Alexandria Materials Database Snapshot: 5.07 Million DFT Crystal Entries (pymatgen Parquet)

Provider
Alexandria (Schmidt, Marques et al., Ruhr University Bochum); snapshot by atom-int-team
License
CC-BY-4.0 (source)
Size
32 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/atom-int-team/alexandria)
Access
Free and public. No account with the data provider is needed.
Category
Chemistry & Materials
Source
https://doi.org/10.1002/adma.202210788

About this dataset

A 2026-01-16 snapshot of the Alexandria Materials Database (Ruhr University Bochum, ICAMS), one of the largest open collections of DFT-computed inorganic crystals, most of them hypothetical compounds proposed by machine learning and relaxed with DFT. Each of the 5,068,744 rows is one Alexandria entry (agm identifiers) with its reduced formula, total energy, energy above the convex hull, band gap, total magnetization, number of sites and the relaxed crystal structure serialized as pymatgen Structure JSON. Files are 30 Parquet shards under data/, about 4.9 GB (measured 2026-11-15); load with Hugging Face datasets or pandas and parse structures with pymatgen.core.Structure.from_str(..., fmt='json'). Typical uses are crystal stability and convex hull analysis, band gap and formation energy prediction, and training crystal generative models (diffusion, flow matching) and graph neural networks. The republishing team kept numeric fields as in Alexandria; consult the Alexandria documentation for units and functionals. Alexandria is released under CC-BY-4.0; cite Schmidt et al., Advanced Materials 2023, and Schmidt et al., Scientific Data 2022.

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("atom-int-team/alexandria")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: data/train-00000-of-00030.parquet

Analyze this dataset without downloading it

On Scigantic, Alexandria Materials Database Snapshot: 5.07 Million DFT Crystal Entries (pymatgen Parquet) 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