[Wannier] vasp-wannier supercell symmetry

[Jonathan Yates]

Aron,

 I don't see a fundamental reason why using a supercell would not work. However, when you change the symmetry of the cell (and hence k-point mesh) you have changed the maths of the problem, and the calculations are now only equivalent in the limit of perfect k-point sampling. Of course the starting guess for the primitive cell should still be perfectly good for the supercell. Have you tried setting guiding_centres = .true. in the win file? That's always a good idea when you see very large spreads from a chemically reasonable guess.

 I am a little surprised that you need to break the symmetry to get the kmesh routines to work. One should never need to do that. Can you send me your win file so I can take a look. The kmesh routine is pretty robust - there are certain cases where automatic generation does fail - but we tried to get the code to give the user hints as to how to work around these issues. I'd welcome the chance to make it more robust.

 Yours
  Jonathan


On 16 Jan 2014, at 16:42, Aron Szabo <szaboa at iis.ee.ethz.ch>
 wrote:

> Dear Wannier users,
> 
> I was trying to do some calculations with supercells, and I've noticed 
> that wannier90 won't converge when the supercell has different symmetry 
> than the primitive unit cell. E.g. on a graphene sheet using 4 atoms in 
> a cell instead of 2, I can either define the new lattice vectors just as 
> multiplying one of the original ones with 2, or I can set orthogonal 
> lattice vectors as well. In vasp it won't make any difference in the 
> energies, and also the band structures can be mapped onto each other 
> just considering the different shapes of the Brillouin zone. But while 
> in the first case wannier90 will produce the same Wannier functions as 
> in a primitive cell calculation (just repeating them on the other two 
> atoms), in the second case it won't converge to localized results. 
> Instead of the order of 1, the spreads are in the order of 10 or 100, 
> when starting from the same projections, and doing a few hundred iterations.
> 
> The same happened when building up supercells of zinc-blende GaSb. When 
> keeping the symmetry of the unit cell it works fine, but when building 
> an orthorombic cell, it doesn't converge. First I got negative weights 
> for the k point neighbours, and negative spreads at the end. I found out 
> that it was because instead of just taking the first two neighbours 
> along each axis (as it should be for orthorombic cells) four other k 
> points (along diagonal directions) were taken too, because they were 
> exactly at the same distance as the first neighbors along the axial 
> direction (the sides of the BZ are a, b, and c, with c^2 = a^2+b^2, due 
> to the symmetry of the system). Artificially changing one lattice vector 
> a bit solved this part with the negative weights, but still, the Wannier 
> functions did not converge to a localized state, but the spread remained 
> huge.
> 
> Has anyone met such issues, and had maybe any idea what causes this, and 
> how to solve it, when the use of a supercell with some different 
> symmetry as the primitive cell can't be avoided (e.g. to simulate an 
> ultra-thin-body transistor with some pre-defined transport orientation)?
> 
> Thank you in advance for any help
> 
> Aron Szabo
> PhD student @ ETH Zurich, Integrated Systems Laboratory
> _______________________________________________
> Wannier mailing list
> Wannier at quantum-espresso.org
> http://www.democritos.it/mailman/listinfo/wannier

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