A class of linearized proximal alternating direction methods (Q658552)

The paper is devoted to some new alternating direction methods (ADMs) for the solution of structured separable convex minimization problems. The idea leading to these methods is to linearize the quadratic term in the objective function of either one or both subproblems to be solved and to add a proximal term to the objective function in case it is linearized. Some properties of the three resulting linearized proximal ADMs are presented and convergence of them is proved. Moreover, the analysis of the methods shows that, for these new ADMs, the relaxation factor can be chosen out of the same interval as for the classical ADM.