Proximal alternating direction-based contraction methods for separable linearly constrained convex optimization (Q2430823)

The alternating direction method (ADM) has been well studied in the context of linearly constrained convex programming problems. Recently, because of its significant efficiency and easy implementation in novel applications, ADM is extended to the case where the number of separable parts is a finite number. The algorithm of the extended method consists of two phases. At each iteration, it first produces a trial point by using the usual alternating direction scheme, and then the next iterate is updated by using a distance-descent direction offered by the trial point. The generated sequence approaches the solution set monotonically in the Fejér sense, and the method is called alternating direction-based contraction (ADBC) method. In order to simplify the minimization subproblems in the first phase of the ADBC method, this paper presents some modified methods by adding the proximal terms to the objective function of the subproblems. Furthermore, the subproblems become easier in the linearized versions and the method is much more practical. For the convergence property of the proposed methods, there are various restrictions on the proximal parameters due to various versions of the proximal alternating direction scheme. All the presented algorithms are guided by the general framework of the contraction methods for monotone variational inequalities, and thus the convergence follows directly.