Singular value decomposition (SVD) is a general linear-algebraic tool for dealing with singular or near-singular matrices (Noble & Daniel 1977). It is a generalization of eigenfunction analysis to systems split into two domains, such as the sky and the u,v planes. SVD determines the form of S which has minimum length by discarding the singular or near-singular terms in the formal algebraic solution. Briggs (1995) briefly discussed the use of SVD for deconvolving VLBA data; he showed that for about 3000 pixel intensities and visibilities, SVD produces an image whose quality is subjectively on a par with that of `CLEAN', but at the cost of large memory use and of long running time (on IBM RS/6000-580 workstations).
1996 November 4