A Refined Distributed Parallel Algorithm For The Eigenvalue Problem Of Large-scale Matrix

In view of eigenvalue problems of large-scale matrix, this paper proposes a refined distributed parallel algorithm named RDPC-DTM based on direct transformation method and DPC-DTM algorithm which is a distributed parallel design of direct transformation method. This new method solves the problem that increasing the number of substructure could not effectively enhance the computing efficiency when the scale of matrix is too large. Numerical experiment proves that RDPC-DTM is more efficient than DPC-DTM, especially when calculating eigenvalue of super large-scale matrix. Numerical experiment also demonstrates that RDPC-DTM has higher degree of parallelism and is more suitable for cluster or MPP parallel computer compared to DPC-DTM.