Parallel Solution of Linear Programs Via Nash Equilibria.

The linear programming problem is shown to be equivalent to a game in which primal players minimize the augmented Lagrangian function for the primal problem and dual players maximize the augmented Lagrangian function for the dual problem. Based on that, a parallel solution method is developed in which processors carry out under-relaxed Jacobi steps for the players. Strong convergence of the method is proved and the ratio of linear convergence estimated. Computational results are highly encouraging.