Dominating points for general sequences of random vectors (Q4524748)

Let \(E\) be a topological vector space, \(E^*\) its dual space. Let \(\{\mu_n\}\) be a sequence of probability measures on \(E\). Let \[ \varphi_n(\xi)=\log \int\exp\{\langle\xi,x\rangle\}\mu_n(dx),\qquad \xi\in E^*. \] Then if \(\varphi(\xi)=\lim_{n\to \infty}\frac{1}{n}\varphi_n(n\xi)\) exists for all \(\xi\) and satisfies a sutable differentiability condition, and \(\{\mu_n\}\) is exponentially tight, then \(\{\mu_n\}\) satisfies the large deviations principle with rate function \[ \varphi^*(x)=\sup_{\xi\in E^*}\{\langle\xi,x \rangle- \varphi(\xi)\}, \qquad x\in E. \] Definition. Let \(D\) be an open convex subset of \(E\) such that \(D\cap \text{dom } \varphi^*\not = \emptyset\). A point \(x_0\in D\) is a dominating point for \((D,\varphi)\) if (1) \(x_0\in \partial D\), (2) \(\varphi^*(x_0)=\inf_{x\in D}\varphi^*(x)\), (3) there exists \(\xi_0\in E^*\) such that \(D\subset \{x: \langle\xi_0,x\rangle>\langle\xi_0,x_0\rangle\}\) and \(\varphi^*(x_0)=\langle\xi_0,x \rangle- \varphi(\xi)\). The objectives of the present paper are to prove the existence and uniqueness of a dominating point for \((D,\varphi)\) and to give an initial application to non-logarithmic asymptotic bounds for \(\mu_n(D)\). For the entire collection see [Zbl 0948.00040].