Approaching coupled-cluster accuracy for molecular electronic structures with multi-task learning.

Area of Science:

• Quantum Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Machine learning models for molecular electronic properties often rely on density functional theory (DFT) data, limiting their accuracy.
• Existing models struggle to exceed DFT's predictive capabilities for molecular electronic properties.

Purpose of the Study:

• To develop a unified machine learning method for predicting electronic structures of organic molecules.
• To utilize highly accurate coupled cluster with singles and doubles and perturbative triples (CCSD(T)) calculations as training data.

Main Methods:

• Developed a novel machine learning approach for molecular electronic structures.
• Trained the model using data from gold-standard CCSD(T) calculations.
• Tested the model on hydrocarbon molecules, aromatic compounds, and semiconducting polymers.

Main Results:

• The machine learning model demonstrated superior performance compared to widely used DFT functionals.
• Achieved higher accuracy and computational efficiency in predicting various quantum chemical properties.
• Showcased excellent accuracy and generalization capabilities for complex systems.

Conclusions:

• The developed machine learning method offers a powerful alternative to DFT for electronic structure calculations.
• The model's ability to use CCSD(T) data enables accurate predictions for systems beyond DFT's limitations.
• This approach advances the application of machine learning in quantum chemistry.