NORTH-HOLLAND
MATHEMATICS STUDIES
Notas de Matematica
editor: Leopoldo Nachbin
J. DIEUDONNÉ
NORTH-HOLLAND
NORTH-HOLLAND
MATHEMATICS STUDIES 49
JEAN DIEUDONNE
Pro fesseur honoraire a la Faculte des Sciences
de Nice, France
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording
or otherwise, without the prior permission of the copyright owner.
ISBN • 0444861483
Publishers:
NORTH-HOLLAND PUBLISHING COMPANY
AMSTERDAM • NEW YORK • OXFORD
TABLE OF CONTENTS
INTRODUCTION 1
V
vi TABLE OF CONTENTS
REFERENCES 280
which deal with functions, but that would practically mean all
sisted of functions.
the old idea that infinitesimal Calculus was derived from the
1
2 INTRODUCTION
h z
11 1
= bl
h z + h z = b2
21 1 22 2
(2 )
h z + h z +...+ h z = b
nl 1 n2 2 nn n n
finite.
Its own evolution had been very slow and painful, stretching
tions, namely
INTRODUCTION 3
linear equations
determinants
matrices
would form the model on which all further advances were pat-
that the heart of the matter lay, not in the differential (or
The second step was taken by Hilbert in his 1906 paper, sub-
tial equations.
spaces for any exponent p such that 1 < p < +00, and
theory. The advances which have been achieved during the last
their generalizations).
Frdholm
integral equations I
Duality
\
4r
NW
i ‘
V
&
Algebraic .ronutative
Topology_ Riesz spectral Algebra
Topological Normed theory
spaces spaces Frichel
spaces ,,
Ai
1910-1930
Non linear
ODE and PDE
Locally convex
spaces Von Neumann
spectral theory
Commutative
.:titirmonic Analysis
1930-1945
1945
Linear
ODE and PDE Geometry of
s
Banach spaces Algebras of C -algebras
tmutiot
Non commutative
harmonic Analysis
Number
theory
CHAPTER I
a very hazy one. The domain where it was defined was very
(n)(n-1)\
(1) = F(x,Y9Y-' ,Y" ,.••9Y
of arbitrary order
9
10 CHAPTER I
also the solutions of (2) when the y k were known: the so-
n
lution is written in the form y E zkyk , where the z k
k=1
are unknown functions, subject to n-1 linear relations
linear equation
(4) E k'k
= b(x)
k=1
primitives.
(n-1)
where B is bilinear in (y,y',...,y (n-1) ) and (z,z',...,z ),
constituting a generalization of the classical "integration
order was only begun by Euler and Lagrange after 1770. Euler
order
z az
(6) F(x,y,z, )=0
"characteristic curves".
given by
argument which did not use equation (7), had concluded that
nux miact
u (x,t) = sin cos
n a
be written as a series
a
EZ- 2ux aux
(13) 2 + a 1 cos TTX
a
+ b
1
sin a + a 2 cos a + b 2 sin
a
his time, and no progress was made on this question until the
2 2
u u
2
_ 0
2
.)(.
2
u
(15) = 0
2 ay
.)(.
a , b
n n
TT
r rr
1
(16) a = f(x)cos nx dx, b = - f(x)sin nx dx
n 7 n Tr
-TT -TT
gence.
gral. But before 1920 there was not much contact between that
derstand it.
ential equation
2
u 2 ,11 \
(18)
at
"k 2
ax
+x
lu hu = 0
(19) for x = r and all t,
equation
Xr
Xr = 1-hr.
(21) tg
word)
r
day language).
tions
u1 - qu + Xu = 0, v - qv + 4v = 0,
one deduces
u v - v"u + (X-1.1)uv = 0
"
and as r b
(uuv-vnu)dx = (u l v-v'ud = 0 because of (25),
a
a
one obtains
b
substituting X and a
for [I and v in (26), one obtains
f:b
,2
lu(x)1 dx = 0, contrary to assumption.
The main contribution of Sturm was the proof that there are
ya +
(2 7)
q l (x)Y = 0, Y° + c1 2 (x)Y = 0
= -q(x)y
l' and then one takes as new unknowns two functions
r,9 such that y i = r sin 0, y2 = r cos e, which leads to
the system
where the second equation now is of the first order only (*)
tions
2 , 2
(30) e' = cos 0 + q i (x )sin 0, 0' = cos 2 0 + q 2 (x)sin 2 0
p i (x) < p 2 (x) for OC < x < b. The proof is very simple and
( * ) This
device seems to have first been introduced by H.Prflfer
[180].
20 CHAPTER I
to show that each equation (32) has one and only one solution
y = un , so that p
J_
is replaced by X 121 , he gives a sketchy
b
,
the condition u 2(x)dx = 1). This allows him to prove that
fa n
the series E c u (x) converges, provided the usual Fourier
n n
n=1
series of f converges. He still has to show that, if f
th
The study of celestial mechanics during the XVIII century
the sun, and then to find the deviations of the actual orbits
of other planets; due to the fact that the masses of the plan-
ets are much smaller than the mass of the sun, these deviations
(1) Y i = + E
2
f 2i (x,y1 ,...,yn ) (1 5 i 5 n)
series in e
2
(2) (1 5 i s n),
Yli + e Y2i +
22
THE "CRYPTO-INTEGRAL" EQUATIONS 23
yii = f (x a a )
' 1" n
y / . = F .(x y y )
2i 2i ' 11 ' ln
• (x ,Y 11 p• • • 9Y- O
Y'33. • = F 3i lnr 21 • • •tY)
2n
with observations.
ions)
p
(3
aU = E A i (t,x 1 ,...,x p ) , aU
x ,
)
i
i=1
24 CHAPTER II
U = u, and defining
o
It
(5 y = y + y +...+ y n +
)
1
equations
y (x) = dt f(s)Yn(s)ds.
n+1
a
from which he concludes that the series (5) giving y(x) con-
(6) dt f(s)y(s)ds
Y Yo
a a
x
(which he could also have written y = y o + I (x-t)f(t)y(t)dt),
a
thus giving what is probably the first example of what will
/t
r x
(7 z (x) = y + dt f(s)zn(s)ds
)
n+1
a Ja.
th
During the whole XIX century, the theory of partial dif-
where the right hand sides do not contain any derivative with
28 CHAPTER II
s7 6v 6v -‘1-
,_ 1I _ ,
-
2 2
2x2 ax3l 2x1 x - f(x 3 )
°IL c/x 2 , °
6v dv v
2 = - 2
x2 2xl 6x31 - 2x 2 )(.3
a X1
I
conditions.
2 2
a z
A(x,y) 2 + B(x,y) a x y +
ax
(9) 2
az
+ C(x,y) 14 + F(x,y,z,
az
wR, = 0
aY
by
2
a 2 z a 2 z
2 . The study of general boundary con-
axay or 2
ax y
ditions for hyperbolic equations only begins around 1860 and
1925 (see chapter IX, §5). On the contrary, the various prob-
2 2 2
a 0 0 a 0
(n ) AO = 2 + 2 2
- 0
ax ay az
(12) AO + 4up = 0
was well aware of this connection and of the fact that one
by other means.
theory of what Green for the first time calls potential func-
(10), but also what will later be called simple layer poten-
on their surface.
(15) (uAv - vAu)dw (v - u ,,)da
I)
an
v
where E is a smooth surface limiting a bounded volume V,
2 -
u and v are C in a neighborhood of V, KT
- is the de-
THE "CRYPTO-INTEGRAL" EQUATIONS 33
has been excised, and by letting the radius of the ball tend
,(\
(17) 47v(m) (v r \
_Y)da
with r(P) = MP)
r 6n
was in agreement with what was known at the time for partial
an
and -a-r-T on E arbitrary continuous functions, so that the
1
(18) I T fi a 2 2 v(P)da
v(M) = T
ar
studies, and we have seen that in his 1813 paper on the at-
( * ) The
fact that Gauss also uses the word "potential" with
the same meaning may be attributed to the fact that the
word (in its Latin form) was commonly used in the
XVIII
th century by "natural philosophers".
36 CHAPTER II
4up(M 0 ); all this had been taken for granted without proof
by Greea.
lution.
on E), for which the integral (21) takes its smallest value;
an
-1/MP on E; the Green function is then G(M,P) = u(M,P)+(l/MP),
provided one shows that and is continuous on E.
38 CHAPTER II
(*)
The discovery of that fact is usually attributed to Hada-
mard, who published a similar example in 1906 194,vol.III,
p. 1245-1248].
