[CP2K:1760] Re: Total dipole via LOCALIZATION - bug?

[Teodoro Laino]

Hi Fawzi,
Just for clarity and precision: joost bug fix does not affect dipole  
computed with localization, only the dipole computed via MOMENTS,  
where there was a bug in the definition of the dipole matrix.

Teo

---------------------------------------------
Teodoro Laino
Zurich Switzerland

Contact info:
Tel.:     http://www.jajah.com/Teo
E-mail: teo... at laino.eu
              teodor... at gmail.com
---------------------------------------------

On 4 Feb 2009, at 16:40, Fawzi Mohamed <fa... at gmx.ch> wrote:

>
>
> On 3-feb-09, at 13:35, Fawzi Mohamed wrote:
>
>>
>>
>> On 3-feb-09, at 12:22, Juerg Hutter wrote:
>>
>>>
>>> Hi Fawzi
>>>
>>>> The thing my code did and is not done now is simply to remove jumps
>>>> by
>>>> L/2*pi, and output also the finite derivative of the dipole, this
>>>> could also be added to the present code as an option.
>>>>
>>>> If you use the total dipole of a periodic system setting the
>>>> reference
>>>> to COAC is not so useful, and does not allow you to remove the  
>>>> jumps
>>>> easily.
>>>
>>> Can you explain why the removal of jumps depends on the reference
>>> point?
>>> If you use COAC there is only one non-zero term, meaning there is
>>> also
>>> one term that can jump. For all other cases both contributions can
>>> jump independently.
>>>
>>> Juerg
>>
>> you are right, indeed the simple removal of jumps works also with
>> COAC.
>> I was thinking that using an oscillating reference point might
>> introduce a spurious signal.
>> Without the berry phase as long at the system is neutral the  
>> reference
>> point has no importance, but if one treats only the electrons, the
>> system is charged, and one has to take care how the ions+electrons
>> system if brought back together (only for the whole system there is  
>> no
>> problem).
>> Using the berry phase the thing should be the same, but I already
>> flip-
>> flopped a couple of times on it.
>> I would say that the thing is not immediately obvious for systems
>> distributed on the whole cell, but it might well be that you are  
>> right
>> (it makes sense for neutral systems).
>>
>> Fawzi
>
> following up my own message ;)
> yes indeed the dipole is always independent from the reference point
> for neutral systems (as it should be).
> So Juerg you were right that COAC is a good choice to minimize the
> jumps for neutral systems.
>
> Victoria, using periodic total dipole you should have the same results
> (maybe even better, given joost bugfixes).
> The only difference is that now the last column does not remove the
> jumps in the dipole, whereas the version you had did (these are jumps
> due to one electron jumping from one side to the other of the cell).
> They are easy to remove in post processing (in reduced units), but
> would also be trivial to add (a call to pbc for the delta_dipole when
> do_berry is true).
>
> Fawzi
>
> >