operators
Citation for published version (APA):
Eijndhoven, van, S. J. L., & Kruszynski, P. (1984). On Gelfand triples originating from algebras of unbounded
operators. (EUT-Report; Vol. 84-WSK-02). Technische Hogeschool Eindhoven.
Document status and date:
Published: 01/01/1984
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 EN INFORMATICA
EINDHOVEN uNIVERSITY O~ TECHNoLOGY
THE NETHERLANDS
DEPARTMENT OF MATHEMATICS AND COMPUTING SCIENCE
On Gelfand triples originating from algebras of
unbounded operators
by
S.J.L. van Eijndhoven and P. Kruszyflski
EUT Report 84-WSK-02 ISSN 0167-9708
UNBOUNDED OPERATORS
by
S.J.L. van Eijnhoven and P. KrUszynski*)
*)On leave from the Department of Mathematical Methods in Physics, University of Warsaw, Poland.
SummMY
Il1tJtodu~on 1..
The..
.6pac.e..
SA
~c)
2.The..
.6pac.e
T A
~ ( )3.
The..
p~ng
06
s~CA)
and
T~(A)'
thein duaiity
4.CoWnUOU.6 ,une..aJL mapp,Ln9.6
n
Jtom
SA
"erl-to
SCA
<P1C 1) 4>2 2)
and
n
Jtom
T
A)
"enio
T
A
~1( 1 <!>2( 2)
5.
A
.6UJtve..y
06
Attan'.6
~heoJtyon GB*-alge..bJta.6
6.The.. aiqe..bJta.6
4> (A) Cand
<!> (A) cc7.
A
6unmonat c.alc.td.u6 60Jt the atgebJta
<!>CA)
cc8.
lUu6.tJraUoY/..6
1 9 23 34 41 4752
61 80 87triple S¢(A) c X c family ¢ there are
T¢(A) where S¢(A)
=
U ¢E¢ certain GB*-algebras of¢(A)
(X). Related to the unbounded operators. They extend the cornrnutant ¢(A) I and bicornrnutant ¢(A)" in B(X) in a naturalway. Further, some of them give rise to a description of the inductive limit topology for S¢(A) in terms of explicit seminorms.
- 1
-IYLtJtoductioYl.
A Gelfand triple consists of three ordered spaces
SeX c
T. In thistriple X is a Hilbert space,
S
is a dense subspace of X which carriesa locally convex topology, and T is a locally convex topological vector
space which is in duality with
S.
Many mathematicians have investigatedthis kind of space triplets. We mention Gelfand and Shilov who in fact introduced the notion of Gelfand triple in their illuminating work on
generalized functions ([G5J), and, also, Grossman and Antoine ([AGJ)
who generalized the ideas on countable Hilbert spaces towards the so
called scaled Hilbert spaces. In fact, any distribution theory is based on
some Gelfand triple.
In this paper we continue the investigations which are initiated in [GJ and [ElJ. In those papers two functional analytic frames are created
for two different types of distribution theories. In [GJ, the test~
space is introduced as an inductive limit and the distribution space as a Frechet space. In [ElJ the test space is a Frechet space and the distribution space an inductive limit. In order to clarify the intentions of the present paper we give a rather detailed outline of the theory [GJ.
In [GJ, the Gelfand triple
has been introduced and thoroughly investigated. Here
A
denotes apositive self-adjoint operator in a separable Hilbert space X. The
set
SX,A
is the analyticity domain of the operatorA,
i.e.U e-tA(X)
It is natural to endow SX,A with the inductive limit topology 0
ind
-tA tA
generated by the spaces e (X) with norm lie .II X.
The space TX,A consists of all mappings from the open interval
(0,00)
-TA
into X with the property F (t + T) = e F(t), t'L> O. The elements of TX,A are called the trajectories. The topology T
proj for TX,A is the locally convex topology generated by the seminorms
F>+ IIF(t)lI, F
E:
Tx,A ' t > 0 .It has been proved that Sx,A and TX,A are in duality.
In [G], B(R) denotes the set of all everywhere finite Borel functions f on R which satisfy
sup If (;\) e
-t>..
I
A>O< 00
B (R) denotes the positive part of B(R) •
+
Following the spectral theorem we can define the operator f(A) for each f
E:
B(R)The operator f(A) is unbounded and normal.
For each f E B+(JR) the mapping Sf : SX,A 1+ JR+ defined by
IIf(A)wll , w E S
X,
A
is a seminorm on SX,Ao The seminorms Sf' f ( B+(JR) , generate a locally convex topology G . for S In
[G]
i t has been proved that a ispro] X,A" ind
('quivalent to () .. But S . also equals the intersection
pro]
X,A
n
D(f(A»fEB
(R) +3
-Thus i t can be seen as a projective limit (D(f(A)) denotes the maximal
domain of the unbounded operator f(A). See also [LTJ).
Correspondingly, for each trajectory F E TX,A there are 9 E
B+(E)
and x E X such that F(t) = g(A)e-tA
x,
t >O.
So we haveU g(A)tX)
gEB+(JR)
where g(A) (X) denotes the set of all trajectories F
g,x
where
x
E X. With the inner product-tA
t -+ 9(A) e x ,
g(A) (X) becomes a Hilbert space. The set B (JR) is a lattice. So i t
+
makes sense to introduce the inductive limit topology T. d for In
T A
=
U g(A) (X). From the results in this paper i t follows thatX, 9 B+(JR)
T . is equivalent to T
pro) indO
Since the theories [G] and [EJ are a kind of reverse copies of each other we give only a short summary of the latter. In [EJ, the Gelfand triple
T(X,A) c XC o(X,A)
has been studied. Here T(X,A) D (e), Le.00
A
T(X
,A)
seminorms
The space T(X,A) is a projective limit with topology induced by the
tA
wt+ lie wll
xt
t > O. We observe that T (X,A) is a Fr~chet space.The space 0(X,A) is introduced as
tA
Here e (X) denotes the completion of X with respect to the norm
-tA
lie
.II
X. Hence o(X,A) can be seen as an inductive limit. In [EJ the set of functions F(R) has been defined as follows
¢ E F(R)
So the operators
¢(A)
are bounded (even smoothing). We have the f01-lowing relations T(X,
A)
and o(X,
A)
U</>(A)
(X) ¢H(R)n
X
</>EF (R) ¢Here X</> denotes the completion of X with respect to the norm
II
¢(A) ·11 •
These relations lead to a description of T(X,A) as an inductive limit and o(X,A) as a projective limit. We observe that each element of o(X,A) is in unique correspondence with a mappingF
F(R) ~ X with theproperty F(¢lji) == ¢(A)F(lji) , ¢,ljJ EF(R) . To see thi,s, write formally
tA
F == e y , with t > 0 and y C X and then define
-
tA
F(¢)
== e¢(A)y,
¢E F
(IR)Both in [GJ and in [EJ a central role is played by a semigroup $ of functions which generate the test space in the Gelfand triple as an inductive limit. In particular in [GJ
- ~)
-{cjl \ cjl Borel,
Vt>O
supxE:IR
The distribution space can be seen as the space which consists of all mappings F from ~ into X with the property
F(cjllj!) '" cjl(A)F(I(I) 1(1 (A)F(cjl) cjl,lj! E~.
Related to the function semigroup
~
is the set~*
which consists of all everywhere finite Borel functions with the property thatsup \f(A)cjl(A) \ < 00 for all cjl
E
~.
Each function f (~*
defines anA(JR
unbounded (normal) operator f (A) in X. The seminorms sf : W 1-+
II
f (A)wllinduce a locally convex topology equivalent to the inductive limit topology for SX,A (T(X,A».
