On the asymptotic behavior of weighted \(U\)-statistics (Q5918052)

Let \(X_n,\;n\geq 1,\) be independent and identically distributed random variables. The \(\mathcal U\)-statistics under consideration are given by an expression of the type \[ {\mathcal U}_n =\sum_{1\leq i_1 < \ldots < i_k\leq n} a(i_1, \ldots , i_k) f(X_{i_1},\ldots , X_{i_k}) , \] where \(a : N^k\rightarrow R\) is a bounded symmetric function and \(f : R^k\rightarrow R\) is a measurable and symmetric function such that \[ E f(X_1, \ldots , X_k) = 0, \quad E f^2(X_1, \ldots , X_k) < \infty . \] It is also assumed that \({\mathcal U}_n\) is degenerate, i.e.: \[ E f(x_1, \ldots , x_{k-1}, X_k) = 0, \;\;\forall \;x_1,\ldots\;, x_{k-1} \in R . \] The main theorem states that under some sufficient condition \(n^{-k/2} {\mathcal U}_n \) converges in distribution to some multiple Wiener-Ito integral with respect to a two-parameter Wiener process. A relationship between this sufficient condition and the condition of \textit{K. A. O'Neil} and \textit{R. A. Redner} [Ann. Probab. 21, No.~2, 1159-1169 (1993; Zbl 0774.62018)] for \(k=2\) is presented. The theorems of \textit{P. Major} [ibid. 22, No.~3, 1514-1535 (1994; Zbl 0813.62014)] are derived for \(k\geq 2 \).