Dvoretzky type theorems for subgaussian coordinate projections (Q501834): Difference between revisions

Let \(F\) be a class of functions on a probability space \((\Omega, \mu)\). Let \(X_1, \dots, X_N\) be independent random variables distributed according to \(\mu\) and denote \(\sigma = (X_i)_{i=1}^N\). Consider the set \(P_\sigma F :=\{ (f(X_i))_{i=1}^N : f\in F\}\). Assuming that \(F\) is subgaussian, convex, centrally symmetric and regular (in some sense), the author proves that with high probability there exists a coordinate (in \(\mathbb{R}^N\)) projection of \(P_{\sigma} F\) which can be squeezed between a properly scaled cube and the corresponding ball. The author provides two applications of this theorem. First, he obtains an embedding result for random polytopes in the spirit of \textit{A. E. Litvak} et al. [Adv. Math. 195, No. 2, 491--523 (2005; Zbl 1077.15021)]. Then he proves a Dvoretzky-type theorem in the spirit of \textit{S. Mendelson} and \textit{N. Tomczak-Jaegermann} [Isr. J. Math. 164, 349--364 (2008; Zbl 1183.46010)], where matrices with independent subgaussian isotropic random rows were considered.