Variational inequalities with examples and an application to the central limit theorem (Q1345613)

Let \(X\) be a r.v. with distribution function (d.f.) \(F_X\) and let \[ \rho (F_X, \Phi) = \sup_A \bigl |F_X (A) - \Phi (A) \bigr | \] denote the usual distance in variation between d.f.'s \(F_X\) and \(\Phi\), where the supremum is taken over the class of Borel sets \(A\), and \(\Phi\) is the standard normal d.f. If \(X\) has density \(f\) with interval support, mean \(\mu\) and variance \(\sigma^2\), then the equality \[ \sigma^2 w(x) f(x) = \int^x_{- \infty} (\mu - t) f(t)dt, \quad x \in R, \] defines the corresponding kernel \(w\) associated with \(f\). The main Theorem 1.1 states that for a r.v. \(X\) with \(EX = 0\), \(\text{Var} X = \sigma^2\) and an absolutely continuous d.f. \(F_X\) with an interval support, \[ \rho (F_X, \Phi) \leq 2 \biggl( E \bigl |w(X) - 1 \bigr |+ |1 - \sigma^2 |\biggr). \] More generally (Theorem 1.3), if \(X\) is a r.v. with \(EX = 0\) and \(\text{Var} X = 1\), then \[ \rho (F_X, \Phi) \leq \bigl( 2 + \sqrt {\pi/2} \bigr) \sqrt {1 - J_X}, \] where \(J_X = \inf_{g \in H_1} (\text{Var} g(X)/E^2 g'(X))\), and \(H_1\) is the class of absolutely continuous functions \(g\) such that \(0 < E[g'(X)]^2 < \infty\). Based on these estimates various applications for gamma and \(t\)-distributions are presented, and a simple proof of the CLT for i.i.d. r.v.'s is obtained.