Sections of smooth convex bodies via majorizing measures (Q1372995): Difference between revisions

The main result in this work is the following: Consider vectors \((x_i)_{i\leq n}\) in a Banach space \(X\) and set \(\tau= \sup\{\sum_{i\leq n} x^*(x_i)^2; x^*\in X^*, \| x^*\|\leq 1\}\). Assume that there is another norm \(\|\cdot\|_1\) on \(X\), larger than \(\|\cdot\|\), such that \(\| x_i\|_1\leq 1\), \(i\leq n\), which is 2-smooth, i.e. \(\| x+ y\|+\| x-y\|\leq 2+ C\| y\|^2\) for all \(\| x\|= 1\) and \(\| y\|= 1\), \(C\) being independent of \(x\), \(y\). Consider \(\varepsilon> 0\) and \(m= [n^{1- \varepsilon}/\tau]\). Then for most of the subsets \(I\) of cardinality \(m\), we have for all numbers \((\alpha_i)_{i\in I}\) \[ \Biggl\| \sum_{i\in I} \alpha_i x_i\Biggr\|\leq K\Biggl( \sum_{i\in I} \alpha^2_i\Biggr)^{1/2}, \] where \(K\) depends only on \(\varepsilon\) and \(C\). The author uses tools from probability in Banach spaces in order to obtain sharp probabilistic estimates; he also shows how one can deduce from his result a theorem of \textit{J. Bourgain} on the equivalence of the \(L_p\) and \(L_2\) norms on certain subspaces of \(L^2\) [cf. Acta Math. 162, No. 3/4, 227-245 (1989; Zbl 0674.43004)].