Materials Science · LeMaterial
LeMat-Bulk-MLIP-Hull
Convex hull stability data for bulk materials computed with machine-learning interatomic potentials and DFT.
What LeMat-Bulk-MLIP-Hull is
LeMat-Bulk-MLIP-Hull is a materials science dataset providing convex hull thermodynamic stability data for bulk crystalline materials. Developed by LeMaterial, the resource compiles formation and hull energies calculated using reference density functional theory (DFT) along with several machine learning interatomic potentials (MLIPs).
Determining thermodynamic stability relative to competing phases via the convex hull is essential for discovering synthesisable bulk materials. By comparing values from diverse potential models alongside standard DFT calculations, the collection helps computational physicists, chemists, and materials scientists evaluate model agreement and screen candidate compositions.
The dataset contains between 1 million and 10 million entries, providing both unfiltered energy calculations across bulk systems and curated subsets identifying compounds within a strict energy-above-hull threshold.
How it was made
The dataset tabulates hull calculations across five distinct energy sources: reference DFT calculations, the ORB model, the UMA model, MACE-MP, and MACE-OMAT. Energies are provided in unified tabular Parquet formats, including an unfiltered split containing all bulk materials and specialized splits filtering materials within 0.001 eV/atom of their respective convex hulls.
For each evaluated potential and DFT reference, the release provides filtered dataset tables, sparse composition matrices stored in compressed NumPy format (.npz), and associated metadata containing element mapping information.
Key facts
- Category
- Materials science, chemistry, physics, engineering
- Data format
- Parquet and NPZ
- Modality
- Tabular and text
- Stability threshold
- 0.001 eV/atom above hull
- Evaluated potentials
- DFT, ORB, UMA, MACE-MP, MACE-OMAT
What you can use it for
- Thermodynamic Stability Screening
- Identify synthesisable or metastable bulk materials by selecting compounds located within 0.001 eV/atom of the convex hull.
- Machine-Learning Potential Benchmarking
- Compare energy-above-hull calculations across ORB, UMA, MACE-MP, and MACE-OMAT against baseline DFT reference energies.
- Composition Matrix Analysis
- Perform high-throughput compositional analysis across chemical spaces using the provided sparse composition matrices and element mappings.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.How the columns are distributed
Over 144,127 rows of default/dft
e_above_hull
float- min
- -0
- median
- 0
- mean
- 1.1e-4
- max
- 1.0e-3
energy
float- min
- -2260.06
- median
- -60.87
- mean
- -89.94
- max
- -0.03
immutable_id
string_text- min
- 5
- median
- 12
- mean
- 11.57
- max
- 13
mace_mp_energy
float- min
- -2258.51
- median
- -60.53
- mean
- -89.60
- max
- -0.01
0.00% empty
mace_omat_energy
float- min
- -2252.80
- median
- -60.38
- mean
- -89.21
- max
- 0.08
nsites
int- min
- 1
- median
- 11
- mean
- 15.39
- max
- 320
orb_conserv_inf_energy
float- min
- -2255.07
- median
- -60.42
- mean
- -89.30
- max
- -0.01
orb_direct_20_energy
float- min
- -2255.00
- median
- -60.42
- mean
- -89.30
- max
- 0.06
species_at_sites
list- min
- 1
- median
- 11
- mean
- 15.39
- max
- 320
true_energy
float- min
- -2260.06
- median
- -60.87
- mean
- -89.94
- max
- -0.03
uma_energy
float- min
- -2254.99
- median
- -60.44
- mean
- -89.33
- max
- 0.07
Strengths and limits
Good at
- Contains extensive coverage comprising between 1 million and 10 million entries across multiple bulk material systems.
- Provides direct comparisons across four separate machine-learning interatomic potentials alongside baseline DFT calculations.
- Offers both an unfiltered master dataset and targeted splits filtered to within 0.001 eV/atom of convex hulls.
- Supplies ready-to-use sparse composition matrices and element mappings in NPZ format.
Watch out for
- Threshold-filtered subsets are restricted strictly to a cutoff of 0.001 eV/atom above the convex hull.
- Coverage is restricted to bulk materials and does not evaluate surfaces, 2D systems, or molecular structures.
- Analysis of sparse composition matrices and element metadata requires NumPy and SciPy tooling.
Questions
Can I look inside LeMat-Bulk-MLIP-Hull without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is LeMat-Bulk-MLIP-Hull under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
Which machine-learning potentials are included in this dataset?
Calculations are provided for four machine learning interatomic potentials: ORB, UMA, MACE-MP, and MACE-OMAT, alongside reference DFT energies.
What energy threshold defines the pre-filtered splits?
The pre-filtered splits contain bulk materials that lie within 0.001 eV/atom of their respective convex hulls.
What data formats are provided for the calculations?
The dataset is supplied as tabular Parquet files, accompanied by sparse composition matrices and element mapping metadata saved as compressed NPZ files.
Written from LeMat-Bulk-MLIP-Hull'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.