SHRED: An open-source DFT code for exascale and matter in extreme conditions

Real space and planewave based Kohn-Sham Density Functional Theory codes are critical tools for studying condensed matter, chemical, material, and plasma physics. However, a large basis and the need to orthogonalize large numbers of orbitals/bands leads to computational complexity that scales cubically in both system size and temperatures, in the electron-volt regime. Additionally, significant communication bottlenecks limit parallel scaling across many nodes and/or GPU’s. In this talk, we present the SHRED (Stochastic and Hybrid Representation for Electronic structure by Density functional theory) code which utilizes alternative linear-scaling stochastic, mixed stochastic-deterministic, and orbital-free DFT and TD-DFT algorithms to circumvent orbital orthogonalization and achieve significant acceleration of calculations in a range of simulations. Newly implemented PAW pseudopotentials (based on Abinit’s LibPAW library), progress in GPU acceleration and new correlated sampling techniques, and applications to warm dense matter will be highlighted.