Block Golub-Kahan-Lanczos algorithms for quaternion singular value decomposition
摘要
In this paper, we investigate the singular value decomposition for quaternion matrices. Building on the quaternion Lanczos algorithm and the block Golub-Kahan-Lanczos algorithm, we propose a quaternion block Golub-Kahan-Lanczos algorithm and prove its convergence. By applying the thick restart strategy, we develop a thick-restart quaternion block Golub-Kahan-Lanczos algorithm to accelerate its performance. Numerical experiments demonstrate that the proposed algorithm outperforms existing methods in both computation time and accuracy.