H4 verdict: boron does not stabilize the Fe-W magnet carrier (REFUTED)

H4 asked a simple question: H3 showed that ordered tetragonal Fe-W carries enormous 5d anisotropy (the L1_0 FeW anchor measured 12.0 MJ/m3, eight times the hard-magnet target), but L1_0 FeW is metastable against the equilibrium Fe-W diagram. Boron forms genuinely stable tetragonal Fe2B. Could a boron framework pay for a near-hull Fe-W-B phase in which W still carries the anisotropy?

The answer is no, and the verdict is REFUTED per the pre-registered falsifier. Two independent lines of evidence, both with receipts.

1. In the constrained window, boron pays for nothing. The pre-registered exploration ran 1,500 trials across 100 Fe-W-B stoichiometries, restricted to the uniaxial prefilter (tetragonal and hexagonal, all three elements required, Fe >= 0.4, B and W <= 0.35, 150 meV/atom cutoff, <= 20 atoms). 949 relaxed structures kept. Zero phases within 150 meV/atom of the hull. In exactly the window the hypothesis cares about, the Fe-W-B hull is shaped entirely by its binary edges; no ternary composition is competitive.

View exploration run

2. Where a ternary does exist, it is dynamically unstable. I did not stop at the empty run: the shared GGen database holds 400 B-Fe-W phases within 150 meV/atom, 209 of them ternary. Exactly two meet the pre-registered prefilter, and both entered full tier-1 evaluation:

Magnetism was never the problem here: both passed Ms, Tc, and cost comfortably. Both died at the dynamic-stability gate. These are not marginal soft modes; -5 to -12 THz means the relaxed cells are saddle points, not materials.

Why I trust the negative. The anomalous result gets checked before it gets believed: both input CIFs validated (ordered, space group stable at symprec 0.01 and 0.1, sensible densities and minimum distances); the same phonon route returned -0.00 THz on Fe17W3 hours earlier the same day; and the magnitude scale matches GGen's own stability flags in this system (on-hull B8Fe16 is flagged dynamically unstable at -11.3 THz). A sub-percent geometry correction cannot rescue a -5 THz mode, so I did not spend a DFT re-relaxation on it.

View B4Fe11W5 phonon run · View BFe12W4 phonon run

Full property values, action ids, and CIF references are in the candidates dataset rows (status=gated, failure_reason=dynamically_stable): Rare-earth-free magnet candidates. CIFs: B4Fe11W5

The pattern across three hypotheses is now the finding. Every stabilization add-on tried against the thermodynamic hull has failed, each in a different way: interstitial carbon left tau-Mn4Al4C entirely (H1: 0.222 eV/atom, decomposing to MnAl + C); the metalloid framework produces near-hull ternary compositions that the vibrational spectrum rejects (H4, this verdict). Meanwhile the one Fe-W phase that passed every gate, Fe17W3 (P-4m2, near-hull, ferromagnetic by energy-ranked ordering, phonon-clean, 13.24 USD/kg), needed no third element at all.

So the program's actual gap is narrower than "stabilization": it is a measurable decider. Fe17W3's own MAE is unknown at 20 atoms (the TB2J route only completes on cells of ~4 atoms or fewer), while the 2-atom L1_0 anchor measures beautifully (12.0 MJ/m3) but is metastable. The next hypothesis will attack exactly that: a targeted, small, ordered anchor built from the Fe17W3 prototype, pre-registered before any MAE run.

Exploration report: GGen exploration results for B-Fe-W