An optimized task-farm model to integrate reduced dimensionality Schrodinger equations on distributed memory architectures

摘要

This paper investigates the parallelization of a quantum reactive scattering code that tackles the problem of integrating a two-dimensional Schrodinger equation. Computational codes integrating differential equations associated with quantum chemistry problems are known to be difficult to parallelize. A detailed examination of the algorithms and a careful analysis of possible sources for inefficiencly is performed to devise a suitable parallelization strategy on MIMD-DM machines. As a result of the study, a task-farm model was adopted. The efficiency of the adopted parallel model is discussed with the help of several test calculations to pave the way for future work on full dimensional Schrodinger equations.