High Performance Computing Seminar Series:Electronic Structure Based Discoveries on Highest-Performance Computers
Computational simulations based on the first principles of quantum mechanics are now virtually indispensable as complements to analytical theory and experimental approaches in molecular and materials science. High-performance computing resources play a key role in advancing the reach of electronic structure based simulations, aiming to predict new materials with atomic-scale tailored properties and to provide atomic-scale understanding of new phenomena observed experimentally. The present talk focuses on the FHI-aims code, developed for simulations of system sizes up to thousands of atoms, with benchmark-quality numerical accuracy and excellent scalability on current massively parallel high-performance computers. The current workhorse methods of the field are density-functional theory (DFT) for ground-state properties and many-body approaches to capture excited-state phenomena. A new, open-source infrastructure "ELSI" connects FHI-aims to even larger-scale simulations. We show how different computational aspects of the problem can be designed to efficiently use computing platforms with (ten) thousands of CPU cores, including recent developments to take advantage of new architectures such as Intel's manycore architectures (KNL) or NVidia's GPUs. Finally, we show how these approaches can aid the discovery of new materials, for the particular example of new hybrid organic-inorganic materials incorporating functional organic molecules for light emission.