[Wien] Convergence: fixed-spin-moment or not?

[Peter Blaha]

> A simple question. I'm running an optimization where I know the magnetic
> moment is 4, a big system so its taking days (and days). Does anyone know
> if the convergence (in a run??_lapw cycle) is faster with runsp_lapw or
> runfsm_lapw -m 4. (The result with runsp_lapw is always 4+/-delta.)

For an insulator there should not be much difference between
runsp and runfsm.

For a metallic system, runfsm should be faster, sich a certain solution is
forced and the degrees of freedom are reduced. However, you might NOT
obtain the LDA groundstate (because LDA may favour another solution than
experiment).
There are a few systems where several magnetic solutions are quite close
in energy and a "free"  runsp calculation might not (or hardly) converge,
while with FSM you can fix the moment and converge this particular solution.

PS: Often convergence problems can be overcome with a different (better)
k-mesh or using some smearing (TEMP with 0.010; the larger this number,
the better the convergence, but maybe you loose (decrease) the moment
artificially; while a smaller  number has less effect on the moment (but
maybe also on convergence improvement).

                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------