HILBERT RIGHT SEMIFINITE W
∗-MODULES
CORNELIU CONSTANTINESCU
Let E be a semifinite W∗-algebras. We define the Hilbert dimension for every selfdual Hilbert rightE-module in such a way that two suchE-modules are iso- morphic iff their Hilbert dimensions coincide.
AMS 2000 Subject Classification: 46L08.
Key words: selfdual Hilbert rightW∗-module.
INTRODUCTION
Every Hilbert space H has a Hilbert dimension, say DimH, which is the cardinal number of the orthonormal bases of H. Two Hilbert spaces are isomorphic iff their Hilbert dimensions coincide and, for every cardinal number ℵ,Diml2(ℵ) =ℵ.Moreover, if H and K are Hilbert spaces, then
Dim (H| K) = DimH+ DimK, Dim (H⊗K) = (DimH) (DimK), and the statements below are equivalent.
a) DimH≤DimK.
b) There is a Hilbert space L such thatK =H| L.
c) There is a surjective operator K →H.
d) There is an injective operator H→K.
Let E be a semifinite W∗-algebra. We define the Hilbert dimension for every selfdual Hilbert right E-module and try to extend the above landscape from classical Hilbert spaces to selfdual Hilbert right E-modules.
In general, we use the notation and terminology of [C1]. For W∗-tensor products of W∗-algebras we use [T], for tensor products of Hilbert right C∗- modules we use [La], and for W∗-tensor products of selfdual Hilbert right W∗-modules we use [C2].
We now give a list of some notation used in this paper.
1. Kdenotes the field of real numbers or the field of complex numbers.
The whole theory is developed in parallel for the real and complex case, but the proofs coincide. N denotes the set of natural numbers (0 6∈ N) and for
MATH. REPORTS11(61),2 (2009), 75–137
everyn∈Nwe putNn:={k∈N|k≤n}.R+denotes the set of positive real numbers (0∈R+).
2. IfX is a set, then for every subsetA ofX we denote byeA:=eXA the characteristic function ofA inX, i.e., the function onX equal to 1 onA and equal to 0 on X\A.
3. If X, Y are sets,
f :X→Y, x7→f(x)
a map, and Aa subset of X, thenf|A denotes the restriction off toA, i.e., f|A:A→Y, x7→f(x).
4. For every set A we denote by CardA its associated cardinal number and by Pf(A) the set of finite subsets ofA.
5. For ordinal numbers we adopt von Neumann’s point of view: if α, β are ordinal numbers then α∈β ⇔α < β. In other words, an ordinal number β is just the set of all ordinal numbers strictly less than β.
6. In ordered sets, ∧and ∨ denote the infimum and supremum, respec- tively.
7. If E is a Banach space, thenE#denotes the unit ball of E:
E#:={x∈E | kxk ≤1}.
IfE has a unique (or a distinguished) predual, then we denote by ¨E this pre- dual and byEE¨ the vector spaceE endowed with the locally convex topology of pointwise convergence on ¨E. If (xi)i∈I is a summable family in EE¨, then we denote by
E¨
P
i∈I
xi its sum. It is characterized by the relation
a∈E¨ ⇒
*XE¨ i∈I
xi, a +
=X
i∈I
hxi, ai.
8. If H is a Hilbert space and ξ, η ∈H, then we denote by hξ|ηi their scalar product and puth · |ηiξ:H →H,ζ 7→ hζ|ηiξ;L(H) denotes theW∗- algebra of operators H →H and L′(H) its predual (which may be identified with the set of nuclear operators of L(H)).
9. If I is a set, thenl2(I) denotes the Hilbert space of square summable families inKindexed byI and (ei)i∈I denotes its canonical orthonormal basis.
For i, j∈I,δij denotes the Kronecker’s symbol δij :=
( 1 ifi=j 0 ifi6=j .
10. For every C∗-algebra E we denote by ReE the set of selfadjoint elements of E. A Gelfand C∗-algebra is a commutative C∗-algebra E such
that in the real case all its elements are selfadjoint. We denote by σ(E) its spectrum and byC(σ(E)) theC∗-algebra of continuous scalar valued functions on σ(E). The Gelfand transform
E → C(σ(E)), x7→xb is an isomorphism of C∗-algebras.
11. Let E be a W∗-algebra. We denote by 1E its unit and by PrE the complete lattice of orthogonal projections ofE. For p, q∈PrE we put p∼q (resp. pq) if there isx∈E with
x∗x=p, xx∗=q (resp.xx∗ ≤q).
The notation∼ and will be used also forC∗-algebras. We put Ec :={x∈E |y∈E ⇒xy=yx}.
AC∗-subalgebraF ofEis calledW∗-subalgebra if the inclusion mapFF¨ →EE¨
is continuous; ¨E+ denotes the convex cone of positive elements of the predual of E.
12. LetE be aC∗-algebra andH, K Hilbert rightE-modules. We denote by L(H, K) the Banach space of operators H→K, by LE(H, K) its Banach subspace of adjointable operators, by LbE(H, K) the Banach space of u ∈ L(H, K) such that
(ξ, x)∈H×E⇒u(ξx) = (uξ)x,
by 1H the identity mapH →H (which belongs toLE(H) :=LE(H, H)),and putHb :=LbE(H, E) and
h · |ξ i:H→E, η7→ hη|ξi
for every ξ ∈H. If for every u∈Hb there is ξ ∈H withu:= h · |ξ i,then we say that H is selfdual.
13. Let Ebe aC∗-algebra. ForH, K Hilbert rightE-modules we use the notation H≈K for “H andK are isomorphic”.
14. LetE be aW∗-algebra andHa selfdual Hilbert rightE-module. For a∈E¨ and ξ, η ∈H we put
(a, ξ) :] H→K, ζ 7→ h hζ|ξi, ai, (a, ξ, η) :^ LE(H)→K, u7→ h huξ|ηi, ai. Then H and LE(H) have preduals which are generated by
{(a, ξ)] |(a, ξ)∈E¨×H} and {(a, ξ, η)^ |(a, ξ, η)∈E¨×H×H}, respectively. In this case we denote by H... the predual of LE(H). If p ∈ PrLE(H) then pH := Imp = {pξ | ξ ∈ H} is a closed Hilbert right E- submodule of H. If K is a closed Hilbert right E-submodule of H, then
we put
K⊥:={ξ ∈H |η∈H ⇒ hξ|ηi= 0}.
