Constructing and representing exchange-correlation holes through artificial neural networks.

Area of Science:

• Computational Physics
• Quantum Chemistry
• Materials Science

Background:

• Density functional theory (DFT) approximations for exchange-correlation (XC) energy rely on the XC hole.
• Constructing accurate XC holes often requires significant mathematical intuition and manual effort.

Purpose of the Study:

• To develop a machine-driven approach for constructing approximations to the XC hole.
• To automate the generation of XC energy approximations using physical constraints.

Main Methods:

• Adaptation of machine learning algorithms guided by physical constraints.
• Development of the ExMachina tool for automated generation of XC hole approximations.
• Application of ExMachina to calculate model XC holes.

Main Results:

• Successful implementation of ExMachina for generating XC hole approximations.
• Demonstration of ExMachina's capability to create novel approximations.
• Potential to surpass existing XC hole approximation methods.

Conclusions:

• ExMachina offers a novel, constraint-based machine learning approach to automate XC hole construction.
• This method facilitates the development of improved approximations for exchange-correlation energy in DFT.
• The tool provides a pathway to explore and generate new XC approximations beyond current limitations.