Almost sure oscillation of certain random processes (Q1815789)

For a real-valued random process \(\{X_t: t\in R\}\) and an appropriate normalizing function \(a(\varepsilon)\), \(\varepsilon\geq 0\), define the process \(Z_\varepsilon (t) = (X_{t+ \varepsilon} -X_t)/a (\varepsilon)\), \(t\geq 0\), and the random measure \(\mu(\varepsilon) = {1\over \lambda(I)} \lambda (\{t\in I: Z_\varepsilon(t) \in B\})\), \(B\) Borel set in \(R\), where \(\lambda\) is the Lebesgue measure. The authors show that for a large family of processes the random measures \(\mu_\varepsilon\) converge weakly as \(\varepsilon\to 0\), for almost all elementary events (almost sure) to a nondegenerate measure \(\mu^*\). The results of this kind are proven for the following classes: (1) a class of Gaussian processes including fractional Brownian motion, stationary processes with certain local behaviour, (2) processes with independent increments and symmetric stable law, (3) continuous martingales satisfying some regularity conditions.