2:30 p.m. - 3:30 p.m. Location: FO 3.222
Fast algorithms for large scale eigenvalue problems
Eigenvalue problems are of fundamental importance and arise frequently in many scientific disciplines, including materials science and data mining. Many modern applications can benefit from algorithms that can solve large scale eigenvalue problems more efficiently. In this talk, we first give a brief overview of large scale eigenvalue problems, and survey a few representative algorithms by pointing out their essential common features. Then we discuss some main challenges for solving ever larger scale eigenvalue problems. We present some recent progress on the 'preconditioned' eigen-solvers related to solving the Kohn-Sham equation in density functional theory, both in the plane-wave setting and in the real-space setting. We will also discuss a novel spectrum decomposition algorithm that addresses several main difficulties encountered in state-of-the-art algorithms for solving large scale eigenvalue problems.
Sponsored by the Department of Mathematical Sciences