Random polytopes and the wet part for arbitrary probability distributions (Q2211500)

Let $\mu$ be a general probability measure defined on the Borel sets of $\mathbb{R}^d$ and let $X_n$ be a sample of $n$ independent random points chosen according to $\mu$. A random $\mu$-polytope is defined as $P_n^{\mu} = \operatorname{conv}X_n$. Furthermore, the authors generalise the notion of the wet part of a convex polytope to this general setting by setting \[ W_t^{\mu} = \{ x \in \mathbb{R}^d : \text{ there is a halfspace $h$ with $x \in h$ and $\mu(h) \leq t$} \}. \] The $\mu$-measure of the wet part is denoted by $\omega^{\mu}(t) = \mu(W_t^{\mu})$. It was shown in earlier work [the first author and \textit{D. G. Larman}, Mathematika 35, No. 2, 274--291 (1988; Zbl 0674.52003)] that the measure of the wet part of a convex body captures how well a random polytope approximates a convex body. The authors extend this result to general measures in Theorem 1.2 and prove that for any probability measure $\mu$ in $\mathbb{R}^d$ and $n\geq 2$ \[ \frac{1}{4} \omega^{\mu}\left( \frac{1}{n} \right) \leq \mathbb{E} (1 - \mu(P_n^{\mu}) ) \leq \omega^{\mu} \left((d+2) \frac{\ln n}{n} \right ) + \frac{\varepsilon_d(n)}{n}, \] in which $\varepsilon_d(n) \rightarrow 0$ as $n \rightarrow \infty$ and is independent of $\mu$. The main result of the paper (Theorem 1.3) shows that the $\ln n /n$ factor can not be dropped in the upper bound for $d=2$ which is in contrast to the original result for convex bodies. The main difference between the wet parts of convex bodies and general measures is discussed in Section 2. Finally, the authors use Theorem 1.2 and Theorem 1.3 as well as Efron's formula to derive a related result for the number of vertices of $P_n^{\mu}$ in Theorem 1.4.