A semilocal machine-learning correction to density functional approximations.

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Density Functional Theory (DFT) is crucial for materials and molecular simulations.
• Existing DFT functionals have limitations in accurately predicting various energetic properties.
• Machine learning (ML) offers a promising avenue for refining DFT methods.

Purpose of the Study:

• To develop and validate a machine learning model for correcting density functional approximations.
• To enhance the predictive accuracy of DFT functionals for diverse energetic properties.
• To assess the performance and efficiency of the ML-corrected functional.

Main Methods:

• Constructed an ML model using semilocal descriptors of electron density and its derivative.
• Trained the ML model with accurate reference data for relative and absolute energies.
• Tested the ML-corrected functional on a comprehensive dataset of energetic properties.

Main Results:

• The ML-corrected Becke's three parameters and Lee-Yang-Parr correlation (B3LYP) functional showed substantial improvement in predicting total energies and atomization energies.
• Marginal improvements were observed for ionization potentials, electron affinities, and bond dissociation energies.
• Accuracy for isomerization energies and reaction barrier heights was preserved, with similar computational efficiency to the original B3LYP functional.

Conclusions:

• The developed ML correction significantly enhances the accuracy of DFT functionals for key energetic predictions.
• The ML-corrected functional demonstrates potential for broader applicability and improved performance over standard B3LYP.
• This work represents a significant step towards uniformly superior DFT functionals through machine learning integration.