groups
Citation for published version (APA):
Graaf, de, J. (1983). A theory of generalized functions based on holomorphic semi-groups: part B : analyticity spaces, trajectory spaces and their pairing. (Eindhoven University of Technology : Dept of Mathematics : memorandum; Vol. 8307). Technische Hogeschool Eindhoven.
Document status and date: Published: 01/01/1983
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne Take down policy
If you believe that this document breaches copyright please contact us at: openaccess@tue.nl
providing details and we will investigate your claim.
TECHNISCHE HOGESCHOOL EINDHOVEN NEDERLAND
ONDERAFDELING DER WISKUNDE
TECHNOLOGICAL UNIVERSITY EINDHOVEN THE NETHERLANDS
DEPARl'MENT OF MATHEMATICS
Memorandum 83-07
J. de Graaf
A THEORY OF GENERALIZED FUNCTIONS BASED ON HOLOMORPHIC SEMI-GROUPS
Part B: Analyticity spaces, trajectory spaces and their pairing
In a complex separable Hilbert space with inner product (-,.)
X
we consider an unbounded non-negative self-adjQint operatorA.
This operator will be fixed throughout chapters 1, 2, 3. SinceA
is self-adjoint it admits a spectral resolutionFor details on such spectral resolutions, see [y], [LS]. By O~a<b~oo, a we mean 00
f
Xa,b(A)$(A)dEA -00 with on (-l,b] if a=
0 X b (A)=
t
on (a,b] if a > 0 a, elsewhere. We define-tA
f
-At E e=
e d A t ~ <I: •o
For t € IR the operator e
-tA
is self-adjoint, e-tA
is bounded iff Re t~
o.
-tA .
.
'ff 0-tA..
'bl de 1S um tary 1 Re t
= .
Further e 1.S 1.nverti e an-tA -1 tA
Yt€~ (e )
=
e+ -tA
On ( , i.e. the set of t ~ Q: with Re t > 0, the operators e establish a one-parameter semi-group of bounded injective operators on
X.
We now introduce our test space
SXtA'
This space equals the set of analytic vectors for the operatorA.
Cf. [Ne].Definition 1.1. 5 X,A
=
U { e -tA xI
x € X}=
Re t>O U Re t>O e -tA(X) •SX,A is a dense linear subspace of
X.
From the semi-group properties the following equalities immediately follow for any 0 > 0U 1 n€IN,O<-<o
n
-tA
Each of the spaces e (X), t > 0, can be considered as a Hilbert space. The inner product is
tA tA
= (e u,e v) X ==
The completeness of
e~tA(X)
follows from the closedness of etA e-tA(X) consists of exactly those u € X for which< QO •
B.2
If locally no confusion is likely to arise we suppress as many subscripts as possible. In chapters 1, 2 and 3 we shall write consistently:
5
for5
X,A' Xt -tA
for e (X), ( • , .) for (.,.)
X.
Definition 1.2. The
strong topoLogy
on S is the finest locally convex topology on5
for which the natural injections it : Xt +
5,
t > 0, are all continuous.In other words:
5
is made into a locally convex topological vector space by imposing the inductive limit topology with respect to the family {Xt}t>O' Cf. [SCH] Ch. II.G. Since the natural injections XT + Xt, T > t > 0, are all continuous the inductive limit topology is already brought about by the family {X1/} IN' A subset
U
c5
is open iff for every t > 0 the set-1 n n€
i t (U) == U n Xt is open in X or, equivalently, iff for every n € IN the set
is not strict! A closed subset of X
T when considered as a set in
X
t,o
< t < 'r, is not necessarily closed. We shall see that open sets in S arealways unbounded.
We introduce some notations:
- B denotes the set of everywhere finite real valued Borel functions on IR such that Vt>O the function w(x)e-tx is bounded on [o,~).
- B+ C B contains those W for which there exists € > 0 such that w(x) ~ € > 0,
voor.alle x ~
o.
Let W ~ B. Then weA) is defined by
The domain of w(A}, denoted by D(w(A}, consLsts of exactly those x ~
X
with ~ 2of
w (A)d(EAx,x) <=.
