An efficient method for spatiotemporally resolved aerosol flow modeling: Discrete migration and GPU acceleration

Describing spatiotemporal evolution and characteristics of dispersed systems using the population balance equation (PBE), examples including sectional and moment methods are fraught with numerous issues. Hence, this study develops an accurate method by combining computational fluid dynamics and population balance-Monte Carlo method (CFD-PBMC) with a moderate computational cost. An efficient sub-model for particle migration was proposed to simulate the convection and diffusion processes of particulate flows. A graphics processing unit (GPU)-based parallel computation was performed to accelerate the high-dimensional CFD-PBMC. Several classical cases with analytical or benchmark solutions were simulated, and a comprehensive comparison was made using the classical weighted random walk method. Good agreements were obtained, except in the case of radial migration, the reasons for which are explained in detail. The measured speedups on the GPU showed a factor of ~450 for pure migration and ~50 for the CFD-PBMC method when compared with a standard high-performance computer.     

Numerical simulation of incompressible fluid flows

In this paper we discuss temporal, spatial, and architectural aspects of the numerical simulation of time-dependent incompressible fluid flows. In particular, we consider: high-order operator-integration-factor time-splitting methods; sliding-mesh spectral mortar-element spatial discretizations; and data-parallel distributed-memory medium-grained parallel solution techniques. Numerous flow examples are presented.