The supremum of a Gaussian process over a random interval (Q1771422)

The authors investigate the exact asymptotic behavior of the supremun of a Gaussian process \(P(\sup_{s\in[0,T]} X(s) > u)\) as \(u \rightarrow \infty\), where \(\{ X(t) : t \geq 0 \}\) is a centered Gaussian process with stationary increments and \(T\) is an independent nonnegative random variable with regularly varying tail distribution. In the classical problem in the extreme-value theory of Gaussian processes, \(T\) is a fixed number in general. In this paper the authors study the case where \(T\) is a random variable, independent of \(X(s)\). They show that the qualitative tail behavior of the supremum may crucially depend on the variability of \(T.\) The authors consider the case \(X(s)=B_{\alpha}(s)\), where \(B_{\alpha}(s)\) is a fractional Brownian motion with Hurst parameter \(\alpha \in (0, 1]\) and they get the following result: Theorem. Let \(T\) be a nonnegative random variable with regularly varying tail distribution at \(\infty\) with index \(\nu > 0.\) Then \[ \begin{aligned} P \left( \sup_{s\in [0,T]} B_{\alpha}(s)>u \right) &= P\left( T^{\alpha} \sup_{s\in [0,1]} B_{\alpha}(s)> u \right)\\ &\sim \mathbf{E}\left( \sup_{s\in [0,1]}B_{\alpha}(s) \right)^{\nu/\alpha} P(T>u^{1/\alpha}) \quad \text{as } u \rightarrow \infty.\end{aligned} \] The main object of this paper is to show that the asymptotics in the proposition essentially remain the same for a much larger class of Gaussian processes than FBM, which is the main result.