KIM-Si-training-set

This is the silicon ("KIM-Si") training set used in the forthcoming NeurIPS '22 paper titled "Injecting Domain Knowledge from Empirical Interatomic Potentials to Neural Networks for Predicting Material Properties" by Shui et al. It contains a total of 16,110 atomic configurations in extended XYZ format with total potential energies computed by eight different empirical potentials for each of them. Reference energies computed using (GGA-PBE) density functional theory (DFT) in the VASP ab initio software package are provided for 2,110 of the configurations.

本数据集为Shui等人即将发表于2022年神经信息处理系统大会（NeurIPS 2022）的论文《将经验原子间势的领域知识注入神经网络以预测材料性能（"Injecting Domain Knowledge from Empirical Interatomic Potentials to Neural Networks for Predicting Material Properties"）》中使用的硅基（KIM-Si）训练集。该数据集共包含16110个采用扩展XYZ（extended XYZ）格式存储的原子构型，每个构型均配有8种不同经验原子间势计算得到的总势能。其中2110个构型提供了采用VASP从头算软件包，通过(GGA-PBE)密度泛函理论(DFT)计算得到的参考能量。