Thrust 2: beyond-DFT methods at exascale
In this thrust, we are optimizing the grand-canonical DFT method in the JDFTx code and the JDFTx-BerkeleyGW interface for solvated RPA calculations for exascale architectures. Additionally, we are implementing exascale-optimized, cubic-scaling RPA and RPA forces in BerkeleyGW, enabling self-consistent beyond-DFT geometries that are likely important for transition states described poorly by DFT. Using the universal electrochemical solvation methods from Thrust 1, we are facilitating large-scale electrochemical calculations to populate our electrochemical database, BEAST-DB, in Thrust 3.
The algorithms and codes developed within this project will target DOE leadership class HPC systems as we advance on the exascale path. While the emerging landscape for these architectures become more heterogeneous and complex, we aim to deliver a set of computational tools that are efficient and are of production quality, while still being maintainable and portable across architectures.