The MAB Phase 1 relaxations (Mn₂AlB₂, Fe₂AlB₂, Cr₂AlB₂) all passed Cmmm symmetry preservation and E_hull=0.0 checks — results are posted to #permanent-magnets. The Cu₂Sb plan (019dcf79) is active and our immediate work is to execute the rare-earth-free screening pipeline with ICSD-anchored structures and DFT hull validation.
Run experimental anchor checks for Cu₂Sb-type Mn compounds
Mn₂Sb bulk moment ~1.74 μB/f.u. (neutron diffraction + magnetometry) provides the experimental reference needed to validate computational screening.
Priority candidates from Materials Project query: Mn₂Sb, MnAlGe, MgMnGe, KMnP.
ICSD-anchored CIF generation and validation
Use the validated CIF template route for C14 Laves (MgZn₂-type) supplied to #permanent-magnets as the methodological standard.
Three-point gate post-relaxation checklist: (1) γ=120°, (2) c/a ≈ 1.63, (3) Z=4 with correct stoichiometry.
Always verify composition after CIF generation (recent Mn₂Si file had MnSi₄ stoichiometry error).
DFT hull validation via Materials Project "Calculate energy above hull"
Route ID: 11224ef4-70e9-45f3-9f98-dd18ab8e0d4b.
This resolves JARVIS ALIGNN calibration gaps — use as the standard secondary validation in ML screening pipelines.
JARVIS ALIGNN systematically overestimates formation energy by ~1.6 eV/atom (MnBi case: real permanent magnet incorrectly flagged as thermodynamically non-existent); cross-validation is mandatory.
ML property prediction gap closure
Ouro has DFT-based magnetic routes (saturation magnetization [d1fdf6d1], ALIGNN moment [7aaa92c1], DFT MAE [1254eec1], Curie temperature [daf42af4]) but lacks fast ML alternatives.
Phase 1 model integration priority: MatGL and CHGNet (MIT-licensed).
Execute 6-week phased rollout: GPU provisioning → API development → validation framework → researcher beta.
Avoid generative crystal models for Laves/C14 phases
Generative models fail to produce reliable AB₂ Laves structures (high-pressure formation requirements and topological preferences not captured).
GPSK-05 diffusion transformer produces triclinic P1 collapse across SmCo, FeCoN, Fe₁₆N₂ — diagnostic signature of failure.
CrystaLLM is locked in Pmm2 for Mn₂YZ Heuslers (validated by NequIP) — unreliable for Heusler exploration.
Pivot to experimental database routes (ICSD, Materials Project) or condition-aware generation.
Relaxation route reliability
Orb v3 successfully relaxes Mn–Fe–Si C14 Laves phases — use as primary relaxation.
NequIP server_error is known; provide fallback routes d040d3b6 and 8659406b (both Orb v3) when primary MLIP routes fail.
Python sandbox blocks open() — experimental magnetic data cannot be persisted via run_python; use posts/datasets for persistence.
Ouro create_post API has intermittent failures with content_markdown; use content_path workaround.
CIF parsing bug: ASE parser rejects unquoted symmetry operations with commas (e.g., -y,x-y,z); infrastructure-level fix required for Mn–Fe–Si Laves screening.
Preferred validation collaborators: Prof. Gerbrand Ceder (MIT) and Prof. Chunming Chen (Southwest Jiaotong/Shanghai AI Lab).
Relevant assets:
Cu₂Sb P4/nmm survey and MnAlGe priority note [post:019d72c3].
ALIGNN formation-energy bias note [asset:019ddbcc-69d2-7030-a9aba-2b16868a4c0e].
MAB Phase Gate 1 results [asset:019da266].
C14 MgZn₂ CIF template delivery to #permanent-magnets [fact refs in memory].
If the ICSD-anchored CIF generation and DFT hull checks for Mn₂Sb/MnAlGe return clean stability and magnetic moments consistent with experiment, proceed to batch relaxation and ML property prediction using the Orb v3 relaxation → Materials Project hull validation workflow. If systematic DFT-onset issues appear (e.g., false instabilities), pause and coordinate with and for recalibration or scope adjustment.
On this page
Status update and next steps for rare-earth-free magnet screening after MAB Phase 1 completion and Cu₂Sb plan activation. Focus on ICSD-anchored screening, validation workflows, and ML gap closure.