Dominant measures and Sanov's theorem (Q1201365)

Suppose \(L_ n\) denotes the empirical distribution generated by i.i.d. Polish space valued random variables. The main result of this paper, Theorem 2.1, represents \(P(L_ n\in B)\) for open convex subsets \(B\) of probability measures in terms of the integral with respect to a dominant measure \(q_{n,B}\) on the product of the original space. Applications to results more exact than logarithmic equivalence are mentioned.