IRTG Modern Inverse Problems (MIP)
Lade Veranstaltungen

« Alle Veranstaltungen

SSD Seminar Series with. Dr. Michael F. Herbst

Juli 12 @ 16:00 - 17:00

Dr. Michael F. Herbst – Accelerating the discovery of tomorrow’s materials by robust and error-controlled electronic-structure simulations

Applied and Computational Mathematics (ACoM), RWTH Aachen


Nowadays an established approach to discover novel materials is large-scale computational screening. In this approach the chemical andphysical properties of a large design space of candidate materials is systematically simulated in order to deduce the most suitable subset for further in-depth investigation. A central ingredient in most screening workflows are simulations based on density-functional theory~(DFT). Given that for a screening study commonly millions of DFT calculations are performed, respective DFT methods need to be efficient and robust. Unfortunately the numerical methods to solve the DFT problem require  the selection of a sizable number of parameters. In practice this is done heuristically, which can lead to suboptimal runtimes if parameters are chosen too conservative or to convergence failures if the selection has been too optimistic. This causes not only a waste in terms of computational resources, but also a constant requirement for human supervision despite all attempts to automate key steps in such studies.

As strategies to improve the robustness of state-of-the-art DFT methodsand thus to accelerate the discovery of novel materials, I will discuss in particular two aspects. One is the development of rigorous error analysis for DFT, such that numerical parameters can be chosen tightly adapted to the simulated material [1]. The other is the development of parameter-free and robust numerical strategies for the DFT fixed-point problem. The conditioning of this fixed-point problem is directly linked to the dielectric properties of the simulated material. For this reason common preconditions are either specific to only a small class of materials (e.g. only cover bulk insulators or metals) or involve parameters, which are difficult to choose a priori. In contrast our recent work resulted in a parameter-free preconditioning strategy applicable to a large range of inhomogeneous systems. Importantly this
includes subclasses such as catalytic surfaces or metallic clusters [2]. For cases where no suitable preconditioner is known, we are currently working on adaptive stepsize methods, which focus on improving the robustness of DFT simulations. The main tool for our efforts is the density-functional toolkit (DFTK) [3], a software framework for multidisciplinary research on DFT. In our research this code has been valuable both to support theoretical analysis by numerical experiments, to develop novel algorithms and to connect to application scientists fortesting our methods on real-world simulation problems.

[1] M. F. Herbst, A. Levitt and E. Cancès. Faraday Discuss. 224, 227 (2020).
[2] M. F. Herbst and A. Levitt. J. Phys. Condens. Matter, 33, 085503 (2021).
[3] M. F. Herbst, A. Levitt and E. Cancès. JuliaCon Proc. 3, 69 (2021).


Juli 12
16:00 - 17:00


a link for the Zoom meeting room will be send in the newsletter one week before the seminar starts. If you need any organizational help please contact office@aices.rwth-aachen.de