Delta-Augmented Subsystem Density Functional Theory: A Study Across Diverse Systems

Area of Science:

• Computational chemistry
• Materials science
• Quantum mechanics

Background:

• Density Functional Theory (DFT) methods are computationally expensive for large molecular systems.
• Subsystem DFT approaches offer a way to reduce computational cost but often lack accuracy.
• Machine learning (ML) corrections show promise for improving DFT accuracy.

Purpose of the Study:

• To benchmark and expand the applicability of the Kim-Gordon (KG) method, a subsystem DFT approach with ML corrections.
• To assess the performance of the KG method for complex molecular liquids beyond water.
• To evaluate the transferability of ML-derived corrections.

Main Methods:

• The study utilizes 'delta-learning' to train ML corrections based on Kohn-Sham (KS) DFT data.
• The KG method is applied to condensed molecular systems, including bulk ammonia and methanol.
• The approach is combined with linear-scaling self-consistent field (LS-SCF) techniques.

Main Results:

• The KG method with ML corrections achieves Kohn-Sham DFT accuracy at a fraction of the computational cost.
• Successful application to complex molecular liquids like ammonia and methanol demonstrates broad applicability.
• ML corrections trained on bulk data showed transferability to molecular clusters.

Conclusions:

• The Kim-Gordon method with ML corrections is a computationally efficient tool for molecular dynamics.
• The 'delta-learning' approach significantly reduces the need for training data.
• The method offers a viable pathway for accurate simulations of complex molecular systems.