Subsystem Density Functional Theory Augmented by a Delta Learning Approach to Achieve Kohn-Sham Accuracy

Area of Science:

• Computational physics and chemistry
• Electronic structure theory
• Materials science

Background:

• Kohn-Sham density functional theory (KS-DFT) is a reference method for ab initio molecular dynamics simulations.
• High computational costs of KS-DFT limit system size and sampling in condensed matter simulations.
• Subsystem density functional theory (SDFT) approaches, like the Kim-Gordon (KG) scheme, offer reduced computational expense but have limitations.

Purpose of the Study:

• To improve the Kim-Gordon (KG) scheme for subsystem density functional theory.
• To enable accurate and efficient molecular dynamics simulations of condensed systems.
• To reduce the computational cost of electronic structure calculations while maintaining accuracy.

Main Methods:

• Proposed an enhanced Kim-Gordon (KG) scheme for subsystem density functional theory.
• Introduced a computationally inexpensive correction to the electronic kinetic energy term.
• Utilized a machine learning procedure to determine the correction.
• Implemented the scheme within a linear scaling self-consistent field formalism.

Main Results:

• The enhanced KG scheme matches Kohn-Sham density functional theory (KS-DFT) accuracy for energies and forces.
• Achieved significant reduction in computational time for condensed molecular systems.
• Demonstrated the scheme's applicability through molecular dynamics simulations of liquid water.
• The determined correction showed transferability across different system sizes and temperatures.

Conclusions:

• The proposed Kim-Gordon (KG) scheme offers a computationally efficient alternative to KS-DFT for condensed matter simulations.
• The machine learning-corrected KG scheme maintains high accuracy, enabling larger system sampling.
• This approach facilitates more extensive molecular dynamics studies of complex molecular systems.