Backward error analysis of the Lanczos bidiagonalization with reorthogonalization
Abstract
The -step Lanczos bidiagonalization reduces a matrix into a bidiagonal form while generates two orthonormal matrices and . However, any practical implementation of the algorithm suffers from loss of orthogonality of and due to the presence of rounding errors, and several reorthogonalization strategies are proposed to maintain some level of orthogonality. In this paper, by writing various reorthogonalization strategies in a general form we make a backward error analysis of the Lanczos bidiagonalization with reorthogonalization (LBRO). Our results show that the computed by the -step LBRO of with starting vector is the exact one generated by the -step Lanczos bidiagonalization of with starting vector (denoted by LB()), where the 2-norm of perturbation vector/matrix and depend on the roundoff unit and orthogonality levels of and . The results also show that the 2-norm of and are controlled by the orthogonality levels of and , respectively, where and are the two orthonormal matrices generated by the -step LB() in exact arithmetic. Thus the -step LBRO is mixed forward-backward stable as long as the orthogonality of and are good enough. We use this result to investigate the backward stability of LBRO based SVD computation algorithm and LSQR algorithm. Numerical experiments are made to confirm our results.
Cite
@article{arxiv.2210.10297,
title = {Backward error analysis of the Lanczos bidiagonalization with reorthogonalization},
author = {Haibo Li and Guangming Tan and Tong Zhao},
journal= {arXiv preprint arXiv:2210.10297},
year = {2022}
}