Next Article in Journal
The Effects of Fiscal and Tax Incentives on Regional Innovation Capability: Text Extraction Based on Python
Previous Article in Journal
How Does Advanced Technology Solve Unreliability Under Supply Chain Management Using Game Policy?
Article

The Ideal of σ-Nuclear Operators and Its Associated Tensor Norm

by *,† and
Department of Mathematics and Statistics, Sejong University, Seoul 05006, Korea
*
Author to whom correspondence should be addressed.
These authors contributed equally to this work.
Mathematics 2020, 8(7), 1192; https://doi.org/10.3390/math8071192
Received: 26 June 2020 / Revised: 17 July 2020 / Accepted: 19 July 2020 / Published: 20 July 2020
(This article belongs to the Special Issue Dynamics of Operators and C0-Semigroups)

Abstract

We introduce a new tensor norm ( σ -tensor norm) and show that it is associated with the ideal of σ -nuclear operators. In this paper, we investigate the ideal of σ -nuclear operators and the σ -tensor norm.
Keywords: tensor norm; Banach operator ideal; nuclear operator tensor norm; Banach operator ideal; nuclear operator

1. Introduction

Let X Y be the algebraic tensor product of Banach spaces X and Y. One may refer to [1] (Section 1) for tensor products and their elementary properties. If α is a norm on the tensor product, then the normed space ( X Y , α ) is denoted by X α Y and X ^ α Y is the completion of X α Y . The most classical two norms ε and π on X Y are the injective norm and projective norm, respectively. For u X Y ,
ε ( u ; X , Y ) : = sup | n = 1 l x ( x n ) y ( y n ) | : x B X , y B Y ,
where n = 1 l x n y n is any representation of u and B Z is the closed unit ball of a Banach space Z, and
π ( u ; X , Y ) : = inf n = 1 l x n y n : u = n = 1 l x n y n , l N .
We refer to [1,2] for ε and π . Our main notion is the following concept.
Definition 1.
For n = 1 l x n y n X Y , let
| n = 1 l x n y n | σ : = sup n = 1 l | x ( x n ) y ( y n ) | : x B X , y B Y .
For u X Y , let
α σ ( u ; X , Y ) : = inf | n = 1 l x n y n | σ : u = n = 1 l x n y n , l N .
We call α σ the σ-tensor norm.
A Banach operator ideal [ A , · A ] is said to be associated with a tensor norm α if the natural map from A ( M , N ) to M α N is an isometry for both finite-dimensional normed spaces M and N. Let · be the operator norm on the ideal L of all operators and let F be the ideal of all finite rank operators. A linear map T : X Y is called approximable if there exists a sequence ( T n ) n in F ( X , Y ) such that lim n T n T = 0 . We denote by F ¯ ( X , Y ) the space of all approximable operators from X to Y. Then the ideal [ F ¯ , · ] of approximable operators is a Banach operator ideal.
A linear map T : X Y is nuclear if there exists sequences ( x n ) n in X and ( y n ) n in Y with
n = 1 x n y n <
such that
T = n = 1 x n ̲ y n ,
where x n ̲ y n is an operator from X to Y defined by ( x n ̲ y n ) ( x ) = x n ( x ) y n . The space of all nuclear operators from X to Y is denoted by N ( X , Y ) with the norm
T N : = inf n = 1 x n y n : T = n = 1 x n ̲ y n ,
where the infimum is taken over all such representations. It is well known that [ F ¯ , · ] is associated with ε and [ N , · N ] is associated with π (cf. [2] (Section 17.12)).
Pietsch [3] introduced a natural extended notion of the nuclear operator. A linear map T : X Y is called σ -nuclear if there exists sequences ( x n ) n in X and ( y n ) n in Y such that
T = n = 1 x n ̲ y n
unconditionally converges in the operator norm. We denote by N σ ( X , Y ) the space of all σ -nuclear operators from X to Y and for T N σ ( X , Y ) , let
T N σ : = inf | n = 1 x n ̲ y n | σ : T = n = 1 x n ̲ y n ,
where | n = 1 x n ̲ y n | σ : = sup { n = 1 | x n ( x ) y ( y n ) | : x B X , y B Y } and the infimum is taken over all σ -nuclear representations. Then [ N σ , · N σ ] is a Banach operator ideal [3] (Theorem 23.2.2).
In this paper, we study the Banach operator ideal N σ of σ -nuclear operators and the corresponding σ -tensor norm α σ . In Section 2, we obtain a factorization of operators belonging to N σ and show that the surjective hull and the injective hull of N σ coincide with the ideal of compact operators. It turns out that [ N σ , · N σ ] is associated with α σ . In Section 3, we show that α σ is a finitely generated tensor norm and the completion X ^ α σ Y is identified. An isometric representation of the dual space ( X α σ Y ) is established. In Section 4, we show that
X ε Y = X α σ Y
holds isometrically when X or Y has a hyperorthogonal basis. As a consequence, we show that α σ is neither injective nor projective.

