Seminar by Volker Deringer

Prof. Volker Deringer from the University of Oxford will deliver an invited seminar on Wednesday 25 March 2025 at 10:15-11:00 in hall A304 (Ke2) of the Aalto University School of Chemical Engineering main building, Kemistintie 1, Espoo.

Machine-learned interatomic potentials for materials chemistry

Volker Deringer

Department of Chemistry, University of Oxford

Machine-learned interatomic potentials (MLIPs) are now widely used in atomistic simulations, giving access to length and time scales that are otherwise inaccessible to first-principles methods. Their reliability in practice, however, depends crucially on the quality and coverage of the training data. In this talk, I will discuss data-efficient approaches to constructing MLIPs, including ML-accelerated first-principles molecular dynamics for generating reference configurations, and automated de novo exploration of relevant configurational spaces using the autoplex software. I will also highlight model-distillation strategies (specifically, a teacher–student approach) using accurate but computationally costly models to train fast, task-specific MLIPs for downstream applications. I will illustrate these ideas using examples of structurally complex materials: amorphous silicon, graphene oxide, and phase-change materials that are used in data storage and neuromorphic computing.