'J'hi~; new vi,t'W on tll<' yeneration of tll'~ C!'lLmd I.cipll'~; 5
X
,A
c X c IX,A and T(X,A)c
Xc
o(X,A) inspired us to look for conditions on a set of bounded Borel functions ~ which lead to a generation of Gelfand triples with similar properties. The conditions we impose on~, reflect on the properties of the Gelfand triples it generates.The triples occuring in this paper are denoted by
Here S~(A) '"
u
¢(A) (X) with inductive limit topology, and T~(A)· ¢E~consists of all mappings F from ~ into X such that F(¢lj!)
=
¢(A)F(X)=
I(I(A)F(¢), cjl,lj!E
~.The set
~*
consists of all everywhere finite Borel functions with the property that sup \f(A)¢(A) \ < 00 • The elements of~*
induce seminormsA(:IR
on S,]J (A) and hence a locally convex topology. This topology is equivalent to the inductive limit topology. Also, we impose conditions on ~ such
that each member of T~(A) can be represented as F(¢)
'"
¢ E ~, for some x E X and f
E:
~•
We note that the inductive limit S~(A) is not strict, in general.
The sections
1-4
contain a detailed discussion of the topologicalaspects of the spaces S~(A) and T~(A)' their duality and corresponding
algebras of continuous linear mappings.
In the sections 5-7, the space S¢(A) is used as a core for certain
related algebras of closable operators. Here we consider so called GB*-algebras. The notion of GB*-algebra has been introduced by Allan.
(ef. [Ai 1] and [Ai 2]). It is a useful generalization
of the usual notion of B*-algebra. A summary of Allan's theory is given in Section 5.
In the sections 6-7 we study the algebras <p(A)c and
~(A)cc,
whichSx
,
A
c X c-tA I
{e t >
to the triple
arc subalgebras of L (S,p (A))' (See also [AET].) Here
~
(AJ
c consistsof all operators L E L(S<p(A)) with the property that L¢(A)w
=
¢(A)Lwfor all
¢ E
~
and all wC
S¢(A)' Similarly, L'~
¢(A)cC if and only ifL'Lw = LL'w for all L
E:
~(A)c
and all wE:
S~(A)'
The algebras~(A)c
and ~(A)cC have roused our interest in earlier research with respect
TX,A' We have proved that both
<jJG(A)c~
{e-tAlt > o)c ccO} are GB*-algebras. Moreover the inductive
limit topology for SX,A is generated by the seminorms
\ (w) = IILwll
X ' vJ E
s
X,A
where L
E:
¢G(A)CC. In this paper both aspects of ¢(A)cC will be taken7
-c cc
In Section 6 we prove that ~(A) and ~(A) are GB*-algebras. More-over
~(A)cC
is commutative. The algebras~(A)c
and w(A)cC are extensions of the usual commutant ~(A) I and bi-commutant ~(A)" which areC*-sub-algebras of
B(X).
We haveand
The Gelfand-Naimark theorem for ~(A)" extends to the whole algebra
cc
In Section 7 we develop a functional calculus for ~(A) . In a natural way the set of functions ~ generates a B*-algebra with the usual supre-mum of bounded functions as its Banach norm. Next we consider monotone sequences of functions in the real part of that B*-function algebra. We obtain a so-called Borel*-algebra W~(~) of bounded Borel functions on ~. We prove that
W<!>(A) := {8(A)
10
E W<l>(]R)}=
1>(A)" •A natural extension of
W~ (~)
is the function algebraW~* (~)
which consists of all everywhere finite Borel functions with the property that for all <p E: 1> the function <p.f CW~ (~) . There exist conditions on ~ such thatelements
Q
of ¢(A)cC induce seminorms S on S A) which generate theQ ¢(
inductive limit topology for S¢(A) .
The results of the last part of Section 7 are derived from non-trivial measure theory concerning Souslin spaces and Borel spaces.
9
-In the first sections of this paper we intend to describe a Gelfand
triple S
cjJ(A)
cX
cT
cjJ(A). Let us first explain the meaning of the
indicated symbols.
X
denotes a separable Hilbert space which is
taken fixed throughout the paper, and A denotes an (unbounded)
self-adjoint operator in
X.
The partially ordered set of functions cjJ is
defined as follows.
(1.1)
Ve6inLtion
The set
cjJof generating functions on
lRis a p.o. set with respect to
the usual ordering of functions. Moreover it has the following
prop-erties:
A
I
Each
¢
E
<jlis a nonnegative Borel measurable function bounded
by
1-A I I
V
¢E<Jl
=1-1jJE<Jl
:4>1
,,;:; 1jJA III V V, ::J- :=I
.pE'l>
l)E:lR
XH c>OA IV
V
¢E:<Jl 3ljJ EcjJ 3
c>O,\ C
lR
.
Let
(EA)AClRdenote the spectral resolution of the identity corresponding
to A. Then for any
¢E
cjJthe operator <jJ(A)
=
J
<jJ('\)dE,\ is bounded and
JR
even "smoothing".
(1.2)
Ve6inLtion
Let ¢ E ¢. Then the subspace ¢(A) (X) of X becomes a Hilbert space if we introduce in ¢(A)X the following inner product
-1
<u,w>¢
=
<¢(A) u, u,w E ¢(Al
(X)-1
where
ep(A)
denotes the unbounded self-adjoint operatorf
¢(A) - l dE\supp(¢)
(supp(<P) = {~EJR
I
¢(A) > O}.)Now we define the space Sr[)
(Al
as follows(1.3) Ve6~nition
S
(I) (A)
where the topology 0
ind on S¢(A) is the inductive limit topology
generated by the spaces ~(A) (X) .
We note that the inductive limit topology is the finest locally convex topology for which the inj.::,ct:Lons
it/> :
(4)CAl (X) C+ S<p(A) are cont.inous.Hence a set V c S<p(A) is open iff V
n
¢(A) (X) is open in the Hilbertspace ¢(A) (X) for all ¢ E ¢. Because of A III and A IV in Definition
(1.1), the space S¢ (A) is Q dense A-·invar iant subspace of X.
Together with the set 1) we introduce the set of everywhere finite
- 11
-sup
I
f (A)¢(A)I
< 00 •AElR
We note that
~*
is a function algebra byA II.
Let f E
~*.
Then f is associated to the normal (unbounded) operator f (A) ,Because of the definition of
~*,
the operator f(A)¢(A) is bounded on X for all <p E Qi.(1.5) Ve..M~on
Let f f
¢*.
The seminorm sf on SQi(A) is defined byIIf(A)wll ,
where
II· II
denotes the norm of X.We observe that sf is well-defined on S¢(A).
( 1.6)
The..ahem
1. The seminorms Sf' f E l'
*
, are (lind-continuous on Sl'(A) .II. Let a convex set
n
C SQi (A) be such that for each <p f Qi the set~
n
¢(A)(X)
contains an open neighbourhood of 0 in ¢(A) (X). Thenn
contains a setI. Prom the inequality
the continuity of sf follows.
wE cjJ(A) (X) , II. Put P n n
f
dE>..
+ n-l -n+lJ
dE,\
-n
n E IN . For each n E IN we define r := sup{p np
n(X) A II Pulln < p ] => u F Pn(Q)}From A III i t follows that r is well-defined and non-zero. The n
function f is defined as follows
f (A) 2 2
max(~-
,1) nI>..
I ( [n - 1, n) . ::F We shall prove that f E: <!> •So let cjJ
E:
<!>. Then there exists X ( <!> and c > 0 such that for all nE:
IN 2 n sup ¢(I.) i>..