THE "CRYPTO-INTEGRAL" EQUATIONS 39
using new ideas in what was called his "direct method", was
to show that when one can solve the Dirichlet problem for two
tential:
( 22 )
u(M)
IC p p ( ) ( 2.,
4 D) do
an
however, — s ithe same on both sides.
an
2np n (m) + a 1 1
MP) ug = 0 for n a 1
pn-1 (P) 8n (
(26) o0 )q n-1
L 5 (L
nn
1
t o-1 5 pn-1 (P) < (Ln_, + t n _ i ) for B
n
(L
n-1n-1
)(A 04 B n (m)) s 4710 n (m) s (Ln-1n-1 X-LA
7
ns' l- 1 2 n
00+Bn(M))
44 CHAPTER II
where A (M) and B (M) are the solid angles from which A
n n n
and B are "seen" from M. This implies
n
Lnn 5 (L -4, )q
n-1 n-1
and also
valent ways:
cludes, from the fact that A(M) + B(M I ) > 0 for all choices
is satisfied for all these choices, and does not give a proof
of that assertion valid for all convex sets other than double
apparently satisfied with the fact that for simple convex sets,
book.
CHAPTER III
(1)
2
u 8
2
u a 2u
2 2 - 2
aX by bt
2 2
b v d v2 + Xv = 0 •
2 +
8x by
2 2
(4) F(v) = fI (( a--- ) + ( s7 )
; )dxdy
2
for C functions v in 5, equal to 0 on E and sub-
(5) v2 dxdy
v , = 1.
0
THE EQUATION OF VIBRATING MEMBRANES 49
ditional condition
But he does not try to prove that lim X n = +co, nor that
n4=
functions in 3 1 possess a "Fourier expansion" 7 c nvn de-
n
fined in the same manner as in the Sturm-Liouville problem
without proof).
2 2
v v 2
(7) 2 + 2+X.
--T pv = 0
x ay
lues > 0; his arguments apply for any such function, but in
Weber.
the equation
(8) Aw + F pw = 0
,
n
w = w Iii
(9 )
o + - 1 +...+ wn +...
he takes for w
o the constant function equal to 1, and im-
poses on the wn for n z 1 to vanish on F; they are then
(10) Aw n + pw
n-1 = 0 for n z 1.
equation
Aw + f 0
(13) 1 P(P)wn-l(P)G(MtP)dW.
wn(M) = Ti7
( if fg dw ) 2 f f 2 dw ) g2dw )
D D D
from (13)
2
(14) 4r 2 (w n (M) ) 2 5( f I( p 2 (P)G 2 (M,P)dw) ( 1 I w -1(P)dW)
n
5 A( 1( [ w
2-1 (P)dw)
D
(*) That inequality had been discovered by Buniakowsky in 1859,
but does not seem to have been noticed nor used by many ma-
thematicians before 1885. If is of course a direct generali-
zation of the corresponding inequality for finite sums, which
goes back at least to Cauchy.
CHAPTER Ill
52
(15)
W
n,k = fi pw w
k n-k
dW
(16) wn k
( awk+1 awn-k awk+1 awn-k
)dxdy.
, 6y
fC
2
(17) W 5 W W
n n-1 n+1'
then from (14) that the series (9) is absolutely and uniform-
ing (9) termwise, and that w then satisfies (8) and is equal
to 1 on the boundary r.
But Schwarz goes one step further. He proves that when
(18) Aw + (l/c)pw = 0.
THE EQATUION OF VIBRATING MEMBRANES 53
(1 9) v(M) = g(M) + p(P)G(M,P)v(P)dW
with g(M) = fr G(m,p)p(p)dw.
te) p i ^ ... ^
2n > 0, where each p n is an eigen-
value counted a number of times equal to its multiplicity;
w = E +w /
o dnepn
n
m m
U .14 = E p d p
n n n
n
and therefore the Schwarz series (9) is equal to
05 dn
w= E E( m=0
o = n E g m4 rinl )d nP n = E 1..kin cPn
m=0
-1
provided I< Pi and we have w = U•w + w
o
- w' . For
= 1/4 i ,
m m m m
U 'w o = E m 4 n d n cr) n = E ) d p
n /la 1 n n
m-k 2
W
m
/
= iu .w
o iw o
) = (U •w o IU k •w o ) = E 4 md
n n
that it has a simple pole with residue - 1 .(7 1 (M) at the point
He limits himself to the case in which p = 1 and r is
U
1
w - + v
1- (A1)
v = v
o +
(20)
1 +...+ nv n t
v0 - l Ul = °, Avn + v n-1 = 0 for n z 1
au
(21) a2Au
a Au
at
surface E of V
au
(22) — + hu = 0
an
(23) Av + Xv = 0
(24) an + hv 0 on E.
av 2
(25) F(v) = h ff v 2 da +fff (4)2 2 )c110.
V
58 CHAPTER III
H. Weyl and R. Courant, and even now has not entirely lost its
ff
a v 2 ay 2
( ax
av 2
( ) (—
a z) )
(26) iffy
2
v dw
2
where v is a C function in V, subject to the condition
(27) if v dw = 0
THE EQUATION OF VIBRATING MEMBRANES 59
lower bound
(28) C•vol(V)/(diam(V))5
what we now call "a priori" inequalities (cf. Chap. IX, §4).
conclusion.
in writing
CO
2
(29) u(x,y,.z,t) =n.E,c n exp(-X n a t)v (x,y,z)
n
=ifi
n=1
fv dw. But Poincare, no more than Weber, is not at
cn n
V
that time able to prove that this Fourier expansion converges
S = f ff (u - E c k exp(-X k a 2 t)vk ) 2 dW
n
k=1
V
2 ,
satisfies an inequality S 5 C•exp(-X a t) where C is
n n+1
independent of n and t; in other words, for t > 0, he
(* )
The method of least squares of Legendre-Gauss had led
Tchebychef to define a "best approximation" to a function F,
N
by a linear combination E a.*. of given functions *.
j=1 J J
(15j5N), by thE condition that N
E p(xk)(F(xk) - E a 4 41 i (xk )) 2
k=1 j=1
be minimum, for given points x k (15k5n) and given "weight"
P. Gram, in 1883, generalized the problem by considering
instead of a finite sum, an integral
ra
N
( )
p(x)(F(x) - E a,*i (x)) 2 dx
j=1
and he solved the problem in an original way, by applying to
the * . the "orthogonalization process" usually attributed
to E. Schmidt [89]. He was thus reduced to the case in which
the *. form an orthonormal system for the measure pdx),
whereheshowedthatthea.gi ving the best approximation a
(b
are the "Fourier coefficients" p(x)FIXIII.00dx. He went
Ja.
on to consider an infinite orthonormal system (I ) and in-
n
vestigated under which conditions the minimum value • n of
the integral (+) tends to 0 when n increases to +co; he
was able to see that this was linked to the "completeness" of
the system ('' n ) ' i.e. the fact that no function other than
the constant 0 is orthogonal to all * . It is unlikely
n
that Poincare had any knowledge of Gram's paper.
THE EQUATION OF VIBRATING MEMBRANES 61
fies (23) and (24), use of Green's formula shows that there
ing at the same time the existence of the long sought eigen-
( 31 ) tv + gv + f = 0
he writes it
n
v = Ef,g] = v + gv
o 1 +...+ g v n +...
for small
n
(32) w = [m f + • .+apv p-2'
2 vo +.
=w +w 1 +...+ wn- +...
1 o
Next, applying his lemma for the evaluation of the Schwarz in-
u ] one has
j = [vj-2"
a l v + a 2 u 2 a = w
P uP
v u2 = v o
-
• u - u = v1
(33) 2 3
u v p-2
p-1 - up =
(34) v = P/D
with
al a2 a 3 ... a p-1 m
p
1 0 ... 0 0
D= 2 , dp-1 p
0 1 0 0 = a p -a p-1 + p-2 -...+k-1) a1
0 0 0 1 -
and
64 CHAPTER III
w '2 ... ct ft p
3 p-1
v -g 0 ... 0 0
o
vl 1 -g ... 0 0
P
v p-2 0 0 •• •
n
35) P P + P +...+ P +...
(
o 1 n
the whole complex plane, with simple real and positive poles
where c is a constant.
questions:
2
absolute convergence when in addition Af and A f also
vanish on E.
for the Laplace equation with that new boundary condition, i.e.
what is now called the Neumann problem for the Laplace equa-
(37) f f uAv dW + f
V
i gvdcr = f + hv)uda
solution.
ence than the others. From his work both on the Dirichlet
b1, ?I
(39) an (sX i x1 1(s+) = 0;
'
which forms the central part of his 1896 paper; but the only
the Cramer formulas gave the unique solution when the deter-
minant of the system was not 0, but not much effort was spent
(1 ) y E a jk x k (1 s j 5 m)
k=1
th
had been familiar since the XVIII century (mostly for
Between 1850 and 1880 are proved the main theorems of linear
( )
Jordan was not dealing with matrices having elements in
E or 0, but with matrices having elements in a finite field
([123], p.114-126).
