`atom_site_label` in CIF file are not unique about pymatgen HOT 10 CLOSED

@stefsmeets I just wanted to mention that this might be problematic for very large structures. There are only 26 letters in the alphabet. With exception to this drawback, such a solution would be nice.

from pymatgen.

I'd be happy to have a go at this

from pymatgen.

One of the reasons labels should be unique: they are used to indicate bonding (and angles, torsions, etc). If they are not unique, this information is useless. (And I am working on a PR to add that capability to pymatgen.)

from pymatgen.

I can confirm that this is likely a bug (in terms of having agreement with the CIF spec).

from pymatgen.

The bug may not necessarily be in the CifWriter itself. Pymatgen expands the structure to P1 when reading a cif file (symmetry information gets lost). This includes duplicating the labels.

from pymatgen.

Does pymatgen require labels to be unique? I do not know. If so, the bug is in the symmetry expansion. But what is sure is that labels in a CIF file should be unique (whether pymatgen accepts duplicate labels or not).

from pymatgen.

No, pymatgen does not require unique labels. The duplication of labels happens when the symmetry is expanded in the CifParser:

One way to solve would be to ensure labels are suffixed 'abcde...' when the symmetry gets expanded there. Although this can also be a check on the CifWriter side.

from pymatgen.

@stefsmeets sound great. (cc @janosh)

from pymatgen.

Some options:

1. CifWriter raises ValueError if Structure contains duplicate label -> user is forced to correct it
2. CifWriter raises UserWarning if Structure contains duplicate label -> user can choose to correct it
3. CifWriter overwrites labels to fix the issue -> less hassle for the users, but can lead to unexpected labels in CIF file
4. CifParser.parse_structures() ensures unique labels when reading CIF -> if the user modifies the sites, they may still end up with duplicate labels

I want to avoid agressively/automatically relabelling, because I know that this can lead to unexpected results which can be frustrating. So I'm leaning towards 1 or 2.

To help with 1 or 2, I want to add a method: Structure.relabel_sites() to give users some room to quickly relabel a large number of sites and ensure uniqueness. I think this can be useful to add regardless.

Let me know what you think.

from pymatgen.

• Option 1: sadly, there are CIF files (mostly from computational tools) that have this problem, so I think that is not viable.
• Option 2: sounds nice, but does not account for situations where pymatgen itself is creating label duplicates (when lowering symmetry). How about an option to CifWriter to make labels unique.

from pymatgen.