Large deviations for local empirical measures (Q4524749)

In the last decade the notion of local empirical process has been used to study statistics that are functions of the observations in a suitable neighbourhood of a point. At first Deheuvels and Mason proved functional laws of the iterated logarithm (LIL) for this process and applied them to obtain consistency results for kernel-type density estimators. And then Einmahl and Mason extended the local empirical process within their setup. They established a weak convergence result, a strong invariance principle and a compact LIL for such a process. The author considers the same setup as \textit{U. Einmahl} and \textit{D. M. Mason} [Probab. Theory Relat. Fields 107, No.~3, 283-311 (1997; Zbl 0878.60025) and in: High dimensional probability. Prog. Probab. 43, 75-92 (1998; Zbl 0908.60011)] and is concerned with large deviation principle for local empirical measure \[ \Theta(a_n,\;\cdot):f\in F \mapsto (nv_n)^{-1}\sum_{k=1}^{a_n} f(V_n^{-1}(\xi_k)), \] where \(F\) is a class of measurable \(R\)-valued functions \(f\) defined on a measure space \((\Sigma,S),\) \(\xi_k\) is a sequence of independent \(\Sigma\)-valued random variables with common law \(\mu\) defined on the same probability space \((\Omega, A, P),\) \(V_n:\Sigma \to \Sigma\) is an invertible bi-measurable map and \(v_n=\mu(V_n(\Sigma_0)),\;\Sigma_0 \in S\) (it is assumed that \(v_n>0\) and corresponding probability measure \(\nu_n\) is such that \(\nu_n(B)= \mu(V_n(\Sigma_0\cap B))\)). From this main result (which constitutes a ``non-standard'' counterpoint of Einmahl and Mason and generalizes the result of Deheuvels and Mason) a functional LIL is derived also. For the entire collection see [Zbl 0948.00040].