Scalable Algorithms for Multivariate Integration with ParAdapt and CUDA

Abstract

ParAdapt is a numerical integration software based on a classic global adaptive algorithm, which employs GPUs in providing integral evaluations. Based on adaptive region partitioning strategies developed for efficient integration and mapping to GPUs, ParAdapt renders the framework of the global adaptive scheme suitable for general functions in moderate dimensions, say 10 to 25. Speedup values achieved are in the double and triple digits up to very large numbers of subdivisions. While sequential adaptive strategies are generally considered effective for integral dimensions through about 10 or 12, it is our goal to move this threshold up considerably.