Machine Learning Accurate Exchange-Correlation Potentials for Reducing Delocalization Error in Density Functional...

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Density functional theory (DFT) is a powerful quantum mechanical method for electronic structure calculations.
• Delocalization errors are a significant limitation in DFT, particularly for stretched molecular systems.
• Existing traditional and machine-learned functionals often fail to accurately describe these systems.

Purpose of the Study:

• To develop a machine learning approach for generating accurate exchange-correlation potentials in DFT.
• To address and reduce delocalization errors in electronic structure calculations.
• To improve the prediction of molecular properties for challenging systems.

Main Methods:

• Utilized a deep neural network to solve the Kohn-Sham equations self-consistently.
• Developed a novel machine-learned functional trained to minimize delocalization errors.
• Tested the approach on stretched molecular systems where traditional methods fail.

Main Results:

• The trained functional accurately captures the dissociation limit of stretched molecular systems.
• Achieved excellent agreement with CCSD reference data for electron densities and electric dipole moments.
• Demonstrated accurate prediction of atomic forces, overcoming limitations of existing methods.

Conclusions:

• The proposed machine learning approach effectively reduces delocalization errors in DFT.
• This method provides a robust tool for accurate prediction of molecular properties across diverse chemical species.
• Offers a promising advancement for computational chemistry and materials science applications.