Dx5 dataset

Benchmark dataset and code associated with the article<br><br><i>A Dataset for Benchmarking Electron Density Representations and Data-Driven Density Functional Theory</i><br>The dataset contains molecular configurations of <b>Water</b>, <b>Water tetramer</b>, <b>ethane</b>, <b>butane</b>, <b>ethanol </b>taken from MD trajectories.<br><br>High-quality electronic structure data is provided in the form of <b>spin density matrices</b> (obtained from Hartree-Fock, LDA, PBE, PBE0 and PBE50 using the Psi4 code). Exact exchange energies for the total molecular densities are contained and provide the target for machine learning models. <br><br>For baseline machine learning models and density representations, the predictive accuracy is provided in the article. The dataset is suited to further improve machine-learned, non-local DFT approximations of exchange energy functionals.<br>

本文配套的基准数据集与代码<br><br><i>用于电子密度表征基准测试与数据驱动密度泛函理论的数据集</i><br><br>本数据集包含从分子动力学（Molecular Dynamics, MD）轨迹中提取的<b>水</b>、<b>水四聚体</b>、<b>乙烷</b>、<b>丁烷</b>、<b>乙醇</b>的分子构型。<br><br>本数据集以<b>自旋密度矩阵（spin density matrices）</b>的形式提供高质量电子结构数据（通过Psi4软件包从哈特里-福克（Hartree-Fock）、局域密度近似（Local Density Approximation, LDA）、PBE、PBE0及PBE50方法中计算得到）。数据集包含全分子密度的精确交换能，可作为机器学习模型的训练目标。<br><br>本文中提供了基准机器学习模型与密度表征方法的预测精度。本数据集可用于进一步优化基于机器学习的非局域密度泛函理论（Density Functional Theory, DFT）交换能泛函近似方案。