Stein's method for conditional central limit theorem (Q6102954)

The authors extend Stein's method to give error bounds and rates of convergence in univariate and multivariate conditional central limit theorems. That is, they give explicit error bounds (measured in 1-Wasserstein distance) in Gaussian approximation for random variables of the form \((W|Y=k)\), and analogous multivariate results. Here \(Y\) is assumed to be a discrete random variable uncorrelated with \(W\). The approach the authors take is motivated by the well-known exchangeable pairs approach to the usual central limit theorem via Stein's method, and uses linearity conditions of a type familiar from that approach. Applications given include the appearance of patterns in random binary sequences, and subgraph counts in an Erdos-Rényi random graph. The paper concludes with a discussion of open problems and potential future work.