Juliacon 2024

DFTK.jl: 5 years of a multidisciplinary electronic-structure cod
07-12, 11:00–11:30 (Europe/Amsterdam), Function (4.1)

This is an update on the density-functional toolkit (DFTK), a density-functional theory code modelling the electronic properties of materials. Since our 2021 talk substantial progress has been made to enhance the code's features and integrations both within the Julia ecosystem as well as key domain-specific packages such as the AiiDA workflow engine. Examples illustrate the code's use as a platform supporting researches from multiple disciplines.

Simulating the electronic structure of matter using density-functional theory (DFT) is an established approach in the chemical and physical sciences. On major supercomputers about 30% of the computational time are spent just on DFT simulations. At this scale the importance to improve efficiency and robustness of simulation algorithms cannot be overstated. This frequently requires a multidisciplinary effort involving fields such as mathematics, computer science, physics and materials science. With DFTK we aim to provide a software platform exactly to foster such cross-disciplinary research.

In this talk we look back to 5 years of DFTK with over 30 contributors. We sketch the design of our package and illustrate our integration with both the Julia ecosystem (CUDA, ForwardDiff, AtomsBase) as well as standard high-throughput workflow engines from the application domain. In the past this has been used both to propose novel numerical schemes with strong mathematical guarantees as well as perform in-depth verification testing of DFTK against other more established packages of the field,
such as QuantumESPRESSO. Both mathematical exploration on simplified models as well as state-of-the-art application simulations can thus be performed with DFTK, making it the ideal code for prototyping novel first-principle simulation approaches.

Michael is currently a tenure-track assistant professor for mathematics and materials science at EPFL, Lausanne. There he leads the MatMat research group, which aims to understand simulation error and improve robustness of materials modelling schemes. As part of this effort his group co-develops the density-functional toolkit.

This speaker also appears in: