Computing the Generalized Singular Value Decompositions
Ang S · Journal of Shanghai Normal University · 1989
A stable and fast soheme is desoribed to oompute the generalized singular value decomposition (GSVD) of a matrix pair (A, B). The new algorithm not only avoid the cross product matrix but also is more faster than Van Loan's. Our main stops are that matrices A and B are reduced to upper bidiagonal forms at the same Vitae and an iterative mashed for computing GSVD is constructed. At the first step, the modified algorithm is poposed to avoid the failure of the reduction caused by small elements.Numorical experiment is presented and the amount of multiplication is valued.