Approximating dominant singular triplets of large sparse matrices via modified moments (Q2564505): Difference between revisions

This is an interesting paper on determining a few of the largest singular values and the associated singular vectors of large sparse matrices. The authors relate the singular value decomposition to two equivalence eigenvalue decompositions, and discuss the Chebyshev semi-iterative method and the method of modified moments for eigenvalue problems. By exploiting deflation techniques and modified moments in conjunction with the Chebyshev semi-iterative method, they propose a scheme for approximating eigenpairs of the equivalent sparse eigensystems. Implementation of the proposed algorithm is described in detail for parallel computations. Comparisons with other methods and performance of the algorithm on a network of workstations and on a massively-parallel computer system are discussed.