Tail behaviour of Gaussian processes with applications to the Brownian pillow. (Q1426355): Difference between revisions

Let \(X\) be a real-valued, separable and zero mean Gaussian process indexed by a subset \(M\) of \(\mathbb R^d\) with \(d\) being greater than or equal to two. Moreover, let \(E\) be the space of real-valued functions defined on \(M\) and let \(T\) be a real-valued functional on \(E\) of the form \(T(f)=\sup_{v\in V}\sqrt{Q_{v}(f,f)}\). Here \(V\) is some index set and \(Q_{v}\) is a symmetric bounded bilinear form on \(E\) for every \(v\in V\). As the functional \(T\) is sublinear and positive homogeneous, it follows for the random variable \(S=T(X)\) that there exists a constant \(a\) such that \(\lim y^{-2} \log P(S>y)= -a/2\) for \(y\rightarrow\infty\). The authors establish methods for the computation of the constant \(a\). Moreover, they construct a random variable \(R\) on the same probability space as \(S\) and with a less intricate distribution such that \(R\leq S\) with probability one and such that \(R\) has the same tail behaviour as \(S\). Special interest is taken in the situation where the space \(M\) and the covariance function of \(X\) both have product structure. This is typical for limiting processes occurring in nonparametric testing of multivariate independency and multivariate constancy over time. An important example is where \(M\) is equal to \([0,1]^2\), and \(X\) coincides with the Wiener pillow.