Sample boundedness of stochastic processes under increment conditions (Q916202)

Let (T,d) be a compact metric space and \(\| \cdot \|_{\phi}\) be an Orlicz norm. There are considered separable stochastic processes \(\{X_ t\), \(t\in T\}\) satisfying the increment condition \[ \| X_ t-X_ s\|_{\phi}\quad \leq \quad d(s,t). \] It is proved that sample boundedness of all such processes is equivalent to the existence of a constant \(S=S(T,d,\phi)\), independent of the processes, such that \[ E \sup_{t,s}(X_ t-X_ s)\quad \leq \quad S. \] In the rest of the paper connections between S and the structure of T, expressed in terms of majorizing measures, are investigated systematically. One of the results, which is a generalization of Kôno-Pisier's theorem, is as follows: Let \(\psi\) be the Young function conjugate to \(\phi\) satisfying the \(\Delta_ 2\)-condition (i.e. \(\phi\) grows at \(\infty\) sufficiently fast). Then the existence of a majorizing measure implies the finiteness of S and is equivalent to it if \(\phi\) grows at \(\infty\) faster than \(x^{\alpha \log \log x}\) for some \(\alpha >0.\) For three special cases: the cube \([0,1]^ k\), the unit interval and the product of q-dimensional torus with an ultrametric space, explicit conditions for finiteness of S are evaluated. Some attention is paid to sample continuity, but the results are less complete. This interesting paper gives a considerable progress in the study of general stochastic processes.