On almost isometric ideals in Banach spaces (Q314494): Difference between revisions

The paper contains interesting remarks on, and examples of, almost isometric ideals (a.i.-ideals). It also contains some open questions. Such locally 1-complemented subspaces (= ideals) were defined and studied by the reviewer and his co-authors in [\textit{T. A. Abrahamsen} et al., Glasg. Math. J. 56, No. 2, 395--407 (2014; Zbl 1303.46012)]: An a.i.-ideal is simply an ideal where the local retractions can be chosen as almost isometries. Ideals where the ideal projection has 1-norming range (= strict ideals) are a.i.-ideals, but there are examples showing that a.i.-ideals need not be strict. \(Y\) being an a.i.-ideal in \(X\) is thus somehow that \(Y\subset X\) like \(X\subset X^{\ast\ast}\) as far as the principle of local reflectivity is concerned, except for weak-star denseness of \(X^\ast\) in the triple dual. The motivation behind defining a.i.-ideals was to make big slice phenomena pass down from \(X\) to \(Y\). As an example, the Daugavet property passes down to a.i.-ideals. Let us now turn to the content of the paper. The first result states that the separable Gurariy space \(G\) contains no other proper a.i.-ideals but isometric copies of itself. (Reviewer's remark: The same proof gives that any a.i.-ideal of any Gurariy space must be a Gurariy space.) Since \(G\) is universal, it contains all separable Lindenstrauss spaces. Thus it contains lots of ideals which are not a.i.-ideals. The next result is that \(G\) also contains no other proper \(M\)-ideals than itself. Thus, in \(G\), we have that the a.i-ideals are exactly the \(M\)-ideals, which are again exactly the isometric copies of \(G\). Next, again involving \(G\) and the results above, it is observed that being an a.i.-ideal is not a 3-space property, that \(C(K,G)\neq G\) if \(K\) has more than one element, and that \(G\oplus_\infty G\neq G\). The author now investigates general transitivity principles and obtains that, if \(Y^{\perp\perp}\) is an a.i.-ideal in \(X^{\ast\ast}\), then \(Y\) must be an a.i.-ideal in \(X\). It is stated as a general question whether \(X^{\perp\perp}\) is always an a.i.-ideal in \(X^{(4)}\). After these general results, the author now looks into \(c_0\) and shows that proximinal ideals there of finite co-dimension are a.i.-ideals. In particular thus, \(M\)-ideals in \(c_0\) of finite co-dimension are a.i.-ideals. This highly generalizes Example 1 from the paper of Abrahamsen et al. [loc. cit.]. Indeed, much more might be true; as the author remarks, we don't know whether every infinite-dimensional ideal in \(c_0\) already is a.i. Concerning the question just mentioned, the answer is always no if one takes a Lindenstrauss space with non-separable dual instead of \(c_0\), as the author demonstrates in Proposition 14. For the last part of the paper, recall Fakhoury's theorem that a Banach space is a Lindenstrauss space if and only if it is an ideal in every super space and the result from Abrahamsen et al. [loc. cit.] that it is a Gurariy space if and only if it is an a.i.-ideal in every super space. The author now improves Fakhoury's theorem by showing that a Banach space is a Lindenstrauss space if and only if it is an ideal in every super space where it is a hyperplane, and asks if the above characterization of Gurariy spaces can be improved similarly, at least in the separable case.