Fractal dimensions for some increments of the uniform empirical process (Q1868444)

Let \(\alpha_n\) be the uniform empirical process of sample size \(n\), and for \(0\leq s,t\leq 1\) and \(0<h<1\) set \(\Delta\alpha_n(t,h,s)= \alpha_n(t+hs)-\alpha_n(t)\). The increment function \(\Delta\alpha_n\) is used to define certain sets of exceptional oscillation points of \(\alpha_n\). The precise definition is technically too involved to be repeated here. For `large' increments given by \(h=h_n\downarrow 0\) with \(nh_n\uparrow\infty\) and \((\log(1/h_n))/\log\log n\to d\in[1,\infty)\) and for `small' increments given by \(h=h_n\downarrow 0\) with \(nh_n\uparrow\infty\) and \((nh_n)/\log n\downarrow 0\) it is shown that these sets are random fractals. In the small increment case the exact Hausdorff dimension is calculated, and in the large increment case upper and lower bounds for the Hausdorff dimension are derived.