A note on Cramer's theorem

From MaRDI portal





Let \(\{\xi_n, n\geq 1\}\) be a sequence of i.i.d. r.v.'s taking values in a locally convex Polish vector space \((X,\rho)\) and let \(L_n= \sum^n_{k= 1}\xi_k/n\). The sequence \(\{\mu_n\}\) of Borel probability measures in \(X\) is said to satisfy the large deviation principle (LDP), if (i) there exists a functional \(I: X\to[0, \infty]\) such that \(\{x: I(x)\leq a\}\) is compact for any \(0\leq a< \infty\); (ii) for each Borel set \(A\subseteq X\), \[ -\inf_{x\in\text{int }A} I(x)\leq \lim_{n\to \infty} {\inf\brack\sup} {1\over n}\log\mu_n[A]\leq -\inf_{x\in\text{cl }A} I(x). \] The aim of the note is to prove the following theorem: \(\{P[L_n\in\cdot], n\geq 1\}\) satisfies the LDP as \(n\to\infty\) iff there is a compact convex balanced set \(K\subset X\) such that \(E\exp q_K(\xi_1)< \infty\), where \(q_K(x)= \inf\{\lambda>0: x/\lambda\in K\}\) is the Minkowski functional of \(K\). In this case, the rate function \(I\) is given by \[ I(x)= \sup_{y\in X^*} [\langle x,y\rangle- \log E\exp\{\langle \xi_1,y\rangle\}],\qquad x\in X. \] For the entire collection see [Zbl 0864.00069].











This page was built for publication: A note on Cramer's theorem

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4349228)