2. The Ideal of σ -Nuclear Operators

For Banach spaces X and Y, we denote by
σ ( X , Y )
the collection of sequences ( x n , y n ) n in X × Y satisfying
lim l sup n l | x n ( x ) y ( y n ) | : x B X , y B Y = 0
and let
| ( x n , y n ) n | σ = sup n = 1 | x n ( x ) y ( y n ) | : x B X , y B Y
for ( x n , y n ) n σ ( X , Y ) .
A basis ( e n ) n for a Banach space X is called hyperorthogonal if for every n N | α n | | β n | implies
n = 1 α n e n n = 1 β n e n .
Using a standard argument, we have the following lemma.
Lemma 1.
Let K be a collection of sequences of positive numbers.
If sup ( k n ) n K n = 1 k n < and lim l sup ( k n ) n K n l k n = 0 , then for every ε > 0 , there exists an increasing sequence ( β n ) n with β n > 1 and lim n β n = such that
lim l sup ( k n ) n K n l k n β n = 0 and sup ( k n ) n K n = 1 k n β n ( 1 + ε ) sup ( k n ) n K n = 1 k n .
It is well known that a nuclear operator T : X Y has the following factorization.
X T Y R S C 0 D 1 ,
where R and S are compact operators, and D is a diagonal operator which is nuclear. From a modification of [3] (Theorem 23.2.5), we have a similar form for σ -nuclear operators.
Theorem 1.
Let X and Y be Banach spaces and let T : X Y be a linear map. Then the following statements are equivalent.
(a) 
T N σ ( X , Y ) .
(b) 
There exists ( x n , y n ) n σ ( X , Y ) such that
T = n = 1 x n ̲ y n .
(c) 
There exist Banach spaces Z and W having hyperorthogonal bases, R N σ ( X , Z ) , a diagonal operator D N σ ( Z , W ) with D N σ 1 , and S N σ ( W , Y ) with S N σ 1 such that the following diagram is commutative.
X T Y R S Z D W .
In this case,
T N σ = inf | ( x n , y n ) n | σ = inf R N σ ,
where the first infimum is taken over all such representations of T in ( b ) and the second infimum is taken over all such factorizations of T in ( c ) .
Proof. 
(c)⇒(a) is trivial and T N σ is less than or equal to the infimum for factorizations of T in (c).
(a)⇒(b): This part is a well known result. For the sake of the completeness of presentation, we provide an explicit proof. Let T N σ ( X , Y ) and let ε > 0 be given. Then there exists a representation
T = n = 1 x n ̲ y n ,
which unconditionally converges in ( L ( X , Y ) , · ) , such that
| n = 1 x n ̲ y n | σ ( 1 + ε ) T N σ .
It is well known that a series n = 1 z n in a Banach space Z unconditionally converges if and only if
lim l sup z B Z n l | z ( z n ) | = 0 .
Thus,
lim l sup n l | x n ( x ) y ( y n ) | : x B X , y B Y lim l sup n l | φ ( x n ̲ y n ) | : φ B ( L ( X , Y ) , · ) = 0 .
Hence (b) follows and the first infimum
inf | · | σ | ( x n , y n ) n | σ = | n = 1 x n ̲ y n | σ ( 1 + ε ) T N σ .
Since ε > 0 was arbitrary, inf | · | σ T N σ .
(b)⇒(c): Let ε > 0 be given. By (b), there exists ( x n , y n ) n σ ( X , Y ) such that
T = n = 1 x n ̲ y n
and
| ( x n , y n ) n | σ ( 1 + ε ) inf | · | σ .
By Lemma 1, there exists a sequence ( β n ) n with β n > 1 and lim n β n = such that
( β n 2 x n , y n ) n σ ( X , Y )
and
| ( β n 2 x n , y n ) n | σ ( 1 + ε ) | ( x n , y n ) n | σ .
Let
Z : = ( α n ) n in C : n = 1 α n β n 2 y n u n c o n d i t i o n a l l y c o n v e r g e s i n Y
and
|| ( α n ) n Z : = sup y B Y n = 1 β n 2 | α n y ( y n ) | .
Then ( Z , · Z ) is a Banach space and the sequence ( e n ) n of standard unit vectors forms a hyperorthogonal basis in Z. Let
W : = ( γ n ) n in C N : n = 1 γ n y n u n c o n d i t i o n a l l y c o n v e r g e s i n Y
and
|| ( γ n ) n W : = sup y B Y n = 1 | γ n y ( y n ) | .
Then ( W , · W ) is a Banach space and the sequence ( f n ) n of standard unit vectors forms a hyperorthogonal basis in W.
