Parallel Computation for Stochastic Dynamic Programming: Row Versus Column Code Orientation

摘要

Parallel computations are presented for stochastic dynamic programming problems arising from the optimal control of nonlinear, continuous time dynamical systems, perturbed by Poisson as well as Gaussian random white noise. The numerical formulation is highly suitable for a vector multiprocessor. Column-oriented code can run more than twice as fast as row-oriented code for highly refined meshes, but row-oriented code is more efficient and practical for coarser meshes. Advanced computing techniques and hardware help alleviate Bellman's curse of dimensionality in dynamic programming computations. 10 refs., 3 figs. (ERA citation 14:016857)