Continuity of Gaussian processes (Q1110186): Difference between revisions

The author first gives a generalization of \textit{M. B. Marcus} and \textit{L. A. Shepp}'s [Proc. Sixth Berkeley Sympos. math. Statist. Probab., Univ. Calif. 1970, 2, 423-441 (1972; Zbl 0379.60040)] theorem on the equivalence between sample continuity of a Gaussian process defined on a compact subset of a metric space, and a.s. continuity of the process at each subset point. He then studies the continuity of the isonormal process G restricted in a subset C of a real infinite-dimensional Hilbert space, satisfying \[ \sup \{| \| x\| -\| y\| | /\| x-y\|^ 2| \quad x,y\in C,\quad x\neq y\}<\infty. \] He shows that the convergence of an integral expressed in terms of local metric entropy is sufficient for local a.s. continuity. The proof is based on an unpublished upper bound of \[ P(\sup \{G(x);\quad x\in C\}>\ell_ 0s+\sum^{\infty}_{j=1}\ell_ je_ j), \] where \(s:=\sup \{\| x\|\); \(x\in C\}\), \(e_ j\downarrow 0\) and the positive sequence \((\ell_ j)\) satisfies some regularity conditions. The author concludes his work by showing that his local condition is satisfied in two examples given by \textit{R. M. Dudley} [Ann. Probab. 1, 66-103 (1973; Zbl 0261.60033)] and where the well known integral of the global metric entropy diverges. Hence this new local condition is weaker.