Development and parallelization of a hybrid particle/continuum method for simulating rarefied flow.

摘要

A particle method such as the Direct Simulation Monte Carlo method (DSMC) simulates a gas flow by statistically modeling the behavior of a large number of virtual particles and is necessary for the simulation of rarefied flows for which the Navier-Stokes (NS) equations become invalid due to failure of the constituent relations upon which they are based. Unfortunately, while more versatile than NS, DSMC also requires much more computational effort. Even on a parallel computer, simulation times can become long, and so a less computationally intensive method is desirable.;Often, most of the particles, and hence computational intensity, are contained in a relatively small, dense region of the domain where a very disproportionate fraction of DSMC's effort is expended. Fortunately, however, the region of greatest density and computational intensity also tends to be a region in which NS is viable. A logical approach, then, is to hybridize NS and DSMC, allowing the former to handle regions of higher density and the latter to handle regions of greater rarefaction.;Two fundamentally different approaches to the hybridization have been considered: flux-passing and state-passing. In the former, DSMC and NS interface by passing one-sided fluxes across a shared interface, while in the latter, state variables are passed between the two methods in a small region of overlap.;The flux-passing approach has a number of philosophically, if not pragmatically, appealing characteristics, the greatest of which is the apparent independence of the two domains. It is therefore the approach that was first pursued, and several simple simulations were run successfully. Unfortunately, attempts to solve problems with more complex, multidimensional interfaces failed and revealed a critical liability.;Attention was returned to the state-passing method when a reasonable solution to the problem could not be found. Despite a couple weaknesses, the state-passing method has proven overall to be more robust than the flux-asessing method, and the results have been very encouraging. A successful state-passing hybrid code has been developed, parallelized, and used to simulate several significant problems, including a rectangular solid in Mach 10 flow, a cylinder, and a wedge at Mach 8.