PARALLEL MULTISECTION APPLIED TO THE EIGENVALUE PROBLEM

The results of some parallel multisection methods are applied to the determination of eigenvalues of real symmetric tridiagonal matrices. The first method whih is suitable for Single Instruction Multiple Data computers has been run on the ICL DAP. The size of the processing array limits only the number of eigenvalues that can be used in parallel, but not the size of the matrix. The second method uses Multiple Instruction Multiple Data parallelism and has been run on an asynchronous 4 processor system. This version allows faster processors to move ahead to sample further points without having to wait on slower processors.