Lecture Notes in Mathematics A collectioo of informal reports and seminars Edited by A. Dold, Heidelberg and B. Eckmann,...
31 downloads
889 Views
4MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Lecture Notes in Mathematics A collectioo of informal reports and seminars Edited by A. Dold, Heidelberg and B. Eckmann, Z0rich Series: Forschungsinstitut for Mathematik, ETH, ZUrich • Adviser: K. Chandrasekharan
30 A. FrSlicher- W. Bucher Universit~ de Geneve
Calculus in Vector Spaces without Norm 1966
•
m
Springer-Verlag- Berlin-Heidelberg. New York
Work supported in part by the Swiss National Science Foundation
AU rights, especiallythat of translation into foreign languages, reserved. It is also forbidden to reproduce this book, Other whole or in part, by photomechanicsl means (photostat, microfilm and/or microcard)or by other procedure without written permission from Springer Verhtg. O by Springer-Verlag Berlin. Heidelberg 1966. Library of Congress Catalog Card Numbez 66-30545. Printed in Germany. Title No. 7350
-
I
-
CONTENTS
IV
Introduction
1
§l. ELEMENTARY PROPERTIES OF FILTERS 1.1
Filters and filter-basis
i
1.2
Comparison of filters on the same space
2
1.3
Mappings into direct products
2
1.4
Images of filters under mappings
3
1.5
An inequality between images of filters
4
§2. PSEUDO-TOPOLOGICAL
VECTOR SPACES
6
2.1
Pseudo-topological
2.2
Continuity
8
2.3
Induced structures
8
2.4
Pseudo-topological
2.5
Quasi-bounded and equable fi3tsrs
15
2.6
Equable pseudo-topological
19
2.7
The associated locally convex topological vecfior sp~ce
21
2.8
EquBble continuity
24
2.9
Continuity with respect to the associated structures
30
§3. DIFFERENTIABILITY
spaces
6
vector spaces
12
vector spaces
AND DERIVATIVES
3.1
Remainders
3.2
Differentiability
3.3
The chain rule
3.4
The local caracter of the differentiability
32 32
at a point
35 38
§4. EXAMPLES AND SPECIAL CASES
condition
38
42
4.1
The classical case
42
4.2
Linear and bilinear maps
43
~.3
The special case
44
f: A ~ E
-
4.4
§s.
§6.
II
-
Differentiable mappings into a direct product
FUNDAMENTAL THEOREM OF CALCULUS
50
5.1
Formulation and proof of the main theorem
50
5.2
Remarks and special cases
58
5.3
Consequences of the fundamental theorem
60
PSEUDO-TOPOLOGIES ON SOME FUNCTION SPACES
65
6.1 6.2
The spaces B(E1;E2) , Co(El;E2) and L(E1;E2) Continuity of evaluation maps
65
6.3
Continuity of composition maps
71
6.4
Some canonical isomorphisms
72
THE CLASS OF ADMISSIBLE VECTOR SPACES
82
7.i
The admissibility conditions
82
7.2
Admissibility of E ~
84
7.3
Admissibility of subspaces, direct products and projective limits
7.4
§8.
§g.
a6
Admissibility of
6g
85 B(EI;E2) , Co(EIIE2), Lp(EIIE2)
87
PARTIAL DERIVATIVES AND DIFFERENTIABILITY
9O
8.1
Partial derivatives
8.2
A sufficient condition for (total) differentiability
go gl
HIGHER DERIVATIVES
g3
g.l
f"
g3
g.2
f(P)
and the symmetry of f"(x) for
p R 1
g5
gg
§I0. Ck-MAPPINGS 10.1 The vector space 10.2 The structure of
Ck(EI;E2) Ck(EI;E2)
gg i01
10.3 C.(E1;E2)
104
10.4 Higher order chain rule
105
-III
§ii.
-
THE COMPOSITION OF Ck-mAPPINGS
ii0
ii.i
The continuity of the composition map
lid
ii.2
The d i f f e r e n t i s b i l i t y of the composition map
114
§12. DIFFERENTIABLE DEFORBIATION OF DIFFERENTIABLE mAPPINGS 12.1
The differentiability of the evmluation map
12.2
The linear homeomorphism
c.° (c1,d. (E2,E3)) - c'..(EI ,,E2~E3)
131 131 132
APPENDIX
137
NOTATIONS
14D
INDEX
143
REFERENCES
146
-
IV
-
INTRODUCTION
As emphasized by J. Dieudonn~, with the approximation
(in a neighborhood
calculus primarily deals of some point) of given
mappings of vector spaces by linear mappings. linear map has to be a "goad" approximation
The approximating
in some precise sense:
it has to be "tangent" to the given map. A very useful notion of "tangent" can easily be introduced
for maps between normed vector
spaces; it leads to the notion of "Fr@chet-differentiable" and gives, theory
mappings
in particular for Banach spaces, a very satisfactory
(cf. Chap. VIII of [3]).
It is well known that in this classical of differentiability
and derivative
the given norms by equivalent topologies.
theory the notions
remain unchanged
if one replaces
ones, i.e. by norms inducing the same
It is natural therefore to look for a theory which does
not use the norms, but only the topologies of the considered spaces.
vector
In fact, throwing out something which is irrelevant usually
leads to a clarification
and simplification
on one side, and allows
a more general theory on the other side. In the case of calculus, such a generalization
is indeed desirable in view of applications
to certain function spaces which have a natural topology,
but no
natural norm.
In classical theory, the norm is essentially used at two places:
(1) One defines what maps
r: EI---~E2
are tangent to zero
~t the origin (we simply ca)l them "remainders") fr~chet-oondition:
~
on the vector space
( 1 . |r (x)~
= O:
L(E1;E2) , consisting
maps from E 1 into E2, by taking,
for
by means of the
(2) One defines a norm of the continuous
~ ~ L(EI~E2):~h=
linear
~li~
(x)R.
In order to obtain a similar theory for a class of non-normed topological ~ector spaces, one has therefore to choose essentially two definitions:
(1) What are the remainders from E 1 to E2;
-V-
(2) What is the topology of L(E1;E2). The second definition comes in as soon as one wants to consider second (or higher) derivatives, since the first derivative f' of a (differentiable) map is a map
f: E1---~E2
f': E1--*L(EI;E2)o But all attempts which have been made
along this line gave theories with a very serious deficiency: the composite of twice differentiable mappings did not turn out to be twice differentiable in general; in other words: there was no higher order chain rule. In fact, a look at the classical proof shows that the second order chain rule is a consequence of the first order chain rule and of the differentiability of the composition map c: L(E1;E 2) × L(E2;E 3)
i L(E1;E3). But for non-norm~ble topological
vector spaces E i there seems to be no separmted topology on the spaces L(Ei;Ej) such that the composition becomes differentiable.(*) Nevertheless, a way out of this difficulty was found: independently A. Bastiani and H.H. Keller realized that though there is no satisfactory topology on the spaces L(Ei;Ej) , there exist pseudo-topologies which have the desired properties. The authors are very much indebted to H.H. Keller for having drawn their attention to the fact that pseudo-topologies seem really the proper thing to use at this place.
(*) This statement is not very precise, in particular since it depends on the adopted definition of "differentiable". If, ho~uever, one requires thst "differentiable" shall imply "continuous" end that the natural isomorphism between L(~;Ei) and E i shall be a homeomorphism, then one knows that with topologies one cannot succeed; in fact, the continuity of the composition map
c: L(~;E1) × L(E1;E2)
..... ,L(~;E . 2)
then is
equivalent with the continuity of the evaluation map e: L(E1;E2) × E1----oE2, and for non-normable spaces El, E 2 there is no topology on L(E1;E2) for which this evaluation map e is continuous
ET]).
-
VI
-
The above remarks show that it is not for the sake of greatest possible generality that we develop our theory right from the beginning for pseudo-topological
vector spaces (topo-
logical ones are special cases of these), but simply in order to obtain a satisfactory theory for a class of vector spaces containing at least some non-normable topological ones. In order to
prove certain theorems of calculus,
some restrictions however
will have to be made: a class of pseudo-topological
vector spaces,
called "admissible" ones, will be introduced. This class contains in particular ai1 separated locally convex topological vector spaces.
Since our whole theory works consistently with filters, §l starts with some well known facts concerning filters. For a reader who is familiar with filters, it will be sufficient to have a look at (1.5..2)~ we found that at some places in the literature the inequality stated there was erroneously used as an equality.
§2 presents the basic facts concerning pseudo-
topologies and in particular pseudo-tomological vector spaces. The material of sections 2.5 to 2.9 will not be used for the beginning of calculus and thus can be read later, whenever referred to.
§5 deals with what might be called the "mean value theorem". However, there is no mean value in it; but it is fundamental in the sense that it is used in order to orove ormctically all of the deeper results of calculus. We thus call it "fundamental theorem of calculus".
Intuitively,
it
gives an estimate of the difference between the endpo±nts of a motion of a point in a vector space by means of the velocity of that motion, the estimation being made by means of convex sets. In the case of normed spaces, the theorem yields the well known estimate by means of the norm (cf. (8.5.1) of [3]) provided one chooses as convex set the closed unit ball; but being able to
-
VII
-
take other convex sets, we get better information also in this classical case: We not only can conclude that the point does not get too far if the velocity is not too big, but also that the point does get far, if the velocity is big (in the sense of lying in a multiple of the convex set in question). some consequences in particular, inequalities
For later applications,
of the theorem are established
at the end of §5;
two versions of the theorem in the form of filter
will turn out to be useful. Another consequence
Taylor's formula;
it will be given in a forthcoming
is
publication.
In §7 one finds the definition of the admissible
spaces
and furthermore a result without which the theory would not be satisfactory:
the class of admissible vector spaces is closed under
the constructions
used in calculus,
yielding new spaces out of
given ones, such as L(E1;E2) or Ck(E1;E 2) out of E 1 and E 2. In §8 we show that the relations between partial and total differentiability
of a mapping of a direct product are as
in classical theory; in particular,
partial differentiability
continuity of the partial derivatives
plus
implies total differentiability.
We remark that this theorem uses in a very essential way the choice of the structure of the spaces L(Ei;Ej), pseudo-topologies
since "continuous"
refers to the
of the spaces in question.
The main results of §9 state that the p-th derivative at a point can be identified with a multilinear map which is symmetric, and that the composite of p-times differentiable differentiable
maps is also p-times
(p-th order chain rule).
The notion of a Ck-mapping
introduced in §lO coincides with
the usual notion of a k-times continuously the case of finite-dimensional more restrictive.
differentiable
mapping in
spaces, while in general it is slightly
The vector space consisting
E 1 into E 2 is denoted by Ck(E1;E2)
of the C~mappings
or C~(E1;E2),
from
depending on which of
-
VIII
-
two pseudo-topological structures we consider (we always use one symbol to denote the space and its structure). The important spaces are the spaces .C~(E1;E2) ; but for technical reasons it is useful to define them by means of the spaces Ck(E1;E2) as auxiliary spaces and a general operator "
"
which associates to any pseudo-topology
of a vector space a second one, having in addition a certain important property, called equability. In special cases, the operator " ~ " becomes the identity; in particular, if the spaces E i are finite dimensional, the pseudo-topology of Ck(E1;£ 2) = C~(E1;E2) is nothing else than the topology of uniform convergence on bounded sets of the functions and thair derivatives up to the k-th order. The cass k = ~ is obtained by forming a projective limit. In §ll the differenti~bility and the C -nature of the P composition map of Ck-mappings are investigated, the main results being theorems (11.2.21) and (11.2°26); here, the result stating that the composition map is of class C
is in fact stronger than P just saying that it is p-times continuously differentiable° §12 deals with differentiable families of differentiable
maps, "differentiable" now always meaning "differentiabls of class CN~
Having our theory of differentiation and also a pseudo-topology
on the vector space of differentiable maps from E 1 into E2, one can consider two sorts of differentiable families of such maps: a) A differentiable family of maps (depending, for instance, on a real parameter)is a differentiable map of ~ x E 1 into E2; b) A differentiable family of maps is a differentiable map of ~ into the function space C~ (E1;E2)o The main result of §12 not only says that these two notions are completely equivalent, but even that the structures put on the space of all differentiable families according to either one of the two points of view a) or b) are the same; in other words, there is a canonical linear homeomorphism between C~ (~ x El;E2) and C~ (~;C~ (El;E2)). Moreover, the "parameter space" ~ can be replaced by any admissible equable vector space E. If we consider this iso-
-
IX
-
morphism in the special case E = E l = E 2 = ~, for instance, the space on the left hand side is classically
then
well known, while
on the right hand side we have a new function space, consisting of functions with values in the infinite dimensional
space C.o(~;~).
Repeating this argument one sees that at least for many spaces Ei, Ej the set and the structure of C ~ ( E i ; E j )
are uniquely determined
one requires the following two conditions: are finite dimensional, functions,
if
(1) in case E i and Ej
C ~ ( E i ; E j) is the set of classical Cm-
with the topology of uniform convergence
of the functions and their derivatives;
on compact sets
(2) the linear homeomorphism
(12.2.5) mentioned before shall hold.
Depending on the choice of the two main definitions obtains different theories.
Our approach is different
of A. Basfiiani, H.H. Keller and E. Binz ([1],[6],[2]).
one
from those In order to
develop our theory, we always postulated that the definitions agree with the classical ones in the case of normed spaces, a condition which is not satisfied by the theories of A. Bastiani or of E. Binz. The structure of L(EI;E 2) defined by H.H. Keller for the case of locally convex spaces El, E 2 by means of families of semi-norms (cf. [5]), seems to be the same as the structure of our L@(E1;E2). In [6], H.H. Keller introduces various notions of differentiability and compares them with definitions other authors
that have been suggested by still
(cf. the references given there),
restricting
himself
in that paper to locally convex spaces. At the time being it is difficult to recognize which one of the various theories will eventually
turn out to be the most
useful one. That mainly depends on what theorems one gets and on what applications
one wants to make. An implicit function theorem
has not been obtained so far; in fact it is known that its classical formulation
simply fails to hold. H.H. Keller has also established
and motivated a series of basic properties that should hold in a useful theory of calculus
(cf. [7]); we believe that our theory
-X-
satisfies these conditions.
Throughout this report, we restrict ourselves to certain vector spaces; manifolds modelled on such vector spaces shall be considered later.
Though our notion of differentlability
is a local property,
a non-local condition is imposed on the so-called Ck-mappings ; but this condition becomes trivial in the case of finite dimensional spaces, and, at least, it is not so restrictive as to rule out the identity map, as it would be the case if one had to restrict oneself to maps with compact or bounded support.
The
f~r@~-named author has presented a first version of
calculus for topological vector spaces in a Seminar of Professors A. Dold and B. Eckmann at the Swiss Federal Institute of Technology (ETH), Zurich, in summer lg63; it was not yet satisfactory,
since
there was no higher order chain rule. A part of the present theory was outlined by the same author in a series of lectures at the Forschungsinstitut
for mathematik of the ETH during the 1964/65
winter term.
The present work has been partially supported by the Swiss National Science Foundation.
-1-
§ 1.
ELEMENTARY PROPERTIES OF FILTERS.
Since the whole theory is based on the convergence of filters, we recall here the fundamental facts concerning filters and state an inequality (in 1.5) which will oe used very frequently in the sequel.
i.I.
Filters and filter-basis.
A filter on a space (i.e. set) m is a non-empty set ~ o f
subsets
of m ~uch that
(1) (2) X2
(3)
X1
XI,X 2 ~ ~
~ Xl ~ X 2 ~ ~
.
A filter-basis on M is a non-empty s e t 1 ~ o f subsets of m such that
(1) (2) Sl,S 2
"B
~There exists 8 3 E I ~ with B 3 c a I n B 2 .
Each filter is a filter-basis. Conversely, each filterbasis ~determines a filter
~ = E ~ ] a s follows : ~ consists of
all subsets of M which contain a set of ~ . ~ i s
called the filter
generated by the filter-basis I~. In particular, if B is any nonempty subset of M, I~ = ~B~ is obviously a filter-basis. The filter
(*) ~always denotes the empty set
-
2
-
generated by it consists of all subsets of M containing the fixed set B and is denoted simply by ~B] .
Analogously, if a ~ m,
(a3
denotes the filter formed by the subsets of M containing the point ao
1.2.
Comparison of filters on the same space. The set of all filters on a given space ~I is partially
ordered by the set-theoretic inclusion~
(1.2.1)
~(I ~ X2 ~'~ ~l o 12
(*).
We thus have the notions of infimum and supremum of a family of filters on m : inf ~i and sup ~i" i~ I i~ I
The second always
exists; it is the filter consisting of all sets of the form
(1.2.2)
L~ x i , where X i ~ ~i" i~l The first does not always exist; it will not be used. According to usual notation one also writes : 1.3.
mappings into direct PrOducts. If fi : m i - ~ N i
we denote by
i~I ×
sup ( ~l' ~2 ) = E1 V ~2'
' i~ I
resp. i = 1,2, are mappings,
>~ m i the direct product of the sets Mi~ and by i~I
fi ~ i~I x mi__iXiN i resp. flxf2 : mlXM 2 --~ NlXN 2 the maps
defined as follows:
(*) In [4] Fischer uses the symbol "~ " in the other sense; definition (1.2.1) is the one used e.go by Kowalsky in[ ~] .
-3,,
(1.3.1)
resp. iEl
iEl
i~l
(flxf2)(Xl,X2)
In the special case where M. = ~ for all i~I, X m .
=
(fl(Xl),f2(x2))o
is usually
denoted by m I, and we further denote by
T[f.
:im
-'--'~ ~N.
i61 i
i~l
reap.
Ill,f2] : m ~
NIXN 2
i
the maps defined as follows:
(i .3.2)
(~fi)(x) iGl
= ~fi(x)}
respo
[fl,f2] (x) = (fl(x),f2(x)).
i¢I
These maps are related by means of the diagonal maps d : m
jmI
respo
d : m
~ mxm = m 2
as follows :
(1.3.3)
f. : ( X f i ) , d iEIi iGI
resp. [fl,f2] : (flxf2).d.
1.4. Imaqes of filters under mappings. Let f: m ~f(X) I X E I ~ i s
~N be a mapping and ~ a filter on ~. The set
than a filter-basis on N, which generates a filter
denoted by f ( ~ )
and called the image o f ~ u n d e r
the mapping f.
The use of the same symboi f is justified because the functor in question is covariant :
(1.4.1)
If M
f~ N
g :P, then (g.f)(~) = g(f(~));
and also because
(1.4.2)
f([x3 ) = ff(x)].
-
4
-
The induced mapping for filters is order-preserving
(i.4.3)
Xl ~ ~ Let now ~
(1.4.4)
'
'
:
f(~l)~ f(~)"
be a filter on mi, i = 1,2. Then we define:
~l x ~
is the filter on m I x ~2 generated by
fx xx l If further g : m l X M 2 - - * N
(i.4.5)
g(~,
is a map, then:
~2) denotes the filter generated by the
?o~oo~n~ ~er
~,~X~,X~) I X~.~X~,
~)
It folloms easily that
(i.4.6)
o(I i, ~2) = o(~ I x X2).
1.5. An i neguality between imaqes of filters. Let X be a filter on m and d : m ~ M x ~ 1
be the diagonal
map : d(x) = (x,x). Then
(i.s.i) Proof. Let A~](x~. we have: A ~ X x X ,
Then A ~ X l X X 2, where Xl,X2~l(.
where X m X .
But since XxX~d(X),
With X = Xl4X 2
it follows that
A m d(X) which shows that A cd(~(). Usually, this inequality will be used in combination with mappings.
(i.5.2)
A typical example is as follows.
Let h i : m --,Ni, i = 1,2 and g : NlXN 2 - - ~ p If f : M
be maps.
; P is the map defined by f(x) = g(hl(X),h2(x)),
then f(K) ~ g(hl(@) , h2(~())"
-
Proof.
S i n c e by ( 1 . 3 . 3 )
5
-
we have f - ge [ h l , h 2 ~
= g,(hlXh2).d
the inequality foliows from (1.5.i), using (1.4.3), the equality (hlxh2) ( @ x ~ )
= h1(~)
,
(1.4.6) end
x h2(~).
We shall refer to (1.5.2) whenever we have occasion to use an inequality of this type, even though the situation may be somewhat different.
For example it could be that there are several
variablesp or the right hand side might contain x repeated more than once.
-
§ 2. .
.
.
.
.
PSEUDO-TOPOLOGICAL .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
-
VECTOR SPACES. .
.
.
.
.
.
.
.
.
.
.
For a detailed introduction of the notion of pseudotopological spaces, the reader is referred to [ 4 ]
. We shall
introduce a slightly different notation which will be more convenient for our purpose. introduced,
Some associated structures will be
and questions concerning continuity will be discussed.
It turned out that a condition which is slightly stronger than ordinary continuity will play an important role: the notion of equable continuity.
In particular we investigate the case
of linear and multilinear maps.
2.1.
Pseudo-topological
spaces.
A pseudo-topology
(or limit-structure)
on a space
(i.e. set) M consists in assigning to each x¢ ~ a set of filters on m, such filters being described as "converging to x". The following axioms are supposed satisfied: to x, then so does any smaller filter; to x, then so does their supremum;
(1) If a filter converges
(2) If two filters converge
(3) The filter Ix] converges
to x. A pseudo-topological
space
E
consists of a set
together with a pseudo-topology on m. The set m is called the
-7
-
underlying space and will be denoted by ~.
If a f i l t e r ~ o n
converges to x in the sense of the given pseudo-topology, write
~x
E (~converges
we
to x on E). The axioms can now be
expressed in the following way.
(2.1.1)
For all
(21 ~
(3)
x E E
~x E
[,3
for
one has :
i = 1,2
---@ t l v
t 2 ~x E ;
E.
Topological
spaces can be c o n s i d e r e d as s p e c i a l
topological
spaces.
define
In f a c t ,
if
cases o f pseudo-
E is a topological
space, we
:
X~ x E where ~ x
~ x
is the filter
verging filters s i n c e then
~
'
o f neighborhoods o f x . Knowing t h e con-
on E, c o n v e r s e l y we can r e c o v e r the t o p o l o g y ,
~ . . = sup
~
. Thus, a necessary c o n d i t i o n
for a
X @xE pseudo-topological
(
sup ~ ) ~x E.
space t o be a t o p o l o g i c a l
one i s t h e F o l l o w i n g :
This condition is not in general sufficient:
~x E however we shall see in (2.4) that it is sufficient in the case in which we shall be interested,
i.e. if E is a pseudo-topological
vector space. For details, see [4] .
-8-
Continuity.
2.2.
Let El, E 2 be pseudo-topological spaces and f: ~i--~p~2 a map.
We say that f: EI---pE 2 is continuous at the point
a( E 1 iff (*)
El
f(~) ~f(a) E2 "
f: EI---$E 2 continuous means: continuous at each point. In the topoIogical case the definition is equivalent to the usual one. One further verifies easiIy that the composite of continuous maps is also continuous. f: E I - - ~ E 2 is a homeomorphism means that f: ~ i - - - ~ 2
is bijective
and that f: EI---~E 2 as well as f-l: E2._.~EI are continuous.
2.3.
Induced structures. Une can introduce a partial ordering on the set of
pseudo-topologies of a fixed space.
Let El, E 2 be two pseudo-
topoiogical spaces with --El = ~2"
(2.3.1)
Definition: The structure of E1 is called finer than that of E2, and we write E 1 ~ E2, iff
~x El ~
~ ~x E2"
(*) From time to time we shall make usa of the expression "iff"; as usual it stands for "if and only if".
.
g
,.,,
We also say the structure of E 2 is coarser than that of El . This is equivalent to the condition: the identity map i : E1----PE 2 is continuous. As in the topological case, one can define structures which are induced by mappings. We will use only the following case: Let Ei, i~ I, be pseudo-topological spaces, M a space (without structure), and fi : m ...... ; ~i
, i~ I .
Then there exists on ~ a unique pseudo-topology which is caracterized as being the coarsest pseudo-topology on m such that all maps fi : ~-'--~Ei are continuous.
Denoting m together
with that structure by E, we have (@ being a filter on m) : (2.3.2)
X& x E ~
fi(X) &fi(x) E i
for all i~ I.
In Fact, one easily verifies that with this definition the axioms (2.1.1) are satisfied; and (2.3.2) certainly defines the coarsest structure on m such that all maps fi : m " - * E i get continuous. If all the spaces E i have the property that for each point xE E i one has
sup ~ ~x Ei ' then also E has this property. i~xE i
This is easily verified, using that for any map f : M ~ any family
IXj~ j 6J
of filters on M, one has
N and
-
(2.3.3)
f ( sup
j~J
~j)=
sup
j~3
lO
-
f(~j)
;
which i s an immediate consequence of the s e t - t h e o r e t i c
f (jV xj)
f(xj)
equality
.
We consider now three s p e c i a l cases of s t r u c t u r e s induced by mappings : subspaces, d i r e c t products and p r o j e c t i v e limits. a)
Each subset A of a pseudo-topological
pseudo-topology, i : A ~
E o
subspace of E.
namely the one induced by the inclusion map A together with this structure is called a Denoting the subspace thus obtained by E 1 ,
we have according to (2.3°2)
(2.3.4)
space E has a natural
~ x El ~ b) Given a family
:
i (~)$x E.
~Ei~ i~l of pseudo-topological
call direct product and denote by space whose underlying
spaces, we
X E i the peeudo-topological iGI
set is the direct product of the sets _Ei,
together with the structure induced by the projections "~k : X _Ei ~ i6I
Ek ,
ke I.
