Physics-Based Machine Learning of the Electron Density

Prof. Michele Ceriotti

Laboratory of Computational Science and Modelling (COSMO),
MXG 337 (Bâtiment MXG), Station 12, CH-1015 Lausanne, Switzerland

michele.ceriotti@epfl.ch

Abstract

Statistical regression techniques have become very fashionable as a tool to predict the properties of systems at the atomic scale, sidestepping much of the computational cost of first-principles simulations and making it possible to perform simulations that require thorough statistical sampling without compromising on the accuracy of the electronic structure model.

In this talk I will argue how data-driven modelling can be rooted in a mathematically rigorous and physically-motivated framework, how this is beneficial to the accuracy and the transferability of the model, and how physical-chemical insights can be obtained by a critical application of ML methods. I will focus in particular on the problem of learning the ground-state electronic charge density, as an example that underscores the advantages of a transferable symmetry-adapted machine-learning framework.