The State of Large-Scale Hermitian Eigenvalue Solvers
|Location:||3083 Kemper Hall|
|When:||Fri Oct 23, 2009 15:10|
Hermitian Eigenvalue problems continue to be a largely unavoidable bottleneck in many applications. We discuss current state of the art eigenmethods, their successes and limitations, and revisit Lanczos type algorithms which have fallen out of favor because they cannot utilize multiple initial guesses and preconditioning. We also discuss our promising recent eigCG technique that finds Lanczos quality eigenpairs without keeping the whole Lanczos basis.