Empirical processes and \(p\)-variation (Q4355985)

For any function \(f:R\to R\) and \(p>0\) the \(p\)-variation of \(f\) is defined by \[ v_p(f)=\sup \{\sum_{k=1}^m | f(t_k)-f(t_{k-1})| ^p : -\infty<t_0<t_1<\dots <t_m<\infty,\;m=1,2,\dots\}, \] and the \(p\)-variation norm by \(\| f\| _{[p]}=v_p(f)^{1/p}+\| f\| _\infty\), where \(\| f\| _\infty\) denotes the usual supremum norm. For \(1\leq p\leq 2\) and any distribution function \(F\) on \(R\) with empirical distribution function \(F_n\) it is shown that \[ v_p(n^{1/2}(F_n - F))=O_p(n^{1-p/2}(\log\log n)^{p/2})\;{\text as} n\to\infty, \] which yields \[ \| n^{1/2}(F_n - F)\| _{[p]}= O_p(n^{(2-p)/(2p)}(\log\log n)^{1/2})\;{\text as} n\to\infty. \] \[ \; \] Conversely, \(v_p(n^{1/2}(F_n - F)\geq n^{1-p/2}\) almost surely for all \(n\), if \(F\) is continuous. The upper bound is useful for finding remainder bounds in Fréchet differentiability for functionals of empirical distribution functions. Nearly optimal bounds can be obtained for \((F,G)\mapsto \int FdG\) via Young's inequality and for \((F,G)\mapsto F\circ G^{-1}\) via a bound on the \(L_p\)-norm of the remainder in terms of the \(p\)-variation norm, applied for \(p=2\). As a second application of the \(p\)-variation norm bound on \(n^{1/2}(F_n - F)\), new proofs for the asymptotic distribution of the Cramér - von Mises - Rosenblatt and Watson two-sample statistics are presented when the two sample sizes tend to infinity without any restriction. For the entire collection see [Zbl 0861.00032].