The parallel direct search, or PDS, algorithm only uses function and constraint values to find the constrained optimum of a real nonlinear function of a vector variable. PDS has been used in many applications, and it is consistently robust and effective. The basic algorithm is supported by a strong convergence analysis. PDS was originally designed by John Dennis and Virginia Torczon, and Virginia Torczon provided the convergence analysis and implemented the original version for unconstrained problems. David Serafini, also of CRPC, is presently developing a version for constrained problems that is portable to a wide range of parallel computers.
For more information on the original (unconstrained) version of PDS, or for downloading instructions, visit the PDS Web Site.
For more information on the latest (constrained) version of PDS, contact the author, David Serafini at email@example.com.