
dominance property D0 in case every function /x of 3JÎ is domi-
nated by some real-valued nowhere negative function /x0 of
9JI; the function ju0 may vary with /*. The class Tt has the
dominance property D in case for every finite or infinite se-
quence {tin} of functions of 3JI there is a function /x of 3JI such
that the functions /xn of the sequence {fin} are dominated re-
spectively by certain numerical multiples anii of the function
/x, that is, for every n and p \iin(v) | ^ | #WM(P) |.
A complex number a has a conjugate complex number â.
A function a has a conjugate function a, whose functional
values a(p) are conjugate to the corresponding functional
values a(p), of the function ce. A class 9JI of functions ju has
a conjugate class 9JÏ consisting of the functions /* conjugate to
the various functions /x of the class 9JÎ. A number a is real if
a — a. A function a is real or real-valued if a = a. Simi-
larly, a class 9Jt is rmZ (JS) if 9JÏ = Sft. Thus, the class of
complex-valued continuous functions on a linear interval is
real. In general, a real linear class of functions is a linear
class of real-valued functions or a class of complex-valued
functions whose constituent functions have real and imaginary
components which range independently over a linear class of
real-valued functions.
Consider a functional operation J on a class 9JÎ of functions
fji. The operation J is linear (L) in case ix = aux\ + a2M2
implies J/JL = aiJm + a2JM2- The operation J has the mod-
ular property (M) in case there exists an associated modulus
M, viz., a functional operation M on real-valued nowhere
negative functions fi0 of 5DÎ such that M/x0 is a real non-negative
number, for which a relation | /x(p) | ^ Mo(p) holding for
every p implies the relation | Jfx | ^ M/x0.
1912.] LINEAR INTEGRAL EQUATIONS. 361

13. Postulates for the Theory F.

We secure the general Fredholm theory F of the adjoint
equations G, G based on the respective systems 2 4 ; 2s î S 6 by
postulating that the respective classes 9ft; 9ft; 9ft, 9ft have the
properties LCD D0, and that the functional operation J on
9t s 3»i; fl s (9ft9ft)*; I s (9ft9ft)* has the properties L M;
and for this theory we have the four closure properties
C\ Ci Cz C 4 ; the theory F based on 2 4 however lacks the
closure property (72.

14. Postulates for the Theory H.

We secure the general Hilbert-Schmidt theory H for the
complex-valued hermitian kernels K(H = K) based on the sys-
tem 2 4 or 2 5 , by postulating that the class 21 is the class of
all complex numbers, that the class 9ft has the properties
L C D Do R, and that the functional operation J on the class
9Î s 9ft; or fl s (9ft9ft)* has the properties L M H P PQ. The
operation J is hermitian (H) in case for every two functions
a, /3 of 9ft Jaj8 = J/to, from which, in view of the properties
of 9ft and the properties L M of J , follows the relation J v = J^
or JK = J K for every function v of 9Î or /c of $. The operation
J is definitely positive (P Po) in case for every function // of 9ft
the result J/x/x (for a hermitian operation J necessarily a real
number) is (P) a real non-negative number (P0) vanishing only
if ju = 0. Thus the operation J may be described as a <fe/£-
mtefo/ positive (PPo) foViear ( i ) hermitian (H) operation having
(M) an associated modular operation M.
The general theory H based on the system 2 4 or S 5 has the
four or three closure properties C\ Cz (74 or C\ C2 Cz C4, where in
O2 the function co of ^ is hermitian (w = co) and positively
definite, viz., for every function /x of 9ft J(rs)J«M)pt(r)(o(^)ju(w)
is a non-negative real number vanishing only if fx — 0.
If we postulate that the class 2Ï is the class of all real numbers,
we secure the general Hilbert-Schmidt theory H for (real)
symmetric kernels K(K = ïc) based on the system S 4 or S5, by
imposing the same conditions as before on the class 9ft and the
operation J . However there are certain simplifications. Since
all the functions and the operations J are real-valued, the
class 9ft is necessarily real (R) ; and the hermitian property
(H) of J is the symmetry (S) Jap — Jpa, holding necessarily
362 SHORTER NOTICES. [April,

for the theory based on 2 4 , and for the theory based on S 5 ,

implying JK = J/c; and the property (PPo) of being a definitely
positive operation J is that Jfxfx is (P) a real non-negative
number (P 0 ) vanishing only if fx = 0.
We have specified the bases or terminologies and the postu-
lates of the general theories F and H, and conclude this ad-
dress on the foundations of the theory of linear integral equa-
tions with the expression of grateful appreciation of your so
prolonged attention.
T H E UNIVERSITY OF CHICAGO.

SHORTER NOTICES.
Lectures on Fundamental Concepts of Algebra and Geometry.
By J. W. YOUNG. Prepared for publication with the co-
operation of W. W. DENTON, with a Note on the Growth of
Algebraic Symbolism by U. G. MITCHELL. New York,
The Macmillan Company, 1911. vii + 247 pp.
T H E book contains twenty-one lectures on the logical
foundations of algebra and geometry in substantially the same
form as delivered at the University of Illinois during the
summer of 1909, with an appended note on the growth of al-
gebraic symbolism. " T h e points of view developed and the
results reached are not directly of use in elementary teaching.
They are extremely abstract, and will be of interest only to
mature minds. They should serve to clarify the teacher's
ideas and thus indirectly serve to clarify the pupil's." " T h e
results nevertheless, have a direct bearing on some of the
pedagogical problems confronting the teacher." "Let the
teacher be vitally, enthusiastically interested in what he is
teaching, and it will be a dull pupil who does not catch the
infection. It is hoped these lectures may give a new impetus
to the enthusiasm of those teachers who have not as yet con-
sidered the logical foundations of mathematics." Such is the
purpose of the author.
The first five lectures, of 57 pages, form an introduction
which makes clear the nature of the problems to be discussed
and the point of view from which they are approached.
Euclid's Elements, a non-euclidian geometry, the history of
the parallel postulate, the logical significance of definitions,