Moderately large deviations and expansions of large deviations for some functionals of weighted empirical processes (Q688074)

Let \(\alpha_ n\) be the empirical process constructed by independent sampling from uniform (0,1) distribution function, and the functional \(T\) be defined on \(D[0,1]\). Some asymptotic properties of \(P(T(\alpha_ n)>t_ n)\) in the case \(t_ n=O(n^{1/2})\) (large deviations) and in the case \(t_ n=o(n^{1/2})\) (moderately large deviations) are studied. The authors investigate the case of \(T\) satisfying the condition \[ | T(x)-T(y) | \leq \sup \{| x(t)-y(t)|/ \omega(t):0 \leq t \leq 1,\quad x,y \in D[0,1]\}, \] where \(\omega\) satisfies some mild and natural conditions. Explicit bounds for \(P(T(\alpha_ n) \geq x_ n \sqrt n)\) are obtained for all \(x_ n \in(0,\sigma]\) and \(n \geq n_ 0\), where \(\sigma\) and \(n_ 0\) are also explicitly given. These bounds play a major role in the proofs of the moderate large deviations results and expansions of large deviations of the following shape \[ \lim_{x \to+0}x^{-2} \liminf_{n \to \infty}n^{-1} \log P(T(\alpha_ n) \geq x \sqrt n)=\lim_{x \to+0}x^{-2} \limsup_{n \to \infty}n^{-1} \log P(T(\alpha_ n) \geq x \sqrt n)=-{\alpha\over 2}, \] where the positive constant \(\alpha\) is related with the asymptotic behaviour of \(\log P(T(B) \geq y)\) as \(y \to \infty\) and \(B\) is a Brownian bridge.