Relations between SVD and GSVD of discrete regularization problems in standard and general form (Q2638724): Difference between revisions

An efficient numerical algorithm for transforming a discrete regularization problem in general form into a problem in standard form by \textit{L. Eldén}'s method [BIT 17, 134-145 (1977; Zbl 0362.65105)] is described. The advantage of doing this is that there exist very efficient methods for solving the standard-form problem such as bidiagonalization and truncated singular value decomposition (SVD). A simple relationship between the SVD associated with the standard-form problem and the generalized singular value decomposition (GSVD) associated with the general problem is derived. The accuracy of the GSVD when computed by means of the transformation to standard form is discussed. It is shown that the accuracy of GSVD computed in this manner depends on the condition number of the regularization matrix and is therefore accurate only if it is well conditioned.