Multiplicative perturbation analysis for QR factorizations (Q428984)

The QR-factorization of an \(m\) by \(n\) matrix \(A\) is the following: \(A = QR\), where \(Q\) has orthogonal columns and \(R\) is upper triangular with positive diagonal entries. This factorization is uniquely defined if \(m \geq n\) and \(A\) has full column rank. The question studied here is what happens to the QR factors if A is multiplicatively perturbed. The main results are given here in Chapter 3. If the perturbation is multiplication from the left by an \(m\)-by-\(m\) matrix \(D_L\) then the relative changes in the factors Q and R are bounded by a small constant multiple of the Frobenius norm of \(I-D_L\). If the perturbation is a multiplication from the right then the factors of the first order bounds contain the condition numbers of the scaled R-factor. Several improved versions of these results are given which are better if the perturbations are not so near to the identity. Two sets of numerical examples are shown.