A Fourier set of H is a subsetA of H such that
ξ∈A⇒ hξ|ξi ∈PrE, ξ, η ∈A, ξ 6=η⇒ hξ|ηi= 0.
A Fourier basis ofH is a maximal Fourier set of H.
15. If E is a W∗-algebra and (Hi)i∈I is a family of selfdual Hilbert right E-modules, then we put
|
i∈I
Hi :=
ξ ∈Y
i∈I
Hi |the family (hξi|ξi i)i∈I is summable inE
, W|
i∈I
Hi:=
ξ ∈Y
i∈I
Hi |the family (hξi|ξi i)i∈I is summable inEE¨
. 16. ⊙denotes the algebraic tensor product of vector spaces.
17. If E, F areW∗-algebras andH (resp. K) is a selfdual Hilbert right E- (resp. F-) module, then we denote byH⊗K theW∗-tensor product ofH and K, which is a selfdual Hilbert right E⊗F-module ([C2], Definition 2.3).
18. HK for Hilbert rightC∗-modules H, K: Definition 2.3.
19. H(E), class of selfdual Hilbert rightE-modules: Definition 2.10.
20. Γ(E), 0Γ(E)∈Γ(E),Hp forp∈Γ(E): Definition 3.4.
21. Semi-abelian orthogonal projection, semi-abelianW∗-algebra, type I W∗-algebra: beginning of Section 4.
22. Cd,Cd0, classes of (mainly) cardinal numbers with order relation ≤ and operations +,−: Definition 4.1.
23. Hilbert dimension of Hilbert right W∗-modules: Theorems 4.9, 5.17.
24. Type II W∗-algebra: beginning of Section 5.
25. ∆(E), 0∆(E)∈∆(E), HZ forZ ∈∆(E): Definition 5.3.
1. MATRIX REPRESENTATIONS OF OPERATORS
We start with
Lemma 1.1. Let E be a W∗-algebra, (Hi)i∈I a family of selfdual Hilbert right E-modules, andH:= |
i∈I
Hi, K:= W|
i∈I
Hi ([C1],Proposition5.6.4.1 c), [C1], Proposition 5.6.4.6 b)). For every η ∈ K we have h · |ηi |H ∈ Hb and the map
K →H,b η7→ h · |η i |H is an isomorphism of Hilbert right E-modules.
Proof. It is obvious that h · |ηi |H ∈Hb for every η ∈K. Moreover, for η∈K, x∈E,and ξ ∈H we have
(h · |ηxi)ξ =hξ|ηxi=x∗hξ|ηi= (h · |ηix)ξ, so that
h · |ηxi=h · |ηix.
Let Pf(I) be the set of finite subsets of I,F its upper section filter and for everyJ ∈Pf(I) leteJ be its characteristic function. For η∈K we have
limJ,Fhη−ηeJ|η−ηeJ i= lim
J,F E¨
X
i∈I\J
hηi|ηi i= 0 inEE¨. By ([C1], Theorem 5.6.2.11 f)),
h h · |ξi | h · |ζ i i=hξ|ζ i for (ξ, ζ)∈H×K. By the above,
limJ,Fh h · |ηi − h · |ηeJ i | h · |ζ i i= 0 forη, ζ ∈K inEE¨, so that
h h · |ηi | h · |ζ i i= lim
J,Fh h · |ηeJ i | h · |ζ i i= lim
J,FhηeJ|ζi=hη|ζ i. Thus, we have to prove the surjectivity of the map only.
Let ϕ ∈ H. For everyb i ∈ I, we identify Hi with a Hilbert right E- submodule of H in a natural way. Since Hi is selfdual, there isηi ∈Hi such that ϕ(ξi) =hξi|ηii for everyξi ∈Hi. LetJ be a finite subset of I and
ηJ :=
( ηi ifi∈J 0 ifi∈I\J . Then
ϕ(ηJ) =X
i∈J
ϕ((ηJ)i) =X
i∈J
hηi|ηii=hηJ|ηJ i, kηJk2 =k hηJ|ηJ i k=kϕ(ηJ)k ≤ kϕk kηJk, kηJk ≤ kϕk.
Thus, η := (ηi)i∈I ∈K and ϕ(ξi) =hξi|ηi i = (h · |ηi)ξ for every i∈I and ξi ∈Hi. It follows that ϕ=h · |ηi |H, hence the map is surjective.
Lemma 1.2. Let E, F be C∗-algebras, (Hi)i∈I a family of Hilbert right E-modules, (Kj)j∈J a family of Hilbert right F-modules, and
H := |
i∈I
Hi, K:= |
j∈J
Kj.
Then the linear map H⊙K → |
(i,j)∈I×J
(Hi⊗Kj), ξ⊗η7→(ξi⊗ηj)(i,j)∈I×J can be extended to an isomorphism
H⊗K → |
(i,j)∈I×J
(Hi⊗Kj).
of Hilbert right E⊗F-modules.
Proof. Let us denote by ϕ the first map and let (ξ′s, η′s)s∈S, (ξ′′t, ηt′′)t∈T be finite families in H×K. Put
ζ′:=X
s∈S
ξ′s⊗ηs′, ζ′′:=X
t∈T
ξ′′t ⊗ηt′′.
