Speaker: Michele Ceriotti (Oxford)

Title: Understanding structural complexity with sketch map

Abstract:

Atomistic computer simulations often involve following the concerted evolution of thousands of degrees of freedom. A crucial step to get insight from these simulations, and to make them more effective, consists in defining a small number of order parameters which must be capable to capture the essential underlying physics. In this talk, I'll discuss how machine learning techniques can be exploited to perform this task in an automated way, provided that features specific to atomistic modeling are properly accounted for. I will introduce the methods involved -- which are not dissimilar from those used to draw a map of the curved surface of the earth on flat piece of paper. I will present applications to the description and the accelerated dynamics of a model peptide. Furthermore, I will demonstrate how the very same techniques can be exploited to give a very detailed description of the free energy landscape of a cluster, and discuss how they could be transferred to the study of defects in solids and disordered systems.

Lunch group: 5