Scientific Achievement
- Researchers in the SciDAC 5 Scientific Partnership program state that quantum chemistry on leadership computing is most powerful if 2 conditions are met: this is achieved for 2nd order many-body theory (MP2). The problem formulation should have linear scaling effort with problem size. SciDAC reduced MP2 from 5th order scaling to linear scaling, with a new higher precision approach. Efficient distributed memory parallel implementation is required. Close partnership between computer scientists and domain scientists enabled us to achieve this goal at an unprecedented scale for this domain
Significance and Impact
- MP2 is the building block of double hybrid density functionals, which are the most accurate available. This work paves the way for linear scaling massively parallel double hybrid density functionals. We have extended the method to treat regularized MP2 methods, which offer higher accuracy than MP2 itself. Further extensions are underway
Research Details
- Dual level OpenMP / MPI calculations used as many as 2048 cores on the NERSC Perlmutter machine. The figure shows strong scaling results in the relatively large def2-TZVP basis set for 5 large molecules
Z. Wang, T. Shi, W. Luo, H.J. Kulik, Y.Liu, X.S. Li, M. Head-Gordon, Journal of Chemical Theory and Computation (2025).
Work was performed at Lawrence Berkeley National Lab, University of California Berkeley, and in part at NERSC User Facility.