Then
ϕζ′ ϕζ′′
= X
(s,t)∈S×T
ϕ(ξs′ ⊗η′s)
ϕ(ξt′′⊗η′′t)
=
= X
(s,t)∈S×T
(ξsi′ ⊗η′sj)(i,j)∈I×J(ξti′′ ⊗ηtj′′)(i,j)∈I×J
=
= X
(s,t)∈S×T
X
(i,j)∈I×J
ξsi′ ⊗ηsj′
ξ′′ti⊗ηtj′′
=
= X
(s,t)∈S×T
X
(i,j)∈I×J
ξsi′ ξti′′
⊗ η′sj
ηtj′′
=
= X
(s,t)∈S×T
X
i∈I
ξ′siξ′′ti
⊗ X
j∈J
ηsj′ η′′tj
=
= X
(s,t)∈S×T
ξs′ ξt′′
⊗ ηs′
ηt′′
= X
(s,t)∈S×T
ξs′ ⊗ηs′
ξt′′⊗ηt′′
=
=
*X
s∈S
ξs′ ⊗ηs′
X
t∈T
ξt′′⊗η′′t +
= ζ′
ζ′′
. Thus,ϕmay be extended by continuity to a map
H⊗K → |
(i,j)∈I×J
(Hi⊗Kj)
which is obviously surjective and so it is an isomorphism.
Lemma 1.3. Let E and F be W∗-algebras and G a W∗-subalgebra ofF. Then E⊗G is the closure of E⊗G in (E⊗F)z }| {¨
E⊗F .
Proof. Let H and K be Hilbert spaces such that E and F are von Neumann algebras on H and K, respectively. Then E⊗G and E⊗F are the closures of E ⊗G and E ⊗F in L(H⊗K)L′(H⊗K), respectively. Since L(H⊗K)L′(H⊗K) induces on E⊗F the topology (E⊗F)z }| {¨
E⊗F
, E⊗G is the closure of E⊗Gin (E⊗F)z }| {¨
E⊗F
.
The next result is nothing else but ([T], Section IV.1) generalized to selfdual Hilbert right W∗-modules.
Theorem 1.4. Let E be a W∗-algebra, H a selfdual Hilbert right E- module, I a set, K :=l2(I), and L:=H⊗K ([C2], Definition 2.3), which is a selfdual Hilbert right E⊗K-module (E⊗K=E).
a) For every ξ ∈ W|
i∈I
H, the family (ξi⊗ei)i∈I is summable in LL¨ and the map
W
|
i∈I
H → L, ξ 7→
L¨
P
i∈I
ξi⊗ei is a (canonical) isomorphism of Hilbert right E-modules.
b) IfM is a Hilbert rightE-module andu∈ LE(L, M)withu(ξ⊗ei) = 0 for all (ξ, i)∈H×I, then u= 0.
c) For everyi∈I putui :H →L,ξ 7→ξ⊗ei.Then, for i, j∈I, ξ ∈H, and η∈K we have
ui ∈ LE(H, L), u∗i(ξ⊗ej) =δijξ, u∗i(ξ⊗η) =hη|ei iξ, u∗iuj =δij1H, uiu∗j = 1H ⊗(h · |ej iei), X
i∈I
uiu∗i = 1L (pointwise).
d) For everyx∈ LE(L)andi, j∈I, putxij :=u∗ixuj.Then, fori, j∈I, x, y∈ LE(L), z∈ LE(H), and w∈ L(K) we have
xij ∈ LE(H), x∗ij = (x∗)ji, (xy)ij =X
k∈I
xikykj (pointwise), (z⊗1K)ij =δijz, (1H ⊗w)ij =hwej|eii1H,
(z⊗w)ij =hwej|ei iz, uizu∗j = (z⊗1K)uiu∗j =z⊗(h · |ej iei), x=X
i∈I
X
j∈I
uixiju∗j =X
i∈I
X
j∈I
xij ⊗(h · |ej iei) (pointwise), and the map LE(L)#...
L → LE(H)#...
H,x7→xij is continuous.
e) {uiu∗j |i, j∈I}c ={x⊗1K |x∈ LE(H)}.
f) If F is a W∗-subalgebra of LE(H), then
f1) F⊗ L(K) ={x∈ LE(L)|i, j ∈I ⇒xij ∈F}; f2) (F⊗K)c =Fc⊗ L(K);
f3) (F⊗ L(K))c =Fc⊗K.
Proof. a) By Lemma 1.2 the linear map H⊙l2(I)→ |
i∈I
H, η⊗ζ 7→(hζ|eiiη)i∈I can be extended to an isomorphism ψ : H⊗l2(I) → |
i∈I
H of Hilbert right E-modules. Then
ψb: \|
i∈I
H→
z \}| {
H⊗l2(I), u7→u◦ψ
also is an isomorphism of Hilbert rightE-modules ([C1], Theorem 5.6.2.11 g)).
If we put G:= W|
i∈I
H, then, by Lemma 1.1, the map ϕ:G→ \|
i∈I
H, ξ7→ h · |ξi |( |
i∈I
H)
is an isomorphism of Hilbert right E-modules. By ([C2], Definition 2.3) and ([C2], Proposition 1.3 f)), we haveL=
z \}| {
H⊗l2(I), so thatψb◦ϕ:G→L is an isomorphism of Hilbert right E-modules and thus the map GG¨ →LL¨ defined by ψb◦ϕalso is an isomorphism.
Let ξ ∈ G and for every i ∈ I put ˙ξi := (δijξi)j∈I ∈ G. For (η, ζ) ∈ H×l2(I) we have
(ψϕb ξ˙i)(η⊗ζ) = (ϕξ˙i)ψ(η⊗ζ) =D
ψ(η⊗ζ) ξ˙i
E
=
=h(hζ|ej iη)j∈I|(δijξi)j∈Ii=hζ|eii hη|ξi i=hη⊗ζ|ξi⊗ei i, so that ([C1], Theorem 5.6.2.11 e)), ψϕb ξ˙i = ξi ⊗ei. By ([C1], Proposition 5.6.4.6 c)), ( ˙ξi)i∈I is summable inGG¨ and
G¨
P
i∈I
ξ˙i =ξ. By the above, (ψϕb ξ˙i)i∈I is summable in LL¨ and
L¨
P
i∈I
ψϕb ξ˙i=ψϕξ.b Hence (ξi⊗ei)i∈I is summable inLL¨ and
L¨
P
i∈I
ξi⊗ei=ψϕξ.b Thus, the map
W
|
i∈I
H→L, ξ 7→
L¨
X
i∈I
ξi⊗ei
is nothing else but the isomorphism ψb◦ϕof Hilbert right E-modules.
b) By ([C1], Proposition 5.6.3.4 c)), the map LL¨ → MM¨, ζ 7→ uζ is continuous and the assertion follows from a).
c) For ζ ∈H we have
huiξ|ζ⊗ej i=hξ⊗ei|ζ⊗ej i=δijhξ|ζ i=hξ|δijζ i. By a),ui∈ LE(H, L) and
u∗i(ζ⊗ej) =δijζ, u∗i(ξ⊗η) =X
j∈I
u∗i(ξ⊗ hη|ej iej) =X
j∈I
δijhη|ej iξ =hη|ei iξ.
