Adapting Craig's Method for Least-Squares Problems Home LS-Craig is based on the Golub-Kahan bidiagonalization process, and the iterates it produces are convex combinations of the iterates produced by Craig's method and the algorithm LSQR. It is aimed at minimizing the 2-norm error at every iteration. Although Craig's method minimizes the 2-norm error for consistent problems, it does not converge when run on inconsistent systems. CRAIG+ patches this shortcoming by guaranteeing convergence whenever \(\lambda > 0\)or a nontrivial lower bound to the smallest singular value of \(A\) is known. Downloads: |
[Last modified: 8 Apr 2019]
|