Some multivariate inequalities with applications (Q2495704)

Let \(X_1,\dots,X_n\) be a stationary sequence of centered, uniformly bounded random variables satisfying the dependance condition introduced and investigated by \textit{P. Doukhan and S. Louhichi} [Stochastic Processes Appl. 84, 313--342 (1999; Zbl 0996.60020)]. Associated and negatively associated sequences satisfy this condition while strongly mixing ones do not. Let \(Y_1,\dots,Y_n\) be i.i.d. centered random variables independent of all \(X_i\)'s. For any real function \(f\) on \(\mathbb{R}^n\) with uniformly bounded second and third partial derivatives the quantity \(Ef(X_1,\dots,X_n)-Ef(Y_1,\dots,Y_n)\) is expressed in terms of covariance function of \(X_i\)'s and moments of \(Y_1\) up to some remainder. Suitable choices of \(f\) give firstly some Rosenthal-type inequalities for moments of sums \(S_n=X_1+\dots +X_n\) and secondly allow to evaluate the Dudley distance between the standard normal distribution and the distribution of normalized sums \(S_n\). Moreover, a corollary, suitable for \(X_i,\;Y_i\) with Bernoulli distribution, is obtained. The main tools are the Lindeberg decomposition and an extension of Taylor's formula.