Ouro's shared crystal database now holds GNoME's 554,054 stable materials alongside Materials Project and Alexandria: 6.8 million structures in one place, on one energy scale, behind four free routes.
Ouro's shared crystal structure database: Materials Project, Alexandria and GNoME with PBE energies, structures generated on Ouro, and structures people submitted, with Orb v3 energies, MP2020-corrected. Check a CIF's energy above the hull, check whether it is already known, add it if it is new, or get a published material's CIF by id. GNoME and Materials Project's GNoME-derived records are CC BY-NC 4.0 (non-commercial use only), so only these free routes use them.
Source | Structures | License |
|---|---|---|
Alexandria (PBE, 2025.07.02) | 5,777,916 | CC BY 4.0 |
GNoME (DeepMind) | 554,054 |
6.48M PBE energies, from the sources themselves.
3.38M Orb v3 energies, covering every published structure within 100 meV/atom of the PBE hull (about 3.0M of them).
Every published structure is stored at the geometry its source published, with the source's own distance to its hull.
Raw energies from different databases don't share a hull as they are. What we do:
We store raw energies and apply MP2020 ourselves. Materials Project and Alexandria agree on raw PBE to a median 8 meV/atom on matched structures, but their published corrected energies differ by a median 42 meV/atom. So we recompute MaterialsProject2020Compatibility for Alexandria and GNoME, and keep MP's own correction for MP. On a 3,000-record sample, our MP2020 matches GNoME's own exactly for 99.1% of records; the rest differ through oxidation-state guesses.
GNoME needs a pseudopotential shift. For Ga, Ge, Li, Mg and Na, GNoME used different pseudopotentials from MP, and DeepMind publishes per-atom offsets for them. We checked those offsets against Alexandria's PBE energies for the same crystals: GNoME runs high by 0.104 eV per Ge atom (published 0.104) and 0.091 per Mg atom (published 0.092). Every other system agrees within 2 meV/atom. The importer subtracts the offsets.
The check: placed on our own PBE hull of Materials Project + Alexandria + GNoME, GNoME's materials sit a median 0.0 meV/atom above it (90th percentile 9 meV in the shifted systems, 4 meV elsewhere). That's what you'd expect for materials GNoME published as stable.
MP has recomputed many GNoME crystals with r2SCAN only (the mp-32… and mp-33… ids). Those have no PBE energy to share a hull with, so they're not in the database; GNoME's own records cover the same crystals.
Orb v3 (orb-v3-conservative-inf-mpa, arXiv:2504.06231) through TorchSim is the default MLIP surface. Reference energies are single-points at the published geometry. Relaxing first changes them by a median 0.2 meV/atom (p99 3.9), well below Orb's own ~8 meV/atom scatter against PBE.
Prophet is coming soon as a second energy surface. It's currently at the top of the Matbench Discovery leaderboard. Adding a surface means one registry entry plus a single-point pass near the hull, so it will cover the same ~3M near-hull structures.
Add a relaxed CIF to Ouro's crystal database so later hull and novelty checks (yours and everyone's) include it. It is matched against every stored structure with its reduced formula, published or made on Ouro; if one matches, nothing is added and you get the match back. Otherwise it is scored with Orb v3 and stored, linked to your file. The structure must be relaxed with Orb v3 (max |F| ≤ 0.05 eV/Å): relax it first with the Materials API's relax route. Returns added, exists (with the matching structure) or updated (the same file submitted again).
Energy above the hull: Orb v3 or PBE, MP2020-corrected. Near-hull references that lack an Orb energy are filled inline.
Novelty: StructureMatcher on primitive cells against every published structure with the formula. It reports known, new polymorph or new composition, plus matches among structures already made or submitted on Ouro.
Materialize: get any MP, Alexandria or GNoME material as a CIF on Ouro by id or link. The file carries the record's license and credits its source.
Add: store a relaxed structure if it's new, so everyone's later hull and novelty checks include it.
Each was pulled in with the materialize route, and each sits on its PBE hull.
Materials Project: olivine LiFePO4 and B2 FeCo.
Alexandria: the full Heusler MnCo2Sn, and a layered Fe–Co–Ga phase.
GNoME: the cubic hydride LiSrRhH6, and Li2Pd4SnSb (both CC BY-NC 4.0).
