Grothendieck’s tensor norms

Vandana
Shivaji College, University of Delhi, India

April, 2015

Vandana Grothendieck’s tensor norms

Overview

Grothendieck’s tensor norms

Operator spaces

Tensor products of Operator spaces

Schur tensor product

Vandana Grothendieck’s tensor norms

Grothendieck’s tensor norms

A Banach space is a complete normed space. For Banach spaces X and

Y , X ⊗ Y = span{x ⊗ y : x ∈ X, y ∈ Y }, where x ⊗ y is the functional
on B(X ∗ × Y ∗ , C) given by x ⊗ y(f, g) = f (x)g(y) for f ∈ X ∗ and
g ∈ Y ∗.
For a pair of arbitrary Banach spaces X and Y , the norm on X ⊗ Y
induced by the embedding X ⊗ Y → B(X ∗ × Y ∗ , C) is known as
Banach space injective tensor norm. That is , for u ∈ X ⊗ Y , the
Banach space injective tensor norm is defined to be
n
n X o
kukλ = sup f (xi )g(yi ) : f ∈ X1∗ , g ∈ Y1∗ .
i=1

Vandana Grothendieck’s tensor norms

Grothendieck’s tensor norms

Question is
How can we norm on X ⊗ Y ?
n
X
kx ⊗ ykα ≤ kxkkyk, then, for u = xi ⊗ yi , by triangle’s
i=1
n
X
inequality it follows that kukα ≤ kxi kkyi k. Since this holds for
i=1
n
X
every representation of u, so we have kukα ≤ inf{ kxi kkyi k}.
i=1
For a pair of arbitrary Banach spaces X and Y and u an element in
the algebraic tensor product X ⊗ Y , the Banach space projective
tensor norm is defined to be
Xn n
X
kukγ = inf{ kxi kkyi k : u = xi ⊗ yi , n ∈ N}.
i=1 i=1
X ⊗γ Y will denote the completion of X ⊗ Y with respect to this
norm.

Vandana Grothendieck’s tensor norms

Tensor Products: Properties
For Banach spaces X1 , X2 , X̃1 , X̃2 and X3 ,
projective, For the quotient maps Q : X1 → X̃1 and R : X2 → X̃2 ,
the corresponding map Q ⊗ R : X1 ⊗ X2 → X̃1 ⊗ X̃2 extends to a
quotient map Q ⊗γ R : X1 ⊗γ X2 → X̃1 ⊗γ X̃2 .
commutative, X1 ⊗γ X2 is isometrically isomorphic to X2 ⊗γ X1 ;

associative, X1 ⊗γ (X2 ⊗γ X3 ) is isometrically isomorphic to

(X1 ⊗γ X2 ) ⊗γ X3 ;

If φi : Xi → Yi , i = 1, 2, are bounded maps between the Banach

spaces, then φ1 ⊗ φ2 : X1 ⊗ X2 → Y1 ⊗ Y2 extends to a bounded
map φ1 ⊗γ φ2 : X1 ⊗γ X2 → Y1 ⊗γ Y2 with
kφ1 ⊗γ φ2 k = kφ1 kkφ2 k;

In general, Banach space projective tensor product does not respect

subspaces, that is, if Y1 is a subspace of X1 , then Y1 ⊗γ X2 is not,
in general, a subspace of X1 ⊗γ X2 . However, it does for closed
∗
-subalgebras of C ∗ -algebras. Also, for Banach spaces X and Y ,
X ⊗γ Y is a closed subspace of X ∗∗ ⊗γ Y ∗∗ .
Vandana Grothendieck’s tensor norms
Tensor Products: Properties

The key property of this tensor product is that it linearizes bounded

bilinear maps just as the algebraic tensor product linearizes bilinear
mappings.
Let X, Y and Z be Banach spaces. Then there exists a natural
isometric isomorphism B(X × Y, Z) ∼
= B(X ⊗γ Y, Z) under which
any element B ∈ B(X × Y, Z) is mapped to the element B̃ of
B(X ⊗γ Y, Z) given by B̃(x ⊗ y) = B(x, y) (x ∈ X, y ∈ Y ).
In particular, (X ⊗γ Y )∗ = B(X × Y, C).

