Physics colloquium

Transferable and Scalable Electronic Structure Simulations Powered by Machine Learning

Dr. Lenz Fiedler

CASUS Görlitz

Abstract:

Electronic structure simulations allow researchers to compute fundamental properties of materials without the need for experimentation. As such, they routinely aid in propelling scientific advancements across materials science and chemical applications. Over the past decades, density functional theory (DFT) has emerged as the most popular technique for electronic structure simulations, due to its excellent balance between accuracy and computational cost. However, even with the most efficient implementations, electronic structure simulations are usually restricted to a few thousand atoms.

Machine-learning DFT (ML-DFT) tackles this challenge by providing rapid access to observables of interest. Most current ML-DFT methodologies focus on the mapping between ionic configurations and scalar observables, rather than a full prediction of electronic structure. A recently proposed alternative lies in the prediction of the local density of states (LDOS) as a versatile representation of the electronic structure. LDOS-based ML-DFT models provide access to a range of electronic structure observables, such as the electronic density, DOS and total free energy.
 

In this talk, an overview over LDOS-based ML-DFT models, their mathematical foundation and application is given. It is shown how such models can be used to scale electronic structure simulations to experimentally relevant length scales while maintaining transferability across relevant simulation parameters, such as temperature or phase boundaries. Further, an introduction to the Materials Learning Algorithms (MALA) package, an open-source python package developed for building and applying LDOS-based ML-DFT models, is given. Finally, current research directions, namely applications to multi-species systems and twisted 2D-materials are discussed.