The ab-initio GW method has emerged over the last couple decades as one of the most-trusted standards for predicting the excited state properties of materials - for example quasiparticle band energies, gaps and excitation spectra. The properties are important for the design of novel new materisl for applications in batteries, photovoltaics, LEDs etc. Density Functional Theory (DFT) calculation (even at the Hybrid-functional level) fail to accurately describe such properties for all systems in all environments but are often used because GW methods are thought to be limited to systems with ~100 atoms. We present here large scale GW calculations with systems of 1000’s of atoms (and many thousands of electrons) on the Cori (Intel Xeon Phi based) HPC system at NERSC. We show that the GW method scales well on such systems and can obtain a significant fraction of peak performance. We discuss some of the of the challenges and opportunties in using these emerging computing architectures. The GW approach, in multiple implementations, is particuarly well suited for exascale and pre-exascale systems.

Large scale GW calculations at full scale on pre-exascale HPC systems

254th National Meeting (2017)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.