Thus (2.3.2) yields for this case:
(~.3.5)
&x (× Ei) ~
ITk (~) ~ ~k(X) Ek for all k~.
i~ I In case of finite direct products we write E 1 x...x E n instead of
X E. . i e (l,...,n~ z
-
c)
l l
-
Let a projective system of pseudo-topological
i.e. to each element i of a directed set II,~} gical space Ei is associated, subset of ~i 2 ,
--
E
i6I
a pseudo-topolo-
~uch that for i I ~ i2, ~i I is a
with a continuous inclusion map.
is the pseudo-topological section E = F ]
spaces be given,
Then L = proj.limi( I Ez
space whose underlying set ~ is the inter-
and whose pseudo-topology
is the one induced by
--z
the inclusion maps Ji : ~ --'-* El" In the case of subspaces and direct products, the following two~lemmas will be used later.
(2.3.6)
Lemma.
Let f : EI----~E 2 be a map and suppose that E 2
is a subspace of E3, i : E 2 ---~E 3 being the inclusion map.
Then f : E1----~ E 2 is continuous if and only if
ief : EI----~ E 3 is continuous. Proof.
Necessity is obvious, since i : E 2 ----, E 3 is continuous.
Suppose conversely that i-f is continuous, (iof)(~) $(iof)(a)E3,
or equivalently
By (2.3.4) this yields f(~) ~f(a)
and let
[~a
El" Then
(by (1.4.1)): i(f(~))$f(a)E 3.
E2' which proves the continuity
of f.
(2.3.7)
Lemma. Let X there exist
ZlX) 2.
be a filter on ElXE 2. Then X~xEIXE 2 ~l' ~ 2
with ~i &%i(x)Ei ' i = 1,2, and
-
Proof.
12
-
i ) Since for any subset X© ElXE2 one has Xcg(X)x l~'2(X),
one has for any filter ~ on ElXE2:
(2.3.8)
~Tl(X ) x ~2(~) • Hence, if ~ x E l X E 2 , we can choose ~i = TFi(~)" 2) If, conversely, ~ ~ ~l x ~2' where ~i ~ ~Ti(x) Ei' then ~i(~) ~ ~i ( ~ l x @ 2 ) = Xi, and according to (2.3.5) this p~oves that ~ x E l X E 2 .
2.4. Pseudo-topological vector spaces. Let the underlying space ~ of a pseudo-topological space E be a vector space (*). The pseudo-topology is called compatible with the vector space structure (shortly: compatible) if the maps ExE ~xE are continuous.
+ "
~E, ~E
By IR we always denote the reals, taken with the
usual topology. A pseudo-topoloqical vector space is a vector space together with a compatible pseudo-topology on it. more precisely: it is a triple, consisting of a set together with two structures on it, namely an algebraic structure (of vector space) and a pseudo-topological structure.
(*) Vector space here always means: vector space over the reals ~.
,- 1 3
-
Continuity of addition implies that the translations are homeomorphisms. Therefore: (2.4.1)
~a
E ~-~
~-a &o E
~+ a ~a E ~ = ~ o
By map
or equivalently
E-
~-a we denote of course the image of ~ under the translation x ~--*x-a.
g(x,y) = x-y.
More generally, we write g ( ~ , ~ )
=~-~
if
Then one easily verifies that ~ - a can also be
considered as the image of two filters under the difference map: -a = ~ -
[4
° Similarly for ~ + a : ~ + a =~,[a].
In view of (2.4.1) the pseudo-topology of a pseudotopological vector space E is completely known if we know what filters converge to zero.
Hence we only need one relation, that
of "converging to zero", and we shall simplify the notation by writing
~ b E instead of ~ o
E. Thus " 6°.." simply means :
"converges to zero in...". The continuity of the operations implies the following compatibility conditions: (2.4.2)
(1)
"~I~E , "~2 ~ E
.---m ~'l ~ ' ~ E ~
(2)
~ ~E , ,-)$.IR
~
~1"3['J, E;
(3)
~ ~E
~
|I/."j[ ~I,E;
(4)
x mE
~
IV, x ~- E.
-
14
-
By W we always denote the filter of neighborhoods of OG IR. For the meaning of
~l + ~2' ~ '~' |V,~ and | V . x ,
see the remark
following (2.4.1) above. Conversely, if for a given vector space we say what filters converge to zero, and if the set of these filters not only satisfies the conditions (2.1.1) for x = O, but also the above compatibility conditions (2.4.2), then we obtain, taking (2.4.1) as definition, a unique compatible pseudo-topology on E. Looking at the induced structures studied in section 2.3 for the case of pseudo-topological vector spaces, one easily verifies that for linear maps fi the induced pseudo-topology is also compatible. We therefore have En particular:
Vector
eubspaces, direct products and projective limits of pseudotopological vector spaces are also pseudo-topological vector
spaces. Suppose now that for a pseudo-topological vector space E also the following condition holds:
(2.4.3)
s u p X ~E. By (2.4.1) we have then more generally: ~x
= sup ~E ×
and hence
~
x
E,
-
15
-
Let U ( 1.Lx. Then U = x + V, where V • t t o. Since by (2.4.2) we have % c
rb~° + ~ o '
there exists V'( ~ o with V ~ V '
Hence, with U' = x ÷ V' • ~x' we have U ~ U '
+ V'.
+ V' ~ y + V I •
Y
for all y( U'. We thus have shown:
for each U e ~ x there exists
U'( % x
Or it is well known that this
with U 6 % y
for all y EU'.
implies that the filters respect to a topology.
(2.4.4)
Proposition.
1&x are the neighborhood
This yields the following Condition
2.5.
and equable filters.
On pseudo-topological the property
IV,~dE
quasi-bounded
filters on E.
(2.5.1)
vector spaces,
will be frequently used.
filters • with We call them
The name is motivated by the
result. Lemma.
On a normed vector space E, a filter is
quasi-bounded Proof.
vector space is
vector space.
qMasi-b£unded
following
(cf[4 ] ):
(2.4.3) is necessary and
sufficient in order that a pseudo-topological a topological
filters with
Let | V . ~ E .
if and only if it contains a bounded set.
Hence
hood filter of zero in E. i.e. V : Ix EEI
|xl~
iV.~
~'Tv~', where I~ is the neighbor-
If V denotes the unit ball in E,
1] , then V ¢ ~ ,
hence a f o r t i o r i
V ( I~. ~
o
-
16
-
Each set of~V contains a closed interval where
~0,
and so there exists
In particular,
if x E B ,
~0
then ~
If U s ~ =
E/~
, choose
~ ~0
one has IG
and therefore
~V. ~
Conversely, ~ £1R with
~I 4 ~
hence U ¢ ~V ~
for
suppose I@ contains a
such that x ~ U for
.BcU, ~
and B • I~ such that V ~ I s ' B.
'xjI&l, which yields ~x~I ~ i/~
x ~ B, showing that B is bounded. bounded set B; so there exists
I~ = { ~ s IR i J~I ~ ~
for all x • B.
Uxll ~
~
° Taking
, showing that ~
~IV"
E.
A filter ~ on a vector space is called an equable filter iff it has the property any topological
~V. ~
= ~
. It is well known that on
vector space the filter of neighborhoods
of
zero is an equable filter.
(2.5.2)
Lemma.
A filter ~
on a vector space is equable
if and only if it satisfies the following conditions: (1) Each X ~ ]~
(2) X¢.'~
contains an X' ~ ]~ with 11.X'~I(*)
, ~.1/.0 ~
~; .X e X .
Proof. i) Suppose ~V.~ = ~ . Let X s ~ o Hence X E ~ . ]~, and thus there exist ~ ~ 0 and X 1 s ~ we have X' ~ X shown, X ~ X '
and hence
with X ~ I ~
and I1. X' = X'.
= IiX~, we get:
~X £ " ~ .
. X 1. Taking X' = I s' X l,
Let further
$ ~0.
Since, as
--
2)
Suppose, conversely,
If X e ~
l?
-
that conditions
(1) and (2) hold.
, it follows that X ~ X' = fIX', where X t~ ~ , and
since IIX' £ ~2.~( we have X 6 W" ~( • And if A G ~ V . ~ ~ then there exist
~ ~0 and X 2 ¢ X
and since ~X 2 6 ~(
with A ~ I $
by condition
X 2, hence A ~ ~X 2,
(2) we have A c ~ , which completes
the proof.
(~.5.3)
L emma. Proof.
X I and ~(2 equable ==~ @l + ~(2 equable.
For arbitrary subsets Xl, X 2 of F one has
I~ • (XI + X 2) c Is X l + 1 6x2; one easily verifies that under the hypothesis liX I = XI, llX 2 = X 2 one gets equality: l& (XI +X 2) = I6 X I + Is X 2. Let now A ~ W A > I ~ . ( X I + X2) , where
~>0
and X i •
.(~( I + ~ 2 ). Then
](i" Since
)~i is
equable, we can according to (2.5.2) choose X i such that X i = I1. Xi, so that by the above equality we get: A ~ I ~ which shows that A ~ ~/.l(1 + •. @ 2 "
XI + 16 X2,
We therefore have:
\v. ( ~l + ~2 ) ~- \v. ~l + W. ~2" Combining this with the converse inequality
W'( ~(1 + X 2 ) ~ ' @ l
which is true for arbitrary filters, since it follows from (xI + x 2) = ~ x I + ~ x 2 by (1.5.2), we get
W , (~i +~2 ) = ~V.~l + W . ~ l "
~l + ~2"
+ W.~,
-18.
(2.5.4)
Lemm_,___~a.Let ~i resp. Then Proof.
~l and
~2 be filters on E1 resp. E2.
~2 equable =====~ ~l x ~2 equable.
As before, one has for arbitrary subsets X1 rasp. X 2
of E1 resp. E2 :
I~, (Xz x X2)c I~ Xl x I% X2, while for subsets satisfying IiX i = X i one has equality. Therefore one gets for arbitrary filters
(2.5.5)
W. ( ~ I x ~E2)& ~v.XI x W . ~ 2 , while for equable filters one gets as before also the converse inequality, and hence
W (2.5.6)
(~i x ~2 ) = W ~l x~v
2 = ~i x ~2"
L=emma. Let '}El' ~2 be filters on E. Then ~l and
){2 equable ==-dp ~l
~/ ~2 equable.
Proof. From the set-theoretic equality 18 ,(X l v
X2) = I~
X1 u I & . X 2
one deduces, using (1.2.2), that for arbitrary filters "~l' onE:
(2.5.7)
\v.(X I ~ ~2) =~v. ~lV~V. ~2" From this, the lemma follows at once.
-
2.6.
lg
-
Equable pseudo-topological vector spaces. A pseudo-topological vector space E is called equable
iff for each ~ with ~ i
with
~
E there exists an equable filter ~ ~
E; i.e. iff
(2.6.1)
---,
It follows from the remark preceding lemma (2.5.2), that each topological vector space is equable. However, not all pseudotopological vector spaces are equable (*). Given any pseudo-topological vector space E, we can introduce on ~
a new pseudo-topology, thus obtaining a new
pseudo-topological vector space E I~" . It is defined as follows:
(2.6.2)
(1) (2)
E~
= ~ ;
~ ~ El~'iff there exists ~
with ~ =
~" ~
E.
One has to verify that the set of filters caracterized by the above condition (2) satisfies the conditions (2.1.1) (for x = O) and the compatibility conditions (2.4.2). Of the conditions (2.1.1), we only verify the second one, the others being obvious.
So let
~i~ E~
for i = 1,2.
Hence there exist
filters ~ i with
~ ~- ~i =w.
~ E
, i = 1,2.
(*) Examples will be given in a forthcoming publication.
-
20
-
Using lemma (2.5.6) we get therefore v
Here, ~l V
+
']
~2 ~ E since E satisfies the conditions (2.l.1), and
we see therefore that
~l ~ @ 2 ~ E+k"
We next verify the compatibility conditions. (i)
Let
~i~ E+ for i = 1~2. As before we getp this time using
lemma (2.5.3) :
~z + ~ which shows that
(2)
Lst
~l + X2~EI~' since ~l + ~2 ~
E.
II~E $ , and ~=IR. Then
and thus
which shows that ~ + ~ E ~, since
~ % ~ 4 E.
(3) and (4) follow immediately, since the equality (W'V)
N
: ~'~
~ (~.~)
implies that each filter of the form
~.~
: is
equable. And according to the definition (2.6.2), each filter which is equable and converges to zero in E also converges to zero in E +
. This completes the verifications.
We next remark, that E ~
is clearly an equable pseudo-
topological vector space and that (cf.(2.3.1)) always
(2.6.3)
E~
zt- E,
with equality if and only if E is equable.
-
(2.6.4)
Lemma. Proof.
E l ~ X E2~
I) Let X ~ E l m
21
-
: (E1 x E2 )~ .
x E2@
. Hence, for i : 1,2:
IT.(~z )~ Ei~ , which implies "~.(~)z_z -- ~ i =W ~ i
~ El'
Using (2.3.7) and (2.3.6) we get :
This shows that
~ ~ (E1 x E2 )~
, since by (2.5.4) ~1 x ~2 is
equable.
2) Let ~ ~ (E1 x E2)e. Hence ~ _z.~ = \V ~ ~E 1 x E2. From this we get ]Ti(~ ) ~ -Ei(~) = ]Ti(~V~) = ~V. ~ i ( ~ ) which implies I~(z X) ~ Eir
2.7.
~ Ei ,
(i= 1,2). Therefore ~ ~ El~, xE2~
•
The associate Q locally convex topological vector space. For any filter ~ on a vector space, we denote by ~ or (X)'~
the filter generated by the convex sets of ~ ;these in fact form a filter basis, since the intersection of two convex sets of is again a convex set of ~
(2.7.1)
A ~
~
~
. We thus have:
there exists X e i( , X convex, X C A.
We further define
(2.7.2) From the definitions it is obvious that one has
(2.7.3)
X
°
-
(2.?o4)
Lemma.
22
-
A set A belongs to ~ o
a set X • ~
if and only if A contains
which is convex and satisfies
where ~o,1J is the closed unit interval of
[O,1] • X = X, IR.
Essentially the proof of this lemma only uses the fact that if a set X is convex and contains the point O, then it satisfies Let now E be any pseudo-topological
~O,l~ , X = X.
vector space. We
introduce the following filters on E :
(2.7.5)
r~L = sup ~E
~
;
A
(2.?.6)
tY= I L = ~L °
The equality ~ = ~ o thus
(27.7)
~
~/~0]
= ~
Lemma.
a) b)
Proof.
holds since ~0] ~ E
implies ~
~[0] and
. The filter ~ K h a s
the following properties:
[o] _--..'1/'~ ~V'[x] z_ ~
c)
"~ 1~ ~ ~
d)
W t# ~IY:
a) Since [0] ~E,
for all x ¢ E for all ~
[0] ~ sup
~
= ~
;
IR ;
furth.r
~
~
by (2.7.3). b) Same argument, using that
~V. Ix] ~ E.
c) For ~ = O, this is a). For ~ ~ 0 ~
it follows, since
E if and only if 3 - ~ ~ E and since V is convex if and only
if ~.V is convex.
-
d) Let V e ~ ' . U ~I ~ -~=~,
and thus -U e ~ ,
and satisfies I l ~ /
we get
-
Then, by (2.7.4),
V
cnntains a set
which is convex and satisfies U = [0,~ . U. By c),
V ~ V n(-V) ~ U ~ (-U) ( ~
I~I ~ l
23
and x e U
since U E I~ c~
. Thus we have
. The set ~/= U ~ ( - U )
= ~/. In fact, if z 6 I l . ~
~(-U), i.e. x ( U
and - x ( U o
is convex
, then z = ~x, where Since ~ O , ~ . U = U,
~x = I ~ l (~x) @ U. Thus we have V ~ I l ~ / ,
which shows
that V e ~ V ~ o e) Let V 6 1 ~ , also ½U e ~ ;
and choose U as before. By (c) we have
and since ½U is also convex it follows that ½U ~ .
Using again the convexity of U we have: V ~ U ~ ½ U
+ ½U ~ + ~
,
which shows that V ( I ~ + ~ o We define now on ~ a new structure, and denote ~ together with this new structure by E° :
(2.7.3)
X
°
-
>
°
Lemma (2°7°7) immediately implies that (2°7.8) defines a compatible pseudo-topology on ~ (cfo (2.1.1)) and (2.4.2)). This pseudo-topology is in fact a topology, as follows from (2.4.4) or simply using the well known fact that the conditions of lemma (2°7.7) are necessary and sufficient in order that ~ i s
the neighborhood-filter of zero
for a unique compatible topology on ~ (cf. [Q] ). Moreover, since has, by definition, a basis consisting of convex sets, we have:
-
(2.7.9)
Proposition.
24
-
For any pseudo-topological vector space E,
the space E° defined above is a locally convex topological vector space. We f u r t h e r
remark t h a t
as a consequence o f
(2.7.3)
one
has (2.7.10)
E ~
E° ,
with equality if and only if E is itself a topological locally convex vector space.
2.8.
Equabl~ continuity. f : E1----P E 2 being any map between pseudo-topological
vector spaces, we denote by ~f:
E1 x E1 ---~E 2 the map
defined
by
(2.8.1)
~f(a,h) = f(a + h) - f(a).
(2.8.2)
Definition.
l (2.8.3)
f: E1 ~ E
2 is called equably c o n t i ~
iff
J
Proposition.
If f: E 1 ~
E 2 is equably continuous, then
it is continuous (i.e. continuous at each point a ~ El). Proof. get
Let ~ a
El" Then, since
A f( Ca] , ~ _ Fa] ) vI E 2.
we get, using (1.5.2):
~V.[a] $ E 1 and
~-[a]
~ E l, we
But since Z~f(b,x-b) = f(x)-f(b)
4 f ( #a] , X
-[a])~
f(~)
- f([a]).
-
Therefore f ( [ ) the c o n t i n u i t y
- f(a)~E2,
25
i.e.
-
f([)
~f(a)E2,
which proves
of f at a.
The n o t i o n of equable c o n t i n u i t y c o n v e n i e n t l y w i t h respect to composition,
does not behave the composite of
equably continuous maps not n e c e s s a r i l y being equably continuous. However we w i l l
have t h i s convenient behaviour i f
we add a
supplementary c o n d i t i o n . (2.8.4)
Definition.
A map f : E1
bounded map i f f
it
sends quasi-bounded f i l t e r s
quasi-bounded f i l t e r s ,
~#l ~E 1 ~
~.~ E2 i s c a l l e d a quasi.-
i.e.
into
iff
V ' f ( ~ ) ~ E 2.
We denote by ~
(El;E2) the space of equably continuous
and quasi-bounded maps of E1 into E2. C_o (El;E2) is of course closed under addition and multiplication by scalars and is therefore a vector space.
(2.8.5)
Proposition.
If feC_o(E1;E 2) and g~C_o(E2;E3) , then
g . f E C_~(E1;E3).
Proof. From th~ definition of the operator~ it follows that (&(g,f))(a,h) = Z3g(f(a),Z3f(a,h)). Hence we get by (1.5.2):
z f( ,x )).
-
If we assume that ~
~and
26
-
~,it
follows from the assumptions
made on f and g that the filter on the right side of this inequality converges to zero in E3, hence also
3
which proves that gof is equably continuous. that gof is quasi-bounded; hypothesis (2.8.6)
but that follows at once from the
that f and g are quasi-bounded
Proposition.
It remains to show
(using of course
If E 1 is a finite dimensional
(1.4.1)).
vector space
with its natural topology and E 2 a normed vector space, then C~(E1;E2)
consists
exactly of the continuous maps
from E 1 into E 2. Proof.
We already know ((2.8.3)) that, even for arbitrary El,E2,
the elements of --oC(El;E2) are continuous. Let now~ conversely, a quasi-bounded
filter~on
set. Its closure,
f: E1---JPE2 be continuous. E 1 . By (2.5.1), ~ c o n t a i n s
Consider a bounded
which we denote by A, is then a set of ~
is closed and bounded.
which
BuG such a set in a finite dimensional
vector space is comp~act,
f being continuous,
f(A) is compact. Hence f ( ~ ) by (2.5.1) this implies that
we conclude that
contains the bounded set f(A), and f(~)
is a quasi-bounded
We have thus shown that f is a quasi-bounded show that f is equably continuous.
So
filter.
map. It remains to
let, as before, W ~
~ E1 ,
-
and further
~E
27
-
1. We choose again a compact set A ~ ~
in addition a compact neighborhood V of zero in E1 . ~ i
, and denotes
the neighborhood filter of zero in Ei. The set A + V is compact (since A and V are), and hence the continuous map f is uniformly continuous on A + V. U1 ~
~l
This means: for every U 2 ~ ~ 2
there exists
such that f(y) - f(z)G U 2 for all y, z • A + V with
y-z • UI. We can choose U1 sufficiently small, such that U1 c V, and then me have in particular: all a ~ A ,
x~U1,
~f(a,x) = f(a+x) - f(a) G U 2 for
i.e.:Z3f(A,U1)¢U 2. Therefore U 2 ~ z ~ f ( J ~ , ~ l ) .
Since U 2 6 ~ 2 was arbitrary, we have ~ f ( ~ , ~ J C l ) ~ ~
~i'
we have a f o r t i o r i ~ f ( ~ , ~ ) ~
~2'
~2"
And since
i.e. ~ f ( J ~ , X ) ~
E2.
This establishes the equable continuity of f. Proposition. If ~ : E1---~E 2 is linear and continuous at
(2.8.7)
the origin, then ~ is quasi-bounded and equably continuous,
Proof.
By the linearity of ~ : ~.~(x)
we
= ~ (~.x)
and
~
(a,h) = ~ (h),
get .
and
=
and hence the continuity at the point zero yields the two assertions.
-
28
-
We denote by L (£1;£2) the vector space formed by the l i n e a r continuous maps from E1 to E2. The above lemma therefore says: L_ (Ei;E2)c C_o(q;£2) By (2.8.7) and (2.8.3), as in the topological case, the continuity of a linear map at zero implies the continuity at each point.
Howevsr~ for bilinsar (and multilinsar) maps, the situa-
tion is different: continuity at the origin does not necessarily imply continuity at all points.
(2.8.8)
Lemma.
A bilinear map b: E1 x E2 ~
E3 is equably
continuous if and only if it satisfies
i E2
b(ll,A2)~£3
(2)wail12 ~E2Ei ) ~====9b(Ai'~2)~E3 proof, (2.8.9)
we use that for a bilinear map one has ~b((al,a2),(hl,h2)) = b(al,h 2) + b(hl,a 2) + b(hl,h2).
(1)
In order to prove first that the given conditions are
necessary, suppose b equably continuous and let Ill ~£1; |V A2 ~ £2" We put ~ = [0] x J~2; ]( = ll x GO]. Then, using (2.5.5) and
(2.3.7), we have ~V'~ i El x F2 and
~ i E1 x F2. Hence by the
equable continuity of b: ~ b ( ~ , i ) ~
£3. But this is the f i r s t
condition, since for our choice of ~ and X one h a s A b ( ~ , ~ ) b( )~I'~ 2 )"
=
-29-
The second condition is verified similarly. (2) Let us now suppose, conversely, that the two conditions are satisfied, and let ~V. ~JE I x E2, ~ E filters ~ i = T i ( ~ ) '
1 x E2. Then the
]£i = ~i (K) satisfy~V.~ i ~ E i , ~i ~ El'
i = 1,2. Using (2.3.8), (2.8.9) and (1.5.2) we get:
~b(~,~) 6 ~b(~l, ~2' ~i' ~2 )~b(~l' ~2 )+b( ~I' ~2 )+b(xl' ~2 )" By assumption each of the 3 filters on the right Converges to zero on E3. Therefore also their sum and afortiori the left side, which proves that b is equably continuous.
(2.8.1o)
Lemm__~a. If the spaces El, E2 are equable~ then a bilinsar map b: E1 x E 2 ~
E3 is equably continuous
if and only if it is continuous at the origin. Proof. Necessity of the condition follows from (2.8.3). In order to prove sufficiency, let
~ I ~ E I and
E1 being equable, there exists ~ i with ~i ~
~V. ~ 2 ~ E 2 " ~i = W N I ~
El"
Thus
b(~(l,d12)~,b(W~l,~ 2) : b(~l,kV~ 2) = b(~l~V~2). But by (2.3.8), ~ i x ~V d~2 ~F1 x E2, and from the continuity of b at the origin we obtain b ( ~ l , Jl2)~E3~e thus have shown that the f i r s t of the conditions of lemma (2.8,8) is satisfied; the second one is verified in the same way, and the result follows from the lemma.(2.8.8).
- 30 -
(2.8.i1)
Proposition.
If b: E1 x E2-_..~E 3 is bilinear and
continuous at the origin, then b is quasi-bounded. Proof.
Let W . ~ E
1 x E2. As before, J~i ='r.(~l ) satisfy
~'~i~ Ei' i = 1,2 and ~ ~ ~I x~2. Using that ~ = W.W and ~-/~b(Xl,X 2) = b(~Xl,/~ x2) we get
W b(a)~.b(@ll,
J~2) = ~ . ~ . b ( ~ l , ~2 ) = b(~VJll'W~2 )'
and by the continuity of b at the o r i g i n , the f i l t e r
on the~
r i g h t and hence also the one on the l e f t converges to zero on E3.
2.9.