74 CHAPTER IV
of linear and multilinear Algebra had been found but were ex-
(* )
In 1896, Pincherle reinterpreted Weierstrass's results in
terms of endomorphisms(C173], vol. I, p.358-367).
(**)
In modern linear algebra, the space of endomorphisms of a
finite dimensional vector space E is identified with the
tensor product E * ®E, whereas the space of bilinear forms on
EXE is identified with E*®E*.
THE IDEA OF INFINITE DIMENSION 75
very end vector spaces and linear maps; in other words, the
§2 - Infinite determinants
(2) 1 = E , a m cos(2m-1)y
m=i
1 = E a
m=i
m
,
2
0 = E (2m-1) a
m
(3) m=1
0 = E (am-1) a m
m=1
(k)
a
m+1 2m-1 m+k
(k) 2m+1 m-k
a
, m-1 /
which gives him am = (-1) 4/7k2m-1); when later in his
as
(4) I a. x = b.
jk k j
(j = 1,2,...)
k=1
(*) During that period, Fourier's method was used in two lit-
tle known papers, one by Flirstenau in 1860 on the computation
of roots of an algebraic equation, and another by KOtteritzch
in 1870, for a system (4) in which the a jk are 0 for j > k
(see [184], p.8-12).
THE IDEA OF INFINITE DIMENSION 77
CO
z
the infinite product F(z) = 1 I1- a—) was convergent for
k=1
all complex numbers z.
+0D
(5 ) + ( E O n e nit )w = 0
n=-m
series
+m
i(n+c)t
(6) w = b e
n=-00
+co
2
E 0 b - (n+c) b -m < n < +m.
(7) n-k k n = 0,
k=-m
thata ......1 for all j (one can always reduce (4) to such
JJ
a system by dividing the JJ when
and from the assumption on the diagonal terms, one has also
(8) E a. x = b.
jk k j (-m < j < +m)
k=-m
(*) One must beware of the fact that the Fourier method (when
no condition is imposed on the a jk ) may very well give con-
vergent "infinite determinants", but the values given by the
Cramer formulas may be such that the left hand sides of (4)
are divergent series. An example is given by taking a jk = 0
if j > k, a. k -, lifj kp la.=( -1 ) j ; one finds as a "solu-
j - k
tion" x k = 2(-1) .
THE IDEA OF INFINITE DIMENSION 79
n S S
1 1 sls2
A n =l+ E c ss +
2 E
1!
s=1 sl's2
s s s2s2
2 1
(9) s s
11 sls2 sls3
1
+ s2s1 s2 5 2 s
2 3
sl's2's3
s s
31
6 3 6 2 cs s
3 3
most of the time. Even more widespread was the use, since the
th
XVIII century, of sequences of functions, or of functions
p.276).
ce, which was only quite cleared up around 1850. This was
th
followed in the last third of the XIX century by a deeper
[ 8] .
ly in a special case (he does not quote the Italians) and used
(loc.cit., p.15-37).
gies" was so to speak "in the air", and ready to blossom forth
simple terms. * )
into another set of functions, is also much older than the ge-
bitrary set, which does not seem to have been formulated be-
fore Dedekind's famous "Was sind and was sollen die Zahlen",
aD
by Lagrange as a relation Y(-a) = e between operators, or
n n-1
the factoring of a differential polynomial D + a1D +...+a
with symbols rather than with numbers, and later with the
(see [54], chap. XIII, §III); but they had no perceptible in-
xl-=
f A(x,y)cp(y)dy, where r is a curve in A and A is
(* )
ce), showing that they are integral operators of the form con-
sidered by Pincherle.
(**)
See A. Weil, Oeuvres Scientifiques, vol.II, Commentaires
sur [195 2 e], p.532 of the correct edition (or vol.III, p.450
of the first printing), Springer, Berlin-Heidelberg-New York,
1979.
(***)
This can be said of practically all mathematicians
before 1906.
THE IDEA OF INFINITE DIMENSION 87
and 435 453] and [146] ). But these ideas have not, up to now,
-
The urge to deal with "infinity" has been present from the
ential equations
dv.
dtx + E C ik (v.-v k 1 3_
(10) ) + C.v. = 0 (1 5 i 5 N),
90 CHAPTER IV
2 2
1(111,u2,...,uN) = E ik k + E C.u. .
2 2
(12) = 1 e4C1 1 I‘Tel)N
(13) 2 2 2- 2 2
u +u +...+ uN pi+...+pN
1 2
ly to the quotient
fff V
(e)2 + e) 2 + 0) 2 )dw + hifv 2 dcy
iff v dw
THE IDEA OF INFINITE DIMENSION 91
tothec io j ,theorthogonalityofthep.corresponding to
the relations
i(r
! U U dw = 0 for P
V P q
between the Um. Finally, he realizes that the same ideas
§2).
tion (34) and chap.II, §1, equation (6)) and by Beer and
0
and the problem consisted in finding f when the transform
2
(15) —
f(x) - 7 p(x)cos tx dt
0
where, as usual with Fourier, both formulas are obtained by
1890 (*) ; but it was only in 1894 that Le Roux attacked the
mes that h(y) = H(y,y) does not vanish in [a,b], takes the
(y
(19) h(y)p(y) +
ap (x,y)p(x)dx e(y)
a
and then applies the method of successive approximations which
*)
As these conditions are not satisfied for Abel's equation,
Le Roux's results (which for him are auxiliary properties
which he needs in a study of partial differential equations)
do not directly generalize those of Abel.
94 CHAPTER IV
the solution
iy
CO
(Y) 1
(20) cl9(Y) ( E S.(x,y)W(x)dx
h(y) h(y)
is i=0
the system
b = a x
1 11 1
b2 = a x + a x
12 1 22 2
(2 2)
b
n = a ln x 1 +...+ a nn xn
which the a id
. . and a.. are the analogous of H(x,y) and
11
yk = a + n (b-a) for 1 5 k 5 n
THE IDEA OF INFINITE DIMENSION 95
(23) F = E Si
o
i=0
(24) F -S = S *F
o o o o
(25) S
o
= E (-1) iF.•
i=0
(Y
(27) P(Y) = f(Y) F (x,y)f(x)dx
o
Ja
formulas.
CHAPTER V
of all the ideas and methods which had been slowly accumulat-
th
ing during the XIX century and which we have described in
§1 - Fredholm's discovery
97
98 CHAPTER V
fore him was in fact extremely simple (much simpler than any-
equations).
and C75]).
that from the start, he brushes aside all the particular fea-
terra had done with Abel's equation (chap.IV, .54)) begins with
tems of linear equations and starts right away with the for-
simple ideas:
(2)f(Y.)+ X(13-a)
a ;KOr k ,y.)f(y0=q)(y.)(1 5 j 5 n).
k=1
,y k ) K(y k ,y k )
K(y k i l
1 2
2/ \
X(ba) X k b-a)
1+ E Kk/ y k ,y k ) 22
k 1 2 !n kk
2
,y k ) K(y k ' y k )
K(y k 2 i
2 2
equation (1)
100 CHAPTER V
b b b
s1
A(X) = 1+X K(s, ․ )ds + K( 1
s
'a
(3)
+
Xm r
I •••
)a
K( s 1 s 2 s
m ) ds 1 ds 2 dsm +
a sl s 2 sm
X X X , 000 X
K( m) = K(s,t) K( m)
x x xi xm
1
x ". x x3 x
1 x2 X1 x2
(6) - K(s,x 1 )1i( m) + K(s,x2)K(
t x
2 ... x m t x ... 'c m)
1 3
x i x 2 ... xm
- + (-1) m K(s,xm )K( )•
t x x
1 m-1
A(s,t;X ) = K(s,t) + X I a
b
K(
S
t
X,
—I
He then introduces the function
b
102 CHAPTER V
b
(
(s) + X K(s,W 1 (t)dt = 0;
1 0
a
lution!
s s s
1 5 1 5 2 5m 1 2 m)
Ak ;X) = K(
t i t 2 tm t1 t 2 tm
(13)
m(b si sm x i x
K( dx dx .
ni x n) 1 n
n=1 • a la t t x
1 mln
, s2 ••• s m s2 s 3 ••• s m
K(s l ,t 1 )Ak ;X ) — K(s1,t2)A( Ov•)+.
t 2 tm t t ... t
1 3 m
and
'1
S i Sm r
S2
b
'm
A( ;X) + K(g,t 1 )A( ;04 =
t 1 ... t m 'a t 2 tm
(15)
s s
2 • m , s1 s3
= K(si,t1)A( ;X) - K(s2,t1)A k
t 2 tm t2 t 3 tm
1 9) S S = SR S XK = Td
(
XK R
and Fredholm has thus shown that the necessary and sufficient
neralizes (11) to
-
ib b si sm
dmA(X) I
(2 0) • • • (
I ;k) ds i ds
m
dk m 'a
a A s l sm
for which he shows that they are linearly independent and that
(23) " = A A
AK K K
bounded any more, but such that (x-y) M K(x,y) remains bounded,
(*)
It is reported (by Hellinger) that Hilbert inaugurated a
session of his Seminar by announcing the development of a
method which would lead to the proof of.the Riemann hypothesis:
the problem is to prove that a particular entire function has
all its zeroes on the real line, and Hilbert hoped that this
function would be expressed as the "determinant" of an inte-
gral equation with symmetric kernel. However, nobody has yet
been able to find such an equation.
