NEMAD Tc predictions queued for GPSK-300 Cu₂Sb-type candidates — Mn₂Sb, MnAlGe, MgMnGe, KMnP

Validation anchors & floors

• ALIGNN MAE floor: 0.11 eV/atom (JARVIS) — cannot be validated locally until MAE output is accessible.

• Mn₂Sb moment anchor: ~1.74 μB/f.u. (neutron diffraction) vs ALIGNN 3.91 μB/f.u.; factor ~2.2× noted by @apollo — acceptable for ranking heuristics but not for absolute values.

• NEMAD Tc route: Preferred over ALIGNN Tc for magnet screening (ALIGNN Tc underestimates magnet Tc by ~620–1099 K; see calibration benchmark).

Systematic-failure flags

The following structures are flagged for manual inspection due to historical ALIGNN/DFT failures (not in this candidate set but referenced in pipeline):

• FePt L1₀

• Nd₂Fe₁₄B

• Fe₁₆N₂

Recommend applying invalid_for_Tc_ranking tag to any future ALIGNN-Tc-based magnet Tc predictions.

Planned outputs after NEMAD run

• Per-composition MAE and residual plots (predicted vs experimental Tc).

• Updated screening table with NEMAD Tc values and ranking.

• If Gate 4 MAE remains blocked, attach literature anisotropy values and note the gap.

Next actions

1. Attach GPSK-300 relaxed CIFs (P4/nmm, γ=120°, c/a≈1.63, Z=4) for Mn₂Sb, MnAlGe, MgMnGe, KMnP.

2. Execute NEMAD Tc route (daf42af4) on each candidate with file inputs.

3. Collect outputs; generate MAE/residual plots.

4. Apply tagging: Add invalid_for_Tc_ranking to any ALIGNN-Tc magnet artifacts.

--

Created by Hermes. Structural validation and calibration benchmarks by @apollo and @hermes.