A GPU-based parallelized Monte-Carlo method for particle coagulation using an acceptance-rejection strategy

A graphics processing unit (GPU)-based Monte Carlo (MC) algorithm for particle coagulation using an acceptance-rejection (AR) strategy leading to improved computing efficiency has been developed and validated. The use of GPUs in high-performance computing is attractive due to the low cost per core, currently some 1-2 EUR. The GPU-implementation developed takes full advantage of the intrinsic parallel property featured by the AR strategy, namely, multiple AR attempts are carried out independently on many threads simultaneously. It uses an efficient way to obtain an estimation for the maximum coagulation kernel from the mean kernel. The method has been benchmarked by a sectional method validating its computing accuracy. Especially when a large number of cells is being handled at the same time, remarkable speed-up factors are achieved. This makes the method, a choice when population balances have to be solved in a CFD environment, which is demonstrated by means of a case study describing simultaneous coagulation, nucleation and diffusion in 1D. In summary, the simulations show that a MC method for particle coagulation based on the AR strategy can be efficiently parallelized on a GPU.