I
([n-l,n) Lnf!
Ai
Eln-l,n) ( A) •To see this we observe that following axiom A IV 2 n sup ¢(A) i
>..
I([n-l,n) < c 1n 2for some well chosen
Xl E:
<!> and c1 > O. Applying A IV again, the existence of the indicated X E <!> follows.
- 1 'l
-J
-2 2{u E X(A) (X)
I
'
x
Pc)
d ( EAU,u) < E: } c Qn
X(A) (X)supp(x)
because
Qn
x(A) (X) contains an open neighbourhood of 0 in x(A) (X).
Thus we get that r
> E:inf
X(A) for all n (:
l-lwith
n
IA/EEn-l,n)
supp(~)
n
(En - l,n)
U
(-n,
-n +lJ)
~
0.
Hence
by+ for all
1..,11..1
EEn - l,n),
2
n
f(A)¢(A)
< 2(sup
¢(A))max{-- ,
1}<
IAIEEn-l,n)
r n
n2 -1
<
2(sup
~ (A) ) (- ( inf
X(Ie))+
1) < 2(~ +
1) •!A!H n -l,n)
E:!A!r::[n-l,n)
E:We next show that
(*.)
Ilf(A)ull
< 1 ~u
E:
Q •Let
u
E:
'HA)
(X)for some
,~
c: '\'.
Then
~
liP
ull
2 < ,"and
n=l
n
¢
(**)(~(A)-lp
,u,p u)<
n n Xsup
(¢(A))
I
AI
Hn-l
,n) 2lIulle!>
Because of our assumption
(*)for all n
E
R
we have
2n
2P
uE
Q.n
Now we express u as follows
U =
N 1 00
I
(2n
2
p
u)
+ (I
__
l__)u
where
1 -1
( I
~-)j=N+l 2j2
By (**) and by A II of Definition (1.1) there exists a function ~ E ¢
such that
00
~
4N4I
(¢(A)-l p u,p u)xn=N+l n n
4 _ 2
~ 4N sup (¢(A)) (~ lIuil
IAI~
N¢
2
Since lIu II, -+ 0 as N-+ 00, we obtain uN EQ
n
~(AJ(X) for sufficientlyN 1jJ
2
large N. Moreover for all n E ]['1, we have 2n
P
u Eli. Hence u is an
subconvex combination of elements in the convex set li. So u E Q. It is
clear that u
<:
Qn
¢(A) (X).(1.7)
The-altern
The locally convex topology generate.s by the seminorms sf' f ( ¢* is
equivalent to the topology Ii.
1 f · '
**
1 L ib natura to de lfle c>'
o
x
(.p**:~
X is a Borel function such that V*:
supI
f(A)X(A)I
f f::<t
HlR
- 15
-In the remaining part
of this section we assume the following extra condition on the lattice ¢.(1. 8) A V
Axiom A V gives rise to a description of bounded subsets and null
sequences in S¢(A) as for strict inductive limits, and also to the
proof of its completeness. With emphasis we note that in general the
inductive limit S¢(A) is not strict. But nevertheless the following
results can be proved.
(1. 9)
Lemma
Let
Be X. B
is a bounded subset of ¢(A)(X)
for some ¢E
¢ iffBe
D(f(A))and the set {f(A)xl x E B} is bounded in X for all f E:
<!J*.
*
~) Suppose
B
is a bounded subset of ¢(A)(X).
Then for all fE:
¢ andall wEB
Hence each set {sf(w) Iw E B} is bounded.
=I:
<:) By assumption for all f
E:
¢ there exists Kf > 0 such that V
wEB IIf(A)wll ~ Kf
Now define r :0= sup
II P
wll , where againP
n wEB n X n
define the function X on lR by
n
( f
n-l -n+l +f )
dE>..
-n , andx
(A) nrn for
I
AI (
[n-1,n) .We prove that X E
¢* .
So let f E¢ .
Then we observe thatf
definedby
f(A) n sup
I
f ( A)I '
IAI([n-1 ,n)
I A
I
C [n - 1,n) ,is also a member of ¢ . Indeed, this follows from the estimation
sup I
f (
A)<p (A)I
AfJRsup n ( sup I f (A) I ) ( sup <p (A» ~
nON , AIC[n-1 ,n) IAIE[n-1 ,n)
<
K sup ( sup (I
f (A)!liJ(
A) )neJN I A! EJn-1 ,n)
K sup If(A)\j!(A)
I
< 00 •AOR
The existence ofK > 0 and l/J f ¢ is derived from A IV.
Now we proceed as follows. Let n E :N and A (JR with
I
AI ~ [n - 1 ,n) •Then
I
X(A) f (A) I~
nr supI
f (A)I
nI A!E[n-1,n)
SCll)
ilf
v,CEl**
Hence X E ¢ . By A V, there exists <p f ¢ and c > 0 such that X~ c·<p. It follows that for all wEB.
n c
2
I [__
1
r-2 (
r
+ 2 nJ
n=1, n n-1 r1-0
1:1. \ 1 I !_, p. 1 n -n+1f
)d(EAw,w)]-n
~o B is a boundc~- 17
-(1. 10)
TheM_em
Let
~fulfil
also axiom
A V.
A
subset
BCS(jJ
(A)is bounded iff there exists
~a bounded subset of
~(A)(X).(1.11)
Co~ottahYf
(jJ such that B is
Hence S
-
n
(jJ (A) - f("(jJ*
+Let u
EX. Let u
E
S(jJ(A) iff u
E
D(f(A»
for all f
E
(jJ:.
D(f(A»
.
In the previous result we have introduced the notation (jJ*
+'The subset
<\J*
Ci\l* consists of all positive functions in i\l* which are larger than 1.
+
It clearly makes sense to impose
generated by the spaces D(f(A»,
on S(jJ(A) the projective limit topology
f E <p*. It is not hard to see that
+
D(f(A»
is a Hilbert space with the inner product
(u,w)f
(f(A)u,f(A)w)
,
u,w (" D(f(A»
We denote this projective limit topology on S
A) by
0 ,.From the
(jJ (
pro)
previous results and Theorem (1.6) it follows that the topologies
o. d
~n
and
G,are equivalent.
pro)
( 1. 12)
Lemma
Let
Be
S(jJ(A) be a bounded set containing null. Then there exists
~E (jJ
such that B is a bounded subset of
~(A)(X). Moreover, a net (wa)aEI
cB
converges to null in S(jJ(A) if
IIwall~ -+O.
~)
Let (w )~I
be a net such thatUw
U~ ~
O.
Then for all fE
¢*,a a _ (j. 'j'
IIf{A)w
II
~IIf(A)rjJ(A) IllIw
II~ -+ 0 .(j. a 'j'
~) Since the set
B
is o. d-bounded, there exists ¢E
Q such thatB
In(*)
is bounded in the Hilbert space
<p(A) (Xl.
Let s >O.
Then there is N E IN such that [sup (~( A) 1I
A13N
2
< r . It follows that
[sup (rjJ(A» + s2]
IAj;?N
where the positive constant KB depends on
B.
-~
*
since A ,-+
CPU)
"EXt ) (A) + X{I ( )'
2,U) ] E
(jl -N,N 11 <p p ,?cJ Moreover, there exists aO
( I
such that (**) <pU..) d(E;\wa,w a}1
-1 2 < € [IA<NJ
1\ [¢ 0.)>s
2 ]for all a > 0'.0. Thus i t follows that for all a > a O
By A II there exists ~ ( ~ such that
IIw
II
< [. (1 +21',:,
- 1<)
-From the above lemma the following result
is
immediately clear.( 1. 13)
TheM.em
Let <P fulfil also axiom A V.