It follows that
u∗iujξ=u∗i(ξ⊗ei) =δijξ, u∗iuj =δij1H,
uiu∗j(ξ⊗η) =uihη|ej iξ =ξ⊗ hη|ej iei = (1H⊗(h · |ej iei))(ξ⊗η), so that, by b),
uiu∗j = 1H ⊗(h · |ej iei).
Since P
i∈I
h · |ei iei = 1K pointwise, we have X
i∈E
uiu∗i = 1L pointwise.
d) By c),
xij ∈ LE(H), x∗ij =u∗jx∗ui= (x∗)ji, (xy)ij =u∗ixyuj =u∗ixX
k∈K
uku∗kyuj =X
k∈J
xikykj (pointwise).
For ξ∈H, again by c),
(z⊗1K)ijξ =u∗i(z⊗1K)ujξ=u∗i(z⊗1K)(ξ⊗ej) =u∗i((zξ)⊗ej) =δijzξ, (z⊗1K)ij =δijz,
(1H⊗w)ijξ =u∗i(1H ⊗w)ujξ=u∗i(1H ⊗w)(ξ⊗ej) =
=u∗i(ξ⊗(wej)) =hwej|ei iξ, (1H ⊗w)ij =hwej|ei i1H, (z⊗w)ij = ((z⊗1K)(1H ⊗w))ij =X
k∈I
(z⊗1K)ik(1H ⊗w)kj =
=X
k∈I
δikzhwej|eki=hwej|ei iz.
For k∈I, by c) we have
uizu∗j(ξ⊗ek) =uizδjkξ =δjk(zξ)⊗ei =
= (zξ)⊗ hek|ej iei = (z⊗(h · |ej iei))(ξ⊗ek)
so, by b), uizu∗j =z⊗(h · |ej iei.By c) we have x= X
i∈I
uiu∗i
x X
j∈I
uju∗j
=X
i∈I
X
j∈I
uiu∗ixuju∗j =
=X
i∈I
X
j∈I
uixiju∗j =X
i∈I
X
j∈I
xij ⊗(h · |ej iei) (pointwise).
For (a, ξ, η)∈E¨×H×H we have Dxij,(a, ξ, η)^ E
=h hxijξ|η i, ai=h hu∗ixujξ|ηi, ai=
=h hxujξ|uiηi, ai=D x,
z ^}| { (a, ujξ, uiη)E
, so that the map
LE(L)#...
L → LE(H)#...
H, x7→xij
is continuous.
e) Let x∈ LE(H) andi, j∈I. By c) and d), uiu∗j(x⊗1K) =uiu∗j(x⊗1K)X
k∈I
uku∗k =X
k∈I
ui(x⊗1K)jku∗k=
=X
k∈I
δjkuixu∗k=uixu∗j = (x⊗1K)uiu∗j,
i.e., x⊗1K ∈ {uiu∗j |i, j ∈I}c and {x⊗1K |x∈ LE(H)} ⊂ {uiu∗j |i, j∈I}c. Let y∈ {uiu∗j |i, j∈I}c. Fori, j, k, l∈I, by c),
δklyij =yiju∗kul=u∗iyuju∗kul=u∗iuju∗kyul=δijykl.
It follows that i 6= j ⇒ yij = 0, yii = ykk. Putting x := yii for an i∈ I we have yij =δijx and, by d),
y=X
i∈I
X
j∈I
yij ⊗(h · |ej iei) =X
i∈I
X
j∈I
(δijx)⊗(h · |ej iei) =
=X
i∈I
x⊗(h · |ei iei) =x⊗1K,
y∈ {x⊗1K |x∈ LE(H)}, {uiu∗j |i, j∈I}c ={x⊗1K |x∈ LE(H)}.
f1) Let (x, y)∈F× L(K) and i, j∈I. By d), σij :LE(L)#...
L → LE(H)#...
H, x7→xij
is continuous andσij(x⊗y) =hyej|eiix∈F.Thus,σij(F⊗L(K))⊂F.Since F⊗ L(K) is the closure ofF⊗ L(K) inLE(H)⊗ L(K) =L(L)...L (Lemma 1.3, [C2] Theorem 2.4 d)), we have
(F⊗ L(K))z }|¨ { F⊗ L(K)
= (F⊗ L(K))...
L
([C1], Corollary 4.4.4.9)). Let z∈(F⊗ L(K))#. Since (F ⊗ L(K))# is dense in (F⊗ L(K))#...
L ([C1], Corollary 6.3.8.7)), there is a filterF on (F ⊗ L(K))# converging to zin (F⊗ L(K))#...
L . By the above, zij =σijz= lim
w,Fσijw∈F sinceF is closed inL(H)H.... Thus,
F⊗ L(K)⊂ {x∈ LE(L)|i, j∈I ⇒xij ∈F}.
Now, let x∈ LE(L) withxij ∈F for all i, j∈I. By d), x=X
i∈I
X
j∈I
xij ⊗(h · |ej iei) pointwise, hence with respect to the topology of LE(L)...
L. Thus, x∈F⊗ L(K), {x∈ LE(L)|i, j∈I ⇒xij ∈F} ⊂F⊗ L(K).
f2) Letx∈(F ⊗K)c. Fori, j∈I and y∈F, by c) and d), we have xijy=X
k∈I
xikδkjy=X
k∈I
xik(y⊗1K)kj = (x(y⊗1K))ij =
= ((y⊗1K)x)ij =X
k∈I
(y⊗1K)ikxkj =X
k∈I
δikyxkj =yxij, i.e., xij ∈Fc. By f1),x∈Fc⊗ L(K), (F ⊗K)c ⊂Fc⊗ L(K).