It follows that S is contained in the domain of each operator w.(A) •-tA
Further: VW~B, Vt>O the operators weA)e are bounded and self-adjoint, The sets of operators corresponding to the sets B and B+ will be denoted by B(A) and B+(A), respectively. The operators in B+(A) are all strictly
positive •
Definition 1.3. For each W € B+ we introduce the (semi-) norm Pw by ~ ":!
pw{u)
=
II w(A) ull= {
f
w(A)2 d(EAU,U)} u ~S .
o
Further for W E B+ and € > 0 we define the set
U,J,
=
{u E SI
IIw(A)ull <d .
'I',€
The next theorem is very fundamental, It tells that the strong topology in SX,A is generated by the semi-norms Pw'
B.4
Theorem 1.4.
I. V1jJeB , Ve>o U", is a convex, balanced, absorbing open set in the strong
+ '1" e:
topology. In other words the semi-norms P1jJ are continuous.
II. Let a convex set " c S be such that for each t > 0
"n
Xt contains a neighbourhood of 0 in Xt • Then" contains a set UtjJ,e: with 1jJ e B+ and € > O. Proof.
I. A standard inner product argument shows that U,,, is convex, balanced and 'I',e:
absorbing. For the terminology see [Y] Ch. 1.1. We now show that for each t > 0 UtjJ,e: n Xt is an open set in Xto
By definition
U n X
t = {u
I
u eXt' IItjJ (A) ull < d . 1jJ, e:Because of 111jJ (A) ull :s IItjJ (A) e -tAli lie tA u II and the boundedness of 1jJ (A) e -tA, the norm P1jJ is continuous on X
t "
II. Introduce the operator p... In dE" n € IN. Let r be the radius of the
n n-l A n
largest open ball in P (X) which fits in " n P (X). SO
n n
r = sup {p
I
[u e P (X) 1\ liP ull < p] .. u E P (m} "n n n n
Next define X : IR +IR as follows
X (A) =
o
if A < 02
max(~2
,1)
ifA
E (n-l,n] n X (~) if A ... 0 • n e lNWe prove X e B +" Let t > O. Then there is e > 0 such that
eo
{u
I
feAt d(EAu,u) < e2} c " nX~t
o
because" n X~t contains an open neighbourhood of O.
Thus we find that for all n E lN, r >
€e-~nt.
Hence for A e (n-l/n] n-At
sup e X(A) <: 00 •
A 2:0 We now show
Suppose u € \ for some t > O. Then E:=l
nPnult~
<: <:0 and for some 1', 0 < l' <: t(*) IIPuU 2
s
e- 2 (n-l) (t-1') lIu II 2t • n l'Further, because of our assumption (*)
II P ull
n
-2
s
~ min(n r ,1) nTherefore 2n2 P u €
n
n X for every n E IN.n 1: In
X
we represent u by 1: u=
I
~
(2n2 P u) +(~
_1_)
~
n=l 2n n n=N2i
with ( 01))-1
01)=
L
-L
L
P
u •~
j=N 2j2 n=N n Wi th (*) we calculateSince
n
nX
contains an open neighbourhood of 0 for N sufficiently largeT
u E
n
nX .
N T
Finally we gather that u is a sub-convex combination of elements in
n
n XT
which is a convex set. A posteriori i t is clear that u €
n
n
X
t • Remark. Similar to the proof of part I one proves that the sets
$ E B+, E > 0 ,
are .closed.
B.6
Defini tion 1.5. A subset W c: S is called bounded if for each O-neighbourhood
U
inS
there exists a complex number A such thatW
c:AU.
Cf. [SCH] 1.5. The next theorem characterizes bounded sets inS.
Theorem 1.6. A subset W c
S
is bounded iffand
Proof •
3t>O W c
X
t3M>O 'v'ueW lIullt :S M •
.. ) Let 1/1 E Band u E: W. Then
+
.) Taking 1/1
=
1 we observe that W is a bounded set in X. Denote its bound by P. Suppose the statement were not true, thentA
'v't>O 'IM>O 3UEW lie ull > M •
By induction we define two sequences of real numbers {t }, {N }, t ~ 0,
n n n
N
+
~ as n ~ ~ and a sequence {u } c:W
as follows:n n
n = 1 Choose tl > 0 and M
=
2. Then take N1 > 0 and u1 E:W
such thatn
=
t + 1: Supposeis true. Then
W
is a bounded set inX
t, t s ~tt' becauseco
J
s
p2e2tNt + t + 1 .N~
If our sequence terminates for a certain value of n then
W
is a bounded set. If not, define the function n(A) on (O,~) byneAl
=
eAtn on the interval [N n-1,N nJ ,
n=
1,2, ••••Since neAle-At is a bounded function for t > 0 we have n € B+. Then because
of the assumption the sequence n (A) u should be bounded. However
n
~
IIn(A} unIl2
=
J
n2(A)d(EAun,un) > n + 1 •o
Contradiction!
