On large deviations of Gaussian measures in Banach spaces

From MaRDI portal





The paper studies the limiting behavior of \(t^{-2}\log \gamma(tA)\) for large \(t\), where \(\gamma\) is a full centered Gaussian measure on a separable Banach space and \(A\) is an open set. The upper and lower bounds are provided by known large deviations results and are equal to \(- \inf_{x \in \text{Int} (A)} I(x)\) and \(-\inf_{x \in \overline A}I(x)\), where \(I(x)={1\over 2} | x |_ \gamma\) for \(x\) in the reproducing kernel Hilbert space of \(\gamma\) and \(\infty\) elsewhere. The main result shows that one can construct such a Gaussian measure \(\gamma\) and an open set \(A\) that \(-\inf_{x \in \text{Int} (A)} I(x)\), \(\liminf_{t \to \infty} t^{-2} \log \gamma(tA)\), \(\limsup f_{t \to \infty} t^{-2} \log \gamma (tA)\) and \(-\inf_{x \in \overline A} I(x)\) are all different. On the other hand it is shown that for a large class of open sets the limit exists. Analogous results are obtained for probabilities of moderate deviations, i.e. for \((n/b_ n) \log P[(S_ n/b_ n) \in A]\), where \(n \to \infty\) and \(S_ n\) is a sequence of partial sums for i.i.d. random vectors. For the entire collection see [Zbl 0773.00018].











This page was built for publication: On large deviations of Gaussian measures in Banach spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3139494)