Poisson-Voronoi approximation (Q1024903)

Let \(X\) be a stationary Poisson point process of intensity \(\lambda\). Denote by \(v_X(x)\) the random Voronoi cell of \(x\) with respect to the point set \(X\cup \{x\}\), that is, \(v_X(x) =\{z\in{\mathbb R}^d: \|z-x\|\leq \|z-y\|\,\, \text{{for\, all}}\,\, y\in X\}\). \(x\) is called the nucleus of the Voronoi cell \(v_X(x)\). The set of all Voronoi cells \(v_X(x)\), \(x\in X\) is the Poisson--Voronoi tessellation of \({\mathbb R}^d\). For a given set \(A\subset {\mathbb R}^d\). Let \(v_X(A) = \bigcup\limits_{x\in X\cap A} v_X(x)\) that is the union of all Voronoi cells with nucleus in \(A\). \(v_X(A)\) is called the Poisson-Voronoi approximation of \(A\). In the present paper the quality of this approximation for a convex set \(A\) is considered. In particular, the authors are interested in the convergence of \(v_X(A)\) to \(A\) when the intensity of the Poisson process tends to infinity. The general problem whether \(v_X(A)\) approximates \(A\) for ``complicated'' sets \(A\) seems to be difficult: only partial answers are available see [\textit{E. Khmaladze}, and \textit{N. Toronjadze}, Adv. Appl. Probab. 33, No. 4, 756--764 (2001; Zbl 0993.60011)] and [\textit{M. D. Penrose}, Bernoulli 13, No. 4, 1124--1150 (2007; Zbl 1143.60013)]. The present paper considers the case of a compact convex set \(K\) with nonempty interior, and its approximation \(v_X(K)\). The authors derive precise estimates for the volume \(V(v_X(K))\) and the volume of the symmetric difference of \(K\) and \(v_X(K)\). It is known that for any bounded Borel set \(A\subset {\mathbb R}^d\), one has \(V(A\triangle v_X(A))\to 0\) almost surely as \(\lambda\to \infty\). This was proved for \(d=1\) in [1] and for any bounded Borel set \(A\subset{\mathbb R}^d\) with \(V(A_{\varepsilon})\to V(A)\) for \(\varepsilon\to 0\), where \(A_{\varepsilon}\) is the MInkowski sum of \(A\) in [[3] \textit{J. H. Einmahl}, and \textit{E. V. Khmaladze}, ``The two-sample problem in \({\mathbb R}^m\) and measure-valued martingales. in: State of the Art in Probability and Statistics (Leiden, 1999). IMS Lecture Notes - Monograph Series, 36, 434 -- 463. IMS, Beachwood, OH, (2001)]. The general case was proved in [Zbl 1143.60013]. The main results of the present paper have applications in nonparametric statistics (see [3], Section 3) and image analysis (reconstructing an image from its intersection with a Poisson point process, see [Zbl 0993.60011]).