Accurate and Interpretable Dipole Interaction Model-Based Machine Learning for Molecular Polarizability.

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Machine Learning

Background:

• Accurate prediction of molecular polarizabilities is crucial for understanding atomic and molecular interactions.
• First-principles calculations of molecular polarizabilities are computationally expensive.
• Existing physical and machine learning models struggle with accurate local chemical environment descriptions or require extensive training data.

Purpose of the Study:

• To develop a computationally efficient and accurate method for predicting molecular polarizability tensors.
• To overcome the limitations of existing models in describing local chemical environments and training data requirements.
• To create physically interpretable and transferable atomic polarizabilities.

Main Methods:

• A hybrid model combining a physically inspired dipole interaction model with a neural network approach.
• Precise description of the local chemical environment.
• Ensuring natural fulfillment of rotational covariance.

Main Results:

• The hybrid model accurately predicts molecular polarizability tensors.
• Significantly reduced the number of required training samples.
• Generated physically interpretable and transferable atomic polarizabilities applicable to unseen molecules.

Conclusions:

• The developed hybrid model offers an accurate and efficient solution for molecular polarizability prediction.
• The method's ability to provide interpretable and transferable atomic polarizabilities enhances its practical utility.
• Potential applications include spectroscopic simulations and the development of polarizable force fields.