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

For Banach algebras A and B, A ⊗γ B is a Banach algebra,

It is commutative if and only if both A and B are.
If A and B are Banach ∗ -algebras, then A ⊗γ B is a Banach
∗
-algebra under the natural involution (a ⊗ b)∗ = a∗ ⊗ b∗ for a ∈ A
and b ∈ B.

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

For Banach spaces X1 , X2 , Y1 , Y2 and X3 ,

If ui : Xi → Yi are isometries for i = 1, 2 then the corresponding
map u1 ⊗ u2 : X1 ⊗λ X2 → Y1 ⊗λ Y2 is an isometry too.
X1 ⊗λ X2 is isometrically isomorphic to X2 ⊗λ X1 ;

X1 ⊗λ (X2 ⊗λ X3 ) is isometrically isomorphic to (X1 ⊗λ X2 ) ⊗λ X3 ;

If φi : Xi → Yi , i = 1, 2, are bounded maps between the Banach

spaces, then φ1 ⊗ φ2 : X1 ⊗ X2 → Y1 ⊗ Y2 extends to a bounded
map φ1 ⊗λ φ2 : X1 ⊗λ X2 → Y1 ⊗λ Y2 with
kφ1 ⊗λ φ2 k = kφ1 kkφ2 k;

In general, Banach space injective tensor product does not respect

quotient.

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

For Banach algebras A and B, A ⊗λ B is not a Banach algebra in

general.
Blecher showed that if A and B are unital C ∗ -algebras then k · kλ is
submultiplicative on A ⊗ B if and only if A or B is commutative.

Vandana Grothendieck’s tensor norms

Introduction to Operator Spaces

An (concrete) operator space V is a closed subspace of B(H)

together with the natural norms on Mn (V ) inherited from
Mn (B(H)) = B(H n ).

A normed space V with a sequence of norms

k · kn : Mn (V ) → [0, ∞), n∈N

is said to be an (abstract) operator space if:

(i) kv ⊕ wkn+m = max{kvkn , kwkm } for
all v ∈ Mn (V ), w ∈ Mm (V ),
where v ⊕ w denotes the matrix v0 w0 ∈ Mn+m (V ).
(ii) kαvβkm ≤ kαkkvkn kβk, for all α ∈ Mm,n , β ∈ Mn,m , v ∈ Mn (V ).

Vandana Grothendieck’s tensor norms

Operator Spaces: Morphisms

An operator φ : V → W between operator spaces V and W is said

to be completely bounded (abbreviated as c.b.) if

kφkcb := sup{kφn k : n ∈ N} < ∞,

where φn : Mn (V ) → Mn (W ) is defined by

φn ((xij )) = (φ(xij )) for all (xij ) ∈ Mn (V ).

The set of all completely bounded maps from V into W is denoted

by CB(V, W ).

Two operator spaces V and W are said to be completely

isometrically isomorphic if there exists a completely bounded
isometry φ : V → W whose inverse is also completely bounded.

Vandana Grothendieck’s tensor norms

Operator Spaces

Ruan(1988)
If V is an abstract operator space, then V is completely isometrically
isomorphic to a closed linear subspace of B(H) for some Hilbert space H.

Vandana Grothendieck’s tensor norms

Introduction to Operator Space: Examples

By Gelfand-Naimark theorem, every C ∗ -algebra is an operator space.

For Hilbert spaces H and K, B(H, K) is an operator space by the

embedding B(H, K) ,→ B(H ⊕ K). The matrix norms are given by
Mn (B(H, K)) ∼ = B(H n , K n ).

