Spherical distributions: Schoenberg (1938) revisited (Q2573227): Difference between revisions

The aim of this article is to provide an alternative proof of a fundamental result of \textit{I.J. Schoenberg} [Metric spaces and completely monotone functions. Ann. Math. 39, No. 4, 811--841 (1938)] on the representation of characteristic functions for spherical distributions. An \(m\)-dimensional random vector \(X\) is said to have a spherical distribution if and only if its characteristic function is of the form \(\phi(\| t\|)\), where \(t\in\mathbb R^m\) and \(\|\cdot\|\) denotes the usual Euclidian norm. Informally, this means that the probability density function of \(X\) is constant on real spheres. Let \(\Phi_m\) be the class of characteristic functions \(\phi\) defined above, and let \(\Phi_\infty\) be the class of characteristic functions \(\phi\in\Phi_m\), for all \(m\). Notice that \(\Phi_1\supset\Phi_2\supset\dots\supset\Phi_\infty\). Schoenberg proved the following important result: A characteristic function \(\phi\in\Phi_m\) belongs to \(\Phi_\infty\) if and only if the corresponding spherical distribution \(X\) can be represented as a mixture of normal distributions. \textit{J.F.C. Kingman} [On random sequences with spherical symmetry. Biometrika 59, 492--494 (1972; Zbl 0238.60025)] gave another proof of this result in the context of exchangeability, with a slightly adapted version of \textit{K.-T. Fang, S. Kotz} and \textit{K.-W. Ng} [Symmetric multivariate and related distributions. (1990; Zbl 0699.62048)]. The authors of the present article provide an alternative proof of the basic result of Schoenberg which is less complicated and more elegant, using probabilistic tools to establish a certain global uniform convergence argument.