Three-Dimensional Massively Parallel Computing of Suspensions

Numerical simulations of suspensions often suffer from the fact that the simulated systems are rather small compared to experimental setups. We present a numerical scheme for non-Brownian particle-liquid mixtures in three dimensions at particle Reynolds numbers between 0.01 and 20 and describe its parallel implementation. The fluid equations are solved by a time-explicit pressure-implicit Navier–Stokes algorithm and the particle motion is tracked by molecular-dynamics methods. The two are coupled by imposing no-slip boundary conditions between the particles and the fluid. We integrate the stress distribution on the particle surface numerically to obtain forces and torques. The building blocks of the algorithm are local and scalable and we have reached particle numbers up to 10⁶ (1.41*10⁸ fluid nodes) on a 512 node CRAY-T3E. We compare our simulation results to theoretical predictions and experimental data and find good agreement for particle volume fractions up to 0.30.