In the next theorem we give a characterization for the convergence of sequences {u } in the strong topology of
S.
n
Theorem 1.7. u + 0 in the strong topology of
S
iffn
Proof •
.. ) For any 1/! E B we have 111/!(A)u II S It1/! (A>e-tAnnetAunl[ + 0 as n +
~
because-tA + n
1/!(A)e is a bounded operator on
X
•
• ) Suppose u + O. Then V1/!EB II1/! (A) u " + O.
n + n
o
We conclude that the sequence {u } is a bounded set. So 36>0 VnElN !tu
II
S 6.n n l'
Further, taking tjI = 1, it is clear that l[unll
O + 0 as n + ~. From this we derive
+0 as n + <Xl •
Now we show
II
un" t + 0 for any t < 1'.L
J
e 2At d(EAun,un) +B.S
The second integral can be estimated uniformly
co co
f
e 2At d(C"u,u )=
J
-2A(T-t) 2AT deE ) ~n n e e "un,un L L co -2L (T-t)
J
e2"T
d(E"u ,u ) ~ -2L(T-t) 2 ~ e ee .
n n LBy taking L sufficiently large the second term in (*) can be made smaller than ~€ uni formly in n.
The first term in
(*>
tends to 0 as n -+ "" because of (*).o
Theorem 1.8.
I. Suppose {u } is a cauchy sequence in the strong topology of S. Then n
3t>O {un} c X
t and {un} is a cauchy sequence in Xt •
II. S is sequentially complete, i.e. every Cauchy sequence converges to a limi t point.
Proof.
I. An argument similar to the proof of the preceding section. II. Follows from I and the completeness of
X
t • Next we characterize compact sets in S. Theorem 1.9. A subset XeS is compact iff
3t>Q X c X
t and
K
is compact in Xt • Proof •• ) Let {Qa} be an open covering of
K
inS.
Then {Qa nX
t} is an open covering of K in X
t • Since X is supposed to be compact in Xt there exists a
N
finite subcovering {Qa.}, 1 ~ i ~ N, with U
i=l (Qa~ n
X
t) ~K.
But thenN ~ •
certainly Ui=l Qa ~ K. i
o
.. ) Since Xis compact i t is bounded and therefore i t is a bounded set with bound
e
in XT for some T> O. We show that K is compact in Xt whenever t < T. Consider a sequence {u } cK.
There exists a converging subsequence {unj }J J Put
Unj -
u=
Vj •
{Vj } is a bounded sequence in Xt and II Vj II ~ 0 as j ~=.
Now we find ourselves in exactly the same position as in the proof of the only-if-part of Theorem 1.7. We conclude Ilv jll t ~ 0 whenever 0 S t < T. Our
subsequence {unj } converges to u in Xt-sense. This shows the compactness of
K in Xt for 0 s t < T.
0
The next theorem gives an alternative description of the space SX,A' The simple proof is omitted.
Theorem 1.10. Let u € X be such that u € D(~(A}) for all ~ € B+. Then u € S.
In other words
o
In order to make a link with the literature on topological vector spaces we now describe the properties of our space S by using the standard terminology of topological vector spaces. [sea].
The terminology is explained in the proof. Theorem 1. 11 •
I. SX,A is complete. II. SX,A is bornological. III. SX,A is barreled.
-tA
IV. SXtA is Montel iff for every t > 0 the operator e is compact as an operator on X.
-tA
v.
SX,A is nuclear iff for every t > 0 the operator e is aas
(=Hilbert-Schmidt) operator on X. Proof.
I. Let {Xa } be a Cauchy net. The a's belong to a directed set D. For each neighbourhood Q 3 0 there is y €
D
such that whenever a >y and S7
y onehas xa -
Xs
€ Q. We now prove that there exists an x € S such that xa ~ xin the strong topology. Let x be the limit of xa in X-sense. For each ~ € B+
closed-B.I0
ness of ~(A) one has x~ € D(~(A}) and x~
=
~(A}x. The result follows byapplying Theorem 1.10.
