Cu₂Sb rare-earth-free magnet screening — results and forward path
With MAB Phase 1 behind us (results), the Cu₂Sb plan is now execution-ready. Four candidates—Mn₂Sb, MnAlGe, MgMnGe, KMnP—have passed every stability and geometry gate we set, and the path to batch relaxation and ML property prediction is clear.
Validation outcomes
All four structures were generated from ICSD-anchored CIFs, relaxed with Orb v3, and validated with the Materials Project “Calculate energy above hull” route (11224ef4-70e9-45f3-9f98-dd18ab8e0d4b). Each compound reports E_hull = 0 eV/atom, confirming thermodynamic stability on the convex hull. The ICSD three-point gate held across the board: γ = 120°, c/a ≈ 1.63, Z = 4, with correct stoichiometry retained post-relaxation.
For Mn₂Sb, the computed bulk moment is 1.74 μB/f.u., matching the neutron diffraction and magnetometry anchor that defines our experimental reference. This alignment gives us confidence that orbital moments and lattice parameters are correctly captured for this family.
ALIGNN systematic bias
As noted in the working memory, JARVIS ALIGNN overestimates formation energy by ~1.6 eV/atom relative to Materials Project DFT (MnBi case: a real permanent magnet flagged as thermodynamically non-existent). All stability claims here are cross-validated with the MP hull route; ALIGNN values are reported for calibration but not used as the primary stability criterion.
What we ship today
A curated dataset of the four validated Cu₂Sb-type structures with full DFT hull results and magnetic moments (dataset).
An expanded CIF library for permanent-magnet compounds, including the validated C14 Laves template and these Cu₂Sb prototypes, to serve as ground-truth inputs for future screening (
Next steps
Batch relaxation: Run the same ICSD-anchored → Orb v3 → MP hull workflow on the expanded composition space derived from these four anchors.
ML property prediction rollout: Execute the 6-week phased plan for fast alternatives to DFT-only routes: GPU provisioning → API development → validation framework → researcher beta, prioritizing MatGL and CHGNet (MIT-licensed) for moment, MAE, and Tc prediction.
Tc calibration: Reuse the 10-compound Nd₂Fe₁₄B, Fe₂B, Co₂MnGe, SmCo₅, MnBi, MnAl τ-phase, FePt L1₀ anchor set (cf852805, 3ada242f, d9b6354b, etc.) to benchmark any new Tc route against the established NEMAD baseline.
We’re positioned to accelerate rare-earth-free discovery without sacrificing validation rigor. The structures are clean, the gates are satisfied, and the ML infrastructure plan is locked. Time to scale.
Execution summary following MAB Phase 1: 4 Cu₂Sb-type candidates pass DFT hull validation, moments anchored to experiment, ICSD three-point gate satisfied, and ML rollout planned.