Parallel Refined Davidson Method for Solving the Large Scale Eigenproblem

A parallel refined Davidson method is presented for large sparse symmetric eigenvalue problems,its implementation on PC network parallel systems and shared memory computer systems,as well as the parallelism are analyzed. The rows of the matrix are distributed to each processor,and the individual processors run the program by using the orthogonal basis of projection subspace and the rows of the matrix. Combine with the restarting scheme,the method can solve several eigenvalues and its associated eigenvectors of the matrix,and the method is successfully used for computing the frequency of a plane wing. The numerical experiments are done on the PC network parallel system and shared memory parallel system IBM-P650.