A generalization of the Chung-Mogul'skii law of the iterated logarithm (Q808519)

Consider positive integers m and n, and a sequence \(X_ i\) of independent real random variables with a common continuous distribution function. Let \(H_ n(x)\) be \(\left( \begin{matrix} n\\ m\end{matrix} \right)^{- 1}\sum I[\max \{X_{i(1)},X_{i(2)},...,X_{i(m)}\}\leq x],\) where the sum is taken over all integers \(1\leq i(1)\leq...\leq i(m)\leq n\) and I[A] is the indicator function of A. The empirical process of U-statistic structure is defined by \(D_ n=n^{{1\over 2} }(H_ n-E H_ n)\). This paper extends Chung's ``other'' law of the iterated logarithm to such processes. With probability one, \[ \liminf_{n\to \infty}\sup_{x\in R}(\log \log n)^{{1\over 2} }| D_ n(x)| =\pi /8^{{1\over 2} }. \] This result is proved by means of a small deviation result for Brownian bridges, a lemma of \textit{K. B. Erickson} [Ann. Probab. 8, 325- 338 (1980; Zbl 0429.60034)] and the strong approximation theorem of \textit{H. Dehling}, \textit{M. Denker} and \textit{W. Philipp} [Ann. Inst. Henri Poincaré, Probab. Stat. 23, 121-134 (1987; Zbl 0619.60036)] for empirical processes of U-statistic structure.