(24) K(s,t)x(s)y(t)dsdt E I
1 ocipflik Nt y lI p fli
n 'n - -
a a
•1)
(26) f(s) = K(s,t)g(t)dt
a
for a continuous function g, then the corresponding "Fourier
expansion"
"Parseval identity"
(b
2 2
(28) f(s) ds E (flm n ) .
Ja.
n
(in the sense of mean square value, or, as we would now say,
(26).
(* )
Some of the results of E. Schmidt were also obtained in-
dependently by W. Stekloff [204].
THE CRUCIAL YEARS AND THE DEFINITION OF HILBERT SPACE 109
kernels
b
K
(31) Km(s,t) =
m-1 (sK(t)(1.; for m > 1, with K 1 =K
/a
eigenvalue of K 2 .
them for the time being, as well as most results in his two
last papers on the subject (see chapters VII and IX), to con-
and one of the best papers he ever wrote. By the depth and
i
(32 ) a a
b b
b = cp(s)w (s)ds, x = f(s)w (s)ds
P P P P
fa. a
equations
00
(33) x +Ekx= b (p=1,2,...)
P q=1 Pq q P
The new twist is that, due to the Bessel identity, one has
b
2
E k (s) K(s,t)2dt;
a
(24).
2
Of course, even under the restrictions E x +co, E y 2 <
P P
< +m, the right hand side of (35) is usually meaningless( * );
n n
(36) Kn(x,y) =E Ekxy,
p=1 q=1 Pq P q
(* )
To this rather awkward formulation, Hellinger and Toeplitz
0106] substituted the consideration of "infinite matrices"
(k ) , and of their "calculus" inspired from Fro-
pq lsp,q<+m
benius, but without associating an endomorphism to a matrix.
THE CRUCIAL YEARS AND THE DEFINITION OF HILBERT SPACE 113
2
with E x < +co, he associates the sequence (x / ), where
P
(37) xl E a x (p=1,2,..0)
Pq q
E a2 . 1 E a a
pq pr
0 for q r
q
(38)
E a2 . 1 E a a
pq nq
= 0 for n 4 p
q
as a "Faltung"
(39) K F(x' ,y') K (.,.)0 (., x')0(. ,y' )
K(x,y) =
1 xiyi +
1 1
(40) x2y2 xnyn +
1 '2
will later be called the compactness of the unit ball for the
sesquilinear forms
Pq
but one step removed from what we now call the theory of
th
Until the middle of the XIX century, mathematicians had
did not lie in the nature of the objects, but in the rules to
(42) (xla ) = °n (n=1,2,...)
2
where the a n are arbitrary vectors of f and the c ar-
n
bitrary complex numbers.
with
THE CRUCIAL YEARS AND THE DEFINITION OF HILBERT SPACE 119
(ally (a l la 2 ) (allan)
(a 2 la l ) (a 2 1a 2 ) (a2lan)
Dn =
(44)
0 c ••••
c1 2 n
cl
An = D
n
e
n
of Hilbert.
process).
e([
a,b]) of all bounded integrable functions in [a,b] (con-
changing U.
2
Once the L spaces had been defined, it was a natural ge-
§2).
p 1/q
(4) 1 E akbkl s( E la k )l/p ( E lb l q )
k for + - = 1
-13 q
k=1 k=1 k=1
l/p
(5) ( F la k +b k I P ) 1/13 ( E 1 a k 113)1/P ( E lb I P )
k=1 k=1 k=1
/
(7) ( f lf(x)+g(x)1 P dx) 11 13 5 ( If(x)Ipdx)1/13+(fI
g(x)IPdx)1/P .
equations
(8) f(x)ga(x)dx c
II a
2
gent for all sequences (x n ) in t , then (a n ) itself be-
,
longed to t 2 L106]; Landau proved that, more generally, if
f E L P , then necessarily g E L q .
1/q
(9) E Raca l s M•( E IX a ga (x)I q dx) •
aEH I aEH
remains bounded for all f mch that If(x) IPdx 5 1), the
I
transposed mapping T' defined by the equation
F. Riesz had thus given, for the first time, examples of what
of linear equations
(
b
(12) g (x)4(x) c
a
a
where [a,b] is a compact interval in R, the g are con-
the same problem was later considered when the interval [0,+=[
[ 3] •
(14) IEXcl
aEH a a
M'suPIE
x aEH
,,
a ga ( x )1
(he explicitly observed that the right hand side of this ine-
replacing OM by C([a,b]).
F. Riesz. (*)
subspace E there has been defined a norm Nx11 (he does not
use that name nor the notation) such that: 1) 114 0 and
(x x )
of real numbers u (u
1n ) and a point x
1" n'
n=1
The central problem in Helly's paper is the solution of a
system
(15) (u (V) ,x) = c (v=1,2,...)
given above (for a single equation) shows that there may well
is satisfied.
u E E'.
To treat problem A), Helly assumes the additional condition
1E%c+EtlY+ y
v v v v m+1
v1 v1
(18)
proving that any three of the disks (18) have a common point;
DUALITY AND THE DEFINITION OF NORMED SPACES 135
llull = lu l l + E uk l
is finite, Hull being the natural
k=1
norm on E' ; then if one takes L(u) = lim u k , L is contin-
1(4=
uous on E' but there is no p E E such that L(u) = (p,u).
complete spaces.
( * ) F.
Riesz had already observed that one could define on the
r
space of functions of bounded variation co ntinuous
linear
b
funcionals which were not of the form g,* f(x)d (s) for a
a
continuous function f for instance one ca_a take for f an
increasing discontinuous function ) ([183], vol.II, p.827).
136 CHAPTER VI
(.)
Transfinite induction had been used by analysts ever since
Cantor, but the application of transfinite induction closest
to Hahn's is probably the method by which Banach, in 1923,
had proved the existence on R of a "measure" defined on all
subsets of R and simply additive [13].
DUALITY AND THE DEFINITION OF NORMED SPACES 137
paper of Hahn, duality theory at last has come into its own.
Hahn's paper, published the same theorem with the same proof
138 CHAPTER VI
29 E lu(xi)I 5 1 ;
j=k+1 nk
k-1
39 lu (x k )1 k + E lu n (x j )1-
nk j=1 k
-k \ -1
B : 5 2 • inf
i (Hu h+1)
k n i
j<k
in E; the assumption that (Hu II) is unbounded guarantees
n
the existence of an index n and a point x E B for which
k k k
condition 3s holds; conditions 12 and 22 are then deduced
S (g) for the sum of the first n terms of the Fourier se-
n
ries of a continuous function g, it is possible to find a
"kernels" K
rb
n insuring that the integrals I f(t)K n (t,x)dt tend
a
to f(x) when n tends to +=, for various kinds of function f.
showed that the assumption that the norms (x)II are bound-
n
ed for each x by a number depending on x, implies that the
(* )
For n 1, the same result had been proved two years
earlier by W. Osgood [170].
142 CHAPTER VI
x E E, sup Ilu(x)11 < the function p(x) = sup Hu(x)II is
uE H uE H
lower semi-continuous, and from the Baire theorem it follows
ing results have become two of the most powerful tools in all
tional Analysis the same impact that van der Waerden's book
DUALITY AND THE DEFINITION OF NORMED SPACES 143
144
SPECTRAL THEORY AFTER 1900 145
one (for the strong topology). Now this can be defined for an
4 years later!).
only locally compact ones: one has only to cover the ball
-1
a) the kernel v (0) has finite dimension;
k
The next step is to consider the iterates v of v, the
the F
k a decreasing sequence of closed subspaces of finite
codimension. F. Riesz shows, by contradiction and using re-
Volterra operators.
on integral equations.