Let
R : X Z , R x = ( x n ( x ) ) n ,
D : Z W , D ( α n ) n = ( β n α n ) n ,
S : W Y , S ( γ n ) n = n = 1 γ n β n y n .
To show that R = n = 1 x n ̲ e n unconditionally converges in L ( X , Z ) , let δ > 0 . Choose an l δ N such that
sup n l δ β n 2 | x n ( x ) y ( y n ) | : x B X , y B Y δ .
Then for every finite subset F of N with min F > l δ ,
n F x n ̲ e n = sup x B X n F x n ( x ) e n Z = sup n F β n 2 | x n ( x ) y ( y n ) | : x B X , y B Y δ .
Hence R N σ ( X , Z ) . Since for every x B X and z B Z ,
n = 1 | x n ( x ) z ( e n ) | = n = 1 λ n x n ( x ) z ( e n ) ( | λ n | = 1 ) n = 1 λ n x n ( x ) e n Z = sup y B Y n = 1 β n 2 | x n ( x ) y ( y n ) | | ( β n 2 x n , y n ) n | σ ,
R N σ | ( β n 2 x n , y n ) n | σ ( 1 + ε ) | ( x n , y n ) n | σ ( 1 + ε ) 2 inf | · | σ .
To show that D = n = 1 β n e n ̲ f n unconditionally converges in L ( Z , W ) , where each e n Z is the n-th coordinate functional, let δ > 0 . Choose an N δ N such that 1 / β n δ for every n N δ . Then for every finite subset F of N with min F > N δ ,
n F β n e n ̲ f n = sup ( α n ) n B Z n F β n α n f n W = sup n F β n | α n y ( y n ) | : ( α n ) n B Z , y B Y δ sup n F β n 2 | α n y ( y n ) | : ( α n ) n B Z , y B Y δ sup ( α n ) n B Z ( α n ) n Z δ .
Hence D N σ ( Z , W ) and D N σ 1 , indeed, for every ( α k ) k B Z and w B W ,
n = 1 | β n e n ( ( α k ) k ) w ( f n ) | = n = 1 | β n α n w ( f n ) | = n = 1 δ n β n α n w ( f n ) ( | δ n | = 1 ) n = 1 δ n β n α n f n W = sup y B Y n = 1 β n | α n y ( y n ) | sup y B Y n = 1 β n 2 | α n y ( y n ) | = ( α k ) k Z 1 .
Let f n W be the n-th coordinate functional. In order to show that S = n = 1 ( 1 / β n ) f n ̲ y n converges unconditionally in L ( W , Y ) , we take δ > 0 . Choose an N δ N such that 1 / β n δ for every n N δ . Then for every finite subset F of N with min F > N δ ,
n F ( 1 / β n ) f n ̲ y n = sup ( γ n ) n B W n F γ n β n y n Y = sup | n F γ n β n y ( y n ) | : ( γ n ) n B W , y B Y δ sup | n F γ n y ( y n ) | : ( γ n ) n B W , y B Y δ sup ( γ n ) n B W ( γ n ) n W δ .
Hence S N σ ( W , Y ) and S N σ 1 , indeed, for every ( γ k ) k B W and y B Y ,
n = 1 | ( 1 / β n ) f n ( ( γ k ) k ) y ( y n ) | n = 1 | γ n y ( y n ) | ( γ n ) n W 1 .
Clearly, T = S D R and the second infimum inf · N σ R N σ ( 1 + ε ) 2 inf | · | σ . Since ε > 0 was arbitrary, inf · N σ inf | · | σ . ☐
The surjective hull [ A , · A ] s u r of an operator ideal [ A , · A ] is defined as follows;
A s u r ( X , Y ) : = { T L ( X , Y ) : T q X A ( 1 ( B X ) , Y ) } ,
where q X : 1 ( B X ) X is the natural quotient operator, and T A s u r : = T q X A for T A s u r ( X , Y ) (see [2] (p. 113) and [3] (Section 8.5)).
Lemma 2.
(see Proposition 8.5.4 in [3]) Let [ A , · A ] be a Banach operator ideal and let X and Y be Banach spaces. A linear map T A s u r ( X , Y ) if and only if there exists a Banach space Z and an S A ( Z , Y ) such that T ( B X ) S ( B Z ) . In this case,
T A s u r = inf S A ,
where the infimum is taken over all the above inclusions.
Lemma 3.
[4] A subset K of a Banach space X is relatively compact if and only if for every ε > 0 , there exists a null sequence ( x n ) n in X with sup n N x n ( 1 + ε ) sup x K x such that
K n = 1 α n x n : ( α n ) n B 1 .
The surjective hull of the ideal of nuclear operators is identified in [3] (Proposition 8.5.5).
Theorem 2.
The surjective hull [ N σ , · N σ ] s u r of the ideal of σ-nuclear operators can be identified with the ideal [ K , · ] of compact operators.
Proof. 
