[Pw_forum] problem with convergence during SCF-procedure

[Alexej Mazheika]

On Mon, Oct 20, 2008 at 1:41 AM, Agostino Migliore
<amigliore at cmm.upenn.edu>wrote:

> Hello Alex
>
> Since, you have transition metals in your system, it could help you to
> introduce an electronic smearing and then progressively  reduce it until
> you get the solution without any orbital fractional occupations.
> Then, from the attached file I saw that you are using  mixing_beta =
> 0.7D0. Did you try something like mixing_beta = 0.5 or similar (perhaps
> you already did, but I don't know) and also mixing_mode = 'plain'
> (although I know that the local_TF mode may be good for highly
> inhomogeneous systems).
> Finally, your system includes Ti atoms, which have d electrons. Have you
> tried to use the Hubbard U correction scheme implemented in pwscf? Also
> irrespective of your convergence problem, it could provide you with a
> better description of the electronic structure (indeed, I don't have
> direct experience on Ti, but I have on Fe, and the Hubbard U correction
> term works well).
> I hope something I said can help you.
>
> Best regards,
> Agostino Migliore
> CMM, Department of Chemistry, Philadelphia
>
>
> Quoting Alexej Mazheika <alex.mazheika at gmail.com>:
>
> > Dear all
> >
> > I perform the calculations of surface with adsorbed cluster Ag4/TiO2 and
> > have a problem with convergence during SCF-procedure. It does not
> converge.
> > In order to eliminate it I tried the shown methods in the user_guide: to
> > decrease mixing beta, to increase mixing ndim, to increase ecutwfc and
> > ecutrho. But it didn't help me. Is there any other methods to solve such
> > problem? The input file is here attached.
> >
> > Best regards
> > Alex Mazheika, Research Institute for phys. and chem. problems, Minsk,
> > Belarus
> >
>
>
> I tried to use mixing_beta 0.3, mixing_ndim 10, mixing_mode = 'local-TF'
and I changed the diagonalization to 'diis'. I performed the calculations
with such parameters: occupations = 'smearing', degauss = 3.7D-3, smearing =
'gaussian'. And then the covergance was achieved.
Thank you for recommendations
Best regards
Alex Mazheika, Research Institute for phys. and chem. problems, Minsk,
Belarus
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20081022/a9f08e76/attachment.html>