Infrared Spectra at Coupled Cluster Accuracy from Neural Network Representations

Area of Science:

• Computational Chemistry
• Spectroscopy
• Machine Learning

Background:

• Infrared spectroscopy reveals molecular structure, dynamics, and reactions.
• Accurate prediction of infrared spectra from first principles is crucial.
• High-accuracy methods like coupled cluster theory are limited for large systems and anharmonic spectra.

Purpose of the Study:

• To develop a method for predicting infrared spectra with high accuracy for large molecular systems.
• To overcome limitations of traditional methods in computational spectroscopy.
• To enable the study of complex intra- and intermolecular couplings.

Main Methods:

• Utilized neural network representations for energies, forces, and dipole moments.
• Employed machine learning to achieve "gold standard" coupled cluster accuracy.
• Applied molecular dynamics simulations with machine learning potentials.

Main Results:

• Successfully predicted infrared spectra for protonated water clusters up to the hexamer size.
• Demonstrated "gold standard" coupled cluster accuracy using neural network potentials.
• Showcased the ability to compute finite-temperature infrared spectra for systems beyond explicit coupled cluster calculations.

Conclusions:

• Machine learning significantly expands the limits of accuracy, speed, and system size in theoretical spectroscopy.
• The developed methodology allows for the prediction of vibrational spectra for large, complex systems.
• Opens new avenues for understanding molecular couplings and predicting spectroscopic properties.