Since [ N σ , · N σ ] [ F ¯ , · ] and [ F ¯ , · ] s u r = [ K , · ] ,
[ N σ , · N σ ] s u r [ K , · ] .
To show the opposite inclusion, let X and Y be Banach spaces. Let T K ( Y , X ) and let ε > 0 . Then by Lemma 3, there exists a null sequence ( x n ) n in X with sup n N x n ( 1 + ε ) T such that
T ( B Y ) n = 1 α n x n : ( α n ) n B 1 .
Let us consider the map
E : 1 X , E = n = 1 e n ̲ x n ,
where each e n is the standard unit vector in c 0 . Since
lim l sup n l | α n x ( x n ) | : ( α n ) n B 1 , x B X lim l sup ( α n ) n B 1 n l | α n | x n = 0 ,
in view of Theorem 1, E N σ ( 1 , X ) and
E N σ | ( e n , x n ) n | σ = sup n = 1 | α n x ( x n ) | : ( α n ) n B 1 , x B X sup n N x n ( 1 + ε ) T .
Since T ( B Y ) E ( B 1 ) , by Lemma 2, T N σ s u r ( Y , X ) and
T N σ s u r E N σ ( 1 + ε ) T .
 ☐
The injective hull [ A , · A ] i n j of an operator ideal [ A , · A ] is defined as follows;
A i n j ( X , Y ) : = { T L ( X , Y ) : I Y T A ( X , ( B Y ) ) } ,
where I Y : Y ( B Y ) is the natural isometry, and T A i n j : = I Y T A for T A i n j ( X , Y ) (see [2] (p. 112) and [3] (Section 8.4)).
Lemma 4.
If [ A , · A ] is a symmetric Banach operator ideal, then
[ A , · A ] i n j = ( [ A , · A ] s u r ) d u a l .
Proof. 
The symmetric operator ideal means that [ A , · A ] [ A , · A ] d u a l . Then by [3] (Theorem 8.5.9),
[ A , · A ] i n j ( [ A , · A ] d u a l ) i n j ( [ A , · A ] s u r ) d u a l .
Additionally, since [ A , · A ] s u r ( [ A , · A ] d u a l ) s u r , by [3] (Theorem 8.5.9),
( [ A , · A ] s u r ) d u a l ( ( [ A , · A ] d u a l ) s u r ) d u a l = ( ( [ A , · A ] i n j ) d u a l ) d u a l .
Note that ( ( [ B , · B ] i n j ) d u a l ) d u a l [ B , · B ] i n j for every Banach operator ideal B . Hence the assertion follows. ☐
The injective hull of the ideal of nuclear operators is identified in [3] (Proposition 8.4.5). The following theorem is a consequence of the fact that the ideal [ N σ , · N σ ] is symmetric (cf. [3] (Theorem 23.2.7)).
Theorem 3.
For the ideal of σ-nuclear operators, the following equality is valid:
[ N σ , · N σ ] i n j = [ K , · ] .
Proof. 
Since [ N σ , · N σ ] is symmetric, by Theorem 2 and Lemma 4,
[ N σ , · N σ ] i n j = ( [ N σ , · N σ ] s u r ) d u a l = [ K , · ] d u a l = [ K , · ] .
 ☐
For T F ( X , Y ) , let
T N σ 0 : = inf | n = 1 l x n ̲ y n | σ : T = n = 1 l x n ̲ y n , l N .
Then · N σ 0 is a norm on F [3] (Proposition 23.2.10).
Proposition 1.
Suppose that X or Y is a finite-dimensional normed space. Then
T N σ 0 = T N σ
for every T L ( X , Y ) .
Proof. 
Let T L ( X , Y ) and let δ > 0 be given. Let
T = n = 1 x n ̲ y n ,
be a σ -nuclear representation in Theorem 1(b) such that
| ( x n , y n ) n | σ ( 1 + δ ) T N σ .
If X is finite-dimensional, then there exists an l N such that
sup n l + 1 | x n ( x ) y ( y n ) | : x B X , y B Y δ T N σ / i d X N σ 0 ,
where i d X is the identity operator on X. We have
T N σ 0 n = 1 l x n ̲ y n N σ 0 + n l + 1 x n ̲ y n N σ 0 | ( x n , y n ) n | σ + n l + 1 x n ̲ y n N σ i d X N σ 0 ( 1 + 2 δ ) T N σ .
If Y is finite-dimensional, then i d X can be replaced by i d Y in the above proof. ☐
Corollary 1.
[ N σ , · N σ ] is associated with α σ .
Proof. 
Let X and Y be Banach spaces. Let n = 1 l x n y n X Y . The by an application of Helly’s lemma,
| n = 1 l x n y n | σ = sup n = 1 l | x n ( x ) y ( y n ) | : x B X , y B Y = | n = 1 l x n ̲ y n | σ .
Consequently, for every u X Y , we have
α σ ( u ; X , Y ) = inf | n = 1 l x n ̲ y n | σ : u = n = 1 l x n y n , l N .
Hence the assertion follows from Proposition 1. ☐

