A Bond-Based Machine Learning Model for Molecular Polarizabilities and A Priori Raman Spectra.

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Machine learning (ML) algorithms are increasingly used in molecular simulations, enabling high-accuracy calculations.
• Existing ML algorithms primarily focus on force fields, with fewer options for predicting tensorial properties like electric polarizability.

Purpose of the Study:

• To introduce a novel kernel ridge regression algorithm for machine learning of the polarizability tensor.
• To demonstrate the algorithm's ability to predict tensor components efficiently, comparable to scalar quantities.

Main Methods:

• Developed a kernel ridge regression algorithm based on the bond polarizability model.
• Applied the algorithm to simulate gas phase Raman spectra of biphenyl and malonaldehyde using MACE-OFF23 potential in classical molecular dynamics.

Main Results:

• The developed algorithm efficiently predicts polarizability tensor components.
• Simulated Raman spectra showed excellent agreement with experimental data for biphenyl and malonaldehyde.
• The MACE-OFF23 potential demonstrated high accuracy for these molecular systems.

Conclusions:

• The introduced physics-informed ML algorithm is effective for predicting molecular polarizability tensors.
• The method offers a computationally efficient approach for simulating spectroscopic properties.
• This work highlights the potential of physics-informed ML for developing accurate molecular property prediction tools.