Randomizing properties of convex high-dimensional bodies and some geometric inequalities (Q1608698)

The main result of this announcement is the following: Let \(D\) be the \(n\)-dimensional standard Euclidean ball. Let \(K\) be a convex symmetric body in \(\mathbb{R}^n.\) Let \(\|.\|_K\) be the associated norm. Let \(T_i \subset \mathbb{R}^n, 1\leq i\leq m\), be measurable sets. Suppose that \(K\) and \(T_i\) have the same volume as \(D\). Then for all scalars \(\lambda_i\) and any positive \(t\) we have \[ \text{Prob}\Biggl\{ (x_i)_{i=1}^m \in \prod^m_{i=1} T_i \biggl|\biggl\|\sum \lambda_i x_i \biggr\|_K < t \Biggr\} \] \[ \leq \text{Prob}\Bigl\{ (x_i)_{i=1}^m \in D^m \Bigl|\Bigl\|\sum \lambda_i x_i \Bigr\|_D < t \Bigr\} \leq \left(t \sqrt{\frac{e}{\sum \lambda _i^2}} \right)^n \] when the points \(x_i\) are chosen independently according to the uniform distribution over the sets \(T_i\) and \(D\), respectively. The result shows in particular that the random sums have stochastically larger Euclidean norm than for any other `normalized' norm. The basic tool is the Brascamp-Lieb inequality [\textit{H. J. Brascamp} and \textit{E. H. Lieb}, Adv. Math. 20, 151-173 (1976; Zbl 0339.26020)]. As a consequence, one gets that every finite dimensional normed space has `random cotype' 2.