Some asymptotic formulae for Gaussian distributions (Q1914691)

Asymptotic expansions (in powers of \(r^{-1}\) and \(\tau\)) for the expectation \[ Ef(r\varphi[\xi;r^{-1},\tau],\tau)g(\xi;r^{-1},\tau) \] are derived. Here \(\xi\) is a centered Gaussian vector with values in a separable real Hilbert space \(X\), functions \(g\) and \(\varphi:X\times [0,1]^2\to\mathbb{R}^d\), \(d\geq1\), are smooth in all of their arguments, function \(f(y,\tau)\), \(y\in\mathbb{R}^d\), is smooth in its argument \(\tau\) and satisfies some integrability condition. An asymptotic expansion is typically \(\sum^J_{j=0} \sum^K_{k=0} m_{jk}{\tau^k\over r^j}+\rho_{JK}(r,\tau),\) \(\rho_{JK}\) is a remainder term. Applications to large deviation problems are given. In particular, the calculation of probability \(P\{\Phi(r^{-1}\xi)\}>0\) for a sufficiently regular functional \(\Phi\) is presented. A specific perturbation of an extremal point of Gaussian deviation functional which arises if a sum of i.i.d. random vectors replaces Gaussian vector \(\xi\) is also considered.