Every Banach space X possess a structure of operator space. To see

this, consider Γ = X1∗ , which is compact in the w∗ -topology by
Alogalu Theorem, so that C(Γ) ⊆ B(`2 (Γ)) is a C ∗ -algebra. The
isometric embedding X ,→ C(Γ) via x → fx , where
fx (g) := g(x) for g ∈ Γ, equips X with an operator space structure.

Vandana Grothendieck’s tensor norms

Quantum tensor product-Tensor Products of operator
spaces
For operator spaces V and W and V ⊗ W their algebraic tensor product.
Assumptions:
k · kn on Mn (V ⊗ W ) satisfying Ruan’s Theorem,
k · kn on Mn (V ⊗ W ) are subcross matrix norms, where an operator
space matrix norm k · kµ on V ⊗ W is called a subcross matrix norm
if kv ⊗ wkµ ≤ kvkkwk for all v ∈ Mp (V ) and w ∈ Mq (W ),
p, q ∈ N; if in addition, kv ⊗ wkµ = kvkkwk, then k · kµ is called a
cross matrix norm on V ⊗ W .
For operator spaces V and W , the operator space projective tensor
product denoted by V ⊗W
b , is the completion of the algebraic tensor
product V ⊗ W under the norm

kuk∧ = inf{kαkkvkkwkkβk : u = α(v ⊗ w)β}, u ∈ Mn (V ⊗ W ),

where infimum runs over arbitrary decompositions with v ∈ Mp (V ),

w ∈ Mq (W ), α ∈ Mn,pq , β ∈ Mpq,n and p, q ∈ N arbitrary.

Vandana Grothendieck’s tensor norms

 Quantum tensor product-Tensor Products of operator
spaces
The analogy of ⊗ b and ⊗γ is not completely transparent from the
definition; the following universal properties of ⊗
b confirm the parallelism.

Theorem
[Effros-Ruan and Blecher-Paulsen] If V , W and Z are operator spaces,
then there are natural completely isometric identifications :
cb
b Z) ∼
CB(V ⊗W, = JCB(V × W, Z).
cb
b )∗ ∼
In particular, if Z = C, (V ⊗W = JCB(V × W, C).

where a bilinear map u : V × W → Z is said to be jointly completely

bounded (in short, j.c.b.) if the associated maps
un : Mn (V ) × Mn (W ) → Mn2 (Z) given by



un (vij ), (wkl ) = u(vij , wkl ) , n ∈ N
are uniformly bounded, and in this case we denote kukjcb = sup kun k.
n
Vandana Grothendieck’s tensor norms
Quantum tensor Products

Given operator spaces X ⊆ B(H) and Y ⊆ B(K), the norm

induced on X ⊗ Y via the inclusion X ⊗ Y ⊆ B(H ⊗2 K) is known
as the operator space injective tensor norm.

Vandana Grothendieck’s tensor norms

Tensor Products

For u ∈ Mn (X ⊗ Y ), the Haagerup tensor norm is defined as

kukh = inf{kvkkwk : u = v w, v ∈ Mn,p (X), w ∈ Mp,n (Y ), p ∈ N},

Pp

where v w = k=1 vik ⊗ wkj ij .

For C ∗ -algebras A and B, the Haagerup norm of an element

u ∈ A ⊗ B takes a simpler and convenient form given by

kukh = inf kΣi ai a∗i k1/2 kΣi b∗i bi k1/2 : u = Σni=1 ai ⊗ bi , n ∈ N .



Vandana Grothendieck’s tensor norms

Tensor Products

Like the projective tensor product of Banach spaces and the operator
space projective tensor product of operator spaces, the Haagerup tensor
norm is naturally associated with the completely bounded bilinear maps
through the following identifications:
cb cb
CB(V × W, Z) ∼
= CB(V ⊗h W, Z) and CB(V × W, C) ∼
= (V ⊗h W )∗ .

