Minima of \(H\)-valued Gaussian processes (Q2563933)

Let \(T\) be a separable topological space and let \((H, \langle\cdot|\cdot\rangle, |\cdot|)\) be a real separable Hilbert space with bounded linear operators \({\mathcal L}= {\mathcal L}(H)\). Let \(\{X(t), t\in T\}\) be a centered separable \(P\)-continuous \(H\)-valued Gaussian random field, with respect to a complete probability \(P\), such that \(X(t)\) has variance \(R\in{\mathcal L}\) independent of \(t\). Also let \(R\) have spectrum (eigenvalues) \(\{\lambda_k, k= 1,2,\dots\}\subseteq(0,\infty)\) and choose a complete orthonormal system \(\{e_k, k= 1,2,\dots\}\) satisfying \(Re_k=\lambda_ke_k\). The author studies the asymptotic behavior of the tail \(P\{|X(s_0)|^2<\varepsilon\}\) as \(\varepsilon\downarrow 0\) for a fixed \(s_0\in T\) and derives upper and lower bounds for \(P\{\inf_{t\in S}|X(t)|^2<\varepsilon\}\) as \(\varepsilon\downarrow 0\), for \(S\subseteq T\). Several applications of these results are given. Next, the treatment is specialized to the process \(Y(t)= \sum^\infty_{k=1} \sqrt{\lambda_k}\xi_k(t)e_k\), where \(\{\xi_k(t)\}\), \(k=1,2,\dots\), are independent (real-valued) stationary and standardized Gaussian processes with covariance functions \(\{r_k(t)\}\). The author finds the exact asymptotic behavior of \(P\{\inf_{t\in[0,1]}|Y(t)|^2<\varepsilon\}\) under conditions on \(r_k(t)\) and proves global limits for \(Y\) when \(r_k(t)\to 0\) not too slowly as \(t\to\infty\).