Machine Learning the Physical Nonlocal Exchange-Correlation Functional of Density-Functional Theory.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Materials Science

Background:

• Density-functional theory (DFT) is a powerful quantum mechanical method for electronic structure calculations.
• Traditional DFT approximations suffer from delocalization errors and lack exact consistency between energy and potential.
• Kohn-Sham equations provide a computationally efficient framework but rely on approximate exchange-correlation functionals.

Purpose of the Study:

• To develop a universal exchange-correlation functional for DFT using neural networks.
• To ensure simultaneous accuracy of both exchange-correlation energy and potential.
• To address delocalization errors and maintain computational efficiency in DFT calculations.

Main Methods:

• Training a neural network to serve as the exchange-correlation functional.
• Utilizing automatic differentiation to enforce the exact mathematical relationship between energy and potential.
• Testing the functional on one-dimensional systems with two strongly correlated electrons.

Main Results:

• The neural network functional reproduces exact exchange-correlation energies and potentials.
• The functional exhibits nonlocality while maintaining the computational scaling of semilocal approximations.
• Demonstrated feasibility in challenging systems where standard DFT methods fail.

Conclusions:

• The proposed neural network approach offers a consistent and accurate exchange-correlation functional for DFT.
• This method has the potential to overcome limitations of current DFT approximations.
• The approach maintains computational efficiency, making it suitable for complex electronic structure problems.