Some \(\liminf\) results for two-parameter processes (Q1807269)

Let \(W\) be a standard Brownian sheet on \(\mathbb R^{2}_{+}\), that is, a Gaussian random field \(\{W(s,t)\); \((s,t)\in \mathbb R^{2}_{+}\}\) such that \(\mathbf E W(s,t)=0\) and \(\mathbf E \{W(s,t)W(q,r)\}=(s\land q) (t\land r)\). Let \(b: \mathbb R_{+}\to \left [1,\infty \right [\) be a nondecreasing function, set \(\theta (t) = (\log \log t)^{-1} (2+2\log b(t))\). Assume that \(t\mapsto t^{-1}\theta (t)\) is nonincreasing and \(\theta \equiv \lim_{t\to \infty} \theta (t) <\infty \). Then \[ \liminf_{T\to \infty} T^{-1/2}\theta ^{1/2}(T)\sup \bigl \{W(s,t); st=T, 0\leq s,t\leq T^{1/2}b(T)\bigr\} \] equals almost surely to \(-2\) if \(\theta =0\) and to \(-\theta ^{1/2}\beta (\theta ^{-1})\) if \(0<\theta <\infty \), \(\beta (\nu)\) denoting the largest real zero of the parabolic cylinder function \(D_\nu \). An analogous theorem is proved also for the random field \(|W|\). Proofs employ new precise estimates for the lower tail of a real Ornstein-Uhlenbeck process. Finally, the results for the Brownian sheet are applied to obtain similar \(\liminf\) results for other processes, including the normalized Kiefer process.