3. The σ -Tensor Norm

Let us recall that a tensor norm α is a norm on X Y for each pair of Banach spaces X and Y such that
(TN1)
ε α π .
(TN2)
for operators T 1 : X 1 Y 1 and T 2 : X 2 Y 2 ,
T 1 T 2 : X 1 α X 2 Y 1 α Y 2 T 1 T 2 .
A tensor norm α is said to be finitely generated if
α ( u ; X , Y ) = inf { α ( u ; M , N ) : u M N , dim M , dim N < }
for every u X Y . The transposed tensor norm α t of α is defined by
α t ( u ; X , Y ) : = α ( u t ; Y , X )
for u X Y .
Proposition 2.
α σ is a finitely generated tensor norm and α t = α .
Proof. 
We see that α σ is a norm and satisfies (TN1) on X Y for each pair of Banach spaces X and Y.
To check (TN2), let T 1 : X 1 Y 1 and T 2 : X 2 Y 2 be operators. Let u X 1 X 2 and let u = n = 1 l x n 1 x n 2 be an arbitrary representation. Then
α σ ( ( T 1 T 2 ) ( u ) ; Y 1 , Y 2 ) = α σ n = 1 l T 1 x n 1 T 2 x n 2 ; Y 1 , Y 2 | n = 1 l T 1 x n 1 T 2 x n 2 | σ = T 1 T 2 | n = 1 l ( T 1 / T 1 ) ( x n 1 ) ( T 2 / T 2 ) ( x n 2 ) | σ T 1 T 2 | n = 1 l x n 1 x n 2 | σ .
Hence
α σ ( ( T 1 T 2 ) ( u ) ; Y 1 , Y 2 ) T 1 T 2 α σ ( u ; X 1 , X 2 ) .
To show that α σ is finitely generated, let u X Y and let u = n = 1 l x n y n be an arbitrary representation. Let M 0 = span { x n } n = 1 l and N 0 = span { y n } n = 1 l . Using the Hahn–Banach extension theorem, we have
inf { α σ ( u ; M , N ) : u M N , dim M , dim N < } α σ ( u ; M 0 , N 0 ) sup n = 1 l | m ( x n ) n ( y n ) | : m B M 0 , n B N 0 = sup n = 1 l | x ( x n ) y ( y n ) | : x B X , y B Y .
Hence
inf { α σ ( u ; M , N ) : u M N , dim M , dim N < } α σ ( u ; X , Y ) .
The other part of the assertion follows from the definition of the σ -tensor norm. ☐
We now consider the completion X ^ α σ Y of X α σ Y . When n = 1 x n y n converges in X ^ α σ Y and sup { n = 1 | x ( x n ) y ( y n ) | : x B X , y B Y } < , we let
| n = 1 x n y n | σ : = sup n = 1 | x ( x n ) y ( y n ) | : x B X , y B Y .
Lemma 5.
Let X and Y be Banach spaces and let ( x n ) n and ( y n ) n be sequences in X and Y, respectively. Then
lim l sup n l | x ( x n ) y ( y n ) | : x B X , y B Y = 0
if and only if the series n = 1 x n y n unconditionally converges in X ^ α σ Y .
Proof. 
Suppose that lim l sup { n l | x ( x n ) y ( y n ) | : x B X , y B Y } = 0 . Let δ > 0 be given. Choose an l δ N such that
sup n l δ | x ( x n ) y ( y n ) | : x B X , y B Y δ .
Then for every finite subset F of N with min F > l δ ,
α σ n F x n y n ; X , Y | n F x n y n | σ sup n l δ | x ( x n ) y ( y n ) | : x B X , y B Y δ .
Suppose that n = 1 x n y n unconditionally converges in X ^ α σ Y . Then
lim l sup n l | x ( x n ) y ( y n ) | : x B X , y B Y lim l sup n l | φ ( x n y n ) | : φ B ( X ^ α σ Y ) = 0 .
 ☐
The following lemma is well known.
Lemma 6.
Let ( Z , · ) be a normed space and let ( Z ^ , · ) be its completion. If z Z ^ , then for every δ > 0 , there exists a sequence ( z n ) n in Z such that
n = 1 z n ( 1 + δ ) z
and z = n = 1 z n converges in Z ^ .
Proposition 3.
Let X and Y be Banach spaces. If u X ^ α σ Y , then there exists sequences ( x n ) n in X and ( y n ) n in Y such that
