Algebraic Multigrid Preconditioning for Adaptive-Implicit Black-Oil Simulation

Geoffrey Brown


First, the research introduces the Algebraic multigrid (AMG), the effect of grid hierarchy and its advantages such as faster reduction of low frequency errors. The AMG preconditioners are developed for a black-oil simulation. Numerical experiments for serial and parallel simulations show that preconditioning is required for the speedup of data sets, and ILU iterations using SAMG can be reduced. Row-scaling provides the best speedup using AMG; Dynamic Rowsum (DRS) is the most robust preconditioner currently; AMG still has difficulties with some data sets possibly due to non-parabolic nature of PDE’s, and matrix checking and well scaling is our planned work to attempt to address these issues.