II. Every circled convex subset fl c S that absorbs every bounded subset W c S
has to be a neighbourhood of O. Let
B
t be the open unit ball in
X
t, t > O. Bt is bounded in S, therefore for some E: > 0 one has E:Bt c fl n Xt •We conclude that for every t > 0 the set fl n
X
t contains an open neighbour-hood of O. But then according to Theorem 1.4.11n
contains a set~" •""E: III. A barrel V is a subset which is radial, convex, circled and closed. We have
to prove that every barrel contains an open neighbourhood of the origin. Because of the definition of the inductive topology V n
X
t has to be a barrel in Xt in the Xt -topology, for every t > O. Since Xt is a Hilbert space there exists an open neighbourhood of the origin L with LeV n
X
t • Again the conditions of Theorem 1.4.I1 are satisfied so that V contains a set U,I. •""E:
IV. We have to prove that every closed and bounded subset of S is compact iff
-tA
for every t > 0 the operator e is compact.
-tA
.) Suppose e is compact for every t > O. Let TN be a closed and bounded subset in
S.
ThenW
cX
t for some t > 0 and
W
is closed and bounded inX
t• See Theorem 1.6. Take.,a
< • < t. We claim thatW
is a compact set inX
.•
T
To see this, consider the following commutative diagram where ~ denotes the natural injection:
X
T
r
e -TAThe vertical arrows are isomorphisms. So
G
is a compact map and TN is compact in X • Consider an open covering {C } of TN in S, then {C n X } is an openT a a T
covering of
W
inX
.•
Because of the compactness of TN inX
there is a finiteT •
subcovering:
N
W
c U(Ca.
n
X) •
i=1 l. T., Suppose
S
is Montel, Le. each closed and bounded set is compact. Let {un} be a bounded sequence inX.
Pick any fixed t > O. Consider thesequence {e-tAu }. Consider the closure in
S
of this sequence. This closure is a boundedse~
and, according to our assumption, compact. So {e-tAu }n
.
S
-tAcontains a converging subsequence in : e Un. -+ v. This sequence certainly
_~ J
converges in X. SO e must be a compact operator. -tA
V • • > Since e is HS and self-adjoint for all t > 0 there exists an ortho-normal basis {e } of eigenvectors of A in n .
X.
We order the eigenvalues with repetition according to multipliCity 0 S 1.1 S A2 ::;; 1.3 S •••• We have An -+ QOas n -+ QO. Thus we get the representation
-tA e u
=
~
n=1 -t). e n(u,e)e n n -t).where {e n} is an 12-sequence for each t > 0 and hence an t
1-sequence.
-tA
So e is a nuclear operator for each t > O.
S
is a nuclear space iff for every semi-norm p~, ~ E B+, there is asemi-norm p , X E B+, X ~ ~, such that the canonical injection J : Sp -+ Sp is
X X 1Ji
nuclear. Cf [y]. Here the Banach space Sp is the competion of S with respect to the norm p.
The canonical injection can be written
Note that _
{x.~l(A)ej}
and _{~-l(A)e.}
J are orthogonal bases in the Hilbert spaces Sp respectively X .c:
-p~•
Now suppose there exists a E B+ such that a-1(A) is a nuclear operator. If
we take X
=
alJi then the canonical injection can be writtenJu =
-1
Since {a (Aj
>}
is an 11-sequence J is a nuclear operator.
It only remains to show that a E B+ with the desired properties can be con-structed. To this end define integers N(n), n
=
0,1,2, ••• , such thatB.12
N(O)
= 0,
N(n + 1) > N(n),AN(n)+l > AN(n) and
Q)
-A./n
I
e J<"2'
1j=N(n) n
Next define the sequence {"k} by
N (n - 1) < k
:s:
N (n) , n = 0,1,2, ••••00
It is clear that L
k=l vk is convergent. A
In
Define a step function cr € B+ by setting cr(A)
=
e k on the interval(\-1
2 + \ ' \+ \+1]
2 ' k=
2,3, ••• , and cr(A)=
eA
1 on 0,[1..1
+ 2A2]
•a-leA} is nuclear because it is self-adjoint, it has discrete spectrum, and the eigenvalues vk establish an t
1-sequence •
• ) Suppose S is nuclear. There exists 1/J € B such that the injection
Sp.
-'l- X 1s-1 + 1/J
a nuclear map. Therefore 1/J (A) must be a nuclear operator and consequently also
as.
