Leveraging active learning-enhanced machine-learned interatomic potential for efficient infrared spectra prediction.

Area of Science:

• Computational Chemistry
• Spectroscopy
• Materials Science

Background:

• Infrared (IR) spectroscopy offers real-time molecular insights but interpretation relies on computationally intensive simulations.
• Density functional theory (DFT) based ab-initio molecular dynamics (AIMD) are accurate but limited by system size and complexity.
• Efficient prediction of IR spectra is crucial for understanding catalytic processes and reaction intermediates.

Purpose of the Study:

• To develop a novel, efficient framework for predicting IR spectra of small catalytically relevant organic molecules.
• To reduce the computational cost associated with high-fidelity IR spectra simulations.
• To enable high-throughput prediction of IR spectra for larger and more complex catalytic systems.

Main Methods:

• Implementation of an active learning-based framework named PALIRS.
• Training a machine-learned interatomic potential using active learning.
• Utilizing machine learning-assisted molecular dynamics simulations for IR spectra calculation.
• Comparison with ab-initio molecular dynamics and experimental data.

Main Results:

• PALIRS accurately reproduces IR spectra calculated via AIMD at a significantly reduced computational cost.
• The framework shows excellent agreement with experimental data for both IR peak positions and amplitudes.
• PALIRS demonstrates the capability for high-throughput prediction of IR spectra.

Conclusions:

• PALIRS offers a computationally efficient and accurate method for predicting IR spectra.
• This advancement facilitates the exploration of larger, intricate catalytic systems.
• The framework aids in the identification of novel reaction pathways in catalysis.