Machine Learning for Bridging the Gap between Density Functional Theory and Coupled Cluster Energies.

Area of Science:

• Computational Chemistry
• Machine Learning
• Quantum Chemistry

Background:

• Accurate electronic energies are vital for chemical reaction design and mechanistic studies.
• High-level computational methods like coupled cluster theory provide accuracy but are limited by computational cost for large systems.
• There is a need for fast and accurate methods to compute electronic energies for larger molecular systems.

Purpose of the Study:

• To develop machine learning models for predicting accurate electronic energies of molecular systems.
• To create a comprehensive database of electronic energies for training machine learning models.
• To enable faster and more accessible high-level energy calculations for computational chemistry.

Main Methods:

• A database of ~8000 organic monomers and ~2000 dimers was created with energies computed at B3LYP-D3(BJ)/cc-pVTZ.
• Single-point energies were calculated using various density functional theory and coupled cluster methods.
• Graph neural network models were trained using two distinct graph representations on the generated dataset.

Main Results:

• Machine learning models achieved mean absolute errors of 0.78 kcal mol⁻¹ for predicting CCSD(T)/cc-pVTZ energies from B3LYP-D3(BJ)/cc-pVTZ inputs.
• Models predicted DLPNO-CCSD(T)/cc-pVTZ energies with mean absolute errors of 0.50 kcal mol⁻¹ for monomers and 0.18 kcal mol⁻¹ for dimers.
• The dimer model was validated on the S22 database, and the monomer model was tested on complex molecular systems.

Conclusions:

• Machine learning models, particularly graph neural networks, can accurately predict high-level electronic energies for molecular systems.
• The developed models offer a computationally efficient alternative to traditional high-level methods for large systems.
• This approach has significant implications for accelerating reaction design and mechanistic investigations in computational chemistry.