(2.9.1)
Continuity with respect tp the associated_structures. Proposition. If a linear map [ : E1---~E2 is continuous, then also ~" El~---~E2 ~ and ~ : E1 ° _--~E 2o are continuous. •
Proof. (1) Let ~ E 1
~ o Hence ~ ~ = ~ E 1 ,
t(~)~(~)
=~ (~V~) = W - [ ( ~ ) ~E 2,
~E2~.
which shows that ~ (~) (2) Let ~ i
and we get
= ~st~ Ei ~
, i = 1,2• By (2.3.3) for any
continuous mapping f : E1----~ E 2 with f(O) = 0 we have f(~l)
: f('sup
)( ) : sup f ( ~ )
l
(.sup ~ =~/~2"
LE1
But by the linearity of ~ we get also ~ (${i °) which ends the proof•
=
({(~i)
)o
~ 2 °,
-
(2.9.2)
31.,
Proposition. If b: E1 x E2 ~
E3 is bilinear and
continuous at the origin, then b : El~" x E2~---~E3@" is continuous (and hence equably continuous by (2.6.4) and (2.8.10)). Proof.
According to (2.B.lO) it is sufficient to show that
b: El~ x E 2 ~ E 3 ~ • i~Ei ~ , i = 1,2.
is continuous at the point (0~0). So let Hence ~i ~ ~ i = ~ V ~ i ~ E i , and
b(~l,~2)~-b(~l,~2 ) = b(W~l,~2) shows that b(~l, @2 ) ~ E 3 ~ "
=~V b ( N l , ~ 2 ) ~E 3, which
-
§ 3.
32
-
DIFFERENTIABILITY AND DERIVATIVES.
In this section, the definition of differentiability is given and the most elementary results of calculus are proved.
3.1.
Remainders. Let r : E1 ~
E 2 be a mapping between pseudo-topolo-
gical vector spaces El, E2. In order to formulate the condition which will replace the classical condition of Fr~chet, we associate to r s new map
~r: IRxEI - - ~
~r(~,x) =
(3.l.l)
Ilo/~'r(~x)
E 2 defined by
if ~4"0' if ~ = 0 .
(3.1.2)
Definition.
r: E1 ~
we write r • R(EI;E2)
E 2 is called a remainder, and iff r (0) = 0 and
w ~ ~ E1 =-~ er( W, ~ ) ~ E2.
(3.1.3)
Proposition. r: Proof. of
Let
E1~ ~ ](~EI~.
If
r : E1 4 E
2 i s a remainder, then
E2~I i s continuous at the p o i n t zero. Thus
~ ~, ~ = ~ E
~ r and since r(O) = 0 we have r ( ~ . x )
l,
By the d e f i n i t i o n
=~.~r(~,x),
and hence
using (1.5.2). r ( ~ ) ~ r ( ~ ) = r( ~ - ~ ) ~ V ~ ( W , ~ ) . Here the filter ~ = , . ~ r ( ~ V , ~ )
satisfies } -- ~V.I~ ~E 2 and there-
fore r(~) ~E2~" , which completes the proof.
-
33
-
As corollary we remark that for equable spaces El, E2 every remainder is continuous at the origin.
(3.l.~)
Proposition.
R(EI;E2) is a vector space, i.e. if
ri 6 R(EI;E 2) and Qi G ~ for i= 1~2 then
Xz rz * ~2 r2c~(Ez;C2)" Proof.
It is sufficient to show (l)
r i ~ R(EI;E2) ~=~ r i + r 2 ~ R(E1;E2)
(2)
r I R(EI;E2) , ~ ~ R ====~r = R(EI;E 2)
(i) From the definition (3.1.i) of the operatmr •
it is obvious
that ~)(r 1 + r 2) (~,x) = ~ r l ( ~ , x ) + e r 2 ( ~ , x ). Let now ~ / . ~ E 1. Then we get, using (1.5.2)= e ( r l + r2) ( ~ / , ~ ) ~ = ~ ) r l ( W , ~ )
+ ~)r2(~],l~)
and since each term on the r i g h t converges to zero~ the l e f t side does also. (2) is a special case of the following result.
(3.1.5)
L emma.
r @ R(E1;E2) ~ E L_(E2;E3) Proof.
==~
I~-r 6 R(EI;E 3)
)
The linearity of ~ implies
(e(L.r))
~,x) = ~ (erC~,x)).
-
34
-
Thus (@(~er))(W,
Ib) = ~ ( ~ ) r ( W , ~ ) ) .
Supposing now that ~V.~E1, we have 8r(~/, ~ ) ~ E2 by the assumption made on r~ and hence, using the c o n t i n u i t y of (at the origin) : [( @r( ~V,'~)) ~E3~ which completes the proof.
(3.1.6)
Uemma. r12 e R(Ez;E2) ~ ~(EI~E 2)
r2~ R(E2~E;) Proof.
I
r : r23o(r12+ ~) ~ R(EI~E3).
From the definition (3.1.i) of the operator e and the
linearity of ~ one obtains
: (e (r23o(e Thus by (1.5.2): er(~V,~) ¢~ er23 (W,~(I~) +@r12 (%V,I~)) - er23(\v,~), where we put ~ = ~
(~) ÷ @r12 (~1,1~), which satisfies
~v~ ~ ~v[(~) +~v @rz2 ( W , ~ ) If we assume now that ~ V ~
El, we see that ~V~E2,hence
@rl2(~V,~) ~E 3, and the above inequality for ~r(IV,'~) shows that also
e r ( W , ~ ) ~ E3, which proves that r is a remainder. The next result is only valid if the secono space
is separated.
-
C3.1.7)
35
-
Definition Ccf~4] ) A pseudo-topological
space E is
separated iff it satisfies
~ × E and
X~yE --~ × = y.
For a pseudo-topological vector space E this implies in particular ~ E
(3.1.8)
~
x= 0
(*)
Lemma. If E 2 is separatedp then the only remainder r : E1 ~ Proof.
E 2 which is linear is the zero map.
Let x ~ E 1 . By (2.4.2), ~V. [x]~E1, hence
But since for a linear map r one has that
~r~,x)
@ r ( ~ V , [ x ] ) ~ E 2.
= r(x), it follows
~r( IV, Ix] ) = r(~x]) = ~r(x)] . So we have ~r(x)] ~ E2,
hence r(x) = O.
3.2.
Differentiability at a point. In order to make use of (3.1.8) we assume henceforth
that all spaces El~ E2,... are separated. Let f: E1----~ E 2 be any map of pseudo-topological vector spaces, and a ~ E1 .
(3.2.1)
Proposition.
There exists at most one ~eL(E1;E2)
such that the map r defined by f(a+h) = f(a) + ~(h) + r(h) is a remainder.
(*) This condition is in fact sufficient to make E separated;
of. C4].
-
Proof.
36
Suppose there exist two, ~
-
and e2, such that the maps
r i are remainders, where ri(h) = f(e+h) - f(a) - ~i(h),
i = 1,2.
Then the map r = rI - r2 = ~2 " ~l is by (3.1.4) a remainder and is linear, hence is zero by (3.1.8), which completes the proof.
(5.2.2)
Definition.
If there exists a
~¢
~(E1;E2) such that
the map r defined by f(a+h) = f(a) +
~(h) + r(h)
is a remainder, then the map f: El
~ E 2 is said to
be differentiable at the point a; and the map
e ¢ ~(E1;E2)
which by (3.2.1) is uniquely determined, is then called the derivative of f at the point a. It will be denoted as follows: ~ = Df(a)
C3.2.31
or
~= f'(a).
Example: A constant map K: E1-----~E 2 is differentiable at each point
a G E1 •
It will be convenient to write Df(a).h instead of (Df(a))(h). If f is differentiable at the point a, the uniqueness of the derivative implies the uniqueness of the remainder r.
-
37
-
We use a similar notation :
(3.2.4)
r = Rf(a).
I (3.2.5)
Therefore:
t(h) = Rf(a)(h), r(~h)
and
= (8(Rf(a)))(~,h)
Propositio n .
If f: E1
point a, then f: E1$ Proof.
~
=~Rf(a),(A,h) E 2 is differentiable at the
~ E2¢ is continuous at the point a.
By assumption, f(a+h) = f(a) + [(h) + r(h), where
~(E1;E2)
and r6 R(E1;E2). By (3.1.3), r: El4F-- ~ E2~ is conti-
nuous at the point zero. Further, ~: E~----~ E2b is continuous according to (2.9.1). Now the result is obvious. Remark.
If in definition
(3.2.2) one only requires
linear, one still gets the uniqueness of the derivative, as the proof of (3.1.8) shows. However, it is essential for the theory, to impose the condition that ~:E1----+E 2 is continuous, since otherwise the proof of the chain rule would not work. On the other side, we see no reason to restrict the considerations to mappings f: E1----~E 2 which are continuous;
in view of (3.2.3),
what we need is that f : El~----~E2 ~ is continuous.
For equable
and in particular for topological spaces, this distinction disappears,
since then Ei = Ei~ .
-
3.5
(3.3.1)
38
-
The chain rule. Theorem.
g f ~ E 2 ___--~E 3. Then :
Suppose we have maps E1
f differentiable at a ~ E1 I ~=~Ig,f differentiable at a, g differentiable at b=f(a~
ID(g.f)(a)=Dg(b) • Df(a)
Proof. By assumption one has f(a+h) = f(a) + ~l(h) + rl(h)
g(b+k) = g(b) + ~2(k) + r 2 ( k ) , where tI = Df(a)@~(E1;E2),
rI • R(EI;E2),
~2 = Dg(b) ~ ~(E2;E3) ,
r2 G R(E2;E3).
Composing the.mappings one gets, using the linearity of ~2 : (g,f)(a) = (gof)(a) ÷ ~(h) + r(h),
where
~(h) = ~2(~l(h)) and
r(h) = ~2(rl(h)) + r2(~l(h)+rl(h))
Obviously, ~= ~2° ~l ~ ~(E1;E3)" Further, using (3.1.5), (3.1.6) and (3.1.4), re R(E1;E3) , which completes the proof.
3.4
The local caracter of the differentiabilit~ condition. Since we consider pseudo-topological
spaces E, we have
to say what we mean by "local". We call E-neighborhood of a point x ~ E a set U with U6sup " ~ = ~ x ,
which means: U •
-
39
-
for all ~ with ~ ~x E. By (2.7.3)p each neighborhood of x ~ Ee is an E-neighborhood of x. A set which is an E-neighborhood of each of its points is called an E-open set. In particular, each set which is open in E° is an E-open set. In [ 4 ] Fischer showed that for arbitrary pseudotopological spaces, the E-open sets are the open sets with respect to a certain topology (p. 273). In general, the filter ~x
is strictly finer than the neighborhood filter of x with
respect to the mentioned topology. Fischer states that in the case of pseudo-topological vector spaces (and more generally for pseudo-topological groups) equality holds and he establishes this by showing that ~ o + ~ o
= ~o
(cf'E4 ] ' Satz 6, p. 294).
Since it seems to us that there is a gap in his proof, we do not use the above topology (not knowing whether it is compatible), but only the associated locally convex structure introduced in 2.7.
(3.4.1)
Propgsition.
Suppose that two maps fi: E I " ' ~ E 2
(i = 1,2) coincide in an El-neighborhood U of the point a~ E1 . Then if fl is differentiable at the point a, f2 is also and f~(a) = f½(a).
o
Proof.
40
-
By assumption we have, for ell h G E 1 : fl(a+h) = fl(a) + ~l(h) + rl(h),
where ~I~L(E1;E2)
and rl¢ R(EI;E2).
We define r 2 : E1 -----~E2 by f2(a+h) = f2(a) + el(h) + r2(h). The proposition is proved, if we show that r2~ R(EI;E2). Since we have fl(x) = f2(x) for xG U, we get: r2(h ) = rl(h )
for all
h ~ U - a =W.
Here, by (2.4.1), W is an El-neighborhood of O, which means W~ ~
for all~
with ~ E I.
Let now ~ be a quasi-bounded filter on El: @ @ ~ E1 • Hence W~.I~,
so there exist V ~ W and B ~
with V.BcW. One has
therefore: @rl(~.x ) = er2(~,x )
for
~ ~ V, x ~ B.
But this implies that erl(~V,I~) = @r2(\V,I~). A ~ @rl(~V,~),
In fact, if
then there exist V16 W and Bl~1~
A ~ @ rl(V1,B l) ~ @ r l ( V l n V , B l ~ B ) that A ~ @r2(~V,~).
with
= @ r 2 ( V l n V,Bl~ B), showing
Similarly fore the converse. But now it
is obvious that rI ~ R(E1;E 2) implies r2GR(E1;E2).
-
41
-
Because of the local caracter of the differentiabiiity condition it would be appropriate to consider maps defined on E-open subsets of the respective pseudo-topological vector
spaces E. For a map f :
A ----~E2,
where A ¢ E1 i s an El-open
s a t , one would i n t r o d u c e d i f f e r e n t i a b i l i t y
at a p o i n t a6 A
and f'(@) by choosing any extension ~ of f to El, e.g. by t a k i n g ~ ( x ) = 0 f o r x ~A. However, i n order to s i m p l i f y presentation,
the
we s h a l l c o n t i n u e to consider maps defined on
the whole space.
-42-
§ 4.
4.1.
EXAMPLES AND SPECIAL CASES.
The classical case.
(4.1.i)
Proposition.
If El, E 2 are normed vector spaces, on
which we consider the pseudo-topology
(i.e. topology)
determined by the norm, then the notions of differentiability at a point and derivative of a map f: E1---~E 2 coincide
with the classical notions in the sense of
rr~chet (cf. [3 ]). Proof.
All we have to show is that a map r: E1----~E 2 is a
remainder (i.e. r(R(E1;E2) , cf. definition
(3.1.2)) if and only
if it satisfies the classical Fr~chet condition
(4.1.2)
lim ×-~o
Ur(x) il
llxII
(1) Suppose r satisfies
q(×) =~II
(4.1.3)
= O.
(4.1o2), and let W'B~E 1. We put
llr(x)ll if x÷0,
L"
ifx~O.
Then, by (4.1.2), q is continuous at the point O, and thus q(W'~)~IR,
which means: q(~V.~)~V.
Since one has II~r(~,x)ll = ~xlI' q(~x), one gets using (1.5.2):
II er ( w, S )11 a, II S II • q (w.a)
= U q( ~. ~)" ~
II
II V. 1BII.
But since ~.a~E 1 and the norm is continooos, II~'~IlJJR, hance also e r ( W,S)II ~ ~, whioh yields, by the definition of the pssodotopology induced by the norm: er( W , ~ ) ~ r& R(E1 ; E2) .
E1 • This shows that
-
(2)
43
-
Suppose conversely that r~ R(E1;E2) and let q be as before.
It is sufficient to show that q is continuous at the point Oo So let ~ E l ;
we shall show that then q(~)~IR. We still introduce
the map s: El ~ E
l by I
s(x)
=
l
. x
for x ~ O ,
Itxlt 0
for x =
O.
Since for all x6 E1 , lls(x)~l~l, s ( ~ ) certainly contains a bounded set, hence by (2.5.1) is a ~ o u n d e d
II ~11 ~
4.2. (4.2.1)
w
and q ( x )
= ~
@r(Uxll,s(x))ll
filter on E1 . We further have
•
so ~e get,
using (1.s.2):
Linear an d bilinear maps. Pro ppsition.
A linear and continuous map f: E1 ~
E2
is differentiable at each point a ¢ E1 and f' (a) = f. The proof is obvious: taking e = f and r = 0 one has f(a+h) = f(a) + ~ ( h ) +
(4.2.2)
Lemma.
r(h), and ~ L ( E 1 ; E 2 ) ~ rER(E1;E2).
Let b: E1 x E 2 ~
E3 be bilinear and continuous
at the point zero. Then b~ R(EIxE2;E3). Proof.
One has for all ~61R, (x1,x2)~ ElXE 2 : 8b(~,(Xl,X2)) =~.b(Xl,X2).
Hence
~b(\V,~)
- ~V.b(~).
Further, since ~.~.b(Xl,X2) = b(aXl,/~x2) , one has \V.b(~l, ~2 ) = ~" \V.b(~l,~2 ) = b(~V~l, ~V~2).
-
44
-
Suppose now that~.'B ~E 1 x E 2. One has 1 ~ 6 T l ( ~ ) where
~'l = ]T.(~)I satisfies W ~ i ~ E i ,
eb(W,I~) (4.2.3)
=~.b(~)~.b(
i = 1,2.
x IT2cB), Hence
e l , q~2) = b ( ~ V ~ ,
~1~2)1E 3.
Proposition. Let b: E1 x E2 ----~E 3 be b i l i n e a r and continuous. Then b is differsntiable at each point a = (al,a2)~ E1 x E2, and b'(al,a2)(hl,h 2) = b(hl,a 2) + b(al,h2). Proof. One has, for h = (hl,h2)E E 1 x E 2 : b(a+h) = b(al+hl,a2+h2) = b(al,a2) + b(hl,a2) + b(al,h2) + b(hl,h2). So we have, with ~(h) = [(hl,h2) = b(hl,a2) + b(al,h2) and r(h) = r(hl,h2) = b(hl,h2)s b(a+h) = b(a) + ~(h) + r ( h ) .
is obviously linear, since b is bilinear, and also continuous, since b is continuous. Thus {C~(ElXE2;E3). And r~ R(ElXE2;E 3) by the preceding lemma. This completes the proof. @
4.3.
(4.3.1)
The special case f: IR
• E.
Proposition. If f: ~R ~ E
is differentiable at the
point ~ {~, then the following limit, which we denote by f'(o(), exists:
(4.3.2)
f'(~) Further one
= lim
then has:
f(~ +~)-f(~)
.
-
Proof.
45
We have f((+~) = f(~) + ~(X) + r(~), where ~¢L(IR;E)
and r GR(IR;E). We put [(1) = f'(~).l = a and define q: IR---~ E by q(~)
~f(~÷~) - f(~)
if~,O,
~
~f ~ = 0.
One has
e(~) = [(~.i) = %. t (i) = ~ . a . Hence r(~)
Since q(~)
-- ¢ ( , ~ . ~ )
er(~,~)
:
Or(~,l)
= q(~)
- f(,~)
f(~+~)
W.I~IR we have
k
- ~.a,
" f(~)
-~(.a
- a. (Definition
3.1.2)@r(~V,1)~, E,
hence
- a~ E resp. q( IV)~a E, and that means exactly that
lim q(~) = a ; the proof is complete. ~--~o In the classical case, the converse of this proposition holds. However it seems, that in the general case, our differentiability condition is a little bit stronger. Again some question of equability comes in.
(4.3.3)
Pro~osltion.
If the scalar multiplicationIRxE
~ E
of E is equably continuous (thus by (2.8.10) in patti.(*) cular if E is equable or even topological), then the existence of the differential quotient (4.3.2) is sufficient for the differentiability of f: IR the point ~ . (*) more generally if E is admissible (see appendix (i)).
rE at
-
Proof. Let a = lim ~--~o
r
(w)
46 -
, q being defined as before; we
can write therefore: q( W)-a ~ E. We define ~: fR---* E by : ~(~) = ~.a, and r: IR--~E by r(~) = f~+~) - f(k) . ~(~). Obviously ~ ( I R ; E ) ,
and it only
remains to prove that r~R(IR;E). We havep according to the above de£1nition of r: ~r(~,W)
= ~.(q(k~)-a).
Thus we get, if ~ ,
i.e. if W ~ ~ ~, using (1.5.2):
@r(~,fl@) ~ ~. (q( W ~ ) - a ) ~ ~ . (q(W)-B). On the r i g h t , we have the product of a quasi-bounded f i l t e r onlR with a zero-converging
filter
on E, which converges to zero under
the hypothesis that scalar m u l t i p l i c a t i o n i s equably continuous.
4.4.
(4.4.1)
Differentiable mappings into a direct product. Proposition.
Let fi : F i
~ El, i ~ l , be a family of
mappings of pseudo-topological vector spaces. Then (cf ( i . 3 . i ) ) X f i : ~ < F i - - - ~ X Ei iel ilI i~l is differentiable at the point (xi)i~ I if and only if fi : Fi i~I,
Ei is differentiable at the point x i for all and then ( X fi)'([xi) i~ I ) = ~ f'(xi)° i6I iQI i
-
47
-
In particular:
(flxf2)'(Xl,X2)
=
(q(Xl), f (x2))
The proof is a combination of the following two lemmas. (4.4.2)
Lemma.
Let ~i : Fi
) Ei, i~ I, be maps. Then
X ~ i ~ L ( X Fi; . M E i) ~ icl i~I i~I all il I .
~iaL(Fi;Ei ) for
Proof.
One verifies separately that >~ [i is linear if and ieI only if each ~i is linear and that ~ ~. is continuous if and o n l y i f each ~.1 i s c o n t i n u o u s .
(4.4.3)
Lemma. Let ri : F i
~ Ei, itl, be maps. Then
Mr ~ R(X Fi;XEi) ~ iEI i i~l i~l Proof. Let us denote, for j~I, by 3
ri~ R(Fi;Ei) for all itI : XEi i~I
~E. and J
wj
:XFi---'~FjI the projections, and put r = M r . . ie iGI I One has (4.4.4)
r.
J
ow
.
J
=
a) Let r j 6 R ( F j ; E j ) follows that
"~.
J
or,
for a l l j * I .
From (3.1,6) and (4.4.4) i t
Ir..r6R(X Fi;Ej) for all jeI. Suppose now that J i~ I
WI~ ~ X F.. Then e(1"jor)(W,I~) ~Ej for all j6I. But "~. being i~I i j linear we have ~(Tj.r) = ~'jo@r. Hence Irj(~r( ~V,~))~ Ej for all jeI, which yields
@r(~V,I~) ~ X Ei. This shows that r~[~(>¢ F i ; X El). il I iG I i~ I
-
b)
48
-
Let r ER(Xi~Ifi ;i¢~Ei)I . Then, by (3.1.5) end (4.4.4) we get
rj.,wj IR(~Fi;Ej)i¢l for all jtl. Denoting by ~j: fj ----~>~i61Fi the map caracterized by the conditions
wk -~j . 0 for k ~-j,
wj o~fj = identity, we now conclude from the continuity and linearity of
Lfj by (3.1.6) that rj = rj,(wjo~fj) = (rj,wj)Q~j
belongs to R(Fj;Ej).
(4.4.5)
Proposition. Let fi : E----~E i, i61~ be a family of mappings of pseudo-topological vector spaces. Then (cf.(1.3.2))
~T i~I
f. : E i
~ ~ Ei is differentieble iel
at the point x~E if and only if fi : £ ----~Ei is differentiable at the point x for all iEI, and then:
( IT f i ) , (x) . "it f: (x). i~I i~I In particular:
Efl, f2],(x> ° [q(x>, Proof. a) Let all the maps f1 be differentiable at x. Then by the preceding proposition ~ fi is differentiable at d(x), where d is i~l the diagonal map of E into its 1-fold direct product El. Further, the map d: E ~
E l being linear and continuous, is differentiable
by (4.2.1). Now the differentiability of -~'fi and also the given iEl formula for its derivative follow from the chain rule, (1.3.3) and
(4.2.1).
-
b) Let
49
-
Jl f be differentiable. Since f =IT.o II fi' where i~l x J J i~l
the projection map 7. is linear and continuous, the differenJ tiability of fo follows again from the chain rule and (4.2.1). J
-
65
-
§ 6.
PSEUDO-TOPOLOGIES ON SOmE FUNCTION SPACES.
6.1.
The spaces B(E1;E2), Co(El;E2) and L(EI;E2). The set _B(EI;E2) of all quasi-bounded maps from E1 into E2
(cf.(2.8.4)) is a vector space, since from %V.(fl+f2)(~) & W . f l ( ~ )
+ ~V.f2(@) (cf. (1.5.2)) and
~V.(Af) (%) = A.(~V.f(%))
one deduces easily that any linear com-
bination Al.f I + ~2~f2 of two quasi-bounded maps fl' f2 is also quasi-bounded. We denote by B(E1;E2) this vector space together with the pseudo-topology caracterized by the following condition: (6.1.1)
¢~8(E1;E 2)
Iff
~V.B~El= ~
~(I~)~E 2.
@e claim that this definition yields a compatible pseudotopology on the vector space _B(E1;E2). For this, we have to verify that conditions (2.1.1) are satisfied at the point zero, and that the compatibility conditions (2.4.2) hold.
Of the three condi-
tions (2.1.1), only the second one, which demands that ~l v ~2~B(EI;E2)
if ri ~B(EI;E2) for i = 1,2, is not obvious.
But this follows easily, making use of the equality
(6.1.2)
(FlVr2)(
)
= rl(
)
which is a consequence of the set-theoretic equality (FIWF2)(B) = FI(B) w F2(B).
-
66
-
The compatibility conditions (2.4.2) follow respectively from Ca)
(rl + r 2 ) ( ~ ) ~
(b)
(~.r)(~) = ~.~(~)
(c)
(w.r) (~) = v.Y(~)
(d)
(W.[f])(~)
=
rl(~ ) + r2(~ )
(cf.(i.5.2));
W.f(1~) .
We remark that for the verification of the fourth compatibility condition it is essential that B(EI;E2) is not the space consisting of all mappings from E1 into E2, but only of the quasibounded ones. We next consider the aubspace consisting of the quasibounded and equably continuous maps from E1 into E2, which was already introduced and denoted by C.Co(EI;E2) in section 2.8. We denote now by Co(El;E2) the vector space Co(E1;E 2) together with the pseudo-topology induced by the inclusion. We thus have:
(6.1.3)
F~Co(EI~E2)
iff
V . ~ E1 --~ r(1~) ~ E2.
