A new large deviation inequality for U-statistics of order 2

From MaRDI portal
Publication:4941309





The second-order \(U\)-statistics \(U_n(f)=\sum_{1\leq i \neq j \leq n} f(\xi_i,\xi_j)\) are considered. Let \(U_n(f-E f)=2(n-1)U_n^{(1)}+U_n^{(2)}\) be the decomposition of \(U_n\) to the linear and second-order canonical projections. The upper bound for the moments \(E (U_n{(2)})^{2k}\) is proved. As a corollary the exponential inequality \[ P(U_n^{(2)}\geq C(f,F,n)x)\leq \exp (6-2\sqrt{x}) \] is derived. This inequality is compared with similar inequalities of \textit{M. A. Arcones} and \textit{E. Giné} [Ann. Probab. 21, No. 3, 1494-1542 (1993; Zbl 0789.60031)].











This page was built for publication: A new large deviation inequality for U-statistics of order 2

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4941309)