Reactant Coordinate Based State-to-State Reactive Scattering Dynamics Implemented on Graphical Processing Units

Abstract

A parallel code for state-to-state quantum dynamics with propagation of time-dependent wavepacket in reactant coordinates has been developed on graphical processing units (GPUs). The propagation of wavepacket and the transformation of wavepacket from reactant to product Jacobi coordinates are entirely calculated on GPUs. A new interpolation procedure is introduced for coordinate transformation to decrease the five-loop computation to two four-loop computations. This procedure has a negligible consumption of extra GPU memory in comparison with that of the wavepacket and produces a considerable acceleration of the computational speed of the transformation. The code is tested to get differential cross sections of H+HD reaction and state-resolved reaction probabilities of O+HD for total angular momenta J = 0, 10, 20, and 30. The average speedups are 57.0 and 83.5 for the parallel computations on two C2070 and K20m GPUs relative to serial computation on Intel E5620 CPU, respectively.

Potential Energy Surfaces for the First Two Lowest-Lying Electronic States of the LiH2+ System, and Dynamics of the H+ + LiH ⇌ H2+ + Li + Reactions

“…where k B is the Boltzmann constant. The initial state selected differential cross section (DCS) is calculated as 43,44 ∑ ∑ ∑ ∑…”

“…To improve the accuracy of the new PESs, a wider coordinate range and high level quantum chemistry calculation method and basis set are used in the calculations of ab initio energy points. The dynamics calculations for reactions and on the new first-excited-state PES are carried out by using two recently developed graphics processing unit (GPU) accelerated time-dependent wave packet (TDWP) programs. ,− The dynamics results are presented and compared with available literature results. The rest of this paper is organized as follows.…”

GQSD: The program for the graphic processing units accelerated quantum scattering dynamics

“…Finally, the product functions are given by systems, branching ratio of O 1 HD reaction. [16,17,31] The performances of GQSD/RC and GQSD/PC are shown in Figure 4. The coordinate transformation in GQSD/PC is calculated using Eqs.…”

“…[14] In GPU accelerated time-dependent computation of Pacifici et al, [15] the speedups on order of 2-20 times have been reported in evaluation of the reactive probabilities. In our previous works, [16,17] we presented the implementation of state-to-state reactive scattering dynamics on GPUs with the propagation in both reactant and product Jacobi coordinates. The result shows that the global speedup of times is 83.5 for the parallel computation on two K20 GPUs comparing with serial computation of E5620 @ 2.40GHz CPU.…”

“…The theoretical backgrounds and tests on reactions can be found in the previous publication. [16,17] In this work, we concentrate on the details of the structure of the program and the implementation of parallel computation on GPUs. The article is organized as follows: next section outlines the organization of the program.…”