A Coarse Grained Particle Transport Solver Designed Specifically for Graphics Processing Units

Abstract

This article introduces a novel coarse-grained particle transport solver, designed specifically for streaming processor architectures. The coarse particles are transported using a Monte Carlo algorithm with a locally homogenized collision operator. Local errors introduced by the homogenization procedure and the use of (deterministic) quadratures, are described and analyzed. A brief description of how the simulation is mapped to the streaming processor (Graphics Processing Unit) is also given.

Keywords:

• Monte Carlo
• neutral particle transport
• GPU
• CUDA

Notes

¹Here, and for the remainder of the article, two particles are called coherent, if they follow the same execution branch within the simulation, which requires that they undergo the same physical interactions. For instance, if two particles arrive at a collision site, and the one scatters while the other continues with the same velocity, the particle histories are incoherent.

²The model is actually more flexible and is referred to as a SIMT (Single Instruction Multiple Threads) instruction model.

³ CUda Coarse Grained Particles.