Integral-Direct and Parallel Implementation of the CCSD(T) Method: Algorithmic Developments and Large-Scale...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Electronic Structure Theory

Background:

• Coupled-cluster singles, doubles, and perturbative triples [CCSD(T)] is a high-accuracy method for electronic structure calculations.
• Existing CCSD(T) implementations face challenges with computational cost and memory requirements for large systems.

Purpose of the Study:

• To develop a highly efficient and scalable CCSD(T) implementation.
• To enable accurate electronic structure calculations for larger and more complex molecular systems.

Main Methods:

• Integral-direct, disk I/O, and network traffic economic CCSD(T) implementation.
• Density-fitting approximation and exploitation of permutational symmetry.
• Hybrid MPI/OpenMP parallelization for strong scaling.

Main Results:

• Achieved 60-70% utilization of theoretical peak performance on hundreds of cores.
• Successfully performed some of the largest CCSD(T) calculations for systems of 31-43 atoms.
• Generated 13 correlation energies and 12 reaction energies/barrier heights for benchmark datasets.

Conclusions:

• The developed CCSD(T) implementation is highly efficient and scalable.
• This method facilitates large-scale electronic structure calculations relevant for DFT and ML parametrization.
• The results provide valuable reference data for molecules at the applicability limit of current methods.