Now, let y∈Fc and z∈ L(K). Forx∈F we have
(y⊗z)(x⊗1K) = (yx)⊗z= (xy)⊗z= (x⊗1K)(y⊗z), so that
y⊗z∈(F⊗K)c, Fc⊗ L(K)⊂(F⊗K)c ⊂ LE(H)⊗ L(K).
By Lemma 1.3, we then haveFc⊗ L(K)⊂(F⊗K)c. f3) Letx∈Fc and (y, z)∈F ⊗ L(K). Then
(x⊗1K)(y⊗z) = (y⊗z)(x⊗1K), so that
x⊗1K ∈(F⊗ L(K))c, x⊗1K ∈(F⊗ L(K))c, Fc⊗K⊂(F⊗ L(K))c.
Letx∈(F⊗ L(K))c. By c) and e), there isx0 ∈ LE(H) withx=x0⊗1K. For y∈F we have
(x0y)⊗1K = (x0⊗1K)(y⊗1K) = (y⊗1K)(x0⊗1K) = (yx0)⊗1K, so that
x0y=yx0, x0∈Fc, x∈Fc⊗K, (F⊗ L(K))c ⊂Fc⊗K. Proposition 1.5. Use the notation of Theorem1.4and assumeH =E.
Let (pi)i∈I, (qi)i∈I, and (ri)i∈I be families in PrE, and P := W|
i∈I
piE, Q:= W|
i∈I
qiE, R:= W|
i∈I
riE, S:= W|
i∈I
E, ϕ:P →S, ψ:Q→S, θ:R→S
the inclusion maps. Identify S with L using the map of Theorem1.4 a).
a) ϕ∈ LE(P, S) and ϕ∗ξ= (piξi)i∈I for everyξ∈S. In particular,ϕ∗ϕ is the identity map of P.
b) Let v ∈ LE(P, Q) and x := ψvϕ∗. Then x ∈ LE(L), x∗ = ϕv∗ψ∗, kxk=kvk, andxij ∈qiEpj for alli, j∈I.
c) The map
LE(P, Q)→ {x∈ LE(L)|i, j ∈I ⇒xij ∈qiEpj}, v7→ψvϕ∗ is linear and bijective.
d) If v∈ LE(P, Q), w∈ LE(Q, R), then
(θwψ∗)◦(ψvϕ∗) =θ(wv)ϕ∗.
e)Letv∈ LE(P, Q) andx:=ψvϕ∗.Thenvis an isomorphism of Hilbert right E-modules iff
j, l∈I ⇒
E¨
X
m∈I
x∗mjxml =δjlpj, i, k ∈I ⇒
E¨
X
n∈I
xinx∗kn=δikqi.
f) Let r∈PrE such thatpi =rpir, qi =rqir for alli∈I, and putF :=
rEr. If P and Q are isomorphic, then the Hilbert right F-modules W|
i∈I
piF, and
W
|
i∈I
qiF are isomorphic, too.
g) Assume P and Qare isomorphic and E is aW∗-subalgebra of aW∗- algebraF. Then the Hilbert right F-modules
W
|
i∈I
piF and
W
|
i∈I
qiF are isomor- phic, too.
Proof. a) Easy.
b) By a), x ∈ LE(L). The relations x∗ = ϕv∗ψ∗, and kxk = kvk are easily checked. For ξ ∈E, by a),
ϕ∗ujpjξ=ϕ∗((pjξ)⊗ej) = (pjξ)⊗ej =ϕ∗(ξ⊗ej) =ϕ∗ujξ, so that ϕ∗ujpj =ϕ∗uj and xijpj =xij.By Theorem 1.4 d),
x∗ij = (x∗)ji= (x∗)jiqi =x∗ijqi, xij =qixij, xij ∈qiEpj. c) If ψvϕ∗ = 0 then, by a),
v= (ψ∗ψ)v(ϕ∗ϕ) =ψ∗(ψvϕ∗)ϕ= 0, so that the map is injective.
Now, let x ∈ LE(L) with xij ∈qiEpj for all i, j ∈I and letξ ∈L. For i, j∈I we have
(xij ⊗(h · |ej iei))ξ = (xij⊗(h · |ej iei))
L¨
X
k∈I
ξk⊗ek=
=
L¨
X
k∈I
xijξk⊗ hek|ej iei=xijξj⊗ei. By Theorem 1.4 d),
xξ= X
i,j∈I
xij⊗(h · |ej iei)
ξ= X
i,j∈I
(xij⊗(h · |ej iei))ξ= X
i,j∈I
xijξj⊗ei, so that xξ=ψ∗xξ∈Q,xϕ∗ξ=xξ. Put
v:P →Q, ξ7→xξ.
Then v∈ LE(P, Q) and
ψvϕ∗ξ=ψxϕ∗ξ=ψxξ=ψψ∗xξ=xξ, ψvϕ∗ =x.
Thus, the map is surjective.
d) follows from a).
e) Ifvis an isomorphism, thenv∗v= 1P,vv∗ = 1Qand the conditions are fulfilled by d) and Theorem 1.4 d). Conversely, if the conditions are fulfilled then, by d),v∗v= 1P,vv∗ = 1Q, i.e., v is an isomorphism.
f) Let x ∈ LE(P, Q) be an isomorphism. By b), xij ∈F for all i, j ∈ I and the assertion follows from e).
g) Let v : P → Q be an isomorphism of Hilbert right E-modules and x:=ψvϕ∗.By b) and e),xij ∈qiEpj ⊂qiF pj for all i, j∈I and
j, l∈I ⇒
E¨
X
m∈I
x∗mjxml =δjlpj, i, k ∈I ⇒
E¨
X
n∈I
xinx∗kn=δikqi.
By ([C1], Corollary 4.4.4.9)), we may replace ¨E by ¨F in the above sums and the assertion follows from c) and e) applied to F instead ofE.
