Restarting Strategies for the DQA Algorithm

A scenario-based decomposition algorithm is proposed for large stochastic pro-grams. The subproblem clusters consisting of separable quadratic programs are solved by means of a nonlinear interior point algorithm. Critical implementation issues are analyzed, including restarting and alternative splitting strategies. The approach is suited to a distributed multicomputer such as a network of workstations. Testing with several large LPs (117,000 constraints and 276,000 variables) shows the efficiency of the concepts.