But then e-tA=
1/J-l(A)[1/J(A)e-tA] must beas
since the operatorWe now introduce our space of trajectories T X ,A' The elements in this space are candidates for becoming "generalized functions fl (or fldistributions").
Definition 2.1. TX,A denotes the complex vector space which consists of all mappings F : (0,=) + X such that
(i) F can be extended to an analytic function on the open right half-plane ¢+ •
+
-tA
(11) Vt/'tE:¢ e F('r) = F(t+T) •
Such a mapping F will be called a trajectory. If for some ~ > 0 F1 (~)
=
F2(~)then F1
=
F2• This follows immediately from the semi-group and analyticity properties. We shall use the notation e -t'\ for F (t). It is immediate that
-tA
Vt>O e F € S. In chapters 2 and 3 T
XtA is abbreviated by
T.
-tA Defini tion 2.2. The embedding emb
for all x € X and t > O.
X +
T
is defined by (emb x) (t) = e xSometimes we shall omit the symbol emb and loosely consider X as a subset of T. Thus SeX c
T.
The question arises whether there exist trajectories F € T such that F(t)
does not converge or I even II F (t) I[ t = if t
+
O. The answer is affirmative: simply take x E X \ D(A) and F(t)=
Ae-tAx. More general examples areobtained by taking
~
€ B, u € X and defining G(t)=
~(A)e-tAu,
t > O. The next theorem shows that all elements ofT
arise in this way.Theorem 2.3. For every F €
T
there exists w € X and ~ € B+ such thatF(t)
=
~(A)e-tAw
for every t > O. We write F=
~(A)w.
n Proof. Let F €T.
Let P denote the projection f dE,.n ~1 A
Put a = ileA P F(1) II.
n n Let ~ be defined by { max (a ,1) ~(A)
=
n 1 for A € (n-l/n] I n E IN elsewhere. Then eA P F(1) € X. nB.14
We shall show that ~ E B+. Let t > O. Then for A E (n-1,nJ with an ~ 1 one estimates n -2(n-1)t e
J
d(EAF(t),F(t» Se2tUF(t>l12 n-1Defini tion 2.4. The strong topology in T is the topology induced by the semi-norms p , n E IN,
n
1
p (F) = I!F(-) IfX •
n
n
In other words a base of open O-neighbourhoods is given by
Remark. The strong topology is equivalent to the topology of uniform
con-+
vergence on compacta in f; •
Theorem 2.5.
o
I. T endowed with the strong topology is a Frechet space, Le. i t is metrizable and complete.
II. A base of open sets {OU,t} is given by 0U,t
= tF
I
F(t) EU}, U
open inX,
t > 0 and fixed.(In words: the set of trajections which pass at t through
U.>
Each open set inT
is a denumerable union of sets of this type. Proof.I. The topology. is generated by a countable number of semi-norms. Hence
T is
metrizable. See [yJ.Further, since
T
is metrizable i t is complete iff each Cauchy sequence converges. Let {F } be a Cauchy sequence inT.
m n I ( (l/n) - (l/v»
A
1an element hl/ t
X.
For v > n one has F (-) "" e F (-). Sincen «l/)_(l/»Amv mn
both sides converge and e n v is a closed operator we have
L" D( «l/n)-(l/v»A d h "" «(1/n)-(1/v»A h N define
hl/n <:. e ·1 an l/v e l/n' . ow we
F(t) "" e-(t-(l/v»A hi/V' Re t e
[-V '
1 v-l' , I Clearly F tT
and is the limit of {F }.m
veIN.
II. It is enough to prove that the semi-norms F ..
II F (t)" are continuous. This
follows simply from the estimatefor 0 <
l
< t, nTheorem 2.6. emb(S) is everywhere dense in T.
Proof. For any F €
T
the sequence emb(F(l/n» converges to F in the strongtopology.
o
o
Theorem 2.7. A set
BeT
is bounded iff for every t > 0 the set {F(t)I
F €B}
is bounded inX.
Proof •
.... ) Each continuous semi-norm has to be bounded on B. Therefore p (F) "" IIF (l/n) II
n is a bounded function on B.
--rA
Because of the boundedness of e for each T > 0 i t then follows that {F(t)
I
F € B} is a bounded set for each fixed t > O..) Trivial.
Theorem 2.8. A set
BeT
is bounded iff there exists a bounded setV
cX
and ljI € B+ such that B= ljI(A)
(V) •Proof.
-tA
.) For each t > 0 the set ljI(A)e (V) is bounded in
X.
