[Cp2k-team] &LOCALIZE in &QS section

[Marcella Iannuzzi]

Hi all,

  below it is reported the discussion  about the localization problem.
But first  here are my conclusion after some checks in the code and  
some additional testing:
- yes, the localization procedures are correct and the charge density  
is not disrupted by localization
- the localization  of the MOS should be used to compute properties  
that require  it
    by activating the LOCALIZE sub-section in the corresponding  
section, for example &XAS, &LINRES ...
As suggested by teo, the LOCALIZE section in QS could be moved into  
the DFT&PRINT
- However, I still believe that it might be useful to propagate  the  
localized orbitals along an MD run, just consider all the work on  
localized wavefunctions that one can find in literature.
However, this cannot be done straightforwardly in cp2k, by just   
overwriting the MOS with the
localized MOS. The reason is that by the standard diagonalization  
method the calculation of the forces assumes that the MOS
are eigenstates. This is not the case for the localized MOS.
It would be probably easy to fix this using another method. For  
example one could use OT, providing that an update of the MO_DERIV is  
added after the localization procedure. Or one could use a more  
general method to compute the Pulay forces, which does not assumes  
that the MOS are eigenstates as it is done for ROKS.

bye
Marcella



On 10 Jan 2008, at 12:32, Teodoro Laino wrote:

>
> On 10 Jan 2008, at 12:25, Marcella Iannuzzi wrote:
>
>>
>> Hi,
>>
>> Well the localization procedure should not disrupt the electronic  
>> density distribution,
>> therefore  overwriting the original  MOS with the localized ones  
>> should make no harm,
>> but it is probably not useful either, if there are no properties  
>> related to the localized orbitals to be computed.
> exactly
>
>> For  linres it is indeed necessary have them overwritten, and this  
>> is also true for  core level spectroscopy.
>> I think it is intuitively more appropriate to have the LOCALIZE  
>> section as a subsection that appears where needed, i.e.
>> inside the LINRES section if it is used there, inside XAS if it  
>> used there, and so on.
>> In this way, for example, XAS will decide how and what to localize,  
>> looking in its own input section, being independent
>> from a possible other localization required and done at the end of  
>> the ground-state SCF  in order to print out the wannier centers.
>> I would say that there is only  one LOCALIZE section, that can be  
>> located in different parts of the input, to be used
>> at different levels of calculation.
> I definitely (more than 100%) agree with you. have a LOCALIZE  
> section wherever (and only!) you need it.
> This means maybe few more nested sections in the input but I firmly  
> believe that this is the way to go!
>
> If we agree I can clean the QS%LOCALIZE and introduce a LOCALIZE  
> section in the print statement of the wannier.
> In this way we should have everything more consistent.
>
> right?
>
> Ciao,
> teo
>> Marcella
>>
>> On 10 Jan 2008, at 10:03, Teodoro Laino wrote:
>>
>>> Hi Marcella,
>>>
>>> On 10 Jan 2008, at 09:19, Marcella Iannuzzi wrote:
>>>
>>>> Dear all,
>>>>
>>>> yes there seems to be a problem with the localization.
>>>> The idea of having a LOCALIZED section was to be able to localize  
>>>> the
>>>> MOS when needed and not only in relation with the
>>>> print-out of the wannier centers.
>>>> But it has been done some time ago now and it is probably time to
>>>> unify  the two versions.
>>> Very good!
>>> The only thing that I'm just trying to understand is: is it really  
>>> used by anyone the feature of having LOCALIZED
>>> orbitals during an MD (controlled by the QS%LOCALIZE) (w/o writing  
>>> them) or is it just an unfinished work towards
>>> a more consistent input structure?
>>>
>>> I see that there's also a LOCALIZE section in LINRES section.. so  
>>> everything connected with the linear response+localization
>>> has even yet another control section..
>>>
>>> If it is not needed in the QS section (if it is needed.. for  
>>> what?) it should be deleted.
>>>
>>> That's fine for me the idea of having the MOs not overwritten by  
>>> the localized ones. Especially because that part is
>>> more or less working and used by quite few people. The new one  
>>> (overwriting MOs, maybe for some bug in the code) would give
>>> the wrong number at the moment..
>>>
>>> On the contrary if you decide to go for keeping the section QS 
>>> %LOCALIZE, please remember to get rid of all the keywords
>>> controlling the localization from the DFT%PRINT%LOCALIZATION. That  
>>> would be just  a confusing redundancy..
>>>
>>>> I remember that some time ago I did check the energy  
>>>> conservation.... but, as it
>>>> has been pointed out, there is no regtest that controls this part  
>>>> of the code ...
>>>> too bad.
>>>> I will have a look.
>>> You can check the example that Marcel was posting on the google  
>>> group. It's pretty fast and energy diverges quickly..
>>>
>>> ciao
>>> teo
>>>>
>>>> ciao
>>>> Marcella
>>>>
>>
>