SearcharxivSearch

arXiv subjects

Haskell P. Rosenthal

Publications and source records attributed to Haskell P. Rosenthal.

12 recordsLinked to original sources

M-Complete approximate identities in operator spaces

This work introduces the concept of an M-complete approximate identity (M-cai) for a given operator subspace X of an operator space Y. M-cai's generalize central approximate identities in ideals in $C^*$-algebras, for it is proved that if X admits an M-cai in Y, then X is a complete M-ideal in Y. It is proved, using ``special'' M-cai's, that if $\cal J$ is a nuclear ideal in a $C^*$-algebra $\cal A$, then $\cal J$ is completely complemented in Y for any (isomorphically) locally reflexive operator space Y with $\cal J \subset Y \subset \cal A$ and $Y/\cal J$ separable. (This generalizes the previously known special case where $Y=\cal A$, due to Effros-Haagerup.) In turn, this yields a new proof of the Oikhberg-Rosenthal Theorem that $\cal K$ is completely complemented in any separable locally reflexive operator superspace, $\cal K$ the $C^*$-algebra of compact operators on $\ell^2$. M-cai's are also used in obtaining some special affirmative answers to the open problem of whether $\cal K$ is Banach-complemented in $\cal A$ for any separable $C^*$-algebra $\cal A$ with $\cal K\subset\cal A\subset B(\ell^2)$. It is shown that if conversely X is a complete M-ideal in Y, then X admits an M-cai in Y in the following situations: (i) Y has the (Banach) bounded approximation property; (ii) Y is 1-locally reflexive and X is $λ$-nuclear for some $λ\ge1$; (iii) X is a closed 2-sided ideal in an operator algebra Y (via the Effros-Ruan result that then X has a contractive algebraic approximate identity). However it is shown that there exists a separable Banach space X which is an M-ideal in $Y=X^{**}$, yet X admits no M-approximate identity in Y.

math.OA

On certain extension properties for the space of compact operators

Let $Z$ be a fixed separable operator space, $X\subset Y$ general separable operator spaces, and $T:X\to Z$ a completely bounded map. $Z$ is said to have the Complete Separable Extension Property (CSEP) if every such map admits a completely bounded extension to $Y$; the Mixed Separable Extension Property (MSEP) if every such $T$ admits a bounded extension to $Y$. Finally, $Z$ is said to have the Complete Separable Complementation Property (CSCP) if $Z$ is locally reflexive and $T$ admits a completely bounded extension to $Y$ provided $Y$ is locally reflexive and $T$ is a complete surjective isomorphism. Let ${\bf K}$ denote the space of compact operators on separable Hilbert space and ${\bf K}_0$ the $c_0$ sum of ${\Cal M}_n$'s (the space of ``small compact operators''). It is proved that ${\bf K}$ has the CSCP, using the second author's previous result that ${\bf K}_0$ has this property. A new proof is given for the result (due to E. Kirchberg) that ${\bf K}_0$ (and hence ${\bf K}$) fails the CSEP. It remains an open question if ${\bf K}$ has the MSEP; it is proved this is equivalent to whether ${\bf K}_0$ has this property. A new Banach space concept, Extendable Local Reflexivity (ELR), is introduced to study this problem. Further complements and open problems are discussed.

math.OA

On Differences of Semi-Continuous Functions

Extrinsic and intrinsic characterizations are given for the class DSC(K) of differences of semi-continuous functions on a Polish space K, and also decomposition characterizations of DSC(K) and the class PS(K) of pointwise stabilizing functions on K are obtained in terms of behavior restricted to ambiguous sets. The main, extrinsic characterization is given in terms of behavior restricted to some subsets of second category in any closed subset of K. The concept of a strong continuity point is introduced, using the transfinite oscillations osc$_αf$ of a function $f$ previously defined by the second named author. The main intrinsic characterization yields the following DSC analogue of Baire's characterization of first Baire class functions: a function belongs to DSC(K) iff its restriction to any closed non-empty set L has a strong continuity point. The characterizations yield as a corollary that a locally uniformly converging series $\sum ϕ_j$ of DSC functions on K converges to a DSC function provided $\sum{osc}_αϕ_j$ converges locally uniformly for all countable ordinals $α$.

