REAL W
∗-ALGEBRAS
CORNELIU CONSTANTINESCU
We give a new proof of Theorem 6 in [1] stating that a semi-discrete, semi- continuos real W∗-algebra (Definition 2) is isomorphic to a real W∗-algebra of the formC(T,H) (the set of continuous H-valued functions onT), where T is a compact hyperstonian space andHis the field of quaternions.
AMS 2000 Subject Classification: 46L10, 46L35.
Key words: realW∗-algebra.
1. INTRODUCTION
In general we use the notation and terminology of [2]. We give below a list of some notation used in this paper.
1. R(resp. H) denotes the field of real numbers (resp. quaternions). If T is a locally compact space, then C(T,R) (resp. C(T,H)) denotes the real C∗-algebra of continuous R-valued (resp. H-valued) functions onT.
2. We put
ReE :={x∈E |x∗ =x}, Ec :={x∈E |y∈E ⇒xy=yx}, PrE:={p∈ReE |p2 =p}
for every C∗-algebraE. If E is unital, then 1E denotes its unity.
3. A realC∗-algebraEis called a GelfandC∗-algebra ifE= ReE=Ec. 4. IfE is aW∗-algebra, then ¨E denotes its predual. If (xi)i∈I is a family inE then we denote by
E¨
P
i∈I
xi the element ofE (if it exists) determined by the relation
a∈E¨ ⇒
* E¨
X
i∈I
xi, a +
=X
i∈I
hxi, ai.
Lemma 1. Let E be a C∗-algebra such that ReE is commutative. Then (a) ReE⊂Ec;
(b)x∈E⇒x∗x=xx∗;
REV. ROUMAINE MATH. PURES APPL.,54(2009),2, 125–129
(c) p, q∈PrE, p∼q ⇒p=q;
(d)if E is not commutative, then there are x, y∈E with x∗ =−x, y∗ =−y, xy 6=yx.
Proof. (a)
Step1. x∈ReE,y∈E,y∗ =−y⇒xy =yx.
Putz:=xy−yx.We have successively
z∗ =y∗x∗−x∗y∗ =−yx+xy =z, z∈ReE,
x2y−xyx=x(xy−yx) =xz=zx= (xy−yx)x=xyx−yx2, x2y+yx2 = 2xyx.
Since y2∈ReE, we have xy2=y2x and
z2 = (xy−yx)(xy−yx) =xyxy−xy2x−yx2y+yxyx=
= 1
2(x2y+yx2)y−x2y2−yx2y+1
2y(x2y+yx2) = 0.
Thus,z= 0 and xy =yx.
Step2. ReE⊂Ec
Let x∈ReE andy ∈E. Then (y−y∗)∗=−(y−y∗) so, by Step 1, x(y−y∗) = (y−y∗)x.
Since y+y∗ ∈ReE,
x(y+y∗) = (y+y∗)x, xy=yx, x∈Ec. (b) By (a), since x+x∗ ∈ReE,
x2+xx∗=x(x+x∗) = (x+x∗)x=x2+x∗x, xx∗ =x∗x.
(c) This follows from (b).
(d) SinceE is not commutative, there areu, v∈E withuv 6=vu. Put x:= 1
2(u−u∗), y:= 1
2(v−v∗), a:= 1
2(u+u∗), b:= 1
2(v+v∗).
Then
x∗ =−x, y∗=−y, a, b∈ReE, x=u−a, y=v−b.
By (a),
xy−yx= (u−a)(v−b)−(v−b)(u−a) =
=uv−ub−av+ab−vu+va+bu−ba=uv−vu6= 0.
Definition 2 ([3], Definition 8.1.5). Let E be a real W∗-algebra. A p∈PrEis calledsemi-abelianifp(ReE)pis commutative. Eis calledsemi- discrete if for every z ∈ PrEc \ {0} there is a semi-abelian p ∈ PrE\ {0}
withp≤z. Eis calledsemi-continuousifp∈PrE, pabelian impliesp= 0.
Proposition 3. Let E be a real W∗-algebra such that ReE is commu- tative.
(a) ReE is aW∗-subalgebra of E and it is isomorphic to C(T,R), where T is the spectrum of ReE and a hyperstonian space.
(b) If E is not commutative then there are i, j ∈E and p ∈PrE\ {0}
with
i∗=−i, j∗=−j, i∗i=j∗j=p, ij =−ji.
Proof. (a). ReE is obviously the greatest Gelfand C∗-subalgebra of E.
By [2], Corollary 4.4.4.12 b, it is a W∗-subalgebra of E. The last assertion follows from [2], Corollary 4.4.1.10.
(b) By Lemma 1 (d), there arex, y∈E with x∗=−x, y∗ =−y, xy6=yx.