Any bounded subset B of S¢(Al is homeomorphic to a bounded subset of X , where a suitably chosen
¢
E ~ establishes the homeomorphism.(1.14l Co~ott~y
I. A subset K of S~(A) is compact iff there exists ¢ E ~ such that K is a compact subset of ¢(Al (X). Hence a subset K
c
S~(A) is compact iff i t is sequentially compact.II. A sequence (unlnON in S~(Al is a Cauchy sequence iff there
exists ¢ ( ~ such that (unlnEIN c ¢(Al (Xl is a cauchy sequence in ¢(Al{Xl. Hence S~(A) is sequentially complete.
In the following theorem we use the standard terminology of locally convex topological vector spaces in order to make a link to the general literature on this subject.
(1.15) The.o~em
Let <f> fulfil also axiom A V.
I .
S~(A)
is completeII.
ScHAl is barreled I I I .
IV. S~(A) is Montel iff each operator ¢(A), ¢ ( ~, is compact as an operator from X into itself.
V.
S~(A)
is nuclear iff there exists f E~:
such that f(A)-l is a nuclear operator from X into itself.I. Let (wa)afI be a Cauchy net in S~(A)' Then (f(A)wa)aEI is a
*
Cauchy net in X for each f E ~ . Since X is complete there
+
exists w(f) E X such that (f(A)w ) ( converges to w(f). Take
a a I
,rCA)
=
1, A E JR. Then ,T(A) is the identity on X, and w -+ w(~)a
in X-sense. Since the operators f(A) are closed, one has
w(~) E D(f(A)) and w(f) = f(A)w(~r). The result follows by
applying Corollary (1.11).
II. Let
V
be a barrel, i.e. a radial convex circled and closedsubset of S~(A)' Then
V
n
¢(A) (X) is a barrel in the Hilbertspace ¢(A) (X) for any ¢ E~. SO there exists an open null
neighbourhood L¢ C V
n
¢(A) (X). Then following Theorem (1.6.11),V contains an open set {wlllf(A)wll < s} for some f E
~*
and E: > O.III. Every circled conV2X SUbset. ~! c: S, " d',at abso:ct>s ("very bounded
y (Ai
subset Wc S~(A) is to be a neighbourhood of
O.
LetB¢
denotethe unit ball in ¢(A) (X), ¢ E ~. Then clearly one has E:B¢ c ~
n
X¢for some E: > O. Again applicat lon of Theorem (1.6.11) lectds to the
- L1
-IV.
~) Consider the following scheme.
x
ljJ (A) ) X" (Ai
2j
j"
(AI
1/I(A)2{X) C r 1/1 (A) (X)
From this scheme the following is clear. Let B be a closed and bounded set in S~{A). Because of Theorem (1.10) and axiom A II there exists 1/12 E
~
such that B is closed and bounded in ljJ{A)2{X). Since~(A)-2B
is closed and bounded in X and since 1/1 (A) is a compact operator from X into itself, it thus follows that B is a compact subset of ljJ{A) (X).<=)
V.
~)
Let (u ) (n n -:IN be a bounded sequence in X. Then (1/1 (A) u ) elN is an n,,-bounded sequence in 1/1 (A) (X). SO the closure of the set C
C = {1/1 (A) u
I
n ( IN} nis compact and hence sequentially compact in S~{A). It follows that the sequence (1/I{A)un)nE:IN contains an S~{A)- and hence X-convergent subsequence.
*
-1Suppose there exists f E ~ such that f(A) is nuclear. Then the +
operator A has a discrete spectrum because f- 1 {O(A» = O{f- 1 {A» by the spectral theorem. Let {ej)jClN denote the corresponding
orthonormal basis of eigenvectors and 0,.). its eigenvalues. J ]ON
::j: ~
Let y ( ~+ and let Sg denote the completion of S~(A) with respect
to the seminorm Sg. Put h
=
f.g. Then the injection Ih,g : Sh ~Sgis nuclear, because 1
w
~ h,g)~
j=1 -1 - 1 - 1 EO.) (h(A)w,h(A)(h(A.) ('))<j(,\.) \'. J J J ] ]-1
We note that (h (;..) e ) 1.S an orthonormal basis in Sh and
j j JON
-1
-(g(;".) e.) is an orthonormal basis in S .
] ] g
~) Let S~(A) be a nuclear space. Take g =~. Then there exists f f ~::j: such that the mapping
I
:
Sf y..X-1
23
-Since the p.o. set
¢consists of Borel functions bounded by 1, we
may as well suppose that ¢ is closed under pointwise multiplication.
Then it makes sense to introduce the following space.
The space T¢(Al consists of mappings
F:¢
+X
satisfying
The projective limit topology
Tproj
on T¢(Al is the locally convex
topology induced by the seminorms
The Hilbert space
X
can be embedded into T¢(Al. To this end, we
define ernb ;
X
+T
¢(Al by
ernb(xl
(~) = ~(A)x,
x E X,
~E
<P.( 2 • 2 1
T he.M,em
emb
(X) is adense subspace of T
A .(', ( 1\)
Put
Since
x
(-n,n) ~ c .<p •J J
for suitably chosen c. >
a
and 4>. (: <P, we derive thatJ J
-1
c.<p.(A)) IT ]
] ] n c.F(¢.)] ]
EX.
So i t is clear that G , defined by
n
G
n <p t+ IT F(¢)n
is an element of emb(X), for each n
E
IN. Moreover, for all¢ (
¢IIG (
¢ ) - F (<p)II
=:II
or -
I) F ( <p)II
-+0 ,
n
n
*
Let f E <l> , and let x EX. Then fox defined by
n -+ 00
o
(fox) (¢) ¢ E ¢
is an element of
T
¢ (A). As a conseq'le:nce of the first part of the nexttheorem i t follows that any element of T¢(A) arises in this way.
We introduce the following notation. Let V C XI then foV =: {fox/xC V}
(2.3)
The.oJtem
I.
A
set B C T¢(A) is bounded iff the set {F(<p) IF ( B} is bounded inX
for all <pE
¢.is bounded iff there exists f
II. A set B C T¢(A)
set So c X such that B =: foSO.
*
25
-Observe first that by axiom
II
~)As usual we take P
n
The proof of I and II
~)is trivial and it is omitted.
n
-n+1
(f
+f
)dEA•
n-l
-n
-1A IV there exists
~E
¢ such that
~(A)P
n -1
is bounded on
X.
Put PnF =
~(A) PnF(~)for each F
E
T¢(A). Then
P F
E Xand r
;= sup liP FII
~ II~(A)-lp
II sup
IIF(~)II
< 00n
n
FEB
n
n FEB
Define the Borel function f by
f(A) =nr
n
IA[
E [n - l,n)
,
nElN.
We prove that f E ¢*. To this end notice that for all
A, IAI
E [n - l,n)
and for each
~E ¢
nr
~ (A) ~nr
sup
(~ (A» ~c sup II P F
(X)II
.
n
n
I
A:EI
n
-1,n)
FfB
n
Here
X
E ¢ and c
>0 have to be chosen as indicated in axiom A IV. The
'*
choice does not depend on n E IN. It follows that
fE ¢ .
Now let FEB and put
00
P F
I
n
XF
nr
n=1,
n
r
r!O
n
Then we have
00liP FII
2
00I
n
~I
12 2
2
n=l,
n r
n=l n
r
r!0
n
n
Hence x
F
(
X
and the set B
O
{XF!F E B} is bounded in
X.
