HEABench VASP Raw Data

HEABench provides raw VASP calculation files for 3,109 high-entropy-alloy (HEA) samples. Each sample has a public integer ID from 0001 to 3109 and is distributed as an independent .tar.gz archive. A sample contains three calculation stages—structural relaxation, a static calculation, and a band-structure calculation—together with per-file checksums and provenance metadata.

This repository is intended for users who want to inspect structures and calculation outputs, build downstream materials datasets, or reproduce post-processing analyses without downloading the entire collection.

Dataset at a glance

Item	Value
Samples	3,109
Calculation stages per sample	relax, static, band
Archive naming	data/0001.tar.gz ... data/3109.tar.gz
Compressed archive size	114,596,868,970 bytes (106.73 GiB)
Indexed files inside archives	74,799
Sample index	manifest.csv
Per-file index	files.jsonl

Repository layout

.
├── README.md               # Dataset documentation
├── manifest.csv            # One row per sample/archive
├── files.jsonl             # One row per file inside the archives
├── checksums.sha256        # SHA-256 checksums for archives and metadata
├── summary.json            # Dataset-level counts and sizes
├── verification.json       # Release validation summary
└── data/
    ├── 0001.tar.gz
    ├── 0002.tar.gz
    ├── ...
    └── 3109.tar.gz

Each archive expands into one numbered sample directory:

NNNN/
├── metadata.json
├── relax/
│   ├── POSCAR
│   ├── CONTCAR
│   ├── KPOINTS
│   ├── IBZKPT
│   ├── OSZICAR
│   ├── OUTCAR
│   ├── vasprun.xml         # Optional; see manifest.csv
│   └── XDATCAR
├── static/
│   ├── POSCAR
│   ├── CONTCAR
│   ├── KPOINTS
│   ├── IBZKPT
│   ├── OSZICAR
│   ├── OUTCAR
│   ├── vasprun.xml         # Optional; see manifest.csv
│   ├── CHGCAR
│   ├── DOSCAR
│   └── EIGENVAL
└── band/
    ├── POSCAR
    ├── CONTCAR
    ├── KPOINTS
    ├── OSZICAR
    ├── OUTCAR
    ├── EIGENVAL
    └── CHGCAR              # Optional; see manifest.csv

NNNN is the same four-digit identifier used by the archive name and sample_name in manifest.csv.

Calculation stages

Stage	Purpose	Good starting points
relax	Structural relaxation. It records the evolution of the ionic configuration and the final relaxed structure.	CONTCAR, OUTCAR, vasprun.xml, XDATCAR
static	Static electronic calculation on the relaxed structure. It provides the reference charge density, density of states, eigenvalues, energies, forces, and related outputs.	CHGCAR, DOSCAR, EIGENVAL, OUTCAR, vasprun.xml
band	Band calculation using the k-point path stored in KPOINTS.	KPOINTS, EIGENVAL, OUTCAR

VASP file guide

The descriptions below are brief usage-oriented summaries. Follow the linked VASP Wiki pages for the complete format and interpretation.

File	What it contains and when to use it
POSCAR	Lattice vectors, chemical species/counts, and ionic positions at the start of a stage. Use it to reconstruct the stage input structure.
CONTCAR	Structure written at the end of a calculation, in a format compatible with POSCAR. In relax, this is the final relaxed structure.
KPOINTS	Brillouin-zone sampling. In band, it defines the k-point path used for the band calculation.
IBZKPT	Irreducible k-point coordinates and weights generated or used by VASP.
OSZICAR	Compact record of electronic and ionic iteration energies; useful for quickly checking convergence.
OUTCAR	Main human-readable output containing calculation settings and detailed results such as energies, forces, stress, eigenvalues, and magnetic information when applicable.
vasprun.xml	Structured XML output containing calculation parameters, structures, energies, forces, and method-dependent results. It is convenient for programmatic parsing.
XDATCAR	Ionic configurations written during ionic steps. Use it to inspect the relaxation trajectory.
CHGCAR	Charge density together with PAW one-center occupancies. It can be used for charge-density analysis or as a restart input in an appropriately licensed VASP environment.
DOSCAR	Density of states and integrated density of states from the static calculation.
EIGENVAL	Kohn-Sham eigenvalues for every stored k point. The band/EIGENVAL file is the primary raw input for band-dispersion analysis.
metadata.json	HEABench metadata for the sample, including its public ID, provenance fields, file availability, and SHA-256/size records for every included file. This is not a native VASP file.

