[Wannier] Wannier functions of molecular crystal

[Jonathan Yates]

On 13 Apr 2010, at 14:08, Gianluca Giovannetti wrote:

Dear All,

i`m calculating Wannier functions for a molecular crystal (sites are molecule with C and H ions within each molecule).

i want to plot Wannier functions of the 4 bands at Fermi level.
the system is metallic.
these four bands come from HOMO and LUMO molecular states of the two inequivalent molecules in the unit cell.

Gianluca,

 I'm assuming that these four bands which span the fermi level form an isolated group (so no disentanglement is needed)


My feeling is that the setting in the win file:
begin projections
random
end projections
is not really good as starting point.
Indeed the CONV is 1.0E-03.
HOMO and LUMO are localized on the molecules.
What is the best strategy in this case?

I think using randomly placed Gaussian functions is likely to be particularly bad in this case. If the orbitals are already fairly localised, and the cell is pretty large, then the trial functions might have little overlap with the bloch states. That won't make the calculation wrong - just that the starting point could be poor.
 So if you think that the states are 2p like on Oxygen then try such projections as a starting guess. Sometimes a two-step procedure might help: look at the WF resulting from the random projections - if they appear to have some symmetry character then try again using a projection of that form.

 It might help in this procedure to use a coarser kpoint mesh (4x4x4 or even 2x2x2) - I find that the minimiser can cope better with a poor starting guess in this case. Then when I know what the MLWF should look like, I increase the mesh.

However i get Wannier eigenvalues which are not exactly the same as the one i have calculated Quantum espresso (the discrepancy is not big but finite at some k-points).

 You're obtaining the band-structure by interpolation. At the kpoints which lie in the original (ab-initio) grid the energies will be reproduced exactly. At any other points it is an interpolation. Importantly the quality of the interpolation increases exponentially as the size of kpoint mesh is increased.
 However, it is probably a good plan to get well localised wannier functions first - as this may improve the interpolation.

 Jonathan




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

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/wannier/attachments/20100413/81171a16/attachment.html>