[Pw_forum] Re: Problem with Wannier function centers

[Nicola Marzari]

Hi Kostya,

it could be a bug, but it could also be an intrinsic issue.

The localization functional, in general, has multiple minima, of
which only one is meaningful in the chemical sense we have grown used to
(sometimes there can be more than one meaningful minimum, but that's
a longer story).

In fact, in the general wannier code (www.wannier.org) we use
projections onto trial wavefunctions to start our localization from
the basin of attraction of the minimum we want. The existence of
  "false" minima is ultimately related to the fact that Im ln <u_nk|u_m
k+b> is a multiply-valued function, and the arbitrary phases of the u_nk
that the wannier code tries to optimize can be such that there is
a 2 pi discontinuity, somewhere in the BZ, between u_nk and its
neigbours.

If you use CP/Gamma sampling (only real wavefunctions), in an orthogonal 
cell, this problem has never appeared.

In CP, if the cell is not orthogonal, and you are using more than
3 G vectors to calculate the WF center (it's a sum on a shell of
neighboring G vectors, of Im ln <psi|e(igr)|psi>), you
can still converge to a local minimum.

Yudong Wu had tested the code for all the ibrav, so the Wannier
algebra should be correct - maybe you are converging to a local
minimum.

1) avoid non-orthogonal cells in CP

2) use www.wannier.org, in combination with pwscf, as soon as it is
released (end of April).

3) use projections in CP. We'll add this in the future, but probably
not right away.


				nicola


Konstantin Kudin wrote:
>  Hi Manu & Nicola,
> 
>  The centers produced for an orthogonal cell seems to be fine (so the
> code works in generic sense), however, when the cell vectors are
> general, things do not work.
> 
>  I have peeked briefly at the part of the code that computes the
> centers [wf.f90, near line that has "write( 26" ], it appears that the
> calculation is clearly wrong if the cells are not orthogonal.
> 
>  Kostya
> 
> 
> --- Nicola Marzari <marzari at MIT.EDU> wrote:
> 
> 
>>Dear Manu,
>>
>>do you have any sense if the Wannier part still works ?
>>
>>This is your code in the CP, correct ?
>>
>>			nicola
>>
>>
>>Konstantin Kudin wrote:
>>
>>
>>> Hi all,
>>>
>>> It appears that the Wannier code in CP (recent CVS) outputs the
>>>Wannier function centers incorrectly for cells that are specified
>>
>>by
>>
>>>ibrav. The input producing strange centers for a water molecule is
>>>here:
>>> https://kiev.princeton.edu/failing_example_wannier.j
>>>
>>> With these coordinates:
>>>ATOMIC_POSITIONS angstrom
>>>   O    0.99999987E+00  0.99999761E+00  0.11192157E+01
>>>   H    0.10000039E+01  0.17655545E+01  0.52313723E+00
>>>   H    0.99999715E+00  0.23446441E+00  0.52313869E+00
>>>
>>> wannier centers (fort.26) come out as :
>>>  13.042997  11.149277  10.237439
>>>   1.702475  -0.381710  -1.063049
>>>   1.309338  -0.584178  -0.629999
>>>   2.165164  -0.547665  -1.062867
>>>
>>> This does not make much sense ...
>>>
>>> Kostya
>>>
>>>__________________________________________________
>>>Do You Yahoo!?
>>>Tired of spam?  Yahoo! Mail has the best spam protection around 
>>>http://mail.yahoo.com 
>>>_______________________________________________
>>>Pw_forum mailing list
>>>Pw_forum at pwscf.org
>>>http://www.democritos.it/mailman/listinfo/pw_forum
>>
>>-- 
>>---------------------------------------------------------------------
>>Prof Nicola Marzari   Department of Materials Science and Engineering
>>13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
>>tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu
>>
> 
> 
> 
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around 
> http://mail.yahoo.com 

-- 
---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu