A parallel algorithm for constrained and nonlinear optimization problems

To improve Parallel Variable Transformation method (PVT) for unconstrained optimization problem, different research direction is constructed in every processor. The every research direction is proved to a descent direction. Every minimizer is researched along every descent direction in every processor. A starting vector is updated by selecting the minimizer in every processor in next iteration. In theory, Convergence and rates of convergence of the method is proved. Some numerical results on HP rx 2600 cluster show that experimental results are consistent with the theory, and the efficiency of the algorithm is very high.