Solving nonsymmetric eigenproblems on distributed memory concurrent computers

A.P. Petitet
Proceedings of Scalable Parallel Libraries Conference  
The design and implementation of a parallel nonsymmetric eigensolver is one of the most challenging problems an numerical linear algebra. This paper describes a distributed memory concurrent computer implementation of an algorithm based on the matrix sign function to "spectrally divide and conquer" the matrix. Performance and scadibility results are also presented. Some of the Intel hypercube iPSC/860 and Intel Touchstone Delta system timings presented in section 3 were obtained with older,
more » ... efficient version of the code. Nevertheless these results not only illustrate the numerical behaviour and the parallelizability of the sign function algorithm, but also illustrate the portability and scalibility of all the previously mentioned parallel software including, the routines of the sign function.
doi:10.1109/splc.1993.365564 fatcat:rpdkbpqn6zcy5ft6apexjzkgva