This project aims to develop a new Hartree-Fock (HF) algorithm that will be able to exploit massive parallelism and whose scaling will be O(N2) instead of the traditional O(N4). The algorithm will have many other significant advantages over existing HF implementations. The GPU version will be tested and scaled on the Summit supercomputer at Oak Ridge, with future plans to run it on the Frontier system.
Programming models: C/C++, CUDA/HIP, MPI, OpenMP
C/C++ programming and good mathematical skills are essential.
For CPU implementation: good knowledge of OpenMP and MPI is required.
For GPU implementation: good knowledge of either CUDA, HIP or OpenCL/SyCL is required.
Massively Parallel Algorithm, High-Performance Computing, Algorithm Design, GPU Programming, Parallel Programming, Computational chemistry