The validation of reaction mechanisms, usually combining electronic structure calculations and transition-state theory, stands as one of the most powerful examples of computational chemistry in action, routinely used to rationalize catalytic mechanisms, design new catalytic species, explain stereochemical outcomes and derive macroscopic rate laws. But how do we proceed if a reaction mechanism is unknown? In such cases, automated reaction discovery tools are rapidly emerging as a powerful computational approach to complement and support experimental mechanistic investigations. Here, I will describe our group’s recent work in the field of automated reaction discovery, and show how the challenge of proposing valid reaction mechanisms can be cast as a problem in discrete optimization. Specifically, I will show how the search for plausible multi-step reaction mechanisms can be viewed as a search problem within the space of chemically-allowed elementary reaction steps; we seek the set of elementary transformations which definitively lead from input reactants to target products within a defined maximum number of reaction-steps. Using ideas from graph theory, combined with simulated annealing-based optimization strategies, I will show how one can readily generate plausible reaction mechanisms leading to defined products, even for complex reactive systems containing many different molecules. Drawing on recent examples from homogeneous catalysis, combustion chemistry and astrochemistry, I will show how our optimization approach can generate large reaction-mechanism datasets (up to several thousand different candidate mechanisms, depending on system size) which can be sifted using electronic structure theory to identify the most plausible mechanistic proposals. I will also highlight how our graph-driven reaction discovery approach can be merged with machine-learning predictions of reaction properties, such as activation energies, to further accelerate the identification of plausible many-step reaction mechanisms in a computationally-efficient framework.

Reaction mechanism discovery as a discrete optimization problem

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.