First principles calculation using multiple scattering theory at the exascale.

Current high performance computer systems, such as Frontier at the Oak Ridge Leadership Computing Facility, are providing unprecedented opportunities for the quantitative exploration of complex materials. Here, I will present our implementation of multiple scattering theory for first principles density functional calculations. This approach directly obtains the single particle Green’s function of the Kohn-Sham equation, either in reciprocal space (Korringa-Kohn-Rostocker i.e. KKR) or real space (Localy-Selfconsistent Multiple Scattering i.e. LSMS). The KKR method allows an efficient description of random solid solution alloys using the Coherent Potential Approximation (CPA), while our LSMS code allows for scalable large scale first principles density functional calculations of materials. A fundamental science driver for scalable, large scale, first principles calculations of materials is the need to understand states beyond periodic crystalline lattices. For large simulation cells, needed to describe extended electronic and magnetic orderings, defect states or disorder in alloys, the cubic scaling of traditional first principles methods have prevented direct calculations. The linear scaling nature of the LSMS ab initio code enables the treatment of extremely large disordered systems from the first principles using the largest parallel supercomputers available, such ascalculations for O(10,000 - 1,000,000) atoms on current high performance computing architectures. For exascale systems, we have extended the use of accelerators to enable the efficient calculation for embedding methods and forces. Currently ongoing work focuses on the calculation of electric conductivity in the presence of disorder and defects. We will present scaling results of our LSMS code from single node calculations to the full Frontier system up to O(1,000,000) atom calculations.

These computational capabilities are available in our Multiple Scattering Theory suite (MuST) [https://github.com/mstsuite]