Making the Coupled Cluster Correlation Energy Machine-Learnable.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Materials Science

Background:

• Accurate electronic structure calculations are crucial across scientific disciplines.
• High-accuracy methods like coupled cluster (CC) theory are computationally expensive, limiting their application to larger systems.
• Approximate methods like Møller-Plesset perturbation theory (MP2) are less demanding but also less accurate.

Purpose of the Study:

• To reduce the computational scaling of accurate electronic structure calculations.
• To enable high-quality molecular dynamics simulations for larger systems.
• To integrate machine learning with established quantum chemistry methods.

Main Methods:

• Developed a vector-based representation of coupled cluster (CC) wave functions.
• Utilized potential energy surfaces of small molecules to train a machine learning model.
• Learned the CC correlation energy from the machine learning model using approximate amplitudes from Møller-Plesset (MP2) theory.

Main Results:

• Machine learning efficiently learned the CC correlation energy.
• The approach is effective even when using approximate amplitudes from MP2 theory.
• Demonstrated a potential pathway to achieve CC-quality results with reduced computational cost.

Conclusions:

• Machine learning offers a viable strategy to overcome the computational bottlenecks of accurate electronic structure methods.
• This work paves the way for computationally tractable, high-accuracy molecular dynamics simulations.
• The integration of machine learning with perturbation theory holds significant promise for computational chemistry.