The QRPS algorithm: A generalization of the QR algorithm for the singular value decomposition of rectangular matrices (Q1070767)

For calculating the singular value decomposition \(U\Sigma V^ H=A\) of a real \(m\times n\)-matrix A, the following procedure is proposed: \(A_ 0=A\); for \(K\geq 1\), calculate the QR-decomposition \(A_{K-1}=Q_{K- 1}R_{K-1}\) and then the QR-decomposition \(R^ T_{K-1}=P_{K- 1}S_{K-1}\), and define \(A_ K=S^ T_{K-1}.\) It is shown that \(A_ KA^ T_ K=Q^ H_{K-1}A_{K-1}A^ T_{K- 1}Q_{K-1}\), so that the algorithm is equivalent to the QR-algorithm applied to \(A_ 1A^ T_ 1\). A proof of convergence is given, it seems however that it works only in the case \(m\leq n\), Rang A\(=m\). A geometrical interpretation is given and it is shown that for bidiagonal A this algorithm is related to the Golub-Reinsch algorithm. No numerical experiences are reported. Also the incorporation of shifts, crucial for the numerical success of the QR-algorithm, is not discussed.