Ethereal AI: Infrared Spectra of Polycyclic Aromatic Hydrocarbons with Machine Learning DFT Scaling Factors.

Area of Science:

• Computational Chemistry
• Astrochemistry
• Spectroscopy

Background:

• Polycyclic aromatic hydrocarbons (PAHs) are crucial in astrochemistry, with their infrared (IR) signals providing key observational data.
• Interpreting these IR signals often requires computational chemistry for reference spectra.
• Current methods use density functional theory (DFT) with scaling factors to correct predicted IR frequencies, but these have limitations.

Purpose of the Study:

• To develop a novel machine learning (ML) approach for generating more accurate IR frequency scaling factors for PAHs.
• To improve the deconvolution of complex IR spectra from space-based observations of PAHs.

Main Methods:

• A machine learning model was developed to individually adjust predicted IR frequencies.
• The model utilized computed frequencies, intensities, reduced masses, and force constants as input features.
• The ML-derived scaling factors were compared against traditional scaling factor methods.

Main Results:

• The ML approach achieved a mean absolute error (MAE) of 5 cm⁻¹ and a maximum error of 13 cm⁻¹.
• This represents a significant improvement over the current practice, which yielded an MAE of 10 cm⁻¹ and a maximum error of 23 cm⁻¹.
• The ML model demonstrated enhanced accuracy in predicting IR spectra.

Conclusions:

• Machine learning offers a promising method for refining DFT-calculated IR spectra.
• This approach can enable higher quality predictions for larger and more complex molecular systems.
• The improved accuracy has significant implications for astrochemistry and the study of interstellar PAHs.