An approach to parallel computing in an Eulerian-Lagrangian two-phase flow model

An approach to parallel solution of an Eulerian-Lagrangian model of dilute gas-solid flows is presented. Using Lagrangian treatments for the dispersed phase, one of the principal computational challenges comes in models in which inner-particle interactions are taken into account. Deterministic treatment of particle-particle collisions in the present work pose the most computationally intensive aspect of the simulation. Simple searches lead to algorithms whose cost is O(N_p ²) where N_p as the particle population. The approach developed in the current effort is based on localizing collision detection neighborhoods using a cell-index method and spatially distributing those neighborhoods for parallel solution. The method evaluated using simulations of the gas-solid turbulent flow in a vertical channel. The instantaneous position and the velocity of any particle is obtained by solving the equation of motion for a small rigid sphere assuming that the resulting force induced by the fluid reduces to the drag contribution. Binary particle collisions without energy dissipation or inter-particle friction as considered. The carrier flow is computed using Large Eddy Simulation of the incompressive Navier-Stokes equations. The centre dispersed-phase population as partitioned via static spatial decomposition of the domain as maximize parallel efficiency. Simulations on small numbers of distributed memory processors show linear speedup in processing of the collision detection step and nearly optimal reductions at simulation time for the entire solution.