Modified CifParser.check() as one possible solution for issue #3626

[kaueltzen]

closes #3626. CifParser.check() raises a UserWarning: Incorrect stoichiometry for a cif, if

• the keys _chemical_formula_sum and _chemical_formula_structural are missing and
• crystallographic sites have different multiplicities

For the relative stoichiometry to be checked correctly, the site multiplicities would have to be included by applying the cif's symmetry operations to the crystallographically unique sites (like for construction for the structure).

As this would be no longer a sanity check, i propose skipping the relative stoichiometry check in this case (but still including all the other checks for valid elements etc.).

For the case of skipping the relative stoichiometry check, i appended a message to self.warnings, but I'm not sure if there are better ways to handle this case.

[kaueltzen]

If you are okay with this approach i could adapt the 2 cif tests that are failing now (and for example add a formula key so the checks are not skipped). (Modifying the tests no longer necessary as other checks than for relative stoichiometry were reincluded in case of missing formula keys)

[janosh]

thanks @kaueltzen, this makes sense! also great issue description over in #3626!

if you could add a test with the MCIF file you linked in the issue (or find an existing CIF test file that repros this problem), this is good to go.

pinging @esoteric-ephemera in case he has further comments

[janosh]

can we add a test for this warning?

[esoteric-ephemera]

Thanks @kaueltzen for pointing this out! I like your solution of warning that the stoichiometry check couldn't be accomplished - the fallback I added originally clearly isn't robust enough in general

Adding a test for this (using the magnetic cif you sent with the issue) would be great to move this forward

[kaueltzen]

Thank you for your quick replies! I added a test both for skipping the relative stoichiometry check and appending a warning in the case of missing formula keys. There was already a test file in pymatgen, Li10GeP2S12.cif, that matched the criteria (it has no formula keys and sites with different multiplicities and occupancies).

[janosh]

very nice work! thanks a lot @kaueltzen 👍