On using the first difference in the Stein-Chen method (Q1379711): Difference between revisions

If \(X\) and \(Y\) are random variables and \(g\) a bounded Lipschitz function, then there are two natural bounds for \({\mathbf E}| g(X)- g(Y)|\), given by \(K(g)d_{\text{W}}(X,Y)\) and \(\| g\| d_{\text{TV}}\) \((X<Y)\), where \(K(g)\) denotes the Lipschitz constant and \(\| g\|\) the supremum norm of \(g\), and where \(d_{\text{W}}\) and \(d_{\text{TV}}\) denote the Wasserstein and total variation distances, respectively. In the usual Stein-Chen argument, the former inequality is most often used. The latter is exploited to obtain alternative bounds for Poisson approximation of random variables and point processes. For sums of independent indicators, the resulting bounds often turn out to be a little sharper.