Dimensional Analysis of Lewis Acidity: A Theoretical, Experimental, and Data Driven Approach

路易斯酸的概念是在一百多年前形成的。尽管它具有极高的定性价值，但仍然没有令人满意的模型来定量理解刘易斯酸的“强度”。虽然有些趋势可以用硬酸和软酸的原理来描述，但许多例外都指出了这种经验方法的重大弱点。而在早期，这样的和类似的模型是基于有限的实验数据，现代计算化学或分析方法允许在更大的规模上生成亲和数据（因子bbb1000）。该项目旨在通过统计工具研究刘易斯酸度。起点是生成一个具有大约30,000个亲和关系的数据集。通过适当的训练数据集生成机器学习模型，该模型可以根据化合物的二维Lewis结构作为唯一输入，预测化合物对选定供体的亲和力。与这些理论方法并行，为了避免忽略凝聚态的影响，用等温量热法（ITC）获得了路易斯对形成的热力学数据。在惰性气体条件下建立ITC操作是无机分子化学中的一项新技术。这样就可以获得大约500个Lewis对的实验亲和数据，可以作为第一个子项目的理论方法的基准。最后，通过降维和相关方法处理收集到的路易斯酸对许多路易斯碱的理论和实验亲和力。例如，通过主成分分析，可以得出一个结论，说明需要多少和哪种亲和尺度才能反映尽可能多的关于路易斯酸的信息。通过因子分析，一个物理化学的解释出现了哪些因素是显著负责亲和力路易斯酸。刘易斯酸问题的维数是多少？据我们所知，这项研究代表了第一次用“大数据”方法来评估这个有数百年历史的概念。我们期望在催化、材料研究和生命科学方面获得重要的知识。

The concept of Lewis acids was formulated over one hundred years ago. Despite its extreme qualitative value, there is still no satisfactory model for a quantitative understanding of the "strength" of a Lewis acid. While some trends can be described by the principle of hard and soft acids, numerous exceptions point to significant weaknesses in this empirical approach. While in earlier eras, such and similar models were based on limited experimental data, modern computational chemistry or analytical methods allow the generation of affinity data on a substantially larger scale (factor > 1000). This project aims to investigate Lewis acidity by statistical tools. The starting point is the generation of a dataset with approximately 30,000 affinities. A machine learning model is generated by suitable training data sets, which can predict the affinity of a compound toward selected donors based on its two-dimensional Lewis structure as the only input. In parallel to these theoretical approaches, and to avoid losing sight of effects in the condensed phase, thermodynamic data on Lewis pair formation are obtained by isothermal calorimetry (ITC). Establishing ITC operation under inert gas conditions represents a novelty in inorganic molecular chemistry. In this way, experimental affinity data of around 500 Lewis pairs will become accessible, which can serve as a benchmark for the theoretical methods from the first subproject. Finally, the collected theoretical and experimental affinities of Lewis acids toward numerous Lewis bases are treated via dimensionality reduction and related methods. For example, with principal component analysis, a statement can be made on how many and which affinity scales are needed to reflect as much information as possible about a Lewis acid. By factor analysis, a physicochemical interpretation emerges as to which factors are significantly responsible for the affinity of a Lewis acid. What is the dimensionality of the Lewis acid problem? To our knowledge, this research represents the first "big data" approach to appraise this centuries-old concept. We anticipate critical knowledge gains with implications in catalysis, materials research, and the life sciences.