[Pw_forum] relax of 216 atoms (cubic structure)

[Paolo Giannozzi]

Increase smearing: 0.0005 is very small. Don't use diagonalization='cg'
unless you have a good reason to.

Paolo

On Wed, Aug 5, 2015 at 2:42 PM, Yevgen Melikhov <melikhovyy at yahoo.com>
wrote:

> Dear all users of QE,
>
> I am trying to relax GaAs doped with a little bit of Mn and with some
> point defects and, of course, I have some problems in doing that.
>
> In order to construct my initial structure I took cubic cell GaAs with 8
> atoms and translated it: (I) 3x2x2 to produce a structure with 96 atoms and
> (II) 3x3x3 to produce a structure with 216 atoms. After that I substitute
> Ga with Mn (where necessary) and introduce point defects (where necessary).
>
> The structure with 96 atoms (sometimes 94 or 95 depending on the problem)
> with GAMMA point relaxes quite well (please see the attached input file).
>
> However, I could not relax the structure with 216 atoms (sometimes 214 or
> 215) (please see the attached input and output files).
>
> If I choose GAMMA point only (that is, of course, what I wanted to speed
> up calculations), the error is:
> %%%%%%%%%%%%%%%%%%%%%%%
> Error in routine efermig (1):
> Internal error, cannot bracket Ef
> %%%%%%%%%%%%%%%%%%%%%%%
> which usually happens in the SCF iteration #2 of the first BFGS cycle.
> After checking the forum and playing with the parameters I decided to
> increase the number of k-point.
>
> If I choose more k-points (as in the attached file), the SCF cannot
> converge (usually this happens within the second or sometimes third BFGS
> cycle) for 70-100 cycles and, due to time limit on the cluster, the job is
> not finished. I tried to restart the job but the problem, in my opinion, is
> not in having not enough time to finish SCF calculations but rather in
> oscillations of the scf accuracy (please see the extracted parts of the
> output file). The first BFGS cycle usually is converged within 10-20 SCF
> cycles.
>
> Is my assumption that I need to change conv_thr to lager values correct:
> the number of atoms doubled (96 vs 216), so is the energy (-10300 Ry vs
> -23000 Ry), and therefore, at least I need to double the conv_thr
> (currently it is at its default value 1.0D-6)? For such a big system may I
> even increase it further 1.0D-4?
>
> Thank you in advance for any hints/comments!
>
> Yevgen Melikhov
>
> Lab of X-ray and Electron Microscopy Research
> Institute of Physics, PAS
> Warsaw, Poland
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>



-- 
Paolo Giannozzi, Dept. Chemistry&Physics&Environment,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20150815/c360b0d7/attachment.html>