Stability Analysis of a Householder-Based Algorithm for Downdating the CholeskyFactorization

摘要

In this paper a new algorithm for downdating the Cholesky factorization ispresented and analyzed. The algorithm is based on Householder transformations. It is less expensive than algorithms based on Givens rotations if more than one row is to be removed from the Cholesky product. Rounding error analysis shows that the new method is as stable as the rotation-based methods for solving this problem. Numerical methods, recursive least squares, beamforming.