Machine Learning-Enhanced Orbital-Free Density Functional Theory.

Area of Science:

• Computational Physics and Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Orbital-free density functional theory (OF-DFT) enables large-scale ab initio calculations for millions of atoms.
• Accuracy of kinetic energy functionals (KEF) and pseudopotentials (PP) are key limitations for OF-DFT adoption.
• Existing methods struggle with the computational cost and accuracy for massive atomic systems.

Purpose of the Study:

• To review the current state of machine learning (ML) applications in OF-DFT.
• To explore ML-based construction of KEFs, PPs, and electron density prediction.
• To discuss the impact of ML on expanding OF-DFT's applicability and accuracy.

Main Methods:

• Review of existing literature on ML for KEFs and PPs in OF-DFT.
• Analysis of various ML techniques including neural networks, kernel regressions, and symbolic regressions.
• Discussion on data requirements and challenges in ML model development for OF-DFT.

Main Results:

• ML has shown significant promise in developing accurate KEFs and OF-DFT-suited PPs.
• ML models can effectively predict electron density, further improving OF-DFT calculations.
• These advancements are expanding the scope and reliability of large-scale atomic simulations.

Conclusions:

• ML is a transformative approach for overcoming key limitations in OF-DFT.
• The integration of ML is crucial for the future development and broader application of OF-DFT.
• Future research should focus on diverse ML methods and robust data strategies for OF-DFT.