NONCOMMUTATIVE Lp-SPACE AND OPERATOR SYSTEM

1 downloads 0 Views 160KB Size Report
Jun 28, 2009 - [Te1] M. Terp, Lp spaces associated with von Neumann algebras, Math. ... [Te2] M. Terp, Interpolation spaces between a von Neumann algebra ...
NONCOMMUTATIVE Lp -SPACE AND OPERATOR SYSTEM

arXiv:0904.0518v2 [math.OA] 28 Jun 2009

KYUNG HOON HAN Abstract. We show that noncommutative Lp -spaces satisfy the axioms of the (nonuni1 tal) operator system with a dominating constant 2 p . Therefore, noncommutative Lp 1 spaces can be embedded into B pH q 2 p -completely isomorphically and complete order isomorphically.

1. Introducton A unital involutive subspace of B pH q had been abstractly characterized by Choi and Effros [CE]. Their axioms are based on observations of the relationship between the unit, the matrix order, and the matrix norm of the unital involutive subspace of B pH q. Indeed, the unit and the matrix order can be used to determine the matrix norm by applying

 0 x }x}  inf tλ ¡ 0 : λI ¤ x 0 ¤ λI u for a unital involutive subspace X of B pH q and x P Mn pX q. The abstract characterization of a nonunital involutive subspace of B pH q was completed by Werner [W]. The axioms are based on observations of the relationship between the matrix norm and the matrix order of the involutive subspace of B pH q. Since the unit may be absent, the above equality is replaced by

 0 x }x}  supt|ϕp x 0 q| : ϕ P M2n pX q1, u

for an involutive subspace X of B pH q and x P Mn pX q. We denote by M2n pX q1, the set of positive contractive functionals on M2n pX q. A complex involutive vector space X is called a matrix ordered vector space if for each n P N, there is a set Mn pX q € Mn pX qsa such that (1) (2) (3)

Mn pX q X rMn pX q s  t0u for all n P N, Mn pX q ` Mm pX q € Mn m pX q for all m, n P N, γ  Mm pX q γ € Mn pX q for each m, n P N and all γ

P Mm.,npCq.

One might infer from these conditions that Mn pX q is actually a cone. An operator space X is called a matrix ordered operator space iff X is a matrix ordered vector space and for every n P N, (1) the -operation is an isometry on Mn pX q and (2) the cones Mn pX q are closed.

2000 Mathematics Subject Classification. 46L07, 46L52, 47L07. Key words and phrases. noncommutative Lp -space, operator system. This work was supported by the BK21 project of the Ministry of Education, Korea. 1

2

KYUNG HOON HAN

Suppose X is a matrix ordered operator space. For x P Mn pX q, the modified numerical radius is defined by

 0 x q| : ϕ P M2npX q1, u. νX pxq  supt|ϕp  x 0

We call a matrix ordered operator space an operator system iff there is a k that for all n P N and x P Mn pxq,

¡ 0 such

}x} ¤ kνX pxq.

Since νX pxq ¤ }x} always holds, we can say that an operator system is a matrix ordered operator space such that the operator space norm and the modified numerical radius are equivalent uniformly for all n P N. Werner showed that X is an operator system if and only if there is a complete order isomorphism Φ from X onto an involutive subspace of B pH q, which is a complete topological onto-isomorphism [W]. Hence, the operator system is an abstract characterization of the involutive subspace of B pH q in the completely isomorphic and complete order isomorphic sense. In this paper, noncommutative Lp -space is meant in the sense of Haagerup, which is based on Tomita-Takesaki theory [Te1]. While the noncommutative Lp -spaces arising from semifinite von Neumann algebras are a natural generalization of classical Lp -spaces [S, Ta], noncommutative Lp -spaces arising from type III von Neumann algebras are quite complicated. Recently, the reduction method approximating a general noncommutative Lp -space by tracial noncommutative Lp -spaces has been developed [HJX]. Noncommutative Lp -spaces can also be obtained by the complex interpolation method. There exist a canonical matrix order and a canonical operator space structure on each noncommutative Lp -space. The matrix order is given by the positive cones Lp pMn pMqq , and the operator space structure is given by the complex interpolation Mn pLp pMqq  pMn pMq, Mn pMop  qq 1 . p

We refer to [K, Pi1, Te2] for the details, and the reference [JRX, Section 4] is recommended for the summary. In particular, its discrete noncommutative vector valued Lp -space Spn pLp pMqq is given by Spn pLp pMqq  Lp pMn pMqq.

