Berry-Esseen bounds for generalized \(U\)-statistics (Q2084842)

Consider the generalized \(U\)-statistic defined by \[ S_{n,k}(f) = \sum_{1\leq i_1 < i_2 < \dots < i_k \leq n} f\left(X_{i_1}, \dots , X_{i_k}; Y_{i_1,i_2},\dots , Y_{i_{k-1},i_k}\right) \] where \(\{X_{i}\}_{i=1}^n\) and \(\{Y_{i,j}\}_{1\leq i < j\leq n}\) are two independent sequences of i.i.d. random variables, and \(f\) is a symmetric function. In this paper the author establishes optimal Berry-Esseen bounds for the generalized \(U\)-statistic. The proof uses the exchangeable pair approach of Stein's method. The results are applied to obtain an optimal convergence rate of the normal approximation for subgraph counts in Erdős-Rényi graphs and graphon-random graph. The asymptotic distribution of \(S_{n,k}(f)\) had been previously studied in [\textit{S. Janson} and \textit{K. Nowicki}, Probab. Theory Relat. Fields 90, No. 3, 341--375 (1991; Zbl 0734.60036]). Related bounds obtained by a different approach can be found in [\textit{R. Lachièze-Rey } and \textit{G. Peccati}, Ann. Appl. Probab. 27, No. 4, 1992--2031 (2017; Zbl 1374.60023)].