Memory-Efficient Recursive Evaluation of 3-Center Gaussian Integrals.

Area of Science:

• Computational chemistry
• Quantum mechanics
• High-performance computing

Background:

• Gaussian integral evaluation is crucial for quantum chemistry.
• Existing methods face challenges with memory footprint on modern hardware.
• Density-fitting approximations require efficient computation of 3-center 2-particle integrals.

Purpose of the Study:

• To optimize FLOP-efficient Obara-Saika-based recursive schemes for reduced memory footprint.
• To enhance Gaussian integral evaluation on accelerated architectures.
• To improve the efficiency of Coulomb and 2-particle interactions in density-fitting.

Main Methods:

• Implemented multiquantal recurrences for significant memory savings.
• Utilized register memory for further memory footprint reduction.
• Employed direct compile-time kernel generation with C++/CUDA features.

Main Results:

• Demonstrated substantial memory savings using multiquantal recurrences.
• Achieved performance improvements in both conventional and CUDA-based implementations.
• Validated schemes for integrals with high angular momenta (up to L=6) and varying contraction degrees.

Conclusions:

• The optimized schemes offer significant memory and performance benefits for Gaussian integral evaluation.
• The developed methods are suitable for large-scale quantum chemistry computations.
• The implementation is available in the open-source LibintX library.