Simulating chemical systems: MPI and GPU parallelization of novel SD algorithms

Molecular dynamics is used for simulating chemical systems with the goal of studying a large range of phenomena starting from cell structures to the design of new materials, drugs, etc. A very important component of molecular dynamics is the use of well-suited atomistic and molecular modelling of the chemical systems. This article presents the MPI and GPU-CUDA parallelization of novel stochastic-Langevin dynamics algorithm that is used in molecular dynamics for controlling the temperature of simulated systems. The research has been performed in the Molecular Dynamics Group of the University of Groningen and it is going to be included in the next version of Gromacs tool of molecular dynamics (www.gromacs.org). The Langevin algorithm implemented is original and is based on impulse application of friction and noise. Theoretical background, implementation, an efficiency discussion and relevant simulation results are presented in the different sections of this article. The simulations used Martini water. The parallelization of the algorithms was done in two versions: one in MPI using domain decomposition and another version was done in CUDA.