An improvement of Tusnády's inequality in the bulk (Q2665753)

Strict bounds for the distance between Normal\(({m/2},{m/4})\) and Binomial\((m,1/2)\) distributions are provided by Tusnády's inequality which ``lies at the heart of the modern approach to the proof of the Komlós-Major-Tusnády approximation of the empirical process by a sequence of Bownian bridges''. The bounds are in terms of a standard \(N(0,1)\) distributed random variable \(Z\). The main result is a remarkable improvement of previous bounds if the values allowed for \(Z\) are restricted to \(|Z|\leq\sqrt{\log(m)}\) (the ``bulk'' of \(N(0,1)\)), where \(m\) is subject to usually large lower bound. The proof relies among other rather technical evaluations on a refined continuity correction for the Binomial distribution.