1/a, and this had led Fourier to consider that the 'limiting
sentation by an integral
00
( 4-°3
(1) f(x) =,11 dt f(t)cos u(x-t)du
, -m 0
where the "eigenvalues" would now fill the interval [0,+m[
equation
that
u l( o,x) = 1, 11'1(0,X) = 0
came from it; but it is a far cry from the vague ideas of
P (z) n M.
2n+1 E
(z)
Q 2n+1 i=1 z+x.
where the M. and the x. are real numbers, and he had shown
re°
d“u)
F(z) =
z+u
SPECTRAL THEORY AFTER 1900 151
n n
(4) K n (x,x) .E E k x x
p=1 q=1 Pq P q
2
To each x = (x p ) of f such that x = 0 for p > n,
fact, he only uses these limits for the points x(pp) having
+ + /
v (x;X) = E v (X)x p x cl , v(x;X) = E v - (X)x x
Pq P,q Pq P q
P,q
e(x;X) = E Er(x)
X r <X
and
and the complement of the set where all the c(x;X) are cons-
(10) ( ba P
1
if 4 > 0, one gets a = if p+q 0, a Pq = 0 if
pq p+ q
q = -p. Making x = 0 for p 5 0, one gets the second
x x
example of Hilbert A(x,x) = E p q , and making x = 0
p,q=1
for p 5 0, y = 0 for p ^ 0, one gets still another
p
x y
example of Hilbert A(x,y) = E' P q (where the summation
P,q
P-q
extends to the pairs (p,q) of integers >0 and distinct);
E anx n 2 F b
- in an infinite system of varia-
0 n+1nn+1
n == n 0
bles. What Stieltjes had done was to study directly unlimited
continued fractions of type (F), representing them as
"Stieltjes transforms" and being led to his "problem of
moments" by the problem of determining the Stieltjes measure
corresponding to a given Stieltjes transform; but he had not
considered the relation between the continued fraction and
quadratic forms. This was done by Toeplitz in 1910 [214] for
the case of Jacobi bounded quadratic forms, and later extended
to the general case; it turned out that these forms exactly
corresponded to spectra of multiplicity 1 [207].
SPECTRAL THEORY AFTER 1900 155
s M•1E .
But F. Riesz goes further. If (f n ) is an increasing se-
(* )
This argument is not the one used by Riesz, who deduces
the result by a passage to the limit from the known result
for the "Abschnitte" A of A.
for z g, e„ (A) = A5
„ is defined since e,
is bounded and
5 5
lower semi-continuous. For any pair of vectors x, y the
f(A)) =
r +c° f(g
(13)
-co
-c .1
E 5 f(A) - Ek f6-10( A E - A ) s e-1
E
'k+1 gk
(with^, k
"lc I•c+1)
where A is constant.
not in the point spectrum and is the direct sum of the (ortho-
Xx. - E a = 0 (i=1,2,...) .
ik x k
k
ce, for the form (11), where there is no point spectrum and
chap. IX, §2); but in the case of the form (11), one could
analogy with the finite dimensional case: can one give ne-
his Seminar that the Fredholm theorems might fail when the
5 1.
equation
(* )
This is apparently the first appearance of a relation
between Fourier transforms and Functional Analysis (see §6).
162 CHAPTER VII
K(t,t+), -
that a
-- a K(t,t-) = -1. A routine calculation
x 6x
then shows that in order that a function y be a solution of
b
i
(19) y(x) = - K(t,x)f(t)dt
a
and therefore the solutions of (16), satisfying (17), are
where
the Green function for the Laplacian (chap.III, §1), and the
(23) L(y) Xy = 0
SPECTRAL THEORY AFTER 1900 165
tial conditions
u u2(0) = 0, p(o)u2(o) = 1
1 (0) 1, p(0)u'l(0) o,
L(y) + iy = 0
so that v is a solution satisfying the boundary condition
from 0 to 27.
,
kernel K P o kt,x) which always satisfies the two conditions
(26) rm K c) kt,x)fkt)dt
\ \
0
is then a solution of
mity 0, and the condtion that the real part of (28) vanishes
at extremity
mities.
r
(s,t) E AnXAn and to 0 outside, has a finite integral
b
,2
1K (s 01 dsdt. If one now considers the integral
n '
la
' ab
equation
(b
tions
.1)
Cauchy-Schwarz inequality,
b b
2
(32 ) IT n (s)1 2 ds 5 41 " 2 If(s)12ds.
IX -5;1 C
to
(33) T *f - % (f + U •f)
X
2
in L for non real X, which seemed to contradict the
context.
popping up every few weeks from all over the horizon [121].
(*) As late as 1924, most physicists did not even know that a
finite matrix was!
CHAPTER VII
172
approach.
and methods.
brought back to the Hilbert theory. But von Neumann had the
§5). Of the two examples given in 1), the first is closed but
that dom( T
2 ) = (0). However, using the decomposition of
r( T) and J(r( T*)), von Neumann could prove that for any
CO 3 1] .
point spectrum of T.
( +w
x 2 d(11E(X) • 4 2 )
00
is finite;
( '
as Stieltjes integrals.
by (1
fined above.
,+
Furthermore, if 1 ,, is the orthogonal supplement of F in
may be given in which they take any integral value or are in-
finite.
sions.
the "limit circle" case, and (1,1) in the "limit point" case.
f
b
integral k(s)lf(s)Ids is finite; S is then the adjoint
a
of that operator, which explains the existence of non trivial
x
operator U , suitably restricted, coincides with the
general context.
§5 - Banach algebras
We have seen (§2) that F. Riesz probably was the first ma-
vol.II, p.177-212).
But for von Neumann this was only a beginning. The period
too far away from our main theme to describe in some detail
of this program for what we now call the von Neumann algebras,
most profound and most difficult which von Neumann ever wrote
for algebras of Type II, may be any real number (in [0,1] or
Sp( N ).
theory" of Hellinger-Hahn:
G
k =
P k (E) H
lk
@ H
2k
H
kk
are. The equivalence class
NI--->UNU 1 . -
set e sub-
T .f(N )T -1 .
3
For self-adjoint operators A in E, the connection with
§6 - Later developments
lopments will have to wait till more time has elapsed and has
SPECTRAL THEORY AFTER 1900 191
books.
fined in C
k
. For a general commutative Banach algebra A,
is a regular algebra.
HI = and B
o is also the closure in C(U) of the alge-
19 4 CHAPTER VII
C) Harmonic Analysis.
+=
Similarly, from the formula f(x) = E a(n)e nix , one obtain-
n=,-=
ed
SPECTRAL THEORY AFTER 1900 195
function
-27ixt
(42) 3f( x) = e f(t)dt.
g = g 1* g 2' defined as
+co
(44)
3`(gl*g2) = 3g1.3g2 [148] .
196 CHAPTER VII
ed independently by Cauchy)
and to connect them with the notion of group, via the concepts
matrix U(s)).
(46) ( E s) (E r s) = E g r lt st.
sEG s sEG s (s,t)EGxG s
C[G]-module by
k
(49) *m = a m for 1 5 p,q,r,s 5 n k
mkq
p rs qr ps
k k'
(5o) m *m if k k'.
p q rs . 0
-1
mk.(s ) m.k .(s ) for 1 5 i t j 5 n k , s E G
3 1 13
k k'
m ks)m ks) = 0 unless p=r, q=s, k=k'
P H r,s
sEG
k
E m (s)m k (s) = n k Card(G) for 1 5 p,q 5 n k
sEG Pq PH
2 2 2
(orthogonality relations). One has n + n Card(G)
2 +...+ n h
G
and the expression of any f E C with respect to the basis
( k
.) of that space,
13
k k
(54) f(s) E
13 1J
i,j,k
, , k
(55) c (n k Card(G)) -1 E fks)m . .(s).
ij
sEG
1 k
If one writes M
/ -
(s) the matrix kn m. .(s)), one
k nkxnk k
has
(56) M (st) = M k (s)Mk (t) and M k (s-1 ) = (M k (s)) *
k
near group GL(N,C) which are such that the elements of U(s)
(see [53]); but it is clear that for such infinite groups, all
written M
k
(s) where M (s), as in (56), was a unitary
k
matrix; furthermore, he proved the relations which he rightly
k /
(57) m qks)mks(s)ds = 0 unless p=r, q=s, k=k'
P r
IG
as (54) were out of the question. But for the group SO(1,10=
Y
(61) U(f) = (U(f)) * .
one has h*g E C(G) and h*g 0 for suitable functions h. (*)
of G onto G C216].
204 CHAPTER VII
self, and (62) was just the definition of the Fourier integral.
