On Stein operators for discrete approximations (Q2405126)

Let \(Y\) be a random variable taking values in \({Z_ +} = \{ 0,1,2,\dots\} \), \({\operatorname{E}}Y < \infty \), \(F\) be the set of real valued bounded functions on \({Z_ + }\), and \(GY\) be the subset of \(F\) consisting of functions \(g\), \(g(0)=0\), and \(g(x)=0\) for \(x \notin \operatorname{supp} Y\). Let also \(Z\) be a random variable with values in \({Z_ + }\) and \(f \in F\). The paper establishes bounds for \(\operatorname{E}f(Z) - \operatorname{E}f(Y)\) using Stein's method. For any \(g \in GY\)a linear operator \(A\) satisfying \(\operatorname{E}(Ag)(Y) = 0\) is established. \(A\) is called a Stein operator. The existence of multiple Stein operators is shown in the case of convoluted measures. In the next step, the so-called Stein equation \((Ag)(j) = f(j) - \operatorname{E}f(Y),f \in F\) is solved with respect to \(g(j)\). Taking expectations of this equation, we get \(\operatorname{E}f(Z) - \operatorname{E}f(y) = \operatorname{E}(Ag)(Z)\). Exploiting this result and useful properties of solutions to the Stein equation, the autors establish desired bounds.