u = n = 1 x n y n
unconditionally converges in X ^ α σ Y and
α σ ( u ; X , Y ) = inf | n = 1 x n y n | σ : u = n = 1 x n y n u n c o n d i t i o n a l l y c o n v e r g e s i n X ^ α σ Y .
Proof. 
We use Lemma 5. Let u X ^ α σ Y and let δ > 0 be given. Then by Lemma 6, there exists a sequence ( u n ) n in X Y such that
n = 1 α σ ( u n ; X , Y ) ( 1 + δ ) α σ ( u ; X , Y )
and u = n = 1 u n converges in X ^ α σ Y .
For every n N , let
u n = k = 1 m n x k n y k n
be such that
| k = 1 m n x k n y k n | σ ( 1 + δ ) α σ ( u n ; X , Y ) .
Then for every γ > 0 , there exists an N γ N such that
sup n N γ k = 1 m n | x ( x k n ) y ( y k n ) | ; x B X , y B Y n N γ | k = 1 m n x k n y k n | σ γ .
This shows that
u = n = 1 k = 1 m n x k n y k n
unconditionally converges in X ^ α σ Y . In addition, the infimum
inf { · } | n = 1 k = 1 m n x k n y k n | σ n = 1 | k = 1 m n x k n y k n | σ ( 1 + δ ) 2 α σ ( u ; X , Y ) .
Since δ > 0 was arbitrary, inf { · } α σ ( u ; X , Y ) .
Since for every representation
u = n = 1 x n y n
unconditionally converging in X ^ α σ Y ,
α σ ( u ; X , Y ) = lim l α σ n = 1 l x n y n lim l | n = 1 l x n y n | σ = | n = 1 x n y n | σ ,
α σ ( u ; X , Y ) inf { · } . ☐
Let α be a tensor norm and let M and N be finite-dimensional normed spaces. let
α 0 ( u ; M , N ) : = sup { | v , u | : α ( v ; M , N ) 1 }
for u M N . Then the dual tensor norm is defined by
α ( u ; X , Y ) : = inf { α 0 ( u ; M , N ) : u M N , dim M , dim N < }
for u X Y . The adjoint tensor norm is
α : = ( α ) t = ( α t ) .
If α is finitely generated, then α , α t and α are all finitely generated and ( α ) = α
The adjoint ideal [ A a d j , · A a d j ] is the maximal Banach operator ideal associated with the adjoint tensor norm α .
Lemma 7.
(see Theorem 17.5 in [2]) Let A be the maximal Banach operator ideal associated with a finitely generated tensor norm α. Then for all Banach spaces X and Y,
( X α Y ) = A ( X , Y )
holds isometrically.
Pietsch [3] introduced a stronger notion of the absolutely p-summing operator. A linear map T : X Y is called absolutely τ -summing if there exists a C > 0 such that
n = 1 l | y n ( T x n ) | C sup n = 1 l | x ( x n ) y n ( y ) | : x B X , y B Y
for every x 1 , , x l X and y 1 , , y l Y . We denote by P τ ( X , Y ) the space of all absolutely τ -summing operators from X to Y and for T P τ ( X , Y ) , let
T P τ : = inf C ,
where the infimum is taken over all such inequalities. Then it was shown in [3] (Theorems 23.1.2 and 23.1.3) that [ P τ , · P τ ] is a maximal Banach operator ideal.
Pietsch [3] also introduced the ideal of σ -integral operators as follows;
[ I σ , · I σ ] : = [ N σ , · N σ ] m a x .
It was shown that
I σ a d j = P τ and P τ a d j = I σ
hold isometrically [3] (Theorem 23.3.6).
We now have:
Corollary 2.
For all Banach spaces X and Y,
( X α σ Y ) = P τ ( X , Y )
holds isometrically.
Proof. 
Since
α σ = ( α σ t ) = α σ
is associated with I σ a d j = P τ , by Lemma 7,
( X α σ Y ) = ( X ( α σ ) Y ) = P τ ( X , Y )
holds isometrically. ☐