,, ,
(63) If(x)12 dx = Igf(x)1 2 dx
commutative groups.
algebras!
did not stop with commutative groups. One can still define
modular groups (i.e. those for which left invariant Haar mea-
XE E and y E E
f(s)(U(s)•xly)ds
( 6/i) (1/0-1'xIY) =
G
where ds is (left and right) invariant Haar measure on G.
tisfy (61) and are non-degenerate (i.e. such that the V(f)•x
/
for x E E and f E L 1 (G) generate a dense subspace of E).
sult for all locally compact groups, and the general theory
sion since 1950, and in which many problems are still open;
D) Other developments.
t e
itA , where A is an arbitrary (in general unbounded)
e tA
space, but for such an operator, e usually has no mea-
p = P P .
(65) s+t s t
-1
(66) A•Jc = lim t (P t -1 E .
t40
vergence for that distance. This results from the fact that
210
LOCALLY CONVEX SPACES AND THE THEORY OF DISTRIBUTIONS 211
have had that idea was von Neumann: he defined weak neighbor-
for any closed ball B' in E', V fl B' be compact for that
topology.
212 CHAPTER VIII
Moore and H.L. Smith [164] (N. Bourbaki uses the equivalent
(*)
The bulk of that book [101] is devoted to metric spaces,
and general topological spaces are given a very scanty treat-
ment in 5 pages; it seems that Hausdorff had lost faith in
his ideas of 1914!
LOCALLY CONVEX SPACES AND THE THEORY OF DISTRIBUTIONS 213
the topology of the norm, and observes that for such a sub-
for all x'E V, but (x'0 ,x) 0. Conscious of the fact that
that there always exist elements u'E E' such that, for
every x E E,
and calls any such u' "a limit" of the transfinite sequence
(ui) there may be infinitely many such "limits"!). He then
II.
until 1950.
metric spaces thus obtained are complete. The fact that ad-
and showed that the closed graph theorem was also valid for
p (x-y) p2(x-y)
1 1 1 Pn(x-Y)
d(x,y) -
l+p (x-y) + 2! 1+p2(x-y) +...+. n! l+pn(x-y) +...
1
a single norm.
Hilbert space (§1). But already in 1910 E.H. Moore had put
two norms 114 1 and 114 2 such that the ratios 114 1 /11xH 2
and 114 2 /114 1 are bounded for x g 0, define the same topo-
bitrary metric space, two distances may give rise to the same
LOCALLY CONVEX SPACES AND THE THEORY OF DISTRIBUTIONS 217
incides with the previous one for normed spaces, and only de-
in R
N
(resp. T ) such that E lu x I converges (nowadays
n n
one says that E * is the Kgthe dual of E). One can then
the weak topology a(E,E * ) in the same way as von Neumann 01),
218 CHAPTER VIII
and E* play symmetrical parts, one can also define the weak
for that topology, and one of the chief results of lathe and
Toeplitz is that bounded sets in E are the same for the weak
what we now call locally convex spaces, but the general defi-
proved the remarkable fact that there are always "enough" ex-
Once the locally convex spaces had been defined, the Kgthe-
compact symmetric sets for the weak topology a(F,E) and form
than even Hilbert spaces (with which they have some surprising
cular satisfies
(1 )
(P'f,g) = (ft t P'g)
(2)
d ( 6F - = 0
form
(b
(4) C' ( x ) f x ) dx = 0,
(
a
b
fa
where f(x) = !;7(x,y(x),y'(x)) - .2(t' y(t) ,y' (0)dt is
6Y
LOCALLY CONVEX SPACES AND THE THEORY OF DISTRIBUTIONS 223
c
cally integrable functions in R, the function (x,y);-,.A(x)+
2
u
+ B(y) is a weak
weak solution of - 0, for one has
2
A(x)dx
a6xagy dy = 0 and f x B(y)dy d 0 for
Ox ay
R
2
any C function g with compact support.
(5) (h,g) = (f, t P•g).
(if r
(6) S dxdydz (Pdy A dz + Qdz A dx + Rdx A dy)
V
for any open set V with smooth boundary E. As an example,
J
is to be found in a paper of J. Leray in 1934 [141] (). In
familiar with the idea that two measurable functions which co-
f', and this could throw doubts on the adequacy of the "na-
n
for a C function g, or
forms such as ff---- f (n) (a) on the vector space 49(0) of all
.
C functions with compact support defined in an open set
N
Q c R ; Sobolev's idea was therefore to deal directly with
(9) sup ID(lf(x)1;
Pm K(f) =
la 1 m,*K
distributions are then the linear forms T on b(0), the
functions" [86].
of functions f E L1
such that 5f also belongs to
L
L1 n Lc° , then if two functions cp are such that
n V i
a linear "functional" m,-,.. w in the vector space of dif-
ferential p-forms on R . This leads to the idea of "gene-
ralized varieties" [231] and of "currents" [49].
LOCALLY CONVEX SPACES AND THE THEORY OF DISTRIBUTIONS 229
(-1- m
1 e-1-4 - Lk (x )
E( ,k) - f x dx
(-ix)k
I—co
/ n .
phism of the Fourier transform in goR ), in other words
w(x,y)dI(x,y) = f w(x,x)dx.
'XXX X
CHAPTER IX
of distributions.
233
234 CHAPTER IX
what one calls a fixed point Cor F. But after the firstyears
th
of the XX century, new possibilities of obtaining "fixed
point in B.
also, for separable Banach spaces, that one could replace com-
az 6z
(1) Az = f(x,y,z, ax' 6y)
2
in a domain Q c R with smooth boundary (no connected com-
tegro-differential equation
(2) z(x,Y) =
i( G(x,Y,,r)f(01,z(g.11), U, %)cg dr
hyperbolic equations (* )
2
az az) a z
E ..'xn axax k -
i v k °-2L1
(3)
az az
= A(x l ,...,x n ,z,, x ).
Q 1 n
az az ).
= A(xl,...,xn,z,x'...vb
1 xn
2
L(u) E E Aik(xl,...,xn)x + E
i,k iaxk j
(5)
+ C(x l ,...,x n )u F(xl,...,xn)
(
iP
( E I d/ f(x)I 2 )dx) 1/2 for another value of s; his
ICI l s
APPLICATIONS OF FUNCTIONAL ANALYSIS 237
showed that the theorem was still true in some types of Banach
(7) Az-f(x,y,z,N, t;) = r(x,y)
( )
This is probably the first time the norm of the space H s
makes its appearance.
238 CHAPTER IX
= 0 in 0.
satisfied.
then defined by
(8) (fig) E ( f n E n du •
nE J
2, 2,
F y : LH kX,p) 4 C such that for any function f E LH kX,p), the
,
T.: E. 4 F. = L 2 kY,p s •v)
J J
the function
Ca -- W01 2 E lej(x9012
lsj<w
is v-integrable.
2 d4(y)) 1/2
29 Let r(x) ( I1{,x,y)1
(, which is 4-measu-
r
X
rable and almost everywhere finite; then, for every function
perty: a
WC)1-2 E luj(012
lsj<w
(11) (P-1.(uj))(x) =I (
j 1.Sj<w
E ei(x,Ouj(0)dv(c)•
Y
(b*f)(x) = I b(x-y)f(y)dla(y)
JG
it follows that B is a Carleman operator corresponding to
2 ,
nerally not in L (X)) solution of the partial differential
equation
although Stone had shown in his book [207] how von Neumann's
„D r (pop r u) Dr-1(por-lo,
(15) L: u ) +...+ D(p r _ i Du) + p r u
tions u E H
L which vanish in a neighborhood of a (resp. $).
If one writes the Lagrange "adjunction" formula (chap.I, §1,
formula (5))
cff (C(u,v))
v(L.u) - u(Lov) = i
0(u) = lim C(u,w)(t) (resp. 0(u) = lim C(u,w)(t))
t4a t40
2 (,
operator in L J); one shows that it is a Carleman operator,
AL - CI ; one has
2r-1 2r-1
G(C,t,t-) 2r-1 G(C,t,t+) 1/po(t)•
(17) 2r-1
6s 6s
(18)
G(C,s,t) = 7‘,(, ․ ) * e(C) .ir(C,t) for t > s
4
where ,v (C,t) is the one column matrix (v (C,t))
15j52r
w(C) and /
W+ (S) are two square matrices of order 2r
whichonlydependonc.Furthernore,ifthev.have been
chosensothatforeachtEJ,thefurictionsCI--.ArJC ,
compactsubsetofS;inaddition,onehase(t,) = 0 for
(20) P = ( g ) * (g )
f E 0(J), write
(u.f)( c) =( f(t)v.(C,t)dt)
l5j52r
(one column
matrix);
J
then, for f, g in 0(J), one has
S
and
the equation
(26) P•T = S
ing also to L 2 (12) for ICI p, and this time this space
APPLICATIONS OF'FUNCTIONAL ANALYSIS 251
These properties were the source of what one may call the
it also belongs to /
H r+1 k0). The first idea of this method
the most refined ones [117, P.207]; it concerns what are cal-
pseudo-differential operators
We have seen (chap.II, §3) that from the beginning the idea
1
defined by the kernel - (where lx1 is the eucli-
things were not so simple, for one had to apply the integral
derivatives of f [122].
were explicitly known, he showed how one could prove the exis-
where A = E a D c
cc o
, linear combination of derivatives of the
that
(30 ) A*E = e ;
one has
(31) A*(E*T) = E*(A*T) = T.
