Multivariate normal approximation with Stein's method of exchangeable pairs under a general linearity condition (Q971936)

Stein's method has been suggested to asses the distance between univariate random variables and normal distributions. It makes use of the method of exchangeable pairs [see \textit{C. Stein}, Inst. Math. Stat. Lect. Notes - Monogr. Ser. 7 (1986; Zbl 0721.60016); \textit{A. Röllin}, Stat. Probab. Lett. 78, No. 13, 1800--1806 (2008; Zbl 1147.62017); and \textit{S. Chatterjee} and \textit{E. Meckes}, ALEA, Lat. Am. J. Probab. Math. Stat. 4, 257--283, electronic only (2008; Zbl 1162.60310)]. The paper establishes a multivariate exchangeable pairs approach enabling to asses distances to potentially singular multivariate normal distributions. An embedding method is suggested allowing for a normal approximation even when the corresponding statistics are inappropriate for Stein's exchangeable pairs approach. Examples of runs on the line and double-indexed permutation statistics illustrate the method.