Smooth points in spaces of operators

Smooth points in spaces of operators

Linear Algebra and its Applications 517 (2017) 129–133 Contents lists available at ScienceDirect Linear Algebra and its Applications www.elsevier.co...

223KB Sizes 0 Downloads 66 Views

Linear Algebra and its Applications 517 (2017) 129–133

Contents lists available at ScienceDirect

Linear Algebra and its Applications www.elsevier.com/locate/laa

Smooth points in spaces of operators T.S.S.R.K. Rao Theoretical Statistics and Mathematics Unit, Indian Statistical Institute, R. V. College P.O., Bangalore 560059, India

a r t i c l e

i n f o

Article history: Received 4 November 2016 Accepted 9 December 2016 Available online 14 December 2016 Submitted by P. Semrl MSC: primary 47L05, 46B20

a b s t r a c t For Banach spaces X, Y , in the space of bounded linear operators L(X, Y ), we examine the relation between T ∈ L(X, Y ) being a smooth point versus T ∗ ∈ L(Y ∗ , X ∗ ) being a smooth point. Motivated by some results in the recent paper of Paul et al., we give some sufficient conditions for the validity of such a statement for spaces of operators. © 2016 Elsevier Inc. All rights reserved.

Keywords: Smooth points Spaces of operators Hahn–Banach smooth spaces

1. Introduction Let X be a real Banach space. A non-zero vector x0 ∈ X is said to be a smooth point, if there is a unique unit vector x∗ ∈ X ∗ such that x∗ (x0 ) = x0 . A well-known theorem of Mazur asserts that in a separable Banach space smooth points are dense (see [2] page 171). An interesting question is to study these points for L(X, Y ) and its subspaces. We use standard linear algebra of tensors and basic functional analysis in our analysis. By an operator we mean a bounded linear map. We recall that x ∈ X is said to be Birkhoff–James orthogonal to y ∈ X if x ≤ x + λy for all real numbers λ. It is known that x0 is a smooth point if and only if E-mail addresses: [email protected], [email protected]. http://dx.doi.org/10.1016/j.laa.2016.12.011 0024-3795/© 2016 Elsevier Inc. All rights reserved.

