Koopmans' Theorem-Compliant Long-Range Corrected (KTLC) Density Functional Mediated by Black-Box Optimization and...

Area of Science:

• Computational chemistry
• Materials science
• Quantum mechanics

Background:

• Density functional theory (DFT) is crucial in chemistry, physics, and materials science.
• Accurate molecular property prediction requires precise functional parameterization in DFT.
• Existing DFT functionals often need molecule-specific parameter adjustments.

Purpose of the Study:

• To optimize parameters of range-separated functionals (LC-BLYP, CAM-B3LYP) using Bayesian optimization (BO).
• To ensure optimized functionals satisfy Koopmans' theorem for improved accuracy.
• To develop a machine learning model for predicting optimal DFT parameters.

Main Methods:

• Bayesian optimization (BO) was employed to tune parameters of LC-BLYP and CAM-B3LYP functionals.
• Koopmans' theorem was used as the optimization target criterion.
• A dataset of over 3000 molecules was generated for training a machine learning model.

Main Results:

• Bayesian optimization effectively optimized functional parameters.
• The Koopmans' theorem-compliant LC-BLYP (KTLC-BLYP) achieved accuracy comparable to experimental UV-absorption values.
• A machine learning model was successfully developed to predict LC-BLYP parameters.

Conclusions:

• Bayesian optimization is a powerful tool for refining DFT functionals.
• KTLC-BLYP shows significant potential for accurate molecular property prediction.
• The developed machine learning approach facilitates automated parameter prediction for DFT.