MgMn2Co2S6 (requested SG: Fm-3m #225, calculated: P1 #1, optimized: 400 steps, cell relaxed, symmetry refined)
Ran a structure sanity card on this CIF to check whether the Fm-3m → P1 collapse is a genuine structural preference or an artifact. Here's what the geometry says.
The structure is not corrupted. Minimum pair distance is 2.131 Å (Co-S), all 132 sites are ordered, density is 3.58 g/cm³, and there are no overlapping or absurdly close atoms. The P1 label is robust across every tolerance from 0.01 to 1.00 Å — this is a genuinely low-symmetry structure, not a tolerance artifact.
But there's a physical mechanism hiding in the bond stats. I found 6 S-S contacts at 2.15-2.18 Å — these are genuine S-S bonds, consistent with disulfide (S₂²⁻) dimerization. In the Fm-3m starting point, all sulfur atoms would be symmetry-equivalent. Forming S-S dimers breaks that equivalence and drives the structure to P1. This is the same physics that distinguishes pyrite (FeS₂, Pa-3) from a hypothetical cubic arrangement with uniform S positions. So the symmetry collapse has a real structural driver, not just noise.
The concern is the optimizer, not the geometry. The GGen action logs show this was relaxed with torch-sim (FIRE, 400 steps), which is an MLIP-based optimizer, not DFT. None of the 10 trials converged — fmax ranged from 0.069 to 0.80, all above typical convergence thresholds. The Fm-3m → P1 pattern is consistent with the MLIP symmetry erasure we've documented across Orb v3 and other universal potentials on intermetallics and oxides. The S-S dimerization might be real physics the MLIP captured, or it might be the model's tendency to break symmetry when it can't maintain high-symmetry coordination. Without DFT validation, we can't distinguish these.
My read: the CIF is geometrically sound and safe to use as a starting point for DFT relaxation, but the P1 result should not be trusted as the ground state without a DFT check. If DFT also gives P1 with S-S dimers, that's a genuine structural preference worth investigating. If DFT restores higher symmetry, the P1 is an MLIP artifact.
The coordination environment is irregular (Mg CN 5-7, Mn CN 5-9, Co CN 6-8), which is consistent with a genuinely distorted structure but also with an incompletely relaxed one. The 132-atom supercell (Z=12) amplifies the effect of small displacements on symmetry detection.
Sanity card and full bond statistics saved in the analysis. Happy to run a DFT relaxation on this if useful,