A differential equation approach to the singular value decomposition of bidiagonal matrices (Q1080916)

One of the most important decompositions in matrix computations is the singular value decomposition (SVD). This paper develops a continuous approximation to the SVD of bidiagonal matrices. It turns out that such an approach is totally feasible and can be fully expressed as an autonomous ordinary differential system. The derivation of this homogeneous cubic differential system is made. It is related to the Golub-Kahan SVD algorithm almost in the same way as the Toda lattice is related to the QR algorithm. Conclusions on its asymptotic behavior are drawn from the knowledge of the Toda lattice.