Adapting hybrid density functionals with machine learning.

Area of Science:

• Quantum chemistry
• Materials science
• Computational physics

Background:

• Exact exchange is crucial for electronic states and chemical bonding.
• Current hybrid functionals have limitations due to delocalization errors.
• High accuracy in quantum chemistry is essential for reliable predictions.

Purpose of the Study:

• To develop adaptive hybrid functionals for improved electronic structure calculations.
• To address delocalization errors in existing density functional approximations.
• To enhance the accuracy of quantum chemical predictions for various systems.

Main Methods:

• Utilizing data-efficient quantum machine learning models for on-the-fly optimization.
• Developing adaptive Perdew-Burke-Ernzerhof hybrid density functional (aPBE0).
• Implementing a model uncertainty-based constraint for smooth transition to PBE0.

Main Results:

• aPBE0 shows improved energetics, electron densities, and HOMO-LUMO gaps on benchmark datasets (QM9, QM7b, GMTKN55).
• The method effectively addresses the spin gap problem in open-shell systems like carbenes.
• A revised QM9 dataset (revQM9) with enhanced properties is introduced.

Conclusions:

• Adaptive hybrid functionals offer a promising avenue for accurate electronic structure modeling.
• The proposed aPBE0 method provides a robust and generalizable approach.
• The development of improved datasets like revQM9 is vital for advancing computational chemistry.