Entropy of convex functions on \(\mathbb R^d\) (Q1691434)

Motivated by both theoretical reasons and concrete applications, the authors provide sharp estimates for the \(\varepsilon\)-entropy of the class of convex functions with \(L^r\)-norm bounded by 1 defined on a bounded solid closed convex subset of \(\mathbb{R}^d\) with respect to \(L^p\)-metrics (where \(1\leq p < r\)). Upper bounds for \(d\)-polytopes are derived then as a general scheme of constructing simplicial approximations of convex sets, having as a rather surprising byproduct the observation that while a general convex body can be approximated by inscribed polytopes, the metric entropy rate heavily depends on the shape of the considered convex domain, increasing thus the challenging degree of the considered problem. Applications concerning rates of convergence of nonparametric estimators of high-dimensional shape-constrained functions are also mentioned.