Parallel, Physics-Based Preconditioning and the Jacobian-Free, Newton-Krylov Method: Application to Multicomponent Geochemical Transport
Glenn Hammond, University of Illinois
As today's subsurface reactive transport models become more sophisticated with large heterogeneous flow domains and complex multicomponent geochemical reactions, scientists are relying more and more on parallel supercomputing to address extensive computational demands. I have developed the parallel reactive transport code PARTRAN, which utilizes leading edge solvers and data structures for parallel computation. It is well known that poor scalability in the parallel realm often originates in the decoupling of the system of equations within conventional preconditioners due to parallelization of iterative Krylov solvers. To circumvent this problem within geochemical transport, PARTRAN utilizes the Jacobian-free, Newton-Krylov method with physics-based preconditioning. I provide results illustrating the superior performance of this preconditioning technique as it is applied to modeling enhanced in situ biodegradation of tetrachloroethylene (PCE) within 3D heterogeneous flow.
Abstract Author(s): Glenn E. Hammond and Albert J. Valocchi, University of Illinois at Urbana-Champaign; Peter C. Lichtner, Los Alamos National Laboratory