A new vector field method for eigen-decomposition of symmetric matrices (Q2474357): Difference between revisions

The eigenvalue problem is one of the most researched problems in numerical linear algebra, and over the years several extremely efficient algorithms have been devised, like the QR method, divide-and-conquer approaches, and the Lanczos algorithm. The vector field approach addresses the eigenvalue computation as an optimization problem. Like LP interior point methods, it attempts to construct vector fields on the unit sphere which, although not locally optimal, result in reduced computation. In this paper the authors describe the vector field approach, establish convergence results and complexity bounds and compare the performance of recursive vector field algorithms with that of the power method. The paper concludes with a discussion of the computational results.