H7 (Co-W binary anchor) verdict: REFUTED on the pre-registered moment falsifier. DFT-signed Co moment 0.112 uB/site on the relaxed L1_0 cell, an order of magnitude under the 1.4 uB floor.
Question. H7 (pre-registered in the research ledger before any result was read) asked whether the Co-W couple alone delivers both halves of a hard magnet at anchor level: W 5d anisotropy in a tetragonal crystal field (established on the Fe-W anchors: L1_0 FeW 12.0 MJ/m³, action 01a076ac) and a preserved 3d moment on a Co framework, since W destroys the moment in Fe frameworks (F11: every measured Fe-W compound ≤ ~0.43 T) while Co-dominated Fe-Co-W exceeds 1.2 T.
Answer. Refuted. The DFT-signed Co moment on the DFT-relaxed L1_0 CoW anchor is +0.112 µB/site, an order of magnitude below the pre-registered 1.4 µB/site floor. W quenches the Co moment just as it quenches the Fe moment at 1:1 composition. No MAE was run; per the pre-registration the moment half kills the hypothesis on its own.
Evidence chain.
Anchor generated with SG 123 pinned, 408-step MLIP cell relax, CIF; validated (parse OK, P4/mmm #123 at symprec 0.01/0.1, ordered 1:1, dmin 2.645 Å, ρ 15.26 g/cm³, c/a 1.386).
On the MLIP cell both probes came back with a dead moment: CHGNet Ms 0.0586 T (action 01a08651-0c56) and DFT-signed Co +0.070 µB despite a +3 µB FM seed (action 01a08651-d815).
Anomaly check before believing it: the H3 scar says an over-dense MLIP cell can spuriously quench moments. The known-answer control re-run on H3's compressed FeW cell reproduced the recorded quenched value (0.839 vs 0.716 µB/cell, action 01a08654), so the route is sane and the collapse could have been a compression artifact. The decider therefore moved to a relaxed cell.
DFT variable-cell relax (action 01a08651-0b2b): converged, 17 ionic steps, volume +3.21% (26.42 → 27.27 ų), SG 123 held, forces and stresses converged. The relaxed CIF
Decider, identical settings to the MLIP-cell run, FM seed +3 µB on Co, fresh SCF (action 01a0867d): Co +0.1121 µB, W −0.0282 µB, total 0.0839 µB/cell, Ms 0.0359 T.
The relaxed cell rules out the compression artifact: the cell relaxed outward, stresses are gone, and the moment stayed dead, while the same route returns 2.11 µB/Fe on the relaxed FeW anchor under the same settings. This is an observation about what this route computes on this structure; the interpretation, stated as such, is that W 5d hybridization quenches the 3d moment regardless of whether the framework is Fe or Co at 50 at% W.
What this changes. The moment surviving only at low W fraction (H6's FeCo6W, 12.5 at% W, Ms 1.0291 T) sharpens the program's central tension into a W-fraction trade-off: anisotropy scales with W content, moment with its absence. The next hypothesis must attack the tension directly, carrying anisotropy at low W fraction (a light 5d/4d partner, strain or layering, or a motif where W enters as a small ordered fraction), not with another W-rich binary anchor. The pre-registered Co-W GGen exploration branch is cancelled at verdict: its premise is exactly what the anchor refuted, and a 15-trial scan would spend compute on a dead mechanism.
Full receipt trail: ledger finding F13, candidates dataset row 01a08658-8480 (Rare-earth-free magnet candidates), and the five actions linked above.
error