Highly Efficient Resolution-of-Identity Density Functional Theory Calculations on Central and Graphics Processing...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Density functional theory (DFT) calculations are crucial in chemistry and materials science.
• Evaluating Coulomb potentials and exchange-correlation terms can be computationally intensive, limiting calculation speed.
• Existing methods often face scaling challenges with increasing basis set sizes.

Purpose of the Study:

• To develop and present efficient algorithms for Coulomb potential and exchange-correlation calculations.
• To leverage graphics processing units (GPUs) for accelerating these computations.
• To demonstrate the performance improvements on both central processing units (CPUs) and GPUs.

Main Methods:

• Utilized the resolution of identity (RI) approximation for Coulomb potential evaluation.
• Implemented the J-engine algorithm for further optimization.
• Developed GPU-based algorithms for both Coulomb potential and exchange-correlation terms.
• Evaluated algorithms for potential and nuclear derivative terms.

Main Results:

• The GPU-based RI-algorithm demonstrates high performance and favorable scaling with basis set size.
• The J-engine based optimizations significantly improve CPU-based algorithm performance.
• GPU acceleration of exchange-correlation terms addresses a computational bottleneck in DFT.
• Overall high-performance method for density functional calculations achieved.

Conclusions:

• The presented GPU-accelerated methods significantly enhance the efficiency of DFT calculations.
• The algorithms offer substantial performance gains for Coulomb potential and exchange-correlation evaluations.
• This work provides a pathway for faster and more scalable quantum chemistry computations.