[Pw_forum] Improve convergence, slab and electric field.

[joaquin peralta]

After many time, finally I ran some simulations (cluster problems), the
'jumps' in the energy convergence remain, but reducing the k-points and
cutoffs, was more fast.

The more fast convergence was (and without many jumps) :

-increasing the mixing_beta from 0.1 to 0.3
-NO use local-TF in the mixing_mode
-NO change the mixing_ndim
-Reduce the vaccum.
-Reduce cutoff energy.

When I only change the mixing_mode and mixing_beta from this previous data
(even reduce from 0.1 to 0.09) the convergence problem remain and increase.
I hope that with this, the problem disappear.

Probably always was the vaccum, and the default values (except mixing_beta)
was fine.

After how many SCF cycles these "fluctuations" appear?


Usually each 10 steps. Related with 0.1 of mixing_beta??

And thank you very much for all the answers, and any other suggestion is
welcome.

Best Regards

Joaquin

On Fri, Sep 3, 2010 at 4:05 AM, Tone Kokalj <tone.kokalj at ijs.si> wrote:

> On Thu, 2010-09-02 at 08:38 -0500, joaquin peralta wrote:
> > Thanks, i obtain more fast convergence for 1V/A, but ...
> >
> > If I use the previous results from 1V/A in order to calculate an
> > increase in the Field i obtain fluctuation in the energy.
> >
> >      total energy              =    -261.46819272 Ry
> >      total energy              =    -261.46815798 Ry
> >      total energy              =    -261.46816720 Ry
> >      total energy              =    -261.46260676 Ry
> >      total energy              =    -261.37365516 Ry
> >      total energy              =    -261.46820680 Ry
> >      total energy              =    -261.46821935 Ry
> >      total energy              =    -261.46807352 Ry
> >      total energy              =    -261.46825528 Ry
> >      total energy              =    -261.46820900 Ry
> >      total energy              =    -261.46803348 Ry
> >
> > for my case, is recommendable reduce the precision by default of the
> > energy convergence from 1D-4 to 1D-3?
>
> To what energy threshold are you referring: conv_thr (in &electrons) or
> etot_conv_thr (in &control). According to the value of 1d-3/1d-4 I would
> assume the latter; note however that the latter has nothing to do with
> the SCF convergence, as it is an optimization threshold criteria for
> geometry optimizer.
>
> If you use 1d-4 for the conv_thr, then the value is much too high, it
> should be 1d-6 or, even better, lower than that.
>
> After how many SCF cycles these "fluctuations" appear?
>
> You may try to reduce mixing_beta to 5%, ...
>
> The following considerations will speed-up your calculations, hence you
> will be able to do more SCF convergence tests: from the previous emails
> of yours, I notice that you use over 50 Angs of vacuum, which is HUGE (I
> would assume that about 25 Angs should be sufficient; in this case there
> will be more than 10 Angs spacing from the slab to the dipole-layer). I
> also noticed that ecutwfc/ecutrho of 30/500 ry is a very large number
> for Aluminum.
>
> Regards, Tone
> --
> Anton Kokalj
> J. Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia
> (tel: +386-1-477-3523 // fax:+386-1-477-3822)
>
> Please, if possible, avoid sending me Word or PowerPoint attachments.
> See:  http://www.gnu.org/philosophy/no-word-attachments.html
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>



-- 
----------------------------------------------------
Group of NanoMaterials
----------------------------------------------------
http://www.gnm.cl
----------------------------------------------------
Joaquín Andrés Peralta Camposano
----------------------------------------------------
http://www.lpmd.cl/jperalta

In a world without frontiers,
who needs Gates and Win.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100924/132c0d10/attachment.html>