A Distributed Memory Implementation of the Nonsymmetric QR Algorithm

The QR algorithm is the crux of the serial nonsymmetric eigenvalue problem. Recent eeorts to parallelize this algorithm have made signiicant advances towards solving the parallel nonsym-metric eigenvalue problem. Most methods to date suuer a scalability problem. In this talk we discuss an approach for parallelizing QR which overcomes many of the disadvantages to date. We also give insights into what is necessary for a parallel algorithm to work using these strategies. Performance of a parallel implementation on the Intel Paragon TM system is reported.