APPROXIMATE EIGENSYSTEMS OF LARGE COVARIANCE MATRICES.
Abstract
The paper presents a method for obtaining approximate solutions of the algebraic eigenvalue problem for hermitian matrices with a substantial reduction in computation time. The approach is to apply a standard eigenvalue routine to submatrices of the original matrix and use the results to transform the original matrix into one of much lower dimension having eigenvalues approximately equal to the largest eigenvalues of the original matrix. A method of information compression by intrinsic analysis is described. The eigensystem approximation is applied to the intrinsic analysis computations, and explicit formulas are derived for the additional error introduced by the approximation. Results of two specific applications are given, along with tables of reductions in computation time realized using the approximation. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1968
- Accession Number
- AD0678172
Entities
People
- Robert B. Roper