On quasicomplemented Hilbertian subspaces of Banach spaces (Q2661251)

This is a technical paper on an interesting topic, where ``technical'' means that the authors work on forging new tools and ``interesting'' that there are many places where the tool could be used. The authors deal with the never well understood topic of the existence of subspaces \(A, B\) of a Banach space \(X\) having a certain property and so that \(A+B\) does not. For instance, if \(Z_2\) is the celebrated Kalton-Peck space [\textit{N. J. Kalton} and \textit{N. T. Peck}, Trans. Am. Math. Soc. 255, 1--30 (1979; Zbl 0424.46004)], then \(Z_2\) has no unconditional basis while \(A\cap B=\{0\}\) and \(A+B\) is dense in \(Z_2\) with \(A, B\) spaces with unconditional basis. Two subspaces \(A, B\) as above are called quasi-complemented after [\textit{H. P. Rosenthal}, Proc. Natl. Acad. Sci. USA 59, 361--364 (1968; Zbl 0157.43701)]. Subspaces of separable Banach spaces are always quasi-complemented but, and here is the tricky point, one cannot choose how nice the quasi-complement is or how nice quasi-complemented subspaces are. Observe the following result that is essentially due to [\textit{V.~I. Gurariĭ} and \textit{M.~I. Kadets}, Sov. Math., Dokl. 3, 966--968 (1962; Zbl 0135.34601); translation from Dokl. Akad. Nauk SSSR 145, 256--258 (1962)]: If a separable Banach space contains \(\ell_2\), then it contains two quasicomplemented subspaces isomorphic to \(\ell_2\). In this paper, the authors provide, but this is just a consequence of their results, complete characterizations of Banach spaces having quasicomplemented Hilbert subspaces; see Theorem~2.4: A Banach space contains two quasi-complemented Hilbert subspaces if and only if it is Hilbert generated and contains a Hilbert subspace with the same density character as the whole space. The paper contains many more technical results and, in particular, applications and improvements of Krivine's theorem about the existence of subspaces \(E\) of Hilbert spaces so that the \(\Vert\cdot\Vert_1\) and \(\Vert\cdot\Vert_2\) norms are equivalent on both \(E\) and~\(E^\bot\) [\textit{J.-L. Krivine}, in: Sémin. analyse fonctionnelle, Paris 1983--84, Publ. Math. Univ. Paris VII 20, 21--26 (1984; Zbl 0598.46008)].