Discovering intrinsic catalytic reaction mechanisms: using machine learning to 'crack' rate expressions

开发高效、可持续的催化剂是介于化学和化学工程之间的研究重点之一。对于大多数涉及多相的工业反应，诸如朗缪尔-辛谢尔伍德方程之类的速率表达式一直是定量描述多相催化的长期动力学模型的中心。尽管多相催化的表观动力学可用于工业应用，但本征反应动力学仍未得到很好的理解。这不仅限制了对催化反应现象的基本理解，而且减少了为新催化剂量身定做设计的可能性。通过融合数据分析和动力学理解，该项目将利用机器学习社区的最新进展来分析和分解速率表达式的经验参数和结构，从而发现内在的催化反应机理。此外，在这些新认识的帮助下，该项目将设计一种方法，为特定应用开发具有所需性能的定制催化剂。将重点放在公认的简单反应上将是重要的，因为文献中已经存在许多数据，以提供一个良好的开端。简单的反应将允许更多地强调催化剂的固有性质，特别是在探索多功能催化剂时。根据最近发表在《绿色化学》上的一篇展望文章[1]，酰胺还原似乎是从上述方法中受益的一个合适的候选者。设计一种用于多官能团选择性加氢的高活性多相催化剂是一个相当大的挑战。成功的博士生将有机会学习机器学习和数学编程技术，以及一系列高度复杂的实验技能。[1]绿色化学，2018，20,5082

Developing high-efficiency and sustainable catalysts is one of the research priorities that lies at the interface between chemistry and chemical engineering. For most industrial reactions involving multiple phases, rate expressions such as the Langmuir-Hinshelwood equation have been the centre of long-standing kinetic models to quantitatively characterise heterogeneous catalysis. Although the apparent kinetics of heterogeneous catalysis can be estimated using these expressions for industrial applications, the intrinsic reaction kinetics are still not well understood. This not only limits the fundamental understanding of the catalytic reaction phenomena, but also mitigates the possibility of a tailored-made design of new catalysts. By merging data analytics and kinetic understanding, this project will discover the intrinsic catalytic reaction mechanism by exploiting recent advances from the machine learning community to analyse and decompose the empirical parameters and structures of rate expressions. Furthermore, aided by these new understandings, this project will design a methodology to develop tailored-made catalysts with desired properties for specific applications. It will be important to focus on well-established simple reactions, where many data already exist in the literature to give a good head start. Simple reactions will allow more emphasis on intrinsic catalyst properties, particularly when exploring multifunctional catalysts. Based on a recent Perspectives article in Green Chemistry [1], amide reduction appears to be a suitable candidate to benefit from the above approach. To design a highly active heterogeneous catalyst for the selective hydrogenation of multifunctional amides is a considerable challenge. Successful PhD candidates will have the opportunity to learn both machine learning and mathematical programming techniques as well as a range of highly sophisticated experimental skills. [1] Green Chem, 2018, 20, 5082