Regularization of star bodies by random hyperplane cut off

From MaRDI portal





Let \(\Gamma\) be a class of random \(n\)-dimensional vectors, satisfying certain conditions. In particular, \(\Gamma\) contains normalized Gaussian vectors, vectors uniformly distributed on the Euclidean sphere, and vectors uniformly distributed on the vertices of the cube \([-1/\sqrt{n}, 1/\sqrt{n}]^n\). Let \(K\) be a compact star shaped subset of \(\mathbb{R} ^n\), containing the origin in its interior. Denote the \(n\)-dimensional Euclidean ball by \(D\) and the covering numbers by \(N(\cdot , \cdot)\). The main result of the paper is the following Theorem. {There are positive constants \(c\) and \(C\) such that if \(k\leq n\), \(X_1\), \dots, \(X_k\) are independent identically distributed random vectors from \(\Gamma\), and \(N(K, D)\leq e^{ck}\) ,then \[ \bigcap _{i=1}^{k} \left\{ x\in \mathbb{R} ^n\mid (x, X_i) \leq \frac{C}{4\sqrt{k}}\right\} \cap K \subset C \sqrt{\frac{n}{k}} \;D \] with probability larger than \(1-2 e^{-ck}\). } The authors provide many interesting corollaries and applications. In particular, they obtain global versions of such well-known results of local theory as the ``Low \(M^*\)-estimate and the theorem about Euclidean sections of a body with finite volume ratio. These global versions imply, in turn, the corresponding local results, extending them to a larger class of random variables. Another corollary says that using \(2k\) operations, every \(n\)-dimensional compact convex body can be transferred into a body close to the convex hull of the \(n\)-dimensional Euclidean ball and a \((k-1)\)-dimensional simplex. The operations are intersecting with (random) half-spaces of the form \((\cdot , u) \leq C/\sqrt{k}\) and taking convex hulls with (random) points of Euclidean length \(\sqrt{k}\).











This page was built for publication: Regularization of star bodies by random hyperplane cut off

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4811112)