Materials Science · LeMaterial
LeMat-Bulk
Primary bulk materials database aggregating 1M+ crystal structures with DFT-computed formation energies, band gaps, and elastic properties for materials discovery.
What LeMat-Bulk is
LeMat-Bulk is an aggregated crystal structure database created by LeMaterial that compiles bulk materials data from three primary sources: the Materials Project, Alexandria, and the Open Quantum Materials Database (OQMD). In materials science, training machine learning models or performing screening across disparate density functional theory (DFT) databases is often obstructed by inconsistencies in exchange-correlation functionals, pseudopotentials, Hubbard U corrections, and convergence settings. LeMat-Bulk addresses these discrepancies by standardising records, applying DFT compatibility filtering, and providing structural deduplication across repositories.
By standardising structures into a unified format, it offers computational researchers a consolidated benchmark and training resource for materials discovery, crystal property prediction, and generative materials design.
How it was made
LeMat-Bulk aggregates calculation outputs from structural relaxations across its source repositories. Data fields are mapped to OPTIMADE-compliant specifications, including chemical formulas, periodic boundary conditions, lattice vectors, and Cartesian atomic coordinates. Additional physical properties extracted from calculation outputs include uncorrected energy (in eV), site forces (in eV/Å), full 3x3 stress tensors (in kB, converted from Voigt notation where needed), site magnetic moments, and total magnetization.
To establish compatibility across sources, the creators filter records based on pseudopotential alignment (such as verifying consistency between Materials Project and Alexandria, while accounting for OQMD variations), Hubbard U parameter matching for oxides and fluorides, and spin-polarisation conditions. Structure deduplication is achieved using a material fingerprinting routine: bonding networks are determined via the EconNN algorithm in Pymatgen, converted to a structure graph labelled with species, hashed using the Weisfeiler-Lehman algorithm, and combined with spacegroup symmetry and composition. When duplicate structures are identified, the lowest-energy configuration is retained.
Key facts
- Total materials
- 5,860,446
- Total structures
- 6,725,590
- Compatible PBE structures
- 5,335,299
- Compatible PBESol structures
- 447,824
- Compatible SCAN structures
- 422,840
- Source databases
- Materials Project, Alexandria, OQMD
- Supported functionals
- PBE, PBESol, SCAN
- File format
- Parquet
- Initial release
- December 2024
What you can use it for
- Machine learning interatomic potential training
- Train machine learning potentials on harmonised atomic configurations, Cartesian site forces, and 3x3 stress tensors derived under consistent DFT parameter regimes.
- Crystal property prediction
- Benchmark machine learning models on bulk crystal structures to predict uncorrected total energies, magnetic moments, and magnetization across millions of materials.
- Cross-database materials screening
- Screen vast chemical spaces using pre-filtered subsets, such as Compatible PBE, without mixing incompatible pseudopotentials or inconsistent Hubbard U settings.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from LeMaterial/LeMat-Bulk at the moment you asked. Nothing is cached or stored — every row above came from that request.
How the columns are distributed
Over 5,021,458 rows of compatible_pbe/train
cartesian_site_positions
list- min
- 1
- median
- 6
- mean
- 9.25
- max
- 444
chemical_formula_anonymous
string_text- min
- 1
- median
- 4
- mean
- 4.61
- max
- 18
chemical_formula_descriptive
string_text- min
- 2
- median
- 11
- mean
- 11.52
- max
- 34
chemical_formula_reduced
string_text- min
- 1
- median
- 7
- mean
- 7.31
- max
- 23
cross_compatibility
bool- True100%
dimension_types
list- min
- 3
- median
- 3
- mean
- 3
- max
- 3
dos_ef
float- min
- -14.23
- median
- 4.33
- mean
- 6.31
- max
- 391.85
11.3% empty
elements
list- min
- 1
- median
- 3
- mean
- 3.23
- max
- 9
elements_ratios
list- min
- 1
- median
- 3
- mean
- 3.23
- max
- 9
energy
float- min
- -2749.01
- median
- -32.35
- mean
- -48.48
- max
- 5186.10
entalpic_fingerprint
string_text- min
- 37
- median
- 46
- mean
- 45.85
- max
- 62
0.00% empty
forces
list- min
- 0
- median
- 6
- mean
- 9.05
- max
- 444
functional
string_label- pbe100%
immutable_id
string_text- min
- 4
- median
- 12
- mean
- 11.96
- max
- 13
0.00% empty
last_modified
string_text- min
- 19
- median
- 19
- mean
- 19
- max
- 19
lattice_vectors
list- min
- 3
- median
- 3
- mean
- 3
- max
- 3
magnetic_moments
list- min
- 0
- median
- 6
- mean
- 9.20
- max
- 444
nelements
int- min
- 1
- median
- 3
- mean
- 3.23
- max
- 9
nperiodic_dimensions
int- min
- 3
- median
- 3
- mean
- 3
- max
- 3
nsites
int- min
- 1
- median
- 6
- mean
- 9.25
- max
- 444
species
string_text- min
- 138
- median
- 420
- mean
- 451.67
- max
- 1,254
species_at_sites
list- min
- 1
- median
- 6
- mean
- 9.25
- max
- 444
stress_tensor
list- min
- 0
- median
- 3
- mean
- 2.98
- max
- 3
total_magnetization
float- min
- -340.30
- median
- 0
- mean
- 1.25
- max
- 383.37
1.6% empty
Strengths and limits
Good at
- Harmonises calculations across Materials Project, Alexandria, and OQMD using explicit pseudopotential and Hubbard U compatibility criteria.
- Standardises crystal representations into OPTIMADE-compliant schemas alongside full stress tensors and atomic forces.
- Includes a robust structural deduplication pipeline based on EconNN bonding graphs and Weisfeiler-Lehman graph hashing.
Watch out for
- High-energy configurations and records with high residual atomic forces are present in the initial release without automatic filtering.
- Descriptive chemical formulas are scraped directly from parent databases and are not standardised across entries.
- The initial release excludes non-spin-polarized calculations from OQMD for systems without d or f electrons to preserve parameter compatibility.
Questions
Can I look inside LeMat-Bulk without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is LeMat-Bulk under?
Its authors declare cc-by-4.0. Read the terms on the original page before reusing the data.
What is the difference between the Compatible subsets and the All subset?
The Compatible subsets (Compatible PBE, PBESol, and SCAN) filter out calculations where pseudopotentials, Hubbard U parameters, or spin-polarisation settings conflict across source databases, whereas the All subset contains all aggregated records regardless of DFT parameter alignment.
How are duplicate crystal structures identified and handled?
Deduplication uses a material fingerprint that constructs a structure graph from EconNN bonds, hashes the graph using the Weisfeiler-Lehman algorithm, and incorporates symmetry and composition; when duplicates are detected, only the lowest-energy structure is preserved.
What physical units are used for forces, energies, and stresses?
Total uncorrected energies are reported in electronvolts (eV), atomistic forces in eV/Å, magnetic moments in Bohr magnetons (µB), and full 3x3 stress tensors in kilobars (kB).
How to cite it
Martin Siron, Inel Djafar, Lucile Ritchie, Etienne Du-Fayet, Amandine Rossello, Ali Ramlaoui, Leandro von Werra, Thomas Wolf, Alexandre Duval. LeMat-Bulk Dataset. 2024, doi:10.57967/hf/3762.
Written from LeMat-Bulk's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.