We refer to [Pi2] for general information on noncommutative vector valued Lp -spaces. The purpose of this paper is to show that noncommutative Lp -spaces satisfy the 1 axioms of the operator system with a dominating constant 2 p . As a corollary, we obtain the following embedding theorem: noncommutative Lp -spaces can be embedded 1 into B pH q 2 p -completely isomorphically and complete order isomorphically. 2. Noncommutative Lp -space and operator system If X is a matrix ordered operator space, we regard Spn pX q as an operator space having the same matrix order structure with Mn pX q. That is, we do not distinguish between Spn pX q and Mn pX q as matrix ordered vector spaces. For a noncommutative Lp -space Lp pMq, it is more efficient to describe Spn pLp pMqq rather than Mn pLp pMqq. So, we change the modified numerical radius into a form more

NONCOMMUTATIVE Lp -SPACE AND OPERATOR SYSTEM

3

adequate to noncommutative Lp -spaces. For a matrix ordered operator space X and p x P Mn pX q, we define νX pxq as p νX



pxq  2 p1 supt|ϕp x0 x0

q| : ϕ P Sp2npX q1, u,

where Sp2n pX q1, denotes the set of positive contractive functionals on Sp2n pX q. Note that a functional ϕ is positive on Sp2n pX q if and only if it is positive on M2n pX q because Sp2n pX q and M2n pX q have the same order structure. According to the following lemma, the uniform equivalence of a matrix norm and νX can be confirmed by showing the p uniform equivalence of the Spn -norm and νX . Lemma 2.1. Suppose X is a matrix ordered operator space, and there is a constant p k ¡ 0 satisfying }x}SpnpX q ¤ kνX pxq for any x P Mn pX q. Then we have

}x}M pX q ¤ kνX pxq n

for any x P Mn pX q. Hence, X is an operator system. n Proof. For a, b P pS2p q1 , we have

}



a 0 0 b



}S  p}a}2pS 2n 2p

From [Pi2, Theorem 1.5], the map



x11 x12 x21 x22



n 2p



Þ 2 p1 ϕp a0 b0 Ñ

}b}2pS q ¤ 2 n 2p



1 2p

x11 x12 x21 x22



1 2p

.

a 0 0 b



q

for ϕ P Sp2n pX q1, defines a positive contractive functional on M2n pX q. It follows that





| : ϕ P M2npX q1, u

 



 a 0 0 x  ¥ 2 supt|ϕp 0 b x 0 a0 0b q| : ϕ P Sp2npX q1, , a, b P pS2pn q1u 

0 axb   2 supt|ϕp b xa 0 | : ϕ P Sp2npX q1, , a, b P pS2pn q1 u  suptνXp paxbq : a, b P pS2pn q1 u ¥ k1 supt}axb}S pX q : a, b P pS2pn q1u  k1}x}M pX q for x P Mn pX q. For the last equality, see [Pi2, Lemma 1.7].  Lemma 2.2. Suppose a is a bounded linear operator on a Hilbert space and a  v |a| is its polar Let f be a continuous function defined on r0, 8q with f p0q  0.  decomposition.

| a | a Then is a positive operator and the equality a |a| 



1 |a | a 1 f p|a |q f p|a |qv fp q  2 vf p|a|q f p|a|q 2 a |a| νX pxq  supt|ϕp

0 x x 0

1 p

1 p

n p

n

holds.

