what is the main reason of better convergence rate and more speed of LSMR in comparison to LSQR algorithm for solving linear least squares equations?
contact me via "firstname.lastname@example.org".
I am trying to use your code for the very large overdetermined linear system, which needs to be repeatedly solved many times with a different right hand side. Hence, I wonder if warm-starting is possible (the naive extension with the initial guess renders the algorithm unstable)?
This is a nice short code, but it covers only a very special case and does not tell which one exactly.
I recommend spincalc() from Jonathan Fuller. It is much more powerfull!