[Pw_forum] problem with TPSS (Tao-Perdew-Staroverov-Scuseria) meta-GGA for silicon

Ari P Seitsonen ari.p.seitsonen at iki.fi
Fri Nov 28 11:56:11 CET 2008

Dear Rafael,

   Adding to Axel's comment; you might experience the problem of 
"non-spherical" atom in Si since in the free atom you have two p-electrons 
to put on three degenerate (spin-up) states. Thus there will be a fight 
(= usually no convergence) which one of p_x, p_y and p_z will get them if 
tried straightforwardly. What one can try is to modify the cell slightly 
making it non-cubic, eg tetragonal and then seeing if the calculation 
converges. Additionally you might try adding more mixing of the previous 
charge density.

   Summa summarum, this problem might not be due to the implemenation of 
TPSS but a "pathodological example".

     Greetings from Paris,


PS This topic keeps on popping up on the mailing lists of different codes
   quite regularly. Apparently the people do not know that the topic was
   discussed already decades ago. ;)

"Energies of atoms with nonspherical charge densities calculated with
     nonlocal density-functional theory"
Frank W Kutzler and G S Painter
Phys Rev Lett 59, 1285 (1987)
DOI: 10.1103/PhysRevLett.59.1285

On Thu, 27 Nov 2008, Axel Kohlmeyer wrote:

> rafael,
> please keep in mind, that a single atom in a large box
> is a bit of a pathological case, since you tend to pick up
> all the noise from the "vacuum" area, which is most likely
> the source of your numerical instabilities. if it is just for
> testing, you'll probably get better results with a bulk crystal
> system (and it'll be faster, too).
> cheers,
>   axel.
> On 11/27/08, Rafael Dias Menezes <rdmeneze at yahoo.com.br> wrote:
>> On Nov 23, 2008 at 22:04 , Paolo Giannozzi wrote:
>>> TPSS is quite nasty from a numerical point of view. Try to use lapack
>>> from sources instead of optimized libraries. On my Macintosh I get
>>> this problem when testing tests/metaGGA.in with atlas loaded;
>>> it disappears if I link lapack instead.
>> Thank you, Paolo for the trick.
>> Really this error disappear but the calculation to silicon atom in a big box
>> don't converge. I did 100 interactions and the convergency isn't reach.
>> Can someone get me a direction in this simple calculus? Maybe I'm missing
>> the setup of some input variable do pwscf.
>> thanks,
>> Rafael Dias
>>       Veja quais são os assuntos do momento no Yahoo! +Buscados
>> http://br.maisbuscados.yahoo.com
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum

   Ari P Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/
   IMPMC, CNRS & Université Pierre et Marie Curie
   Tel: +33-1-4427 7542, Fax: +33-1-4427 3785, GSM: +33-6-6736 3820

More information about the users mailing list