In the same way, L(EI;E2) denotes the pseudo-topological vector space whose underlying space is the space ~(E1;E 2) of continuous linear maps from E1 into E2, together with the pseudotopology induced by the inclusion of ~(E1;E 2) in Co(El;E2) or in B(EI;E2). Therefore we have:
(6.1.~)
r~ LCEI~EZ)
iff
~.~ ~ El ~
¢(~) ~ E2 .
-
67
-
We shall also need multilinear maps. Let ~(E!,...,En;E~ be the vector space consisting of the equably continuous multilinear maps from EIX...×E n into E. We have ~(E1,...,En;E ) c~(EIX...XEn;E),
and
thus we can consider on ~(EI,...,En;E) the pseudo-topology induced by that of B(ElX...×En;E ). Together with this structure we denote the space of equably continuous multilinear maps by L(E1,...,En;E), and we have therefore:
(6.1.5)
~ * L(EI,...,En;E )
v. ~i I E i
iff
for i = i,...,o ~
The case E 1 = E 2 . . . .
F(~l,-..,~n)
~ E.
= E n will be of special interest
and we will use the shorter notation
(~.l.6)
Ln(E1;E) = L(EI,...,E1;E). Besides these structures we can consider the structures which are associated to them by means of the operators " ~'' or ,,.,,. It will be convenient not to write the operator ~t the end, but immediately after the B,Co,L or Ln; e.g.
(~.I.?)
Co$ (E1;E2) = (Co(El;E2))~ In special cases, the structure of B(EI;E2) can be topological:
(~.l.B)
Proposition.
If E l is a normed vector space snd E 2 a
topological one, then the pseudo-topology of B(EI;E2) is the topology of uniform convergence on bounded sets.
-
Proof.
68
-
Denoting by B*(EI;E2) the space B(E1;E 2) together with
the topology of uniform convergence on bounded sets, we use that this topology is caracterized as follows:
F~B*(E1;E 2) if and
only if for each bounded subset B of El,
~(B) ~ E 2. Using lemma
(2.5.1), one shows immediately that
r ~ B.(q ~E2) ~
f ~ e(q ~E2),
which completes the proof. Combining this result with Proposition (2.8.6) we get now immediately
(6.1.9)
Proposition.
If E1 is a finite dimensional vector
space with its natural topology and E2 a normed vector space, then Co(El;E2) is the space of continuous maps from E1 into E 2 with the topology of uniform convergence on bounded sets.
(6.1.10)
Proposition.
If El and E 2 are normed vector spaces,
then L(E1;E2) is the space of continuous linear maps from [~ into E2 together with the topology induced by the usual norm (cf [ 3]) on ~(E1;E2). Proof. One combines (6.1.8) with the well known fact that the normtopology on ~(E1;E2) is the topology of uniform convergence on bounded sets.
-
(6.i.n)
Remark.
69
-
However, if we assume E1 and E2 to be topolo-
gical, not even L(E1;E2) is topological in general. In fact, as it was shown by H.H.Keller (cf.[5 ] ), there does not exist for non-normed vector spaces a topology having the properties which we shall need and which we shall verify in the sequel. The structure used in our theory on L(E1;E2) is different from the one used by Bastiani [I ] and Binz ~2 ] . In the case of locally convex topological vector spaces, our structure is related to a pseudotopology introduced by H.H.Keller in [5 ] , using families of seminorms, as follows: the structure of Keller is that of our L f (El;E2). A detailed discussion will be given in m forthcoming paper.
6.2.
(6.2.l)
Continuity of evaluation maps. Proposition. The evaluation map e: Co(El;E2) x £I ~
E2'
defined by e(f,x) = f(x), is continuous. Proof.
We show continuity at (f,x). So let ~ ~f Co(£1;£2) and
X ~x £1" Then
~-
f~Co(E1;E2) and %/. X ~E I. By (1.5.2) we get
TC ) Cf-
+
-.
"]'0
-
The first term on the right side converges to zero on E2, the second (using that f is continuous) afortiori
to f(x); hence the sum and
the left hand side converge to f(x): ~ ( ~ )
(x) E2 '
which completes the proof. We remark, that we did not use the equable continuity of f here; but continuity was essential.
(6.2.2)
Corollary. The following evaluation maps are conti~UOUS:
e: L(EI,...,En;E ) x E I x ... x E n e: Lp(E1;E ) x E1 x ... x E 1 ~
• E; E.
Since evaluation of multilinear maps is a multilinear mapping, we get by the generalisation of (2.g.2) to multilinear maps:
(6.2.3)
Proposition.
The evaluation map
e: Lp ~ (El;E) x El$ x ... x El$-------~ E ~ is continuous. The evaluation map of (6.2.1) is not, however, bilinear. But since (2.5.3) implies Co~ (E1;E 2) x E1 ~ Co(El;E2) x El, we get"
(6.2.4)
Proposition.
The evaluation map
e: Co~ (El;E2) x El is continuous.
~ E2
-
71
-
One of the difficulties of the theory is due to the fact that these evaluation maps, such as e.g. the map e: L(EI;E2) x EI---gE2,
6.3.
(6.3.l)
fail to be equably continuous in general.
Continuity of composition maps. Proposition. The composition map
c: B(El;E2) x L(E2;E3)---~B(E1;E3), defined by c(f,t) = ~°f, is continuous. Proof. Let ~ ~gB(EI;E2) and£ ~kL(E2;E3). The map c being bilinear, we have, for f, g¢ B(EI;E2) and ~, k•L(E2;E3): ~.f-
k,g : ( ~ - k ) . ( f - g )
+ k,(f-g)
+ (t-k).g.
This i m p l i e s , using ( 1 . 5 . 2 ) : c(~ ,~) - c ( g , k ) ~ (~=k).(~-g) Let n o w ~ . ~ E
+ k'(~-g)
* (~-k),g.
1. Then, again using (1.5.2):
(c(~,Z)-c(g,k))(~)~(~-k)(~
-g)(~)+k((~-g)(~))+(~
-k)(g(~)).
Since each of the three terms on the r i g h t hand side converges to zero on E3~ also the left hand side does. This implies:
- c(g,k)
B(EI
E3),
which proves the contiffuity of c at the point (g,k).
-
72
-
Using (2.3.6) we get:
(6.3.2)
Cqrollary.
The following composition maps are conti-
nuoue:
c$
L(EI,...,En;E)xL(E;F)
c=
Lp(E1;E2)xL(E2;E3 )
L(EI,...,En;F);
; Lp(E1;E3).
Again, these composition maps fail to be equably continuous in general. However, (6.3.2) implies by (2.g.2) the following
(6.3.3)
Proposition. The following composition maps are equably continuous: c: L ~(EI,...,EnIE) x L ~ (EIF)
L $ (EI,...,En;F);
c: Lpf(E1;E2) x L4F(E2;E3)--.---~LplP(EI;E3).
6.4.
Some canonical isomorphisms. We first consider the vector spaces L(IR;E) and E,
which are isomorphic, and we investigate whether they are also homeomorphic.
(6.4.1)
Lemma.
(I):
(a) The canonical isomorphism
L(IR;E) ~
E,
defined by #(e) = [(l), is continuous. For its inverse
map ~ ,
caraoterized by (~(x))(~) =~.x, we have
(b) 1~,r: Ee...-$L(IR;E)
is continuous;
(c) ~ : E
is continuous provided that
~L(IR;E)
scalar multiplication of E is equably continuous.
-
73
-
Proof.(a) ~ being the evaluation at l~ its continuity follows from (6.2.Z). (b) Let ~ E ~
Hence • ~ ~ = W . ~ ~E. Thus we get for any
with ~.~ ~IR:
which shows that ~ ( ~ ) ~ L ( I R ; E ) .
The assertion now follows by
(2.a.7). (c) Similarly we get, if ~ ~ E and ~.I~ ~ ]R:
and we deduce from (2.8.8) that this converges to zero on E. The rest goes as before. From (a) and (b) of the above lemma we get for an arbitrary E, using (2.9.1):
(6.4,2)
Proposition. There is a natural (*) linear homeomorphism
(IR;E)
E*
From (a) and (c) of lemma (6.4.1) we get:
(6.4.3)
Proposition. If scalar multiplication of E is equably continuous, in particular (cf.(2.9.2)) if E is equable, then we have a natural linear homeomorphism: L(~;E) ~ E.
(*) We do not discuss the categorical meaning of "natural" or "canonical", so the statement simp3y means: the isomorphism which we are considering is a homeomorphism.
-
74
-
This in particular implies that L(IR;E) is equable if E is equable. We further mention, that from (2.8.8) and (2.5.1) it follows easily, that scalar multiplication on E is equably continuous if and only if X~E implies Il.~ ~ E; this condition will be satisfied by the so-called admissible spaces considered in § 7 and later.
(6.4.4)
Proposition. If E2 is equable, then we have a natural linear homeomorphism: L(E1;L(E2;E3)) ~
L(EI,E2;E3).
Proof. (a) Let ~L(E1;L(E2;E3) ). We consider the map ~ ~--~(~), where ~(~) = b is the map from ElXE 2 into E3 defined by
(6.4.5)
b(Xl,X 2) = (~(~))(Xl,X 2) = ( t ( X l ) ) ( x 2 ) . This map b: ElXE2
p E3 is obviously b i l i n e a r . We show Further-
more, that b satisfies the conditions
(6.4.6)
& l ~E1; ~V.1B2~E2 ~
(6.4.7)
\V'1@l~E1;
~(2~E2 ~
b()[1,1~2) ~E3; b ( 1 ~ l , ~ 2 ) ~ E 3-
In fact, (6.4.6) follows easily: since ~:E 1
~LCE2;E3) is
continuous, ~1~[1 implies ~(~l)$L(E2;E3) , and hence (~(~[1))(~ 2) ~ E3. But by (6°4.5), (~(Z1))(~2)
= b(IEl; I~2).
In order to establish (6.4.7), we have to use that E 2 is equable. Therefore
~2 ~E2 implies
~2 ~- ~2 "- V . ~ 2 ~ E 2 ,
b ( l ~ l , ~ 2) = ( [ ( 1 3 1 ) ) ( l [ 2 ) _ z - ( ~ ( ~ l ) ) ( \ V ~ 2 ) =
and we obtain
(~(tV.~l))(~2).
-
75
-
Here, the filter on the right side converges to zero on E3, since t(W.1~l) ~ L(E2;E 3) by the continuity of ~ and s i n c e ~ V . ~ 2 ~ E 2. From (6.4.6) and (6.4.7) it follows now by (2.8.8) that b: ElXE2--.--~E 3 is equably continuous, and hence b~L(E1,E2;E3). We have shown so far that the map ~--~W(~)
= b has its image in
L(E1,E2;E3), i.e. we have
(6.4.8)
~: (b)
L(E1;L(E2;E3))---eL(E1,E2;E3)-
In order to show that ~ i s bijective, we now construct a map
~: L(E1,E2;E3) ---*L(EI;L(E2;E3)).
(6.4.9)
So let b ~L(EI,E2,E3). a map
We first define, for any fixed Xl~ El,
~Xl : E2.___~E 3 by
Xl(X2) b(Xl,X2). =
Since b: ElXE 2 ___~E 3 is continuous, and we have: ~xl~L(E2;E3).
~Xl : E2
We define e=~(b)
~E 3 is continuous, : E1
~L(E2;E3) by
~(x l) = ~Xl , and we thus have caracterized #(respectively by the equation
((~(b)(Xl))(x2)
~(b))
= b(xl,x2).
= ~(b) is obviously a linear map from E1 into L(E2;E3). We want to show that
~ :E1
~L(E2;E3) is continuous. So let
~l ~El" We have to show that ~(~l)~L(E2;E3) , which is equivalent to the condition \V.'I~2~E 2 ,,,,,~(~-(~i))(~ 2) ~E 3,
-
76
-
and this holds by (2.8.8), since ~(~i)(I@2) = b(~l, ~2 ) by (5.4.5). We remark that at this point it was essential that b~L(E1,E2;E3) implies according to the definition of the spaces L(E1,E2;E3) that b is equably continuous. Having shown that ~= ~(b) is linear and continuous, we know now that in fact t = ~ (b) ~ L(E1;L(E2;E3)), and that therefore ~ i s a map as stated in (6.4.9). It is easy to verify that 4(~(b)) = b for all b£L(E1,E2;E3) and ~ (~({)) =~
for all {¢L(E1;L(E2;E3)) , and
thus ~ is bijective, ~ being its inverse. (c)
We show that the map ~ of (6.4.8) is continuous. Since
is linear, we have to verify its continuity at the origin. So let Z ~L(EI;L(E2;E3)). In order to show that ~(Z)~L(E1,E2;E 3) , we have (cf.(6.1.5))to form ~ ( ~ ) ( ~ l , ~2 ) and to verify that this converges to zero on E3 provided that ~.I~i ~Ei, i = 1p2. But by (6.4.5) we have ( ~ ( ~ ) ) ( ~ 1 , ~ 2 )
= (~(~1))(~2),
and
here Z ( @ l ) ~L(E2;E3) and thus (~ (~1))(I~2) i E3. (d)
We finally show that the map ~ =~'l of (6.4.9) is continuous.
In fact, let ~L(EI,E2;E3).
to
W . ~ IEl ~
# (r) ~ L(E1;L(E2;E3)) is equivalent
(~(~))(~i)
~L(E2;E3),
and this again is equivalent to W~I~EI,
~325E2
~'~((F(T))(~I))(~
2) ~E 3,
- 77 -
and this holds since by the equation caracterizing ~ one has
((
=
I 2)"
This completes the proof of (6.4.4).
(6.4.1o)
Lemma. L~(EI~;E2) = LIF(EI~;E21r). Proof. (a) We first show that the underlying spaces are the same. If~IL'(EI';E2"), then ~: EI~_....~E2( is continuous, and thus afortiori ~: E1~ ----.~E2 is continuous, showing that t ¢ L#(EIf;E2 ).
Conversely, if ~Lf(Ell;E2) , we deduce from
the continuity of ~: Ela-.._~E 2 and (2.9.1) the continuity of [
:
(b)
E#.-
,~ E2¥.
We now show that the structures are the same. One part is
immediate: if ~ $ Lf(El~;E2~), then ~
Lf(El#;E2 ). Suppose con-
versely that~LW(El#;E2 ). Then ~ m ~ = ~/.)~$L(El*;E2). Let ~/'~l~Ele" Then )~(~i) IE 2. But since )I(151) = (W.~)(~S1) = ~V.~('S1), we have ewen:~(~l)~E2#.
This shows that]~L(E#;E2~), and since
~ ~ = \V.~ it follows that ~ IL~(E1t ;E2~).
(6.4.n)
Proposition.
If E1 and E2 are equable, then we have
a natural linear homeomorphism:
Llr(E1;t ~(E2;E3)) = Lf(EI, E2;E3).
-?8-
Proof. The map ~ of (6.4.8) being linear and continuous, we deduce by (2.g.1) that also
L"(EI;L(E2;E3))---, Lt (EI,E2;E3) is linear. The same holds for the map
~ = ~-l. Now using that
E = El@ by assumption, the result follows, since by lemma (6.4.10) we have Lf(E1;L(E2;E3)) = L~(EI;L~(E2;E3)).
(6.4.12)
Proposition.
If E1 is equable, then there are natural
linear homeomorphisms as follows: Lp(EIILq(EIIE2))~Lp+q(EIIE2); Lp$(E1;Lqg(E1;E2))~ L~+q (El;E2). The proofs are completely analogous to those of propositions (6.4.4) and (6.4.11). Instead of (2.8.8.) one has to use the analogous result which states that a multilinear map : E1 x ... x En ---~E is equably continuous if and only if R(~l'''"
@n ) ~ E provided that one of the filters @l'''''@n
converges to zero, the others being quasi-bounded. And at the place of lemma (6.4o10) one has to establish the corresponding equality
The map which we call natural is of course the following: to EQLp(E1;Lq(E1;E2))
we associate ~((~)~Lp+q(EI;E2), Q((R) being
ceracterized by (~e)(Xl,...,Xp+q)
= (~ (Xl,...,Xp))(Xp+l,...,Xp+q).
-
(6.4.13)
79
-
Proposition. The map (fl,f2) i
~l,f2]
(cf.(1.3.2))
yields natural linear homeomorphisms as follows:
CoCE;EI) x CoCE~E2) ~ CoCE~EI×E2); Lp(E;EI) x Lp(E;E2)=Lp(E;ElXE2). Proof.(a)
We first show that (fl,f2) ~ Co(E;E l) x Co(E;E 2) implies
[fl,f2] & Co(E;ElXE2). By (1.3.3) and (1.5.1) We have Ill,f2] (~) = (flxf2)(d(1~)) ~(flxf2)(~x ~) = fl(l~) x f2(B). Thus, if W.~ ~ E~ the filter on the right hand side and hence a fortiori [fl,f2] (~) is quasi-bounded, which shows that ~fl,f2] is quasi-bounded. From the equality
~[fl,f2] (a,x): (~fl(a,x), ~f2(a,x)) we deduce similarly that
A[fl,f2] (~,~) ~ ~fl(~,~) x Af2(~,~), and from this it is obvious that the equable continuity of fl and f2 implies that of Ill,f2] . We thus have shown that (fl,f2) l
(6.4.1~)
~Ffl,f2] induces a map ~:
Co(E;E1) x Co(E;E2)--~-~Co(E;ElXE2).
(b) Let, conversely, f ( Co(E;EIXE2). Then we conclude by (2.8.?) and (2.8.5) that ~iof, where TFi : ElXE 2 -.--~Ei is the projection, lies in Co(E;Ei), i = 1,2. Thus f ~('Ir~,f,
(6.4.15)
~F2.f ) yields a map
~: Co(E;EIXE2) ---~Co(E;EI) x Co(E;E2),
-80-.
and ~
is obviously the inverse of ~ . Both maps ~,~
are
linear, and so it only remains to show that they are continuous, or equivalently: (c)
continuous at the origin.
L e t ~ C o ( E ; E l) x Co(E;E2). Then ~ ~ ~l x ~2' where
~i~Co(E;Ei)"
If W . ~ E ,
further, since ( ~ r ) ( 1 5 )
we have therefore
&~(l~)
This shows that ~ C o ( E ; E l X E 2 )
~i(1~)$Ei, and
x ~ 2 ( ' 6 ) , also ( ~ ' ) ( ~ ) ~ E L x E 2.
and hence the map (6.4.14) is
continuous. (d)
Let ~Co(E;ElXE2).
We have to show that
# ~ICo(E;E1) x Co(E;E2) , or equivalently: w i ( ~ ) ~ C o ( E ; E i ) , where ~i' i = i~2, are the projection maps of the product
Co(E;E1) x Co(E;E2). Since Wi(g) =~i.g we get
which ends the proof of the First homeomorphism. The second result, concerning p-linear maps, is a corollary; one only has to remark that Ill,f21 is p-linear if and only if fI and f2 are p-linear and that Lp(E;Ei) has the structure oF subspace of Co(EX...xE;Ei).
(6.4.16)
Propositio.~. IF ~:E 1
~E 2 is linear and oontinuous,
then ~: Co(E;EI)----~Co(E;E2) defined by ~.(f) =~ -f also is linear and continuous.
-
B1
-
By (2.8.5) and (2.8.7) ~,f~ Co(E;E2). ~ is obviously linear. Let no. T~Co(E~EI) , W ~ E. Then ~ ( ~ ) ~ E I anO ~(~(15)) = (t',~')(~)~E2, hence ~o~= t . ( ~ ) ~ Co(£;E2) .
(6.4.17)
~orollary.
I f [z E1 ~
E2 is a linear homeomorphism,
then so is ~.: Co(E;EI)---~ Co(E;E2)
-
§ 5.
50
-
FUNDAMENTAL THEOREM OF CALCULUS. =
The value of a vector-valued function will be estimated by means of its derivative.
Since no norm is available, estimation
is formulated by means of convex sets, which is advantageous also in the normed case.
5.1.
Formulation and proof of the main theorem. Suppose there are two maps given: f:[~,~]
~ E and
~, E being a pseudo-topological vector space, such that the following conditions are satisfied
(51.1)
E ° separated;
(5.1.2)
f:
(5.l.3)
B a closed and convex subset of E°;
(5.14)
for almost all (*) t ~ ~ ]
[~,~@] ~
E ° and
~: ~o(,M
; ~ continuous;
, f and ~
are differen-
tiable at t and satisfy
f'(t) (5.l.5)
s ~t
..... ~ ( s ) ~
~ ~'(t).B; ~(t).
* Throughout this paper, "almost all" is never used in the sense of measure theory~ but always means: "all with at most a denumerable infinity of exceptions".
-
(5.i.6)
51
-
Theorem. Under the above hypothesis one has
f(~ ) - f(~)~ ( ~ ( ~ ) -~(~)).a. Proof.
Part i. In this part, we reduce the general case to
the following special case :
(5.i.7)
~=0
;
~ (0)
= 0 ;
f(o)
: o ;
o ¢
In order to do this, we choose a fixed point p ( B
B.
and we define
~I' ~I' ~i' ~i' Bl as follows:
ql(t) = q ( t + m )
-q(~)
~l(t) = f(t+~)
- f(~)
for t &[~l, @l] -ql(t).p
;
for t ~[~l,~l]
;
BI=B-p. One easily verifies that the validity of the conditions (5.1.1) to (5.1.5) for
e(, ~, 5o, f, B implies their validity for o(1, ~l' ~l'fl'B1"
In fact, let us check (5.1.4), the others being obvious. By the chain rule, we have: q'l(t) = w ' C & + ~ )
;
fi(t) = f ' ( t + ~ ) - q [ ( t ) . p
= f'(t+~)
Hence one concludes from f'(t) ~ ~'(t).B
fi(t)e
- W'(t+o().p.
for t G [o(,~]
~'(t+~,).B -W'(t+~,).p =M,~(t).B1
for t e
Suppose now the theorem holds for the special case.
:
[~i' l@l] "
-
Then we have f 1 ( ~ l ) ¢
52 -
~I(~I).BI .
But this, using the defini-
tions of ~ l ' ~ I ' f l ' Bl' yields immediately:
f(~) . f(01)
(~(~) -t~(e()).pE(~(~)-~(ot)).(B-p),
.
which is equivalent to the assertion of the theorem.
Having
reduced t h e g e n e r a l case t o t h e s p e c i a l
we hence-
forth
assume t h a t c o n d i t i o n s
P a r t 2.
We suppose i n t h i s
a neighborhood o f z e r o ,
(5.l.S)
f(/~)
£(4(~)
(5.1.?)
case ( 5 . 1 . 7 ) ,
hold.
pact of the proof that
and we show t h a t + ~'#
B is,
i n E°~
then one has f o r a l l
~ >0:
+~).e.
Let #'i' JP2' t3'''" be an enumeration of the points where possibly (5.1.4) does not hold (*), and let us define an auxiliary function : IR by
(5.l.g)
~(S)
=~(s)
* E.S
+ E.
Let
7n C- S
f(s) G
O b v i o u s l y one has •
0¢I.
(½)n. ~(s).B
Let ~ = sup I.
for 0 4=s ~ t ~
If 0 ~ t
z g,
there
exists t I ~ I with t At) ~ ~ ; hence f(s) G ~(s).B for 0 ~ $ ¢- t I, thus in particular for 0~s_~t,
which shows that t & I end that
f(t) •
~ (L).B. But this shows that also
fore:
I - G0,~]
~
I, and we have there-
.
* We can always take ~ and ~ among the points ~n; then we do not have to bothsr what differentiability would mean (cf.3.4).
in the endpoints ~ ,
-
53
-
We claim now that f ( ~ ) £ ~ ( ~ ) . B . I f ~= O, nothing has to be proved. If
~ > O , we use that
~
B
Since f:[~,~]-
for O , t
; E° is continuous and ~ continuous from the
left t it follows, using that B is closed in E°, that
t~ which means
(S.l.lO)
f(~ ) • Since I : [ 0 , ~ ]
~(~).B. ~ [0,~]
, we have of course
~--~.
(S.I.II)
In order to show that here equality holds, we proceed i n d i r e c t l y : we assume that
~ ~.
Then either ~÷~n for a l l n or ~ = ~m" We
show that none of these cases is possible. Case l: ~ ~ ,
V ~& ~n for all n.
This means that ~ is not an
exceptional point: f and ~ are differentiable at ~ and (5.1.4) holds for t = ~ . Thus we get f(~+h) = f(~ ) + h.f'(~) + rl(h) , ~(~+h)
=~(~)
+ h.~'(~)
f ' ( V ) ~ 4"(~).B.
+ r2(h),
where rl~
R( IR;E°);
where r 2 C
R(IR; IR);
- 54 -
Since II = I-l,1] certainly satisfies W.I I~IR we conclude: ~rl(W, [II])~E m. which means ~rl(\V,[ll]) @ ~ , ~.Tbeing the neighborhood filter of zero in E°. Since in this part 2 of the proof we assume that B ¢ ~ ,
we have also e/2.B ~ ].Fc @rl( W, [Ii]),
GI~O such that ~ r l ( I ~ ,
and hence there exists
I1) C t/2 B, and
thus (cf (3.I.i)):
l>l+'~l
for
~l(~ )G +/2. ~ .a
In the same way we conclude: there exists ~2
~2(~) ~
1,2(~,)1
~/2. A. z I
0 such that
for 1~1-~2,
-'- "12.1;~1
i.e.
for IM~'~2-
Let now ~= Sin ( ~I' ~ 2 ' ~ ' ~ )" Then we have for O ~ H
~6:
f(~+h) = f(~) + hof'(~) + rl(h) E ~(~).B + h. ~'(~).B + h. E/2.B -~(~).B Here,
the c o e f f i c i e n t s
+ ( ~ ( ~ +h)-~(~)-r2(h)).B
~(~), q(~ +h)- +(~)-r2(h)
= h.