130

T.S.S.R.K. Rao / Linear Algebra and its Applications 517 (2017) 129–133

whenever x0 is orthogonal to y and x0 is orthogonal to z implies x0 is orthogonal to y + z. In [3], the authors relate, in some special cases, the orthogonality of operators in K(X, Y ) to the orthogonality of vectors in Y at the points where the operators attain the norm. Since T → T ∗ is an isometry (surjective only when Y is reflexive, see [2] Exercise 3.9 on page 192), it is an interesting question to consider when smoothness gets preserved in K(Y ∗ , X ∗ ) under this into isometry. Throughout the paper we consider a Banach space X as canonically embedded in its bidual and recall that for a continuous linear map T on X, T ∗∗ = T on X. We first note the easy implication that if T ∗ is a smooth point of K(Y ∗ , X ∗ ), then since it is also a smooth point of {S ∗ : S ∈ K(X, Y )}, T is a smooth point of K(X, Y ). For x∗∗ ∈ X ∗∗ and y ∗ ∈ Y ∗ , we denote by x∗∗ ⊗ y ∗ the linear functional defined on the space of operators by (x∗∗ ⊗ y ∗ )(T ) = x∗∗ (T ∗ (y ∗ )), for any operator T and note that x∗∗ ⊗ y ∗  = x∗∗ y ∗ . We denote a rank one operator by x∗ ⊗ y, so that (x∗∗ ⊗ y ∗ )(x∗ ⊗ y) = x∗∗ (x∗ )y ∗ (y). Let y0 ∈ Y be a unit vector and a smooth point. Suppose T = x∗ ⊗ y0 is a smooth point of K(X, Y ), then if T ∗ = y0 ⊗ x∗ is a smooth point of K(Y ∗ , X ∗ ), it is easy to see that y0 is a smooth point of Y ∗∗ . Such points are called, very smooth points in the literature, see [6]. Thus one needs additional conditions to deduce the smoothness of T ∗ from that of T , even for rank one operators. This affects Corollary 4.2.1 in [3]. See also [4]. An interesting question that arises in this context is to give sufficient conditions on X or Y to ensure that T ∈ K(X, Y ) (T ∈ L(X, Y )) is a smooth point implies T ∗ ∈ K(Y ∗ , X ∗ ) (T ∗ ∈ L(Y ∗ , X ∗ )) is a smooth point. We recall that X is said to be a Hahn–Banach smooth space, if under the canonical embedding, functionals in X ∗ have unique norm-preserving extension in X ∗∗∗ . See [1] for several examples of such spaces. We show that if Y is a Hahn–Banach smooth space and T ∈ K(X, Y ) is a smooth point, then T ∗ is a smooth point of K(Y ∗ , X ∗ ). We prove a local version of the above result by showing that if T is a very smooth point of K(X, Y ), then T ∗ is a smooth point of K(Y ∗ , X ∗ ). To answer this question for L(X, Y ), we need stronger geometric assumptions. For a Banach space X, let X1 denote the closed unit ball and let ∂e X1 denote the set of extreme points of X1 . We recall from Chapter I of [1] that a closed subspace M ⊂ X is said to be an M -ideal, if there is a linear projection P : X ∗ → X ∗ such that ker(P ) = M ⊥ and x∗  = P (x∗ ) + x∗ − P (x∗ ) for all x∗ ∈ X ∗ . This geometric condition implies that ∂e X1∗ = ∂e M1∗ ∪ ∂e M1⊥ . Also functionals in M ∗ have unique norm preserving extension in X ∗ . The monograph [1] also studies spaces X, Y for which K(X, Y ) is an M -ideal in L(X, Y ). Let Y be a Hahn–Banach smooth space. Suppose K(Y ∗ , X ∗ ) is an M -ideal in L(Y ∗ , X ∗ ). Let T ∈ L(X, Y ) be a smooth point that attains its norm. We show that if d(T, K(X, Y )) < T , then T ∗ is a smooth point of L(Y ∗ , X ∗ ). Thanks are due to the referee for several suggestions that led to this improved version.

