Data set and SevenNet-Polar models for ZrO2, Li3PO4 and perovskites

This Zenodo recod contains: Born effective charges for ZrO₂, Li₃PO₄, and perovskite oxides, provided in extended XYZ format. The datasets for ZrO₂ and Li₃PO₄ additionally contain energies, forces, and stress tensors. SevenNet-Polar model checkpoints (specialized [PS] and multitask [PM], with different model sizes [S/M/L]) Dataset Contents Three subsets are included: ZrO₂: 10,103 configurations Li₃PO₄: 17,900 configurations Perovskites: 1,224 configurations License and Data Provenance All datasets, trained model parameters, checkpoints, and accompanying documentation contained in this Zenodo record are made available under the Creative Commons Attribution 4.0 International License (CC BY 4.0). The ZrO₂ and Li₃PO₄ datasets included in this record originate from simulations performed by the authors and their collaborators. The ZrO₂ dataset was originally published in the following GitHub repository: AugustinLu/BEC-NN The original GitHub repository is distributed under the MIT License. The copy of the ZrO₂ dataset included in this Zenodo record is additionally released under CC BY 4.0. The perovskite dataset is derived from the following CC BY 4.0-licensed dataset: Alex Kutana, Koji Shimizu, Satoshi Watanabe, and Ryoji Asahi, “Equivariant graph convolutional neural network for predicting tensors of atomic Born effective charges (Equivar). Pretrained models and scripts, and datasets of ab initio tensors of atomic Born effective charges of perovskites, Li₃PO₄, and ZrO₂,” Mendeley Data, Version 1 (2024).https://doi.org/10.17632/hx8kcpxh84.1License: CC BY 4.0. The perovskite data were extracted from the original archive and reformatted. No endorsement by the original dataset authors is implied. References and Citation For details regarding the generation, methodology, and application of these datasets, please refer to the publications below. Users of these datasets should cite the article or articles corresponding to the data they use, together with this Zenodo record. SevenNet-Polar Models & Main Reference Lu, A., Arai, S., Park, Y., Han, S., Miyazaki, T., and Watanabe, S. (2026). "SevenNet-Polar for MultiTask Prediction of Energy, Forces, Stress, and Born Effective Charges: Development and Application to ZrO₂, Li₃PO₄, and Perovskites." arXiv preprint arXiv:2607.14827. https://doi.org/10.48550/arXiv.2607.14827 ZrO₂ dataset Lu, A., Maekawa, N., Ikeda, A., Shimizu, K., Masuda, H., Yoshida, H., and Watanabe, S. (2026). “Study of the ion mobility in defect-laden ZrO₂ under an electric field using neural network with predictions for Born effective charges.” Physical Review Materials, 10, 066001.https://doi.org/10.1103/jcsd-dbl2 Li₃PO₄ dataset Shimizu, K., Otsuka, R., Hara, M., Minamitani, E., and Watanabe, S. (2023). “Prediction of Born effective charges using neural network to study ion migration under electric fields: Applications to crystalline and amorphous Li₃PO₄.” Science and Technology of Advanced Materials: Methods, 3(1), 2253135.https://doi.org/10.1080/27660400.2023.2253135 Perovskite dataset Kutana, A., Yoshimochi, K., and Asahi, R. (2025). “Dielectric tensor of perovskite oxides at finite temperature using equivariant graph neural network potentials.” Science and Technology of Advanced Materials: Methods, 5(1), 2497254.https://doi.org/10.1080/27660400.2025.2497254 The original perovskite data source should additionally be cited as: Kutana, A., Shimizu, K., Watanabe, S., and Asahi, R. (2024). “Equivariant graph convolutional neural network for predicting tensors of atomic Born effective charges (Equivar). Pretrained models and scripts, and datasets of ab initio tensors of atomic Born effective charges of perovskites, Li₃PO₄, and ZrO₂.” Mendeley Data, Version 1.https://doi.org/10.17632/hx8kcpxh84.1