Please note
that these are MATLAB files. They do not
include mex
files. Therefore the speed is misleading when compared to other
packages
that are written in compiled languages. In addition, these files are
under constant improvement. The CPU times are significantly better
than those reported in the above cited Research Report.

The algorithm is based on an inexact Gauss-Newton (GN) method using
preconditioned conjugated gradients (PCG) to solve the least squares
problem.

Important features: q-quadratic convergence and high
accuracy.

We use the PCG-type
algorithm LSQR written (in MATLAB) by Michael Saunders and available on his web page.