Improving Density Functional Prediction of Molecular Thermochemical Properties with a Machine-Learning-Corrected...

Area of Science:

• Computational chemistry
• Materials science
• Quantum mechanics

Background:

• Density Functional Approximations (DFAs) are crucial for electronic structure calculations.
• Machine Learning (ML) integration into DFAs shows promise but is often limited to model systems.
• The Perdew-Burke-Ernzerhof (PBE) functional is a widely used semilocal DFA.

Purpose of the Study:

• To develop a machine learning (ML) correction for the Perdew-Burke-Ernzerhof (PBE) functional.
• To enable ML-based density functional approximations for real-world molecular systems.
• To improve the accuracy of thermochemical property predictions.

Main Methods:

• Constructed a semilocal mapping using electron density and reduced density gradient.
• Developed a machine learning correction to the PBE exchange-correlation energy density.
• Applied the ML-corrected PBE functional to calculate heats of formation.

Main Results:

• The ML-corrected PBE functional significantly improves heats of formation for real molecules.
• The ML-PBE approach maintains accuracy for other thermochemical and kinetic properties.
• Demonstrated the applicability of ML-DFAs to complex chemical systems.

Conclusions:

• Combining data-driven ML with physics-based derivations enhances DFA accuracy.
• The ML-corrected PBE functional offers a pathway to achieving chemical accuracy.
• This work paves the way for more accurate and broadly applicable ML-based computational chemistry methods.