4

KYUNG HOON HAN

Proof. The positivity follows from [Pa, Exercise 8.8 (vi)]. By using the Weierstrass theorem and the right polar decomposition a  |a |v, it is sufficient to show that the equality   n 

|a | a  2n1 |a|n |a|n1a a |a| |a|n a |a |n1 holds for n ¥ 1. We proceed by mathematical induction. Since a paa qk a  pa aqk 1 for k ¥ 0, we have n1 a |a |n1a  pa aq 2 1  |a|n 1. Similarly, apa aqk  paa qk a implies a|a|n

It follows that





 |a|na.

|a|n |a|n1a a |a|n a |a |n1  n 1 n 1

a | | a | | a |n a a|a|n | n1  2 a|a|n |a|a|a|n1 a|a|n1a |a|n 1  n 1 n

n |a |  2 a|a|n ||aa|n| a1 .

|a|

a |a|

2 



n 1

 In order to prove that the Schatten p-class Sp is an operator system, Lemma 2.2 is sufficient. However, in order to prove that a general noncommutative Lp -space is an operator system, an unbounded version of Lemma 2.2 is needed. We denote the Dirac measure massed at the zero point by δ0 . Lemma 2.3. Let a be a closed densely defined ³8 operator on a Hilbert space. Suppose a  v |a| is a polar decomposition and |a|  0 tdE is a spectral decomposition. Then

1 1 vv  v   2 2 δ0  1 v I  12 vv 2   2 |a | a . is a spectral measure corresponding to 12 a |a| Proof. Let 

1 vEv  vE 1 E  . 2 Ev  v  vE Since v  v  E pp0, 8qq, v  v commutes with E p∆q for any Borel set ∆ in r0, 8q. For all Borel sets ∆1 and ∆2 in r0, 8q, we have E 1 p∆1 qE 1 p∆2 q 



1 vE p∆1 qv  vE p∆1 q vE p∆2 qv  vE p∆2 q  4 E p∆1qv vvE p∆1q E p∆2qv vvE p∆2q

 1 vE p∆1 qv  vE p∆2 qv  vE p∆1 qE p∆2 qv  vE p∆1 qv  vE p∆2 q vE p∆1 qv  vE p∆2 q  4 E p∆1 qvvE p∆2qv vvE p∆1 qE p∆2 qv E p∆1 qvvE p∆2 q vvE p∆1qvvE p∆2q

 1 vE p∆1 X ∆2 qv  vE p∆1 X ∆2 q  2 E p∆1 X ∆2qv vvE p∆1 X ∆2 q E 1p∆1 X ∆2 q. r1 E



vEv  vE Ev  v  vE





I

NONCOMMUTATIVE Lp -SPACE AND OPERATOR SYSTEM

5

Taking ∆1  ∆2 , we see that E 1 is projection valued. The countable additivity with respect to strong operator topology is obvious. Since E pt0uq  I  v  v, we have E 1 pt0uq  0.

r satisfies all the conditions of a spectral measure. From these, it is easy to check that E We still need to show that 

  » 8 1 |a | a h h x | y  zdErξ,ξ  a k k | a | 2 0 

for ξ

 hk and h P Dpaq, k P Dpaq. For the left-hand side, we compute    

x |aa | |aa| hk | hk y  x|a|h|hy xak|hy xah|ky x|a|k|ky  xv|a|vh|hy xv|a|k|hy x|a|vh|ky xvv|a|k|ky »8 »8 tdEk,v h  tdEv h,vh 0 0 »8 »8 tdEv h,k

0

For the right-hand side, we compute

 1 vE p∆qv  1 Eξ,ξ p∆q  x 

 

vE p∆q v  vE p∆q

tdEk,v vk .

