LMX_Dataset

Pretty name: LMX Dataset
Tags: materials-science, atomistic-simulations, extxyz, ase, mliap
License: CC BY 4.0
Dataset type: splits (train/val/test) Associated manuscript: A Database for Design and Optimization of Halide Based Solid Electrolytes for Lithium Ion Batteries — Chiku Parida*, Arghya Bhowmik, and Juan Maria García Lastra*

Summary

This dataset provided in .extxyz format for training and benchmarking machine-learning interatomic potentials (MLIAPs).

Subset	Chemistry	Description
dataset_aliovalent.extxyz	LMX + aliovalent dopants	Tetravalent cation (e.g. Zr⁴⁺) substitution on the trivalent M site, creating Li vacancies and modifying ionic conductivity
dataset_aliovalent_antisite.extxyz	LMX + aliovalent dopants + antisite defects	Aliovalent-doped structures with additional antisite disorder on the cation sublattice
dataset_isovalent.extxyz	LMX + isovalent substitutions	Same-valence cation replacements (M³⁺ → M′³⁺) that tune lattice chemistry without altering Li stoichiometry

All frames carry DFT-level energies, forces, and stress tensors under consistent VASP/PBE settings, making this dataset directly suitable for fitting modern MLIAP architectures (e.g. MACE, NequIP, CHGNet, ALIGNN-FF).

• Format: ASE-compatible EXTXYZ with keys REF_energy, REF_forces, REF_stress
• Units: energy (eV), forces (eV/Å), stress (eV/ų)
• Total frames: ~400,000
• Systems: Ternary and quaternary Li-metal-halide solid electrolytes (LiₐMXᵦ families; M = trivalent / tetravalent metals; X = Cl, Br, I)
• DFT code / functional: VASP / PBE

Files

data/
├── dataset_aliovalent_antisite.extxyz   # aliovalent-doped + antisite-defect structures
├── dataset_aliovalent.extxyz            # aliovalent-doped structures
├── dataset_isovalent.extxyz             # isovalent-substituted structures
├── train_10k.extxyz                         # training split  (sampled 10k)
├── idd_test_2k.extxyz                          
└── ood_test_3k.extxyz                         
index.json                              

Schema (per-frame)

Each frame in every .extxyz file exposes the following properties:

Key	Location	Shape	Unit	Description
REF_energy	atoms.info	scalar	eV	Total DFT potential energy of the cell
REF_forces	atoms.arrays	(N, 3)	eV/Å	Atomic forces on all N atoms
REF_stress	atoms.info	(6,)	eV/ų	Cell stress tensor in Voigt order

Standard ASE atom properties (positions, species, cell, periodic boundary conditions and other metadatas in atoms info) are also stored in the usual EXTXYZ fields.

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="LMX_dataset",
    local_dir_use_symlinks=False,
)
print(f"Dataset downloaded to: {local_dir}")

Python: Load a split with ASE

from ase.io import read

# Load training split (all frames)
train = read("LMX_dataset/data/train.extxyz", index=":")
print(f"Training frames: {len(train)}")

# Access DFT labels for the first frame
frame = train[0]
energy = frame.info["REF_energy"]       # eV
forces = frame.arrays["REF_forces"]     # (N, 3) eV/Å
stress = frame.info["REF_stress"]       # (6,) eV/ų  — Voigt

Python: Fetch a single file

from huggingface_hub import hf_hub_download
from ase.io import read

path = hf_hub_download(
    repo_id="cparidaAI/LMX_dataset",
    repo_type="dataset",
    filename="data/ood_test_3k.extxyz",
)
test = read(path, index=":")
print(f"Test frames: {len(test)}")

Benchmarks / Intended use

Primary task: Supervised training and benchmarking of machine-learning interatomic potentials (MLIAPs) for Li-metal-halide solid electrolytes. Recommended evaluation metrics:

Property	Metric	Units
Energy	MAE per atom	meV/atom
Forces	MAE per component	meV/Å
Stress	MAE per component	GPa or eV/ų

Downstream applications:

• Long-timescale MD simulations of Li⁺ diffusion in halide electrolytes
• Screening of aliovalent and isovalent substitution effects on ionic conductivity
• Calculation of activation energies and diffusion coefficients at finite temperature
• Foundation for active-learning workflows to extend the chemical space of LMX electrolytes

Provenance & generation

Parameter	Value
Code	VASP (Vienna Ab initio Simulation Package)
Exchange–correlation functional	PBE (Perdew–Burke–Ernzerhof)
Plane-wave energy cutoff (ENCUT)	650 eV
k-point sampling	Γ-point only (supercell calculations)
Energy convergence (EDIFF)	(to be detailed in companion manuscript)
Force convergence (EDIFFG)	(to be detailed in companion manuscript)

Full INCAR and data generation code will be found in GitHub: https://github.com/chiku-parida/LMX_data

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

How to cite

If you use this dataset in your work, please cite:

Dataset manuscript (under review):

Chiku Parida, Arghya Bhowmik, and Juan Maria García Lastra. A Database for Design and Optimization of Halide Based Solid Electrolytes for Lithium Ion Batteries.

This HuggingFace repository:

Parida, C. (2025). LMX Dataset [Data set]. Hugging Face.
https://huggingface.co/datasets/cparidaAI/LMX_dataset

BibTeX (update DOI upon publication):

@dataset{parida2026lmx,
  author    = {Parida, Chiku and Bhowmik, Arghya and {Garc\'ia Lastra}, Juan Maria},
  title     = {{A Database for Design and Optimization of Halide Based Solid Electrolytes for Lithium Ion Batteries}},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/cparidaAI/LMX_dataset},
  note      = {}
}

Changelog

• v1.0.0 — initial release.

---

Contact: Chiku Parida, DTU Energy,