GPU-Accelerated Graph-Based Semiempirical Quantum Chemistry

Area of Science:

• Computational chemistry
• Materials science
• Quantum mechanics

Background:

• Graph-based electronic structure theory offers scalability for large atomistic systems.
• Distributed and hybrid computing platforms are crucial for complex simulations.

Purpose of the Study:

• To couple graph-based linear scaling electronic structure theory (SEDACS) with semiempirical quantum chemistry methods (PySEQM).
• To leverage Graphics Processing Unit (GPU) acceleration for enhanced computational efficiency.
• To enable scalable electronic structure calculations on distributed and hybrid platforms.

Main Methods:

• Integration of the Scalable Ecosystem, Driver, and Analyzer for Complex Chemistry Simulations (SEDACS) with the PySEQM code.
• Implementation of Graphics Processing Unit (GPU) acceleration within the coupled framework.
• Analysis of parallelization efficiency, computational accuracy, and communication overheads.

Main Results:

• Achieved efficient and scalable electronic structure calculations across multiple nodes.
• Demonstrated an order-of-magnitude speedup for systems up to 10,000 atoms.
• Validated the reduction in computational cost and effective harnessing of parallelism.

Conclusions:

• The combined SEDACS and PySEQM approach with GPU acceleration provides a powerful tool for large-scale electronic structure studies.
• This method significantly reduces computational expense for complex atomistic systems.
• The findings pave the way for more accessible and efficient computational chemistry research.