Uniform estimation of an unknown probability distribution density (Q1079895)

Let \(X_ 1,...,X_ n\) be independent random variables with common density p(x) and let \(\{\psi_ i(x)\), \(i=1,...\}\) be a complete orthonormal system of functions on \(R_ 1\). Let \(p_{n1}(x)\) be the estimator of the unknown density p(x) constructed in the following way \[ p_{n1}(x)=\sum^{l_ n}_{i=0}(s_ n^{-1}\cdot \sum^{s_ n}_{j=1}\psi_ i(X_ j))\cdot \psi_ i(x),\quad if\quad | x| \leq K_ n, \] and equals 0 otherwise, where \(\{s_ n\}\), \(\{l_ n\}\) and \(\{K_ n\}\) are sequences of positive constants. Let \(p_{n2}(x)=p_{n1}(x)/\int^{\infty}_{-\infty}p_{n1}(y)dy\). Conditions are found under which \[ \sup_{p\in U}P_ p\{\int^{\infty}_{-\infty}| p_{nj}(x)-p(x)|^ idx>\gamma_{nij}\}=O(n^{-q}),\quad n\to \infty,\quad i=1,2,j=1,2, \] for all \(q>0\), where U is a family of known densities and \(\gamma_{nij}\) are constants. Using the same approach, the end of this paper is devoted to the construction of an estimator of the density p(x) of iid rv's \(\xi_ i\) based on the observations \(X=\theta +\xi_ i\), where \(\theta\) is an unknown parameter, p(x) belongs to a known family of densities, and to the derivation of its properties.