Porphyrin

This dataset contains 100,000 molecular structures sampled from a quantum molecular dynamics trajectory of a porphyrin molecule, generated using Density Functional Tight Binding (DFTB). The initial structure was obtained from ChemSpider (ID: 4086) and optimized using density functional theory (DFT) at the B3LYP/6-31G* level of theory. Subsequently, 150,000 femtoseconds (fs) of ground-state molecular dynamics at 300 K were performed using the DFTB+ package. From the final part of this trajectory, 100,000 snapshots were extracted at 1 fs intervals. For each structure, excitation energies were computed using time-dependent long-range corrected DFTB (TD-LC-DFTB). This dataset was used for evaluating uncertainty estimation methods for Gaussian process regression-based machine learning interatomic potentials, as described in the accompanying publication. In the .npz file the key for the coordinates is 'coordinates', the key for the nuclear charges is 'charges' and the key for the excitation energies is 'ES_energies'.本数据集包含100000个分子结构，这些结构源自采用密度泛函紧束缚（Density Functional Tight Binding, DFTB）方法生成的卟啉分子量子分子动力学轨迹。初始结构取自ChemSpider数据库（编号：4086），并在B3LYP/6-31G*理论级别下通过密度泛函理论（Density Functional Theory, DFT）完成结构优化。随后，使用DFTB+软件包开展了300K下时长150000飞秒（fs）的基态分子动力学模拟。从该轨迹的末尾部分，以1飞秒的间隔提取了100000个快照结构。针对每个分子结构，采用含时长程校正密度泛函紧束缚（Time-dependent long-range corrected DFTB, TD-LC-DFTB）方法计算了激发能。本数据集被用于评估基于高斯过程回归（Gaussian Process Regression, GPR）的机器学习原子间势的不确定性估计方法，相关细节详见随附的研究论文。在该.npz格式文件中，坐标数据的键名为"coordinates"，核电荷数据的键名为"charges"，激发能数据的键名为"ES_energies"。