Machine learning potentials in chemistry: Uses, opportunities, and challenges

Machine learning potential energy surfaces (ML-PES) have now become established as a powerful and versatile technique in computational chemistry, connecting the accuracy of quantum-mechanical methods with the large system sizes and extensive sampling required by many realistic applications, especially for macromolecular systems and amorphous or defective materials. However, their potential applications in chemistry are only beginning to be explored.

In this talk, I will present a few examples of recent innovations that dramatically expand the types of systems and experimental observables we can model with ML techniques, along with my ideas for applications of these innovations in research areas of interest within the Department. I will also discuss a few challenges that have persisted throughout the development of ML-PES techniques, and the work I (and others) am doing to address these challenges, allowing ML-PES techniques to be applied to more types of systems with higher accuracy and reliability than was possible before.

Hosted by: Mat Evans (mat.evans@york.ac.uk)