Linear-scaling self-consistent field calculations based on divide-and-conquer method using resolution-of-identity...

Area of Science:

• Computational chemistry
• Quantum chemistry
• Materials science

Background:

• Graphics Processing Units (GPUs) are increasingly utilized in computational chemistry for methods like Hartree-Fock (HF).
• Ab initio calculations for large molecules face challenges including slow CPU-GPU memory access and limited GPU memory.
• Existing methods struggle with computational cost and memory constraints for large-scale molecular simulations.

Purpose of the Study:

• To implement a Divide-and-Conquer Resolution-of-the-Identity Hartree-Fock (DC-RI-HF) code on GPUs.
• To overcome computational bottlenecks in ab initio calculations of large molecules.
• To enhance the efficiency and scalability of quantum chemistry computations.

Main Methods:

• Implementation of the DC-RI-HF method on GPU architecture.
• Utilization of optimized math libraries for GPU compatibility and performance.
• Application of the divide-and-conquer strategy to partition molecular systems.
• Incorporation of the resolution-of-the-identity approximation to reduce computational cost.

Main Results:

• Significant acceleration of Hartree-Fock (HF) calculations using the GPU-implemented code.
• Demonstrated accuracy of the DC-RI-HF method on GPUs comparable to traditional methods.
• Successful mitigation of memory access and memory shortfall issues.
• Validation of the code's compatibility with future GPU architecture developments.

Conclusions:

• The GPU-accelerated DC-RI-HF code offers a powerful approach for large-scale molecular calculations.
• This implementation effectively addresses computational and memory limitations in ab initio chemistry.
• The developed method provides a scalable and accurate solution for quantum chemistry simulations.