More Numerical Precision for Less Compute Cost: Optimizing a Local Correlation Algorithm for Second Order...

Area of Science:

• Quantum Chemistry
• Computational Physics
• Theoretical Chemistry

Background:

• Electron correlation in molecules is computationally intensive.
• Second-order Møller-Plesset (MP2) theory is a key method for electron correlation.
• Local MP2 methods aim to reduce computational cost by exploiting spatial localization.

Purpose of the Study:

• To significantly improve a localized orthogonal orbital-based local MP2 approach.
• To enhance accuracy and computational efficiency for MP2 calculations.
• To develop a more robust and scalable quantum chemistry method.

Main Methods:

• Implemented a novel embedding correction for MP2 amplitudes.
• Utilized localized orthogonal orbitals for occupied and virtual spaces.
• Optimized algorithms including local fitting, sparse maps, and on-the-fly BLAS evaluation.

Main Results:

• Achieved an order of magnitude improvement in accuracy.
• Demonstrated significant reductions in memory usage and improved compute efficiency.
• Outperformed domain-localized pair natural orbital (DLPNO-MP2) in accuracy for a given time.

Conclusions:

• The enhanced local MP2 method offers a superior balance of accuracy and computational cost.
• The novel embedding correction is crucial for improving the performance of local MP2.
• This approach provides a more efficient pathway for complex molecular electronic structure calculations.