A Krylov-Schur Method for Computing Singular Values
Open Access
Author:
Narthana, Amuthan Easwara
Area of Honors:
Computer Science
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Dr. Jesse Louis Barlow, Thesis Supervisor Dr. Jesse Louis Barlow, Thesis Honors Advisor Kamesh Madduri, Faculty Reader
Keywords:
Singular Value Decomposition Krylov-Schur Method Golub-Kahan-Lanczos bidiagonalization
Abstract:
Given a large sparse matrix, it is often useful to compute a small number of its singular values and vectors. This computation is based on Golub-Kahan-Lanczos bidiagonalization and the Krylov-Schur method for finding singular values and vectors of a matrix. The Krylov-Schur approach to finding a small number of singular values will be described, and various reorthogonalization strategies will be discussed. Numerical tests will compare variations of this method on several matrices, and the results will be analyzed.
