On Nov 24, 2:07 pm, "Nargol " <x...@x.com> wrote:
> Hello,
>
> I have a large sparse nonsquare linear system, I need to solve. Back slash operator(mldivide) does not help due to memory problems. I have also used lsqr, but it is does not return a good enough solution (back slash works better, but it is not useful for large problems).
It would be worth while to understand how LSQR is failing. If LSQR is
converging, but the solution is not acceptable because Axb is too
large, then you have a more fundamental problem, because LSQR is
solving the least squares problem and you're just not satisfied with a
least squares solution even though there isn't an exact solution to
your problem.
On the other hand, if LSQR is failing to converge (maximum iterations
exceeded), than this is an indication that the least squares problem
is illconditioned. In that case, some kind of regularization would
typically be the appropriate thing to try.
