Large-scale electronic structure theory in FHI-aims and ELSI

Volker Blum
Duke University

This talk outlines new developments in the FHI-aims code and in the ELSI infrastructure that enable large-scale simulations of complex systems in production. The focus is on density functional theory (DFT), still the primary production method of practical electronic structure theory, but many-body methods will also be discussed as an outlook. FHI-aims enables fast calculations yet high numerical precision of an all electron code for complex systems. The ELSI infrastructure connects FHI-aims and other codes to several high-perfornance eigenvalue and density matrix solvers, including the massively parallel ELPA solver and distributed-parallel GPU support, as well as a proof of concept study of O(N^3) DFT on Google’s Tensor Processing Units. In FHI-aims, an optimized implementation of hybrid DFT enables full all-electron calculations up to thousands of atoms. In conjunction with spin-orbit coupling, this implementation unlocks practical simulations that were previously out of reach, such as direct simulations of effectively separated defects in perovskite semiconductors, using supercell sizes well over 1,000 atoms and hybrid DFT.

Presentation (PDF File)

Back to Workshop I: Increasing the Length, Time, and Accuracy of Materials Modeling Using Exascale Computing