Optimal rates of convergence of matrices with applications
Heinz H. Bauschke, Yunier Bello-Cruz, T. T. A. Nghia, Hung M. Phan, Xianfu Wang · arXiv (Cornell University) · 2014
We present a systematic study on the linear convergence rates of the powers of (real or complex) matrices. We derive a characterization when the optimal convergence rate is attained. This characterization is given in terms of semi-simpleness of all eigenvalues having the second-largest modulus after 1. We also provide applications of our general results to analyze the optimal convergence rates for several relaxed alternating projection methods and the generalized Douglas-Rachford splitting methods for finding the projection on the intersection of two subspaces. Numerical experiments confirm our convergence analysis.