On the supremum of certain families of stochastic processes (Q973167)

Let \((T,d)\) be a complete separable metric space with finite diameter and let \(\left(X_t^{(\varepsilon)}\right)_{t\in T}\) be a family of real-valued, centered, square-integrable stochastic processes on \(T\), indexed by \(\varepsilon>0\). This is the main result of this article: Suppose that \(\lim_{\varepsilon\downarrow 0}E\left(\left(X_{t_0}^{(\varepsilon)}\right)^2\right)=0\) for some \(t_0\in T\) and that \(E\left(\left|X_t^{(\varepsilon)}-X_s^{(\varepsilon)}\right|^\beta\right)\leq B_\varepsilon d(s,t)^{1+\alpha}\) for arbitrary \(s,t\in T\) with some \(\alpha,\beta>0\) and \(B_\varepsilon>0\) satisfying \(\lim_{\varepsilon\downarrow 0}B_\varepsilon=0\). Then, for any \(\delta > 0\), \(\lim_{\varepsilon\downarrow 0}P\left(\sup_{t\in T}\left|X_t^{(\varepsilon)}\right|< \delta\right)=1\), if \((T,d)\) satisfies an additional entropy condition. An application of this result to a stochastic integral with respect to compensated Poisson random measures completes the article.