Recurrence relationships for the mean number of faces and vertices for random convex hulls (Q848666): Difference between revisions

Consider the experiment where points \(P_1\), \(P_2\),\dots, \(P_n\) are placed randomly and independently in \({\mathbb R}^d\), \(d\geq1\) according to a common probability law. This law is defined by the (induced) probability measure \(\mu\) defined on \({\mathcal B}_d\), the Borel sets of \({\mathbb R}^d\). Let the convex hull of the \(n\) points be denoted by \({\mathbf H}_n\) (\({\mathbf H}_n\) is a convex polytope), while \(\mu_n=\mu({\mathbf H}_n)\) be the content (that is, \(\mu\)-measure) of \({\mathbf H}_n\). Denote by \(F_{n,i}\) the number of \(i\)-faces of \({\mathbf H}_n\). Combining the power of three distinct ideas -- the Efron-Buchta identity [\textit{C. Buchta}, Discrete Comput. Geom. 33, No.~1, 125--142 (2005; Zbl 1065.52003)], an integral technique of Rényi-Sulanke [\textit{R. Schneider} and \textit{W. Weil}, Stochastic and Integral Geometry. Probability and its Applications. Berlin: Springer (2008; Zbl 1175.60003)] and an identity involving \(\mu_n\) that follows from recent work of the author [\textit{R. Cowan}, Adv. Appl. Probab. 39, No.~3, 630--644 (2007; Zbl 1127.60011)] -- the author derives various new partial recurrence identities for \({\mathbf E}[F_{n,i}]\) for \(i=0,d-1, d-2\) (\({\mathbf E}\) stands for the expectation). Mostly the author focus on the case where \(\mu\) gives zero probability mass to every \(j\)-dimensional flat (for all \(j< d\); a \((d-1)\)-flat is called a hyperplane) and, in particular, to any one-point set. This implies that the polytope \({\mathbf H}_n\) is almost surely simplicial with dimension equal to \(\min(d, n-1)\). This constraint on \(\mu\) is called the hyperplane condition. It is weaker than absolute continuity with respect to Lebesgue measure. A general recurrence identity, in the case where \(\mu\) satisfies the hyperplane condition, for any \(0\leq i< d\) is conjectured. Note that the author have proved the identity for all \(d\) when \(i\) equals \(0,d-1, d-2\), and additionally for \(i=1\) when \(d=4,5\) and \(i=2\) when \(d=5\). The paper concludes with a brief discussion of the difficulties that arise when \(\mu\) does not satisfy the hyperplane condition. The discussion suggests that this condition is the most appropriate for the current study.