It is clear
We can replace ¢* in the above theorem by ¢*. +
(2.4)
Lemma
Let B be a bounded subset of T¢(A) which contains zero. Then there
exists 9 E ¢* with the following properties
+
I. There exists a bounded subset
8
0
cX
such that B = gOBO
.
II. Let (F ) E- be a net in B. Then (F ) c converges to zero iff
(J. a _I a (J.cI
F gox , a E:' I and
II
xII
+o.
a
a
a
In virtue of Theorem 2.3 there exists f E
~*
and a bounded subset+
B
1
cX
such that B = foB1
•
Put g(A)2 2 -1
(1 + A ) f (A) and put B0
=
(1 +A)
B1 .Then clearly B gOB
O·
For each a E I there exist xaE
Band y0 aE
B1
suchthat F
a
foya goxa .2 -1
Now let f: > 0, and take N E IN so large that (1 + A) < € for all A, with
(* )
IAI ~ N. Then we obtain
f
d(EAxa,xa)IAI~N
2 (; Ef
d(EAya,y a) (;IAI~N
Next observe that followi'l:) dxi2El p, III dTlU .A PI l:her~; ,"'dst" constants
c. > 0 and functions ¢. E ~ such that
J J -1 g(A) X(-N,N) (\) (; We thus derive ":J=1. ~;-6 .
,
i (** )r
~
d(E,x,x) <c, /\ a a IA <N L jII. A
- 27
-2 Hence there exists a
O
E I
such that A<N5
d(EAxa IXa) < E as soon asa > aO. Together with
(*)
we getfor all a > aO.
Now we obtain by the previous lemma.
(2 • 5)
TheM.em
Each bounded subset B of T~(A) is homeomorphic to a bounded subset
of X. The homeomorphism is established by a well chosen 9 E
~:.
(2.6) COJtOUM!1
I. A subset K of
T~(A)
is compact iff there exists f (~:
and acompact subset V of X such that K
=
f-V. Hence K c T~(A) iscompact iff i t is sequentially compact.
sequence (Fn)nElN c T~(A) is fundamental iff there exists
*
f E ~ such that (F ) c is a Cauchy sequence in foX.
+ n n,-lN
Hence T~(A) is sequentially complete.
(2.7)
TheoJteJn
The space
T$(A)
with topology lproj is complete.net in X for all ¢ E <P. Hence there exists x¢ ~ X, ¢ E <P , with Fa (,~)+ x¢.
Now put F : ¢ 0-+ x<p, <P (<P • Then
lim F (¢oljJ)
a a ¢ (A) lim Fa a (ljJ) ¢ (A)F(ljJ) .
Hence F E T¢(A) and the space T¢(A) is complete. o
In the remaining part of this section we assume that the lattice ¢
satisfies axiom A V, i.e.
::J
c" 0
x
~ c¢The function space ¢* consists of all functions in ¢* which are +
larger than 1. It can be seen easily that (¢*)* = ¢**.
+
*
Let f
E
<P+.
We recall that foX denotes the subspace of all F f T¢(A)for which there exists x ( X such that F(¢) = f(A)¢(A)x , ¢
E
<P.The space fo X becomes a Hilbert space with the inner product
(2.8) (x,y) F,'G (.-= £
X
-1 -1
f(A) F and y - f(A) G iff fox F and goy G.
From Theorem 2.3 we obt2in
- 29
-It IIIdkl':; :;('11:;(' LUlnlruducl' III<' l.ndIICLi.vl.' .limit I Ul)()J(Hlyl , d ull
~n
TW(A)
generated by the Hilbert spaces foX. So a set VC T~(A) is:j:
open iff
vn
foX is open in foX for all f f W . + The following theorem is similar to Theorem (1.6).(2.9)
The-ahem
1. The seminorms t¢1
¢
~W,
are continuous with respect to the inductive limit topology Lind on T~(A)'*
II. Let a convex set ~ c
TW(A)
be such that for each f E ~+ the set ~n
foX contains an open neighbourhood in foX. Then ~ con-tains a setv
<PIE:·= {F E T~
(A)
IIIF(<p) II <d
for some E: > .0 and ¢E:
~.I. Let <p E ~. Then for all F C foX
From this observation the wanted continuity of t¢ follows. II. The proof of this part is similar to the proof of part II of
Theorem (1.6).
For each n C ~, put
p
n
and define
f
dE'
r "= sup{p > OI[F C emb(P (X)) /\ liP FII < pJ ~ P FEst}.
n n n n
We recall that P F can be interpreted as an element of
X
in then
following way: P F
=
~(A)-l
PF(~)
where~
€
~
is taken such thatn
n
(X[ + X J) ~ eli; for some c > O. Define the function G as
n-l,n) (-n,n+l
follows
G (:\) i f !AIE [n - 1,n) .
(ef. the
To this end, let f
E
¢*.
Then f+
- *
element of ¢ .
+
We shall prove that
G E
¢**
sup (f(A)) is an
I
AI
EJn-l,n)proof of Theorem (1.6)). So there exists 0 > 0 such that
- 2
defined by f(A)
=
n{F E
f·
XIII FIIf
<o}
C $(,n
f·Xbecause
n n
~·X contains an open neighbourhood ofO.
Thus we find2 > cS n that r n So for each sup (f(A)) IAIEIn-l,n)
n E :IN and all
I
AI
Efor all n C IN.
[n - 1,n) 2n2
o
(A)flIt) ~ r n 2 sup ( f ((p )) ~ ().1,\
JEIn-l ,n) ::f::4:.It follows that 8 E ¢ . By axiom A V there exists c > 0 and
¢
E ¢such that G ~ c¢.
We next show that i t is enough to put s 1
c i.e. we prove'
1
II F (¢)II < - ~ F E ,L c
\1 ~
Let F E
foX
for some f E~:.
Thenw
I
/lPnF/I~
< 00 n=l andII
A
2)-3PnFllf 2 -311 II
(*) (I + ~ (1 + (n - 1» F f 1Because of our assumption /IF(<j>)/I < - we get c
(**) 2n 211P FII
~
r cIIP F (¢)II~
rn n n n
for all n
E
IN. Now we express F as followswhere F N
L
1 (2n2p F) + (L
~l~)F
n=l 2n2 n j=N+1 2n2 N 00 00 Put g(A)( I
~)-1(
I
P
F) j=N+1 2j n=N+1 n(1 +
A
2
)3 f(A).
Then gE
~*,
because of A IV. By (*) we get4 2 -3
~F /I ~ 4N (1 + N) /lF~f N g
and so i t follows that F
N
E
~n
goX
for sufficiently large N E ~ .Moreover, by (**) for all n E :IN , 2n P F E2 [2.
n
Hence F is a subconvex combination of elements in the convex set ~.
So F ~ ~. It is clear that F E Q
n
foX.
An immediate consequence of the previous theorem is the following interesting result.
III.
(fog) 0 X
bOUllded in T
cp(A) - 'l'hcn 8 is closed and bounded ¢*. From the above scheme i t is clear that B is
+
is Montel iff there exists f
E
¢* such that f(A)-l+ is barreled
is bornological
is a compact operator from X into X.
and hence Hilber-Schmidt.
T¢(A) is nuclear iff for all ¢
E
¢ the operator ¢(A) is nuclearf(A)-l X ~~---~~---~+ X q(A)
1
1
f(A)g(A) II. I. (2.11)The.OfLe.m
I I I . (2. 10)The.OfLe.m
The topology 'proj on T¢(A) is equivalent to the topology Lind.
in yoX for some g ( Let B be closed and IV.
The proof of I and II runs completely the same as the proof of
Theorem (1.15. II and III). Therefore i t is omitted.