0

 

h h | y k k

2 E p∆qv 1 pxvE p∆qvh|hy xvE p∆qk|hy xE p∆qvh|ky xvvE p∆qk|kyq 2 1 pEv h,vhp∆q Ek,vhp∆q Ev h,k p∆q Ek,vvk p∆qq. 2 

Lemma 2.4. Suppose M is a semifinite von Neumann algebra with a faithful semifinite normal trace τ and a is a τ -measurable operator with its polar decomposition   a  v|a|. |a | a is a Let f be a continuous function defined on r0, 8q with f p0q  0. Then a |a| positive self-adjoint τ2 -measurable operator and the equality 1 fp 2



|a| a



a |a|

q



1 f p|a |q f p|a |qv 2 v  f p|a |q f p|a|q



holds. Proof. Because we have



|a|



1 τ2 pE˜ ppn, 8qq  τ p vE ppn, 8qqv  q 2  τ pE pn, 8qq Ñ 0,

1 τ p v  vE pn, 8qq 2

a  a |a| is τ2-measurable. Suppose E is a spectral measure corresponding to |a|. We let an

 aE pr0, nsq.

6

KYUNG HOON HAN

Then we have

|an|  pE pr0, nsaaE pr0, nsq  |a|E pr0, nsq. Since an v   v |a|E pr0, nsqv  ¥ 0 and pan v  q2  aE pr0, nsqv  vE pr0, nsqa  aE pr0, nsqa , 1 2

we have

an v 

We compute  1 |a | 2 a  1 |a |  4 a



 paE pr0, nsqaq  |an|. 1 2



a r |a| E pr0, nsq



a |a|



vE pr0, nsqv  vE pr0, nsq E pr0, nsqv  v  vE pr0, nsq



 |vE pr0, nsqv  aE pr0, nsqv  |a|vE pr0, nsq av  vE pr0, nsq  14 a|a vE pr0, nsqv |a|E pr0, nsqv avE pr0, nsq |a|vvE pr0, nsq  

1 |an | an  2 a |an| . n

Since |an | is bounded, we can apply Lemma 2.2. The partial isometry in the polar decomposition of an is vn : vE pr0, nsq. It follows that 1 fp 2

Since v  |an |k v





|a|

a |a|

a





 qErpr0, nsq  f p 12 |aan| |aann| q  n

1 f p|an |q f p|an |qvn  2 vf p|a |q f p|an|q . n n

 vpanvqk v  |an|k for any k ¥ 1, we have v  f p|an |qv  f p|an |q.

Similarly, we also have

vf p|an |qv   f p|an |q. The spectral measure for |a | is vEv  δ0 pI  vv  q. It follows that





1 f p|a |q f p|a |qv r E pr0, nsq 2 v  f p|a |q f p|a|q

 1 2f p|a |qvE pr0, nsqv  2f p|a |qvE pr0, nsq  4 vf p|a|qvE pr0, nsqv f p|a|qE pr0, nsqv vf p|a|qvE pr0, nsq f p|a|qvvE pr0, nsq

 1 2f p|a |vE pr0, nsqv q 2f p|a |vE pr0, nsqv qvn  4 vf p|a|vE pr0, nsqvq vvf p|a|E pr0, nsqqv vf p|a|vE pr0, nsqvqv f p|a|E pr0, nsqq n n





   12 vffp|p|aan|q |q ffp|p|aan|qn|qvn . n n

” rpr0, nsq is τ2 -dense, we obtain the desired result from [Te1, Because the set n81 ranE Proposition I.12].  Lemma 2.5. Suppose M is a von Neumann algebra with a faithful semifinite normal weight ϕ. If a is an element of Haagerup’s noncommutative Lp -space Lp pMq, then we have



    a | a |a | a P LppM2 pMqq | and } a |a| }LppM2 pMqq  2}a}LppMq . a |a|

NONCOMMUTATIVE Lp -SPACE AND OPERATOR SYSTEM

7

Proof. Let θ be the dual action of R on M σϕ R. Then idM2 b θ is the dual action of R on M2 pMq σtrbϕ R. Because

pidM

2

   θθssp|paa q|q     e |a |

  b θ q |a | a

s

s p



we have

a From Lemma 2.4, it follows that

  } |a | a



a |a|

}



|a|

a |a|



a |a| ,

a

P LppM2 pMqq

.

