On pointwise estimates of positive definite functions with given support

Abstract

The following problem originated from a question due to Paul Turan. Suppose $\Omega$ is a convex body in Euclidean space $\RR^d$ or in $\TT^d$, which is symmetric about the origin. Over all positive definite functions supported in $\Omega$, and with normalized value 1 at the origin, what is the largest possible value of their integral? From this Arestov, Berdysheva and Berens arrived to pose the analogous pointwise extremal problem for intervals in $\RR$. That is, under the same conditions and normalizations, and for any particular point $z\in\Omega$, the supremum of possible function values at $z$ is to be found. However, it turns out that the problem for the real line has already been solved by Boas and Kac, who gave several proofs and also mentioned possible extensions to $\RR^d$ and non-convex domains as well. We present another approach to the problem, giving the solution in $\RR^d$ and for several cases in $\TT^d$. In fact, we elaborate on the fact that the problem is essentially one-dimensional, and investigate non-convex open domains as well. We show that the extremal problems are equivalent to more familiar ones over trigonometric polynomials, and thus find the extremal values for a few cases. An analysis of the relation of the problem for the space $\RR^d$ to that for the torus $\TT^d$ is given, showing that the former case is just the limiting case of the latter. Thus the hiearachy of difficulty is established, so that trigonometric polynomial extremal problems gain recognition again.