Machine learning potentials accurately reproduce vibrational dynamics in complex environments.

Area of Science:

• Computational Chemistry
• Spectroscopy
• Machine Learning

Background:

• Vibrational spectroscopy offers detailed molecular insights but requires accurate simulation models.
• Machine learning interatomic potentials (MLIPs) combine high accuracy with computational efficiency.

Purpose of the Study:

• To benchmark the Universal Model of Atoms (UMA), a novel MLIP, for predicting vibrational spectroscopy observables.
• To assess UMA's performance against established methods for molecular simulations.

Main Methods:

• Utilized the Universal Model of Atoms (UMA) for molecular dynamics simulations.
• Computed infrared absorption spectra and frequency fluctuations.
• Compared UMA predictions with empirical frequency maps and GFN2-xTB.

Main Results:

• UMA accurately predicted experimental vibrational spectra and frequency fluctuations for an ester carbonyl.
• UMA's accuracy was comparable to traditional empirical and semiempirical methods.
• UMA demonstrated broader generality and computational efficiency.

Conclusions:

• The Universal Model of Atoms (UMA) is a viable and efficient tool for predicting vibrational spectroscopy.
• MLIPs like UMA offer a promising direction for transferable and accurate molecular simulations.