Interstellar infrared bands: a machine learning model calculates 144,111 spectra of aromatic molecules to interpret JWST data

An open library covers 48,037 polycyclic aromatic hydrocarbons across three charge states, with accuracy approaching that of quantum chemistry at a fraction of the cost.

The aromatic infrared bands observed in the interstellar medium are attributed to polycyclic aromatic hydrocarbons (PAHs), but their interpretation depends on libraries of reference spectra that cover only a small part of the expected chemical diversity. Three researchers present a massive database to fill this gap for small and medium-sized PAHs.

Their method, called ML-DH (double harmonic by automatic learning), combines force fields with neural networks with electronic dipolar moment models: it approaches the accuracy of quantum chemistry calculations at a much lower cost, and has been validated against independent DFT calculations and reference bases. Result: 144,111 infrared spectra for 48,037 benzenoid structures, in neutral, cationic and anionic states, each with an estimated uncertainty. Spectres, optimized geometries and code are published in open access to help model the observations of the James Webb space telescope and future infrared missions.