Robust, black-box algorithms for density-functional theory simulations

by Prof. Michael Herbst (EPFL)

Monday 4 Dec 2023, 11:00 → 12:00 Europe/Zurich

OFLG/402

With the advent of sophisticated tools for high-throughput materials discovery running thousands of first-principle simulations is becoming increasingly seamless. At this scale both the robustness and the efficiency of employed simulation algorithms is of unprecedented importance.

In this talk I will provide an overview of the research in the Mathematics for Materials Modelling group, where we employ recent ideas from mathematical analysis to understand simulation errors and improve the reliability of first-principle methods. In particular I will discuss recent black-box methods for density-functional theory (DFT) simulations as well as a Schur-complement-based approach for response methods and property calculations.

Our methods are implemented and tested using the density-functional toolkit (DFTK), a novel Julia-based density-functional theory code, which is accessible to mathematical research (only 7500 lines of code), but at the same time well-integrated into workflow libraries such as AiiDA.