4. Non-Injectiveness and Non-Projectiveness of the σ -Tensor Norm

Proposition 4.
Let X and Y be Banach spaces. If X or Y has a hyperorthogonal basis, then
X ε Y = X α σ Y
holds isometrically.
Proof. 
Suppose that Y has a hyperorthogonal basis ( e i ) i . Let ( e i ) i be the sequence of coordinate functionals for ( e i ) i . Let u = n = 1 l x n y n X Y and let U be the corresponding w e a k to w e a k continuous finite rank operator for u, namely, U = n = 1 l x n ̲ y n : X Y . Then for every x X ,
U x = i = 1 ( e i U x ) e i
and e i U X X for every i N . Moreover, since U ( B X ) is relatively compact,
lim l ε i = 1 l e i U e i u ; X , Y = lim l i = 1 l e i U ̲ e i U = lim l sup x B X i = 1 l ( e i U x ) e i U x = 0 .
Consequently,
u = i = 1 e i U e i
converges in X ^ ε Y . We will use Lemma 5 to show that the series unconditionally converges in X ^ α σ Y .
Let η > 0 be given. Let { U x k } k = 1 m be an η / 2 -net for U ( B X ) . Choose an l N so that
i l ( e i U x k ) e i η 2
for every k = 1 , , m . Let x B X and y B Y .
Let k 0 { 1 , , m } be such that
U x U x k 0 η 2 .
Then we have
i l | ( e i U x ) y ( e i ) | i l | ( e i U ( x x k 0 ) ) y ( e i ) | + i l | ( e i U x k 0 ) y ( e i ) | i l γ i ( e i U ( x x k 0 ) ) y ( e i ) + i l δ i ( e i U x k ) y ( e i ) ( | γ i | = 1 = | δ i | ) i l γ i ( e i U ( x x k 0 ) ) e i + i l δ i ( e i U x k ) e i i = 1 ( e i U ( x x k 0 ) ) e i + i l ( e i U x k ) e i U x U x k 0 + η 2 η .
By Proposition 3 and the above argument,
α σ ( u ; X , Y ) | i = 1 e i U e i | σ = U = ε ( u ; X , Y ) .
The other part of the assertion follows from α σ t = α σ and ε t = ε . ☐
A tensor norm α is called right-injective (respectively, right-projective) if for every isometry I : Y Z (respectively, quotient operator q : Y Z ), the operator
i d X I ( respectively , i d X q ) : X α Y X α Z
is an isometry (respectively, a quotient operator) for all Banach spaces X , Y and Z. If α t is right-injective (respectively, right-projective), then α is called left-injective (respectively, left-projective).
An operator ideal is said to be surjective if [ A , · A ] s u r = [ A , · A ] . According to [2] (Theorem 20.11), a maximal operator ideal is surjective if and only if its associated tensor norm is left-injective.
Example 1
(Non-injectiveness of α σ ). We show that
I σ s u r I σ .
For every separable Banach space X and every Banach space Y,
I σ s u r ( X , Y ) = L ( X , Y ) .
Indeed, according to [3] (Theorem 23.3.4), an operator T : X Y is σ-integral if and only if i Y T is factored through some Banach lattice, where i Y : Y Y is the canonical isometry. Consequently, T q X I σ ( 1 , Y ) for every T L ( X , Y ) .
On the other hand, there exists a separable Banach space Z such that i d Z I σ ( Z , Z ) (cf. [5] (p. 364)). Hence
I σ s u r ( Z , Z ) = L ( Z , Z ) I σ ( Z , Z ) .
Example 2.
(Non-projectiveness of α σ ) The following argument is due to the proof of [2] (Proposition 4.3). Let q 2 : 1 2 be the canonical quotient operator. Consider the map
i d 2 q 2 : 2 ^ α σ 1 2 ^ α σ 2 .
By Proposition 4,
2 ^ α σ 1 = 2 ^ ε 1 = K ( 2 , 1 ) and 2 ^ α σ 2 = 2 ^ ε 2 = K ( 2 , 2 )
hold isometrically. Consequently, the map i d 2 q 2 can be viewed from K ( 2 , 1 ) from K ( 2 , 2 ) .
Now, let T : 2 2 be a compact operator failed to be Hilbert–Schmidt. If i d 2 q 2 would be surjective, then there exists an R K ( 2 , 1 ) such that
q 2 R = i d 2 q 2 ( R ) = T .
This is a contradiction because q 2 R is Hilbert–Schmidt.