By a completely bounded (in short, c.b.) bilinear map, we mean a

bilinear map u : V × W → Z for which the associated maps
un : Mn (V ) × Mn (W ) → Mn (Z) given by

X

un (vij ), (wkl ) = u(vik , wkj ) , n ∈ N
k

are uniformly bounded, and in this case we denote kukcb = sup kun k.
n

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

For operator spaces V1 , V2 and V3 , the product ⊗

b is
cb
commutative, that is, V1 ⊗V
b 2 = V2 ⊗V
b 1;
cb
associative, that is, V1 ⊗(V
b 2 ⊗V
b 3) = (V1 ⊗V
b 2 )⊗V
b 3;

functorial, that is, if φi : Vi → Wi , i = 1, 2, are completely bounded

maps between the operator spaces, then
φ1 ⊗ φ2 : V1 ⊗ V2 → W1 ⊗ W2 extends to a completely bounded
map φ1 ⊗φ
b 2 : V1 ⊗V
b 2 → W1 ⊗W b 2 with kφ1 ⊗φb 2 kcb ≤ kφ1 kcb kφ2 kcb ;

projective, that is, for closed subspaces Wi ⊆ Vi , i = 1, 2, the tensor

map V1 ⊗ V2 → (V1 /W1 ) ⊗ (V2 /W2 ) extends to a complete quotient
map V1 ⊗V
b 2 → (V1 /W1 )⊗(V b 2 /W2 );

In general, operator space projective tensor product is not injective,

that is, for subspaces Wi ⊆ Vi , i = 1, 2, the induced map
W1 ⊗Wb 2 → V1 ⊗V b 2 is not a complete isometry. However, for
C ∗ -algebras, it is injective for finite dimensional C ∗ -subalgebras.

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

Haagerup tensor product is associative, projective, injective and

functorial. However, it is not commutative.

Vandana Grothendieck’s tensor norms

Algebraic structure of A⊗B
b

(Kumar, Itoh) For C ∗ -algebras A and B, A⊗B b is a Banach

∗-algebra, and is a C ∗ -algebra if and only if either A or B is C.

(Kumar) However, the natural involution is an isometry on A ⊗h B

if and only if A and B are commutative.

Vandana Grothendieck’s tensor norms

Schur tensor product of operator spaces

For operator spaces V and W , and elements x = [xij ] ∈ Mn (V ) and

y = [yij ] ∈ Mn (W ), we define an element x ◦ y ∈ Mn (V ⊗ W ) by
x ◦ y = [xij ⊗ yij ].
x ◦ y = [e11 , e22 , e33 , · · · , enn ](x ⊗ y)[e11 , e22 , e33 , · · · , enn ]t
Each element u in Mp (V ⊗ W ), p ∈ N, can be written as
u = α(x ◦ y)β for some x ∈ Mn (V ), y ∈ Mn (W ), α ∈ Mp,n , and
β ∈ Mn,p , n ∈ N, and we define

kuks = inf{kαkkxkkykkβk}

where infimum is taken over arbitrary decompositions as above. Let

V ⊗s W = (V ⊗ W, k · ks ), and define the Schur tensor product
V ⊗s W to be the completion of V ⊗ W in this norm.
Note: If u ∈ Mp (V ⊗ W ) there exist n ∈ N, v ∈ Mn (V ), w ∈ Mn (W ),
α ∈ Mp,n and β ∈ Mn,p such that u = α(v ◦ w)β.

Vandana Grothendieck’s tensor norms

Schur tensor product of operator spaces

Theorem
For operator spaces V and W , k · ks is an operator space matrix norm on
V ⊗ W.

Given operator spaces V , W and Z, a bilinear map ϕ : V × W → Z is

said to be Schur bounded bilinear map if the associated maps
ϕn : Mn (V ) × Mn (W ) → Mn (Z) given by



ϕn (vij ), (wij ) = ϕ(vij , wij ) , n ∈ N

are uniformly bounded, and in this case we denote kϕksb = sup kϕn k.
n

Vandana Grothendieck’s tensor norms

Schur tensor product of operator spaces

Proposition
If V , W and X are operator spaces, then there is a natural isometric
identification
CB(V ⊗W,
b X) = SB(V × W, X).

