[Wien] thebest rkmax
Stefaan Cottenier
Stefaan.Cottenier at UGent.be
Tue Sep 6 15:08:42 CEST 2011
> I work on b-Ga2O3 and want find the best rkmax for later computation
> I read Density Functional Theory andthe Family of (L)APW-methods.
> I set rmt automatically 3 percent and I find the best kpoint=1500 and
> for finding RKmax I draw ENE and EFG DIAGRAM WITH THREE METHOD: LAPW
> , APW+lo( only for d orbital) ,APW+lo
It's always nice to see someone doing decent convergence testing!
> EFG and ENE diagram of LAPW and all of EFG diagram converge 8.5--9
> but EFG diagram of LAPW and APW+lo(d) first drop monotonically until
> rkmax=8 then start going up and then in 8.5-9 coverge again
> why?
> can I choose the best rkmax=8.5 or I must choose rkmax in 6-8?
> is the best method APW+lo in this case?why?
> please guid me
I've put graphs of your data here, for easier interpretation:
http://dl.dropbox.com/u/10829484/converge.pdf
1) APW+lo is in most cases just fine. Unless you have particular reasons
not to use it, you may stick to this choice.
2) In your energy graph, you see that LAPW behaves somewhat differently
from the APW+lo variants. That is expected. In your EFG graph, however,
two curves are nearly identical while the third one (APW+lo) is
different. And this is the slowly converging one. Strange. Didn't you
mix up some data? I would expect the green curve to be LAPW.
3) Provided the green curve is indeed LAPW, you see that even RKMAX=6
gives you a fair value for the EFG (within 0.1 of the most accurate value).
4) You are most likely not interested in EFGs. Are there forces in your
cell? Plotting forces as a function of rkmax gives you another clue (and
probably one that is easier to interpret) on your convergence.
5) The total energy will always go down with rkmax. What you really want
to know is whether the energy *difference* of two calculations is
converged. Your could repeat these tests for a cell that has e.g. a 2%
smaller volume, and plot the difference of the two total energies as a
function of rkmax. Compare this with the result of 4).
Stefaan
More information about the Wien
mailing list