Automating the search for new chemical reactions is one of the ‘grand challenges’ in computational chemistry. While characterizing reaction energy pathways has been routine for practitioners, efficiently exploring complex reaction mechanisms remains time consuming and a highly non-systematic endeavour. Moreover, access to high-quality methodologies remains impractical for the study of reactions mechanisms in the condensed phase.<br/>This talk will present our team’s efforts to tackle these challenges by introducing automation and machine-learned (ML) potentials. I will describe novel developments of our tool, autodE, enabling us to explore increasingly challenging organic and organometallic reaction classes. To address the challenges faced when studying chemical reactions in solution, which often requires considerable human intervention and computational resources, we have introduced an efficient protocol to generate reactive machine-learned (ML) potentials. We show how this can be achieved with very few QM calculations within the Gaussian Approximation Potential (GAP) framework. Our work demonstrates that by leveraging active learning, inter-intramolecular decomposition, and hierarchical QM reference methods, reactive ML potentials can achieve ab initio accuracy at a much lower cost than related strategies.<br/>The potential of these approaches and their combined use is illustrated with a diverse range of examples, for which expensive AIMD simulations would otherwise be needed.

Exploring reactions mechanisms through automation and machine learning

ACS Spring 2022

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