Empirical U-statistics processes (Q1200626)

Let \(\xi_ 1,\xi_ 2,\dots\) be independent and identically distributed random elements defined on a probability space \((\Omega,{\mathcal E},P)\) with values in a measurable space \(X_ 2\) with \(\sigma\)-field \(\mathcal A\) and let \(h: X^ q\to R^ r\) be an \(({\mathcal A}^ q,{\mathcal B}^ r)\)- measurable mapping \((q,r\in N)\) (\(h\) being symmetric in its \(q\)- arguments). Let \(J_ n(q)\) be the set of all subsets of \(q\)-different elements from \(\{1,2,\dots,n\}\) so that \(\# J_ n(q)={n\choose q}\). Let \(X_ I=h(\xi_{i_ 1},\dots,\xi_{i_ q})\) for \(I=\{i_ 1,\dots,i_ q\}\in J_ n(q)\) and define \[ \hat H_ n(t)={n\choose q}^{- 1}\sum_{I\in J_ n(q)}I_{(-\infty,t]}(X_ I),\qquad H(t)=E I_{(- \infty,t]}(X_ I), \] for \(t\in R^ r\) where \((-\infty,t]=(-\infty,t_ 1]\times\dots\times(-\infty,t_ r]\) and \(t=(t_ 1,\dots,t_ r)\in R^ r\). The empirical \(U\)-statistics process indexed by multidimensional time is defined by \[ U_ n(t)=n^{1/2}(\hat H_ n(t)-H(t)),\;t\in R^ r. \] Let \(U_ n\{s,t\}\) be the increments of the process. The authors obtain a sharp exponential probability bound for \(U_ n\{s,t\}\) and using this inequality obtain the exponential bound on the fluctuations of the weighted \(U\)-statistics and a weak convergence theorem of the weighted process. The final result generalizes a result of \textit{B. W. Silverman} in Ann. Probab. 11, 745-751 (1983; Zbl 0514.60040).