A modular, composite framework for the utilization of reduced-scaling Coulomb and exchange construction algorithms:...

Area of Science:

• Quantum Chemistry
• Computational Chemistry
• Materials Science

Background:

• Multiple algorithms exist for calculating Coulomb (J) and exchange (K) contributions to Fock-like matrices.
• Integrating diverse J and K construction algorithms into a unified framework is beneficial for computational efficiency.

Purpose of the Study:

• To implement a flexible framework, "CompositeJK," for combining different J and K construction algorithms in quantum chemistry.
• To enhance Psi4's capabilities by integrating novel computational methods and hardware acceleration.

Main Methods:

• Implementation of the "CompositeJK" formalism in the Psi4 quantum chemistry software.
• Interfacing Psi4 with the sn-LinK implementation from the GauXC library.
• Leveraging both Central Processing Unit (CPU) and Graphics Processing Unit (GPU) for Fock matrix element construction.

Main Results:

• The CompositeJK framework allows modular combination of J and K algorithms for various quantum chemistry methods.
• Integration with sn-LinK provides the first non-commercial GPU support for Fock matrix construction in Psi4.
• CPU-based sn-LinK shows up to 2.8x speedup over existing Psi4JK methods for large systems.
• GPU-accelerated sn-LinK achieves up to 7.0x performance gains.

Conclusions:

• The CompositeJK framework offers an intuitive and modular approach for developing and utilizing advanced computational methods in quantum chemistry.
• The integration of sn-LinK, particularly with GPU acceleration, significantly enhances the performance of Fock matrix calculations in Psi4.
• This advancement facilitates larger and more complex quantum chemistry simulations, benefiting fields like materials science.