- II
-a comp-act subset of (f·g)·X. Cf. Coroll-ary (2.6).
~) Let (x )n nE:IN be a bounded sequence in X. Then the sequence (xn )nElN is also bounded in T<p (A)' and hence ~I n E :IN} is a
compact subset of T<p(A)' SO for each ~ E <p the sequence (~(A)xn)nEm contains a converging subsequence. It follows that each operator ~(A), ¢ E (~ , is compact. Hence the projection nn
=
X (-n,n] (A) is a compact operator, cf. the proof of Theorem (2.2). Thus we obtain that A hasdiscrete spectrum without accumulation points. Put f(A) = 1 +
;.2,
AE:
lR.*
-1Then f E <P and f(A) is a compact operator on X. +
IV
=» Assume that all operators ¢(A), ¢
E
<P, are Hilbert-Schmidt. It follows that A has a discrete spectrumand eigenvectors (e.). ON' Let ~ E <P. Then ]
J-with eigenvalues (A.) ''IN
J J-'
J :
t
¢~
t
;p2,
whereT
~
2-and T~ denote the completion of T<p(A) with respect to t¢2 and t~, respectively. Since] can be written as
JG
co
I
~(A) (G,~(A
)-l e)~ ~(A
)-2 en=l n n n ~ n n
-1
where (G,¢(A) e ) = (¢(A)G,e ),
J
is a nuclear map.n n <p n
=:l
For each XE:
<P the embeddingJ
Then there exists ~ E <P such thatx
e+t
is bounded. Let <p E <P Xis nuclear. Hence the injection X e+T¢ is nuclear. It follows that
In order to make S¢(A) and T¢(A) into a dual pair, we introduce the sesquilincar form <,> on S,]>(A) x T<j'(A) by
~lcre ~ E ¢ is such that u f ~(A) (X). This definition does not depend
on the choice of ~. To see this, we note that for any u C ~(A) (X)
n
1jJ(A) (X)The sesquilinear form is non-degenerate. I f <u,F> 0 for all F C T<P(A) ,
then u = 0 because
X
c T<p(A)' Conversely i f <u,F> 0 for all u E:S<p(A)'
tlwn (f(A)u,x) - 0 for all u E
S¢(A) and F = fox. Since D(f(A)) is
dense in
X,
we obtain x 0 and hence F =O.
(3.1) Theo~0m (Riesz-Fischer)
1. A linear functional 9, on
S
is continuous in the topology o;nd,]>(A) L
iff there C'X~f' \1(' Lhe,,!:
9,(u) = <u,G> ,
II. A linear functional m on
T
, i s¢ (A) c:on-;:.~_~c.nuous in the topology Tpro].
of T¢(A) iff there exists w
E S
stech that- 35
-1. =» Let GET¢ (Al. Then for all w E~ (Al (Xl
Hence £ defined by £(w) <w,G> is continuous on each ~(A) (X) and hence on (S<j>(Al' oind) .
1. <=) Let Q, be a continuous linear functional on (S1l (A) , o. d)l.n ~ Then £ 0 ~(A) is a continuous linear functional on X for all ~ f ¢. So
there exists x~
E
X such that(2 0 ~ (A) ) (x)
=
(x,x¢)Put G ~ ~ x~, ~
E
¢. Then i t is easy to see that G(~~) Hence G (T¢(A). Moreover1/I(A)G(~).
<u,G> .
II. =» Let w 1/1(Alx with x E: X and 1jJ E ¢. Then we have
So the linear functional F ->- <w,F> is continuous on (TI' (A)' T .)
( pro]
II.<=) Let
m
be a continuous linear functional on (T¢(A) ,Tproj)' Since 1l is a p.o. set there exists c >
a
and a function ¢ E ¢ with(*) !m(F)
I
~cIIF(¢)IIX·f
-1f
-1Put
P
dEA
and ¢(A)¢(A)
dEA
.
It is clear that ¢ sUPP(<jl) supp(<jl)¢(A) (X) is a dense subspace of
P
¢ (X) (a Hilbert subspace of Xl. Moreover, the linear functional m 0¢(A)-l
is bounded on¢(A)
(Xlby (*l. Hc'ncC' there exists y (
r
(Xl such t.hat the continuous 0xt"c'nsion'I'
-1 -1
m
0 ¢(A) ofm
0 ¢(A) toP¢
(X) satisfiesm
0 (x,y) ,x
EX.
-1 .
Hence m(F) =
m
0 ¢ (A) (F(¢)) = (y,F(¢))Now put w
=
¢(A)y. Then m(F)=
<w,F>.From the previous theorem i t follo\,s that the spaces S<jl(A) and T<jl(A) a dual pair. Thus it makes sense to introduce the weak topologies
are []
o (Sq, (A) ,Tq, (A)) and oCTq, (A) ,Sq, (A)) on Sq; (A) and T<jl(A)' respectively. I t is
a natural question whether weak boundedness of a subset of Sq,(A) (Tq,(A)) implies its boundedness.
(3.2)
TheOJteJn
(Banach-Steinhaus)I . Each weakly bounded set B in
Sq,(A) is bounded. II. Each weakly bounded set
V
inTq, (A) is bounded.
I. Let B be a weakly bounded set in S i~\' So for each G ET<p(A) there exists M
G > 0 such that
'*
Now let f E <p • Then fot d.ll. x E X IJe have
- 37
-From the Banach-Steinhaus theorem for Hilbert spaces i t follows that there exists M
f
>
0 such thatwEB.
*
Since f ( ¢ is taken arbitrarily, i t follows that
B
is bounded.II. Let V be a weakly bounded set in T¢(A)' Then for each u E S¢(A)
there exists M > 0 such that
u
V
G(B
I
<u,G>I
~MuLet
¢
C
W. Then for all x E X we haveV GCB
So there exists K¢ > 0 such that
G (
V .
o
In the remaining part of this section we assume that the p.o. function
set ¢ satisfies axiom A V, see (1.8). Then we have the following theorem
which is similar to Corollary (1.14).
(3.3)
The.altern
Let ~ fulfil also axiom A V.
1. A sequence (un)nCN C S~(A) is weakly convergent to zero iff there
exists
¢
E ~ such that the sequence (u ) ( converges to zeron n ]N
weakly in ¢(A) (X), i .e. for all x E: X.
( ¢ (
A)
-1 u , x ) -+ 0 . nII. A sequence (G )
E:
eTA
is weakly convergent to zero iff nnlN ¢ ( )there exists f E ¢* such that (G ) c
foX,
Gfox,
and (x ) ON+ n n n
nn-tends to zero weakly in
X.
1. Since (un) nON is weakly bounded in S¢(A) i t is also strongly
bounded in S¢(A). SO by Lemrrla (1.9) there exists
¢ (
¢ such that(u ) EJN is a bounded seql1ence in the Hilbert space ¢(A)2 (X). It
n n
follows that there is a positive constant M such that for all
n CIN
r
-4J
¢(A)
d(EAun,u n) supp(¢) Now take AO
so large that forIAI
> AO
'
¢(A)
< s, where S > 0 istaken arbitrarily. Put
A
= {A EJRI(I"A!
