scientific article; zbMATH DE number 16088
[For the entire collection see Zbl 0716.00011.] Consider a separable Banach space \(B\), and let \(\mu\) be a mean zero Gaussian measure on \(B\). Denote by \(H_ \mu\subset B\) the unique Hilbert space such that \(\mu\) is determined by considering the pair \((B,H_ \mu)\) as an abstract Wiener space. Assume that \(H_ \mu\) is infinite- dimensional. If \(X\) is a \(B\)-valued random vector with law \(\mu={\mathcal L}(X)\), then large deviation results imply that \[ \lim\sup_{\varepsilon\downarrow0}\varepsilon^ 2\log P(\varepsilon X\in A)\leq-\inf_{x\in \bar A}I(x), \leqno (1) \] where \(\bar A\) denotes the \(B\)-closure of \(A\) and \(I(\cdot)\) is the rate function \(I(x)=(1/2)\| x\|^ 2_{H_ \mu}\), \(x\in H_ \mu\); \(=+\infty\), else. The authors establish a probability estimate which can be viewed as a refinement of the inequality (1). The latter estimate is useful in proving upper bounds for the rate of convergence in functional laws of the iterated logarithm (LIL), but it is also of independent interest in different settings. Some applications to investigating integrability properties of certain random variables related to the LIL are presented, too.
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3972736)
