Gaussian measures in function space

Abstract

Two Gaussian measures are either mutually singular or equivalent. This dichotomy was first discovered by Feldman and Hajek (independently). We give a simple, almost formal, proof of this result, based on the study of a certain pair of functionals of the two measures. In addition we show that two Gaussian measures with zero means and smooth Polya-type covariances (on an interval) are equivalent if and only if the right-hand slopes of the covariances at zero are equal. The H and J functionals* Two probability measures μ0 and μx on a space (Ω, &) are called mutually singular (μ0 _L μd if there is a set ΰe & for which μo(B) — 0 and μλ{Ω — B) = 0. The measures are called mutually equivalent (μ0 ~ μ ±) if they have the same zero sets, i.e., μo(B) = 0 if and only if μ^B) = 0. Setting μ = μ0 + μx we may define the Radon-Nikodym derivatives,