MPtrj Relaxation Trajectories and Materials Project 2023 Structures (Matbench Discovery Files)
- Provider
- Deng, Ceder et al. (MPtrj); Riebesell et al. (Matbench Discovery); Materials Project; redistributed by Emulated Inc
- License
- CC-BY-4.0 (source)
- Size
- 5 files
- Format
- JSON Lines, JSON (in sampled files)
- Stored at
- Hugging Face (https://huggingface.co/datasets/Emulated-Inc/materials-project-2023-mptrj)
- Access
- Free and public. No account with the data provider is needed.
- Category
- Chemistry & Materials
- Source
- https://doi.org/10.1038/s42256-023-00716-3
About this dataset
Two core training files of the Matbench Discovery benchmark redistributed unchanged or near-unchanged from its figshare record 22715158. mptrj.extxyz.zip is MPtrj, the Materials Project trajectory dataset built for CHGNet: frames of Materials Project GGA and GGA+U (VASP PBE) relaxations, about 1.58 million structures from roughly 146,000 materials, each with energy, forces, stress and magnetic moments, stored as one extended-XYZ file per material (readable with ASE). mp.jsonl holds 142,666 Materials Project 2023-02-07 computed structure entries with at most 80 sites, each with lattice, species, fractional coordinates, raw and MP2020-corrected energy per atom, formation energy, energy above hull, run type and Hubbard U values. A build report records sha256 checksums of the source files; the repository is about 1.7 GB (measured 2026-11-15). MPtrj is the standard training set for universal interatomic potentials such as CHGNet, MACE-MP-0, M3GNet and SevenNet and for crystal stability prediction. The figshare record is CC-BY-4.0; cite Deng et al., Nature Machine Intelligence 2023 (CHGNet), Riebesell et al., Nature Machine Intelligence 2025, and Jain et al. 2013.
How to load this dataset in Python
from datasets import load_dataset
ds = load_dataset("Emulated-Inc/materials-project-2023-mptrj")
print(ds)Data structure
File types in sampled files: JSON Lines, JSON
Files
Example file: mp.jsonl
Analyze this dataset without downloading it
On Scigantic, MPtrj Relaxation Trajectories and Materials Project 2023 Structures (Matbench Discovery Files) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.
This page is also available as Markdown and as JSON.
Related Chemistry & Materials datasets
- Gistbench
- WS22 urocanic acid: Wigner-Sampled Conformations with PBE0 Energies and Forces, ColabFit Conversion
- Materials Project Electronic Band Structures (VASP)
- Oxygen-Deficient SrMnO3 and SrCaMnO3 Manganite Heterostructures: Vibration-Induced Ferrimagnetism DFT (Kaviani et al. 2025, Materials Cloud), ColabFit Conversion
- q-AQUA Water Many-Body CCSD(T) Interaction Energies: 2-, 3- and 4-Body Water Clusters (Yu et al. 2022), ColabFit Conversion
- NiCoCr Medium-Entropy Alloy DFT Data for Misfit Volumes and Yield Strength (Yin and Curtin 2020), ColabFit Conversion
- CoCrFeNiPd High-Entropy Alloy (HEA) Strength: DFT Misfit Volumes (Yin and Curtin 2020, Materials Cloud), ColabFit Conversion
- MSnLib: Multi-Stage (MSn) Tandem Mass Spectral Libraries of 4.5 Million Orbitrap Spectra (IOCB Prague, Parquet)