GNoME and Materials Project's GNoME-derived records are licensed for non-commercial use only. Only the free Crystal DB routes above use them. Paid routes that read the database (GGen, Matra-Genoa
Stored PBE hull distances are each source's own, measured against its own hull. The e-hull route builds a fresh hull from our MP2020-corrected energies, so its result is the one to compare across sources.
"New" means new to these databases, not to the literature.
Generated structures are stored once per crystal. Copies of a published or earlier structure are linked to it rather than saved again.
The code is open: crystal-db (MIT) holds the schema, importers, corrections, matching and the add-a-source checklist.
Please cite the sources you use: Materials Project (Jain et al. 2013), Alexandria (Schmidt et al. 2023) and GNoME (Merchant et al. 2023).
Materials Project (GGA/GGA+U) | 164,368 | CC BY 4.0 (194 GNoME-derived records: CC BY-NC 4.0) |
Generated on Ouro (GGen, Matra-Genoa) | 288,738 |
Submitted through Ouro's routes | 3,996 | the submitter's |
Assess the thermodynamic stability of a crystal structure by computing its energy above the convex hull of every structure in Ouro's crystal database: Materials Project, Alexandria and GNoME phases, structures generated on Ouro, and structures submitted earlier. Choose the energy surface: Orb v3, where every reference is scored with the model that scores your structure, or PBE, the published DFT energies with your structure scored by Orb v3. MP2020 corrections are applied on both. Upload an already-relaxed CIF so the result refers to that geometry; inputs with max |F| > 0.05 eV/Å are rejected unless is set (which relaxes with Orb v3 first). A structure the database does not have yet is added to it. Also rejects overlapping-atom CIFs unless is set. Returns eabovehull (eV/atom), decomposition products and formation energy; set for an interactive phase diagram too.
Match a CIF against every published structure with the same reduced formula in Materials Project, Alexandria, GNoME and the Crystallography Open Database (COD, experimental structures), using pymatgen's StructureMatcher on primitive cells with volume scaling (so DFT- and MLIP-relaxed versions of one crystal still match, as do supercells). Returns known (with the matching entries, closest first), new polymorph (the formula is published, this structure is not) or new composition (no published structure has the formula), plus any matching structures already made or submitted on Ouro. says whether a match was observed experimentally: a COD entry, or a Materials Project entry MP counts as experimental (with the ICSD collection codes MP matched it to). New means absent from these databases, not from the literature. Rejects overlapping-atom CIFs unless is set.
Look up a material by its Materials Project, Alexandria, GNoME or Crystallography Open Database id (e.g. mp-149, agm005272041, 0000a1b2c3, COD 1000041, or a link to its page) and save its crystal structure as a CIF file on Ouro, at the geometry the source published. Returns the formula and space group and, for the DFT databases, the source's PBE energy and the energy above the PBE hull of every structure in Ouro's crystal database. The file carries the record's license: CC BY 4.0 for Materials Project and Alexandria, CC BY-NC 4.0 (non-commercial use only) for GNoME and Materials Project's GNoME-derived records, CC0 for COD (whose file credits the structure's paper). The CIF credits its source.
Crystal structure of LiFePO4 (Pnma) from Materials Project (mp-19017), at its published PBE geometry. 28 atoms per cell. PBE energy above the hull: 0.0 meV/atom.
Crystal structure of FeCo (Pm-3m) from Materials Project (mp-2090), at its published PBE geometry. 2 atoms per cell. PBE energy above the hull: 0.0 meV/atom.
Crystal structure of MnCo2Sn (Fm-3m) from Alexandria (agm002289224), at its published PBE geometry. 4 atoms per cell. PBE energy above the hull: 0.0 meV/atom.
Crystal structure of Fe2Co3Ga (P-3m1) from Alexandria (agm005782395), at its published PBE geometry. 12 atoms per cell. PBE energy above the hull: 0.0 meV/atom.
Crystal structure of LiSrRhH6 (F-43m) from GNoME (01066bc53f), at its published PBE geometry. 9 atoms per cell. PBE energy above the hull: 0.0 meV/atom.
Crystal structure of Li2Pd4SnSb (R-3m) from GNoME (00e7359459), at its published PBE geometry. 8 atoms per cell. PBE energy above the hull: 0.0 meV/atom.
554k GNoME materials alongside Materials Project and Alexandria: 6.8M structures on one MP2020-corrected energy scale, with free hull, novelty and CIF routes.