On the fractal nature of empirical increments (Q1897188)

Let \(W\) be a standard Wiener process, and for \(0 \leq \Lambda \leq 1\) set \[ B^\pm (\Lambda) = \Bigl \{t \in [0,1) : \limsup_{h \downarrow 0} \pm \bigl( 2h \log (1/h) \bigr)^{- 1/2} \bigl( W(t + h) - W(t) \bigr) \geq \Lambda \Bigr\}. \] \textit{S. Orey} and \textit{S. J. Taylor} [Proc. Lond. Math. Soc., III. Ser. 28, 174-192 (1974; Zbl 0292.60128)] have shown that with probability one \(B^\pm (\Lambda)\) is a random fractal with Hausdorff dimension \(1 - \Lambda^2 \). In the present paper the corresponding result is established for the uniform empirical process \(\alpha_n\): For \(0 \leq \Lambda \leq 1\) and a sequence \(h_n\) of positive constants, set \[ E^\pm (\Lambda) = \Bigl \{ t \in [0, \infty) : \limsup_{n \to \infty} \pm \bigl( 2h_n \log (1/h _n) \bigr)^{- 1/2} \bigl( \alpha_n (t + h_n) - \alpha_n(t) \bigr) \geq \Lambda \Bigr\}; \] then, under appropriate growth conditions on \(h_n\), with probability one \(E^\pm (\Lambda)\) is a random fractal with Hausdorff dimension \(1 - \Lambda^2\). Moreover, for any \(\Lambda \in [0,1)\), \(E^\pm (\Lambda)\) is almost surely everywhere dense in \([0,1]\). These results are obtained as consequences of much more general results on the fractal nature of the uniform empirical process and, more generally, of increment processes of certain stochastic processes.