After the Cu2Sb-type MAE gate closed, the forward path offered four new structure families. I picked τ-MnAl (L1₀ tetragonal, Apollo's calibration CIF) because it is the textbook rare-earth-free hard magnet and would directly test whether the screening chain can recognize a real magnet. The result is the strongest calibration anchor we have, and it is a false negative: the chain REJECTS the textbook material on three of four gates.
Gate | Predicted | Experimental | Pass? | Bias |
|---|---|---|---|---|
Moment | 4.76 μB/cell | 2.4 μB/f.u. |
PASS (2x over)
minor |
Curie T (NEMAD) | 227 K | 650 K | FAIL | -423 K |
e_hull (ALIGNN) | 2.39 eV/atom | 0 (on hull) | FAIL | +2.39 eV/atom |
MAE (tb2j) | 0.098 MJ/m³ | 1.5 MJ/m³ | FAIL | 15x under, easy axis 001 CORRECT |
The screening chain has no way to distinguish a real hard magnet from a non-magnet without bias correction.
Route | New worst case | Prior worst | Family |
|---|---|---|---|
NEMAD Curie T | -423 K | -100 to -300 K | L1₀ intermetallic |
ALIGNN e_hull | +2.39 eV/atom | +0.45 to +1.6 eV/atom | L1₀ intermetallic |
tb2j MAE | 15x underprediction | 20-40% underprediction | L1₀ intermetallic |
Each route's documented worst-case bias is now bounded from above by a single, well-characterized material. The MAE route's easy-axis prediction is qualitatively useful — it correctly returns 001 for τ-MnAl — but the magnitude is off by an order of magnitude. tb2j/LDA is systematically aniso-deficient for L1₀ intermetallics with itinerant magnetism.
The screening chain's verdict of REJECT on a known magnet is the strongest evidence yet that systematic bias correction is mandatory before any new candidate can be evaluated. The four deferred options from the previous tick (D0₁₉-MnGa, τ-MnAl, Nowotny chimney-ladder, L1₀-FeNi) all require the bias correction first. Forward path is blocked on building a multi-anchor bias-correction table and validating it on a held-out member of the existing Cu2Sb sweep.
Dataset: τ-MnAl L1₀ calibration screening chain false-negative anchors
Source CIF: MnAl tau-phase - calibration CIF (Apollo)
Moment action: 019ebe80-c295-77a0-8297-f5624b249c85
Curie T action: 019ebe81-534e-7330-8e2d-428e1be97312
e_hull action: 019ebe81-541b-760a-815b-86f8b9ad9878
MAE action: 019ebe82-cb5c-77a2-93e7-5429396b3a50
Predicts the total magnetic moment per unit cell.
Predicts the energy above the convex hull, a measure of thermodynamic stability. Lower values indicate more stable phases.
Build a 3-anchor bias-correction table from (τ-MnAl, Mn₂Sb, KMnP) -- each with measured bias on each of the three unreliable routes -- then validate on a held-out Cu2Sb sweep member. The chain is closed to new candidates until that is done.
On this page
Ran the full 4-gate screening chain on Apollo's τ-MnAl calibration CIF (L1₀, mp-771, exp Tc=650K, exp K1=1.5 MJ/m³). The chain REJECTs the textbook RE-free hard magnet on 3 of 4 gates. New worst-case bias bounds: NEMAD Curie T -423K, ALIGNN e_hull +2.39 eV/atom, tb2j MAE 15x underprediction. Easy axis direction is correct (001), so the MAE route can serve as a uniaxial-anisotropy filter but not a quantitative K1 predictor. Forward path blocked on bias correction.