The parallel programming project I am working on is to parallelize MISER. MISER is a soil-remediation simulation program which models soil chemistry and biochemistry in the presence of gas and water transport. It is a large EPA-funded project directed by Prof. Linda Abriola, head of U. Michigan's Environmental Studies program. MISER is a multistage program, taking several passes at each timestep to solve as many as a dozen or so coupled systems of sparse nonlinear PDEs. The preliminary parallel version currently uses the AZTEC library as its solver. While it now runs on an IBM SP2, the use of standard packages (MPI, AZTEC) should make it portable to other parallel systems.

While the MISER program is currently used by many groups, the complex chemistry involved makes MISER infeasible for field-scale testing on workstations: either too little resolution, or too little volume, can be simulated to run such large-scale simulations, despite assuming radial symmetry to cut down on the workspace. Further since radial symmetry of soil composition and pollutant distribution tends to be rare outside the laboratory, a truly 3-dimensional simulation is keenly needed, which necessitates the development of a very high-performance parallel version.

In the short term, I would like to better understand the AZTEC package, to insure that I am making best use of it. Simultaneously I would also like to become familiar with other packages for solving systems of equations in parallel, to see if another might be better-suited to the task at hand. In the longer term it will be necessary to use dynamic regridding for the model when we move to a fully 3d model, if not sooner. Tools for doing so efficiently, with reasonable load balancing, would therefore be very useful.

It is important that the parallel MISER code uses standardized tools that will continue to supported, such as those in the ACTS program. This is because the MISER program has users throughout the nation and is expected to be in use for many years, with ongoing development to continually add new aspects to the simulation. We will need to simultaneously support flexibility and high-performance.lectic structures, various symmetries.