Using a Filter-Based Sequential Quadratic Programming Algorithm in a Parallel Environment

A parallel, filter-based, sequential quadratic programming algorithm is implemented and tested for typical general-purpose engineering applications. Constrained engineering test problems, including a finite element simulation, with up to 512 design variables are considered.Theaccuracyandserialperformanceofthefilter-basedalgorithmarecomparedagainst that of a standard sequential quadratic programming algorithm. The parallel performance of the algorithm is evaluated, using up to 52 cores on a Linux Cluster. The results indicate that thefilter-based algorithm competes favorably with a standard sequential quadratic programming algorithm in a serial environment. However, the filter-based algorithm exhibits much better parallel efficiency due to the lack of a one-dimensional search.