Bug in Wannier90 interface?
From commit 8ea262a4 the Wannier90 interface sometimes gives unreasonable results with WF:s with negative spread that are not atom-centered. See e.g. attached scripts and compare the initial and final spreads and centres in the end of the .wout files from before commit 8ea262a4 (w_TiCl2.wout_old) and after commit 8ea262a4 (w_TiCl2.wout_new). This seems to be a bug in the wannier90 interface. Specifically the overlaps (.mmn) files are different for the different versions which leads to different Wannier functions. A possible fix could be to roll back to the old version of the Wannier90 interface. get_wannier.py
Run scripts in the following order:
- gpaw python gs_TiCl2.py
- mkdir wannier && cd wannier
- gpaw python gs_fullbz.py
- GPAW python get_wannier.py