Published November 12, 2026 | Version v1

Polaron migration barriers and features of Li₄Ti₅O₁₂ (LTO).

  • 1. University of bayreuth
  • 2. ROR icon University of Bayreuth

Description

Overview

This supplementary dataset provides the computational data, machine-learning models, feature-generation scripts, and analysis results associated with the study of polaron migration barriers in bulk Li₄Ti₅O₁₂ (LTO).

The dataset is organized into three main directories:

  • dft_data/ – DFT-calculated atomic structures in extended XYZ (.extxyz) format, used for model development and barrier analysis.
  • ml_models/ – Machine-learning implementations, training datasets, notebooks, and model evaluation results.
  • symmt_features/ – Structural distance features transformed using radial symmetry functions based on the Behler–Parrinello method.

1. Directory Structure

.
├── dft_data/
│   ├── all_pol_geom.extxyz
│   └── barrier_struct.extxyz
│
├── ml_models/
│   ├── GPR/
│   ├── SVM/
│   └── xgb_regression/
│       ├── analysis/
│       ├── model_training/
│       ├── utils/
│       └── pyproject.toml
│
└── symmt_features/
    ├── add_en.ipynb
    ├── add_en.py
    ├── apply_radial_symm.py
    ├── barrier_feature_gen.py
    └── generate_features.py

2. DFT Data (dft_data/)

This directory contains the atomic structures obtained from density functional theory (DFT) calculations. The structures are provided in extended XYZ (.extxyz) format.

3. Machine-Learning Models (ml_models/)

This directory contains three regression approaches used for predicting polaron migration barriers:

  • Gaussian Process Regression (GPR): Implementation, model evaluation, and kernel comparison.
  • Support Vector Machine (SVM): Regression models developed using different kernel functions.
  • XGBoost Regression: Model training, feature-generation utilities, and subsequent analysis of predicted migration barriers.

The respective subdirectories contain the Jupyter notebooks, trained models, datasets, and graphical results associated with each approach.

The analysis/ directory within xgb_regression/ contains post-processing workflows and results, including barrier distributions, feature importance, configuration statistics, and temperature-dependent analyses.

4. Symmetry-Based Features (symmt_features/)

This directory contains the scripts and datasets used to generate structural descriptors for the polaron migration barriers. Radial symmetry functions based on the Behler–Parrinello method are applied to the structural distance features to obtain symmetry-based descriptors.

The directory also includes the associated feature-generation and energy-processing scripts, along with the resulting datasets.

5. Software and Dependencies

The computational workflows are implemented in Python and use the following libraries:

  • NumPy
  • SciPy
  • Pandas
  • Matplotlib
  • Scikit-learn
  • XGBoost
  • Jupyter Notebook
  • ASE (Atomic Simulation Environment)

The specific package versions may vary depending on the computational environment used for the original calculations.

Files

supplementary_data.zip

Files (20.0 MB)

Name Size
md5:421323e492da6facd600b01a656f2f86
20.0 MB Preview Download

Additional details

Funding

Deutsche Forschungsgemeinschaft
MultiTrans CRC 158
Deutsche Forschungsgemeinschaft
OPTEXC IRTG 2818
Bavarian State Ministry for Science and Art
SolTech
Deutsche Forschungsgemeinschaft
Festus 523317330
Deutsche Forschungsgemeinschaft
NHR@FAU Fritz 440719683