On the rate of convergence of the centering walk (Q1897904): Difference between revisions

For a bounded measurable set \(B \subset R^k\) of nonzero measure the author defines the centering walk to be the sequence \[ C_{n + 1} \triangleq C_n + \varphi_n (X_n - C_n) \cdot \chi_B (X_n), \quad n = 0,1,2, \ldots, \] where \(0 \leq \varphi_n \leq 1\), \(C_0\) is an initial approximation, \(X_n\) is a sequence of normally distributed random vectors with mean vector \(C_n\) and unit covariance matrix. The author proves that if \(\sum^\infty_{n=0}\varphi_n = \infty\) and \(\sum^\infty_{n = 0} \varphi^2_n < \infty\), the sequence \(C_n\) converges to the set of points, satisfying the centering condition \(C_+ = E(X_+ \cdot \chi_B (X_+))/E (\chi_B (X_+))\), where \(X_+ \sim N (c_+, I)\). He also obtains, for \(\varphi_n = \min (b,a/n)\), the estimate of the rate of convergence which has the order \(O(n^{- 1})\).