The fundamental gap is that no reliable magnetic anisotropy energy (MAE) data exists for MnFeSi-C14 and Fe₂Si-C14 Laves phases, the two most promising quaternary candidates for rare-earth-free permanent magnets. This absence stems from systematic model failures and an experimental literature gap rather than structural impossibility.
ALIGNN systematic bias: ~1.6–2.0 eV/atom overestimate of formation energies produces false thermodynamic instability flags (e.g., MnBi incorrectly predicted non-existent)
GPSK-05 generative failure: Produces P1 triclinic collapses instead of correct symmetry for all tested permanent-magnet prototypes (FePt L1₀, Nd₂Fe₁₄B, Fe₁₆N₂, SmCo)
Orb v3 relaxation artifacts: C14 structures collapse (Z drops 4→2; c/a distorts to 2.36–2.90) during relaxation; useful only for initial generation, not final validation
Experimental silence: Mn₂Si/MnFeSi/Fe₂Si quaternary phases are essentially undocumented in the experimental literature
A curated ICSD-anchored dataset exists and is immediately usable:
Asset: c14_mgzn_type_icsd_calibration_dataset
Contents: 4 experimental ICSD references (TiMn₂, Fe₂Ti, Mn₂Ti, Co₂Ti), 1 binary centroid for quaternary geometry seeding, 4 Orb v3 collapsed negative controls, 3 validated quaternary rebuilds (MnFeSi-C14, Fe₂Si-C14; Mn₂Si excluded)
Three-point geometric gate (post-relaxation validation):
γ = 120°
c/a ∈ [1.60, 1.68]
Z = 4 (correct stoichiometry)
Current DFT-based energetics (via MP energy-above-hull):
MnFeSi-C14: +3.506 eV/atom above hull (unstable by ALIGNN/DFT proxy)
Fe₂Si-C14: +3.271 eV/atom above hull
Note that these energetic values reflect parent-phase decomposition references; the geometric validation confirms the structures are sound C14 lattices suitable for property calculation.
Platform routes ready for submission with the validated CIFs:
DFT MAE calculation: route 1254eec1
Curie temperature prediction: route daf42af4
Saturation magnetization: route d1fdf6d1
ALIGNN moment (use with caution): route 7aaa92c1
Action: Submit MnFeSi-C14 and Fe₂Si-C14 CIFs from the calibration dataset directly to the DFT MAE route. These structures have passed the geometric gate and are ready for physics calculation.
Replace ALIGNN with models that avoid its systematic offset:
MatGL and CHGNet (MIT-licensed; prioritized for Phase 1 deployment per platform guidance)
Materials Project energy-above-hull route as a correction layer for stability screening
Magnetic Structure Analyzer (MSA) or equivalent if available for direct MAE
Rationale: These models align better with Materials Project ground truth and do not exhibit the ~1.6 eV/atom formation-energy inflation that invalidates ALIGNN-based stability ranking.
Mine underexplored sources for isostructural reference data:
Expand ICSD queries to isostructural ternary and quaternary relatives (e.g., Mn₂FeSi-type, Fe₂MnSi-type variants)
Query Materials Project formation energies for ternary Mn–Fe–Si phases not in the current calibration set
Search NOMAD and institutional repositories for unpublished DFT or experimental results on related compositions
Key insight: MnFeSi and Fe₂Si inherit structural stabilization from the Fe₂Ti/Mn₂Ti parent prototypes, making them more viable than binary extrapolation suggests.
If GPSK-300/GPSK-05 must be used, implement a strict validation pipeline:
Pre-relaxation symmetry check: enforce P6₃/mmc, Z = 4, c/a ∈ [0.968, 0.974] (for Th₂Ni₁₇-type gates) or C14 equivalents before relaxation
Post-relaxation gate: discard structures where c/a > 2.0 or Z ≠ 4
Re-relax fallback: Orb v3 for generation only, then re-relax with DFT or high-fidelity MLIP; do not trust MLIP-relaxed properties directly
Calibration anchoring: Every new structure must be geometrically benchmarked against the ICSD calibration set
Critical note: GPSK-05 currently fails across all tested permanent-magnet prototypes, yielding P1 collapses. Treat its output as suspect unless the symmetry gate is enforced.
Today: Run DFT MAE route (1254eec1) on MnFeSi-C14 and Fe₂Si-C14 CIFs from the calibration dataset.
This week: Validate MatGL/CHGNet predictions against the 4 experimental ICSD references to establish baseline reliability before applying to quaternary candidates.
Documentation: Create a decision log tracking which models/routes produce convergent vs. divergent results for these specific compositions.
Risk | Mitigation |
|---|---|
ALIGNA bias propagates to MAE | Use MP formation energy as a correction factor; treat ALIGNN MAE as qualitative only |
GPSK-05 generates invalid structures | Enforce mandatory symmetry validation gate before any property calculation |
Experimental literature gap | Prioritize DFT for these specific compositions; flag for experimental follow-up |
Orb v3 relaxation artifacts | Apply automatic geometric filtering (c/a < 2.0, Z = 4) to discard collapsed structures |
Short-term: Obtain converged MAE values (≤0.1 meV/atom tolerance) for MnFeSi-C14 and Fe₂Si-C14 via DFT route.
Medium-term: Establish a property-prediction pipeline (structure → symmetry gate → DFT/ML → MAE) with ≥95 % geometric validity rate.
Long-term: Generate 10–20 new quaternary candidate structures with predicted MAE > 1 MJ/m³ and thermodynamic stability (E_hull < 0.05 eV/atom after MP/DFT correction).
The dataset and validation framework exist—skip generative structure generation and proceed directly to DFT property calculation on the already-validated MnFeSi-C14 and Fe₂Si-C14 structures. The absence of MAE data is a consequence of systematic model failures, not structural impossibility. Use the ICSD-calibrated geometric gate to filter any new structures, and rely on DFT or higher-fidelity ML models (MatGL/CHGNet) rather than ALIGNN for property prediction. This focused approach is likely to yield actionable MAE numbers faster and more reliably than repairing generative pipelines at scale.
On this page
Systematic gap analysis and actionable alternatives for obtaining reliable magnetic anisotropy energy (MAE) data for MnFeSi-C14 and Fe₂Si-C14 Laves-phase permanent magnet candidates.