Radial positive definite functions generated by Euclid's hat (Q1290939): Difference between revisions

Let \(\Phi_n\) be the class of all radial (depending only on the distance from the origin) characteristic functions in \({\mathbb R}^n\). Let \(H_n\) be the class of all scale mixtures of Euclid's hat function, the latter being defined as the self-convolution of the indicator function of the unit ball in \({\mathbb R}^n\) scaled to equal one at the origin. Then, clearly \(H_n\subset \Phi_n\) for all \(n\geq 1\). The author shows that the inclusion holds: \(H_1\supset H_2\supset H_3\dots,\) and gives a complete characterization of the classes \(H_n\) and \(H_{\infty}=\bigcap_{n\geq 1} H_n\). This characterization is particularly nice for odd \(n\) and for \(n=\infty.\) In particular, a function \(\varphi(\|x\|)\) belongs to \(H_{\infty}\) if and only if the function \(-\varphi'(\sqrt{t})\) is completely monotone. These results lead to the following analogue of the famous Pólya's criterion. If a continuous function \(\varphi: [0,\infty)\to{\mathbb R}\) is such that \(\varphi(0)=1, \varphi(t)\to 0\) as \(t\to\infty\), and the function \((-1)^k\frac{d^k}{dt^k}[-\varphi'(\sqrt t)]\) is convex when \(k\) is equal to the greatest integer less than or equal to \((n-3)/3,\) then \(\varphi(\|x\|)\) is a characteristic function in \({\mathbb R}^n.\) The author also presents examples showing that \(H_n\) is a proper subclass of \(\Phi_n,\) for all \(n\), and that there is a function which belongs to all \(\Phi_n\) but does not belong to any of the classes \(H_n, n\geq 2.\) Relations are discussed between the classes \(H_n\) and the classes studied earlier in the works of Askey, Mittal and Berman. Finally, the author explains why his results can be useful in geostatistics.