T.S.S.R.K. Rao / Linear Algebra and its Applications 517 (2017) 129–133

131

2. Main results We recall from [5] that ∂e K(X, Y )∗1 = {x∗∗ ⊗ y ∗ : x∗∗ ∈ ∂e X1∗∗ , y ∗ ∈ ∂e Y1∗ }. Theorem 1. Let Y be a Hahn–Banach smooth space. For any smooth point T ∈ K(X, Y ), T ∗ is a smooth point of K(Y ∗ , X ∗ ). Proof. Let T ∈ K(X, Y ) be a smooth point and suppose T  = 1. Since T ∗ is compact, ∗∗ ∗∗ ∗ ∗ ∗∗ let y0∗  = 1 = T ∗ (y0∗ ). Let x∗∗ 0 ∈ ∂e X1 be such that x0 (T (y0 )) = 1. Thus (x0 ⊗ y0∗ )(T ) = 1. Since T is smooth, it is easy to see that y0∗ ∈ ∂e Y1∗ . As Y is Hahn–Banach smooth, y0∗ ∈ ∂e Y1∗∗∗ . Let K = {Λ ∈ K(Y ∗ , X ∗ )∗1 : Λ(T ∗ ) = 1}. K equipped with the weak∗ -topology is a compact, convex, extreme subset of K(Y ∗ , X ∗ )1 . We have from the result of Ruess ∗ and Stegall quoted above that y0∗ ⊗ x∗∗ 0 ∈ ∂e K. Clearly T is smooth if and only if K ∗ ∗∗ is the singleton {y0 ⊗ x0 }. Suppose K is not a singleton. Then by the Krein–Milman theorem, as K is an extreme subset of the dual unit ball, there is another extreme functional, which by the result of Ruess and Stegall, again will be of the form, τ ⊗ x∗∗ , with τ (T ∗∗ (x∗∗ )) = 1, for some τ ∈ ∂e Y1∗∗∗ and x∗∗ ∈ ∂e X1∗∗ . Let y ∗ = τ |Y . We now consider the functional x∗∗ ⊗ y ∗ ∈ K(X, Y )∗1 . As T is a compact operator we note that T ∗∗ maps X ∗∗ into Y . Thus (x∗∗ ⊗ y ∗ )(T ) = 1 and hence x∗∗ ⊗ y ∗  = 1 = y ∗ . Since ∗ ∗ ∗ T is smooth, x∗∗ ⊗ y ∗ = x∗∗ 0 ⊗ y0 on K(X, Y ). If y and y0 are linearly independent ∗ ∗ ∗ on Y , let y0 be such that y (y0 ) = 1 and y0 (y0 ) = 0. Fix x ∈ X ∗ such that x∗∗ (x∗ ) = 1. Now consider the rank one operator S = x∗ ⊗ y0 ∈ K(X, Y ). We have (x∗∗ ⊗ y ∗ )(S) = 1, ∗ ∗ ∗ whereas (x∗∗ 0 ⊗ y0 )(S) = 0. This contradiction shows that y = αy0 and |α| = 1. By ∗∗ ∗∗ replacing x0 by −x0 , if necessary, we assume without loss of generality that α = 1. Again as Y is Hahn–Banach smooth, we have, τ = y0∗ . Similarly using linear dependence, ∗ ∗ ∗ one can also show that x∗∗ = x∗∗ 0 . Hence T is a smooth point of K(Y , X ). 2 In the next proposition, we give a local condition, to ensure that smoothness of T implies that of T ∗ . We recall from [6] that a smooth point x ∈ X is said to be very smooth, if x continues to be a smooth point of X ∗∗ . In this case the x∗ ∈ ∂e X1∗ with x∗ (x) = x, is a point of weak∗ –weak continuity for the identity map on X1∗ , see Lemma III.2.14 in [1]. Proposition 2. Let T ∈ K(X, Y ) be a very smooth point. Then T attains its norm and T ∗ is a smooth point of K(Y ∗ , X ∗ ). Proof. Let T  = 1 and as in the proof of Theorem 1, let x∗∗ ∈ ∂e X1∗∗ , y ∗ ∈ ∂e Y1∗ be such that (x∗∗ ⊗y ∗ )(T ) = 1. Since T is a very smooth point, by what we have remarked above, x∗∗ ⊗ y ∗ is a point of weak∗ –weak continuity for the identity map on K(X, Y )∗1 . We next claim that x∗∗ is a point of weak∗ –weak continuity for the identity map on X1∗∗ . Since X1 is weak∗ -dense in X1∗∗ , this in particular shows that x∗∗ = x ∈ X1 and thus T attains

132

T.S.S.R.K. Rao / Linear Algebra and its Applications 517 (2017) 129–133

∗∗ ∗∗ ∗∗ its norm. Let {x∗∗ in the weak∗ -topology. For α }α∈Δ ⊂ X1 be a net such that xα → x ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ any S ∈ K(X, Y ), we have (x∗∗ ⊗ y ∗ )(S). α ⊗ y )(S) = xα (S (y )) → x (S (y )) = (x ∗∗ ∗ ∗ ∗ Thus the net {xα ⊗ y }α∈Δ ⊂ K(X, Y )1 , converges in the weak -topology to x∗∗ ⊗ y ∗ . Therefore by our hypothesis, we get that this convergence is also in the weak topology. We next note that the injective tensor product, X ∗ ⊗ˆ Y ⊂ K(X, Y ) and thus by duality, (X ∗ ⊗ˆ Y )∗∗ ⊂ K(X, Y )∗∗ . It is easy to see that X ∗∗∗ ⊗ˆ Y ⊂ (X ∗ ⊗ˆ Y )∗∗ such that under the canonical embedding, (τ ⊗ y)(x∗∗ ⊗ y ∗ ) = τ (x∗∗ )y ∗ (y), for τ ∈ X ∗∗∗ , y ∈ Y . See ∗∗ page 265 of [1]. This implies that x∗∗ in the weak topology of X ∗∗ . Thus x∗∗ is a α →x ∗ point of weak –weak continuity. Similarly one can show that y ∗ is a point of weak∗ –weak continuity for the identity map on Y1∗ . By using Lemma III.2.14 of [1] again, we get that y ∗ has unique norm preserving extension in Y ∗∗∗ . Proceeding as in the proof of Theorem 1, we see that T ∗ is a smooth point of K(Y ∗ , X ∗ ). 2

