Computing singular value decompositions of parameterized matrices with total nonpositivity to high relative accuracy (Q1704844)

The authors are concerned with accurately computing singular value decompositions (SVDs) of matrices. They present a new algorithm to compute SVDs of nonsingular totally nonpositive matrices to high relative accuracy. The new algorithm is developed by taking into account the property that there are \(n\times n\) independent parameters for any nonsingular totally nonpositive matrix of order \(n\) which determine all its singular values and singular vectors to high relative accuracy. Numerical results are presented to demonstrate the high relative accuracy of the proposed algorithm.