The ABINIT Users Mailing List ( CLOSED )

[Vincent Chevrier <vincent.chevrier@dahn.phys.dal.ca>]

Thanks for your replies!
I don't have administrative access to the cluster and can't conveniently
run the parallel tests. However, the administrator has since recompiled
abinip with "less agressive optimization flags", the recompile was
prompted by problems with some PAW jobs, and should not have changed
anything for me. However an identical job with ngkpt 10 10 10 now returns
values which are consistent.

The occopt = 4 with tsmear = 0.04 now also works, and convergence in ecut
and kpt are indeed independent as expected.

It's great that it now works, but a little worrisome that it is not clear
to me why that is. Is it possible to get different results for identical
jobs?

Finally, with respect to the ghost states, I am using the LDA FHI (TM)
pseudopotentials from the website. Looking back to the .out file I shared
I can see some abnormally low eigenvalues at certain kpts. If I no longer
see this behaviour can I rule out the ghost state?

Thanks again for all your help!

Vincent

> Hello Vincent,
>
> I had a close look at your output file, and it seems that you are facing a
> typical case of a ghost state. If you look at the eigenvalues output, you
> will see for certain k-points abnormally low values which are a sign of a
> ghost state. Moreover, your scf convergences are very fast, 7 or 8
> iterations, this is also a sign of a ghost state. I suggest that you check
> your pseudopotentials.
>
> Best wishes,
>
> Michel
>
>
> Le 28/05/07 06:28, « Anglade Pierre-Matthieu » <anglade@gmail.com> a
> écrit :
>
>> Hi,
>>
>> Usualy convergence in ecut is considered independant from convergence in
>> kpt.
>> You can trust your 2 2 2 grid about that I think. It looks like there is
>> a
>> problem with the parallel version of abinit you use. Does it stand the
>> parallel tests correctly ?
>> By the way have you tryed to see what's happening if you treat your
>> supercell
>> has metallic (occopt 3/4.. and tsmear)?
>> What do you mean when you say that Abinit crashes with occopt parameters
>> different from zero ?
>> regards
>>
>> Pierre-Matthieu Anglade
>>
>> On 5/28/07, vincent.chevrier@dahn.phys.dal.ca <
>> vincent.chevrier@dahn.phys.dal.ca
>> <mailto:vincent.chevrier@dahn.phys.dal.ca> >
>> wrote:
>>> Hi,
>>> I've recently begun using Abinit and have gone through the basic
>>> tutorials.
>>> I'm trying to run an energy calculation of Li15Si4. The primitive cell
>>> is
>>> sizeable (30 Li, 8 Si) and I'm having difficulty establishing the
>>> appropriate
>>> ecut and ngkpt (or nkpt).
>>>
>>> Here's a table of energies (in Ha). If anyone could tell me why the
>>> energy
>>> starts varying so wildly as the k points are increased or if anyone has
>>> any
>>> recommendations as to which parameters I should use I would really
>>> appreciate
>>> it!
>>>
>>> ecut\ngkpt (nkpt)
>>>
>>>         2 2 2 (2)       4 4 4 (6)      10 10 10 (44)   12 12 12 (68)
>>> 30      -42.500550      -42.500085      -56.940709      -43.130515
>>> 35      -42.500777      -42.500314      -61.734328       -42.554030
>>> 40      -42.500897      -42.500449      -86.955397      -149.902062
>>> 45      -42.500918      -42.500504      -42.604657      -260.973168
>>> 50      -42.500920      -42.500457      -42.566132      -158.776883
>>> 55      -42.500928      -42.500670      -44.279878      -48.285954
>>> 60      -42.500936      -42.500489      -75.221022      -43.411153
>>>
>>> The .in, .out and log files, of the ngkpt 12 12 12 case are at the
>>> following
>>> address (if you're really interested!). The other energies were
>>> generated
>>> with the exact same input files with only the ngkpt changed.
>>> http://myweb.dal.ca/chevrier/abinit/ecut_LDA_22.in
>>> http://myweb.dal.ca/chevrier/abinit/ecut_LDA_22.out
>>> <http://myweb.dal.ca/chevrier/abinit/ecut_LDA_22.out>
>>> http://myweb.dal.ca/chevrier/abinit/log_Li15Si4_ecut_LDA_22
>>>
>>> I'm running this with the parallel version.
>>> Thanks!
>>>
>>> Vincent Chevrier
>>> PhD Candidate
>>> Physics Department
>>> Dalhousie University
>>
>>
>
>
> --
> ***************************************************************
>   Michel Cote                          tel: +1 (514) 343-5628
>   Professeur agrégé                   fax: +1 (514) 343-2071
>   Département de physique
>   Université de Montréal
>   C.P. 6128, succ. Centre-ville
>   Montréal (Québec) H3C 3J7          Michel.Cote@umontreal.ca
>   Canada                   www.phys.umontreal.ca/~michel_cote
> ***************************************************************
>
>
>