CAREER: CDS&E: Predictive Discovery of Complex Reaction Mechanisms

密歇根大学的保罗·齐默尔曼得到了化学理论、模型和计算方法计划以及计算和数据使能科学与工程(CDS&amp；E)计划的支持，以开发预测化学反应结果的新工具。计算化学在使用量子力学原理创造工具方面有着悠久的历史，这些工具为各种各样的化学过程提供了详细、准确的解释。这些工具中的许多已经达到了足够的精度，可以应用于发现化学反应，而不需要事先从实验中了解情况。不幸的是，这些方法的高昂计算成本阻碍了它们在预测化学反应中的广泛使用，特别是在化学非常复杂或了解很少的情况下。这笔赠款支持开发完全计算、低成本和高精度的方法，这些方法能够仅从原料分子开始预测化学反应的结果。这些方法的应用重点是催化反应，为研究可应用于工业水平的高价值化学产品提供了一个非常有用的工具。这些工具可供更广泛的计算化学社区使用，使开发能够最大限度地影响化学反应中的大量问题。由于这项研究的潜在应用的多样性，参与该项目的学生不仅获得了算法开发能力，还获得了对控制分子行为的过程的基本见解、尖端计算研究经验和解决问题的能力，这些都是应对21世纪挑战所必需的。建议的方法被转化为教育战略，将入门实验室练习与真实世界的研究相结合，从荣誉有机化学的试点研究开始。该研究计划建立在齐默尔曼小组的工作基础上，该小组表明，化学反应可以用少量局部的、非调和的反应坐标来描述。这些反应坐标由原子间的距离、角度和扭转组成，是描述原子运动的可靠的、可转移的基础。通过使用先进的单端状态链优化算法沿着这些坐标搜索可信的化学反应，这种研究方法可以有效和可靠地预测反应事件，而不需要化学直觉的指导。正在进行的工作的主要目标是将该方法扩展到涵盖过渡金属元素，实现在具有软性自由度的系统中进行有效的构象搜索，并开发机器学习算法来自动处理化学数据并大大提高计算效率。总而言之，这些新的反应搜索技术能够在各种大型、高度复杂的化学系统中发现预测性反应，在这些系统中，仅有化学知识尚不足以做出准确的预测。这些方法正被应用于具有挑战性的催化案例，在这些案例中，未确定的副反应严重阻碍了有效产品的形成。这些不必要的反应途径的不断发现导致了提高反应选择性和催化剂稳定性所需的化学洞察力。这些工具允许化学初学者假设和评估在硅胶中的反应，从而为学生在学习的早期阶段进行研究提供了一种手段。这样的教学工具被分发给密歇根大学以外的教育工作者，以最大限度地发挥其影响。

Paul Zimmerman at the University of Michigan is supported by the Chemical Theory, Models and Computational Methods Program and the Computational and Data-Enabled Science and Engineering (CDS&E) Program to develop new tools to predict the outcome of chemical reactions. Computational chemistry has a long history of using the principles of quantum mechanics to create tools which provide detailed, accurate explanations for a wide variety of chemical processes. Many of these tools have reached sufficient accuracy that they can be applied to discover chemical reactions without requiring prior insight from experiment. Unfortunately, the high computational cost of these methods has prevented their broad use in predicting chemical reactivity, especially in cases where the chemistry is highly complex or poorly understood. This grant supports the development of fully computational, low-cost and highly accurate methods that are able to predict the outcome of chemical reactions starting only from the feedstock molecules. The application focus of these methods is catalytic reactions, providing a highly useful tool for research into chemistries that can be applied at an industrial level to create high-value chemical products. These tools are available to the wider computational chemistry community, enabling maximized impact of the developments to a great number of problems in chemical reactivity. Due to the diversity of potential applications for this research, the students involved in this project gain not only algorithm development abilities, but also fundamental insights into processes governing molecular behavior, cutting-edge computational research experience, and problem-solving abilities that are needed to address the challenges of the 21st century. The proposed methods are transformed into educational strategies that merge introductory laboratory exercises with real-world research, starting with a pilot study in honors organic chemistry. The research program builds upon work by the Zimmerman group that shows chemical reactions can be described in terms of a small number of localized, anharmonic reaction coordinates. These reaction coordinates consist of interatomic distances, angles, and torsions, and are a reliable, transferable basis for describing atomic motion. By employing advanced single-ended chain-of-states optimization algorithms which search along these coordinates for plausible chemical reactions, this research methodology can efficiently and reliably predict reactive events without guidance from chemical intuition. The main objective of ongoing work is to expand this method to cover transition metal elements, enable efficient conformation searches in systems with floppy degrees of freedom, and develop machine learning algorithms to automatically process chemical data and provide great enhancements in computational efficiency. In sum, these new reaction search techniques enable predictive reaction discovery in a wide variety of large, highly complicated chemical systems where chemical knowledge alone is not yet sufficient to make accurate predictions. These methods are being applied to challenging cases in catalysis where uncharacterized side reactions are severe impediments to efficient product formation. Ongoing discovery of these undesired reaction pathways lead to the chemical insight required to improve reaction selectivity and catalyst stability.These tools allow beginning chemistry students to hypothesize and evaluate reactions in silico, resulting in a means for students to perform research at an early stage of their studies. Such pedagogical tools are distributed to educators outside of the University of Michigan to maximize their impact.