We next study smooth points of L(Y ∗ , X ∗ ). Theorem 3. Let Y be a Hahn–Banach smooth space. Suppose K(Y ∗ , X ∗ ) is an M -ideal in L(Y ∗ , X ∗ ). For any smooth point T ∈ L(X, Y ) that attains its norm and such that d(T, K(X, Y )) < T , T ∗ is a smooth point of L(Y ∗ , X ∗ ). Proof. As in the proof of Theorem 1, let T  = 1 and let K = {Λ ∈ L(Y ∗ , X ∗ )∗1 : Λ(T ∗ ) = 1}. Let K be equipped with the weak∗ -topology. Let x0  = 1 = T (x0 ) and let y0∗ ∈ ∂e Y1∗ be such that y0∗ (T (x0 )) = 1. Since T is a smooth point of L(X, Y ), it is easy to see that T (x0 ) is a smooth point of Y . As K(Y ∗ , X ∗ ) is an M -ideal in L(Y ∗ , X ∗ ), by the properties of M -ideals recalled in the introduction, x0 ⊗ y0∗ ∈ ∂e K. Proceeding as in the proof of Theorem 1, suppose there is a Λ0 ∈ ∂e K such that Λ0 (T ∗ ) = 1. Again by the M -ideal assumption, we have two possibilities. Suppose Λ0 ∈ K(Y ∗ , X ∗ )⊥ . Then Λ0 (T ∗ ) = 1 = d(T ∗ , K(Y ∗ , X ∗ )) ≤ d(T, K(X, Y )) < 1. Thus we only have the possibility, Λ0 ∈ ∂e K(Y ∗ , X ∗ )∗1 . Hence Ruess and Stegall’s theorem ensures that Λ0 = τ ⊗ x∗∗ , as in the proof of Theorem 1. Since in the proof of Theorem 1, we are only manipulating with rank one operators, the conclusion follows. 2 Remark 4. It is clear from the above proof that we only need to assume, extreme points of Y1∗ which attain their norm at a smooth point of Y , have unique norm-preserving extension in Y ∗∗∗ . This assumption is equivalent to the assumption, every smooth point in Y is a very smooth point. Remark 5. We note that if the T in the above theorem is a compact operator and smooth point only in K(X, Y ), then by Theorem 1, we have that T ∗ ∈ K(Y ∗ , X ∗ ) is a smooth point. Since K(Y ∗ , X ∗ ) is an M -ideal in L(Y ∗ , X ∗ ), it is easy to see that T ∗ is a smooth point of L(Y ∗ , X ∗ ). In particular T is also a smooth point of L(X, Y ). It is also easy to see that the same proof goes through, if we assume T ∗ attains its norm and d(T ∗ , K(Y ∗ , X ∗ )) < T . We recall that by Proposition 4 in Section 4 of [7], one has that {T ∈ L(X, Y ) : T ∗ attains its norm} is a dense subset of L(X, Y ).

T.S.S.R.K. Rao / Linear Algebra and its Applications 517 (2017) 129–133

133

References [1] P. Harmand, D. Werner, W. Werner, M -Ideals in Banach Spaces and Banach Algebras, Lecture Notes in Mathematics, vol. 1547, Springer-Verlag, Berlin, 1993, viii+387 pp. [2] R.B. Holmes, Geometric Functional Analysis and Its Applications, Graduate Texts in Mathematics, vol. 24, Springer-Verlag, New York–Heidelberg, 1975, x+246 pp. [3] K. Paul, D. Sain, P. Ghosh, Birkhoff–James orthogonality and smoothness of bounded linear operators, Linear Algebra Appl. 506 (2016) 551–563. [4] T.S.S.R.K. Rao, Review of ‘Birkhoff–James orthogonality and smoothness of bounded linear operators’, Zbl 06617602, 2016. [5] W. Ruess, C. Stegall, Extreme points in duals of operator spaces, Math. Ann. 261 (1982) 535–546. [6] F. Sullivan, Geometric properties determined by the higher duals of a Banach space, Illinois J. Math. 21 (1977) 315–331. [7] V. Zizler, On some extremal problems in Banach spaces, Math. Scand. 32 (1973) 214–224 (1974).