+ h.
e/2.B.
~'(~)
and h. ~/2 are, using (5.1.5), ali non-negative. Using that for e convex set B and non-negative coefficients/u, Y ,G" one has
/~.8 +Y.B + ~'.B c(/~+~+G).B~we get therefore: f(~+h) E ( ~ ( ~ )
+~(~+h)-
= (~(~+h)
. +. ~
~(~)-r2(h)
(½)n+ +.+ _r2(h) + h ./2) .B
C (W( V +h) ++. ~.,+.>"'+ f,,,, <.p,,
= ~(~
+h).B.
. ho E/2).B
+.<8, +h)J.B
-
55
-
Thus one would have f(s) e ~(s).B for O&s ~
and ~6s 6 ~ +6 ,
hence ~ + ~ G I, which contradicts the definition of ~ as supremum of I.
This shows that case 1 is not possible.
Case 2: ~ z.~,~ = ~m" Since f is continuous at the point 8m, there exists
#l > 0 with
Analogously, since ~ is continuous and monotonic: there exists 2 > 0 with
~(~) -~(~)
•
1
~ y2.~
for ~ _ ~ ÷
Putting again ~ = Sin ( ~j, ~2' # ' y ) '
we have for
S2. ~ ¢. ~ ~ ~ + ~ :
f(~) = (f(~) - f(~)) + f(~) 1 E { . ~.B, (~(~) . ~ , ~ 7__ l
As b e f o r e i n case 1~ i t contradiction either. tion
would f o l l o w
with the definition
The consequence o f t h i s
of
from t h i s
~,,hat ~ + ~ E
I~ i n
~.
Hence case 2 i s not p o s s i b l e
indirect
argument i s t h a t t h e assump-
~ ~ # is impossible, and by (5.1.11) the equality ~ = #
must hold.
But then, (5.1.10) yields exactly (5.1.8), i.e. what we wanted to show in this part of the proof. Part 3.
We still suppose, as in part 2, that B is a neighborhood
of zero in E°~ and we show, that then the conclusion of the theorem holds, namely
(5.i.12)
f(~)
~ 4(~).B.
-
If ~ ( ~ )
~0,
56
-
this follows immediately, using (5.1.8) and the
fact that B is closed, by passing to the limit for ~----~0: =
lim ~0
f(~l
4(~) +~+~
~ B.
t>O So l e t
us c o n s i d e r the c a s e ~ ( ~ )
= O. S i n c e ~ ( O )
= O, t h e mono-
tonicity of ~ implies that then q(t) = 0 for all t ~ ~0,~]. Hypothesis (5.1.4) then yields: f'(t) = 0 for almost all t ~ [0,~] . Thus for any closed, convex neighborhood C of the origin in E°, the five points of the hypothesis of the theorem are satisfied (with B replaced by C), and according to Part 2 which is already proved one has f ( ~ )
£ (~+~).C,
or in particular
f-qL l c This being true for any closed, convex neighborhood of zero in E °, it follows from (5.1.1) that f( +l ~ ) = O, and since 0 • B, we have again (5.1.12). Part. 4
Having reduced in Part 1 the general case to the special
case (5.1.7), we still assume that ~5.1.7) holds. It only remains to get rid of the additional hypothesis made in Part 2 and Part 3, namely that B is a neighborhood of zero in E°. So let now B be any closed convex set containing zero. If ~ ( ~ ) = O, we already know that f ( ~ )
= 0 and thus certainly f ( ~ ) ¢ ~ ( ~ ) . B .
-
57
-
In order to show that this also holds in the c a s e ~ ( ~ )
@ O, we
assume (indirect proof) the contrary:
Since B is closed (always in E°), we can choose a neighborhood U of z with U ~ B
= ~.
By the continuity of the addition, since
o - o + z = z G U, we can choose a neighborhood V of 0 such that V - V + z ¢ U. The topology of E ° being locally convex, we can choose a V which is convex. Now,
(z + V ) ~ (B + V) = ~ . In fact, if this intersec-
tion were not empty, there would exist points Vl, v 2 ¢ V and b ~ B with z + v I = b + v2, thus b = v I - v 2 + z G V - V + z c U, which contradicts U m B
= ~.
Hence z + V is a neighborhood of z which is disjoint from B + V, and therefore z ~ @ + V = B*. The set B* is a closed, convex neighborhood of zero in E °, and since B* ~ B, the hypothesis of the theorem are a f o r t i o r i
satisfied for B* instead of B, and
according to Part 3 we conclude that f ( ~ ) ~ valently: assumption
~(~).B*,
or equi-
z ~ B*, in contradiction with z ~ B*. This implies that (5.1.13) is impossible, which concludes the proof of
the fundamental theorem.
•,
5.2.
(5.2.l)
58
-
Remarks and special cases. Proposition.
The fundamental theorem (5.1.6) is also
valid if the hypothesis that f: [~,~]-----~ E° is continuous and differentiable at almost all points t e E~,~] f: ~ , ~ ]
is replaced by the hypothesis that ~ E is continuous and differentiable at
almost all points. Proof. Since by (2.7.10) E ~ E°~ the identity map i: E ....~ E° is continuous,
thus (being linear) also differentiable.
Hence the
statement follows from the fact that the composite of continuous maps is continuous and the chain rule. We further remark that f'(t) is then the same element of ~ = E° whether we consider
f: [ ~ , ~ ] ~ E (5.2.2)
or f: [ ' , # ] ~ E
Corollary. Let ~:
°. 7 |R be continuous and satisfy
~,~]
for almost all t ¢ ~ , ~ ] is monotonic:
s4t ~
the inequality ~'(t) ~ O. Then ~(s) ~
~(t).
Proof. We put B : IR+ = {x ~ IR Ix ~ o} , and define an auxiliary function ~ :
[M,~]
~ IR by
can be applied with functions choosing any interval and closed and
Is,t] c
?4-is monotonic.
(t) - ~(s)
?+ (t) : t. Then the main theorem ~,~+ ~,~]
in the place of f, ~ , and , noting that B is convex
It yields:
~ (t-s).B,
which implies the result, since (t-s).B = B.
-
(s.2.3)
5g
-
Remark.For normed vector spaces E, the "mean value theorem" in its modern formulation of theorem
(cf.(8.5.1)
in ~ 3 ]) is a special case
(5.1.5). One simply chooses B to be the closed unit ball
of E. We have then E = E °, and (5.1.1) and (5.1.3) are satisfied. Hypothesis
(5.1.2) and (5.1.4) are maintained,
(5.1.4) reducing now to the inequality
the condition of
I I f ' ( t ) ~ ~°(t).
From (5.2.2)
follows,
that now also hypothesis
finally,
the conclusion of the theorem reduces also to an inequality:
IIf(~)
- f(~)ll~(~)
-~(~).
(5.1.5i becomes superfluous.
And
So we are exactly left with the
so-called mean value theorem for normed vector spaces. However,
theorem
spaces more interesting
(5.1.5) is also in the case of normed vector then the classical
formulation which uses
the norm (i.e. taking as B the unit ball), since it not only says that if for a motion of a point in E the velocity is not too big, then the point does not stray too far, but also that if the velocity is "big" in a certain sense, the point is much displaced.
However,
the condition that the velocity has to be "big" cannot be expressed in terms of a norm-inequality
of the form
ilf°(t)~
.... , since
this would involve a non-convex set. The velocity has to be "big" in the sense that f'(t) lies in a convex set B (which does not have to contain the origin),
or more generally in ~'(t).B.
-
60
-
As an example, let us consider a continuous function ~: [N,~]---~IR, differentiable at almost all points of [ ~ , ~ ] , and suppose that m ~
~'(t) ~ m for almost all t ( ~ , ~ ]
. We can
apply the theorem, taking as convex set B the interval [m,m] and as comparative function the identity function ~(t) = t, and we get immediately: m ( ~ - ~ )
~ ~(~)
-~(g)
~m(~-~).
~xamples in spaces of dimension greater than 1 are of course more interesting.
5°3.
(s.3.1)
Consequences of the fundamental theorem. ..P....rp.position. L e t ~ : E1
~ E2 b e c o n t i n u o u s
at all,
and
differentiable at almost all points of the segment S from 0
a to a + h, where a, h ~E1; let E 2 be separated;~¢L(E1;E2) ; o
B a convex subset of E 2 which is closed in E2. Then we have:
~'(X)for almost(h) - ~all(h)xEcBs I - ~ ( a + h ) Proof.
- ~(a) - ~(h) £ B.
We first consider the special case
~ = O. Let f: -~rO,1]-'-~E2
be the map defined by where g(t) = a + th. g: [0,i]-
~ E1 is differentiable and the chain rule gives
f'(t) = f'(t)(1) = (~'(g(t))og'(t))(1)
=~'(g(t))(g'(t))
=~'(g(t))(h).
-
61
-
Since g(t)~ S, we have by assumption: f'(t)~ B
for almost all t G~O,1] .
Hence the fundamental theorem implies
f(1) - f(O) G B, which is exactly what we want to show, since 4 (a+h) - ~(a) = f(1)-f(O). The general case ~ 0 introducing the map ~: ~(x) =4(x)
is reduced to the special case by
EI----~E 2 defined by
- ~(x).
Then
4'(x)(h) = 4 ' ( x ) ( h ) - e(h). By applying the result for the special case we have
V(a+h) - ~(a)~ B, which ts precisely what we wanted to show.
(s.3,2)
Remark. Proposition (5.3.1) is also valid under the following weaker hypothesis concerning continuity and differentiability of ~ : Assume that ~ : E1 .---~Eo2 is continuous at a11 and differentiable at almost a l l points x ( S . In fact, g being as before, it followe easily (cf (2.6.2)) that g: [0,I]
~El~ is differentiable and continuous. As before,
we apply the fundamental theorem to the map f which is the composite of the two maps
tO,l] 0----~El'----4--~ E~.
-
62
-
Since our approach to calculus is established by means of filters, the appropriate result for later applications is a filter-theoretic version of the fundamental theorem, respectively of proposition (5.3.1.). It will combine in a filter-inequality the main operators of calculus: differentiation,
the operator
~, the taking of closures and of convex hulls. In order to formulate it~ we first introduce and discuss some notations.
(s.3.3)
Definition.
If A ¢ ~
is any subset of a pseudo-topological
vector space E~ we denote by ~ its closure in E° (remember that E° is topological). gously denote by ~IX
~ ~},
~
If ~ is a filter on E, we analo-
or (~)-- the filter generated by
which is a filter basis, since XlnX2CX--lnX-- 2.
Let us denote by D(E1;E2) the vector space consisting of the differentiable functions f from E 1 into E 2 which satisfy f(O) = O. The I! @tl
differentiation operator
gives a mapping from D(E1;E2) into
the space of mappings from E1 into L(E1;E2). Therefore we can, for any f i l t e r ~ o n
D(E1;E2) , consider the filter~",
which is
It I I!
the image o f ~
under this map (or operator)
. Similarly,
~
is the image of ~ under the operator ~ introduced in (3.1ol); thus
@ ~ is a filter on the space of maps from
IR x E 1 into E 2.
-
(5.3.4)
63
-
Proposition. L e t ~ b e a filter on the space D(E1;E 2) definited a b o v e ; ~ a n y
filter on
R; ~ any filter on El;
~O,1] the unit interval of IR, respectively the filter generated by it; "@",
" e " and "- " the operators defined
in (3.1.1), (2.7.2) and (5.3.3). Then
(~'(II.~,~ .~) (~))°'.
e ~ (~,~) ~
Proof. Let m be any set belonging to the filter on the right hand side of this inequality. Then there exist VE ~ ,
F@ ~ and X ~
such that the set B = (F'(V.[O,1].X)(X)) °- is contained in m:
m~(F,(V./0,1] .x)(x))°': B~ F,(V.[0,i].X)(X). Hence
f'(~. k.x)(Ix~.8
for O~ ~
i, f~ F, I ~ V,x~X.
Since ~.B is closed in E ° and convex, we can now apply (5.3.1), which yields (since f (0) = 0 by assumption): f(A .x)~ ~
.B
for f ~
multiplying this by i/~ f(~,x)Q B
F, ~
V, x ~ X .
if ~ ~ 0, respectively using that 0~ B, we get
for f~ F, ~
V, x ~ X , or equivalently:
B ~ e F(V,X). This shows that B (thus afortiori m)belongs to the filter ~(~,~),
and the proof is complete.
We shall use (5.3.4) in two special cases, taking ss1~either one of
-
the filters
64
-
%V or Ill. In the first case we get~.
[0,I] .)~: IV.X.
In the second case we shall assume that [O,1] .X=9(and we will have:
~.
[ o , g . ~ = ~ and a ~ c[l],~ : ~ > .
We thus h a v e , ~ being as before, the following two corollaries of (5.3.4):
(5.3.5)
cor o!lary ,!-
(s.3.63
corollary 2.
If.~" ;Co,O . x , then
-
§ 7.
82
-
THE CLASS OF ADMISSIBLE VECTOR SPACES.
In order to obtain many of the deeper results of calculus in pseudo-topological vector spaces, some restrictions concerning the spaces are necessary.
We discuss these conditions
and call "admissible" those pseudo-topological vector spaces which satisfy them.
Since calculus makes use of various constructions
yielding new spaces from given ones (such as e.g. L(EI;E2) or go~(El;E2 ) from EI,E2) , it will be important to know whether these constructions yield admissible spaces if applied to admissible ones.
7.i.
We shall verify that this is the case.
The admissibilit~ conditions. We recall (cf.(5.3.3) and (2.7.I)) that for any filter
X on a pseudo-topological vector space E, we denoted by ~
or (~)--
the filter generated by the closures (in E°) of the sets of ~ , and by ~ or ( ~ ) ^ the filter generated by the convex hulls of the sets of ~ .
(7.1.1)
Definition.
A pseudo-topological vector space E is
called admissible iff it satisfies the following three conditions:
-
83
-
E° is separated;
(b) A
(c) From (2.7.2) and (2.7.3) it follows at once that we have the following
(7.1.2)
Lemma.
Condition (c) is equivalent to the condition
(c,) We further remark that condition (a) implies that E is separated, because by (2.7.10) ~ E
implies ~ x E°. Hence the admissible
spaces satisfy in particular the condition made at the beginning of 3.2. Not even topological vector spaces always satisfy condition (b). To see this, choose a separated topological vector A
space E for which ~26= [E] , ~ d e n o t i n g
the neighborhood filter
of zero in E (cf. e.g. [8 ] , p. 161). Then [0] ~E; but T ~ E is not true (note that closures have to be taken, according to the definition of ~ w
(7.1.3)
with respect to E°, so that ~ = E).
Proposition. Each locally-convex separated topological vector space E is admissible.
-
Proof.
-
We first remark that for topological
condition
vector spaces,
(c) is equivalent to the classical condition of
local convexity.
Hence E satisfies
E = E ° , also (a) is satisfied. E = E° it is well knowl zero in E satisfies
(7.1.4)
84
Finally
(b) holds, since for
that the neighborhood
~=
Corollary.
(c); and since (cf.(2.7.10))
filter ~ o f
~. If E is admissible,
then also E° is
admissible.
7.2. .......Admissibility ..
(7.2.1)
Lemma. Proof.
Let M E T
of E I.
If~V.~ : W.~
:~
. X
~.~
o Then M m V . X ,
Since scalar multiplication M ~ V.X ~ V
, then
: ~. where Vc~V and X ~
of E ° is continuous, ~
.
we get
. ~,
which shows, since ~V = ~V, that m¢~V. ~.
Suppose, conversely,
that m E ~V. ~ . Then M ~ I~.X, where by (2.5.2) X can be chosen such that X = ll.X. Since multiplication
by ~ is a homeomorphism
of E°, we get : ~I$.Y
~ ~.II.X
=
E. Il.X
= I6'X c~V.~'
= ~ ,
which shows that M ~ ~ . A
(?.2.2)
Lemma.
If
D/.~
= X,
then
W.l~
= ~ .
-
85
-
Proof. Using standard arguments concerning convexity, one first shows that ll.X = X implies ll.~ = X.
From this, the result follows
easily, again making use of (2.5.2).
(7.2.3)
Proposition. If E is admissible, then E ~ i s Proof.
also admissible.
We verify that each of the three admissibility conditions carries
over from E to E ~ . (a) From (2.6.3) we get (E~)o & Eo. Hence E° separated implies (ES)° separated. (b) For AcE, we denote by ~ the closure of A with respect to E° and by ~ ~
the closure with respect to (E~)o. for a filter ~
~
~
=~
on E.
Correspondingly
Let X i £ ~ . Hence there is ~
~
and
with
~E. Using (7.2.1) and the admissibility of E, we get:
w
This shows that
(7.2.4) and afortiori that
X
E#
(C) Let X~E4~ • As before: = ~V.
7.3.
(7.3.1)
, where
,
because
~
6
~"
]( & ~ = \V.A~ ~E, and by (7.2.2) we get:
E since E is admissible. Therefore we have
Admissibility of subspaces~ direct products and projective limits. Lemma.
Supose that the spaces Ely i £ I, are admissible;
that the maps fi : E
~E i , i ~ I
-
86
-
are linear and such that for each x ~ O, x ~ E , there exists
~cI with fk(x) ~ O; and that E has
the structure induced by these maps. Then also E is admissible. Proof.
We show that each of the admissibility conditions
carries over from the spaces E i to E. (a)
The maps fi : E
~E i being continuous and linear, we
know by (2.g.1) that the maps Eo
Eo
•
i
are also continuous. fk(x) = x k ~ O.
Let x~ E, x ~ O..Choose
~ G I such that
E °k being separated, we can in E°k choose a
neighborhood U k of x k with 0 ~ U k.
Then U = fkl(Uk ) is a
neighborhood of x in E°, and 0 ~ U
o This shows that E ° is
separated. (b)
Let ~ E .
Then fi ( ~ ) ~ Ei for all i ~I, and hence we have
fi-~~Ei
for all i ~ I.
But the continuity of fi : E° ~ and therefore fi(-~-) ~ f i - ~ '
E °i
implies that fi(~) c 7
closures being taken in E° rasp.
E~. We thus have afortiori: l
fi
Ei
which yields (cf.(2.3.2))
for all i ~ I, ~E.
-87..
(c)
Let
X~E. Then we have fi ( @ ) ~ Ei and
(fi(x)) ~Ei
for all i ~ I.
Now the linearity of fi : E ---~E i implies that fi(X)¢(fi(X)) ~ and therefore fi(~) ~(fi( X))~ and we have afortiori
fi( ) showing that
(7.3.2)
Ei
for all i
I,
~E.
Proposition.
Subspaces, direct products and projec-
tive limits of admissible spaces are also admissible. Proof.
One easily checks that in each of these cases, lemma
(7.3.1) can be applied.
7 .~B
(7.4.1)
Admissibility of B(E1;E2) , Co~l;E2), L ~ l ; E 2 ) . Proposition. If E2 is admissible, then B(E1;E 2) is also admissible.
Proof.
Again we show separately that each of the admissibility
conditions carries over from E 2 to B(E1;E2). (a) Let fo ~B(E1;E2)' f ~ O. Yo = fo(Xo ) ~ O. Yo ~ U.
Then we can choose x o~ E 1 with
In E o2 there exists a neighborhood U of 0 with
Let ~ be the evaluation at Xo, i.e. the map defined by
(f) = f(Xo). If ~ B ( E 1 ; E 2 ) , then ~ ( ~ )
= ~ ( X o ) ~ E2 since
W.x ° ~E 1 . ~ being
linear, we conclude by (2.8.7) that ~ : B(E1;E2)-----~E 2 is continuous, and hence (cf.(2.g.1))
-
(?,4.2)
~: B°(E1;E2 )
88
-
o
~ E2
is also continuous (*). Therefore V = ~'l(u) is in B°(E1;E2 ) a neighborhood of O.
Furthermore, fo ~ V, because fo & V would
yield Yo = fo(Xo ) = ~(fo ) ¢ ~(V) = ~(~'l(u)) contradicts Yo ~ U. (b)
c U, which
This proves that B°(E1;E2) is separated.
We first establish, for any filters ~ resp.~- on E1 resp.
B(E1;E2) , the following relation:
(?.4.3) which follows if we show that for subsets X resp. F of E1 resp. B(E1;E2) one has
(7.4.4) So let y ~ ( X ) .
Hence y = fo(Xo), where fo ~ ~
and x° ~ X.
Using again the continuity of the map (7.4.2), we have ~ ( ~ ) C W - - ~ . We thus obtain y = fo(Xo) =~(fo) ~ ~ ( ~ ) ¢
~-~
= F--~¢F--~,
which establishes (7.4.4). Assume now that ~ B ( E 1 ; E 2 ) .
Then we have, for any quasi-bounded
filter 15on E l , ~ ( ~ l ~ E 2. Since E2 is admissible, we conclude that ~ I E 2
and hence by (7.4.3) also ~(13):E~.
Rut this
J
shouJs that ~'~B(El;E2).
(*)
We remark that the evaluation e: B ( E 1 ; E 2 ) × E I ~ E not continuous in general (cf.(6.2.1)).
2 is
-
(c)
89
-
~ , ~ , F and X being Qs before, we first remark that
~(x) ~ (F(x)) ~, a relation concerning convex hulls which is easily checked. From that we obtain
(7.4.5)
s~ (~) ~ (,I'(~)) ~. Let now~'~B(EliE2).
Then we have, for any quasi-bounded filter
1~on El, ~'(~)~E2, (7.4.5) afortiori
from which we get ( ~ ( ~ ) ) ~ E 2 and thus by
~ ( ~ ) ~ E2.
This shows that ~B(E1;E2),
the proof of proposition (7.4.1) is complete.
(?,4.6)
Proposition.
If E2 is admissible, then also the
spaces Co(El;E2) , Co~(E1;E2), Lp(E1;E2), Lp~(E1;E2) are
admissible.
Proof. Co(El;E2) resp. Lp(EI;E2) are subspaces of B(EI;E 2) resp. B(EIX...xEI;E2); hence their admissibility results from (7o4.1) and (7.5.2). The rest follows from (?.2.5).
and
-
§ 8.
8.1.
90
-
PARTIAL DERIVATIVES AND DIFFERENTIABILITY.
Partial derivatives. Let f : ElXE 2 .__.~E3.
For any fixed (al,a 2) e ElXE 2,
we can consider the partial mappings Xl ~ x2 ,
~f(al,x2).
cf(xl,a2) and
If these mappings are differentiable at
a I resp. a2, f is called partially differentiable at (al,a2), the derivatives being denoted by Dlf(al,a2) rasp. D2f(al,a2).
(8.i.1)
Proposition.
If f: ElXE 2 ----~E3 is differentiable
at (al,a2), then f is partially differentiable at (al,a2),
(8.1.2)
Df(al,a 2) • (tl,t 2) = Dlf(al,a2).t I + D2f(al,a2).t2; and continuity of Df implies continuity of Dlf and D2f. proof,
xI ~ ~f(xl,a2) is the composite of the two mappings
x I ~----~(xl,a2) and (xl,a2) ~
:f(xl,a2).
The first is differen-
tiable by [4.2.1), (3.2.3) and (4.4.5), it's
derivative at any
point being i I : t I ~----~(tl,O). Hence we get by (3.3.1):
(8.1.3)
Dlf(al,a2) = Df(al,a2) o i 1.
Analogously
D2f(al,a 2) = Df(al,a2) o i2, where i2(t 2) = (O,t2). Since (tl,t2) = il(tl) + i2(t2) , we get (8.1.2). The continuity assertion for Dlf follows from (8.1.3), using that by (6.3.2) the mapping c : L(E1;Et×E2) x L(ElXE2;E3) Similarly for D2f.
~L(E1;E 3) is continuous.
-
8.2.
91
-
A sufficient condition for (total) differentiability. In the preceding paragraph the definition of admissible
vector spaces was given. They now enter essentially in the proof of the following theorem whose classical version is well known.
(a.2.1)
Theorem.
Let U be an ElXE2-neighborhood of (0,0)~ ElXE 2
and E3 an admissible vector space. Suppose that f: ElXE2----~E 3 is partially differentiable in (al,a2) + U and that D2f: ElXE 2 ----~L(E2;E3) is continuous at (al,a 2) (*). Then f is differentiable at (al,a2) and Df(al,a 2) • (tl,t 2) = Dlf(al,a 2) • tI ÷ D2f(al,a2) • t 2. Further continuity of Dlf and D2f implies continuity of Dr. Proof. (8.2.2)
f(al+Xl, a2+x2) - f(al,a2) = Dlf(al,a2~ • xI + O2f(al,a2) • x 2 + (r2+rlol]"l) • (Xl,X2),
where rl(Xl) = f(al, xl,a2) - f(al,a2) - Dlf(al,a2) . xI and r2(xl,x 2) = f(al+xl,a2+x 2) - f(al+xl,a2) - D2f(al,a 2) • x2. Since by assumption rI E R(EI;E3), we get rl o ~ 1 e R(ElXE2;E3) by (3.1.6).
(*) (1) For (Xl,X2)~(al,a2) (see also 3.4.1).
+ U, one can define D2f arbitrarily
(2) E and E ~have the same quasi bounded filters. Hence the following proof also shows that the theorem is true under the weaker hypothesis: U is an El~XE2~-neighborhood and Dtf: EI~XE2~--_-_~L(E2;E3 ) is continuous.
-
92
-
In order to show that r2 E R(ElXE2;E 3), let ~VI~ ~EIXE 2. Then ~ ~i x~2' where(for i = 1,2) I~i = I. ~i (~),
I the interval
[O,1] c ~. But ~V~ i = W.I. ~'i(~ ) =~'i(IV.1@)~ E i by continuity of the projections, hence by the continuity of D2f at (el,a2):
02f(a I + w
a2
Let ~ x l ( x 2 )
" 02( i'e2)
= r2(xl,x2).
Then
for
L(E21E3)"
(Xl,X2)£U, ~Xl'(X2)
=
D2f (a 1 + xl,a 2 + x 2) - D2f(al,a2). Further ~xl(O ) = O, hence by (5.3.5)
(~VI~ 1 ~V@2) • 1~2 )0- > l e
~lV~l
(IV,152)-
Observe now that (~
l~l~V 1~2) • 15 2 )0- = ((D2f(al+W~l,a24V~2)
by the admissibility at
- D2f(al,a2) ). ~2)°-~E3
assumption of E3 and the continuity of D2f
(al,a2).
Further
e ~Vl~l.(%V , ~2 ) ~ 8 r2(~J,~lXl@2) ~ -~ r2(|V,15) which proves that r2 E R(EIXE2;E3).
Hence by (3.1.4):
r2 + rlo~f 1 C R(ElXE2;E3) , and (8.2.2) shows that f is differentiable at (al,a2). The continuity assertion for Df follows from (6.3.2), because
Df(Xl,X 2) = Dlf(Xl,X 2) elTl * D2f(xl,x2 ) • 1" 2 .
-
93
-
§ g.
HIGHER DERIVATIVES.
9.1.
f" and the s~mmetry of f"(x). If u : ElXE 2
~ E 3 is bilinear and continuous, we already
know that u is differentiable at every point (4.2.3), but we are not able to prove that Du : ElXE2----~L(ElXE2;E 3) is continuous unless E1 and E 2 are equable vector spaces.
(9.l.l)
From now on we suppose that E, El, E2,... are always equable and admissible vector spaces.
(g.l.2)
Definition.
A map f: E1----~E 2 of equable and admissible
vector spaces is called twice differentiable at a point a, iff Df: E1-----~Lf(EI;E 2) exists in a E-neighborhood of a (cf. 3.4) and is differentiable at a. We write D2f(a) instead of D(Df)(a) and remark that thus D2f(a) ~ L~(E1;L~(E1;E2)).
The element which corresponds to
D2f(a) in the natural isomorphism
(6.4.2) is denoted by f"(a),
and we shall write f"(a).(s,t) instead of (f"(a))(s,t).
f"(a) is
a bilinear map : f"(a) E L2f(E1;E2).
(9.1.3)
Proposition.
If f:El-~-~E 2 is twice differentiable at a,
then f"(a), i.e. the bilinear mapping (s,t),-~f"(a).(s,t), is symmetric
(*).
n
* The same is true under the weaker condition that f: EI----~E ~_ is twice differenti~ble.
-
Proof.
Let (s,t) e ElxE 2,
g : [O,1]
of ao
-
~clR and consider the mapping
~ E2 defined by g (~)
Let I1 = [O,1] .
94
= f(a+~.s+
At) - f(a+ ~ . s ) .
By assumption Df exists in a El-neighborhood U
Hence a+\V(Il.S+t) ~ U, because
~V [Il.S+t ] ~ ~Vs+ ~Vt~ E1 .
In other words, there exists @l > 0 such that a + ~.s
+ ~.t £ U for
I ~ ~ ~l and
g is differentiable in [O,l 3 for each
~ ~ [O,q
, and therefore
I~I _z-i1.
By (3.3.1) and (4.3.2), we get g'(~)
= CDf(a+~s+
~t) - Dr(a+ ~ s ) ~ . ~ s .
o Inside any neighborhood of zero in E 2 we can choose a neighborhood W which is absolutely convex and closed; this means that -
=W.
-W=W=
We show that there is ~2 ~ 0 with
(9.1.4)
(Df(a+ ~ s +
e2 ~ ~i such that for
~t) - Of(a) - O 2 f ( a ) . ( ~ s +
and ( D f ( a + ~ s )
~t)).s
- Df(a) - D 2 f ( a ) . ( ~ s ) ) . s
I~I ~ e2"
( &W
e &W.
From the differentiability of Df at a we conclude also that eRDf(a)
. (\V, [Ii.s+t ~ )~ L(E1;E 2) hence, because
~RDf(a)
. (W,
[Il.S + & ] )
• s _~W]which yields (9.1.4).
Now we get by multiplying with ~ g'(~).l
~VS~El~
and subtracting
- ((D2f(a). ~t) o u Xs).l e 2 ~2.W where u &s : IR ----?E2
is defined by u ~s (~)
= ~ . ~s.
-
95
-
By (5.3.1) we get g ( 1 ) - g(O) - ( D 2 f ( a ) . g(l)
~3
~t).~
s ¢ 2A2.W .
- g(O) b e i n g symmetric i n s~t~ we can f i n d
~ ~2 ~ such that for each I~I ~
g(l) - g(0) -
s).
t
2
e3 > 0 w i t h
~3 : 2W;
hence by subtraction (D2f(a). ~t) . ~ s - (D2f(a). ~s). ~t ~ 4 ~2W for each
I~I ~-~3" Finally
(D2f(a).t).s
- (D2f(a).s).t
: f"(a).(t,s)
- f"(a)o(S,t)
~ 4 W.
This proves the symmetry, since E02 is separated. 9.2.
f(P) for p ~ 1. By induction on p we define p-times differentiable mappings
as follows : A map f: E1 ~
E2 of equable and admissible vector
spaces is called p-times differentiable at a, iff f(p-l): E1 _ _ ~ L~p-l(E1;E2) exists in a El-neighborhood of a at a.
and is differentiable
Then by definition D(f(P-l~(a) ~ L~(E1;Lp~I(E1;E2) ). By (6.4.12)
this vector space is linearly homeomorphic to L~(E1;E2). We write f(P)(a) for the p-linear mapping thus corresponding to Df (p-1)(a).
(9.2.l)
proposition.
If f : E1 _-~ E2 is p-times differentiable
at a point a, then for any fixed (s2,s3,...,Sp) , the deri~t~ e_.._~f (p_l) vative~JBf the mapping x (x).(s2,s3,...,ep) of E1 into E 2 is the linear map sI
~f(P}(a).(Sl,S2,...,Sp).
-
Proof. Let UCLn(EI;E2).
96
-
Then the mapping u ~
?u(tl,t2,...,tn) ,
where (tl,...,tn) is fixed, is linear and continuous by (6.2.2). The result follows from (3.3.1), (4.2.1) and the equality
(Df(P-1)(a).sl) . (s 2, .. . ,Sp) (9.2.2)
Proposition.
=
If f: E1
f(P)(a)
. (Sl,S2,...,Sp) (see(6.4.12)).
~E 2 is p-times differentiable at
a ~ El, then the p-linear mapping f(P)(a) is symmetric. This is a consequence of the two preceding propositions. The induction proof can be found e.g. in [3] p.177.
(9.2.3)
Proposition. If f: E l S E
2 is p-times differentiable and
f(P): E1----~L~(E1;E2) is q-times differentiable, then P f: E I ~ E Proof.
f(P): E1
2 is (p+q)-times differentiable. ~ L~(E1;E2) being q-times differentiable,
(f(p))(q-1): E1 _.._~Lq~_I(E1;L~(E1;E2)) by (6.412) also f(p+q-1): El_. ~
is differentiable. Hence
L p+q-1 ~ (El;E2) is differentiabie,
which ends the proof. We call a mapping f : E 1
~E 2 infinitely differentiable~
~ff it is p-times differentiable for all p ~ . (9.2.4)
Example. If u : ElXE 2
~E 3 is bilinear and continuousp
then u is infinitely differentiable. We already know (see 4.2.3) that Du exists throughout ElXE 2 and is linear.
- 97 -
To show that Du : ElXE 2 ~ L $ ( E l X E 2 ; E 3 ) is continuous, let ~l x ~2 ~ E3xE2' \V(J~lX~2 ) ~ ElXE 2. Then
Ou( ~j., ~2 )
.
(,,a.I, ~2) :
by (g.l.l))
(2.g.2),
u(~(z,,,~2) +
u(,A.z, ~.2) ~, E3
(2.8.8) and (2.3.7).
Hence by (4.2.1) Du : ElXE 2 ~ L m ( E l X E 2 ; E 3 )
is diffsrentiable, and
(Du)'(Xl,X2) = Du for each (Xl,X2) E EIXE 2.
Therefore
u" : ElXE 2 -----~L2~(ElXE2;E3) is a constant mapping and thus differentiable by (3.2.3). The same is true for u (k), k ~ 2 , which ends the proof. (g.2.s)
Proposition. If fl: E ----~El and f2: E ----~E2 are p-times differentiable at a, then the mapping Ill,f2] : E
7EIxE 2
is p-times differentiable at a, and [fl,f2](P)(a) = [fl(P)(a), f2(P)(a)l. Proof.
For p = 1 this is (4.4.5).
Suppose (g.2.5) for p and let
fl,f2 be (p+l)-times differentiabls at a.
We have to prove that
x ~-'-~[fl,f2~(P)(x) is differentiable at a. By the induction hypothesis we have [fl,f2] (P)(x) = fl (P)(x), f2(P)(x)], which is an element of L p (E;ElXE2). This space is linearly homeomorphic to Lp(E;E1)XLp(E;E2) by (6.4.13), the corresponding element being the pair (fl (p)(x)' f2 (p) (x)).
-
98
-
It is therefore sufficient to prove that x ~----~(fl(P)(x), f2(P)(x)) is differentiable, which is now a consequence of (4.4.5), because (fl(P)(x), f2(P)(x)) = ~fl (p), f2 (p)] (x) by (1.3.2), and ~fl (p), f2(P)j '(x) = [(fl(P))'(x),
(f2(P))'(x)] corresponds to
[fl(P+l)(x), f2(P+l)(x)] .
(9.2.6)
Theorem. If f : E1
~ E2 is p-times differentiable at a~
and g : E2---~E 3 is p-times differentiable at b = f(a), then g6f : EI _.--~E3 is p-times differentiable at a. Proof. For p = 1 this is the chain rule (3.3.1). Suppose (9.2.6) proved for p, and let f and g be (p+l)-times differentiable. By (9.2.3), gof is (p+l)-times differentiable, if we show that (gof)' is p-times differentiable. write (gof)' = c o ~f',g'°f]
By (1.3.2) and (3.3.1) we can
, where the bilinear map
c: L~(E1;E 2) x L~(E2;E3) ----~L~(E1;E3 ) is infinitely differentiable by (6.3.3), (gol.1) and (9.2.4).
The mapping g'of is
p-times differentiable by the induction hypothesis. Further by (g.2.5), the same is true for [f',g'.f] • Finally, applying the induction hypothesis completes the proof.
to the maps [f',g'of] and c, one
-
§ i0.
99
-
Ck-mAPPlNGS.
In § 2 the vector space Co(El;E2) was introduced, and in § 6 endowed
with a pseudo-topology. We recall the definition:
f~Co(E1;E 2) iff : W~,@~E
l :~f(~),
$ co(E l
Z 3 f ( ~ , X ) ~ E2;
iff :
~ V ~ E l . . ~ ~(~:L)~ E 2, We now introduce, always assuming that El, E2 are equable and admissible (cf. (9.l.l)), a class of mappings, called Ck-mappings; in the case of finite dimensional spaces, these are exactly the k-times continuously differentiable mappings.
i0.io
The vecLo r space Ck(EI;E2).
(lO.l.l)
Definition: For any k ~ o
(.) we call f: E1-----~E2 a Ck-map
and write f~Ck(E1;E2) , iff f is k-times differentiable in El and f(k)e Co(E1;L~(EI;E2)).
(io.i.2)
Propqsition. If E1 is finite dimensional and E2 normabla, then f~Ck(EI;E2)
if and only if f is k-times continuously
differentiable. This follows by (2.8.6), (2.6.1), (2.6.2), (2.5.2) and (2.8.3).
:
0,I,2,...
; for k : O, we define f(o) : f and
Lo~(EI;E 2) = E2# f o r k = i ,
f(1) = f, and LIQ(EI;E2 ) = L@(E1;E2).
-
(I0.I.3)
I00
-
Proposition. If f(Ck(EI;E2) , then f(n) ~ Co(EI;~(EIIE2) ) for n = O,...,k. Proof. The case k = 0 is trivial. k = I.
We prove in detail the case
We then know that f'e Co(EI;L$(E1;E2 )) and we have to show
5o l e t \VJ~E 1•
that fG Co(El;E2). and ~= [ O , l ] . ~
.
Let us put fo(X) = f(x) - f(o)
Since f ' = f ' is quasi-bounded and~.~ = W . ~ E I , o
~.f~(~)~L¥(EI;E2).
Hence, using (2.6.3)
(\V.f~(1~))(~)
= (W.fo)'(~).~E
2.
Since by (1.5.2) and (5.3.6) we have,
~.f(~)
~ W . f o ( ~ ) + w.f(o) ~ (W.fo)(~) + W.f(o)
~((Wfo),(~).~) °-
+ w.f(o)
we conclude, E2 being admissible, that W . f ( ~ ) ~ E2.
This shows
that f is quasi-bounded; i t remains to show that i t is equably continuous. Since
So let
W A, ~ ~E I.
~f(a,x) = f'(a).x + Rf(a).x,(1.5.2) yields ~f(~,~)
~ f,(~).~
+ Rf(~).~
.
Here, f'(w~).~ ~ E2 by (6.2.2), (g.l.l), (2.g.2) and (2.8.B), since
~.f'(~) ~ L(E1;E2).
The second term is estimated by means
of (5.3.6), which can be applied to the filter Rf(~) = ~ s i n c e (Rf(s))(o) = 0 for all a ( E1 .
We thus obtain, using that
(Rf(a))'(x) = ~f'(a,x) and that ~ ~ ~ = W ~
= [O,1] . ~
~ E1
-
i01
-
by the equability of E1 :
(Rf(~))(~) ~ (Rf(~)).~ ~(~f,(~,~)(~))o-. The equable continuity of f' and the admissibility of E2 implie that the right hand side converges to zero on E2.
This completes
the proof for k = i. If k > l~ we deduce from f(k)( Co(El,.LW[ k~EI;E2) ) by means of the linear homeomorphisme (6.4.12) and (6.4.17) that (f(k-1)),~Co(El;L,(El;Lk_~(E 1;E2))), and the same arguments used before yield f(k-l)~ Co(El;Lk~l(El;E2))"
Repeating this
k-times one completes the proof (see also (7.4.6) and (7o2.3)).
(lO.l.4)
.Cp,rollary, C_k+n(EI;E2)¢ C.Ck(EI;E2), For k > i, the proof is the same in view of the linear homao-
morphism Co(EI;L~(EI;E2)) ~w Co(EI;L'(E 1 ;Lk~l (El;E2))) where (f(k-l)), corresponds to f(k) and Lk~_I(EI;E2 ) is an admissible vector space.
For details see (6.4.12), (6.4.17), (7.4.6) and
(7.2.3).
10.2.
(lO.2.l)
The structure of Definition.
Ck(EllE2),
~Ck(El;E2)
iff :
[(n) ~ Co(EI~,,(EI~E2)) for n -- 0,I,2,.. ,~k.
-
102
-
We remark that by (I0.1.3) ~(n) is in fact a filter on the indicated space. Lemma. The inclusion (cf. (10.1.4))
(10.2.2)
Ck+n(E1;E 2) c Ck(E1;E 2) is continuous. This follows at once by (10.2.1).
(lo.2.3)
Lemma. The mapping f ~_.~f(n) of Ck+n(E1;E2) into Ck(E1;Ln~(E1;E2 ) is linear and continuous for each n ~ ~o. Proof. By (6.4.12) and (6.4.17), the corresponding element of f(n+k) E Co.(EI;Ln~k~(EI;E2 )) is (f(n))(k)~Co(E1;L~(E1;L~(E1;E2)))" hence f(n) ~ Ck(EI;Ln~(E1;E2)).
Let now ~'~CK+n(E~;E2).~
Then
by definition ~(P)~Co(E1;Lp@(E1;E2) ) for p = 0,1,...,k+n. By (6.4.17) and (6.4.12) we have
~o~ ~ ' ~ ~ p = n,n+l,...,n+k.
~Co(~~ o ~ ~" ~ ~ ~) ~o~
Hence (~(n))(P-n)~Co(E1;Lp~n(E1;Lg(E1;E2) ~
for p = n,n+l,...,n+k, which shows that~(n)~Ck(E1;Ln@(E1;E2)). This proves the continuity.
(lO.2.4)
Proposition.Ck(E1;E 2) is an admissible vector space. Proof. We have to veri~y that Ck(E1;E 2) satisfies the compatibility conditions (2.4.2) and the admissibility conditions (cf.(7.1.1)). By (9.1.1), (7.4.6) and (7.2.3), Co(E1;Ln~(EI;E2 )) is admissible for n
= O,...,k°
-
103
-
Observing that the mapping f ~----)f(n) is linear, one easily verifies the compatibility conditions for Ck(E1;E2).
We next
verify the three admissibility conditions. o a) The inclusion i : Ck(El;E2) 9 nuous by (10.2.~) and (2.9.1). Let f ~
-Co°(El;E2) is contiCk(E1;E2), f ~ O.
Then f E Co(El;E2) , and by (7.4.6) there is a convex neighborhood V of
rl
o 6 Co(E1;E 2) with f ~
V.
Hence by continuity
i-l(v).
i'l(v) is a convex neighborhood of o ~ Ck(E1;E2), and f Hence Ck(E1;E 2) is separated• b) Let ~ C k ( E 1 ; E 2 ) .
By (lO.2•l) and (7.4.6) we have
~---~) ~ Co(E1;Ln~(E1;E2)), n = O,...,k. map f ; ~(n)
~f(n) ((10.2.3)) implies that (~-)(n)z-~-~), because
C -F~ )
for each F ~qJ7 (see also (2.9•)~). Hence ~ C
c) Let ~ C k ( E 1 ; E 2 ) . f ~
(10.2.s)
But continuity of the
~ f(n)
k(E l~E2).
By the linearity of the mapping
it follows that (~) (n) = ~ )
Hence ~ C k (E1 ;E2 )
Remark• It seems that in general Ck(E1;E2) is not equable, not even if E1,E 2 are topological.
For this reason we later on
consider Ck(1;E2), ~E if the equability condition is required.
-
i04
-
Combining (6.1.8) and (10.1.2) we obtain the following generalization of
(6.1.g)
(10.2.6)
Proposition. If E1 is a finite dimensional vector space with its natural topology and E2 a normed vector space, then Ck(E1;E2) is the space of k-times continuously differentiable maps from E1 into E2 with the topology of uniform convergence on bounded sets of the derivatives of order O~l,...,k.
103
c ~ (ELSE2) The admissible vector spaces Ck(E1;E2) furnish a projective
system of pseudo-topological vector spaces (cf. (2.3.5)), where [I, ~ =
I~°,~.~
We define C ~
and the inclusion maps are continuous by (10.1.4).
(EI;E 2) to be the projective limit of this system.
Hence by (2.3.5) and (7.3.2)
(lO.3.1)
C~(E l~F2) : ~
C#E l~F2)
~" ~ C ~ ( E 1;E 2)
iff
k=o
(lO.3.2)
~ Ck(E1;E 2) for each k E ~ ° .
(lO.3.3)
Remark. It is readily verified that the propositions (lO.1.2), (10.1.3), (10.2.2) and (10.2.3) are also true for k = ~ .
-
(lO.3.4)
...proposition. .
105
-
If E1 is a finite dimensional vector space
with its natural topology and E2 a normed vector space, then C=~(EI;E2) is the topological projective limit of the topological vector spaces Ck(EI;E2). Proof.
In view of (I0.2.6), it is sufficient to show that the
pseudo-topological projective Iimit of any projective system of topologicaI vector spaces Ei, i £ I, is the same as the topological projective Iimit, which means that the coarsest pseudo-topology on E =
(~ iEI
Ei for which the incIusions fi : E
nuous, i s a topology ( * ) . we have
~ E i are conti-
By (2.3.3) and since @~E ~
fi(t[)~Ei,
:
ftCsup
= sup
Hence
sup
10.4.
Higher order chain rule.
~£,
(X))
= sup
which by (2.4o4) proves the assertion.
In order to prove the chain rule for Ck-mappings ~i0.4.7~,
we need some auxiliary results.
(lO.4.l)
Proposition. Let f: E1 ----~E 2 be C k.
Then :
f is Ck+ p ~_..~f(k) is Cp.
(*)
For inductive limits, the situation is different: for topological vector spaces, one has to distinguish between the topological inductive limit and the pseudo-topological inductive limit, the latter not necessarily being topological.
-
Proof. Let f ~ C k+p(E1,"E2).
106
-
Then f is (k+p)-times differentiable
and from f(k+p)~ Co(E1;Lkf+p(El;E2) ) we deduce by the linear homeomorphisms (6.4.12) and (6.4.17), that
(f(k)) (p) ~ Co(E1 ;Lp~(E 1 ;Lk~ (E1 ;E2) )) and that
f(k) : E1 ___.~Lk~(E1;E2) is p-times differentiable.
Conversely, if f(k) is Cp, then f is Ck+ p for the same reasons and by the assumption.
(lO.4.2)
proposition.
Let ~
be a filter on Ck+p+l(El;E2). Then:
~ Ck+p+l(E1;E 2) ~----~[~'~Cp(E1;E2)(*)
and
%
proof. Let • ~ Ck+p+l(E1;E2).
Then • $ Cp(E1;E2) by (10.2.2) and
~(p+l) ~Ck(E1;Lp+l~(E1;E2) ) by (10.2.3).
Conversely, ~
Cp(£1;£2)
means that ~(m) ~ Co(E1;L:(EI;E2)) for m = O,l,...,p and ~(p+l) ~ Ck(E1;Lp+14r(E1;E2)) implies ~(p+l+n)~ Co(E1; Lp .~,~(E.;E2)) ?~?ll
&
for n = O,1,...,k by (6.4.12) and (6.4.17), which ends the proof.
(10.4.3)
Propositio n. If ~ ~ L(E1;E2) , then ~, : Ck(E;E1) ---~Ck(E;E2) defined by Proof.
~.(f) = [of is linear and continuous.
For k = 0 this is (6.4.16).
k and let f ~ Ck+l(E;E1).
Suppose the assertion for
First we show that ~of ~ Ck+l(E;E2) .
By (3.3.1) we have (~of)'(x)
= ~of'(x).
The linear mapping
~. : L~(E;E1 ) ---~ L#(E;E 2) is continuous by (6.3°3)°
(*)
To be exact, one should write: i(~) ~ Cp(E1;E2) , where i : Ck+p+l(E1;E2)
~ Cp(E1;E2) is the inclusion.
-
i07
-
Further f'E Ck(E;L~(E;EI)) by (i0.2.3). Applying the induction hypothesis to ~. and f', we get (~of)' = ~.o f'~ Ck(E;L~(E;E2)), hence ( ( t . f ) ' ) ( k ) ~ Co(E;Lk~CE;L~(E;E2))~ and (eof)(k+l)~ Co(E;Lk~I(E;E2)) by the linear homeomorphisms (6.4.12) and (6.4.17). Let ~Ck+l(E;E1)o
Then ~.~is a f i l t e r on Ck+I(E;E2) by the prece-
ding result ~o~Co(E;E2) by (6.4.16) and (10.2.2)~ (Eo~)'
=
~. o ~ ' ~ Ck(E;L$(E;E2)), because ~'~Ck(E;L~(E;E2)) by
(10.2.3) and g :
~.og = (~.).(g) of Ck(E;L$(E;E1)) into
Ck(E;L~(E;E2) ) is continuous by the induction hypothesis. The induction proof is completed by (10.4.2).
C10.4.4)
Corollary. I f ~: El
~ E2 is a linear homeomorphism, then
so is e. " Ck(E;EI)--'-'~Ck(E;E2)"
(lo.4.s)
Proposition. The mapping ~: Ck(E;E1) x Ck(E;E2)-----~Ck(E;ElXE2) defined by ~(fl,f2) = [fl,f2~
(see (1.3.1)) is a linear homeomorphism.
Proof. [fl,f2~ is k-times differentiable by (g.2.5). By definition (lO.l.1): (f~k) f~k))~ Co(E;Lk~(E;EI) ) x Co(E;Lk$(E;E2)). By (6.4.13) thi~ space is linearly homeomorphic to Co(E;L~(E;E l) x L~(E;E2)), hence also to Co(E;Lke(E;EIXE2 )) by (10.4.3), the corresponding
element being Ell'f2](k).
-
10B
-
In order to prove the continuity of K , morphisms and (6.4.14).
one uses the same homeo-
The continuity of ~-l
follows from
(10.4.3), since ~ - l ( f ) = (~TlOf, ll-20f).
(i0.4.6)
Proposition. continuous,
If u : EIXE 2
~ E 3 is bilinear and
then u is Co..
Proof. By (2.8.10) and (g.l.1), u is Co . nitely differentiable,
by (2.8.7).
By (9.2.4), u is infi-
u' is linear and continuous, hence C o
u (k) is a constant map for k ~
2, hence obviously
Co .
