Speaker: Prof. Yin Zhang
Department of CAAM, Rice University, USA
(Joint work with Zaiwen Wen, Chao Yang and Xin Liu)
Time: 3:30-4:30 Dec, 6
Place: ROOM 510
As massive parallelism is increasingly becoming a necessity
in today's scientific computing, new parallel algorithms are
acutely needed even for ``solved" fundamental problems such
as eigenvalue problems. In this work, we investigate a simple
formulation, based on the classic Courant penalty function,
for computing large-scale eigenspaces of huge sparse matrices.
The primary motivation is to avoid, as much as possible,
orthogonalization of many vectors which has often become the
bottleneck in computation. We will give theoretical properties
of the proposed model, and present numerical results.