Abstract:

Hydrogen Atom Transfer (HAT) is one of the fundamental chemical reactions in nature, and accurate prediction of its reactivity and selectivity is essential for the rational design of related chemical reactions. One important approach is to study the reactivity and selectivity by predicting the activation energy barrier of the reaction. This paper reviews the current research progress in predicting the activation energy barriers from the perspectives of both empirical models and machine learning models. Empirical models are based on experimental data and chemical laws of known reactions and are fitted using empirical formulas (e.g., linear equations), which have good interpretability, but have some limitations in terms of applicability and accuracy. Machine learning models, on the other hand, are capable of handling much larger amounts of data and more complex reaction mechanisms, and have more potential for accurately predicting activation barriers, but the predictions are dependent on the quality of the data and are less interpretable. Finally, this paper provides an outlook on how to develop more accurate and interpretable activation energy barrier prediction models in the future, and looks forward to improving the understanding of the factors influencing reactivity by improving the interpretability of activation energy barrier prediction models.

Key words: hydrogen atom transfer, radical, computational chemistry, reaction, model, machine learning

摘要：

氢原子转移（hydrogen atom transfer，HAT）是自然界中的基本化学反应之一，准确预测其反应性和选择性对于合理设计相关化学反应至关重要。其中一种重要方法是通过预测反应的活化能垒来研究其反应性和选择性。本文从经验模型和机器学习模型两个角度综述了当前预测活化能垒的研究进展。经验模型基于已知反应的实验数据和化学规律，采用经验公式（如线性方程）进行拟合，具有较好的可解释性，但在适用性和准确性方面存在一定局限性。而机器学习模型则能够处理更大量级的数据和更复杂的反应机理，在准确预测活化能垒方面更有潜力，但是预测效果依赖于数据的质量，并且可解释性较弱。最后，本文对未来如何开发更准确且可解释的活化能垒预测模型进行了展望，并且期待通过提高活化能垒预测模型的可解释性进而提高人们对反应活性影响因素的理解。

关键词: 氢原子转移, 自由基, 计算化学, 反应, 模型, 机器学习