Use of parallel SuperLU in large-scale scientific calculations. |
|
|
|
|
|
|
| Firstly, we present the main algorithmic features and software design in SuperLU_DIST, a distributed-memory sparse direct solver. We present the numerical results of the static pivoting strategy, the parallelization strategies, and the solver's performance on the parallel machines. Secondly, we describe two scientific applications in which SuperLU_DIST has played a critical role. The first application is in the solution of a long-standing problem of scattering in a quantum system of three charged particles. This requires solving the complex, nonsymmetric, and very ill-conditioned linear systems. The largest system solved is of order 8 million. The second application is in the solution of the interior eigenvalues of a sparse generalized eigenvalue problem arising from the accelerator design. The largest system solved is or order 7.5 million. |
|
|
algweb@cerfacs.fr Last Update: Apr 4, 2003 |