A QR-method for computing the singular values via semiseparable matrices (Q706231)

A new iterative algorithm for computing the singular value decomposition of a real matrix is proposed. In the first phase of the algorithm the matrix is reduced into an upper triangular semiseparable matrix by using orthogonal transformations. In the second phase an iterative method is applied to the matrix obtained in the first phase in order to reduce it into a block diagonal matrix. It is shown that each iteration step is equivalent to one iteration step of the QR-method. An advantage of the algorithm in the first phase is that after a few steps of this algorithm the largest singular values are already computed with high precision. Numerical experiments show that the new method can compete with the standard method.