Almost square Banach spaces (Q891418)

For real Banach spaces the properties almost square, weakly almost square and locally almost square are defined and investigated. The abbreviations are ASQ, WASQ and LASQ, respectively. Here is an attempt to motivate the definition of the three properties: Given \(n\) points \(x_1,x_2,\dots,x_n\) in the unit sphere \(S_X\), suppose we want to look for the infimum of \(\|x_i\pm y\|\) over \(y\in S_X\). Depending on the geometry in \(X\), the solution is a number between \(1\) and \(2\). If the solution is \(1\), no matter of the choice of points \(x_1,x_2,\dots,x_n\), \(X\) is ASQ. In other words, ASQ means that \[ t(X)=\sup_{x_1,x_2,\dots,x_n\in S_X}\inf_{y\in S_X}\max_i\|x_i\pm y\|=1. \] \(t(X)\) is known as Whitley's thinness index. If the solution to the above problem, but with \(n=1\), is \(1\), then the space is LASQ. Now, let \(y_k\) be such that \[ \sup_{x_1,x_2,\dots,x_n\in S_X}\max_i\|x_i\pm y_k\|\to 1, \] which exist in an ASQ space. It is a basic, but not at all easy, observation (Theorem~2.8) that we may assume \(y_k\to 0\) weakly. On their way to this result, the authors discover that ASQ-spaces contain an asymptotically isometric copy of \(c_0\). The authors have not been able to prove that one may assume \(y_k\to 0\) weakly in the LASQ case (Question 3.4), but leave this as the property WASQ. However, WASQ turns out to be very interesting in itself. LASQ spaces belong to the ``big slice world'', i.e., every slice of \(B_X\) has diameter \(2\) (local diameter \(2\) property [LD2P]). \textit{D. Kubiak} [J. Convex Anal. 21, No. 1, 189--200 (2014; Zbl 1300.46016)] proved this and also that WASQ implies that every non-void relatively weakly open subset of \(B_X\) has diameter \(2\) (diameter \(2\) property [D2P]). The authors prove that ASQ implies that every finite convex combination of slices of \(B_X\) has diameter \(2\) (strong diameter \(2\) property [SD2P]) by proving (Proposition 2.5) that \(X\) ASQ implies that \(X^\ast\) is octahedral (which is known to be equivalent to \(X\) having the SD2P). This ends the description of the first two sections of the paper. Section 3 contains illustrating examples of WASQ- and ASQ-spaces. In particular, it is observed that \(c_0(X_i)\) is always ASQ. More difficult is the result that the Cesàro function space \(C_{p,\omega}\) (shown by Kubiak to be WASQ) is not ASQ because it does not contain any copy of \(c_0\). In Section 4, we learn that (L)ASQ passes down to almost isometric ideals, and that non-reflexive \(M\)-embedded Banach spaces are ASQ. In Section 5, stability with respect to direct sums of (L)[W]ASQ is studied. LASQ and WASQ are stable with respect to \(\ell_p\)-sums for all \(1\leq p\leq\infty\), while ASQ always breaks down under \(\ell_p\)-sums unless \(p=\infty\), in which case it is stable. In the final Section 6, the connection of ASQ with the intersection property (IP) of \textit{E. Behrends} and \textit{P. Harmand} [Stud. Math. 81, 159--169 (1985; Zbl 0529.46015)] is discussed. Denote by \(B_{(\varepsilon,1]}\) all elements of \(B_X\) with norm \(>\varepsilon\). Look at the index \[ \gamma(X,\varepsilon)=\sup_{x_1,x_2,\dots,x_n\in S_X}\inf_{y\in B_{(\varepsilon,1]}}\max_i\|x_i\pm y\|. \] The authors say that \(X\) \(\varepsilon\)-fails the IP if \(\gamma(X,\varepsilon)=1\) and then describe ASQ as \(X\) \(\varepsilon\)-fails the IP for all \(0<\varepsilon<1\) (Proposition 6.1). Following the question of Behrends and Harmand [loc. cit.] whether all dual spaces have the IP, the authors ask whether any dual space can be ASQ (Question 6.6).