[Wien] Minimization: acceptable approximations

[Peter Blaha]

> I'm running some fairly large surface minimizations, and simultaneously
> looking for any tweaks that can be used in the port minimizer to improve
> its performance. For a large minimization one obviously wants to reduce
> the accuracy of the calculation (for speed) -- then step it up close to
> the minimum. What I'm currently using is:
>
> 	1) Small number of k-points (TETRA mode)
> 	2) Small RKMAX (6)
> 	3) .fulldiag (the extra CPU time is minimal with mkl6.1)
> 	4) -fc 1.0 (larger seems to be too inconsistent)

If I'm very far from the minimum (forces above 50) I test my convergence
(on the same geometry) and accept force "errors" up to 5-10 mRy/bohr. Thus
I use often an even samller RKMAX (5), # of k-points adjusted to the available
nodes for parallel calculations, (so that I do just 1-k point/node) and do
only  -fc 5  (to save in #of scf cycles).

Once the forces go down to 10, I stop the minimization and restart with
better parameters (eventually doing it a 3rd time once I'm close to forces
<2 mRy/bohr).

(So far I would have always "deleted" the history,
i.e. rm case.tmpM, because of my "background" with "BROYD" (and also the scf
cycle convergence), that when one changes anything which will lead to a
different minimum, one should remove the "history", otherwise BROYD got
"crazy".)



> improving the accuracy of the force calculation (R0 ?), and what
> parameters effect this. (It looks to me like in some cases the forces
> could be slightly inconsistent with the energy.)

I've also seen such inconsistencies between forces and E-tot from time to
time (E-tot min does not agree with F=0). Besides the above parameters,
sometimes a smaller R0 value is necesary, eventually one has also to
change RMT rations,...

Sorry, this is a little "diffuse", but for me it did not happen very often.

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