(32 ) )•3E = 1;
Q•T = 1.
ing the values of the solution and its first m-1 normal de-
(33), one may say that the operator Q is such that PQ= 1+ R
and it is not surprising that after Levi not much work was
p. 207-343).
fkz)dz
(35) (H•f)(x) = lim
z-x
E 40 L
e
where x E L and L is the part of L for which the arc
e
of L joining x and z has length >c, exists for each
1
C function f defined on L, and is written
f(z)dz
vp I ; if L is a simple closed curve, the boundary
z-x ;
L
of a bounded open set 0, the usual line integral
1
f(z)dz is defined for x L, and is in 0 a holo-
217i z x
-
L -
morphic function F + / kx), and in the exterior C - 0 another
F 4- (t) =
1 f -C( Z )d.Z
f(t) 2rivP z-t
L
1 1 fiWgz
F(t) = -f f(t) + 2 fli vp
z—f— •
(u
ing the invariant measure on S n _ 1 ). Then, for any function
P.
one has
where x E L and E R
n are arbitrary, and is inde-
Ca$1,
pendent of x E L and
n
g E R ; the main difference is that
here m (the order of a, or of P) is an arbitrary real
writes a
o
The main properties of pseudo-differential operators are
2
for all u, v in gi(0) (scalar product of L (0)), which
0
(4o)
0
a
= gp.
P*
o = o o
(41) aPQ aQP PaQ
in the whole of Oxf 2. One says that the symbol a (and the
-
262 CHAPTER IX
is 0.
a — a + a +...+ a +...
o l k
of a. If P
are the pseudo-differential opera-
APPLICATIONS OF FUNCTIONAL ANALYSIS 263
P — P o + P +...+ P k +
1
Q 1P = I- P 1, where P
1
has order 5 -1, and one is reduced
that result!
(42) 11P.uil s s
for any uE 49(0), the norms being those of H s (0) and
tor such that c (x,g) > 0 for large It I ; then an easy in-
ductive argument determines (by an asymptotic expansion of
of an increasing sequence (X
n ) of real eigenvalues of fini-
te multiplicity, tending to +..; the corresponding eigen-
2
functions (suitably normalized) form a Hilbert basis of L (0)
where (-1)I v l a (x) Z 0 for all IVI < p, then one may even
V
take b = 0 in the Garding-Vis ik theorem this is the case in
a p ( x,g) is a homomorphism Ex 4 F
x of the vector spaces,
hold.
PP* are both elliptic and equal to their adjoints; the study
smooth?
borhood
o
of C), has even order 2p Z 2, and possesses
C2 0 - 5. Then P.0
0
is well defined as a distribution on 0
o,
and it is easy to check that one can write
(e(r)) 2p
,
u) which to every C function in associates a
parametrix Q.
ping
At this point, one might think, from the example (47), that
ever, this is not always the case, and there are examples for
definition
r(
(P.u)(x) = I exp(27i(x-ylt))a(x,g)u(y)dydt
JJ CD<R n
a(x,y,t).
2 ,2 2
u to uN
(53) ❑u = 2 - k 2 +...+ 2 ) •
tx1
ax n
at( 0,x) = g (x) are given functions, had already been solved
1
by Cauchy; the explicit formula he gave for the solution can
1 (gl
Y)
(54) u ( -t,x)= 2l fexp(27i((x-y It )+1 I t)))(g o (Y) ± 27i It ' )dYcg
variables t k n
a m u ,a
x ( am -j/
(55 P•u _ + E E c (t,x) L')
)
m
j=1 Ja 6tm
be written
m-1
(t o ,x 0 )
m-1 m-1
(59) v = E E.(t o )*g i L q( 1-1- 0*(f -
j=0 j=0
equation of order 1
APPLICATIONS OF FUNCTIONAL ANALYSIS 275
az 3z ) = 0
(60)
°-"- 1
in which /
is obtained from the principal symbol a (x,g)
az ,
by / az
by replacing the vector
); such hyper-
xi
surfaces are called characteristic for the operator P.
into m equations
az
(61) 7 1. --- q j (t,x,grad x z) = 0, 0 5 j 5 m-1
(with grad z = (
ax "'" 6z
ax n ))• For the wave operator (53)
x
1
the equations (61) are
with solutions
at
at
= ±Igrad zl
x
z = 27((xM t IgIt)
troduces the m
2
operators Fjh (s) defined by
(P jh (s)-u)(t,x) =
(62)
= 1
i UxtR. n
exp(i0j(t,s,x,)-27(ylg)))ajh(t,s,x,y,u(y)dyd
(63) = 27(xM
m-1
(R h (s)og)(t,x) = E (t k!
s) (Qhk (s).g)(x)
-
k=0
and E h (s) = F h (s) - R h (s).
expansions
(6 4) a. — E a (t)
jh jh
t=0
u)(t,x) =
(FjhN(s).
N
= E I exp(i(Ilij(t,s,x,g)-2rr(yV))a(iht)(t,s,x,y,u(y)dyd
t=0
UxIR n
then:
t ( t , ( t ) , , x n ( t ) , p o (t) pn ( t ) )
dxk aqi
dt as k (t ' x 1"'" x n' 13 1" . " 13 n )
(65) 1 5 k 5 n
dp aq.
k it
dt axk 'xl"'"xn'131"."13n)
yk ± (t-s) (1 5 k 5 n).
)
(N,
In the general theory, each a. kt s,x is taken inde-
jh
pendent of y, and is obtained by integrating, along each
tion of the first order (in the variable t), whose coeffi-
be distinct.
characteristics.
280
REFERENCES 281
284 REFERENCES
ax
+
12
ay
z = 0,
C95] H. HAHN, 'Ober die Integrale des Herrn Hellinger und die
orthogonalinvarianten der quadratischen Formen von
unendlichvielen Verdnderlichen, Monatsh. fir Math.
und Phys., 23 (1912), p.161-224.