math.FA

The complete separable extension property

This work introduces operator space analogues of the Separable Extension Property (SEP) for Banach spaces; the Complete Separable Extension Property (CSEP) and the Complete Separable Complemention Property (CSCP). The results use the technique of a new proof of Sobczyk's Theorem, which also yields new results for the SEP in the non-separable situation, e.g., $(\oplus_{n=1}^\infty Z_n)_{c_0}$ has the $(2+\ep)$-SEP for all $\ep>0$ if $Z_1,Z_2,...$ have the 1-SEP; in particular, $c_0 (\ell^\infty)$ has the SEP. It is proved that e.g., $c_0(\bR\oplus\bC)$ has the CSEP (where $\bR$, $\bC$ denote Row, Column space respectively) as a consequence of the general principle: if $Z_1,Z_2,...$ is a uniformly exact sequence of injective operator spaces, then $(\oplus_{n=1}^\infty Z_n)_{c_0}$ has the CSEP. Similarly, e.g., $\bK_0 \defeq (\oplus_{n=1}^\infty M_n)_{c_0}$ has the CSCP, due to the general principle: $(\oplus_{n=1}^\infty Z_n)_{c_0}$ has the CSCP if $Z_1,Z_2,...$ are injective separable operator spaces. Further structural results are obtained for these properties, and several open problems and conjectures are discussed.

math.OA

On an inequality of A.~Grothendieck concerning operators on $L^1$

In 1955, A.~Grothendieck proved a basic inequality which shows that any bounded linear operator between $L^1(μ)$-spaces maps (Lebesgue-) dominated sequences to dominated sequences. An elementary proof of this inequality is obtained via a new decomposition principle for the lattice of measurable functions. An exposition is also given of the M.~Lévy extension theorem for operators defined on subspaces of $L^1(μ)$-spaces.

math.FA

On wide-$(s)$ sequences and their applications to certain classes of operators

A basic sequence in a Banach space is called wide-$(s)$ if it is bounded and dominates the summing basis. (Wide-$(s)$ sequences were originally introduced by I.~Singer, who termed them $P^*$-sequences). These sequences and their quantified versions, termed $λ$-wide-$(s)$ sequences, are used to characterize various classes of operators between Banach spaces, such as the weakly compact, Tauberian, and super-Tauberian operators, as well as a new intermediate class introduced here, the strongly Tauberian operators. This is a nonlocalizable class which nevertheless forms an open semigroup and is closed under natural operations such as taking double adjoints. It is proved for example that an operator is non-weakly compact iff for every $\varepsilon >0$, it maps some $(1+\varepsilon)$-wide-$(s)$-sequence to a wide-$(s)$ sequence. This yields the quantitative triangular arrays result characterizing reflexivity, due to R.C.~James. It is shown that an operator is non-Tauberian (resp. non-strongly Tauberian) iff for every $\varepsilon>0$, it maps some $(1+\varepsilon)$-wide-$(s)$ sequence into a norm-convergent sequence (resp. a sequence whose image has diameter less than $\varepsilon$). This is applied to obtain a direct ``finite'' characterization of super-Tauberian operators, as well as the following characterization, which strengthens a recent result of M.~González and A.~Mart{\'ı}nez-Abejón: An operator is non-super-Tauberian iff there are for every $\varepsilon>0$, finite $(1+\varepsilon)$-wide-$(s)$ sequences of arbitrary length whose images have norm at most $\varepsilon$.

