[Wien] mixing vs. lapw2 CPU time

[Peter Blaha]

> Q: Why does changing the value of the broyd mixing parameter make a big
> difference to the CPU time used by lapw2?
>
> I have two identical cases, except that one has BROYD=0.4 and doesn't
> convere, while the other has BROYD=0.1 and does converge.
>
> The curious thing is that with the smaller mixing parameter lapw2 takes
> over 2 times longer, and is the most time-consuming step of each SCF
> cycle.
>
> I have 10 atoms (6 non-equiv.) in a long (38 bohr) and skinny (5 x 5
> bohr^2) primitive unit cell.

No, the mixing parameter does NOT have any influence on the runtime of
lapw2.

Most likely the different times come from   TOT  vs. FOR in the two case.in2(c)
files.
(Once you converge with  run_lapw -fc 1     FOR will be put into case.in2
in the last iteration (and lapw2 takes longer).
However, the next time you restart an scf cycle, you should do it with

run_lapw -I -fc 1

and the -I changes FOR back to TOT, making lapw2 fast again.

Another possible reason would be the presence (lack) of case.recprlist, which
usually is created just at the first call of lapw2 and kept.

                                      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/
--------------------------------------------------------------------------