Put
a:=xy−yx, b:=xy+yx, c:=xyxy+yxyx.
Then
a∗=−a6= 0, b, c∈ReE.
By Lemma 1 (a),
a∗a= (yx−xy)(xy−yx) =yx2y−yxyx−xyxy+xy2x= 2x2y2−c, b2= (xy+yx)(xy+yx) =xyxy+xy2x+yx2y+yxyx=x2y2+c, so,
06=a∗a= 4x∗xy∗y−b2≤4x∗xy∗y.
Using (a), we identify ReE with C(T,R) and find p ∈ PrE\ {0} and ε > 0 with
{p >0} ⊂ {a∗a > ε} ∩ {x∗x > ε}.
Define
f :T →R, t7→
( √ 1
(x∗x)(t) ifp(t) = 1 0 ifp(t) = 0 , g:T →R, t7→
( 2
q(x∗x)(t)
(a∗a)(t) ifp(t) = 1 0 ifp(t) = 0
,
h:T →R, t7→
( 1
√
(a∗ax∗x)(t) ifp(t) = 1 0 ifp(t) = 0 , i:=f x, j:=bhx+gy.
Then f, g, h∈ReE,hx∗x= 12g, and, by Lemma 1 (a), i∗=−i, j∗ =−j, i∗i=f2x∗x=p,
j∗j= (bhx∗+gy∗)(bhx+gy) =b2h2x∗x+bhgx∗y+gbhy∗x+g2y∗y=
=b2h2x∗x−2b2h2x∗x+ 4h2(x∗x)2y∗y =x∗xh2(4x∗xy∗y−b2) =x∗xh2a∗a=p, ij+ji=f bhx2+f gxy+bhf x2+gf yx=
= 2bhf x2+f gb=−2bhf x∗x+ 2f bhx∗x= 0.
Theorem 4 ([1], Theorem 6). The statements below are equivalent for every real W∗-algebra E.
(a)E is semi-discrete and semi-continuous.
(b)E is isomorphic toC(T,H), whereT is a compact hyperstonian space.
Proof. (a)⇒(b). Let (pλ)λ∈Λbe a maximal orthogonal family in PrE\ {0}such that for every λ∈Λ there areiλ, jλ ∈E with
i∗λ =−iλ, jλ∗=−jλ, i∗λiλ =jλ∗jλ =pλ, iλjλ=−jλiλ. Put (see [2], Proposition 5.6.3.20 b⇒a)
p:=
E¨
X
λ∈Λ
pλ, i:=
E¨
X
λ∈Λ
iλ, j:=
E¨
X
λ∈Λ
jλ, k:=ij.
If p 6= 1E then by Proposition 3 (b) there are p0 ∈ PrE \ {0} and i0, j0 ∈ E with
p0 ≤1E−p, i0∗=−i0, j0∗=−j0, i0∗i0=j0∗j0 =p0, i0j0 =−j0i0 and this contradicts the maximality of (pλ)λ∈Λ.
By [2], Proposition 5.6.3.20 c,
i∗ =−i, j∗=−j, i∗i=j∗j= 1E, ij=−ji.
It follows that
k∗ =−k, jk=−kj=i, ki=−ik=j, k∗k= 1E. Thus, we may identify C(T,H), where T is the spectrum of ReE, with
{a1E +bi+cj+dk|a, b, c, d∈ReE}.
By Proposition 3 (a), T is a compact hyperstonian space.
Letx∈E. We have to show that x∈ C(T,H). Since ReE ⊂ C(T,R)⊂ C(T,H), we may assume x∗ =−x. Put
a:= 1
2(ix+xi), b:= 1
2(jx+xj), c:=ai+bj+x, d:=ijc.
Then a, b∈ReE, c∗ =−c, and we deduce (by Lemma 1 (a)) that ic+ci=ai2+bij+ix+ai2+bji+xi=−2a+ 2a= 0,
jc+cj =aji+bj2+jx+aij+bj2+xj=−2b+ 2b= 0, d∗ =c∗j∗i∗=−cji=ijc=d, d∈ReE,
d=aiji+bij2+ijx=aj−bi+kx, kd=−ai−bj−x, x=−ai−bj−kd∈ C(T,H).
(b)⇒ (a) is obvious.
REFERENCES
[1] Sh.A. Ayupov, A.A. Rakhimov and A.Kh. Abduvaitov,RealW∗-algebras with Abelian Hermitian part.Math. Notes71(2002),3, 432–435.
[2] Corneliu Constantinescu,C∗-Algebras. Elsevier, 2001.
[3] Bingren Li,Real Operator Algebras. World Scientific, 2003.
Received 9 November 2007 Bodenacherstr. 53
CH 8121 Benglen constant@math.ethz.ch