See also the VASP Wiki's overview of output files.

Quick start

1. Find a sample

manifest.csv is the main sample index. For example, inspect the first few rows with pandas:

import pandas as pd

manifest = pd.read_csv("manifest.csv")
print(manifest[[
    "sample_id",
    "sample_name",
    "archive_path",
    "archive_bytes",
]].head())

2. Download one archive

from huggingface_hub import hf_hub_download

archive = hf_hub_download(
    repo_id="Steve360b/HEABench",
    repo_type="dataset",
    filename="data/0001.tar.gz",
)
print(archive)

To download from the command line:

hf download Steve360b/HEABench data/0001.tar.gz \
  --repo-type dataset \
  --local-dir .

3. Extract the sample

mkdir -p samples
tar -xzf data/0001.tar.gz -C samples

The extracted files will be under samples/0001/.

4. Read common outputs

Read the final relaxed structure with pymatgen:

from pymatgen.core import Structure

structure = Structure.from_file("samples/0001/relax/CONTCAR")
print(structure.composition)
print(structure.lattice)

Read energies and structures from vasprun.xml:

from pymatgen.io.vasp.outputs import Vasprun

run = Vasprun(
    "samples/0001/static/vasprun.xml",
    parse_projected_eigen=False,
)
print(run.final_energy)
print(run.final_structure)

For large XML files, consider streaming XML parsing or disabling parser features that are not needed.

Metadata and indexes

manifest.csv

One row per sample. Important columns include:

Column	Meaning
sample_id	Integer sample ID (1–3109).
sample_name	Zero-padded public name (0001–3109).
original_dataset_entry_id	Stable identifier of the original dataset entry.
selected_source, source_record_id	Provenance fields for the selected calculation record.
band_chgcar_included	Whether band/CHGCAR is present.
missing_optional_files	Optional files absent from this sample.
archive_path	Archive location in this repository.
archive_sha256, archive_bytes	Archive checksum and compressed size.
raw_bytes, file_count	Uncompressed byte count and member-file count.

files.jsonl

One JSON object per file inside the archives. Each record contains the sample ID/name, archive path, relative file path, byte count, and SHA-256 checksum. Use this index when a workflow needs to inspect archive contents without opening every archive first.

Other repository metadata

• checksums.sha256: checksums for all archives and root metadata files.
• summary.json: dataset-level sample, file, size, workflow, and availability counts.
• verification.json: validation scope and release status.

File availability

Most stage files are present for all 3,109 samples. The known optional exceptions are recorded in both manifest.csv and summary.json:

• relax/vasprun.xml: present for 3,103 samples.
• static/vasprun.xml: present for 3,107 samples.
• band/CHGCAR: present for 191 samples.

Code should use missing_optional_files or check for file existence instead of assuming that every optional file is present.

Integrity checks

Compare a downloaded archive with the published checksum:

sha256sum data/0001.tar.gz
grep 'data/0001.tar.gz$' checksums.sha256

For individual files inside an archive, use files.jsonl or the sample's metadata.json.

Citation

An accompanying paper and arXiv record are forthcoming. Until the formal citation is available, please cite the dataset name and repository:

HEABench VASP Raw Data. https://huggingface.co/datasets/Steve360b/HEABench

Scope and licensing notes

This release is designed for analysis and post-processing; it is not a complete standalone VASP rerun bundle. It does not include INCAR, POTCAR, POTCAR.*, WAVECAR, or PROCAR. In particular, no PAW potential payload is redistributed. Access to and use of VASP and PAW potential files remain subject to the applicable VASP license.

The calculation results and HEABench-authored metadata in this repository are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). This license applies to the dataset contents released here.