Machine-learning-enhanced density functional theory calculations.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Machine Learning

Background:

• Density Functional Theory (DFT) is widely used in computational chemistry but has limitations in accuracy.
• Coupled Cluster (CC) methods offer higher accuracy but are computationally expensive.

Purpose of the Study:

• To develop a machine learning post-correction model to enhance DFT energy accuracy.
• To calibrate DFT total energies towards Coupled Cluster (CC) accuracy.

Main Methods:

• Training a machine learning model on energy differences between DFT and CC methods.
• Utilizing the G2 dataset comprising 56 small molecules for training.
• Applying a single post-processing correction step after standard DFT calculations.

Main Results:

• Reduced absolute energy errors from 358.7 kcal/mol (DFT) to 1.3 kcal/mol.
• Demonstrated significant error reduction in relative energies (atomization energies, ionization potentials, etc.).
• Showcased strong model transferability across various datasets.
• Achieved minimal additional time cost (0.69 s on average per G2 molecule).

Conclusions:

• The developed machine learning model offers a systematic and efficient approach to improve DFT accuracy.
• The method enhances the reliability of DFT for various energy-related calculations.
• This approach bridges the accuracy gap between DFT and high-level quantum chemistry methods cost-effectively.