论文部分内容阅读
Graphics processing unit (GPU) computing has been successfully applied in some scientific simulations.In this work we developed an efficient GPU version of a time-dependent wave-packet code for atom-diatom state-to-state reactive scattering processes.Both the reactant Jacobi coordinates and product Jacobi coordinates approach are implemented,and the efficiency and convergence are compared.Since we use the split operator method to propagate the wave-packet,most of the runtime on each time step in the program is consumed in two parts: (1) the transformation of wave-packet between discrete variable representation (DVR) and finite basis representation (FBR) and (2) the actions of the several parts of Hamiltonian on the wave-packet.Both of these parts are actual matrix multiplication and are therefore very viable to realize paralleling in GPU with careful structuring of the matrices and manipulation of multiplication.