Numerics of Gram-Schmidt orthogonalization (Q1319993)

The paper surveys the numerical properties of the classical and the modified Gram-Schmidt (MGS) orthogonalization procedures. The key observation is the numerical equivalence of the modified Gram-Schmidt procedure to the Householder QR factorization of the matrix \(A\) augmented by an \(n \times n\) zero matrix on top. This result is used to derive bounds on the loss of orthogonality in MGS. A backward-stable algorithm based on MGS is developed. The use of reorthogonalization and iteration is also investigated. Block Gram-Schmidt algorithms are presented.