Quantitative and Qualitative Tools for Predicting the Products and Mechanisms of Chemical Reactions

化学家通过制造具有有用性质的分子和材料来解决问题。有时，化学问题可以通过找到一种廉价和环保的方法来制造已知物质来解决。计算化学软件允许化学家在开始耗时的分子制造之前探索可能的合成途径。有时解决一个化学问题需要创造一种全新的物质。系统地测试所有可能的物质是不可能的，所以化学家们使用定性理论规则和定量计算来导航天文数字的可能性，直到他们发现具有理想性质的物质。*我开发理论模型和计算软件，用于预测化学反应路径，描述化学现象，并设计具有理想性质的分子。在最基本的层面上，我开发了新的计算方法来描述电子如何将原子结合到分子中。在解决这一问题的传统方法中，假定电子的运动是准独立的，因此可以清楚地识别单电子构型。这个近似不适合于探索化学反应活性，因为它对应变的分子几何结构(化学过渡态、伸缩的键等)是不可靠的。由于传统的多组态方法计算昂贵且相对难以使用，我的团队正在开发一种替代方法，该方法具有单一组态方法的一些特征(计算速度、易用性)，但不偏向任何特定的电子组态。我们的方法将使我们能够可靠地预测分子结构的能量，从而预测化学反应如何发生。我的团队还开发了一些概念工具，可以预测分子中最活跃的部分，并允许我们理解为什么有些反应很容易发生，而另一些反应不容易发生。这些洞察力导致了化学家可以用来指导寻找新现象的直观规则。这些工具也可以用于机器学习方法，即计算机(而不是人类化学家)获得有关已知分子的特征和性质的数据，发现组织这些数据的规则，并使用这些规则预测具有所需性质的新分子。这些方法是作为免费和开源软件传播的，通常作为Horton包的一部分。

Chemists solve problems by making molecules and materials with useful properties. Sometimes a chemical problem can be solved by finding an inexpensive and environmentally friendly way to manufacture a known substance. Computational chemistry software allows chemists to explore possible synthetic pathways before embarking on the time-consuming quest to make a molecule. Sometimes solving a chemical problem requires creating an entirely new substance. It is impossible to systematically test all possible substances, so chemists use qualitative theoretical rules and quantitative computations to navigate through the astronomical number of possibilities until they find a substance with desirable properties. ******I develop theoretical models and computational software for predicting chemical reaction pathways, describing chemical phenomena, and designing molecules with desirable properties. At the most fundamental level, I develop new computational methods to describe how electrons bind atoms into molecules. In conventional approaches to this problem, electrons are assumed to move quasi-independently, so that a single electron configuration can be clearly identified. This approximation is unsuitable for exploring chemical reactivity because it is unreliable for strained molecular geometries (chemical transition states, stretched bonds, etc.). Because traditional multiconfiguration methods are computationally expensive and relatively difficult to use, my group is developing an alternative approach, which shares some of the features (computational speed, ease-of-use) with single-configuration methods, but which is not biased towards any specific electron configuration. Our method will allow us to reliably predict the energy of molecular structures and, thereby, predict how chemical reactions occur. My group also develops conceptual tools that predict the most reactive parts of a molecule and allow us to understand why some reactions occur readily, while others do not. These insights lead to intuitive rules that chemists can use to guide the search for new phenomena. These tools can also be used by machine-learning methods, where a computer (rather than a human chemist) gets data about the features and properties of known molecules, discovers rules for organizing this data, and uses these rules to predict new molecules with desirable properties. These methods are disseminated as free and open-source software, usually as part of the HORTON package.