Noncovalent Quantum Machine Learning Corrections to Density Functionals

Area of Science:

• Computational chemistry
• Quantum mechanics
• Machine learning

Background:

• Density functional theory (DFT) approximations often contain systematic errors.
• Accurate modeling of noncovalent interactions is crucial for many chemical systems.
• Existing dispersion corrections have limitations in describing complex interactions.

Purpose of the Study:

• To introduce noncovalent quantum machine learning (QML) corrections for density functionals.
• To recover missing nonlocal and nonadditive physical effects in DFT.
• To improve the accuracy of predicting molecular interactions and properties.

Main Methods:

• Developed QML models to learn and correct errors in density functionals.
• Applied corrections to six physically motivated density functionals.
• Trained models on sufficient data to capture complex interaction patterns.

Main Results:

• QML corrections successfully recovered missing physical effects.
• Accurate predictions of dissociation curves were achieved.
• Improved description of molecular two- and three-body interactions, especially in water clusters.
• Provided insights into density functional errors at an atomic resolution.

Conclusions:

• QML corrections offer a flexible and accurate approach to improve DFT.
• The method effectively captures various noncovalent interactions, including hydrogen bonding.
• This work advances the development of more accurate density functionals for computational chemistry.