Two-sided bounds and leading term for rates of convergence in the multivariate central limit theorem (Q1079864)

For a sequence \(X_ 1,X_ 2,..\). of i.i.d. random vectors with values in \(R^ k\), upper and lower bounds are obtained for the discrepancy between the probability measure \(P_ n\), induced by their normalized sum \(n^{-1/2}(X_ 1+...+X_ n)\) and the standard normal probability measure \(\Phi\) on \(R^ k\). The authors study the expansion \[ (*)\quad P_ n(B)=\Phi (B)+Q_ n(B)+R_ n(B),\quad B\in {\mathcal E}, \] where \({\mathcal E}\) is the collection of all convex Borel subsets of \(R^ k\), \(Q_ n(B)\) is a ''leading term'' and \(R_ n(B)\) is a remainder term. One of the main results - Theorem 2.4 - states that under the conditions \(EX_ 1=0\), \(E| X_ 1| =1\) \[ \{\sup_{B\in {\mathcal E}}| Q_ n(B)| \}/\delta_ n \] is bounded away from zero and infinity as \(n\to \infty\), where \(\delta_ n\) is a particular function of the distribution of \(X_ 1\). Theorem 2.1 states that \[ | P_ n(B)-\Phi (B)| \leq c(\delta_ n+\lambda^ 2n^{-1/2}),\quad B\in {\mathcal E}, \] where \(c=c(k)\) is a universal constant and \(\lambda\) is a number, depending on the distribution of \(X_ 1\). L. V. Osipov was the first mathematician who introduced and studied the expansion (*) for the case \(k=1\).