LA/Opt Seminar: A New Analysis of the Symmetric Lanczos Algorithm in Finite Precision, Professor Ming Gu, Dept of Mathematics, UC Berkeley

LA/Opt Seminar

Title: A New Analysis of the Symmetric Lanczos Algorithm in Finite Precision
Speaker: Professor Ming Gu, Dept of Mathematics, UC Berkeley
Date: November 13, 2017
Time: 4:30pm
Location: Y2E2 111

Abstract:

Since C. Paige's finite-precision analysis half a century ago, the
Symmetric Lanczos Algorithm and its finite-precision variants have been
the main workhorse for computing leading eigenvalues of a large sparse
matrix. We review Paige's analysis and discuss new insights into the
finite-precision behavior of the SLA. In particular, we propose a
simple scheme to detect eigenvalue convergence and a modification
procedure to make SLA numerically stable without too much additional
reorthogonalization.

Date: 
Thursday, November 16, 2017 - 4:30pm to 5:30pm
location: 
Jerry Yang and Akiko Yamazaki Environment and Energy Building (Y2E2), 473 Via Ortega, Stanford, CA 94305, USA