Parallel Predictor-Corrector Interior-Point Algorithm of Structured Optimization Problems

摘要

A novel parallel predictor-corrector interior-point algorithm (P-PCIPA) is proposed. The algorithm uses technique of predictor-corrector to improve the centrality of the iterate point to reduce the number of iterations. Corrector equation, which has special structure, of our parallel algorithm for a class of large scale optimization problems has been presented. And this structured corrector equation can be solved in parallel with our parallel factorization-backsolve technique. The good speedup, which nearly equaling to the number of processors, can be got from our analysis of time complexity.