> AO) v (¢(A) <
s)L
AO,
E:Then we have
2 2 < E: r.~
Moreover, let x E X be taken arbitrarily. Then there exists N E IN
~II
Hence
I
«I)(A)-lu ,x)I "
c(l + Mllxll) for all n " N.n
The converse of statement I is trivial, i.e. any weakly convergent
sequence (un)nE~ c ¢(A) (X) is a weakly convergent sequence in S~(A)'
II. Let (Gn)nEIN C T~(A) be weakly convergent to null. Then the sequence
=l=
(G)n nE:-IN is Tpro].-bounded. So there exists g E ~+ such that (G ) isn
bounded in g-X. Put G ~ goy. Then there exists D > 0 such that
n n k
Ilynll
~
D for all n E::IN. Put Ilk=
f
dEA, k E::IN. Then
-k
2 -1
Now let € > 0, and take k E IN so large that (1 + k ) D < €.
By A IV of Definition
(1.1),
there exist constants c1
, ...
,cjk and functions ¢ 1 " " ' ¢ ' ( ~ such that]k
(*) (1 + A)2 -1g(A)-1X(-k,k) (A)
~
c.¢.(A) ,] ] A E JR •
Now let x E X. By (*) we obtain
So there exists NEill such that for all n > N
< €
Put x
+
With f
:( c(l +
Ilxll)
2
A~ (1 + A )g(A), thG proof is complete. LJ
As a consequence of the previous result we obtain.
(3.4) COfLOUMIj
1. Each bounded sequence in S~(A) contains a weakly convergent sub-sequence.
ones,
II. Each bounded sequence in T¢(A) contains a weakly convergent sub-sequence.
Comparing the results of this section with the results in the previous i t is not hard to see that
a.
d ~ (J on S a n d T. . ~ T. d~n proj ~(A) proJ ~n
on T¢(A) are equivalent to the Mackey topology on each of these spaces. We note that this observation seems to be no longer true if axiom A V is dropped.
- 41
-Let 1>1 and 1>2 be two generating function lattices as introduced in
Definition (1.1), and let
Al
andA
2 be self-adjoint operators in the
Hilbert spaces Xl and
X
2
,
respectively. In this section we characterizethe continuous linear mappings from S~
(A )
into S~(A )
and from1 1 2 2
T~
(A )
into T~(A ).
Also we introduce the notion of extendibility1 1 2 2
for linear mappings from
S
into S~(A ).
~
1(A
1) 2 2( 4 • 1)
T
he.O!te.m
-~----Let R be a locally convex topological vector space, and let Lbe a
linear mapping from S~
(A )
into R. The mapping1 1
L~1 (AI) : Xl ~ R is continuous for all ~1 E ~1·
L
is continuous iffBy the definition of inductive limit topology, ~1(AI) (Xl) is
continuous-ly injected into 51>
(A ).
So ifL :
51>(A )
~ R is continuous i t follows1 I 1 1
that L~l (AI) : Xl ~ R is continuous.
~) Let L~ denote the restriction of
L
to ~1(A )
(Xl). Then L~ is1 1 '1'1
continuous on ~l(AI) (Xl). So if ~ is an open null neighbourhood in R,
then
L-1(~)
n
~1
(AI) (Xl)=
L;l(~)
is an open neighourhood in1
~1
(AI) (Xl)· I t follows that L-1(~)
is open in 5q, (A ). 0In the following theorem we characterize the space
L(S¢
(A
),S¢
(A ))
1 1 2 2
of continuous linear mappings fnyu) Scp1(Ai) into S¢2(A 2
) •
(4.2)
The.OfLe.m
I. Let
P
be a linear mapping fromS¢ (A )
intoS¢ (A ).
ThenP
is1 1 2 2
continuous iff for all ¢1 E CPl and for all f
2 E
¢i,
the linear II. Suppose ¢2 satisfies axiom A V.Let
P
be a densely defined operator in Xl wi-thP :
D(P) -+ X2 and D(P) :::J
S<jJ (A )' Then P is a continuous linear mapping from ScjJ (A)
1 1 1 1
into S iff one of the following conditions is fulfilled. cjJ2(A
2)
(i) For all ¢1 E ¢1 and for all f 2
E:
<1>2 the linear operatorr/'
(1\)
1',1'
1(1\)
It· ( )I~IXl
iIII ( )X;>
i ,; I )()lllld,·<I .(ii) For each ¢1 E '1>1 there exists ¢2 € cjJ2 such that the linear operator -1
¢2
(A
2) P¢ 1(Ai) is a bounded linear operator from X1 into X2·
I. 'l'he proof of thispar' olLy"c; >;''lTk;:',:i" f J'( from the previous
theorem and the fact that the inductive limit topology on 5 '1>2
(A
2)
is equivalent to the locally convex topology generated by the seminorms Sf
\'1~"
il C ..
(A,.,)II,
f E <1'*' 2 ';::
2
-) Let wI
= ¢
(A)x
E S A ) ' Then for all
f2
C
~~2_
itfolloWA
1
1
1
'1'1 (
1 "that
P¢l
(A 1)x 1
E
D(fZ(AZ))' So by Corollary (1.11)
P¢1
(A
1
)x
1
E S¢
(A )'
2 2
Hence
P
maps
S¢
(A ) into S¢ (A ). Applying Part I of this
1 1 2 2
theorem, i t follows that
P
is continuous.
II.il . •) The set
{P¢l(Al)xl~xl E
X
1
,Ox
l
O= 1}
is bourided In S¢l(A
1
)'
So there exists ¢z
E ¢2and c
> 0following Lemma (1.9) with
operator from Xl into
X 2•¢2
E
¢
such that the operator
2 .
'*
Then for all f
2 f 4'2
the operat.or
-J
If for each ¢1
E
¢l
there exists
-1
A
¢2(AZ)
~Pl ( 1) is bounded.
o
(4. 3)
COfWUaJtlJ.
Let
Q :
Xl
+X
z
be a closable operator. If
D(Q)
~
S
and if
<t> 1 (A
1
)
Q(S<t> (A ))
CS~
then
Q
is a continuous linear mapping from S
1 1
""z
(A
Z) <Ill(AI)
into S¢ (A ).
z z
*
Let ¢l E <PI and let
fZ E <P Z
,+Then the operator f
2
(A
Z
)Q¢l (AI) is an
everywhere defined operator from Xl into X
operator (f
2
(A
2)Q<P 1
(A
1
»*
is densely defined (its domain contains-1 f
2
(A
2)(D(Q*)}),
It follows that f2(A
2)Q<P
1(A
l) is closable and :I'hence a bounded operator from
Xl
intoX
2
·
Since<P
1
E
~1 and f2
E
~2,+are taken arbitrarily, the result follows from the previous theorem, - 0
Next we characterize linear mappings from
(4,4)
TheoJtem
of continuous
Let
R
Xl
7X
2
,
ThenR
can be extended to a continuous linear mappingfrom such
T~
(A )
into T~(A )
iff for all<P2 E
~2 there exists<P l E
~l1
1
2
2
-1
that the operator
<P
2
(A
2
) R<P
l(A
l) is bounded inX
10n<P
l(A1)X
l .If the function lattice ~l satisfies axiom
A V,
then the previouscondition can be replaced by: for all
~) Let
<P
2
E
~2' Then there exists <P1
E
~1 and c >0
such thatinto
X2'
=-) Wu definc~
I<
on T,p1(A,L 1
45
-This definition does not depend on the choice of
¢1
E<I>1.
It is clear that RF ET<I>2(A
2
).
The continuity ofR
follows from the inequalityII~) We define R in
T<I> (A )
as follows 1 1<1>2
E
<I>2
is the continuous extension to . Xl of the operator
¢ 2
(A2)
Rf~
(Ai)· Hence R mapsT<I> (A )
intoT
1>(A ).
