Optimal rates of linear convergence of relaxed alternating projections and generalized Douglas-Rachford methods for two subspaces (Q312177)

The authors establish conditions under which convergent matrices attain the optimal convergence rate. Based on this result, the optimal linear convergence rates of the relaxed alternating methods and the generalized Douglas-Rachford splitting methods with parameters for two linear subspaces are analyzed. The optimal linear convergence rate is explicitly given in terms of the relaxation parameter and principal angles between two subspaces. A nonlinear map that helps to accelerate the convergence of alternating projection methods is introduced. Some numerical experimental results are provided to illustrate the convergence theory developed.