2. SELFDUAL HILBERT RIGHTW∗-MODULES
We start with
Lemma 2.1. Let E be a W∗-algebra, (pi)i∈I a family in PrE, H :=
E⊗l2(I), F :=LE(H), and ϕ: W|
i∈I
E →H, ξ7→
H¨
X
i∈I
ξi⊗ei
the canonical isomorphism of Hilbert right E-modules (Theorem 1.4 a)). For every i∈I put ωi :=h · |eiiei ∈Pr(L(l2(I))).Then
p:=
F¨
X
i∈I
pi⊗ωi∈PrF and Imp=ϕ W
|
i∈I
piE
.
Proof. (pi ⊗ωi)i∈I is a family in PrF such that (pi⊗ωi)(pj ⊗ωj) = 0 for all distincti, j ∈I, so thatpis well-defined and belongs to PrF. Fori∈I and ξ ∈ W|
i∈I
E, by ([C1], Proposition 5.6.3.4 c)),
(pi⊗ωi)ϕξ=
H¨
X
j∈I
(pi⊗ωi)(ξj⊗ej) =
H¨
X
j∈I
(piξj)⊗(ωiej) = (piξi)⊗ei. For a∈E¨ and η∈H we have
Dpϕξ, (a, η)] E
=D
p,(a, ϕξ, η)^ E
=X
i∈I
Dpi⊗ωi,(a, ϕξ, η)^ E
=
=X
i∈I
D
(pi⊗ωi)ϕξ, (a, η)] E
=X
i∈I
D
(piξi)⊗ei,(a, η)] E , so that
pϕξ =
H¨
X
i∈I
(piξi)⊗ei and Imp=ϕ W
|
i∈I
piE
.
Lemma 2.2. Let E be a C∗-algebra, H a Hilbert right E-module, p ∈ PrLE(H), and K:= Imp.
a) Kerp=K⊥. b) K⊥⊥=K.
c) If H is selfdual, then K is selfdual, too.
Proof. a) Take ξ ∈ Kerp. Then hξ|ηi = hξ|pηi = hpξ|ηi = 0 for η∈K, so that ξ∈K⊥.
Conversely, assume ξ ∈K⊥. Then 0 = hξ|ηi =hξ|pηi =hpξ|ηi for η∈K, so that pξ∈K⊥. Since pξ ∈K, we have pξ = 0,ξ ∈Kerp.
b) K ⊂K⊥⊥ is trivial. Let ξ ∈K⊥⊥ and put η:=ξ−pξ ∈K⊥⊥. Then pη=pξ−p2ξ = 0, η∈Kerp=K⊥
by a). It follows thatη= 0 and ξ =pξ ∈K.
c) Let u ∈ LbE(K, E). Put v := u◦p ∈ LbE(H, E). Since H is selfdual there is η∈H withv=h · |ηi.For ξ∈K we have
uξ=up2ξ=vpξ=hpξ|η i=hξ|pηi, so that u=h · |pηi and K is selfdual.
Definition 2.3. LetEbe aC∗-algebra andH, KHilbert rightE-modules.
Write HK if there is a Hilbert rightE-moduleL such thatK ≈H| L.
Proposition 2.4. Let E be a C∗-algebra, H a Hilbert right E-module, F :=LE(H), p, q∈PrF, and i:pH →H, j:qH →H the inclusion maps.
a)We have i∈ LE(pH, H). Withi∗:H →pH,ξ 7→pξ, we have ii∗=p and i∗i= 1H.
b) For every u ∈ qF p and v ∈ pF q put u¯ := j∗ui, ¯v := i∗vj. For u∈qF p, v∈pF q, and ξ∈pH we have
¯
u∈ LE(pH, qH), u¯∗=u∗, uξ¯ =uξ, uv = ¯u¯v.
c) The map
qF p→ LE(pH, qH), u7→u¯
is an isometry of Banach spaces. If p = q then it even is an isomrphism of C∗-algebras.
d) Foru∈qF pwe have u∗u=p iffhuξ¯ |uη¯ i=hξ|ηi for allξ, η∈pH.
In this case, pq.
e) For u∈qF p, u¯ is an isomorphism of Hilbert right E-modules iff u∗u=p and uu∗=q.
f) The statements below are equivalent.
f1) pq.
f2) There is an r∈PrF such that qH ≈(pH)| (rH).
f3) pH qH.
g) IfE is aW∗-algebra andH is selfdual, then the statements below are equivalent.
g1) pH ≈qH.
g2) pH qH, qH pH.
Proof. a) For (ξ, η) ∈ pH×H we have hiξ|ηi = hξ|ηi =hpξ|ηi = hξ|pη i,so thati∈ LE(pH, H),ii∗ =p and i∗i= 1H.
b) By a),
¯
u∈ LE(pH, qH), u¯∗ =i∗u∗j=u∗,
¯
uξ=j∗uiξ=j∗uξ=j∗quξ =quξ =uξ,
¯
u¯v=j∗uii∗vj =j∗upvj=j∗uvj =uv.
c) For u∈qF p and ξ∈H, we have uξ=upξ, so that by b), k¯uk= sup
ξ∈(pH)#
kuξk= sup
ξ∈H#
kuξk=kuk, i.e., the map preserves the norms.
Letw∈ LE(pH, qH). Putu:=jwi∗.By a),u∈qF pand ¯u=j∗jwi∗i= w,so that the map is surjective. The last assertion follows from b).
d) By b),
huξ¯ |uη¯ i=huξ|uη i=hu∗uξ|ηi.
If u∗u = p then huξ¯ |uη¯ i = hξ|ηi. If huξ¯ |uη¯ i = hξ|ηi for all ξ, η ∈pH, then
hu∗uζ|ηi=hu∗upζ|ηi=hpζ|ηi, u∗uζ=pζ, u∗u=p forζ ∈H. The last assertion follows fromp∼uu∗=quu∗ ≤q.
e) follows from d).
f) f1)⇒f2) There isr ∈PrF withp∼r≤q.By e), qH ≃rH| (q−r)H =pH| (q−r)H.
f2)⇒f3) is trivial.
f3) ⇒ f1) There is a Hilbert right E-module K and an isomorphism ϕ:qH →(pH)| K of Hilbert right E-modules. Put
ψ: (pH)| K →pH, (ξ, η)7→η.
Then
ψ∗:pH →(pH)| K, ξ7→(ξ,0), ψ∗ψ: (pH)| K→(pH)| K, (ξ, η)7→(ξ,0),
ψψ∗ :pH →pH, ξ7→ξ.
