Parallel subgradient methods for convex optimization. (Q2768025)

A subgradient method for minimizing a sum of convex functions on a convex set is given. To generate a search direction, each iteration employs a subgradients of a subset of a subset of the objectives evaluated at the current iterate, as well as past subgradients of the remaining objectives. A stepsize via ballstep level controls for estimating the optimal value. It is established the global convergence of the method. When applied to Lagrangian relaxation of separable problems, the method allows for almost asynchronous parallel solution of Lagrangian subproblems, updating the iterates as soon as new subradient information becomes available.NEWLINENEWLINEFor the entire collection see [Zbl 0971.00058].