On the Work Needed to Factor a Symmetric Positive Definite Matrix.
Abstract
When comparing different row ordering strategies, the measure often used is the fill-in, or the number of additional non-zeroes elements. This report proposes an another measure: the number of arithmetic operations necessary to factor the matrix. Two classical ordering strategies: Minimum Degree and Minimum Local Fill-in are compared with respect to this measure and Minimum Local Fill-in usually produces better results than Minimum Degree. Also, an application is presented where the number of arithmetic operations may be more interesting measure of performance is presented: Karmarkar's Linear Programming Algorithm. Keywords: Heuristic methods; Numerical linear algebra; Matrix computations; Graph theory. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1987
- Accession Number
- ADA185781
Entities
People
- Marcio De Carvalho
Organizations
- University of California, Berkeley