MAS1.1: SPARSE

This PhD project deals with sparse grid methods for time-dependent PDE problems. Advection-diffusion-reaction problems from air quality modeling form an important test case. Joint project with MAS2. The project started in January 1998. The planned duration is four years.

Project Description:

The aim of this 4-year PhD project is to make significant progress in the numerical solution of large-scale transport problems: systems of partial differential equations of the advection-diffusion-reaction type. Such problems are frequently encountered in applications. They play a prominent role in the mathematical modeling of pollution of atmospheric air, surface water and groundwater. Advanced models are three-dimensional in space. Their 3D nature and the necessity of modeling transport and chemical exchange between different components over long time spans, In the past, much research has been done on developing efficient solvers, notably advection schemes, tailored integrators for stiff systems of ordinary differential equations and other time stepping techniques. This has already led to significant progress. However, for advanced 3D modeling, computer capacity (computing time and memory) still is a severe limiting factor. This limitation is felt in particular in the area of global air pollution modeling where the 3D nature always leads to huge numbers of grid points in each of which many calculations must be carried out.

This project aims at taking up a new line of research: application of sparse-grid techniques in order to reduce the number of grid points, of course without loss of accuracy and in combination with an appropriate, efficient time stepping process. Sparse grids were introduced by Zenger in the early nineties to reduce the degrees of freedom in finite element calculations. Recent literature in the field of reactive flow problems and multigrid methods for stationary differential equations confines this very promising development. Sparse grids can be considered through multivariate extrapolation techniques. We plan to investigate these techniques in order to further enhance the efficiency of solving advection-diffusion-reaction problems. By error analysis and experimentation, the supposed advantage of sparse grids must be shown. To assess the newly developed algorithms for truly practical purposes, at the end use will be made of a prototype of a global atmospheric air pollution model. Implementation aspects, including vectorization and parallelization, must be taken into account in this assessment.

People involved:


Last update 98/02/17.