Efficient Composite Infrared Spectroscopy: Combining the Double-Harmonic Approximation with Machine Learning...

Area of Science:

• Computational Chemistry
• Molecular Spectroscopy

Background:

• Vibrational spectroscopy, particularly infrared (IR) spectroscopy, is crucial for molecular characterization.
• Computational methods are increasingly vital for investigating molecular materials and predicting spectral properties.

Purpose of the Study:

• To assess the predictive accuracy and computational efficiency of gas-phase IR spectra calculations.
• To establish a standard protocol for efficient IR spectra prediction using modern computational techniques.

Main Methods:

• Utilized a composite approach based on the double-harmonic approximation for IR spectra prediction.
• Employed harmonic vibrational frequencies and squared derivatives of the molecular dipole moment.
• Systematically tested various methods including semiempirical quantum mechanics (xTB), charge equilibrium models, and machine learning potentials (MACE-OFF23).

Main Results:

• Evaluated the accuracy and efficiency of combining semiempirical quantum mechanical and machine learning potentials for IR spectra prediction.
• Focused on the MACE-OFF23 machine learning potential to overcome limitations of conventional low-cost methods.
• Assessed a diverse dataset of organic molecules to identify suitable computational protocols.

Conclusions:

• The study provides a framework for efficient and reliable computational prediction of IR spectra.
• Aims to facilitate rapid identification of unknown compounds and advance automated high-throughput analytical workflows in chemistry.