Affordable and accurate large-scale hybrid-functional calculations on GPU-accelerated supercomputers.

Area of Science:

• Computational physics and chemistry
• Materials science
• Quantum mechanics

Background:

• High accuracy hybrid functional calculations for large condensed matter systems are computationally intensive.
• Existing methods often require approximations or are limited to smaller systems.
• Efficient computation is crucial for advancing materials discovery and understanding.

Purpose of the Study:

• To present a computationally efficient implementation for hybrid functional calculations.
• To enable high-accuracy density-functional theory (DFT) for systems with up to 1000 atoms.
• To reduce the time-to-solution for complex condensed matter simulations.

Main Methods:

• Developed a highly optimized multiple graphics processing unit (GPU) implementation of the exact exchange operator.
• Enabled fast hybrid functional DFT calculations using systematic basis sets without approximations.
• Integrated the method into a portable, open-source library.

Main Results:

• Achieved significant speed-up for hybrid functional DFT calculations on large systems.
• Enabled calculations for up to 1000 atoms with high accuracy.
• Reduced the time-to-solution to be comparable to traditional semilocal-GGA functionals.

Conclusions:

• The presented GPU-accelerated method makes high-quality hybrid DFT calculations accessible for large condensed matter systems.
• This advancement facilitates more accurate simulations on state-of-the-art supercomputers.
• The open-source library promotes wider adoption and further development in computational materials science.