A Lanczos-like reduction of symmetric structured matrices to semiseparable form (Q2504981)

A Lanczos-like algorithm is proposed to transform a symmetric matrix to a similar semiseparable one by making use of a product of the original matrix times a vector at each step. The algorithm can be used efficiently if the original matrix is sparse or structured. All the techniques developed to prevent the loss of orthogonality in the Krylov basis can be used in this case as well. An extension for reducing a matrix to an upper semiseparable form to compute the singular value decomposition is shown to be straightforward. Two numerical examples are given to demostrate the effectiveness of the proposed method.