VBM and CBM parsing inconsistency in Pyiron

[ahmedabdelkawy]

In pyiron we parse the occupancy matrix from the outcar (last scf of each ionic step) and vasprun (last scf of last ionic step). However we parse them differently and in case when we use Methfessel-Paxton smearing (and also negative occupancies), that messes up with the outcar parser resulting in different values!

from pyiron import Project
pr = Project('.')
j = pr.create.job.Vasp('test', True)
j.structure = pr.create.structure.bulk('Al', cubic=True)
j.calc_static()
j.run(run_mode='queue')

es = j.get_electronic_structure()     # from vasprun -> last ionic step
es.vbm

j['output/generic/dft/vbm_list'] # from hdf which is from outcar -> one value per ionic step

I will add more to this later!

[ahmedabdelkawy]

@ligerzero-ai probably this is what you faced when comparing values from pymatgen:

es = j.get_electronic_structure()
es.vbm

This should give the same values as pymatgen but you also need to add vr.separate_spins = True for the vasprun pymatgen object in case of spin polarized calculation to get the same results!

[ligerzero-ai]

thanks @ahmedabdelkawy, does this mean that we currently default to using a specific valence band max from the spin up electrons only?

From my (very limited understanding) is this not problematic since the band gap computed by pyiron does not account for the dependency of the band gap on the spin-up vs spin down channel for electrons?

[ahmedabdelkawy]

@ligerzero-ai No we currently save two values in the case of spin-polarized calculations and one value in the case of the spin-nonpolarized calculation, which is the correct behavior. The value, however, is different as mentioned earlier in case you access the value parsed from the outcar or vasprun. The value stored from vasprun is correct. We store both values but we store the wrong one from the outcar because we deduce it wrongly from the occupation and get the corresponding eigenvalue. It is not readily printed out in the output files. It is deduced correctly however from the vasprun.

for the band gap, you will have two values (usually they are the same or similar except for the case you have a half-metal system or charged defects) In the case of the spin-polarized calculations and one value in the nonpolarized case.
Yes, the band gap value is affected by the absolute values of the vbm and cbm as we compute the band gap as the difference between the cbm and the vbm for each spin channel (see get_band_gap function from the get_electrostatic_potential object). Luckily, this is also done electronic.py (in pyiron_atomistics/dft/waves) which uses the cbm and vbm values from vasprun, which are correct and this is the adapted code from pymatgen.