Law of the iterated logarithm for U-statistics (Q583710): Difference between revisions

The note contains a law of iterated logarithm for the well-known U- statistics \[ U_ n=\left( \begin{matrix} n\\ m\end{matrix} \right)^{- 1}\sum_{1\leq i_ 1<...<i_ m\leq n}\Phi (X_{i_ 1},...,X_{i_ m}). \] The proof is based on a fundamental decomposition by \textit{W. Hoeffding} [The strong law of large numbers for U-statistics (1961)]: \[ U_ n-E \Phi (X_ 1,...,X_ m)=\sum^{m}_{c=1}\left( \begin{matrix} m\\ c\end{matrix} \right)U_{nc}, \] where \(U_{nc}\) is a U-statistic with \[ U_{nc}=\left( \begin{matrix} n\\ c\end{matrix} \right)^{-1}\sum g_ c(X_{i_ 1},...,X_{i_ c}) \] and \(g_ c\) is an appropriate kernel [see also \textit{V. S. Korolyuk} and \textit{Yu. V. Borovskikh}, Theory of U- statistics (Russ.) (1989)].