Map
Every quantity is laid out by tier. Select one to see its formula, what it is built from, what it feeds, the codes that compute it, and its Lean dimension where one exists.
Open the mapA versioned map of physics.
OpenMaterials is a typed graph of physical quantities and the formulas that relate them. Representations of 33 codes and workflows connect their outputs to its nodes, and values from simulations and measurements attach to those nodes with their provenance and a content hash.
- Map version
- 808a954e386e
- Quantities
- 137
- Codes
- 33
- Values
- 90, 6 of them measured
Fig. 1Silicon near 300 K, every value on the map
Where it fits
- Materials Project, AFLOW, and JARVIS
- Each publishes a database of computed materials properties. You can contribute a value from any of them, with its source, when its quantity is on the map and the database's license permits it.
- NOMAD
- A repository of uploaded calculations and experiments, with their files. The map stores no raw files, and a record can point to wherever their owners keep them.
- OPTIMADE
- A standard API for querying structures across materials databases. OpenMaterials publishes the typed quantities that values attach to, with their formulas, as JSON files.
Tools
Tracer
Select a quantity to list each step that builds it and each quantity it feeds. Shift-click a second quantity to get the shortest path between them.
Open the tracerPlayground
Open a lineage record, resolve a method name to map nodes, or compare two materials by composition.
Open the playgroundLearn a paper
Drop a PDF and the parser proposes its reported values, each with a verbatim quote and page number, for you to accept or reject.
Open the Learn tabEvidence
- Cross-code agreement
- Values of one quantity from different methods or codes, with the spread of each group.
- Lineage tour
- One QHGK thermal conductivity result for amorphous silicon, from its derivation on the map to its lineage record.
- Lean
- 85 operator formulas carry a compiled Lean proof under stated conditions, anchored to Physlib where it has a model, and 132 node dimensions are declared, in one package at one pinned Physlib revision.
- Formalization roadmap
- Every operator on the map, rated by what a Lean theorem about its formula would take: a closed-form definition, a formula that needs analysis, or an external code output with nothing to prove.
Build
- Guide
- Install the omai kernel, resolve a label, load the graph, and parse a paper.
- Document
- The full specification, in HTML and PDF.
- Codes
- A page for each code the map cites: what it is, what it computes, its submap, the papers that describe it and the studies that use it. The list carries each code's interface, citation, DOI and license.
- Data files
- The main JSON files under
/data/, with their shapes.
The map
- Quantities
- 137 (123 observable, 11 hidden, 3 parameter)
- Links
- 313 (270 formula, 43 parameter)
- Operators
- 134
- Tiers
- 18
Each quantity is a typed node. An observable carries no gauge freedom, so codes given the same inputs should agree on it. A hidden quantity depends on a gauge choice (an eigenvector phase, or the basis inside a degenerate subspace). It is an intermediate, like the eigenvectors, or an approximation, like the RTA thermal conductivity, and codes can differ on it. An operator computes a quantity from its inputs: formula links carry its symbolic formula, and parameter links supply inputs such as the cell volume. The nodes are grouped into tiers, such as Sources, Harmonic, Transport, Ground state, and Thermochemistry.
A code representation records how one code's outputs map onto these nodes, with units and conventions. Representations exist for kALDo, phono3py, ShengBTE, LAMMPS, GPUMD, Quantum ESPRESSO, VASP, and the other codes on the codes page, each with its own page.
Values
Values attach to nodes without changing the graph. A value is a small record: the node, the material, the conditions, the number with its units, and the source. Its id is the SHA-256 hash of the record's lineage object, so the same claim always gets the same id. When values from different methods or codes answer the same question, the agreement page reports their spread. Values come from published papers and from runs of kALDo, phono3py, Quantum ESPRESSO and xtb.
Seed set: 90 values, 84 from simulations (mostly the maintainers' papers and example runs), 6 measurements.
Protocol
Every node, formula link, and value has a content hash. A change to the graph is appended to the map's change log after the gates check identity, connectivity, duplicates, and the dimensions of each formula whose symbols are registered, and after a maintainer reviews it. The map version hashes the head of that log together with the record-format rules, so citing a version pins both.
Contribute a value
- 01Write the file by hand, or call
omai.map_data.record_instance, which writes it and computes the id. - 02Add it under
docs/data/instances/and open a pull request. - 03Automated checks confirm that the node exists on the map and that the id recomputes from the lineage, and a maintainer reviews the source. The map version does not change.
The playground's Learn tab exports the values you accept from a paper as JSON.
{
"id": "f735a05c14db39676b46e7f1100c55152afb7bf49cea79c9e1f61b8de075c619",
"kind": "simulation",
"lineage": {
"node": "ThermalConductivity[bte_solver=direct_inverse]",
"material": "Si (diamond)",
"conditions": {"T": 300, "statistics": "quantum", "method": "direct inversion of scattering matrix", "q-grid": "19x19x19", "potential": "DFPT LDA"},
"values": {"value": 147.0, "units": "W/(m K)"},
"source": "paper:kaldo-2020-barbalinardo"
},
"source": {
"kind": "simulation",
"ref": "paper:kaldo-2020-barbalinardo",
"detail": "which gives κQ,Inv = 147 Wm−1K−1 (p. 6). G. Barbalinardo, Z. Chen, N. W. Lundgren, D. Donadio, Efficient anharmonic lattice dynamics calculations of thermal transport in crystalline and disordered solids, J. Appl. Phys. 128, 135104 (2020), doi:10.1063/5.0020443, arXiv:2009.01967."
}
}
Governance
OpenMaterials is stewarded by OpenMaterials-AI, a foundation in formation. The initiative holds the map, the protocol, and the openmaterials.ai name, and changes them only through the gated contribution process. The rules are public in GOVERNANCE.md.
- Maintainer
- Giuseppe Barbalinardo
- Contact
- [email protected]
- Decision making
- A small set of scientific authors reviews changes behind the gates. GOVERNANCE.md names the next step once external contributors arrive: a technical steering group drawn from contributing groups.
- Licenses
- Map data is CC BY 4.0 and code is Apache 2.0.
- Contributions
- Contributors grant OpenMaterials-AI a non-exclusive license to redistribute each record under CC BY 4.0 and keep every other right. There is no copyright assignment and no contributor license agreement.
- Credit
- Reuse credits the map version and, through each record's provenance, the original source.
- Removal
- A rights holder who believes a record entered in error opens an issue; the record leaves the current view of the map, and the append-only log records the retraction.
- Openness
- Anyone can clone the repository, rerun the gates, and rebuild the Lean proofs.
- Tools
- Da Vinci Labs builds the site, the map views, the tracer, the playground, the codes bibliography, the paper parser, and the codebase readers. Proposals from the parser and the readers enter the map only after the gates and a maintainer accept them.
- Compute
- MaterialsCodeGraph, from Da Vinci Labs, runs these calculations for AI agents over MCP, with a binding token quote before every run and a permanent record after it.
Cite the map version you use: 808a954e386e.
Barbalinardo, G. openmaterials.ai: a versioned, content-addressed map of physics. Map version 808a954e386e. https://openmaterials.ai, accessed 2026-10-04.
@misc{openmaterials-808a954e386e,
author = {Barbalinardo, Giuseppe},
title = {{openmaterials.ai}: a versioned, content-addressed map of physics},
url = {https://openmaterials.ai},
note = {Map version 808a954e386e, accessed 2026-10-04}
}