Smoothness of distribution function of \({\mathcal F}{\mathcal L}\)-statistic. I (Q753333)

Let \(U_ 1,U_ 2,...,U_ n\) be iid uniform on (0,1) random variables, and let \(h_{jn}(x)\) for \(j=1,2,...,n\) be Borel functions on (0,1). Then an \({\mathcal F}{\mathcal L}\)-statistic is defined as \[ L_ n:=\sum^{n}_{i=1}h_{jn}(U_{j:n})+R_ n, \] where \(U_{j:n}\) are order statistics and \(R_ n\) is nonrandom. Examples show that \(L_ n\) form a general class, which includes L-statistics, and in particular statistics known as Durbing-Nott, Cramér-von Mises, and Anderson- Darling. The objective of the paper is a study of the smoothness of densities for \({\mathcal F}{\mathcal L}\)-statistics. This is done via construction of majorants of their characteristic functions over the complement of the (0,1) interval. The author remarks that this kind of analysis of characteristic functions allows a study of convergence rates for distributions of the statistics, however he does not develop further his remark. The results are only stated in this paper. Proofs are presented in part II of this paper, see the following review, Zbl 0716.62029.