News & Announcements

April 27, 2012

"Argonne's Sou-Cheng Choi Wins Linear Algebra Prize"

Media Contacts: Gail Pieper at pieper@mcs.anl.gov

Sou-Cheng Choi, a computational mathematician, together with colleagues Christopher Paige and Michael Saunders, has won the 2012 SIAM Activity Group on Linear Algebra (SIAG/LA) Prize, for the paper “MINRES-QLP: A Krylov Subspace Method for Indefinite or Singular Symmetric Systems,” which appeared in the SIAM Journal on Scientific Computing in 2011.

The SIAG/LA prize, awarded every three years by the Society for Industrial and Applied Mathematics (SIAM) for the best scientific work published in linear algebra, will be presented on June 21 at the SIAM Conference on Applied Linear Algebra in Valencia, Spain, where Choi will present a talk on the award-winning paper. Past recipients include several members of the National Academies (U.S. and foreign) as well as a Turing Award winner.

According to the SIAG/LA Prize committee, “The authors improve on MINRES, an elegant, efficient and widely used iterative method for linear systems, achieving optimal accuracy and extending the algorithm to the solution of least squares problems.”

Choi is a computational mathematician in the University of Chicago’s Computation Institute and holds a joint appointment at Argonne National Laboratory. She received her Ph.D. in computational and mathematical engineering from Stanford University in 2007. Paige is a McGill University emeritus professor, and Saunders is a research professor at Stanford University.

Motivating their work was the fact that traditional methods often break down or fail to find the minimum-length solution to a singular symmetric or Hermitian system. In other cases, the solvers converge but require more than twice the number of iterations to reach a solution. This new method involves two phases: it initially uses iterations from the original minimum-residual MINRES algorithm of Paige and Saunders from 1975, and it then transfers to using MINRES-QLP iterations when the problem becomes moderately ill-conditioned.   

“An ill-conditioned problem is one in which the outputs can change significantly in proportion to small changes in the inputs,” said Choi. “Our new method works well with these problems, avoiding a potential instability in MINRES.”

The announcement is available on the SIAM website at: http://connect.siam.org/?p=2079?p=2079