Estimation of oxidation numbers and icsd_vb.yaml

[futofuji]

@computron
Thank you for your great efforts to pymatgen.
We are now trying to estimate oxidation numbers for Ti4O7. Then, pymatgen gives us the following solution:

from pymatgen import Composition
comp = Composition("Ti4O7")
print(comp.oxi_state_guesses())
print(comp.add_charges_from_oxi_state_guesses().as_dict())

({'Ti': 3.5, 'O': -2.0},)
defaultdict(<class 'float'>, {'Ti2+': 1.0, 'Ti4+': 3.0, 'O2-': 7.0})

On the other hand, see mp-12205 of the MATERIALS PROJECT:
https://next-gen.materialsproject.org/materials/mp-12205?formula=Ti4O7
It is stated to be composed of Ti3+ and Ti4+.

We are wondering how to resolve this inconsistency.

In my understanding, the estimation of oxidation numbers by pymatgen is based on "occurrence" written in the "pymatgen/analysis/icsd_bv.yaml",
c.f.,

occurrence: 
  Ti2+: 39
  Ti3+: 119
  Ti4+: 665

In the case of "Ti2+" + 3 * "Ti4+" the summation of the occurrence is 39 + 3 * 665 = 2034, while the case of 2 * "Ti3+" + 2 *"Ti4+" is 2 * 119 + 2 * 665 = 1568. Therefore, {'Ti2+': 1.0, 'Ti4+': 3.0, 'O2-': 7.0} is returned by pymatgen.

Anyway, which set of oxidation numbers should be used? Besides, how the "occurrence" in icsd_bv.yaml was calculated?

Any comments are welcome. Best regards.