Efficient biased random bit generation for parallel lattice gas simulations

Abstract

Lattice gas methods are often used to model the kinetics of a variety of diffusive systems. One of the main advantages of these methods is the ease at which they can be parallelized using simple bit vector operations. However, to describe the kinetics of the lattice gas in a randomly biased fashion, it is necessary to efficiently generate a randomly biased bit vector. If one generates a random floating point number per bit, then this is very costly. In this paper, an efficient algorithm is developed that leads to a fully bit vector implementation of a lattice gas automation while significantly reducing the amount of needed generated random floating point numbers. The bit vector algorithm using the new random biased bit vector algorithm is tested on a lattice gas method whose solution is modeled by the solution of the 1-D Burgers' equation. This new lattice gas method is then implemented on the BBN TC2200 and CRAY 90 parallel processors.