Accelerating modified Shepard interpolated potential energy calculations using graphics processing units

The potential energy surfaces constructed with the modified Shepard interpolation scheme have been widely used in studies of chemical reaction dynamics. However, computational costs of interpolation increase rapidly with the size of the system and the number of data points needed to achieve a given accuracy. In this work, we present a naive Graphics Processing Unit (GPU)-accelerated algorithm for modified Shepard interpolated potential energy calculations and its implementation with the PGI CUDA Fortran language. The benchmark tests on a NVIDIA Tesla C2050 using four interpolated potential energy surfaces (one for H+H2O↔H2+OH, two for H+NH3↔H2+NH2 and one for H+CH4↔H2+CH3) demonstrated a speedup of 50-fold over the original CPU implementation on an Intel E5620 processor and the speedup increases with the system size and the number of data points. This work presents a promising GPU application in the field of chemical reaction dynamics.