Extremes of Gaussian processes with maximal variance near the boundary points (Q5926522)

The authors deal with a family of Gaussian processes \( \{ X_n(t) : t \in [0,1] \} , u>0 \), with mean zero and nonconstant variance \( \sigma_n^2(t) \). Suppose for each \( u>0 \) there exists a unique point \( t_u \) of maximal variance \( \sigma_n^2(t_u)=1 \) such that \( t_u \rightarrow 0 \) as \( u \rightarrow \infty \). The main result of this paper gives the exact asymptotic behaviour of \( P(\sup_{t \in [0,1]} X_u(t) > u) \sim g(u) , u \rightarrow \infty\). The form of the function \( g(u) \) is given in the case when the covariance function tends ``slower'' to \( 1 \) than the correlation function in the vicinity of \( t_u \) as well as in the converse case. Finally, some applications are mentioned and the simulated probabilities are compared with the asymptotic formula.