Screening methods for linear-scaling short-range hybrid calculations on CPU and GPU architectures

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Short-range hybrid density functionals are increasingly used in computational chemistry.
• Efficient calculation of exchange interactions is crucial for accurate electronic structure methods.
• Existing methods may not be optimal for the specific decay properties of short-range functionals.

Purpose of the Study:

• To develop efficient screening schemes for short-range exchange calculations.
• To adapt existing linear-scaling methods (LinK and PreLinK) for short-range hybrid functionals.
• To enable accurate and fast electronic structure calculations on both CPU and GPU architectures.

Main Methods:

• Implementation of screening methods based on LinK and PreLinK algorithms.
• Accounting for the Coulomb operator decay in short-range hybrid functionals.
• Discussion of the GPU implementation for short-range electron repulsion integrals.

Main Results:

• Achieved speedups of up to 7.8 times compared to the underlying linear-scaling algorithms.
• Maintained full numerical control over accuracy.
• Demonstrated efficiency for both CPU and GPU architectures.

Conclusions:

• The developed screening schemes significantly reduce computational cost for short-range exchange calculations.
• These methods are applicable to a growing number of short-range hybrid functionals.
• The computational savings will benefit researchers using CPU and GPU computing resources.