Data supporting the publication: Fine-Tuning Universal Machine-Learned Interatomic Potentials for Applications in the Science of Steels

This dataset contains the supplemental data for the study "Fine-Tuning Universal Machine-Learned Interatomic Potentials for Applications in the Science of Steels". The primary research objective is to enhance the accuracy of foundational models—specifically CHGNet, MACE or SevenNet to better predict the properties of bcc Fe system. The dataset was generated using VASP and it was subsequently used to fine-tune these potentials. The archive includes the complete fine-tuning outputs, model checkpoint files (`*.pth.tar`), and validation results, offering a comprehensive benchmark for evaluating MLIP performance on iron-based materials.

本数据集为研究《面向钢铁科学应用的通用机器学习原子间势微调》（Fine-Tuning Universal Machine-Learned Interatomic Potentials for Applications in the Science of Steels）的补充数据。本研究的核心目标是提升基础模型的预测精度，具体涵盖CHGNet、MACE及SevenNet，以更准确地预测体心立方铁（bcc Fe）体系的各项性质。本数据集通过维也纳从头算模拟包（Vienna Ab initio Simulation Package，VASP）生成，并随后用于上述原子间势的微调工作。该归档文件包含完整的微调输出、模型检查点文件（`*.pth.tar`）以及验证结果，可为评估铁基材料的机器学习原子间势（Machine-Learned Interatomic Potentials，MLIP）性能提供全面的基准测试支撑。