Unmasking the physical information inherent to interstellar spectral line profiles with machine learning

I. Application of LTE to HCN and HNC transitions

¹ Dept. Ciencias Integradas, Facultad de Ciencias Experimentales, Centro de Estudios Avanzados en Física, Matemática y Computación, Unidad Asociada GIFMAN, CSIC-UHU, Universidad de Huelva, Spain
² Lyon College, Batesville, AR, USA
³ Planetário Juan Bernardino Marques Barrio, Instituto de Estudos Socioambientais, Universidade Federal de Goiás, Brazil
⁴ Centro de Estudios Avanzados en Física, Matemática y Computación, Universidad de Huelva, Spain
⁵ Andalusian Research Institute in Data Science and Computational Intelligence, Universidad de Huelva, Spain
⁶ Dept. Tecnologías de la Información, Escuela Técnica Superior de Ingeniería, Centro de Estudios Avanzados en Física, Matemática y Computación, Universidad de Huelva, Spain
⁷ SRON Netherlands Institute for Space Research & Kapteyn Astronomical Institute, University of Groningen, 9747 AD Groningen, The Netherlands
⁸ Instituto Universitario Carlos I de Física teórica y computacional, Universidad de Granada, Spain

^★ Corresponding authors: edgar.mendoza@dci.uhu.es; miguel.carvajal@dfa.uhu.es

Received: 27 September 2024
Accepted: 25 April 2025

Abstract

Context. Physical and chemical properties, such as kinetic temperature, volume density, and molecular composition of interstellar clouds are inherent in their line spectra at submillimeter wavelengths. Therefore, the spectral line profiles could be used to estimate the physical conditions of a given source.

Aims. We present a new bottom-up approach, based on machine learning (ML) algorithms, to extract the physical conditions in a straightforward way from the line profiles without using radiative transfer equations.

Methods. We simulated, for the typical physical conditions of dense molecular clouds and star-forming regions, the emission in spectral lines of the two isomers HCN and HNC, from J = 1–0 to J = 5–4 between 30 and 500 GHz, which are commonly observed in dense molecular clouds and star forming regions. The generated data cloud distribution has been parametrised using the line intensities and widths to enable a new way to analyse the spectral line profiles and to infer the physical conditions of the region. The line profile parameters have been charted to the HNC/HCN ratio and the excitation temperature of the molecule(s). Three ML algorithms have been trained, tested, and compared aiming to unravel the excitation conditions of HCN and HNC and their abundance ratio.

Results. Machine learning results obtained with two spectral lines, one for each isomer HCN and HNC, have been compared with the local thermodynamic equilibrium (LTE) analysis for the cold source R CrA IRS 7B. The estimate of the excitation temperature and of the abundance ratio, in this case considering the two spectral lines, is in agreement with our LTE analysis. The complete optimisation procedure of the algorithms (training, testing, and prediction of the target quantities) have the potential to predict interstellar cloud properties from line profile inputs at lower computational cost than before.

Conclusions. It is the first time that the spectral line profiles are mapped according to the physical conditions charting the ratio of two isomers and the excitation temperature of the molecules. In addition, a bottom-up approach starting from a set of simulated spectral data at different physical conditions is proposed to interpret line observations of interstellar regions and to estimate their physical conditions. This new approach presents the potential relevance to unravel hidden interstellar conditions with the use of ML methods.