On Berry-Esseen theorem for some functions of \(U\)-statistics (Q1194005)

Consider the class \(G\) of all real differentiable functions \(g\) such that \[ | g'(x)-g'(y)|\leq L| x-y|^ \delta, \qquad 0<\delta\leq 1, \] \(L\geq 0\) being a positive constant. Let \(U_ n\) be a \(U\)-statistic based on \(n\) i.i.d. observations \(X_ 1,X_ 2,\dots,X_ n\) such that \[ U_ n={n\choose 2}^{-1} \sum^ n_{{i,j=1} \atop {i\neq j}} h(X_ i,X_ j), \] \(L\geq 0\) being the symmetric kernel function with \({\mathcal E}h(X_ 1,X_ 2)=\theta\). Under \({\mathcal E}| h(X_ 1,X_ 2)|^{2+\gamma}<\infty\), \(0<\gamma\leq 1\), the rate of convergence of the d.f. of \(W_ n:=\sigma(U_ n)^{-1} {{g(U_ n)- g(\theta)} \over {g'(\theta)}}\) to the standard normal d.f. has been given. A similar result concerning \(W_{N_ n}\) where \(\{N_ n\}\) is a sequence of positive integer-valued r.v.'s which are independent of \(X\)'s and have finite (constant) mean and variance has been given. Rates of such convergence in respect of \(W_ n'\), where \(W_ n'\) is as \(W_ n\) with \(\sigma(U_ n)\) replaced by \(\sigma(\hat U_ n)\), \(\hat U_ n\) being the projection of \(U_ n\) are also proved.