REFERENCES 297
N. ABEL: 5,92,93,94,98
L. ALAOGLU: 212
A. ALBERT: 183
P. ALEXANDROFF: 212
P. APPELL: 76
ARCHIMEDES: 88
ARISTOTLE: 79
E. ARTIN: 183
C. ARZELA: 81
G. ASCOLI: 81
M. ATIYAH: 7,184,268
R. BAIRE: 138,141,142,210
S. BANACH: 6,128,130,134,135,136,137,141,142,143,155,156,160,
182,185,186,187,190,191,192,193,194,205,206,207,
208,211,212,213,214,215,216,218,219,220,233,234,
238
R. BEALS: 279
A. BEER: 5,41,42,66,69,91,97,98,102
D. BERNOULLI: 13,14,30,88,89
F. BESSEL: 21,108,109,110,163
G.D. BIRKHOFF: 234
S. BOCHNER: 23
B. BOLZANO: 45,82
G. BOOLE: 79
E. BOREL: 212
N. BOURBAKI: 212
C. BOURLET: 85,86,121
R. BRAUER: 183
L.E.J. BROUWER: 234,236,237,238
F. BROWDER: 242,246
V. BUNIAKOWSKY: 51
A. CALDERON: 259,270
G. CANTOR: 79,151
T. CARLEMAN: 160,168,169,170,171,181,238,239,240,242,243,245
E. CARTAN: 183,199,223
299
300 AUTHOR INDEX
A. CAUCHY: 11,14,23,24,26,27,28,31,3,51,73,119,156,169,195
196,226,235,252,253,256,258,268,270,273,274,275
B. CAVALIERI: 231
A. CAYLEY: 3,72,123,178,181,196
CHARPIT: 11
G. CHOQUET: 220
A. CLAIRAUT: 14
C. COULOMB: 31
R. COURANT: 58
G. CRAMER: 10,63,71,76,78,100
J. d'ALEMBERT: 10,11,12,88
P. DANIELL: 229
R. DEDEKIND: 84
A. de MOIVRE: 195
U. DINI: 81
P. DIRAC: 171,187,225,227,255
G. DIRICHLET: 14,37,38,39,40,41,42,46,48,50,53,57,66,67,68,
70,81,82,91,97,98,102,110,196,256,257,266,268,
271
P. du BOIS REYMOND: 38,97,222,223,250
L. EHRENPREIS: 255
EUDOXUS: 88
L. EULER: 11,12,13,14,82
C. FEFFERMAN: 279
P. FERMAT: 231
E. FISCHER: 119,120,123,125,128,172,201
J. FOURIER: 4,13,14,15,16,20,28 ,49,56,59,60,64,65,69,75,76,
77,78,83,88,89,92,107,108,109,110,112,120,139,140,
148,149,161,166,194,195,200,204,205,228,229,230,
231,242,251,259,260
M. FR1CHET: 5,97,116,117,121,124,130,145,210,213,215,216,220,
230,232,268
I. FREDHOLM: 5,41,53,79,97,98,99,100,101,102,104,105,106,107,
108,109,115,119,120,144,145,160,163,164,208,254,
255,257
K. FRIEDRICHS: 224,236,251,264
G. FROBENIUS: 3,73,112,113,167,196,197
E. FtRSTENAU: 76
AUTHOR INDEX 301
L. GIRDING: 242,246,265,266,267,268
C.F. GAUSS: 31,33,35,36,37,4 2 , 6 0,71,196
I. GELFAND: 160,184,185,186,187,188,191,205
I. GLAZMAN: 243
H. GOLDSTINE: 214
J. GRAM: 60,83
H. GRASSMANN: 72,73,85
G. GREEN: 32,33,34,35,36,37,41,48,51,52,58,59,61,65,66,87,
163,164,165,235,245,253,256,265,266,270,271
A. GROTHENDIECK: 220
A. HAAR: 194,202,204,206
J. HADAMARD: 38,83,86,87,100,120,121,128,191,249,253
E. HAHN: 6,128,130,134,135,136,137,138,141,159,188,213,219,
229
H. HAMBURGER: 129
W. HAMILTON: 72
G. HARDY: 194
C. HARNACK: 41
H. HASSE: 183
F. HAUSDORFF: 117,129,210,211,212,214,217,219
E. HEINE: 154
E. HELLINGER: 106,112,125,140,156,159,167,188,190,191
E. HELLY: 6,130,131,132,133,134,135,136,211,213
H. HELMHOLTZ: 47,61
G. HERGLOTZ: 255
E. HILB: 164,167
D. HILBERT: 5,6,39,53,54,60,82,91,96,97,98,103,105,106,107,
110,111,113,114,115,117,118,119,120,124,125,140,
144 , 14 5,1 4 7, 148 , 1 5 0 ,151,153,15 4 ,155,157,158,159,
160,163,164,166,167,168,169,170,171,172,173,176,
178,181,182,188,189,191,200,202,206,207,208,211,
213,216,220,228,231,238,239,240,256,259,265,268
G. HILL: 77,78,149
E. HILLE: 207
W. HODGE: 86
O. HOLDER: 31,124
L. HORMANDER: 256,279
B. HOLMBOE: 92
302 AUTHOR INDEX
E. HOLMGREN: 255
A. HURWITZ: 199
C. JACOBI: 154
C. JORDAN: 72,73,119
O. KELLOGG: 234
W. KELVIN: 37
A. KNESER: 38
K. KODAIRA: 243
G. KOTHE: 217,218,219,220
T. K6TTERITZCH: 76
A. KOLMOGOROFF: 217
S. KOWATEWSKA: 26,27,28,274
M. KREIN: 219,220
L. KRONECKER: 78
E. LANDAU: 125,126
J.L. LAGRANGE:10,11,20,30,33,83,85,126,170,221,222,245
P. LAPLACE: 26,28,29,30,31,34,39,65,66,85
P. LAX: 274
H. LEBESGUE: 5,45,46,68,97,119,120,139,140,212,224,227,272
G. LEIBNIZ: 84,231
A. LEGENDRE: 60,82,163,196
J. LERAY: 224,237,238
J. LE ROUX: 5,91,93,94,98
E. LE ROY: 70
E.E. LEVI: 68,255,256,257,258,266,271
H. LEWY: 27,236,264,279
A. LIAPOUNOFF: 70,195
S. LIE: 199,200,206,207,224
J. LIOUVILLE: 4,5,16,20,21,24,25,49,50,53,84,91,149,160,163
164,166,167
S. ZOJASIEWICZ: 256
Y. LOPATINSKI: 271
G. MACKEY: 219
B. MALGRANGE: 255
F. MAUTNER: 239,240,243
S. MAZUR: 215,218
D. MILMAN: 219,220
AUTHOR INDEX 303
H. MINKOWSKI: 124,131,132,219
M. MITTAG-LEFFLER: 98
T. MOLIEN: 183
G. MONGE: 11
E.H. MOORE: 212,216
F. MURRAY: 183,184
M. NAIMARK: 187,188,243
C. NEUMANN: 5,26,39,41,43,45,4 6 ,53,55,65,6 6 ,67,69,91,97,98,102
I. NEWTON: 231
L. NIRENBERG: 272,279
E. NOETHER: 183
L. NORHEIM: 171
W. ORLICZ: 215
W. OSGOOD: 141
J. RADON: 227
D. RAIKOV: 205
J. RICCATI: 18
B. RIEMANN: 31,37,38, 39,48,49,80,82,84,96,99,106,119,229,253
F. RIESZ: 6,119,120,1 21,123,124,125,126,127,128,129,130,131,
133,135,144 ,145,146,147,148,152,154,155,156,157,
160,169,171 ,175,176,182,185,188,190,191,201,211,
213,231,242
G. ROBERVAL: 231
304 AUTHOR INDEX
S. SAKS: 141
J. SCHAUDER: 234,235,236,237,238,250,251
E. SCHMIDT: 60,108,117,118,119,120,125,126,130,133,166,168,
170,201,202
I. SCHUR: 182,199,200
L. SCHWARTZ: 226,230,231
H.A. SCHWARZ: 26,39,41,47,49,50,51,52,53,54,55,56,61,62,65,
69,70,71,109,156,163,169,201,202
I. SINGER: 7,184,268
H.L. SMITH: 212
S. SOBOLEV: 7,226,227,230,248,250,269
H. STEINHAUS: 128,141,142
W. STEKLOFF: 70,108
T. STIELTJES: 122,129,150,151,154,177,229
G. STOKES: 236
M.H. STONE: 172,186,192,207,243
C. STURM: 4,16X,18,20,24,49,50,51,91,149,160,163,164,166
B. TAYLOR: 13,84
P. TCHEBYCHEF: 59,60,195
W. THOMPSON: 37,38
H. TIETZE: 68
E. TITCHMARSH: 248
O. TOEPLITZ: 112,125,140,154,157,191,217,218,219,220
F. TRgVES: 279
A. TYCHONOFF: 212
P. URYSOHN: 68,212
A. VANDERMONDE: 76
B.L. VAN DER WAERDEN: 3,142
E. VAN KAMPEN: 203,205
V
M. VISIK: 266,267,268,271
V. VOLTERRA: 2,5,25,53 85,86,91,93,95998,99,109,147,253,274
;
Eigendifferential: 159
Eigenfunction: 17,48,67,106
Eigenvalue: 17,48,67,90,91,106
Elementary solution: 253
Elliptic operator: 254,263,267
Elliptic partial differential equation: 29
Equation of vibrating membranes: 47
308 SUBJECT INDEX
Infinite determinant: 78
Infinite matrix: 112
Integral equation: 25,91,97
Integral equation of the first kind: 148
Integral equation of the second kind: 96,98
Integral operator: 103
Inversion of a definite integral: 93
Involution: 187
Involutive algebra: 182
Irreducible representation: 198
Iterated kernels: 109
Kernel: 95
Kernel distribution: 232
Kernel theorem: 231
Kgthe dual: 217
Laplace equation: 30
H. Lewy's example: 27
Linear representation: 196
Locally convex space: 218
Neumann function: 66
Neumann problem: 66
Norm: 117,131
Normal operator: 167,175
Normed algebra: 184
Normed sequence space: 130
310 SUBJECT INDEX
Operator: 84
Order of a pseudo-differential operator: 260
Orthogonal transformation: 112
Orthogonality, orthogonality relations: 16,20,48,91,117
Orthogonalization process: 60
Oscillating integral: 272
Parametrix: 256,263
Parseval's formula: 14
Plancherel theorem: 205
Point spectrum: 153,158,176
Poisson equation: 30
Poisson formula: 34
Potential: 30,32
Principal symbol of a pseudo-differential operator: 261
Principal value of an integral: 258
Principle of choice: 115,125
Pseudo-differential operator: 260,267
Pseudo-differential operator of proper type: 261