A Blocked Jacobi Method for the Symmetric Eigenproblem
Abstract
A block matrix generalization of the Jacobi rotation method for computing the eigen decomposition of a symmetric matrix is presented. This Blocked Classical Jacobi (BCJ) algorithm selects for block rotation at each step the off-diagonal block(s) of largest mass. The BCJ algorithm exhibits substantially shorter runtimes than another blocked Jacobi method, but is slower than a scalar Jacobi algorithm on random matrices with i.i.d. uniform elements. A probabilistic analysis of the BCJ selection method is presented. Timings and other data are presented from experiments on random matrices.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1989
- Accession Number
- ADA206553
Entities
People
- David E. Foulser
Organizations
- Yale University