On the nontrivial projection problem (Q1016473): Difference between revisions

The nontrivial projection problem is the following: Do there exist a constant \(C \geq 1\) and a sequence of integers \(k_n\), growing to infinity, such that, on every \(n\)-dimensional normed space \(X\), there is a projection \(P\) with (i) \(\| P \| \leq C\) and (ii) \(k_n \leq \text{rank}\, P \leq n - k_n\)? This conjecture is connected to a version of the uniformly complemented \(\ell_p^n\) conjecture of Lindenstrauss (and the cotype-type conjecture) and is implied by the conjecture of the existence of generalized Auerbach systems. We refer to the paper for the statements of these conjectures. In the paper under review, the authors prove a weak version of the nontrivial projection problem, with (i\(^\prime\)) \(\| P \| \leq C\,(1 + \log k_n)^2\) instead of (i). The proof is based on a dichotomy which yields either: (a) a ``good enough'' complemented copy of \(\ell_2^{k_n}\), or (b) a good copy of \(\ell_\infty^{k_n}\) in the space \(X\) or in its dual \(X^\ast\). The difficult part is (b), which needs various tricky results obtained since the 1980s. In the last part, the authors prove that their main result is ``almost optimal'', even for the spaces \(X = \ell_q^n\). Fortunately for non-specialists, they give proofs of some estimates on factorization constants which are considered as well-known, but whose proofs have not explicitly appeared before (and ``seem to require combining formulae from several sources'').