Fe17W3 beside its anisotropy anchors: what the 12.0 MJ/m³ L1_0 FeW MAE does and does not establish

Question. The GO-branch decision for Fe17W3 rests on a tier-2 MAE that was measured not on the candidate but on a 2-atom L1_0 FeW anchor (12.0 MJ/m³, action 01a076ac). What exactly does that anchor buy, and what is still missing before the 20-atom candidate's anisotropy is known?

Answer. The anchors establish three things: the MAE route resolves real anisotropy with a numerical floor three orders of magnitude below the program target, the W-5d mechanism clears the 1.5 MJ/m³ target in two independent small cells, and the route is calibrated against a literature-known value. They establish nothing about Fe17W3's own MAE, easy axis, or ferrimagnetic sublattice behavior, all of which differ structurally from every anchor. That missing number is one large-cell MAE run, specified exactly in the capability request to@apollo.

The comparison

Measured route outputs only. Blank means not run, not zero, not estimated. Every value carries its producing action in the receipts section.

All MAE runs share settings: tb2j, DZP PBE, ecutwfc 65, kspacing 0.16, scf_thr 1e-6, mp smearing 0.05. The Fe10Ni6 survivor never earned an MAE: it was gated at phonons (min frequency -3.43 THz, action 01a0721a), which is exactly what the gate is for.

What the anchor MAE establishes

1. The measurement chain works. On identical settings the route returns 0.00047 MJ/m³ on a cubic cell where symmetry forbids uniaxial anisotropy and 0.65 to 12.0 MJ/m³ on tetragonal cells. Four orders of magnitude of dynamic range, and the artifact floor (5×10⁻⁴ MJ/m³) is three orders below the 1.5 MJ/m³ program target. Any candidate MAE above ~0.01 MJ/m³ cannot be a numerical artifact of the route.

2. The W-5d mechanism is real in small cells. Two independent W-containing tetragonal anchors, L1_0 FeW (11.9975 MJ/m³) and the Fe3W-motif cell (6.4815 MJ/m³), exceed the 1.5 MJ/m³ target by 4-8x. This is what the pre-registered H5 falsifier tested, and it is the reason the GO branch was selected at the 2026-09-07 checkpoint.

3. The route is not systematically inflated. The L1_0 FeNi anchor returns 0.6477 MJ/m³, inside the literature band for L1_0 FeNi (0.5-1.3 MJ/m³; Paulevé 1968 experiment ~1.3 extrapolated to S=1, uncorrected GGA-DFT 0.47-0.56). A route that overestimated anisotropy would not land there.

What the anchor MAE does not establish

1. Fe17W3's MAE. It has never been measured and no value is imputed. The 12.0 MJ/m³ belongs to a 2-atom P4/mmm cell; Fe17W3 is a 20-atom P-4m2 ferrimagnet with a different W coordination. Magnitude, and even the sign of the anisotropy constant, are open.

2. Easy-axis transfer. In the anchor itself the easy axis flipped: 001 on the compressed MLIP cell (7.2509 MJ/m³, action 01a07666), 010 on the relaxed cell (11.9975, action 01a076ac), tracking c/a. Fe17W3 has a different c/a and W ordering, so its easy axis is unknown.

3. Ferrimagnetic sublattice behavior. In Fe17W3 the W moments sit antiparallel to the Fe sublattice (-0.81 to -0.86 μB, perturbed pair). Neither W anchor contains an antiparallel W sublattice, and sublattice contributions to MAE can partially cancel. Nothing in the anchor set tests this.

4. Size convergence. MAE per volume is not guaranteed converged between 2-4 atom and 20-atom cells; the effective k-mesh density per atom differs and no convergence study spans the gap.

5. A stress-free absolute value. The 12.0 MJ/m³ anchor carried 39.3 kbar residual anisotropic stress after relaxation. The robust statement is the bracket, 7.25 to 12.0 MJ/m³ across two cells, not any single number.

Blocked entry, per the citation rule. A kspacing-0.08 mesh check on the Fe3W anchor infra-failed at K=1410 (action 01a07853) and is excluded from the evidence base: both anchors are cited at kspacing 0.16 only. The disk-full failure that blocked it was cleared by @mmoderwell; the check has not been re-run and nothing here depends on it.

Where this leaves Fe17W3

The candidacy currently rests on tier-1 evidence (e_hull 12.9 meV/atom, CHGNet Ms 1.74 T, Tc 780 K, $13.24/kg, clean MLIP phonons at zero margin, FM ordering preference of 0.140 eV/atom confirmed under numerical perturbation) plus a mechanism argument from the anchors. The missing number is one MAE run on the validated CIF. The capability request

Receipts

• Fe17W3: e_hull GGen 01a07478, independent replication 01a077e1; Ms/Tc/cost 01a074af

Full candidate state lives in Rare-earth-free magnet candidates; the evidence dossier is here