Development of hybrid MPI/Open parallelization strategies for quantum chemistry software
Title: Development of hybrid MPI/Open parallelization strategies for quantum chemistry software
DNr: SNIC 2020/5-691
Project Type: SNIC Medium Compute
Principal Investigator: Zilvinas Rinkevicius <rinkevic@theochem.kth.se>
Affiliation: Kungliga Tekniska högskolan
Duration: 2021-01-01 – 2022-01-01
Classification: 10407
Homepage: https://veloxchem.org
Keywords:

Abstract

Main objective of this proposal to develop and implement novel hybrid parallelization strategies for quantum chemistry software, which go beyond "master-worker" scheme employed in conventional quantum chemistry packages. In previous year, we successfully implemented hybrid OpenMP/MPU parallelisation for self-consistent field method (good scaling achieved up to 16 000 CPU cores on Beskow [1]). The next step in this project, is implementation of distributed storage scheme for Fock matrices in self-consistent field and linear response calculations, which would allow us to extend applicability of VeloxCHem code beyond limit of 10000 contracted basis functions. These developments will enable efficient computations of electronic structure and properties of molecular systems on modern HPC systems. References: [1] Zilvinas Rinkevicius, Xin Li, Olav Vahtras, Karan Ahmadzadeh, Manuel Brand, Magnus Ringholm, Nanna Holmgaard List, Maximilian Scheurer, Mikael Scott, Andreas Dreuw, and Patrick Norman. "VeloxChem: a Python-driven density-functional theory program for spectroscopy simulations in high-performance computing environments", WIREs Computational Molecular Science, 2019;e1457. https://doi.org/10.1002/wcms.1457