A new iterative scheme for obtaining eigenvectors of large, real- symmetric matrices (Q1122939)

This work is directed towards obtaining an algorithm for matrix eigenvalues that arise from the discretization of self-adjoint partial differential equations. The most time-consuming parts of the proposed algorithm are a matrix multiplication and a Gauss-Seidel relaxation step which are performed on each iteration. The described method represents a reasonable compromise between simplicity, efficiency and robustness; it can handle matrices that exceed the physical memory size without serious performance degradation.