The above identification yields a new formula for the Schur norm :

kuks = sup{|ϕ(u)| : ϕ ∈ SB(V × W, C), kϕk ≤ 1}.

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

For operator spaces V1 , V2 and V3 , the product ⊗

b is
cb
commutative, that is, V1 ⊗s V2 = V2 ⊗s V1 ;

functorial, that is, if φi : Vi → Wi , i = 1, 2, are completely bounded

maps between the operator spaces, then
φ1 ⊗ φ2 : V1 ⊗ V2 → W1 ⊗ W2 extends to a completely bounded
map φ1 ⊗s φ2 : V1 ⊗s V2 → W1 ⊗s W2 with
kφ1 ⊗s φ2 kcb ≤ kφ1 kcb kφ2 kcb ;

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

projective, that is, for closed subspaces Wi ⊆ Vi , i = 1, 2, the tensor

map V1 ⊗ V2 → (V1 /W1 ) ⊗ (V2 /W2 ) extends to a complete quotient
map V1 ⊗s V2 → (V1 /W1 ) ⊗s (V2 /W2 );
Like the operator space projective tensor product, the Schur tensor
product is not injective, that is, for subspaces Wi ⊆ Vi , i = 1, 2, the
induced map W1 ⊗s W2 → V1 ⊗s V2 is not a complete isometry.
However, it behaves well for completely complemented subspaces. In
particular, if E and F are finite dimensional C ∗ -subalgebras of the
C ∗ -algebras A and B, respectively. Then E ⊗s F is a closed
∗
-subalgebra of A ⊗s B. Also, for von Neumann algebras M and N ,
Z(M ) ⊗s Z(N ) is a closed ∗ -subalgebra of M ⊗s N .

Vandana Grothendieck’s tensor norms

Tensor Products: Properties

Associative, For finite dimensional operator spaces X, Y and any

operator space Z, X ⊗s (Y ⊗s Z) is bi-continuously isomorphic to
(X ⊗s Y ) ⊗s Z.
For commutative C ∗ -algebras A and C with identity and a finite
dimensional C ∗ -algebra B, A ⊗s (B ⊗s C) is bi-continuously
isomorphic to (A ⊗s B) ⊗s C.

Vandana Grothendieck’s tensor norms

Schur tensor product of completely contractive Banach
algebras

Theorem
For completely contractive Banach algebras A and B, A ⊗s B is a Banach
algebra, and it is ∗ -algebra provided both A and B have completely
isometric involution. Furthermore, if A and B possess bounded
approximate identities then A ⊗s B has a bounded approximate identity.

A ⊗s B is a C ∗ -algebra if and only if either A = C or B = C.

A ⊗s B is an operator algebra if and only if either A or B is finite
dimensional.

Vandana Grothendieck’s tensor norms

Related Books

D. P. Blecher and C. L. Merdy, Operator algebras and their

modules-an operator space approach, vol. 30, London Mathematical
Society Monographs, New series, The Clarendon Press, Oxford
University Press, Oxford, 2004.
E. G. Effros and Z-J. Ruan, Operator Spaces, Clarendon Press,
London, 2000.
G. Pisier, Introduction to Operator space theory, London
Mathematical Society, Lecture Note Series 294, Cambridge
University Press, Cambridge, 2003.
Paulsen, Completely bounded maps and operator algebras,
Cambridge Studies in Advanced Mathematics, 78. Cambridge
University Press, Cambridge, 2002.
Rajpal, V, Kumar, A., and Itoh, T., Schur tensor product of operator
spaces, To appear in Forum Mathematicum (2014), DOI:
http://dx.doi.org/10.1515/forum-2013-0142, Available on
arXiv:1308.4538v1 [math.OA].

Vandana Grothendieck’s tensor norms

THANK YOU

Vandana Grothendieck’s tensor norms