B.16
-) See [ETh], Ch.II cor. 2.5. It is not difficult to give an ad hoc proof in
the spirit of Theorem 2.3.
o
Theorem 2.9. A set K c T is compact iff for each t > 0 the set {F(t)
I
F ~ K} is compact in X.Proof.
-) If
K
is compact then each sequence {F } cK
has a convergent subsequence.n
This means that in the set K
t
=
{F (t)I
F € K}, t fixed, each sequence has aconvergent subsequence, which says that K
t is compact in X.
-) Let {F } be a sequence in K. We must prove the existence of a converging
n
subsequence. Consider the sequence {F
n(l)} c
Kl
c X.Kl
is compact, therefore a convergent subsequence inKl
exists. Denote it by {F~(l)}. The sequence{F~(~)}
hasa convergent subsequence inK~.
Denote it by{F~(~)}.
Proceeding in this way we arrive at sequences {Fm} c K such that {Fm} c {FR.}
ml n . n n
for m > R. and {Fn(m~} converges to an element in K
1/m• The "diagonal
sequence" {F~} has the property that {F~(t)} converges to F(t) E K
t• But then
also Fn + F in the strong topology.
0
n
Theorem 2.10. A set
K
c T is compact iff there exists a compact setW
c.X and ~ ~ B+ such thatK
=
~(A)(W).
Proof •
• ) For each t > 0 the operator
~(A}e-tA
is bounded. Therefore~(A}e-tA(W)
is a compact set in X.-) See [ETh], Ch. II cor. 2.13.
In the last theorem of this chapter we describe the properties of our topological vector space T in the standard terminology of topological vector spaces [SCH].
Theorem 2.11.
I. TX,A is bornological. II. TX,A is barreled.
operator on X.
IV. TX,A is nuclear iff for every t > 0 the operator e-tA is a
as
operator on X. Proof.I,ll. T is bornological and barreled because it is metrizable. For a simple proof see [SCB] II.S.
III • • ) Suppose T is Montel. Take a bounded sequence {x } c X. The sequence {F }
n n
defined by F (t)
=
e-tAx is a bounded set in T. Because of our assumptionn n
the closure of this set is compact. But then, by theorem 2.9, for each t > 0 the sequence F (t) contains a converging subsequence. This shows the
compact--tA
nness of e
-tA
.>
Suppose e is compact for every t > O. LetB
be a closed and bounded set. Then the setB
t
=
{F(t)I
F €B,
t > 0 and fixed} is bounded. Since Bt+T=
e-tA(B~)
it follows that Bt+T is precompact in X. Now take any sequence{G
n} cB.
Because of the pre compactness of eachB
t the "diagonal procedure" of the proof of theorem 2.9 yields a converging subsequence of {G }. This subsequence converges to an element in nB
becauseB
is closed. We conclude that B is compact.IV • • ) Suppose T is a nuclear space. Take n € IN. There exists a semi-norm
Pm ~ Pn such that the natural injection fpm ~ Tpn is nuclear.
This natural injection is realized by the map e-(l/n-l/m)A which must therefore be nuclear. It follows that e-
tA
isas
for each t > O.-tA
.) Suppose for each t > 0 e is HS. Because of the seud-group property e-
tA
is also nuclear. Let m > n. The natural injection Tp~
Tp
is realized-(l/n-l/m)A . . m n
by e and 1S therefore nuclear. This is enough for T to be nuclear.
CHAPTER 3. The pairing of SX,A and TX,A
We now consider a pairing of S-and T. It turns out that S and T can be considered as each others strong duals.
Defini tion 3.1. On Sx ,A x T X ,A we introduce a sesquilinear form by tA
<u,F>X
=
(e u,F(t»X'Note that this definition makes sense for t > 0 sufficiently small. Note
B.18
-tA also that because of the semi-group property and self-adjointness of e
the definition does not depend on the choice of t. We remark that <UO,F>
= 0
for all F € T implies Uo
=
0 and <u,FO>
=
0 for all u € S implies FO=
O. These two facts easily follow by taking F=
emb(uO)' respectively the denseness of emb(S) in T.
Theorem 3.2.
I. For each F € T the linear functional <u,F>, which maps S onto ~, is
con-tinuous in the strong topology of S.
