rik@cs.washington.edu (Rik Littlefield) (10/09/90)
I am looking for information on state-of-the-art in parallel eigensolvers. These matrices are real symmetric or complex Hermitian, dense or widely banded, and we need all the eigenvectors. The target computer is MIMD distributed memory, and the processor count may be large compared to the matrix size, e.g., 512 processors applied to matrices of size 200x200 .. 1000x1000. On serial machines, our applications (quantum chemistry) often use QR, sometimes Jacobi. We have developed a factored Jacobi that scales nicely into the many-processor regime, but would prefer a faster basic algorithm. If anyone has a parallel QR or similar algorithm that works well with lots of processors, we'd like to find out about it. Please email responses. I will summarize to the net if there is interest. Thanks much! -- Rik Littlefield Tel: (509) 375-3493 Molecular Sciences Research Center email: d39135@pnlg.pnl.gov Battelle, Pacific Northwest Lab or rik@cs.washington.edu Richland, WA 99352