[Wannier] Wannier function centers at wrong positions

[Jonathan Yates]

Jan,

 I just wanted to add a few things to Nicolas's reply:

Given that you are using the Gamma-point only routines and your cell is orthorhombic I would expect that any initial projection would work fine - and the minimisation would go rapidly to the minimum. Of course you can test this statement by trying different projections. If you find this not to be true I'd be interested to know.

I assume that you want MLWF to analyse some bonding properties of your material. A common task would be to obtain MLWF for the complete set of valence states (I'm assuming here that your material has a gap). In which case 606 MLWF for 234 atoms seems rather low (so for 234 atoms of SiO2 or GeO2 we'd have 1872 MLWF).
  Are you trying to obtain MWLF for a subset of the valence bands? This will only work if those 606 bands form an isolated set. If not you have to look into the disentanglement routines.

'centers hang somewhere around'
  I'm not quite sure what you mean by this. I assume it is not something as simple as needing to translate by a lattice vector? 

Hopefully that has given you some things to try. I think to make any more comments I'd need to see the *.wout file, and know a bit more about your system.

 Yours
  Jonathan


  



On 9 Apr 2010, at 17:22, nicolas poilvert wrote:

> Dear Jan,
> 
> I see that in your input file that you use exactly 606 bands for 606 Wannier Functions.
> It would be good to maybe include more bands in your calculation and use the disentanglement
> procedure (look at the Wannier User Guide for that).
> Also, if you cell is quite big, using random projections may not be optimal. So if you have an
> idea of what those Wannier Functions should be, there is definitely advantage to input some
> carefully chosen initial projections. In the random guess case, the code may get stuck in a local
> minimum that does not correspond to the proper physical minimum you seek.
> As a final note, I would set the convergence threshold to 1.0d-7 at least to both disentanglement
> and wannierisation and use a lot of steps just in case (num_iter > 2000 and dis_num_iter > 2000).
> 
> good luck,
> 
> Nicolas
> 
> On Fri, Apr 9, 2010 at 12:02 PM, Jan Felix Binder <janfelix.binder at epfl.ch> wrote:
> Hello,
> I used the W90 code with PWscf (QE 4.1.2) to calculate MLWF for a
> amorphous structure (gamma point only, 234 atoms, 606 wf).
> All WF convered to reasonable values. Most of the centers are at the
> right positions, but approximately 10 centers hang somewhere around.
> What can be wrong?
> 
> I add the imported lines of the input file:
>  gamma_only = .true.
> num_iter         =  100
> dis_num_iter     =  200
> conv_window = 10
> conv_noise_amp = 1
> conv_noise_num = 10
> dis_mix_ratio = 0.3
> guiding_centres = .true.
> 
> iprint           =    2
> num_dump_cycles  =   10
> num_print_cycles =   10
> 
> length_unit      =  Ang
> 
> num_wann = 606
> num_bands=606
> 
> exclude_bands : 607-614
> 
> Begin Projections
> random
> End Projections
> 
> translate_home_cell=.true.
> write_xyz=.true.
> 
> begin unit_cell_cart
>  bohr
> 47.65304  0.00000  0.00000
>  0.00000 23.82652  0.00000
>  0.00000  0.00000 23.82652
> end unit_cell_cart
> 
> Thank you for help!
> Felix Binder
> 
> EPFL Lausanne

-- 
Department of Materials, University of Oxford, Parks Road, Oxford, OX1 3PH, UK
tel: +44 (0)1865 612797                http://users.ox.ac.uk/~oums0549/