Efficient implementation of effective core potential integrals and gradients on graphical processing units.

Area of Science:

• Computational Chemistry
• Materials Science
• High-Performance Computing

Background:

• Accurate calculation of core potential integrals and gradients is computationally intensive.
• Scaling of computational cost with system size limits the study of large molecular systems.

Purpose of the Study:

• To accelerate core potential integral and gradient evaluations using graphical processing units (GPUs).
• To develop efficient screening and workload balancing strategies for GPU-accelerated computations.
• To assess the performance, scalability, and accuracy of the GPU implementation.

Main Methods:

• Implementation of core potential integral and gradient calculations on GPUs.
• Development of two formulas for estimating integral upper bounds for screening.
• Design of a sorting strategy to balance workload across GPU threads.
• Combination of screening and sorting methods for performance optimization.

Main Results:

• Significant performance improvements and reduced scaling with system size were observed.
• The method is highly efficient for systems up to 10,000 basis functions.
• GPU implementation maintained calculation precision.
• Accurate ground state Hartree-Fock energies were obtained for CdSe and ZnTe nanocrystals.
• Energy conservation was demonstrated in ab initio molecular dynamics simulations.

Conclusions:

• GPU acceleration with screening and sorting strategies offers a highly efficient approach for large-scale electronic structure calculations.
• The method provides accurate results and is suitable for studying complex systems like nanocrystals.
• This approach enhances the feasibility of ab initio molecular dynamics simulations for larger systems.