Efficient diagonalization of oversized matrices on a distributed-memory multiprocessor (Q922639)

Discussion of strategies for the parallel computation of the symmetric eigenvalue decomposition by the Jacobi method on distributed memory systems. Distribution of the \(n\times n\) matrix onto the p processors: For load balancing either introduction of zero-columns or assigning excess columns to the first processor if mod(n,p)\(\neq 0\). Comparison of theoretical model with implementation on an NCUBE gives excellent agreement.