[Wannier] several questions about wannier calculation and settings

[Jonathan Yates]

On Sun, 22 Oct 2006, lan haiping wrote:

> one question is about kmesh setting:
>  when i use 3x3x3 (or 4x4x2) mp grids , wannier could generate data
> tio2.nnkp without complain. But when i tried another mp grids such as 4x4x4,
> wannier would complain  " kmesh_get: something wrong, found > 12 nearest
> neighbours"  and exit without generating tio2.nnkp data file.   would you
> please  explain what the problem it is related to ?   From mannual , i guess
> this error is due to kpoints' shell, but i donot follow much idea about this
> setting, would you please explain more about it ?

  If you go back to the original formulation in the Marzari-Vanderbilt MLWF 
paper you see that rather than use a real space representation of the 
position operator we use a k-space finite difference approach.
  In 1-D a finite difference formula is trivial. For a general 3-D lattice 
it is a little more complicated; this is discussed in detail in Appendix B 
of M-V.

  Wannier90 takes your crystal lattice and finds the reciprocal lattice. It 
them looks for shells of kpoints away from the origin. Next it decides 
which combination of shells gives a good finite difference representation 
(it might be just the 1st shell, or the 1st and 2nd shells, or maybe the 
1st, 4th and 7th shells). One thing we know from Appendix B of M-V is that 
we shouldn't need more than 12 kpoints to give a good finite difference 
representation. If the code finds more than 12 kpoints it aborts with the 
warning you found. This is normally an indication something is wrong with 
the lattice, or the kpoints.
  One way in which this can happen is if you use a 3x3x3 (or 6x6x6) mesh 
and don't give the kpoints to sufficient precision. For this reason we 
supply a script kmesh.pl in ./utilities/ - this gives the kpoints to the 
required accuracy.

  In your particular case I don't think this is the problem. In fact I 
could run your input file with a whole series of mp grids (inc 4x4x4) with 
no trouble at all.
  Maybe this a problem with your particular setup? [ie buggy compiler, or 
lapack library]

You can find my exact input and outputs on the following webpage
http://www.tcm.phy.cam.ac.uk/~jry20/wannier90/


> Just with(4x4x2 mp mesh) , successfully finishing pw2wannier calculation ,
> but wannier.x calculation gave this complain: "param_read: mismatch in
> tita.eig"  .  Does it mean scf calculation is not converged or wrong
> parameters for wannier.x ?

It's nothing to do with the scf calculation.

  Wannier90 has found a different number of bands or kpoints to what it was 
expecting.

  In the example you gave (num_wann=24, mp_grid = 4 4 2), *.eig should have
24x4x4x2=768 lines.

I suspect you might have included more than 24 bands in your pwscf nscf 
calculation. But can check this by looking at the *.eig file. Each line of 
of *.eig is
  band#  kpoint#   eigenvalue.


> another stupid question about projection setting:    the setting Nicola gave
> for PbTiO3 in last mail  is according its band's structure , and the
> projection functions are s/p type-like orbitals.  i wanna know   whether
> there are some guidelines for setting up initial projection functions ?

  Ok, I'll come back to this latter (or maybe someone else will say 
something).

  I will add that Nicola's article on MLWF for perovskites is available on 
cond-mat.
  http://arxiv.org/abs/cond-mat/9802210
  I wasn't aware of this until recently; the actual article is from an AIP 
conference proceedings, but I couldn't find this electronically.


  Yours
    Jonathan


-- 
Dr Jonathan Yates         |    Theory of Condensed Matter Group
Corpus Christi College    |    Cavendish Laboratory
Cambridge, CB2 1RH, UK    |    Cambridge, CB3 OHE, UK
email jry20 at cam.ac.uk     |    Tel +44 (0)1223 337461