(i0.4.7)
Theorem. If f ~ Ck(EI;E2) and g (Ck(E2;E3) , then gof ~ Ck(E1;E3). Proof. For k = 0 this is (2.8.5). let f and g be Ck+ 1. tiable.
(lO.4.B)
Suppose the theorem for k and
By (9.2.6), gof is (k~l)-times differen-
We assert that (gof)' ~ Ck(EI;L~(EI;E3)).
By (3.3.1) we get (gof), = c off' , g'6f], where c :
L$(EI;E2) x L¢(E2;E3)~L#(E1;E3) is C~. by (6.3.3) and (I0.4.6). g' ~ Ck(E2;L~(E2;E3)) and f' ~ Ck(EI;L~(EI;E2)) by (10.2.3).
f ( Ck(EI;E2) by (10.1.4).
Applying the induction hypothesis
to f and g', we get g'of ~ Ck(EI;L$(E2IE3) ).
Hence
~f', g'of] ~ Ck(E1;LC(E1;E2 ) x L~(E2;E3) ) by (10.4.5).
-
log
-
Applying once again the induction hypothesis to c and [f',g,ef] ,
we
get (I0.4.8).
From this we conclude
((gof),)(k) ¢ Co(E1;Lk#(EI;L$(EI;E2))~, hence
(g.?)(k+l)
Co(EI;Lk+I~(EI;E2)) by the linear homeomorphisms
(6.4.12) and (6.4.17).
One could establish a formula expressing
(g.f)(k) by means of the derivatives of f and g; the formula is the same as in the classical theory.
-
§ II.
llO
-
THE COMPOSITION OF Ck-mAPPINGS.
It will be shown that the composition map c : CJ(E1;E 2) x C~(E2;E 3) ......... > C~.~(E1;E3) is not only continuous (*), but even C ~ o We thus get a non trivial example of a
C_
_mapping between spaces which in general are infinite
dimensional and not topological.
Since the notion of a Ck-mapping
was only defined for maps between equable spaces, we have to consider the spaces C~(Ei,Ej)
(cf.(i0.2.5) and (2.6.4)).
They coincide with the spaces C ~ (Ei;E j) if e.g. the spaces E1,E2,E 3 are finite dimensional (cf.(lO.3.4)).
II.I.
The continuity of the composition map.
(ll.l.l)
proposition.
Let uGL(EI,E2;E3).
mapping ~ : Ck(E;EI) x Ck(E;E 2)
Then the bilinear ~ Ck(E;E3) defined
by ~(f,g) = u .~,g] is equably continuous. Proof.
By (10.4.5), (10.4.6) and (10.4.7), u~(f,g) e Ck(E;E3).
To prove the continuity of "~, u we show
(1) ~Ck(E~EI), W~ ~Ck(E~E2) ~
~(~,~) ~Ck(E~E3).
(2) W~Ck(E~E1), ~ C (E~E)
0(~,
(*)
Ck(E;E3).
This continuity statement is not eas~y comparable with the continuity result of (ll.l.~); however the latter is used in order to obtain the differentiability of c.
-
III
-
which will end the proof by (2.8.8). Let k = O, IVJ~E. Then by ( 1 . 5 . 2 ) and ( 2 . 8 . 8 ) we have in both cases (~(~,~))(J~) -~ u (~'(~), ~ (J~))~E 3 because either
~(~)~EI,\V~(~)~E 2 or W~(~)~EI,~(~)~E 2 Suppose (ll.l.l) for k and let us consider the first case:
~Ck+I(E~E1), Then by ( 1 0 . 2 . 2 )
~v ~Ck+I(E~E2)
and the p r o o f f o r k = 0 i t
0(~, ~)~Co(E;E3). (ll.1.2)
follows
that
We assert that
@ ( ~, ~ ))' ~ Ck(E;L~(E;E3)) By (6.4.11) and (g.l.l) we have the linear homeomorphisms
(ll.l.3)
L$(E1,E2;E3) ~ L$(EIIL~(E2;E3))" L$(E1,E2;E3) ~
L$(E2;L~(E1;E3)).
We denote by uI resp. u2 the linear mappings thus corresponding to u.
~
This means U(Xl,X2) = Ul(Xl).X 2 = u2(x2).x 1
o~oo~
u~
~o ~
@(~ ~))
o ~.hu~
+ 0.~
u~ ~]
where b i s the c o m p o s i t i o n map o f l i n e a r
and c o n t i n u o u s mappings,
i.e.
which i s b i l i n e a r
b(u,v)
= you, d i s c u s s e d i n
(6.3.3),
equably continuous. By (1.5.2) the above equality yields
( ~ ( ~ , ~ ) ) , ~_~(~,,UlO~) + ~(~ ,u2. ~) where IVy' = (IVy)' ~Ck(E;L~(EIE2)) by (10.2.3)
and
-
112
and UlOqJr= (Ul).(~-)$Ck(E;L~(E2;E3))by (i0.2o2),(ii.I.3) and (10.4.3). For the same reasons ~' ~Ck(EIL~(E;EI)) and yd(u24~) = u2°\V~Ck(E;L~(EI;E3)). By (6.3.~,(2.8.8) and the induction hypothesis, applied to the bilinear mapping b, it follows that b(~lJUl°~) ~Ck(E;L~(E;E3 )), and ~(~',u2o ~) ~Ck(E;L~(E;E 3)). Hence the above inequality yields (10.4.9). Applying (10o4.2) for p s 0 this proves that ~ ( ~ , ~ ) ~Ck+l(E;E3). If W~Ck+I(E;E1),~Ck+I(E;E2),
then exactly the same
proof shows that "U(~-,q) ~Ck.I(E;E3). (11.l.4)
Theorem. The mapping c: Ck(E1;E2) x Ck(E2;E3)--~Ck(E1;E 3) defined by c(f,g) = gof is continuous. Proof. We use induction on k. Observe first that
(11.1.5)
÷
(l)
k
=
o.
Let ~ x ~ C o ( E I ~ E
2) x Co(E2;E3),Wd~E1. By (1.5.2),
the above equality yields
By assumption g is Co, hence ~g(fCd~),~'(~z[)) ~E3, because ~Vf(~)~E 2
end
~ ( ~ ) ~ E 2. Further ~V(f(d~) + ~7(d~)) -~-
\Vf(J~) + ~V~(~) by (1.5.2), hence also ~ ( f ( ~ ) + ~ ( ~ ) ) ~ E 3 , which ends the proof for k = O.
-
(2)
113
-
Suppose the theorem for k, snd let
x Ck÷l(El,E2) x
Ck+I(E2;E3).
We assert that
(n.l.6)
(z~c((f,g),(~ ,~ ))),j,Ck(EI,d(EI~E3)). We again denote by b : L~(E1;E2) x L$(E2;E3) ~ the composition map discussed in (6.3.3).
L¢(E1;E 3)
Using (3.3.1) and
(ll.l.1) one gets
(n.1.7) where ~ is the map associated to b according to (ll.l.1) and c in the right hand side expression is the composition map: Ck+l(El;E2) x Ck(E2;L~(E2;E3))----gCk(E1;L~(E2;E3)). Applying the induction hypothesis to this map and using (i0.2.2)
and (I0.2.3), one gets
(n.1.8)
Ac((f,g'),(~, ~')) ~ Ck(EI;L' (E2;E 3))o ~Vf~Ck+l(E1;E 2) by the compatibility condition (4) of (2.4.2), hence (Wf)' = ~Vf'$Ck(EI;L*(EI;E2)) by (10.2.3). ~ being equably continuous by ( l l . l . 1 ) , this proves that
(n.l.9) Using the equality ( g ' + ~ ' ) , ( f + ~ ) = A c ( ( f , g ' ) , ( ~ , ~ f ' ) ) + g'of and (1.5.2) one gets
~v.(g,+ ~,)o(f+~) ~
-
114
-
by the compatibility conditions of (2.4.2), by (10.4.7) and
(ll.l.B). ~' ~ Ck(EI;L~(EIIE2)) by (10.2.3), hence also
(11.1.1o) The equality (ii.I.7) yields
by (11.1.9) and (ii.i.10), which proves the assertion (ii.I.6). The proof for k = 0 also shows that
~ c((f,g),(~,(] ))$ Co(E~E3), hence applying (10.4.2) for p -- 0, one completes the induction proof.
11.2.
The differentiability of th e composition map. In order to prove the general theorem, we establish some
auxiliary results.
(ii.2.1)
Proposition. Let g~Ck+l(E2;E3). g.: Ck(E1;E 2)
Then the mapping
~Ck(E1;E 3) defined by g.(f) = gof
is differentimble throughout Ck(E1;E2) , and
(g.)'(f).~
: ~(g'of,~).
-
Proof.
115
-
Consider the expression rg(~) = g.(f+~) - g . f -
where e : L~(E2;E 3) x E2 ~ in (6.2.3).
song'.f, ~ ,
E3 is the evaluation map discussed
According to (ll.l.1) we use the notation eo [g',f,~] = ~(g'of,~).
By (ll.l.1) the linear mapping ~ ~ ( g ' o f , ~ )
is continuous,
g,of being an element of Ck(EI;L~(E2;E3)) by (10.4.7) and (10.2.3). The proof is complete if we show that
(ll.2.2)
N
~ R(Ck(EI;E2); CM(EI;E3))"
We use induction on k. (A) Let k = O, ~VJ%~Co(E1;E2) and \VIS~E 1.
Then by (5.3.5),
(g.l.l), (1.5.2) and using the equality
(11.2.3)
@rg g, (f(x)).
Hp
because (Rg(f(x)))'(h) = A g,(f(x),h) and (Rg(f(x))~(O) = 0 for each x,h 6 E1 • This p r o v e s that Org(~V,~) ~Co(EI;E3). (B) Suppose (ll.2.1) for k and let g ~ Ck+2(E2;E3), \VJ~Ck+I(E1;E2).
-
116
-
We a s s e r t :
(ll.2.~)
(Org(~/, J~)) ' ~ Ck(EI;L~(EI ;E3)), which will end the proof by (A) end (10.4.2)o Using (11.2.3) and the chain rule one finds that the derivative of the mapping x ~ e r g ( ~ , linear map
•
~). x at a point x E E 1 is the
-~ ((~,.~~,,~)~x~ -
(o,.~)~x~ ~,~x~.~
- ~',(~x)). (~, ~).~, 4 ~xO. By
(g.l.3) one g e t s :
~,,(~x)).(~,(x).o,4 ~x)) =
~(D2g of, ~)(X)of' (x)~ .s, hence
(e,g(r).( ~, 4))' (x) =
(g,o(f+~) - g,of. ~(o2g,r,~))(x)of,(x)
+ (g'-(f+~) -g'-f)(x)o~'(x),
which yields
where (g').(f) = g'of end (11.2.5)
rg,(V) = Q'o(f+v) - g'of - ~((g')'of,~). By (I.5.2) we obtain therefore
(ll.2.6) Since g' ~Ck+ I(E2;L~(E2;E3)), the induction hypothesis applied to (g'), yields : erg,(W,v~)~ Ck(EI;L~(E2;E3))I further by (3.2.5) and (2.6.3):
-
117
-
Applying (Ii.I°i) to the bilinear map b and using (2.8.8) one proves (11.2.4) by the inequality (11.2.5).
(11.2.7)
Proposition. The mapping c: Ck(E1;E 2) x Ck+I(E2;E3)-----~ Ck(E~;E3) is differentiable throughout Ck(E1;E2) and c'(f,g).(~,~ ) : ~(g'o?,~) +~of Proof.
We verify the assumptions of (8.2.1).DlC(f,g).~ = ~(g'°f,~)
by (11.2.1). D2c(f,g ).~ = ~of, because c is linear in the second variqble and continuous by (ll.l.4). Ck(E1;E 3) is an admissible vector space by (10.2.4). DlC : Ck(E1;E2) x Ck+l(E2iE3) tinuous, because W~Ck(E1;E2)
The mapping
,L(Ck(E1;E2);Ck(E1;E3)) , is conand ~'x~ ~Ck(E1;E2) x Ck+l(E2;E3)
imply (DlC(f+~-,g+~) - Dlc(f,g))(~ ) = ~ c ( ( f , g ' ) , ~ x
~'),~
Ck(E1;E3) by (ll.l.1) and (2.8.8). (11.2.8)
Lemma.
If E2 is an admissible but not necessarily
equable vector space, f : EI___.-~E 2 is differentiable and f' :El~-----*L(EI~ ;E2~ ) is continuous, then f: Ele Proof.
~E2* is differentiable.
Let a ¢ E1 . By assumption f(a+x) = f(a) + ~(x) + r(x) where
¢ L(E1;E 2) and r E R(Ei;E2). Since by (2.g.1) we have L(EI;E 2) c L(Ela;E2~), it only remains to show that r~R(El~;E2~).
-
118
-
So let ~V15~ EI~. Since r (0) = 0 and r'(x) = Af'(a,x), we deduce °-
convo oo
zero on E2~ , because by the continuity of f' we have 4f'(a, ~V ~) ~L(El~;E2 m) and because E2~is admissible (cf.(7.2.3)).
(ll.2.g)
Lemma.
If ~VJ~Ck(EI;E 2) and ~Ck(E2;E3),
then
° ~ i Ck(EI~E3)" Proof. Let k = 0, WI~ ~E 1. Then (~oJ[)(~) = ~ ( ~ ( ~ ) ) ~ E 3 , because \V(J%(~)) = (~VJ%)(I~)~ E2. let ~V~Ck+I(EI;E2),
~ ~Ck+I(E2;E3).
Suppose the lemma for k and We have (g6f)' = ~(f',g'of),
hence by (1.5.2) (~oJ~)' _~ b (J[', ~'o~) which converges to zero on Ck(E1;L$(E1;E3) ) by (10.2.3), (ll.l.1), (2.B.8) and the induction hypothesis.
The proof for k = 0 also shows that ~oJ[~Co(E1;E3),
hence the result by (10.4.2).
Since ~ equable implies ~oJ[ equable,
we have :
(il.2.10)
Corollary.
If \V~CMN(E1;E2) and ~ ~ Ck~(E2;E3), then
~o~cM~(El~E3). (ll.2.1Z)
Lemma.
If ~Ck+I(E2;E3)
and ~V~l, IVJ~2~Ck(E1;E2 )'
then @R~.(Jll).( ~],~2 ) ~ Ck(E1;E3). Proof.
The equality Zl(g').(f, t~) = g'o(f+~) - g'of yields
~ ( ~')*(J~l' ~V~2) -~- ~'°(J~l+\V~t2) " ~'~ J~l which converges to zero on Ck(E1;L~(E2;E3)) by the preceding lemma and (10.2.3).
-
From
119
the equality ( c f . ( l l . 2 . 1 ) )
one deduces:
-
d(g.)'(f, ql )
A(~,)'(dtl,~VA2).~ 2
:
"~2
= ~<~(g')*(f'91 )'~2 )
~(~(~').(~l,~/A2),d~2),
and this converges to zero on Ck(£i;£3) by the preceding statement, (ll.l.1) and (2.8.8). We further have R g.(f).O = 0 and
(Rg.(f))'(~) hence by (5.5.5)
= Z~(g.),(f,w),
eR (~.(~1).( \V, J l 2 ) - ~ ( A ( q . ) ' ( d ~ 1, \V~2).d~2)O"
which converges to zero on Ck(E1;E3) since Ck(E1;[3) is admissible by
(10.2.4).
Lemma. Ir ~v~ ~ Ck+l(E2~E 3) and ~ , ~V~ ~ Ck"(q,E2),
(n.2.1z)
then ~ ~ . ( a ) . ~ ProoF.
One easily verifies the following equalities:
(1) Rg.(f).(A~)
= @R(A~).(f).(A,~I)
(2) A @ R g . ( f ) . ( ~ , ~ ) If ~
$Ck~(q,~).
= @R(Ag).(f).(~,~).
Ck~(EI;E2) , then there is
~1 ~Ck*(£1;£2)"
By
(1), (1.5.2)
R~.(d~) ( ~ ) ~ R ~ . ( ~ ) . ( l u l l )
~i with ~ ~ I
=~i and
and (ll.2.10) we therefore have
~ eR( XV~ ) . ( J t ) . (~V,~'~l) ~ Ck(E1;E3).
By (2) one gets @R(~V ~ ) . ( ~ ) . ( ~ V , ~ : l )
= I V ( @ R ~ . ( ~ ) . ( W , ~ l ) ) which is an equable
Filter and thus converges to zero even on Ck~(EI;E3).
-
(11.2.1~)
Lemma. Let c2=
120
-
Ck~(Ei,E2)
;L(Ck$+p+I(E2;E3);Ck*(Ez;E3))
be the mapping defined by c2(f) = f*, =here f*(z~) =~=f for any ~ ~ Ck+p+I(E2;E3). Then c2 is Co.
P~oof. Then
(a) Let \VJ[~Ck~(EI;E2), \ V ~ C* k+p+l(E2;E3)" Wc2(J~).C~ = ~/L~J~Ck~(EI;E3) by (11.2.10), which shows
that c 2 is quasi-bounded.
(b) Let
~VJ:I.,'~~CkW(E1;E2)
'
W(~ ,,~ Ck~+ p + l (E1;E 3)
~ "E £ -~k+p+l t 2'E 3) we have by (11.2.1),
"
For any
(10.1.4) and (2.6.2)
:
(11.2.14) Hence by (1.5.2)
~c2(~,~)
~_L~(~,o~,~)
+ R~.(~) ~
=here ~(~'o~,~)
~C~(Ez~E3) by (6.2.3), (11.1.1), (2.e.e), (10.2.33 and (11.2.10). R~,(~).~r~Ck4r(EI;E3) by the preceding lemma, hence C2 is equably continuous. (a) and (b) prove the lemma.
(11o2.15)
Lemme. For any f , ~ ICk(EI;E2) r f ( ~ ) : Ck+p+2(E2;E3)_..__>Ck(EI;E3) denotes the linear map defined by
(11.2.16)
rf(~).~ = R~.(f).V
=~o(?+~)-~of-~(~'.f,~).
Then we have: If ~V~, W~-~Ck~(E1;E 2) and
~v~ ck+p+2(E2; ~ E3), then ~r~. ~v,~ ) . ~ ~ck'(Ei,E 3)
-
Proof.
121
-
One easily verifies the following equalities:
(1) ~ ( ~ ' ) . ( f , ~ )
= ~c2(f,~).~, ~
(2) ~(~.),(f,~) = ~R ~.(f)),(~)~ (3) ~(~(~')*(f'~l)'~2) = ~(~*)'(f'~l)'~2" By (10.2.3) we get ( ~V~)' - W ~'~ C'I'K+p+±'(E~;L~(E2;E3))" Hence by (11.2.13) and (I) :
~(~').((~}", ~V~)I CJ(EI;L~(E2;E3))and by (3), (6.2.3), (ii.I.'I), (2.8.8) and (11.2.1): ~(Z3(~').(~, ~V~),~) = Z3(~.)'C~F, ~V~).#~Ck~CE1.E3).
Using (2), (5.3.5), (i0.2.4), (7.2.3), (ii.2.15) and the above result we get
_~(~(~.),(~,~).~)o-''~Ck#tE1;E3), hence the assertion of the lemma.
(n.2.1~)
Lemma.
The mapping (cf, (ii.2.13))
c~, Ck~ (EI~E2)
c~ E3)~C~(EI~E3) ) ~L~k+p+2(E2;
is differentiable. Proof. By (11.2.14) we have (c2(f+q)-c2(f)). ~ = ~(~,f,~)+R@.(f).~ . For any f~Ck(E1;E2)
~f : ek~(El~E2)
' L(c~k÷p÷2(E2 ~E3)~Ck'(EI~E3)}
denotes the mapping caracterized by
~f(~).~ = ~(~,of,~).
-
122
-
Similarly we had defined rf (cf.(II.2.16)) by rf(~).1# = R~.(f).u~. Obviously ~f is linear~ it remains to show that ~f is continuous and that rf is a remainder. Let ~r~ Ck~(E1;E2), ~V~Ck~+p+2(E2;E3).
Then
tf~.~ = ~(~,of,~;) ~Ck~(El~E3) by (6.2.3), (ll.~.l), (~.~.8) and (11.2.10).
Obviously %Vf~Ck~(El;E2) , hence by (11.2.15):
@rf(W,~t).% ~ CkW(E 1;E3) , if %V~t~Ck'(El;E2),which completes the proof.
(n.2.1a)
Lemma.
The mapping
c2: Ck~(E1;E 2)
~L w (Ck+p+2(E2;E3 ~ ) ;Ck~(E1;E3))is Co •
Proof. (a) Let \VJ[~ Ck'(E 1;E2).
If
W~ ~C:+p+2(E2;E3), then
~Voc2(A). ~ = \V~ od{~Ck~(E1;E3) by (ll.2.10). Since furthermore \V.c2(w&) is an equable filter, it follows that c 2 is quasi-bounded. (b) By the differentiability of c 2 we have
~c2(f, q) = c2(f+q) - c2(f) = c2'(f), q + Rc2(f). ~ Let ~V#[,~/~Ck*(E1;E2). Then by (1.5.2) (11.2.19)
Ac2(~,~)
& c2'(~).~"
+ Rc2(4&).q.
~/c2'(~) ~L~k~(EI;E2);L(C:+p+2(E2;E3);Ck~(E1;E3))~, because Wo13i ~Ck~(E1;E2),~VI~2~C:+P+2(E2;E3) implies ~Vc2'(~t).1~l).15 2 = ~ (I~2'oJ~ , W1~l) ~Ck~(El;E3 ) by (6.2.3),
(n.l.i), (2.s.s) and (n.2.10).
-
Since we can suppose ~ - =
The equality Rc2(f).(~) Rc2(J~).~ This f i l t e r
123
~ p we get
= ~. eRc2(f). ( ~ , ~ )
yields by (1.5.2)
= R c 2 ( J % ) . ( ~ : ) ~ W. ~Rc2(J~). ( ~ , ~ ) .
being equable, i t
remains to show that f o r
W~Ck+p+2" (E2;E3) one has ( e R c 2 ( J ~ ) . ( W , ~ ) ) . ~ C k ~ ( E 1 ; E 3 ) . But t h i s follows at once from (11.2.15), (Rc2(f).~).~ (cf.(ll.2.17)). inequality
(n.2.20)
= rf(~).
because
~J(
Now the equable c o n t i n u i t y of c 2 follows from the
(11.2.19).
Lemma. The mapping c2: Ck~(E1;E2)
Proof.
,L"(Ck~+p+2(E2;E3); Ckat(E1;E3)) is Cp.
For p = 0 this is (ll.2.18).
Suppose the assertion for p
and consider the mapping
We make use of the f o l l o w i n g abbreviations=
E4 = Ck~(E1;E2);
E7 = Ck+~+2(E2;L$(E2;E3));
E5 = Ck+p+3 ~ (E2;E3);
E8 = Ck~(E1;L~(E2;E3)).
E6 = Ck$(EI;E3);
-
124
-
We assert that the mapping (A)
c2' : E4-----~Le(E4;Le(E5;E6))
is Cp.
Observe that this mapping exists by (11.2.17), (2.g.1) and (5.4.10).
The assertion follows from (6.4.11) end (10.4.4),
if we show: (8)
The mapping c21 : E 4 .... ~ L~F(E4, ES; E6)
defined by c21(f).(L~,~ ) = (c2'(f). ~ ) . ~ We denote by ~l : E4 x E 5 the projection (~#,1~).
is Cp.
~ E4 resp. 3T2 : E4 x E 5 ~ ,~
E5
resp. ( ~ , ~ ) - - - - - ~ z ~ ,
Then we have c21(f) = ~* ~c 2(f)ooolT2,~l ~ D : E5 ~
= ~(z~'af,~)
where
E 7 is linear end continuous by (10.2.3),
c 2 : E 4 ----~ Le(E7;E8 ) is Cp by our induction hypothesis and : E8 x E4 -.---~E6 is Cp by (ll.l.1) and (10.4.6). Applying (ll.l.1) to ~
which is an element of L(EB, E4; E6) ,
we conclude that e : L~(E4xE5;E4) x L~(E4xE5;E 4) ----~L~(E4xE5;E 6) is bilinear and continuous by (2.3.6). One has of course to verify that for any k ~
L~r(E1;E2) has the structure induced by its
inclusion in Ck~(EI;E2).
-
Let ~'l = E4
125
-
> L#(E4xE5;E 8) respo 4 2 : E4-----~L~(E4xE5;E 4)
be the mappings defined by ~ z ( f ) = c2(f),Do iF2 resp. q~2(f) = ITl • T2 is • constant map, hencs obviously Cp. ~ l = (D= 3F2) * oc2 is the composite of Cp mappings hence also
Cp by (10.4.7).
We have therefore = c21 = ~e o [q~l ' • 2 ~ P and
t h i s i s the composite of Cp-mappings by (10.4.5), which proves (B) and hence (A). By (ii.2.17) we know that c 2 : E4
~L(£5;£6) is diffsrentiable.
By (A), (2.6.3) and (2.8.3), c 2' : £4----~L(E4;Lff(E5;E6) ~ is continuous.
L(E5;E6) is an admissible vector space by (7.4.6).
Hence from (11.2.8) it follows that c 2 : £4----~L~(£5;£ 6) is differentiable.
One deduces from (A) that c 2 is C 1 and c 2' is Cp
and thus by (10.4.1) c 2 is Cp+ I which ends our induction proof of (11.2.20). (11.2.21)
Theorem.