II. For each strongly continuous linear functional 1 on
S
there exists G € Tsuch that t{u)
=
<u,G> for all u € S.III. For each v €
S
the linear functional <v,G>, which maps Tonto C, is con-tinuous in the strong topology of T.IV. For each stronglt continuous linear functional m on T there exists w €
S
such that m(F)
=
<v,F> for all F € T.Proof.
I. The function u ~ <u,F> is continuous on
S
iff for a l l t > 0 it is continuous in the Xt-norm when restricted to this subspace. IndeedII.
I
<u,F>I
=I
{e tAU,F (t)}I
s
[e tAu II IIF (t) II=
IIF (t) II lIull t • -tAFor fixed t > 0 the mapping e : X +
S
is continuous. Therefore the mappingtee-tAx) : X +
~
is continuous. From Riess' theorem follows the existence ofX
-tA
-tA
bt € such that tee x)
=
(x,btake G such that G(t) = bt• Then t(u) = <u,G> for all u E S.
III. Since T is metrizable it is sufficient to prove conti nul.. ty for sequences in
T. Let G + 0 in the strong topology. Then for sufficiently small t we have
n
fA
<v,G >
=
(e v,G (t» + 0 because G (t) + 0 in X-sense. IV.n n n
+
Take ~ E B • Take a sequence {x } c
X,
x + 0 in X. Define ~(A}x ET
b~-tA n An n
(~(A)x ) (t)
= ~(A)e x. Since ~(A)e-t is a bounded operator ~(A)x
+ 0n n n
strongly in
T.
By taking ~=
1 we conclude first that the restriction of m toX
has the Riess representation m(x)=
(x,a) for some a EX.
Secondly we conclude that m(w(A)x) is a continuous function onX
for eachW
E B+. Usingthe self-adjointness of w(A) it follows that a e D(~(A» for each
W
E B+. But then, by Theorem 1.10, a e S. Finally, asX
is dense inT
therepresenta-tion m(F)
=
(F,e)=
<e,F> is valid for all F ET.
0
Definition 3.3. The weak topology on S is the topology induced by the semi-norms PF(u)=
l<u,F>!, F eT.
The weak topoLogy on T is the topology indiced by the semi-norms
pv(G}
=
l<v,G>!, v e S.A simple standard argument, [CH] II § 22, shows that the weakly continuous
functionals on S are all obtained by pairing with elements of
T
and vice versa. Together with Theorem 3.2 it follows then that Sand T are reflexive both in the strong and the weak topology.In the next theorem we show that in both spaces Sand
T
weakly bounded sets are strongly bounded.Theorem 3.4 (Banach-Steinhaus).
I. Let ::: c
T
be such that for each g ES
there exists M > 0 such that for gevery Fe::: one has l<g,F>1 S M , then for each t > 0 there exists C t > 0 g
-such that for every FE::: one has II F (t)1I S C
t , So weakly bounded sets in T are strongly bounded,
II. Let 9 c S be such that for every F e
T
there exists ~ > 0 such that for every f E 9 one hasI
<f IF>I
S ~, ~ there exists, > 0 and c > 0 such that 8 c XT- and IIf II, < C for all f e 9. So weakly bounded sets in S areB.20
Proof.
I. From the assumption it .follows that for each h € X and each t > 0 there
exists a constant
~,t
> 0 such that for every F € E,l<e-t~,F>1
s
~,t
orI
(h,F(t)}I
s
~,t' From the Banach-Steinhaus theorem in Hilbert space it then follows that the set {F(t)I
F €E},
t fixed, is a bounded set inX.
II. Let tjJ E: B+. Let w E D(tjJ(A». Then tjJ(A) 2 w, defined by (l/J(A)2w) (t)
=
2-tA
== tjJ (A) e w belongs to T. From our assumption it follows that for every
f E:
a
From the Banach-Steinhaus theorem in Hilbert space it then follows that
a
is a bounded set in the Hilbert space XtjJ' i.e. the completion of $ with respect to the norm IItjJ (A) • II. But this means thata
is bounded in $, since each semi-norm PtjJ' tjJ € B+, is bounded on it.In the next two theorems we give characterizations of weak convergence of sequences both in
S
andT.
Theorem 3.5. u + Q in the weak topology of $ iff n
3t>Q {u } c
X
n t
tA
Proof. Weak convergence of {un} in Xt means weak convergence of e un in X.
-) Let F E:
T.
