Parallel implementation of 2D LGA and MD simulations

Lattice gas automata (LGA) and molecular dynamics (MD) have become alternative computational methods for studying complex dynamics problems at the microscopic level. In this paper we present efficient parallel programs for 2D LGA and MD simulations with large number of particles (106+). The LGA algorithm is based on the 2D 7-bit FHP model with a rest particle, 76 nondeterministic collision rules and special sites: reflectors, absorbers and injectors. Multispin coding was used for efficient calculations. In the MD program we have applied the Verlet list to speedup calculations of Lennard-Jones interactions and Hockney-cells based distributed sorting of particle data to generate the neighbor lists. Parallel algorithms for both LGA and MD simulations are based on domain decomposition (i.e. computational box is divided into layers), on the same virtual topology of communication, and similar load balancing (LB) algorithm. The aim of the LB procedure is to reallocate data adaptively among processors of a computing system, taking into account changing load of processors, in a way which reduces the execution time of the program. Both programs are optimized for the execution time and memory requirements. LGA and MD parallel programs are scalable and they were successfully implemented on networked workstations and multicomputers under PVM and applied to wide class of computer experiments. We consider this work to be the first step towards development of an integrated environment for efficient simulation of processes for which classical continuous models are not sufficient. Such an environment should enable intercomparison of results of different particle techniques like LGA, lattice Boltzmann, MD, and dissipative particle dynamics.