Automated prediction of ground state spin for transition metal complexes

Predicting the ground state spin of transition metal complexes is a challenging task. Previous attempts have been focused on specific regions of chemical space, whereas a more general automated approach is required to process crystallographic structures for high-throughput quantum chemistry computations. In this work, we developed a method to predict ground state spins of transition metal complexes. We started by constructing a dataset which contains 2,032 first row transition metal complexes taken from experimental crystal structures and their computed ground state spins. This dataset showed large chemical diversity in terms of metals, metal oxidation states, coordination geometries, and ligands. Then, we analyzed the trends between structural and electronic features of the complexes and their ground state spins, and put forward an empirical spin state assignment model. We also used simple descriptors to build a statistical model with 97% predictive accuracy across the board. With this, we provide a practical and automated way to determine the ground state spin of transition metal complex from structure, enabling the high-throughput exploration of crystallographic repositories.

预测过渡金属配合物的基态自旋是一个颇具挑战性的任务。以往的研究主要集中于化学空间的特定区域，而处理晶体结构以进行高通量量子化学计算则需要一种更为通用的自动化方法。在本研究中，我们开发了一种预测过渡金属配合物基态自旋的方法。我们首先构建了一个数据集，其中包含从实验晶体结构中提取的2,032种第一行过渡金属配合物及其计算得到的基态自旋。该数据集在金属、金属氧化态、配位几何形状和配体等方面展现了广泛的化学多样性。随后，我们分析了配合物的结构特征和电子特征与其基态自旋之间的趋势，并提出了一个经验自旋态分配模型。此外，我们还利用简单的描述符构建了一个统计模型，该模型在所有情况下均实现了97%的预测精度。凭借此方法，我们提供了一种从结构确定过渡金属配合物基态自旋的实用且自动化的途径，从而实现了对晶体结构库的高通量探索。