Regularization, GSVD and truncated GSVD (Q1824353): Difference between revisions

The linear least-squares problem \(\min \| Ax-b\|_ 2\) for very ill-conditioned matrices A is considered. First of all, the generalized singular value decomposition (GSVD) and its truncated form (TGSVD) are outlined, and then it is explained why the Tikhonov regularization and the TGSVD approach produce smooth solutions. In the sequel a perturbation analysis for the two methods is presented in the most important case of a perturbed right-hand side and the proper choice of the regularization and truncation parameters is discussed. Finally, the author shows an efficient procedure of solution of the given problem by means of a transformation suggested by \textit{L. Eldén} [ibid. 17, 134-145 (1977; Zbl 0362.65105)] to avoid the explicit computation of the GSVD.