The Kitchin Research Group

Machine learned potentials are revolutionizing molecular simulations. In this talk, I will introduce what machine learned potentials are, how we think about them, and how we create them. Then, I will show an example of how we use them to model segregation in a metal alloy surface at many different bulk compositions. Finally, I will show how automatic differentiation, which is one of the foundations of machine learning, can be used more broadly in scientific programming with derivatives.

This was the second talk I did, but it is somewhat of an introduction to the next talk below.

In this talk I talk about how we use machine learning to build cheap and accurate surrogate models of alloy catalyst surfaces in the dilute limit. I will show how we use this to simulate acrolein adsorption on dilute Ag-Pd alloys.