By c), there isu∈qF pwith ¯u=ϕ−1◦ψ∗.By b) and ([C1], Corollary 5.6.1.12)), u∗u=u∗u¯= ¯u∗u¯=ψ◦ϕ◦ϕ−1◦ψ∗ =ψ◦ψ∗,
uu∗= ¯uu∗ = ¯uu¯∗ =ϕ−1◦ψ∗◦ψ◦ϕ≤ϕ−1◦ϕ, so that, by c),
u∗u=p, uu∗ ≤q, pq.
g) By ([C1], Theorem 5.6.3.5 b)), F is a W∗-algebra. By f3) ⇒ f1) we have pq and q p,so that p∼q and the assertion follows from e).
Lemma 2.5. Let E be aC∗-algebra, H a Hilbert right E-module, andH the set of complementable submodules of H ordered by the inclusion relation.
a) The map
PrLE(H)→H, p7→pH := Imp is an isomorphism of ordered sets.
b) Forp, q∈ LE(H),
p∼q ⇔pH ≈qH, pq ⇔pHqH.
Proof. a) By Lemma 2.2 a), pH ∈H for every p∈ LE(H). The map is surjective since every element of His complemetable. Let p, q∈PrLE(H). If p ≤ q then qξ = qpξ = pξ = ξ for ξ ∈ pH, so that pH ⊂ qH and the map preserves the order. Conversely, ifpH ⊂qH thenqpξ =pξ forξ ∈H, so that qp=p and p≤q. Thus the map is an isomorphism of ordered sets.
b) By Proposition 2.4 e), p ∼ q ⇔ pH ≈ qH. If p q then there is r ∈ LE(H) withp∼r≤q. By the above and a),
pH≈rH ⊂qH, qH =rH| qH∩(rH)⊥
, pH qH.
IfpH qH then there is a Hilbert rightE-moduleKsuch thatqH ≈K| pH.
Using this isomorphism we obtain an L ∈ H with pH ≈ L ⊂ qH. By a), pq.
Lemma 2.6. If E is a C∗-algebra, then the statements below are equiva- lent for all Hilbert right E-modules H and K.
a) There is a Hilbert right E-moduleL such that K≈H| L.
b) There is a surjective u∈ LE(K, H).
c) There is an injective v∈ LE(H, K) with closed range.
Proof. a)⇒ b) and a)⇒ c) are obvious.
b)⇒a). Sinceuis surjective,u∗is injective. By ([C1], Corollary 5.6.1.12)), u∗u∈ LE(K)+ and put |u|:= (u∗u)1/2 ∈ LE(K)+.Let ξ ∈ K with|u|ξ = 0.
Then u∗uξ=|u|2ξ = 0 anduξ= 0, since u∗ is injective. Thus, the map w: Im|u| →H, |u|ξ 7→uξ
is well-defined. For ξ, η ∈ K we have h |u|ξ| |u|ηi = hu∗uξ|ηi =huξ|uηi, i.e.,wpreserves the scalar products. Being surjective, Im|u|is a closed Hilbert E-submodule of K and Im|u| ≈H. By ([W], Corollary 15.3.9)), Im|u| is complementable and we get
H= Im|u| | (Im|u|)⊥≈H| (Im|u|)⊥.
c) ⇒a). By ([W], Corollary 15.3.9), Imv is complementable, i.e., K= Imv| (Imv)⊥.
By b) ⇒ a),H ≈Imv, so thatK ≈H| (Imv)⊥.
Proposition 2.7. If E is a W∗-algebra then the statements below are equivalent for all selfdual Hilbert right E-modules H andK.
a) H≈K.
b) HK and K H.
c) LE(H, K) has a bijective map.
d) LE(H, K)has a surjective map and an injective map with close range.
Proof. a)⇒ b) and a)⇒ c)⇒ d) are trivial.
d) ⇒ a) follows from Lemma 2.6, b) ⇒ a), c)⇒ a).
b) ⇒ a). Put L:=H| K.By Lemma 2.5 a), there arep, q∈PrLE(L) with H=pL,K=qL.By Lemma 2.5 b), we get successively
pq, qp, p∼q, H≈K.
Remark. The example
K→K, ξ 7→2ξ
shows that a bijective map ofLE(H, K) is not necessarily an isomorphism.
Corollary 2.8. If E is a W∗-algebra then the statements below are equivalent for all Hilbert right E-modules H, K, L.
a) K ≈H| L.
b) There is an exact sequence
0→H→K →L→0 with adjointable homomorphisms.
Proof. (a)⇒ (b) is trivial.
(b) ⇒ (a). Let u∈ LE(H, K), v ∈ LE(K, L) be the homomorphisms of b). It follows from Imu = Kerv that u has closed range. The restriction of v to (Imu)⊥ = (Kerv)⊥ is bijective and so, by Proposition 2.7, c) ⇒ a), we have
H ≈Imu, L≈(Imu)⊥. Thus,K ≈Imu| (Imu)⊥≈H| L.
Lemma 2.9. LetE be aC∗-algebra,r ∈PrE,F :=rEr, andϕ:F →E the inclusion map.
a) If p∈PrE,q ∈PrF, and pϕq, then there isq′ ∈PrF with q′ ≤q, p∼ϕq′.
b) If p, q∈PrF then
p∼q⇔ϕp∼ϕq, pq ⇔ϕpϕq.
c) If E is a W∗-algebra then c1) ϕ defines an injective map
ψ: (PrF/∼) → (PrE/∼);
c2)ψ(PrF/∼)is a hereditary set of(PrE/∼), to mean thatq ∈(PrE/∼) belongs to ψ(PrF/∼) if there is p∈(PrF/∼) withq ψ(p);
c3) the map (PrF/∼)→ ψ(PrF/∼) defined by ψ is an isomorphism of ordered sets;
c4) if (PrE/∼) is order complete, then so is (PrF/∼), and ψ comutes with the suprema and infima.