  p |a | a qq  } a |a| }L pM pMqq  p p

p1  2 } v|a|a||p |a|a||pv }L pM pMqq  2p1p}|a|p}L pMq }|a|p}L pMq q  2p}a}pL pMq .

p Lp M2 M

p p



θs paq θs p|a|q

a |a|

1

2

1

1

2

1

p

 Theorem 2.6. Suppose M is a von Neumann algebra with a faithful semifinite normal weight ϕ. Its noncommutative Lp -space Lp pMq is an operator system with

}a}M pL pMqq ¤ 2 νL pMq paq, a P MnpLppMqq. Proof. Let a P Lp pMn pMqq with its right polar decomposition a  |a |v. Because }a}L pM pMqq  }|a|v}L pM pMqq ¤ }a}L pM pMqq , the involution is an isometry on Lp pMn pMqq, which can be identified with Spn pLp pMqq. From [Pi2, Lemma 1.7], the involution is also an isometry on Mn pLp pMqq. The H:older inequality and [Te1, Proposition II.33] show that the positive cone Lp pMn pMqq is 1 p

n

p

p

p

n

p

p

n

n

closed. Hence, the noncommutative Lp -space is a matrix ordered operator space. Let a P Spn pLp pMqq and 1p 1q  1. From Lemma 2.5, it follows that

}a}L pM pMqq  supt|trMpbaq| : b P Lq pMnpMqq1 u  suptRe trMpb aq : b P Lq pMnpMqq1u  supt 12 ptrMpb aq trMpba qq : b P Lq pMnpMqq1u   

1 |b |a : b P Lq pMnpMqq1u  supt 2 trM pMq |ba  b|a b a

   1 | b | b 0 a  supt 2 trM pMq p b |b| a 0 q : b P Lq pMnpMqq1u 

0 a ¤ suptϕp a 0 q : ϕ P LppM2npMqq1, u p

n

2

2

2

1 p

νLp p pMq paq.

8

KYUNG HOON HAN

Thus, from Lemma 2.1, we obtain the desired result.



By combining this with [W, Theorem 4.15], we obtain the following embedding theorem. Corollary 2.7. Noncommutative Lp -spaces can be embedded into B pH q 2 p -completely isomorphically and complete order isomorphically. 1

For a C  -algebra A, we define an involution on its dual space A as ϕ paq  ϕpa q ,

ϕ P A , a P A.

Along the line of operator space duality, we identify Mn pA q and Mn pAq algebraically by using the parallel duality pairing

xrϕij s, raij sy 

¸

ϕij paij q,

ϕij

P A , aij P A.

i,j

We can now define the positive cone Mn pA q . However, the duality between the von Neumann algebra Mn pMq and its noncommutative L1 -space L1 pMn pMqq is given by the trace duality pairing

xraij s, rbij sy  trM b trMpraij srbij sq 

¸

n

trM paij bji q.

i,j

Hence, we must be more careful. For an operator space V , its opposite operator space V op is defined by

}rvijops}M pV q  }rvjis}M pV q. op

n

n

op

Note that V and V are the same normed spaces. For a matrix ordered operator space X, we define its opposite matrix ordered operator space X op as the opposite operator space with the matrix order

rxopijs ¥ 0  rxjis ¥ 0.

A functional ϕ : Mn pX q Ñ C can be written as ϕ  rϕij s,

ϕprxij sq 

¸

ϕij pxij q.

i,j

Let ϕ˜  rϕjis. Then we have

ϕprxij sq  ϕ˜prxji sq. A functional ϕ : Mn pX q Ñ C is positive and contractive if and only if ϕ˜ : Mn pX op q Ñ C is positive and contractive. From this, we see that X is an operator system if and only if X op is an operator system. Corollary 2.8. The dual space of a C  -algebra is an operator system. Proof. The bidual A of a C  -algebra A can be identified with the enveloping von Neumann algebra of A. The opposite matrix ordered operator space pA qop is completely isometric and completely order isomorphic to L1 pA q by [Te1, Theorem II.7]. Hence, the dual space A is an operator system.  1

Finally, we show that the constant 2 p in the inequality

}a}L pM pMqq ¤ 2 p

n

1 p

νLp p pMq paq

NONCOMMUTATIVE Lp -SPACE AND OPERATOR SYSTEM 1

9

is the best one. However, this does not imply that 2 p is the best constant in Theorem 2.6. Unexpectedly, it is achieved by the 2-dimensional commutative case. We consider an element p1, 1q in ℓ2p . It is sufficient to show that 1 p

1 q

νℓp2p pp1, 1qq ¤ 1.

