Arbitrary Angular Momentum Electron Repulsion Integrals with Graphical Processing Units: Application to the...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• The Hartree-Fock (HF) method is a fundamental tool in quantum chemistry for approximating the electronic structure of atoms and molecules.
• Efficient computation of electron repulsion integrals (ERIs) is a bottleneck in HF calculations, especially for large systems.
• Leveraging the parallel processing power of graphical processing units (GPUs) offers a promising avenue for accelerating these computations.

Purpose of the Study:

• To develop and present a resolution of identity (RI) based implementation of the Hartree-Fock method optimized for GPUs.
• To enable the calculation of integrals with arbitrary angular momentum, extending the applicability of GPU acceleration.
• To demonstrate significant performance improvements over traditional CPU-based implementations.

Main Methods:

• Implementation of the resolution of identity approximation for electron repulsion integrals.
• GPU acceleration of integral calculations using OpenCL for broad hardware compatibility.
• Hybrid CPU-GPU approach to handle integrals of varying angular momentum.
• Support for multiple GPU cards and parallelization using MPI or OpenCL.

Main Results:

• Achieved speedups of up to a factor of 30 compared to state-of-the-art serial and parallel CPU implementations.
• Successfully performed benchmark calculations using over 3500 contracted basis functions (def2-SVP and def2-TZVP basis sets).
• Demonstrated the capability to handle basis functions up to (ff|f) angular momentum on the GPU.

Conclusions:

• The presented GPU-accelerated Hartree-Fock implementation offers a substantial performance enhancement for quantum chemistry calculations.
• The hybrid approach effectively balances computational load between CPU and GPU, accommodating higher angular momentum integrals.
• The implementation's broad compatibility (OpenCL) and scalability (multi-GPU) make it a valuable tool for computational chemistry research.