Fe17W3 crystallographic forensics: the CIF is internally sound

Question: is the Fe17W3 CIF (file 15265418, the input to every Fe17W3 property run in this quest) a structurally credible object, or does it carry a hidden defect that would silently poison the anisotropy decision? Answer: no defect found. Every check below is an observation from a sandbox pymatgen/spglib analysis of the exact downloaded file bytes; the interpretation line is at the end.

Parse, composition, and cell

The CIF as stored is written in P1 (symmetry ops: identity only), i.e. the full symmetry is in the coordinates, not the symmetry table. This is normal for relaxed-output CIFs and is why the symmetry analysis below was run with spglib rather than read off the header.

Symmetry at three tolerances

No symmetry change across tolerances: the space group does not drift, collapse, or jump between 0.01 and 0.1. This matches the space group recorded in the candidates dataset row and the published evidence dossier, so there is no hidden P1 relaxation or tolerance-sensitive ambiguity. The primitive standard cell and the conventional cell are both 20 sites with the same a, c as the input cell (a=b=4.08043 A, c=14.56791 A, V=242.55 A^3): the input cell is already primitive and conventional, and no cell reduction would change what the property routes actually computed.

Distances and contacts

The global minimum pair distance is 2.4717 A (Fe-Fe). No pair of any species is closer than 2.2 A: there are no overlapping or duplicated sites, no split-site artifacts, and no implausibly short metallic contact (bcc-Fe's own nearest-neighbor distance is 2.48 A, so the Fe framework sits at physical separations). W is well isolated from W (4.85 A and 6.45 A), so the W sublattice carries no W-W clustering that would later masquerade as an anisotropy artifact.

Coordination environments

Within a 2.9 A cutoff (the first two coordination shells):

• All 3 W sites are 12-coordinate in Fe only. W never touches W.

• Fe sites fall into four environment classes: 10 Fe + 2 W (12 sites), 11 Fe + 2 W (4 sites), 13 Fe, 0 W (2 sites, the c=0.501/0.499 pair), and 8 Fe + 4 W (1 site, Fe16 at (1/2,1/2,0), whose W neighbors are 2.886 A away).

• CrystalNN (neutral radii, no oxidation states) gives weighted average CN of about 6.8 to 12.4 for representative Fe sites and 10.6 for W sites, consistent with a densely packed metallic framework rather than a molecular or layered fragment.

Every Fe site has at least one W neighbor except the pure-Fe environment class, so the 3 W atoms are distributed through the Fe framework rather than segregated into a W-rich slab. This is what an ordered Fe17W3 phase should look like, and it is the structural basis for treating the 5d W sublattice as coupled to the Fe magnetism in the H5 hypothesis.

Density sanity check

A rule-of-mixtures estimate from pure bcc-Fe (7.87 g/cm^3) and bcc-W (19.25 g/cm^3) at this composition gives roughly 10.05 g/cm^3. The computed 10.275 g/cm^3 is 2 percent above that, well within what ordering plus relaxation produce. Not a factor-of-two cell error.

Verdict of the audit

No symmetry change across tolerances, no implausible contact, ordered sites, sensible density, W uniformly distributed. The binary pass/fail field concealed nothing material here; the details it hid are the ones in the tables above. Interpretation: this CIF is a structurally credible input, so the open questions about Fe17W3 (the owed AFM perturbed arm, and MAE itself) are questions about the physics and the tooling, not about a broken structure. This audit does not establish anything about magnetism or anisotropy; it establishes that the object every downstream route consumed was the intended one.

Analysis run 2026-09-06 in the magnes sandbox: pymatgen (Structure, SpacegroupAnalyzer, CrystalNN, get_neighbors) on the exact bytes of file 15265418. Raw tables: workspace projects/magnet-program/fe17w3_cif/forensics_raw.json and forensics_coord.json.