Implementation of a proximal algorithm for linearly constrained nonsmooth optimization problems and computational results

This paper describes an implementation of the so-calledproximal point algorithm for solving convex linearly constrained nonsmooth optimization problems. Contrary to other previous implementations of the same approach (which solve constrained nonsmooth problems as unconstrained problems via exact penalty function techniques), our implementation handles linear constraints explicitly (linear constraints being incorporated into the direction-finding subproblem). The relevance and efficiency of the approach is demonstrated through comparative computational experiments on many classical test problems from the literature, as well as on a series of large constrained dual transportation problems introduced and studied here for the first time.