Massively Parallel Implementation of Explicitly Correlated Coupled-Cluster Singles and Doubles Using TiledArray...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• Coupled-cluster singles and doubles (CCSD) calculations are essential for accurate molecular electronic structure.
• Existing implementations face challenges with computational cost and memory for larger systems.
• Explicitly correlated (F12) methods improve basis set convergence but add complexity.

Purpose of the Study:

• To present a novel distributed-memory, massively parallel implementation of CCSD and explicitly correlated CCSD (CCSD(F12)).
• To leverage the TiledArray tensor framework for efficient data management and computation.
• To enable routine, accurate calculations of coupled-cluster basis set limits for molecules with 20+ atoms.

Main Methods:

• Developed a massively parallel implementation of CCSD and CCSD(F12) with O(N^6) complexity.
• Utilized the TiledArray framework for distributed memory management.
• Incorporated density fitting and integral-driven formulations to reduce memory footprint.

Main Results:

• Demonstrated excellent strong scaling on various computing architectures.
• Achieved competitive performance against established CCSD codes (ORCA, Psi4) and superior performance over NWChem.
• Successfully applied the method to recalculate the CCSD binding energy of the uracil dimer.

Conclusions:

• The new implementation facilitates accurate basis set limit estimations for larger molecular systems.
• Provides valuable benchmarks for emerging reduced-scaling coupled-cluster methods.
• The revised CCSD correlation binding energy for the uracil dimer differs from previous estimates, highlighting the importance of accurate high-level calculations.