 1, we let

 a1  b1    a2  b2 Æ 4  ¯b1  d1  Æ P pSq q1, .  ¯b2  d2 We have   a1  b1  1  1    a2  b2 ÆÆ   1  1ÆÆq  0 ¤ trp ¯b1  d1  1  1   ¯b2  d2  1  1  a1 a2 d1 d2  b1  b2  ¯b1  ¯b2 . It follows that   a1  b1    1    a2  b2 ÆÆ     1ÆÆq  b1 b2 ¯b1 ¯b2  trp ¯b1  d1  1     ¯b2  d2  1   ¤ 12 pa1 a2 d1 d2 b1 b2 ¯b1 ¯b2 q   a1  b1  1  1    a2  b2 ÆÆ   1  1Æ  12 trp ¯b1  d1   1  1  Æ q  ¯b2  d2  1  1  1  1    1  1ÆÆ ¤ 12 }  1  1  }S  1  1  1 1   1 1   ÆÆ  12 }     1 1 }S   1 1 2 . For

4 p

4 p

1 p

In the same manner, we can also show that

  a1  b1    1   Æ  a2  b2    trp 1   ¯b1  d1  Æ  ¯  b2  d2  1 

We obtain the inequality

νℓp2p pp1, 1qq ¤ 1.

1Æ Æ  q ¥ 2 



1 p

.

10

KYUNG HOON HAN

References [CE] M.-D. Choi and E. Effros, Injectivity and operator spaces, J. Funct. Anal. 24 (1997), 156–209. [HJX] U. Haagerup M. Junge, and Q. Xu, A reduction method for noncommutative Lp -spaces and applications , Trans. Amer. Math. Soc., to appear. [JRX] M. Junge, Z-J. Ruan and Q. Xu, Rigid OLp structures of non-commutative Lp -spaces associated with hyperfinite von Neumann algebras, Math. Scand. 96 (2005), 63–95. [K] H. Kosaki, Applications of the complex interpolation method to a von Neumann algebra: Noncommutative Lp -spaces, J. Funct. Anal. 56 (1984), 29–78 [Pa] V.I. Paulsen, Completely Bounded Maps and Operator Algebras, Cambridge Studies in Advanced Mathematics, 78. Cambridge University Press, Cambridge, UK, 2002. [Pi1] G. Pisier, The operator Hilbert space OH, complex interpolation and tensor norms, Mem. Amer. Math. Soc. 122, No. 585, 1996. [Pi2] G. Pisier, Non-commutative vector valued Lp -spaces and completely p-summing maps, Ast´erisque 247, 1998. [S] I. E. Segal, A non-commutative extension of abstract integration, Ann. Math. (2) 57 (1953), 401–457. [Ta] M. Takesaki, Theory of Operator Algebras II, Encyclopaedia of Mathematical Sciences, Vol. 125, Springer-Verlag, 2002 [Te1] M. Terp, Lp spaces associated with von Neumann algebras, Math. Institute, Copenhagen University, 1981. [Te2] M. Terp, Interpolation spaces between a von Neumann algebra and its predual, J. Operator Theory 8 (1982), 327-360. [W] W. Werner, Subspaces of LpH q that are -invariant, J. Funct. Anal. 193 (2002), 207–223. Department of Mathematical Sciences, Seoul National University, San 56-1 ShinRimDong, KwanAk-Gu, Seoul 151-747, Korea E-mail address: [email protected]