The PRISM Project: Recent Progress on Algorithms for Parallel Eigensolvers

Anna Tsao

In the first part of this talk, work done as part of the PRISM (Parallel Research on Invariant Subspace Methods) Project will be presented. The PRISM project is composed of researchers from Argonne National Laboratory, Duke University, the Supercomputing Research Center, the University of California at Berkeley, and the University of Kentucky, who are developing algorithms and software for solving large-scale eigenvalue problems based on the Invariant Subspace Decomposition Approach (ISDA) originally suggested by Auslander and Tsao. Algorithms based on the ISDA derive most of their parallelism from highly scalable primitives, such as dense matrix multiplication, as well as encompassing a divide-and-conquer component that is useful in load balancing. After a brief review of the mathematical framework for the ISDA, we will discuss progress in developing two symmetric eigensolvers based on this approach.

In the second part of this talk, I will briefly discuss my experiences as a mathematician, trained in a pure field, but working as an applied research mathematician/computer scientist outside of academia.

Anna Tsao received a Ph.D. in mathematics, with a specialty in univalent function theory, from the University of Michigan in 1981. She spent a few years in academia at the U. S. Naval Academy and Texas Tech University, after which she took the plunge into industry, where she worked at Hughes Aircraft Company and AT&T Bell Laboratories before taking her current position at the Supercomputing Research Center in 1988. Her industrial experience has included writing functional requirements and software for the F15E radar system, systems engineering for underwater sonar systems, and the design and development of parallel algorithms for solving eigenvalue problems.