Simple Automatized Tool for Exchange-Correlation Functional Fitting.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Materials Science

Background:

• Density functional theory (DFT) is a cornerstone of computational chemistry, offering a balance between accuracy and computational cost.
• Existing DFT functionals, while numerous, often require complex and time-consuming design processes.
• Developing new functionals for specialized tasks remains a significant challenge despite available quantum-chemical data.

Purpose of the Study:

• To introduce a novel, efficient approach for designing and optimizing density functional theory (DFT) functionals.
• To simplify and accelerate the development of new DFT approximations.
• To demonstrate the creation of specialized DFT functionals that surpass existing popular options.

Main Methods:

• A Bayesian search strategy combined with stochastic sub-sampling was employed.
• The approach incorporates the 'history' of fitting steps to improve efficiency.
• Techniques to reduce computational time per step and prevent overfitting to training data were implemented.

Main Results:

• The proposed Bayesian search method significantly accelerates the design of DFT functionals.
• Specialized DFT functionals were successfully trained, demonstrating superior performance compared to widely used functionals.
• The developed approach was validated through general efficiency testing and specific application examples.

Conclusions:

• The novel Bayesian search approach offers a simplified and accelerated pathway for DFT functional development.
• This method enables the creation of high-performing, specialized DFT approximations for specific chemical tasks.
• The freely available code, coupled with reference databases, empowers researchers to construct tailored DFT approximations.