GPU-Based Implementations of the Noniterative Regularized-CCSD(T) Corrections: Applications to Strongly Correlated...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Calculating electronic energies for strongly correlated systems is computationally demanding.
• Existing methods struggle with accuracy in the presence of strong quasidegeneracy effects.
• The regularized CCSD(T) (Reg-CCSD(T)) method was recently introduced to address these challenges.

Purpose of the Study:

• To detail the graphical processing unit (GPU) implementation of the computationally intensive (T)-part of the Reg-CCSD(T) method.
• To assess the scalability and performance of the Reg-CCSD(T) GPU implementation.
• To demonstrate the accuracy of the Reg-CCSD(T) method for challenging molecular systems.

Main Methods:

• Graphical processing unit (GPU) implementation of the (T)-part of the Reg-CCSD(T) method.
• Parallel performance tests on various molecular systems (methylfluoride, dodecane, Spiro cation, uracil).
• Analysis of performance based on spinorbital domain partitioning (tiling structure).

Main Results:

• The Reg-CCSD(T) GPU implementation exhibits very good scalability for the triples part.
• Performance is analyzed concerning tiling structure parameters.
• Reg-CCSD(T) significantly improves the accuracy of ground-state energies in systems with strong quasidegeneracy.

Conclusions:

• The GPU implementation of Reg-CCSD(T) is efficient and scalable for strongly correlated systems.
• The method demonstrates high accuracy, outperforming standard approaches in challenging electronic structure cases.
• Reg-CCSD(T) is a valuable tool for accurate electronic energy calculations, especially for systems with quasidegeneracy.