A regularizing Lanczos iteration method for underdetermined linear systems (Q1971820): Difference between revisions

Underdetermined linear systems \(Ax=b\) lead to linear equations of the form \(Bx=b\) where \(B=A^T A\) and \(b\) is redefined. If \(B\) is an ill conditioned matrix, a possible regularization is \[ \left(B+{1\over \beta} I\right) y=b. \] Of course, the result heavily depends on the choice of the regularization parameter \(\beta\). If the data are contaminated by some error, the value of \(\beta\) may be chosen such that the regularization term \(\beta^{-1} y\) has the same order of magnitude as the contamination. Its determination in a Lanczos type iteration is in the center of the paper.