Friday, March 14

1:30 PM-3:30 PM
Nicollet D2

MS4
Recent Progress in Sparse Direct Methods

The robust solution of large, sparse linear systems requires efficient direct solvers which employ some form of sparse matrix factorization (Cholesky, LU, or QR). The central idea in sparse direct solution is that of limiting "fill" (zeros in the original matrix that become nonzero in the factor) by ordering the rows and columns of the matrix. The actual numeric factorization is performed in a subsequent step using the reordered matrix.

The solution of sparse linear systems is the main computation in a vast number of scientific and engineering applications. Parallel sparse direct solution is therefore of significant interest to application developers. Furthermore, the development of parallel sparse direct solvers involves new graph algorithms and data-partitioning techniques and should be of interest to those concerned with algorithm design for parallel unstructured computation.

Organizer: Padma Raghavan
University of Tennessee, Knoxville

1:30 Effective Sparse Matrix Ordering: Just Around the Bend

Bruce Hendrickson, Sandia National Laboratories, Albuquerque, and Edward Rothberg, Silicon Graphics, Inc.

2:00 Parallel Supernodal Method for Sparse Gaussian Elimination

Xiaoye S. Li, Lawrence Berkeley National Laboratory, James W. Demmel, University of California, Berkeley, and John R. Gilbert, Xerox Palo Alto Research Center

2:30 A Parallel Sparse Direct Solver for Least Squares Problems

Chunguang Sun, Cornell University

3:00 Sparse Direct Solution on Distributed Memory Machines

Michael T. Heath, University of Illinois, Urbana-Champaign, and Padma Raghavan, Organizer

PP97 Homepage | Program Updates|
Registration | Hotel Information | Transportation | Program Overview | Program-at-a-Glance