5. Discussion

We introduce a new tensor norm and associate it with an operator ideal. This work continues the study of theory of tensor norms and we expect that several more results on tensor norms and operator ideals can be developed. We introduce one of the important subjects. For a finitely generated tensor norm α , a Banach space X is said to have the α -approximation property ( α -AP) if for every Banach space Y, the natural map
J α : Y ^ α X Y ^ ε X
is injective (cf. [3] (Section 21.7)). We can consider the α σ -AP and the following problems.
Problem 1. Does every Banach space have the α σ -AP?
Problem 2. For every Banach space X, if X has the α σ -AP, then does X have the α σ -AP?

Author Contributions

Conceptualization, J.M.K.; methodology, J.M.K. and K.Y.L.; software, J.M.K. and K.Y.L.; validation, J.M.K. and K.Y.L.; formal analysis, J.M.K. and K.Y.L.; investigation, J.M.K. and K.Y.L.; resources, J.M.K. and K.Y.L.; data curation, J.M.K. and K.Y.L.; writing—original draft preparation, J.M.K.; writing—review and editing, J.M.K.; visualization, J.M.K.; supervision, J.M.K.; project administration, J.M.K.; funding acquisition, J.M.K. All authors have read and agreed to the published version of the manuscript.

Funding

The first author was supported by NRF-2018R1D1A1B07043566 (Korea). The second author was supported by NRF-2017R1C1B5017026 (Korea).

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Ryan, R.A. Introduction to Tensor Products of Banach Spaces; Springer: Berlin/Heidelberg, Germany, 2002. [Google Scholar]
  2. Defant, A.; Floret, K. Tensor Norms and Operator Ideals; Elsevier: Amsterdam, The Netherlands, 1993. [Google Scholar]
  3. Pietsch, A. Operator Ideals; Elsevier: Amsterdam, The Netherlands, 1980. [Google Scholar]
  4. Grothendieck, A. Produits tensoriels topologiques et espaces nucléaires. Am. Math. Soc. 1955, 16, 1–185. [Google Scholar]
  5. Diestel, J.; Jarchow, H.; Tonge, A. Absolutely Summing Operators; Cambridge University Press: Cambridge, UK, 1995. [Google Scholar]
Back to TopTop