Acceleration of High Angular Momentum Electron Repulsion Integrals and Integral Derivatives on Graphics Processing...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• High-Performance Computing

Background:

• Ab initio self-consistent field (SCF) calculations are fundamental in quantum chemistry.
• Efficient computation of electron-repulsion integrals (ERIs) and their derivatives is crucial for SCF methods.
• Current GPU implementations have limitations in handling higher-order ERIs and derivatives.

Purpose of the Study:

• To develop and implement an efficient GPU-accelerated algorithm for ab initio SCF energy and gradient calculations.
• To extend GPU capabilities to include f-type ERIs and d-type ERI derivatives.
• To analyze the performance and accuracy of the GPU implementation compared to traditional CPU methods.

Main Methods:

• Utilizing recurrence relations and machine-generated code for ERI calculations on CUDA-enabled GPUs.
• Porting SCF gradient calculations to GPUs, optimizing the computation of ERI first-order derivatives.
• Developing a novel algorithm to compute f-type ERIs and d-type ERI derivatives on GPUs.

Main Results:

• Achieved significant speedups of 10-18 times for ERI and ERI derivative computations using GPUs compared to CPUs.
• Successfully implemented the computation of f-type ERIs and d-type ERI derivatives on GPUs.
• Double-precision calculations confirmed satisfactory overall accuracy for most applications.

Conclusions:

• The developed GPU implementation offers a substantial acceleration for ab initio SCF calculations.
• The extended algorithm enables efficient computation of higher-order ERIs and derivatives on GPUs.
• This approach enhances the feasibility of large-scale quantum chemistry simulations.