The mapping
c: Ck~(E1;E2) x Ck+p+l ~ (E2;E3) ----v Ck~(EI;E3) is Cp. Proof. (A) p = 0. 8y (11.2.13) we have
=4c2(?,4 ).g + "4. Let ~J~,~-~C:(EI;E2)
end ~ , ~ C k +
Ifr (E2;E3). Then by (1.5.2) +
where
Z ~ c 2 ( J L , ~ ) . ~ C k ~ ( E I ; E 3) by (11.2.13) and ~ o ( ~ + ~ ) by (11.2.10); hence c is equably continuous.
ICk~(E1;E3 )
-
Since V . c ( ~ x ~ )
= W.~o~
125
-
= 6V~)o ~ ~Ck~(E1;E3) by (11.2.10),
c is quasi-bounded. (8) Suppose the theorem for p end consider the map c: Ck~(EI;E 2) x C~+p+2(E2;E3)----~Ck~(EI;E3 ) We assert:
(ll.2.22)
c' : Ck~(EI;E2) x Ck~f+p+2(E2;E3)
d(Ck~(EI~E 2) x C~K+p+~'(E~E3)'~ ck'(ElsE3)~ is cp Observe f i r s t
that the map exists by (11.2.7),
(2.6.2) and
(2.9.1). We use some abbreviations=
E 4 = Ck~(Ez~E2),
E5 = Ck+p+2(E2;E3), ~ E 5 = Ck*(E1;E3), E7 = Ck~+p+I(E2;L~(E2;E3) ) , E8 = Ck~(E1;L~(E2;E3). We denote by T I
= E4
x
E5
....~E 4 and by
IT2 : E4 x E5 ~
E5 the projections, by
~T; : L~(E4;E? )
>L~(E4XEs;ET) and by
~ ; : Lf(Es;E7)
~L~(E4XEs;ET) the associated l i n e a r map~.
- 127 -
w
Further by c : E4 x E T - ~ - - ~ E 8 the composition map, and as
before by c 2 : E4 ----~L~(E5IE6) the mapping f ; Finally D = E 5 and id : E 4
~f*.
) E? denotes the map defined by O(f) = f' ) E 4 the identity.
The evaluation map s : Lq(E2;E 3) x E2-----gE 3 is bilinear and continuous by (6.2.3). Hence by (ii.I.I) and (2.9.2) =e have e
L~(E8, E4; E?).
6
Le(E8, E4; E7) ~
By (6.4.11)
=e have
L¢(E8; L~(E4~-?))-
We denote by ~ I the element thus corresponding to e". III' ~2 and id are obviously linear and continuous. Ill, ~r2 ere continuous by (6.3.31 D is linear and continuous by (I0.2.3). c is Cp by our induction hypothesis, c 2 is Cp by (11.2.20). A linear map is obviously Cp. Hence also idxD by (4.4.2). As a consequence of the formula c ' ( f , g ) . ( ~ , ~ ) -
(see (Ii.2o7)) =e have c' = ~
= ~ (g'of,~)+~=f .
6~l=c o(idxO) + ~r 2 oc 2 o W l which
is the composite of Cp-mappinge~ hence (ii.2.22) by (10.4.7). We further assert that (11.2.25)
c
:
E4
x
E5 - -
~
E6 i s C1.
From (11.2.?) me conclude t h a t the mapping c : Ck(EI;E2) x Ck+p+2(E2;E3)----~Ck(EI;E3) is differentiable. The spaces are eli admissibie by (I0.2.4).
-
128
-
The mapping c m = E4 x E5 ----~L(E4xE5;E6) (10.1.4) and ( 2 . 6 . 3 ) . (11.2.8). c'
= E4 x
is Co by (11.2.22),
We have v e r i f i e d the assumptions of
Hence c I E4xE5 ~ E
6 is diffsrenttable
by ( 2 . 6 . 4 ) .
E5 -----~Le(E4xE5;E6) being Co by (11.2.221 and ( 1 0 . 1 . 4 ) ,
=e get (11.2.23). Applying (11.4.1) to (11.2.22) and (11.2.23) one completes the i n d u c t i o n proof of the theorem. (11.2.24)
Latona. Let f: If i k o f
Eo----~ E = proj. lira Ei . t~I
= E° ~ E
f : Eo ~
k is Cp for each k 6 I~ then E is Cp.
Proof. Let p = O; ~VJ~,IE~Eo. Then ~J(ikof)(J~) = ik(~f(J~)) ~ Ek
f o r each k E I . ~f(~,
Hence ~Vf(J~)~ E. S i m i l a r l y one shows that
X ) ~F.
Suppose the lemma for p and assumes ik=f : E° -----~Ek is Cp+ l for each k E I. Then by (10.1.3) (ik=f)' = ikof' : Eo----~L4 (Eo;Ek) is Cp for each k E I. Since the inclusion i k = E
~ Ek is linear and continuous, so
i s ( t k ) . : L(Eo;E)-----~L(Eo;Ek) L~(Eo;E)-----~L~(Eo;Ek)
(11.2.25)
by ( 6 . 3 . 3 ) ,
by ( 2 . 9 . 1 ) .
=s assert
L(Eo,E) = p r o j . lira L ( E o ; E i ) . i£I
and also ( i k ) . :
-
129
-
One first verifies that the underlying sets are the same. We remark however, that the projective system L(Eo;Ei) , i ~
I, is
slightly more general than those considered in 2.3, since the maps Jk! : L(Eo;Ek) ik~: E k
J L(Eo;E|) induced by the inclusions ~ Et
,
k
~
~,
are not inclusions in the strict sense. This implies that the underlying set of proj. lim L(Eo;Ei) is not the intersection of i ~ I the sets L(Eo;Ei) , i ~ [, but has to be constructed in the usual manner. For this one verifies that the maps Jkt
satisfy the
transitivity condition jemOJkt = Jkm and that they are continuous. Both conditions are easily verified, because Jkl = (ik~)*" The structure of proj. lim L(Eo;Ei) is the coarsest for which the i ~ I induced maps Jk = (ik)* : p~oj. lim L(Eo;Ei) i~I tinuous. Furthermore: ~ L ( E o ; E ) , S ik(~(J~)) i k. ~ =
L ~(jq.)~ E for
= (ik.~)(j~) ~ E k for k ~ I and
i L(Eo;E k) are con-
W~Eo-'-'--~. WJ~&Eo~
(ik).(~) ~ L(Eo;Ek) for k ~ I.
This proves that also the structures of L(Eo;E ) and proj. lim i £ I L(Eo;Ei) are the same, hence (11.2.25). By appendix (5), (g.l.1) and (7.4.6) we get Le(Eo;E ) = proj. lim L~(Eo;Ei). i c I Applying the induction hypothesis to i' k- f ' one concludes: f' : E °..... ~Le(Eo;E) is Cp.
: E
o
; L~(
Eo;Ek)
-
130
It remains to show that f: E° ~
-
E is differentiable. We have
by assumption (ikef)'(x) = ikof'(x ) ~ L~(Eo;Ek) and R(ik, f)(x) a R(Eo;Ek) for each k 6 I.
Let @ ~ E o. Then
(ikof'(x)). ~E = ik(f'(x).~)~ Ek for each k • I, hence f'(x).~IE
and thus f'(x) ~ Le(Eo;E).
Let ~ V ~ E o. Then (of. proof of (3.1.5))
e~(ikof)(x).(W,~) = ik(eRf(x).(W,~)) ~ Ek for each k 6 I. It follows that
eRf(x).(~V,~)~ E, hence
Rf(x) & R(Eo;E). Since f' is C o by the induction hypothesis and (10.1.3), the assertion of the lemma is a consequence of (10.4.1).
(11.2.25)
Theorem.
The composition map
c, C~(EI~E 2) ~ c~• (E2;E3) ~
C~(EI~E 3) is c ~ .
Proof. By appendix (5), (10.2.4) and the definition given in I0.3 we have C~(E1;E2)
= proj.lim~, Ck#(EI;E3). The inclusion
~ (E2;E3) is continuous C$.. (EIIE2) x C p (E2;E 3) c Ck~(E 1;E2) x C k+p+l
and thus by (11.2.21) c: C~(E1;E2)xC~(E2;E3)----~Ck ~ (E1;E 3) is C
P
for each p, k 6 h~ °. From the lemma (11.2.21) one concludes
~ ~ c: C~(EI;E 2) x C~(E2;E3)-----~C~(E1;E3) hence the assertion of the theorem.
is Cp for each p e N
o
-
§ 12.
12.1
131
-
DIFFERENTIABLE DEFORMATION OF DIFFERENTIABLE ~APPlNGS.
Th~ differentiability of the evaluation map.
(12.1.i)
Lemma.
The mapping
Ul: Co~OR;E)----~E defined by ul(f) ~ f(I) is linear and continuous. Proof. The llnearity is obvious. Furthermore ~.I~IR, hence the result.
(12.1.~)
Lemma.
The mapping
u: E
~C:~R;E) defined by u(x).~ = ~.x is linear
end continuous. Proof.
By (6.4.2) and (g.l.l) we have E~L~(~;E). Obviously
L$(~;E) c Cee~R;E), and L~OR;E) has the structure induced by its inclusion in Co~(R;E). Hence the result by (2.3.6).
(12.1.~)
Theorem. The evaluation map e: Ck~+I(EI;E2) x E1 -----~E 2 is Ck. Proof.
The composition map
~: CkI'÷I(EI;E2) x Co~POR;EI)----'~Co~(IR;E2) (*) is Ck by (ii.2.21). Hence e = UlO ~ 6 (idx~) is Ck by (12.1.1), (12.1.2) end (10.4.?).
(*)
~(f,g) = c(o,f) = f.g
-
Because t h e i n c l u s i o n
132
CJ(E1;E2)
c
-
C kq+ l ( E 1 ; E 2 )
is continuous
for
each k G R~ o we g e t o b v i o u s l y :
(12.1.4)
Theorem. The e v a l u a t i o n f
e" C,= (E1;E 2) x E1
12.2
The l i n e a r
(12.2.1)
) E2 i s Cw
this
First
~E denote the map defined by
= x+y and ~:
by ~ ( x ) . y Proof.
E---~C~
= x+y. Then ~
we show t h a t ~ ( x )
i s immediate because ~ ( x )
Furthermore ~(x+h) The mapping h ~ - - - ~
- ~(x)
(E;E) t h e map c a r e c t e r i z e d
is C~
.
i s an element o f C ~ ( E I E ) . is a translation
= ~ where ~ ( y )
of E into
i s a c o n s t a n t map f o r of C~(E;E).
•
homeomorphism C ~ ( E I I C ~ ( E 2 . L E 3 ) ) ~ c ; ( E l x E 2 1 E 3 ) .
Lemma. Let s: ExE s(x,y)
map
C~ (E;E)
= h for
is linear
= ~ for
each x •
each x ~ E.
each y = E. and c o n t i n u o u s .
each h • E, hence o b v i o u s l y
T h i s shows t h a t ~ i s d i f f e r e n t i a b l e
and t h a t ~ ' ( x ) . h
for
an element
throughout
E. C o n s e q u e n t l y ~ '
Lemma. The mapping el: Cw(E2;E3)
E
is a cons-
t a n t map of E into L(E;C~ (E;E)) and thus C~ • (12.2.2)
But
~C=.(C==(EI; E2);C~(E1 E3))
defined by cl(g) = g. is linear and continuous.
-
133
-
Proof. The linearity of c I is obvious, g. is an element of ~ E3) C=.(C=.(EI;E2) x C=~(EI;
by (11.2.26) since g, is a partial
mapping of the composition map.V~C~.(E2;E3). We assert: (12.2.3)
~ . ~ Ck(C.. (E1;E2);C ~(E1;E3)) for each k • R~ o.
Let k = O, ~VJ~,~CJ(E1;E2). Then ~.(J~) =~'~:~c(J~xO,Ox~)~ Ca(El;E3) by (11.2.25) Suppose now the assertion for k. By (10.2.3), (10.3.3) and
(2.g.l) we have ~'~C. (E2;Lf(E2;E3)) Fromthe induction hypothesis one deduces
( ~ , ) . ~ Ck~C~ (EI~ E2) ~co(q $ ;L* (E2~E3))>. The evaluation map e: L$(E2IE3 ) x E2-----)E 3 is continuous by (6,3.3). Using (ii.I.I) and (2.9.2) one easily verifies that
e'l C~I(EI;L~(E2;E3)) x Ce(EI;E2))-----~C $ . -
..(E1;E3)
is continuous. The same is true for the corresponding map
el: C..~(EI;L*(E2;E3))
,L (C~.(E1;E2);C~.(EI;E3))
(cf.(6.4.11)). From (10.4.3) and (11.2.7) one concludes that #
(~.)' = (~i).((~').) ~Ck(C~(EI;E2);L$(C~(EI;E2);C=-(El;E3)~Further we had (case k =
0),~.~Co(C£(E1,E2),C.(EI~E3)).
By (10.4.2) it follows that ~. ~Ck+I(C"~(EI;E2)
; C.~ (El ;E3))
which finishes the induction proof of (12.2.3). From (10.3.2) and (2.g.1) one concludes the assertimn of the lemma.
-
(12.2.4)
134
-
Lemma. If f e Co.(El;E2), then
f*
L(c; (E ;E3) ,,.C"(El;E3)) .
Proof. Let ~$C'.~ (E2;E3), then f*(~) =~,f~C~.(EI;E3) by (11.2.26), because ~or =mc(fx0,0x~) and c ie equably continuous.
(12.2.5)
Theorem. There is a natural linear homeomorphism ~: C~ (ElXE2;E 3) ~
C.~(E 1;C. (E2;E3)), the map ~b-
being caracterized by (~g)(xl).X 2 = £ (Xl,X2)" Proof. Let if: E1 jections Xl~
~ElXE 2 resp. i2:E2-----~EIxE2 be the in-
~(xl,O) resp. x2~
~(O,x2).
Then
g(xl,x 2) = g(il(x I) + i2(x2) ) = (g,~(il(Xl))-i2)(x2) where s: (EIXE2) x (EIXE2)
~EIXE 2 is the map discussed in the
lemma (12.2.1). Hence (~g)(xl) = g.s(il(xl))oi2 = (12og,os°~l)(Xl) where i I • E1-----~ElxE2 is linear and continuous, ~." ElXE 2---~C~(EIxE2~EIxE2) is C, by (12.2°i), g.: C~(ElXE2;ElXE2)~ ~C~(EIxE2;E3) is C ~ which follows easily from (Ii.2.21), g. being a partial mapping of the composition map; finally i .2 * • C~ (EIxE2;E 3)
>C~(E2;E3) is Coo for the same reasons.
By (10.4.7) it follows that ~z-g is an element of C.~ (E1 •,C ~ (E2;E3). Let
E4 = Coo (ElXE2; E3) , E5 = C~. (ElXE2;ElXE2), E6 = C..( E l",C~(E2;E3) ).
- L35 -
Then V #
~' (i~). o(~oi 1) * o cI whets cI ( L . tL' - 4.C ..(EasE4) ) by
(12.2.2), ("s . i l ) * ¢ L(C~(Es,E4),C~" (ELSE4)) by ( 1 2 2 4 ) . (i2) . ~ L(C~(EI;E4);E 6) by (10.4.3), (2.9.1) and the same arguments as in the proof of (11.2.26). Hence ~Tr is the composite of linear and continuous maps and thus obviously linear and continuous. Conversely let ~ : C~(E1;C~-~(E2;E3) )
~Ce6- (EIXE2;E3) denote
the linear map defined by (~f)(xl,x2) = f(xl).X 2. f = e o[f.iTl, IT2~
where e: C~..(E2;E3) x E 2 ~ E
Then 3 is the
evaluation map and ~l reap. qt2 are the projections of ElXE 2 into E1 resp. E2.
Hence by (10.4.5) and (10.4.7)
~ f is an
element of C~(EIXE2;E3). We shall prove the continuity of the mapping ~=
~
by the formula
e. o ~os(i2(1T2))oilo IT 1 and using the abbreviations
~6
=
c*-"(ElSe-" *
(E2;E3)) '
E7 = C ~ (EIXE2;C ~ (E2;E3)), E8 = C~, (EIXE2;E2) , E9 = C% (ElXE2;C ~ (E2;E3)xE2). One has ~rI E C~.(E6;E 7) by (12.2.4), iI E C.~(E7~E7xE8), ~(i2(Tf2) ) C C~.(E7xEs;E7xE8) by (12.2.1), c~ ~ C~(E7xE8;Eg) by (10.4.5) and e. ~ C@(E9;C@~(ElXE2;E3) ) by (12.1.4) and (11.2.26).
-
From (I0.4.7) it follows that ~
136
-
is C ~ and thus obviously con-
tinuous. Furthermore ~ o ~
and ~ o ~
which ends the proof.
ere the corresponding identities
-
137
-
APPENDIX (1) Proposition. If E is an admissible vector space, then the scalar multiplication:
IR x E---~E is equably
continuous. Proof. Let X ~ E .
Then by (2.1.i),
(2.4.2) and (7.i.I)
: (Xv(- X))'LE. If Y £ ~
, then Y ) (X•(-X)),
X ~]E. Thus for any xGX the
segment [ - x , x ] ¢ (Xu(-X))" and consequently
~.xC(Xv(-X)) ~
for each #~#il and x (X, hence IIX ~Y where I 1 = [ - I , i ] , therefore 11 )E i ~
LE. By (2.4.2) also
and
~ . I 1 X = I~ X ~ E for
each ~# O. This proves the equable continuity of the scalar multiplication by (2.8.8) and (2.5.1). (2) Proposition. Let ]( be a filter on a vector space. Then the filter X * =
W.sup (~.~()
is the finest one
among all equable filters coarser then Proof. Obviously X'is equable. Let M E ~*. Then m~V.A, V ~ W
'
A ~ ~su
for all
Choose o( ~ V,
~
~(
= ~ (W~)
= (~V)~
where
,
~ ~ is coarser then ~( .
be any equable filter coarser then ~ ~ # 0 :
.
~ O. Since A ¢ ~ (
~.A (~( , and therefore m ( ~
A. This proves that
Then, for • ~
•
~ ~ D, we have
because M ~ ~ Let now ~
(~ X )
~
= W.~ = ~
;
hence
: ~i
~ = I;~
-
~up ( ~ )
" ~
and thus
~"
138
-
= V.sup ( ~ )
which completes the proof. Corollary. For any pseudo-topological vector space E one has:
E
X' E.
(3) Proposition. Let E be any pseudo-topological vector space. Then
( ~ ~ E implies sup ~
E equable ~
Proof. (e) Suppose E equable, and let ~ E . with
~
•
~
= V. ~ LE. Since •
for
~
E).
Then there exists
is coarser then ~ and equable,
~ ~ O; hence su
(~)~
i
(b) Suppose the condition satisfied, and let
~IE.
showing that ~1.~
me have
~
m
~
= W. ~
Then
~E, which proves that E is equable.
(4) Proposition. Let El, i gI be a projective system of equable and admissible vector spaces (cf.(2.3.5)(c)). Then E' = proj.lim E i is equable and admissible. i¢ I Proof. The admissibility is proved in (7.3.2). Let ik(~) L E k for each k g I. But sup ~ ik(~) ~$0
one has
U ~ ~0
ik(X~ ) =
such union. Hence sup ~ ~ E J$O
proposition.
U ~0
~ ~E. Then
: ik(sup ~ ~ ), since 8~0
[k(~X~)
= ik( ~tJ 0 ~ ~
for each
and thus E is equable by the preceding
-
139
-
(5) Proposition. For each projective system of admissible vector spaces, (proj.lim Ei )e = proj.lim £i$. i~l iel Proof. (a) Let )&~(proj.Iim £i)°. Than there is ~ with itl ~( " ~ = W~&proj.lim El" Since ik(K)"ik(~) = ik(V~) = Wik(~), i&I one gets ik(K)~ £kf for each k ~I and thus K&proj.lim £~. l~I i (b) If X proj.lim Eit , then ik(X ) &E k for each k6I by ilI (2.6.3). But by (4) and (2.6.1) we can suppose that ]( is equable, hence ~(proj.lim Ei )$ . i~I
-
140
-
NOTATIONS
~ , $', X
filters
@
empty set
[8],[A],[a]
generated filters
i.I I.I 1.1
comparison of filters
(1.2.1)
][converges to x on E
2.1
~converges to zero on E
{Xi' i.I sup
i~I
~
family of filters
2.4
1.2
least upper bound of filters
1.2
l
XlvK2
sup (~(l' "~2 ) SupX
Xo
1.2
(2.4.3)
X
2.7 2.7
,(xF
(5.3.3)
E, E l, E 2
pseudo topological spaces
i
underlying set
Ee
eouable space associated to E
E°
localiy convex space associated to E
E1 z_ E2
comparison of structures
E1 x E2, x E . i~I I
direct product of pseudo-topological spaces 2.3
2.6 2.7
2.3
-
141
-
direct product of filters ( on a direct
~I x X 2
product of two sets) 1,4 the reels (with the natural topology) neighborhood filter of zero in closed interval in closed interval [ - ~ , ~ ] , ~ •o
Ij
{ 1,2,3,.-- l N°
{0,1,2,...I
proj. lim Ei i~l R(EI;E 2)
projective limit
Ln(EI~E 2)
space of n-linear maps
Ck(EIIE2)
space of Ok-mappings
10.2
C,~(EI ;E2 )
space of C,,-mappings
10.3
L'(EI,E2)
instead of (Ln(EI;E2)) °
(6.1.7)
c:(q E2)
instead of (Ck (E I ;E2)~
(6.1.7)
C~(EI ;E2 )
instead of (C~(EI;E2)) m
f : El,,,
mapping f of E 1 into E 2
xs
set of remainders
m E2
(anonymous) map
fl x f2'
(3.1.2) (6.1.6)
x is sent into y under the considered
;y
[fl,f2 ] ,
2.3
fi
1.3
x f.
1.3
i~l
i~l z
-
IT
k
142
-
k-th projection
c
composition map
e
evaluation map
2.3
map associated to u
(ii.i.i)
U°X
instead of u(X)
~f(x,y)
abbreviation of f(x+y)-f(×)
ef(i,x)
abbreviation of
{~0
~or
~~ 0
for
~
=0
(4.3.2)
f'(a)
differential quotient
f'(a), Df(a)
derivative of f at the point a
Rf(a)
remainder belonging to 8 map f which is differentiable at a
f(k)(a)
see
Dlf(al,e2),D2f(al,a2 )
partial derivatives
f
f*(g) = gof
f.
f. (~) = f-0
,,,,,,@
implies
(5.2.4)
9.2
if and only if linearly homeomorphic
8.I
(3.2.2)
-
143
-
I N D E X
admissible
7.i
almost all
5.1
associated locally convex topological vector space
2.7
canonical isomorphisms
6.4
chain rule
3.3
coarser
2.3
compatible
2.4
composition map
6.3
continuous
2.2
continuous with respect to associated structures
2.9
Ck-mapping
iO.l
derivetlve
3.2
diagonal map
1.3
differentiable at e point
3.2
differentiable map into e direct product
4.4
differential quotient
4.3
direct product
2.3, 4.4
equable continuity
2.8
equable filter
2.5
equable pseudo-tooological vector space
2.6;appendix~
evaluation map
6.2
filter
i.i
filter-basis
i.i
finer
2.3
Fr~chet condition
4.1
function spaces
6.1,6.2
fundamental theorem of calculus
5.1
-
144
-
higher derivatives
g.l
higher order chain rule
g.2, 10.4
homeomorphism
2.2
images of filters under mappings
1.4
inclusion map
2.3
induced structures
2.3
infimum of filters
1.2
infinitely differentieble
g.2
local cerscter
3.4
mappings into direct products
1.3
mesn value theorem
5.2
netursl
6.4
neighborhood, E-neighborhood
3.4
open, E-open
3.4
partial mapping
8.1
partial derivatives
8.1
projective limit
2.3
pseudo-topological space
2.1
pseudo-topological vector space
2.4
pseudo-topology
2.1
quasi-bounded filter
2.5
quasi-bounded map
2.8
remainder
3.1
separated
3.1
subspece
2.3
supremum of filters
1.2
-
145
-
symmetry of f"(×)
9.1
underlying space
2.1
uniform convergence on bounded sets
6.1
-
146
-
REFERENCES
[1]
Bastiani A.
"Applications diff~rmntiables et vari~t@s
:
diff~rentiables de dimension infinie"~ Journal d'Analyse Math~matique XIII (1964) p.l-ll4. [2]
Binz E.
: "Ein Differenzierbarkeitsbegriff in limitLerten Vektorr@umen", Comm. Math. Helv. 41 (to appear).
[3]
Dieudonn~ J. : "Foundations of modern analysis", Academic Press 1960.
[4]
Fischer H.R. : "Limesrgume", Math. Annalen 137 (1959) p.269-303.
Is]
Keller H.H.
: "R~ume stetiger multilinearer Abbildungen als Limesr~ume", Math. Annalen 159 (1965) p.25g-270.
[6]
Keller H.H.
: "Differenzierbarkeit in topologischen Vektorr~umen", Comm. math. Helv. 3B (1964) p.308-320°
[7]
Keller H.H.
:
"Uber Probleme die bei einer Differentialrechnung in topologischen Vektorr@umen auftreten", Nevanlinna Festband, Springer (to ~ppear).
: "Topologische linearB R~ume", Springer 1960.
[8]
K~the G.
[q]
Kowalsky H.J.: "Topologische R~ume"~ Birkh~user 19ill.
Offsetdruck: Julius Belt~ Weinheim/Bergstr.