<u ,F>=
(etAu ,F(t}). Since etAu + 0 weakly inX
andn n n
F(t) €
X,
it follows that <u ,F> + O.n
o
->
First we remark that weak convergence in $ implies weak cORvergence inX.
Therefore for any w €X
any L > 0, t > 0,+ 0 as n + co •
{u } is
n a bounded set in S. Therefore 3'£>0 39>0 Vn€lN lIu n '( II s
e.
We shall CII)prove that u -,)0 0 weakly in X • Denote
n A T ~, llL commutes with e'( • Taken w E:
X,
the projection operator Lf dEA by then
TA
We have IIL e un S
e
and II IIL wll -+ 0 as L -+ QO. Therefore if we take L largeenough the second term in (*) is smaller then ~€ uniformly for all n. As we have just seen the first term in (*) can be made smaller than ~€ by taking n large enough. This finishes the proof.
COroll¥X 3.6.
I. Strong convergence of a sequence in
5
implies its weak convergence. II. Any bounded sequence in5
has a weakly converging subsequence.Thearem 3.7. F n -+ 0 in the weak topology of Tiff
Proof.
-tA
.) For any v to:
X,
<e v,F>=
(v,F (t)}O -+ 0 as n -+ 00.n n
.) For any , t
5
and t sufficiently smalltA
because e . ~ EX.
COrollary 3.8.tA
=
(e ~,F (t» -+ 0 nI. Strong convergence of a sequence in T implies its weak convergence.
II. Any bounded sequence in T has a weakly convergent subsequence. Cf. Theorem 2.9.
It looks reasonable to conjecture that the weak topology on
S
is theinduc-t i ve limit topology with respect to the Hilbert spaces
X
t , now endowed with the weak topology. It is easily seen that the weak topology on T is induced by the semi-norms Pt , t > 0, V €
X,
and Pt (F)
=
I
(v,F(t})I.
We will not,v tV
pursue these things further here. The next theorem deals with the question: When does weak convergence of a sequence imply its strong convergence?
o
I.
II.
III.
B.22
Theorem 3.9. The following three statements are equivalent. For each t > 0, e -tA is a compact operator on X.
Each weakly convergent sequence in S converges strongly in S. Each weakly convergent sequence in
T
converges strongly inT.
Proof.I - II. A weakly convergent sequence in S converges, for some t > 0, weakly in Xt • See Theorem 3.5. Because of the assumption the natural injection X
t c Xa, 0 < a < t, is compact. Cf. the proof of Theorem 1.11. But then our sequence converges strongly in Xa'
II - I. Take any sequence {f }
-tA
neach F €
T
we have <e f ,F> + O.c X, f + 0 weakly in X. For each t > 0 and
n -tA
-tA
nThus e f + 0 weakly in S. Because of
n -tA .
the assumption e f + a strongly in S.
-tA
n And so lie fII + a. This shows n
that e must be compact.
I - III. Let {F } c
T.
Suppose VgES <g,F > + O. Thenn n
Vh€X Va>O
<e-a~,F
>=
(h,F (a» + 0 •n n
This means that Va>O F (a) n B > 0, we find that F (a + n Vt>O IIF (t) II + a. n -SA
+ a weakly in X. Using the compactness of e ,
B)
=
e-B~
(al + 0 strongly in X. Thereforen
III ... I . Let {v } c X
n be such that v n + 0 weakly in X. Then v n + 0 weakly in
T
because Vg€S <g,v >n :: (g,v )
-tA
n + O. Now the assumption says v n + 0 strongly in T. This means Vt>O e v + 0 strongly inX.
The compactness ofn
follows for each t > O.
-tA
e
o
-tA
One might wonder whether in the case that e is compact for each t > 0 the strong and weak topologies coincide. This however cannot be true since a set which belongs to a weak base of a-neighbourhoods always contains a closed subspace of finite codimension. Generally speaking such "large" sets do not fit in a strongly open O-neighbourhood.
B.23
It is relatively simple to show that both in $ and
T
the weakly compact sets are just the (weakly) closed and bounded sets. Then i t is not difficul t to prove that $ andT
with their strong topologies areMackey spaces.
That is the strong topology is the finest locally convex topology for which the dual of $ (respectvelyT)
isT
(respectively $). (All spaces which are either barreled or bornological are Mackey. See [SCS] IV.4.)REFERENCES. See Part A.
ACKNOWLEDGEMENT The author thanks Dr. S. J • L. van Ei j ndhoven for his contribution to the revision of the manuscript.