The ABINIT Users Mailing List ( CLOSED )

[Anglade Pierre-Matthieu <anglade@gmail.com>]

PS : Sorry, I guess I wasn't clear at all in py previous post.
Know, I thing I understand clearly why your calculation can not
converge : the SCF problem has an infinite eigenvalue if you insist on
populating only one of the two crossing bands.
But unless one of the removed nonsymmorphic operation reduce the
overlap of those two bands to zero exactly, I can't understand how the
case with full symmetries can converge.

regards

PMA


On Thu, Mar 5, 2009 at 2:15 PM, Anglade Pierre-Matthieu
<anglade@gmail.com> wrote:
> Hi Matthieu
>
> (k = 1/3 1/3 0)
>
> I don't know how things are handled exactly within Abinit but:
> -  in the case you have  two equal eigenvalues the occupancy should be
> the same. Otherwise the polarisability of the system gets to infinity
> which leads to divergence of the SCF cycle.
> - when I run abinit with only k and all symmetries no problem happens
> and the convergence is fine although the two bands gets very different
> occupancies.
> - if I remove nonsymmorphic sym. operations I get the now usual divergence
> - When I want to check wether the system polarizability lead to the
> divergence I use iprcel and it compute eignevalues. They tend to
> infinity as expected ! And iprcel makes your system converge.
> - With more kpoint, "iprcel" is missing the problem and gets some low
> usual eigenvalues for graphene. We keep our usual SCF divergence.
> - I can check that using a metallic style occupancy (occopt 3)
> together with your small tsmear solves perfectly the problem since
> both bands gets the same occupancy.
>
>
> Is it possible that using the non-symmorphic sym. operation your two
> bands gives the same potential ? So that you get no difference in the
> coulomb potential from populating one or the other. While removing
> those symmetries, the two bands are no more equivalent and you should
> populate them at equal levels ? In such a case, the non-convergence
> with occopt 1 and nonsymmorphi 0 would not be a bug but something
> pretty natural.
>
> Conversely, if there is no such equivalent by symmetry operations, I
> can't understand how Abinit can reach convergence of your system with
> two bands at the exact same level, one empty and the other full.
>
>
>
>
>
> On Thu, Mar 5, 2009 at 10:39 AM, Zeila Zanolli
> <zeila.zanolli@uclouvain.be> wrote:
>> Yes, in my experiece it is exactly like that:  position of the kpoints in
>> the reciprocal lattice + symmetries.
>> Indeed, a similar situation happens with the SIESTA code, where the
>> symmetries are not 'recognized' as in ABINIT.
>> The convergence of whatever "something" will oscillate with the k-mesh and
>> to get it right one has to go to really high value of k-mesh.
>> The convergence of the scf cicle is more difficoult for k-meshes not
>> including the 1/3 1/3.....
>> To achieve the convergence of the scf cicle I'm using a very tiny mixing
>> factor in the pulay mixing scheme.
>> Zeila
>>
>> On 5 Mar 2009, at 00:37, matthieu verstraete wrote:
>>
>> Hmmm, this is very interesting! It looks like it depends on the kpoint
>> sampling including the K point or not (1/3 1/3 in recip space)... This is 
>> of
>> course the only point on the Fermi surface. However the residuals are not
>> only bad at K... So it's the combination of having the K point and not
>> having the appropriate symmetrizations, but I don't know how this plays in
>> to the conditionning of the convergence.
>>
>> Matthieu
>>
>> On Wed, Mar 4, 2009 at 6:08 PM, Anglade Pierre-Matthieu <anglade@gmail.com>
>> wrote:
>>>
>>> Hi Matthieu,
>>>
>>> Have you noticed that there is two case depending on your k-point mesh ?
>>> In the wrong case, obtained with meshes like 36 36 1, 12 12 1, 9 9 1,
>>> 6 6 1, the potential residual you get is a function of the number of
>>> k-point you use :The less k-point,  the bigger the vres2.
>>> Yet If one uses a single k-point or some other meshes (tried 211, 221,
>>> 441, 551, 771, 881, 35 35 1)  everythings converge fine.
>>>
>>> If one or a few k-point were getting a  "shift"  because of bad
>>> application of symmetries in the nonsymmop case,  don't you think it
>>> could lead to this behavior ?
>>>
>>> Have you already explored this idea ?
>>>
>>> Regards
>>>
>>> PMA
>>>
>>>
>>
>>
>>
>> --
>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> Dr. Matthieu Verstraete
>>
>> European Theoretical Spectroscopy Facility (ETSF)
>> Dpto. Fisica de Materiales,
>> U. del Pais Vasco,
>> Centro Joxe Mari Korta, Av. de Tolosa, 72,   Phone: +34-943018393
>> E-20018 Donostia-San Sebastian, Spain        Fax  : +34-943018390
>>
>> Mail : matthieu.jean.verstraete@gmail.com
>> http://www-users.york.ac.uk/~mjv500
>>
>> ---------------------------------------------------------------------------------------------
>> Dr. Zeila Zanolli
>> Université Catholique de Louvain (UCL)
>> Unité Physico-Chimie et de Physique des Matériaux (PCPM)
>> Place Croix du Sud, 1 (Boltzmann)
>> B-1348 Louvain-la-Neuve, Belgium
>> Phone: +32 (0)10 47 3501
>> Mobile: +32 (0)487 556699
>> Fax: +32 (0)10 47 3452
>> e-mail: zeila.zanolli@uclouvain.be
>> web: http://www.nano2hybrids.net/
>> ---------------------------------------------------------------------------------------------
>>
>>
>>
>>
>
>
>
> --
> Pierre-Matthieu Anglade
>



-- 
Pierre-Matthieu Anglade