The ABINIT Users Mailing List ( CLOSED )

[Luigi Genovese <luigi.genovese@gmail.com>]

Hi Everybody,

I examined the question of the water with the standalone version of BigDFT.
The problem which I would like to address is related to the fact that
with the direct minimization scheme the occupation number are always
given in the semiconductor-like scheme.
This may cause problems with open shell or highly degenerate systems,
in which the electrons tend to distribute on equally energetic levels.
For this reason, in the BigDFT standalone version there is the
possibility to specify at hand the occupation number of the orbitals.
In the ABINIT version, presumably this functionality can be also
imported from standard ABINIT input files.

I did three runs for a sample water molecule (not fully relaxed indeed):
1) Non spin polarised run (The one I suppose PMA did)
2) Spin-averaged run (a collinear spin treatment with the last
orbitals populated with half electrons)
3) Spin polarised run (polarization 1)

What happened is that run 1) and 2) gave the same results, while run
3) is lower in energy (about 0.8 eV).
This means that the semiconductor-like population which is imposed in
a non-spin polarised run forces the solution to be higher that the
ground state, or, alternatively, that the ground state of the charged
system becomes spin-polarised.

Of course, for a non-spin polarised run there is no way, also
conceptually, to come out from the fake ground state.
Such problem would persist also for a spin-polarised run with the
wrong choice of the occupation number (the case 2)).
Fortunately, for this case the semiconductor-like population of BigDFT
seems to be the good answer.

Let me know your impressions

Ciao

Luigi


On Tue, Mar 10, 2009 at 10:05 PM, Anglade Pierre-Matthieu
<anglade@gmail.com> wrote:
> Dear Thierry (and Damien),
>
> I'm jumping on the wagon of charged system and bigdft : Is there some
> variables usable from Abinit to help Bigdft converge a charged system
> with open shells. For instance suppose you remove an electron from
> water or whatever. My experience is that it takes forever to converge
> with bigdft 1.2.0. I was in fact not able to reach the true ground
> state of the system. Eventhough the calculation seemed converged
> (1e-13 on WF). I guess there is some options to deal with that on the
> wavelet code. But I didn't found them.
>
> regards
>
> PMA
>
> On Tue, Mar 10, 2009 at 5:58 PM, Thierry Deutsch <thierry.deutsch@cea.fr> 
> wrote:
>> Difficult question...
>>
>> Assume that we can calculate exactly the ground state of O + H far away 
>> each
>> other.
>>
>> What is the ground state of this system?
>>
>> I guess it is O- and H+.
>> So there is a charge transfer even if you have no interaction. This is
>> correct because we want the ground state and the oxygen is more
>> electronegative than H.
>>
>> I did the calculation with BigDFT (wavelet part of ABINIT) forcing the
>> initial configuration to 1 electron to H. I found effectively H+ and O-
>> (separation to 20 bohr).
>>
>> I think it  is not related to functional because this comes from
>> electrostatic effect.
>>
>> Best Regards
>> 6671011@163.com a écrit :
>>>
>>> Dear Thierry,
>>>
>>> I'd like to question further on this issue with an example.
>>>
>>> For a system with two atoms: H and O, which are separated by very large
>>> distance (say several hundred Angstroms). Almost all of the present
>>> programs
>>> show that there is a charge transfer between them. Experimentally,
>>> however, it
>>> is absolutely wrong.
>>>
>>> Someone said this problem is beyond the Born-Oppenheimer approximation. Is
>>> that
>>> true?
>>>
>>>
>>> Sincerely,
>>> Guangfu Luo
>>>
>>>
>>>
>>>
>>> ----------------------------------------------------------------------------------------------------
>>> Hi,
>>>
>>> ABINIT can use periodic Poisson solver based on plane waves or Poisson
>>> solver
>>> for isolated system. The last one is based on wavelet (exactly
>>> interpolating
>>> scaling functions). If you use plane waves to represent wavefunctions,
>>> there is
>>> no difference for neutral systems. In this case, the dependence of the
>>> kinetic term versus the size of the box is
>>> predominant (plane waves delocalize).
>>> For charged system, (as H3O+), the Poisson solver based on wavelet is more
>>> accurate and uses very very less memory.
>>> If you calculate NH3 and H3O+ together in a big box, you only need to
>>> specify
>>> the total charge. The system should localize an electron on H3O. Because
>>> you have molecules, you should have 1 for occupied states
>>> (spin-polarized case).
>>> An alternative is also to use the whole wavelet part of ABINIT... The
>>> advantage
>>> is to use less memory.
>>> Best Regards,
>>>
>>> Thierry
>>>
>>> -----------------------------------
>>> Thierry Deutsch Laboratoire de simulation atomistique (L_Sim) INAC/SP2M
>>> Tél:(33) 04 38 78 34 06
>>> C.E.A.Grenoble Fax:(33) 04 38 78 51 97
>>> 17, Avenue des Martyrs mailto:Thierry.Deutsch@cea.fr
>>> 38054 GRENOBLE CEDEX 9 FRANCE http://inac.cea.fr/L_Sim
>>>
>>>
>>>
>>> ----------------------------------------------------------------------------------------------------
>>> Quan Phung Manh a écrit :
>>>
>>> After calculating NH4+, I have more question.
>>> If I have a big box with a molecule NH3 and an ion H3O+ for example. How
>>> can I
>>> simulate it? How can I sure that the charge + is in H3O, not NH3? Can I
>>> solve
>>> this by using occ?
>>> Best regarded
>>>
>>>
>>
>>
>> --
>> -----------------------------------------------------------------------
>> Thierry Deutsch
>>  Laboratoire de simulation atomistique (L_Sim)
>> INAC/SP2M                                       Tél:(33) 04 38 78 34 06
>> C.E.A.Grenoble                                  Fax:(33) 04 38 78 51 97
>> 17, Avenue des Martyrs                    mailto:Thierry.Deutsch@cea.fr
>> 38054 GRENOBLE CEDEX 9 FRANCE                  http://inac.cea.fr/L_Sim
>> -----------------------------------------------------------------------
>>
>>
>
>
>
> --
> Pierre-Matthieu Anglade
>
>