Combining low-cost electronic structure theory and low-cost parallel computing architecture.

Area of Science:

• Computational chemistry
• Electronic structure theory
• High-performance computing

Background:

• Electronic structure methods are computationally intensive.
• Heterogeneous computing architectures, like GPUs, can accelerate these calculations.
• TeraChem software has been developed since 2008 for GPU acceleration.

Purpose of the Study:

• To implement and evaluate three low-cost electronic structure methods (HF-3c, PBEh-3c, ωB97X-3c) on consumer-grade GPUs.
• To assess the performance benefits of mixed-precision integral handling in TeraChem.
• To explore the combination of these methods with the hh-TDA formalism for excited states.

Main Methods:

• Implementation of HF-3c, PBEh-3c, and ωB97X-3c methods in TeraChem.
• Leveraging mixed-precision integral handling for Gaussian integrals with angular momentum l < 3.
• Combining 3c methods with the hh-TDA formalism for excited state calculations.

Main Results:

• Significant performance improvements observed for HF-3c, PBEh-3c, and ωB97X-3c on consumer GPUs.
• ωB97X-3c achieves higher accuracy with the current GPU integral library limitations.
• New, efficient low-cost multi-configurational excited states methods are enabled via hh-TDA combination.
• Benchmarking demonstrates effective description of lowest vertical excitation energies.

Conclusions:

• Combining efficient electronic structure methods with affordable parallelized hardware offers an optimal cost-to-accuracy ratio.
• Consumer-grade GPUs can be effectively utilized to accelerate quantum chemical calculations.
• The implemented methods provide accessible and accurate computational tools for researchers.