\(L\)-embedded Banach spaces and measure topology (Q2339737): Difference between revisions

The author introduces a topology which ``looks like'' the topology of convergence in measure in \(L^1\) for spaces which ``look like'' \(L^1\) and proves that many, but not all, non-trivial facts about the topology of convergence in measure in \(L^1\) remain true in this setting. These spaces are those which are \(L\)-embedded, that is, the spaces \(X\) whose bidual can be written as \(X^{\ast\ast} = X \oplus_1 X_s\), the sum being an \(\ell_1\) one, as it is the case for \(L^1\). To define the suitable topology, the author uses the following consequence of the Kadeč-Pełczyński theorem: A bounded sequence in \(L^1\) converges to \(0\) in measure if and only if every subsequence has a further subsequence which converges in norm or spans \(\ell_1\) almost isometrically (such a characterization had also been proved by the author for preduals of finite von Neumann algebras [\textit{H. Pfitzner}, J. Oper. Theory 47, No. 1, 145--167 (2002; Zbl 1029.46092)]). In fact, the author [Proc. Am. Math. Soc. 129, No. 5, 1367--1373 (2001; Zbl 0983.46007)] had proved that a sequence in an \(L\)-embedded Banach space which spans \(\ell_1\) almost isometrically contains a subsequence spanning \(\ell_1\) asymptotically isometrically. The main technical lemma (Lemma~3.3) asserts that for every net \((x_\gamma)\) in an \(L\)-embedded space \(X\) \(w^\ast\)-converging in \(X^{\ast\ast}\) to \(x^{\ast\ast} \in X^{\ast\ast} \setminus X\), there is a bounded sequence \((y_n)\) of convex combinations of the \(x_\gamma\)'s such that the sequence \((y_n - Px^{\ast\ast})\) spans \(\ell_1\) asymptotically isometrically. The author then shows (Theorem~5.2) that every \(L\)-embedded Banach space admits a unique abstract measure topology, i.e., a topology \(\tau_\mu\) having the following properties: {\parindent=0,6cm\begin{itemize}\item[(1)] it is a sequential topology (i.e., sequentially closed sets are closed) in which every convergent sequence has a unique limit; \item[(2)] it is coarser than the norm topology;\item[(3)] it is translation invariant; \item[(4)] each bounded sequence in \(X\) spanning \(\ell_1\) almost isometrically converges to \(0\) for \(\tau_\mu\), and, conversely, every \(\tau_\mu\)-null sequence is bounded and contains a subsequence which is norm-null or spans \(\ell_1\) almost isometrically. \end{itemize}} An abstract version of the Bukhvalov-Lozanovskiĭ theorem is then proved (Theorem~6.3): If \(X\) is \(L\)-embedded and \(P\) is the \(L\)-projection from \(X^{\ast\ast}\) onto \(X\), then for every norm bounded convex set \(C \subseteq X\), \(C\) is \(\tau_\mu\)-closed if and only if \(P \big({\overline C}^{w^\ast}\! \big) = C\). It follows that a subspace \(Y\) of an \(L\)-embbedded space \(X\) is itself \(L\)-embedded if and only if its unit ball \(B_Y\) is \(\tau_\mu\)-closed in \(X\). Then, an abstract version of the characterization of the subspaces of \(L^1\) which are dual of an \(M\)-embedded space given in [\textit{G. Godefroy} and \textit{D. Li}, Math. Scand. 66, No. 2, 249--263 (1990; Zbl 0687.46010)] is proved in Corollary 7.3: If \(X\) is \(L\)-embedded, then \(X\) is (isometrically isomorphic to) the dual of an \(M\)-embedded Banach space \(Z\) if and only if the subspace \(X^\sharp\) of the continuous linear forms on \(X\) whose restriction to \(B_X\) is \(\tau_\mu\)-continuous separates the points of \(X\), and when this is the case, \(Z\) is isometrically isomorphic to \(X^\sharp\). The abstract version of (part of) the main result of [\textit{G. Godefroy} et al., J. Reine Angew. Math. 471, 43--75 (1996; Zbl 0842.46008)] is obtained in Proposition 7.7: If \(X\) is a separable \(L\)-embedded Banach space, then the following assertions are equivalent: {\parindent=0,6cm\begin{itemize}\item[(1)] \(B_X\) is \(\tau_\mu\)-sequentially (Hausdorff) compact and locally convex; \item[(2)] \(B_X\) is \(\tau_\mu\)-(Hausdorff) compact and locally convex; \item[(3)] for every \(\varepsilon > 0\) there is a subspace \(Z_\varepsilon\) of \(c_0\) such that \(d_{BM} (X, Z_\varepsilon^\ast) \leq 1 + \varepsilon\). \end{itemize}} As negative property, it is shown that there is in general no version of the Komlós theorem. The paper ends with several questions; two of them are particularly intriguing: Is \(\tau_\mu\) Hausdorff? Is \(\tau_\mu\) a vector topology? To end this review, I should say that, though the author kindly quotes me for some unpublished notes (which actually go back to the late 1980s, and not to 2000), these notes, except the idea to define such a topology, contain no interesting result.