Proof. a) By the hypothesis, there isp′ ∈PrEwithp∼p′ ≤ϕq.SinceF is a hereditaryC∗-subalgebra ofE ([C1], Example 4.3.4.2)), there isq′∈PrF with q′ ≤q, ϕq′ =p′ ∼p.
b) Assume ϕpϕq. Then there is u ∈E with u∗u=ϕp,u u∗ ≤ϕq. It follows that
u=u ϕp∈Er, u=u u∗u= (ϕq)u u∗u= (ϕq)u∈rE, u∈rEr=F, pq.
The implication
ϕp∼ϕq⇒p∼q is proved similarly. The other implications are trivial.
c) follows from a) and b). More precisely, c1) follows from b).
c2) follows from a).
c3) follows from b) and c2).
c4) follows from c2) and c3).
Definition 2.10. For every W∗-algebra E we denote by H(E) the class of all selfdual Hilbert right E-modules (with identification of the isomorphic ones) endowed with the associative and commutative addition | and the order relation(Proposition 2.7, b)⇒a)). {0}is the neutral element for the addition and the smallest element with respect to the order relation.
Theorem 2.11. Let E be a W∗-algebra and for every setI consider the selfdual Hilbert right E-module
E⊗l2(I)≈ W|
i∈I
E (Theorem1.4 a)) and the W∗-algebra
EI :=E⊗ L(l2(I))≈ LE(E)⊗ L(l2(I))≈ LE⊗¯K(E⊗l2(I))≈ LE W
|
i∈I
E
([C2], Theorem2.4 d)).
a) ForI ⊂J the natural imbedding ψIJ : W|
i∈I
E → W|
j∈J
E
produces the map
ϕIJ :EI→EJ, u7→ψIJu ψIJ∗ .
ϕIJ is an injective W∗-homomorphism and its image isrIJEJrIJ, where rIJ :=
E¨J
X
i∈I
1E⊗ h · |eiiei.
b) (PrEI)I and the associated maps (ϕIJ)IJ form an inductive system.
For p ∈ PrEI, q ∈ PrEJ write p ∼ q if there is a set K ⊃ I ∪ J with ϕIKp∼ϕJKq. ThenϕILp∼ϕJLq for every L⊃I∪J.
Denote by PR(E)the class of S
IPrEIwith the identification of the ele- mentsp∼q;PR(E)is in some sense the inductive limit of the above inductive system.
c) For p ∈ PrEI, q ∈ PrEJ, write p q if there is K ⊃ I ∪J with ϕIKpϕJKq. Then ϕILpϕJLq for every L⊃I∪J and is an order relation on PR(E).
d) For every p∈PrEI we have Imp=p
W |
i∈I
E
∈ H(E) and, for p∈PrEI and q∈PrEJ,
p∼q⇔Imp≈Imq, pq⇔ImpImq while the deduced map
PR(E)→ H(E), p7→Imp
is bijective and we may identify PR(E) and H(E) as ordered classes.
e) If PR(E) is order complete (equivalently: H(E) is order complete) then (PrE/∼) is order complete and may be identified as ordered set with a hereditary subset of PR(E).
f) Fix a set I0. For every I ⊃I0 we have
EI=E⊗ L(l2(I))≈E⊗ L(l2(I0))⊗ L(l2(I\I0))≈EI0⊗ L(l2(I\I0)).
By this relation, PR(E) and PR(EI0) may be identified.
Proof. a) We have
ψIJ∗ ψIJ = 1l2(I), ψIJψIJ∗ =rIJ, ψIJ∗ rIJ =ψ∗IJ, rIJψIJ =ψIJ and, for u∈EI,
ψIJ∗ (ϕIJu)ψIJ =ψ∗IJψIJu ψIJ∗ ψIJ =u, rIJ(ϕIJu)rIJ =rIJψIJu ψIJ∗ rIJ =ψIJu ψIJ∗ ,
i.e., ImϕIJ ⊂rIJEJr∗IJ.Ifv∈rIJEJrIJ thenψ∗IJv ψIJ ∈EI and ϕIJ(ψIJ∗ v ψIJ) =ψIJψ∗IJv ψIJψIJ∗ =rIJv rIJ =v, i.e., ImϕIJ=rIJEJrIJ.Moreover, foru∈EIand (a, ξ, η)∈E¨×W|
j∈J
E×W|
j∈J
E we have
D
ϕIJu,(a, ξ, η)^ E
=h h(ϕIJu)ξ|ηi, ai=h hψIJu ψIJ∗ ξ|ηi, ai=
=h hu ψIJ∗ ξ|ψIJ∗ ηi, ai=D u,
z ^}| { (a, ψ∗IJξ, ψ∗IJη)E
, i.e.,
ϕ′IJ(a, ξ, η) =^
z ^}| { (a, ψ∗IJξ, ψ∗IJη)
and ϕIJ is a W∗-homomorphism ([C1], Theorem 5.6.3.5 b) and Proposition 4.4.4.6)).
b) It follows from a) (the last assertion of b) from the last assertion of a)).
c) It follows from a) and b).
d) By Lemma 2.2, Imp ∈ H(E). The relations follow from Proposi- tion 2.4 c), e), f), g). By Proposition 2.1 and ([C1], Theorem 5.6.4.10 a)), the map is surjective.
e) It follows from a) and Lemma 2.9 c).
f) This is obvious.
3. SOME TECHNICAL RESULTS
We start with
Lemma 3.1. Let E be a W∗-algebra, α an ordinal number, (pγ)γ∈α a family in PrE, β the smallest element of α with W
β∈γ∈αpγ 6≤ pβ, F :=
E⊗ L(l2(α)), and ωγ := h · |eγ ieγ ∈ PrL(l2(α)) for every γ ∈ α. Then there is a family (qγ)γ∈α such that
1. γ∈β ⇒qγ =pγ, 2. pβ ≤qβ,
3. β∈γ ∈α⇒qγ≤pγ, 4. γ≤β ⇒W
γ∈δ∈αqδ≤qγ, 5.(qβ−pβ)⊗ωβ ∼
F¨
P
γ∈α\{β}
(pγ−qγ)⊗ωγ, 6.
F¨
P
γ∈α
pγ⊗ωγ ∼
F¨
P
γ∈α
qγ⊗ωγ.