Supervised learning of a chemistry functional with damped dispersion

Area of Science:

• Computational Chemistry
• Quantum Chemistry

Background:

• Kohn-Sham density functional theory (KS-DFT) is a cornerstone of computational chemistry.
• Existing functionals struggle to accurately predict the full spectrum of chemical properties.

Purpose of the Study:

• To develop a highly accurate singly hybrid functional for broad chemical applications.
• To improve upon the performance of existing non-doubly hybrid functionals.

Main Methods:

• Optimization of a flexible functional form (CF22D) combining global hybrid meta-nonseparable gradient approximation with damped dispersion.
• Utilized a large, combined database for training via performance-triggered iterative supervised learning.

Main Results:

• CF22D demonstrates superior across-the-board accuracy compared to most existing non-doubly hybrid functionals.
• Validated performance across diverse chemical benchmarks: barrier heights, isomerization energies, thermochemistry, noncovalent interactions, radical/nonradical chemistry, and transition-metal chemistry.
• Effective for both small and large, simple and complex systems.

Conclusions:

• The CF22D functional represents a significant advancement in KS-DFT accuracy for chemical predictions.
• Its flexible design and rigorous optimization yield a versatile tool for computational chemists.