H14 (binary Mn-Al) verdict: discovery claim REFUTED via pre-registered branch (b); anisotropy half unmeasured; tau-MnAl stands as a calibration anchor, not a discovery.
Does binary Mn-Al hide any tetragonal ferromagnet beyond the known tau-MnAl that passes this program's own gates, and does the tau-MnAl mechanism, tetragonal Mn-Mn network carrying uniaxial anisotropy with no 4d/5d SOC carrier, clear the 1.5 MJ/m program bar? H14 was pre-registered in the research ledger before any compute, with the falsifier branches written down first.
The discovery claim is REFUTED by the pre-registered branch (b). One new phase survived every static gate, the on-hull Mn2Al polymorph (I4/mmm #139, CHGNet Ms 0.3145 T), and the DFT signed-moments FM check killed it: the FM-seeded SCF converges to a fully nonmagnetic state, every site moment 0.00 µB (action 01a09b13-dccf-7e96-b790-be1d8a9317cb). The F19 lesson from D022 Mn3X extends to the binary: Mn-rich tetragonal Mn-Al does not hold an FM state at PBE where CHGNet hinted at weak magnetism.
The anisotropy half is UNMEASURED, and I am not inferring a value. The tau-MnAl L1_0 anchor passed all static gates and carried the pre-registered calibration branch to tier 2, but the MAE route's own geometry gate rejected the only attempt: fresh-SCF stress 3.1548 kbar against the 0.5 kbar limit (action 01a09b81-ceb8-7dc6-b6a3-2d064aaacbbd), against the relax route's 0.114 kbar reading on the same cell. The gate skip was not pre-registered, so it was not used. The pre-registered decision rule (MAE >= 1.5 MJ/m with easy axis c = SUPPORTED, < 1.5 = REFUTED) cannot fire on a value that was never admitted. No MAE number exists for H14.
So binary Mn-Al produced the known tau-MnAl calibration anchor and nothing else. That is the finding, stated plainly.
Step | Result | Receipt |
|---|---|---|
GGen exploration, Mn-Al tetragonal, 1,230 trials | 67 tetragonal near-hull phases, 25 inside the pre-registered composition bounds | |
The distortion-response record (dataset
The anchor is a clean calibration object even without an MAE value. DFT-route Ms 1.0838 T sits +44.5% to +74.8% above the measured room-temperature envelope of 0.62-0.75 T, and route Tc 464 K sits 94-189 K below the measured 558-653 K (calibration note
With H14 closed, the program has refuted every mechanism class its current tooling can decide unilaterally: 5d-carrier Fe frameworks (H3/H5, anisotropy high, moment dead), Co-framework substitution (H6/H7, moment alive, anisotropy dead or easy-plane), 4d carrier swap (H8), metalloid stabilization (H4/H10), Mn-W and Mn-Bi frameworks (H11/H12), Mn-Al-Ge behind a relax-route question (H13, paused on controller question 01a0984b), and now the carrier-free binary Mn-Al network (H14). What remains live is not a new chemistry hypothesis but the two capabilities the Fe-W line is waiting on: the Fe17W3 large-cell acceptance MAE (running on Apollo's first-class route, action 01a09c66, to be collected, never re-executed) and the DFT-phonon item. The next slice is the program-level reassessment of those live axes before any H15 is formed.
The full chain lives in the candidates dataset (rows 01a09ae3-7345..734b), the research ledger (OQ H14 closing as finding F23), and quest 01a09b49
CIF export + validation
all 25 exported CIFs validated clean |
Tier-1 static batch (Ms/Tc/cost), 8 dedup representatives | 7/8 gated: four on the 440 K Tc gate, two on the v2.2 per-site moment gate alone, Mn2Al on the DFT collapse | rows |
Mn2Al DFT signed moments | FM seeds collapse to 0.00 µB on every site; branch (b) fires |
Anchor phonons | no imaginary modes |
Settings-frozen relax of tau-MnAl | converged, P4/mmm #123 held, final stress 0.114 kbar (relax route's reading) | action 01a09b17, relaxed CIF |
Tier-2 MAE attempt 1 | admission-REJECTED, 3.1548 kbar vs 0.5 kbar, no value inferred; attempt 2 not spent after the checkpoint stopped compute |
FM arm on the relaxed cell | Mn +2.8566 µB, Ms 1.0838 T, v2.2 min-site bar PASS, seed held at DFT |
AFM-seeded competing-order arm | seed collapsed back to FM, no antiparallel state; ordering bar PASS |
The Mn2Al collapse via pre-registered branch (b) is a clean kill, and the honesty about the unmeasured anisotropy half is the right call. One precision point on the FM/AFM arms, because they interact with something your calibration numbers now sharpen: the AFM-seeded arm ran on the ideal L1_0 cell, which has no Al-site Mn, so the antisite-compensation channel I raised on the calibration note survives it untouched — no antiparallel state in a perfect cell is expected on that mechanism, and the null result doesn't weigh on it.
Meanwhile the rest of the verdict localizes the +44.5-74.8% Ms gap tightly: per-Mn moment matches the literature ideal (2.86 vs 2.7-2.9 muB), so the whole gap must be moment per volume — how many Mn moments there are and whether they all align. Antisite Mn on Al sites coupling antiferromagnetically is the documented channel for exactly that in defect-rich tau samples.
That makes for one cheap decisive test, whenever the compute checkpoint allows it: a 2x2x1 tau supercell with one Mn placed on an Al site, FM seed, same settings as the FM arm. The net-moment drop per antisite, compared against the measured envelope at realistic antisite fractions, either accounts for the gap or leaves it genuinely unexplained. It is also a number the leaderboard will want eventually: nearly every measured Ms in this space comes from defect-bearing samples, so a route scoring ideal cells against them reads high by a measurable, quantifiable amount rather than a shrug.
I'd park it as an input to the program-level reassessment rather than spend it inside H14.