Regularized Second-Order Møller-Plesset Theory: Linear Scaling Implementation and Assessment on Large-Molecule...

Area of Science:

• Computational chemistry
• Quantum chemistry
• Method development

Background:

• Second-order Møller-Plesset (MP2) theory, while widely used, exhibits limited accuracy for large intermolecular interactions and lacks a lower energy bound.
• Regularized MP2 methods offer potential improvements in accuracy and robustness without altering the fifth-order scaling.

Purpose of the Study:

• To develop and implement a linear scaling algorithm for two regularized MP2 methods: kappa-MP2 and size-consistent second-order Brillouin-Wigner perturbation theory (BW-s2).
• To assess the performance and accuracy of these regularized methods for large-scale molecular systems and intermolecular interactions.

Main Methods:

• A linear scaling algorithm with shared and distributed memory parallel implementations was developed for kappa-MP2 and BW-s2.
• The algorithm's accuracy was controlled by a single numerical threshold, and its time-to-solution was evaluated across various system sizes (up to 1800 atoms), thresholds, and processor counts (up to 2000 cores).
• Performance was benchmarked against canonical implementations and coupled cluster data for intermolecular interaction datasets (IONPI19, L7, S12L, C60ISO).

Main Results:

• The new implementation demonstrates effective linear scaling for kappa-MP2 and BW-s2.
• BW-s2 shows an earlier crossover to favorable scaling compared to MP2 and kappa-MP2.
• kappa-MP2 and BW-s2 significantly outperform MP2 for large intermolecular interactions, with comparable performance to MP2 for short-range amino acid interactions.

Conclusions:

• The developed linear scaling algorithm provides an efficient and accurate means to evaluate kappa-MP2 and BW-s2.
• Regularized MP2 methods like kappa-MP2 and BW-s2 represent significant advancements over standard MP2 for studying large molecular systems and intermolecular interactions.
• These methods offer a more robust and accurate alternative for computational chemistry applications.