math.FA

Differences of bounded semi-continuous functions, I

Structural properties are given for $D(K)$, the Banach algebra of (complex) differences of bounded semi-continuous functons on a metric space $K$. For example, it is proved that if all finite derived sets of $K$ are non-empty, then a complex function $φ$ operates on $D(K)$ (i.e., $φ\circ f\in D(K)$ for all $f\in D(K)$) if and only if $φ$ is locally Lipschitz. Another example: if $W\subset K$ and $g\in D(W)$ is real-valued, then it is proved that $g$ extends to a $\tilde g$ in $D(K)$ with $\|\tilde g\|_{D(K)} = \|g\|_{D(W)}$. Considerable attention is devoted to $SD(K)$, the closure in $D(K)$ of the set of simple functions in $D(K)$. Thus it is proved that every member of $SD(K)$ is a (complex) difference of semi-continuous functions in $SD(K)$, and that $|f|$ belongs to $SD(K)$ if $f$ does. An intrinsic characterization of $SD(K)$ is given, in terms of transfinite oscillation sets. Using the transfinite oscillations, alternate proofs are given of the results of Chaatit, Mascioni and Rosenthal that functions of finite Baire-index belong to $SD(K)$, and that $SD(K)\ne D(K)$ for interesting $K$. It is proved that the ``variable oscillation criterion'' characterizes functions belonging to $B_{1/4}(K)$, thus answering an open problem raised in earlier work of Haydon, Odell and Rosenthal. It is also proved that $f$ belongs to $B_{1/4}(K)$ (if and) only if $f$ is a uniform limit of simple $D$-functions of uniformly bounded $D$-norm iff $\osc_ωf$ is bounded; the last equivalence has also been obtained by V.~Farmaki, using other methods.

math.FA

On Functions of Finite Baire Index

It is proved that every function of finite Baire index on a separable metric space $K$ is a $D$-function, i.e., a difference of bounded semi-continuous functions on $K$. In fact it is a strong $D$-function, meaning it can be approximated arbitrarily closely in $D$-norm, by simple $D$-functions. It is shown that if the $n^{th}$ derived set of $K$ is non-empty for all finite $n$, there exist $D$-functions on $K$ which are not strong $D$-functions. Further structural results for the classes of finite index functions and strong $D$-functions are also given.

math.FA

A characterization of Banach spaces containing $c_0$

A subsequence principle is obtained, characterizing Banach spaces containing $c_0$, in the spirit of the author's 1974 characterization of Banach spaces containing $\ell^1$. Definition: A sequence $(b_j)$ in a Banach space is called {\it strongly summing\/} (s.s.) if $(b_j)$ is a weak-Cauchy basic sequence so that whenever scalars $(c_j)$ satisfy $\sup_n \|\sum_{j=1}^n c_j b_j\| <\infty$, then $\sum c_j$ converges. A simple permanence property: if $(b_j)$ is an (s.s.) basis for a Banach space $B$ and $(b_j^*)$ are its biorthogonal functionals in $B^*$, then $(\sum_{j=1}^n b_j^*)_{n=1}^ \infty$ is a non-trivial weak-Cauchy sequence in $B^*$; hence $B^*$ fails to be weakly sequentially complete. (A weak-Cauchy sequence is called {\it non-trivial\/} if it is {\it non-weakly convergent\/}.) Theorem. Every non-trivial weak-Cauchy sequence in a (real or complex) Banach space has either an {\rm (s.s.)} subsequence, or a convex block basis equivalent to the summing basis. Remark : The two alternatives of the Theorem are easily seen to be mutually exclusive. Corollary 1. A Banach space $B$ contains no isomorph of $c_0$ if and only if every non-trivial weak-Cauchy sequence in $B$ has an {\rm (s.s.)} subsequence. Combining the $c_0$ and $\ell^1$ Theorems, we obtain Corollary 2. If $B$ is a non-reflexive Banach space such that $X^*$ is weakly sequentially complete for all linear subspaces $X$ of $B$, then $c_0$ embeds in $X$; in fact, $B$ has property~$(u)$.

math.FA

On Weakly Null FDD's in Banach Spaces

In this paper we show that every sequence (F_n) of finite dimensional subspaces of a real or complex Banach space with increasing dimensions can be ``refined'' to yield an F.D.D. (G_n), still having increasing dimensions, so that either every bounded sequence (x_n), with x_n in G_n for n in N, is weakly null, or every normalized sequence (x_n), with x_n in G_n for n in N, is equivalent to the unit vector basis of l_1. Crucial to the proof are two stabilization results concerning Lipschitz functions on finite dimensional normed spaces. These results also lead to other applications. We show, for example, that every infinite dimensional Banach space X contains an F.D.D. (F_n), with lim_{n to infty} dim (F_n)=infty, so that all normalized sequences (x_n), with x_n in F_n, n in N, have the same spreading model over X. This spreading model must necessarily be 1-unconditional over X.

math.FA