CrystalNN bug with atom bonded to its own images

[fxcoudert]

Python version

3.11.2

Pymatgen version

2024.6.10

Operating system version

macOS 14.5

Current behavior

Take a structure with only one atom per cell, which bonds to its own images in neighboring cells. Many simple metals do that, like mp-2647057.

Apply CrystalNN to get a bonded structure. The structure for mp-2647057 looks like this:

Structure Graph
Structure: 
Structure Summary
Lattice
    abc : 2.428889488469988 2.428889488469988 2.428889488469988
 angles : 109.47122063449069 109.47122063449069 109.47122063449069
 volume : 11.030656874268672
      A : -1.40232 1.40232 1.40232
      B : 1.40232 -1.40232 1.40232
      C : 1.40232 1.40232 -1.40232
    pbc : True True True
PeriodicSite: Co0 (Co) (0.0, 0.0, 0.0) [0.0, 0.0, 0.0]
Graph: bonds
from    to  to_image    
----  ----  ------------
   0     0  (1, 1, 1)   
   0     0  (1, 0, 0)   
   0     0  (0, 0, 1)   
   0     0  (0, 1, 0)   

The bonds graph is incorrect. The Co atom should be 8-connected, and there should be bonds to atoms at positions (-1, -1, -1) (-1, 0, 0) (0, 0, -1) (0, -1, 0). These missing bonds have the exact same length and characteristics as the ones present, in fact they are equivalent by symmetry.

If you make a 2x2x2 supercell and apply CrystalNN, you get the correct coordination.

Expected Behavior

The Co atom should be 8-coordinated.

Minimal example

crystallnn = CrystalNN()
from pymatgen.analysis.local_env import CrystalNN
from pymatgen.ext.matproj import MPRester

with MPRester(apikey) as mpr:
    mp_data = mpr.materials.summary.search(
        material_ids=["mp-2647057"],
        fields=["material_id", "formula", "formula_pretty", "nelements", "structure"]
    )

structure = mp_data[0].structure
bonded_struct = crystallnn.get_bonded_structure(structure)
bonded_struct

### Relevant files to reproduce this bug

_No response_