James Demmel

Digital Library

ACM Paris Kanellakis Theory and Practice Award

USA - 2014

citation

For contributions to algorithms and software for numerical linear algebra used in scientific computing and large-scale data analysis.

Jim Demmel introduced a geometric view of error that could be applied to a broad class of known but unstable eigenvalue (and other) computations to produce methods that worked efficiently and had provably small error. This work served as an important view in the field for others as well for his own work. His theoretical work also identified and analyzed the impact of communication in finding optimal methods in numerical analysis. All the while, Professor Demmel was crucially involved in the implementation of provably reliable methods in LAPACK, the definitive and central code used whenever accuracy and speed is desired. Professor Demmel was deeply involved at various levels ranging from fast and accurate algorithms for the singular value decomposition to floating point issues. Professor Demmel's theoretical work in scalable systems is evident in his extensive involvement in scalable implementations of eigenvalue computations in ScaLAPACK. His joint work on the implementations of matrix factorizations central to solving sparse linear systems is embodied in SuperLU, the thesis work of his PhD student, Xiaoye Li. These libraries have been used in many scientific research projects, including some cover articles in Nature and Science.
Press Release

ACM Fellows

USA - 1999

citation

For outstanding contributions to scientific computing, parallel processing and software engineering.