A Hybrid Approach For Solving Sparse Linear Least Squares Problems

  • Date: 12/01/2009
Lecturer(s):
Esmond G. Ng (Computational Research Division, Lawrence Berkeley National Laboratory)
Location: 

University of British Columbia

Description: 

We propose a hybrid method for solving large sparse linear least squares problems. The method is iterative in nature, as it is based on preconditioned LSQR. However, the preconditioner comes from an orthogonal factorization of a submatrix of the original matrix associated with the least squares problem; the construction of the preconditioner is based on well-known techniques in sparse direct methods. In this talk, we will discuss the quality of the preconditioner and the choices of the submatrix.

Schedule: 

12:30pm - 2:00pm, WMAX 216.

Sponsor: