Multilevel Schwarz Preconditioners for Two- and Three-phase Flow in Porous Media.

Chris Kees

Department of Environmental Sciences and Engineering
University of North Carolina, Chapel Hill, NC 27599-7400


Abstract

Differential-algebraic systems can be derived from multiphase model equations via the method of lines. These systems are solved effectively with variable order, variable time-step backward difference formula methods. Maintaining efficiency in two and three dimensions, however, requires the parallel iterative solution of large nonlinear systems of equations. For this purpose we investigate Newton-Krylov solvers and domain decomposition preconditioners. We present model formulations, method implementation details, and results from numerical experiments along with a discussion of the trade-offs between efficiency and robustness for a variety of preconditioners. We use cell-centered finite differences to discretize the model equations in space and then advance the solution in time using an adaptive backward difference scheme that controls the local truncation error of the temporal discretization. This method produces, at each step, a nonlinear system of equations, which is solved with a modified Newton-Krylov method. The Krylov method is scaled, preconditioned, Bi-CGSTAB. We compare two-level additive and multiplicative Schwarz preconditioners to one-level additive preconditioners for the scaled Bi-CGSTAB iteration.