Build the definitive dataset of rare-earth-free permanent magnet candidates on the Ouro platform, with references to CIF files and all existing calculated properties.
Wide dataset — one row per unique compound/CIF, with file reference columns to both the unrelaxed and relaxed CIF (if a relaxed version exists).
Prefer relaxed CIFs — if a relaxed version of a CIF exists on the platform, use it as the primary file reference. Keep the unrelaxed version in a separate column. Only report properties from relaxed structures when relevant.
Magnetic properties — from Estimate magnetic moments and Ms from a CIF and MAE from Calculate magnetic anisotropy energy. Only from existing route executions — do NOT run new calculations.
Stability — energy above the convex hull from Calculate energy above the convex hull. Dynamical stability from Calculate phonon dispersion and band structure if they exist.
NO ALIGNN predictions — exclude any results from the ALIGNN route (de2d96c5). Only use: Estimate magnetic moments/Ms, Calculate MAE, Calculate energy above hull, Calculate phonon dispersion, Relax a crystal structure, and the new DFT routes (Ground-state SCF, Magnetic moments, Magnetic anisotropy energy, Exchange couplings TB2J, Band structure, Band gap, DOS, PDOS).
Properties will be sparse — that's expected. Collect everything we already have; don't run new calculations.
Complete platform CIF compilation — 4,319 unique RE-free magnetic formulas from 12,291 CIF files
CIF calculated properties — 88,050 rows of extracted route action results
Use ouro-py to:
Query the CIF compilation dataset for all compounds and their CIF file IDs
For each compound, identify if a relaxed version exists (filename contains "relaxed")
Query the calculated properties dataset for non-ALIGNN route results on those CIFs
Build the final wide dataset with: formula, compound, space group, crystal system, team, unrelaxed_cif_id, relaxed_cif_id, Ms (emu/g or A/m), magnetic moment (µB), MAE (eV/formula), energy above hull (eV/atom), phonon stability (stable/unstable), Curie temp (K), and action_id references for each property.
3 open3 of 3 resolvedOpenedClosed after 19 minutes
Audited both source datasets. CIF compilation (019f68b5) has 4,324 unique RE-free magnetic compounds from 12,291 CIF files. Calculated properties (019f6b44) has ~88,050 route action result rows. After filtering out ALIGNN (route de2d96c5), 86,430 non-ALIGNN rows remain across 17 route types. Identified that 3,586 compounds have relaxed CIF versions and 4,260 have unrelaxed versions. The Estimate magnetic moments and Ms from a CIF route (d1fdf6d1) was not captured in the calculated properties dataset, but the DFT MAE route (1671b2d5/331d9faf) provides totalmagneticmomentuB and saturationmagnetizationAper_m for 2,070 compounds.
Built the comprehensive dataset with 4,324 rows and 29 columns. Created as dataset 019f7111-6b38-7c6d-a962-ffacd979a94c on permanent-magnets team. Includes 3 file reference columns (unrelaxedcifid, relaxedcifid, primarycifid) and 5 action reference columns (momentsactionid, maeactionid, hullactionid, phononactionid, curieactionid). All ALIGNN predictions excluded. Properties are sparse as expected: 47.9% have magnetic moments, 47.8% have MAE, 58.9% have energy above hull, 59.1% have Curie temperature. Fixed column type issues from CSV inference.
Published dataset to permanent-magnets team. Verified schema: all file and action reference columns have proper FK constraints. Spot-checked 10 rows with full property coverage (magnetic moment + MAE + hull energy + Curie temp). Data looks correct — values match expected ranges for RE-free magnet candidates. Reply posted to request post 019f7102.
cc:
Quest complete — all 3 items done.
The definitive RE-free permanent magnet candidate dataset is live: RE-free permanent magnet candidates: properties from route calculations
4,324 compounds × 29 columns, joining the complete platform CIF compilation with non-ALIGNN route calculation results from the CIF calculated properties
3 file reference columns (native FK): unrelaxed_cif_id, relaxed_cif_id, primary_cif_id — relaxed preferred when both exist (3,586 of 4,324 compounds have a relaxed version)
5 action reference columns (native FK): moments_action_id, mae_action_id, hull_action_id, phonon_action_id, curie_action_id — each links to the specific route execution
10 property columns: magnetic moment, Ms, MAE + easy axis, energy above hull, Curie temperature, band gap, SCF energy, phonon stability, thermodynamic stability
ALIGNN route de2d96c5 fully excluded (1,620 rows filtered from 88,050 total)
No new calculations run — everything harvested from existing route executions
~48-59% coverage for core PM properties (moment/MAE/hull/Curie). Phonon, band gap, and SCF columns are empty (those routes were barely used on platform CIFs) but ready to populate.
CIF compilation: 4,324 rows from ouro-py files.search() across 13,501 platform CIF files
Audit details are in the item 1 completion note. Full dataset description and caveats are in the prior comment on the request post.