1 1 .2 2
The continuity of R follows from the definition of inductive limit
*") Trivial.
o
(4.5)
CoJWliaJl.lj
Let
<I>2
satisfy axiom A its dual mappingP' :
v. Then P :
S<I> (A )
~S<I> (A )
is1 1 2 2
T1>
(A )
~ Tq,(A )
is continuous.2
2
1
1
continuous iff
Finally we introduce the notion of extendibility.
(4.6)
VeOi.-iUUon.
mappings by
E(S<I>
(A ) ,
5<I>(A
».
1 1 2 2
T"
(A )
such thatff
=E
I 2 . :>. 5<I' (A) •
1 1
We denote the space of extendible
A continuous linear mapping
E
from 5<I>
1(Ai) into 5<I>2 (A
2
)
is calledex-tendible if there exists a continuous linear mapping
E
fromT<I>
(A )
into1 1
(4.7) Theoltern
Let
E
be a continuous linear mapping from S~l(A
l )
~l satisfy axiom A V. Then
E
E E(S~(A )
,S~(A ))
*
1 1 2 2and
E
(S~
(A ))
cS~
(A ).
2 2 1 1
~) Observe that
E*
is contir.'lOl by Corollary (4.3). Putclearly
E
is continuous by-
*
1 E = (E ) : Corollary T~(A )
~ T~(A )'
Then 1 1 2 2 (4.5) and E~S (A) 00 E • I <P1 1~) For x
E
D(E) and wE
S~(A )
we have2 2 (w,Ex)X 2 <w,Ex>X 2
-*
- I*
It follows that E ~ E and
E
(S~(A ))
cS
2 2
~l(Al)
o
(4.8) COltOliCULlj
E E E(S", (A)' S", (A ,) iff O(E) :) SI; '.~. \ '
'" 1 1 '" 2 'L ' ,co .1· 1·
E(S~
(A)) cSq> ( A ) ' F tScP (A)) c SCP1(Al) .
47
-5.
A
~~vey06
Atta~'~ theo~y O~ GB*-atgeb~The and
c space S¢(A) establishes a natural core for the *-algebras ¢(A)
cc c
¢(A) of closable operators. Here ¢(A) denotes the commutant in L(S¢(Al) of the set ¢(A), and similarly ¢(A)cC its bicommutant. In Section 6, precise definitions of ¢(A)c and ¢(A)cC will be given and i t will be proved that both *-algebras are GB*-algebras. The notion of GB*-algebra has been introduced by Allan in his papers [Al 1J and [Al 2].
We give a short survey of his theory here.
Let A be a locally convex topological *-algebra over the field of complex numbers. So the mappings p~ pq, p~ qp and p~ p* are
continuous in A. An element pEA is said to be bounded if there exists s E
~
, {O} such that the set {(sp)nln E~}
is bounded in A. Theset of bounded elements is denoted by A O.
Let
B
denote the family of all bounded absolutely convex closed subsets of A with the property that for allBEB
(S.l)
(5.2)
B
=B*.
Now a GB*-algebra is introduced as follows.
(5.3) Ve6~nition
A locally convex topological *-algebra A with unit ~ is called a GB*-algebra if the folloWing conditions are fulfilled.
(i) A is sequentially complete.
(ii) A is symmetric, i.e. for every pEA the element (~ + p*p)
is invertible in A and
(~
+ p*p)-l is bounded.(iii) The family
B
defined by (5.1) and (5.2) has a maximal elementB
O with respect to set inclusion.
Remcvdz
Allan uses the notion of pSE:\l<:lncompleteness in his definition of a GB*-algebra. However for our aims i t is enough to consider sequential completeness which implies pseudo-completeness.
Let A be a GB*-algebra. Then the *-algebra
is a B*-algebra with respect to the Minkowski norm induced by B O. (Cf. [Ai 2], Theorem 2.6.)
(5.4) P~opo~~on ([Ai 2], Proposition 2.9)
is also a GB*-algebra. The maximal element of the family
{B c C , B bounded, absolutely convex closed, B2c B, B*
is just the set
B,
~ = 8l.__0n c.
(ii) If A is commutative then A
For a commutative GB*-algebra Allan has proved a "non-bounded"
extension of the Gelfand-Naimark theorem. In order to show this,
let A denote the set of all multiplicative linear functionals on
AO
A(B O)' i;e. the spectrum of the commutative B*-algebra A
O
.
So A is a compact set in the topology o(A,A
O
). The Gelfand
transform on A
O
is an isometric *-isomorphism from A
O
onto the
B*-algebra of continuous complex valued functions C(A), with norm
II
fll '" sup
I
f (X)
I .
Each element X E: A has an extension to the whole
XfA
algebra
A.
(5.5) P~aro~~t{on
([Ai 21, Proposition 3.1)
Let A be a commutative GB*-algebra with identity and let A be the
spectrum of AO. Then corresponding to each X E A there is an
ex-tending complex-valued function X
e
: A
~ ~*'"
~U
{oo}such that
(i) (ii)
e
X
is an extension of X
for pEA,
~E
~with the convention 0
0 00o.
e e
provided that X
(Pi)and X
(P(iv) for Pi ,P2 E A, provided that (Xe (Pl) ,X8(P 2))
~
(00,0) or (0,00). (v) for pEA, e X (p*) Xe (p)with the convention 00 = 00
(vi) the set N = {X E Alxe(p) ~ oo} is a nowhere dense closed subset of p
A.
The explicit formula for Xe can be obtained as follows. Let pEA be hC'rmitGi.m. Then Allan proved that there exists tl E a: such that (p,-r - p)
A -1 e X ( (tl,-r -1 -1
is invertible in and (tl,-r
-
p) (" AO'
We put X (p) = tl-
p) )-1 e
if X(tl,-r - p) ) ~ 0, and X (p)
=
otherwise. So we can extend X to all hermitean elements of p.. ,md next t,y 1 Lnearity to the whole of A.(5.6) Ve6~~a~
A collection F of a:*-valued functions on a LOpOlOglcdL space f is
called a *-algebra of functions if each f E F takes the value 00 on an at most nowhere dense subset of
f.
Moreover for any f,g (F
anda,S
r-
a: the functions o.f+ 13g, 1'-g and f* = ar", pointwise well defined ona dense subset of f on which f and g are finite.
Each of all those functions has a unique extension to a a:*-valued continuous function jp •
- 51
-(5.7)
The-altern
([Al 2], Theorem 3.9)
Let A'be a commutative GB*-algebra with identity and let
A
be the
spectrum of A
O
. Define the mapping
on A,
p H·p,
pfA, by
P
(X)
coXe (p)
~
where X E A. Then
~is a *-isomorphism of A onto a *-subalgebra A of
continuou~;<r:*-Vil1ued funct.ions on II. 'l'hp Illappinq ~ ('xt('n(l~; the \1Slli11 Gelfand transform for Alet <Jl(A)
G.
The. aXgeb!Lcu
1'(A)cand
1'(A)cCLet l' be the function lattice as introduced in Definition (1.1), and
{ ¢(A)
I
¢ E 1)}. Further, ll't L(5 A)) denote the set of(I) (
continuou~; Unc:ilr mappinqs on 5(1)(A). We note' tlwt 'I)(AJ c L(S,!)(AJ)·
In the following definition we introduce two subalgebras of L(Sw(A))
which are connected to w(A) in a natural way. As a main result we
prove in this section, that these snbalgebras are GB* in the sense
of Allan. In the next section we shall construct a functional calculus
for one of these GB*-algebras.
(6.1) Ve6i~~on
is called its bicommutant.
(6.2)
Ve6inition
The uniform topology
T
u for L(S<j)(AJ) is the topology induced by the
seminorms r f ,¢.
II
f(A)
L
¢(A) II ,
where f E