LMX_Dataset

Pretty name: LMX Dataset
Tags: materials-science, atomistic-simulations, extxyz, ase, mliap
License:
Dataset type: splits (train/val/test)

Summary

This dataset provides three .extxyz splits (train/val/test) for training and benchmarking machine-learning interatomic potentials (MLIAPs).

• Format: ASE/EXTXYZ with keys: REF_energy, REF_forces, REF_stress
• Units: energy (eV), forces (eV/Å), stress (eV/ų)
• Systems: Halide-based Solid electrolytes
• Provenance: DFT/MD details: functional, code, cutoffs, k-points, thermostat/barostat, etc.

Files

• data/dataset_aliovalent_antisite.extxyz — <1> frames
• data/dataset_aliovalent.extxyz — <1> frames
• data/dataset_isovalent.extxyz — <1> frames
• data/train.extxyz — <1> frames
• data/val.extxyz — <1> frames
• data/test.extxyz — <1> frames
• index.json — split counts, checksums, and schema

Schema (per-frame)

• Energies stored in info["REF_energy"] (float, eV)
• Forces stored in arrays["REF_forces"] (shape: (N,3), eV/Å)
• Stress stored in info["REF_stress"] (Voigt 6 or 3x3; eV/ų). Specify exact convention below.

Stress convention: <Voigt order and sign convention; e.g., [σ_xx, σ_yy, σ_zz, σ_yz, σ_xz, σ_xy]>

Quickstart

Python: Download entire repository

from huggingface_hub import snapshot_download
local_dir = snapshot_download(
    repo_id="cparidaAI/LMX_dataset",
    repo_type="dataset",
    local_dir="dataset",
    local_dir_use_symlinks=False
)

Python: Load a split with ASE

from ase.io import read
train = read('hf://datasets/<your-username>/my-mliap-dataset/data/train.extxyz', ':')

Python: Fetch a single file

from huggingface_hub import hf_hub_download
path = hf_hub_download(
    repo_id="cparidaAI/LMX_dataset",
    repo_type="dataset",
    filename="data/train.extxyz"
)
from ase.io import read
train = read(path, ':')

Benchmarks / Intended use

• Primary task: supervised learning of energies/forces/stresses for MLIAPs.
• Suggested splits: train/val/test as provided.
• Recommended metrics: energy MAE (meV/atom), force MAE (meV/Å), virial/stress MAE (GPa or eV/ų).

Provenance & generation

• Code: VASP/PBE
• Settings: ENCUT, K-mesh, smearing, thresholds, etc.
• MD Sampling: NVT/NPT, T/P ranges, timesteps
• Post-processing: relaxations, filters, deduplication, unit conversions

License

This dataset is released under will be here. See LICENSE.

How to cite

Please cite:

1. <Your article / preprint / DOI>
2. This dataset (see CITATION.cff)

Changelog

• v1.0.0 — initial release.

---

Contact: Chiku Parida, DTU Energy,