Unmasking the physical information inherent to interstellar spectral line profiles with Machine Learning. I. Application of LTE to HCN and HNC transitions

Abstract: Physical and chemical conditions (kinetic temperature, volume density, molecular composition) of interstellar clouds are inherent in their mm-submm line spectra, making spectral line profiles powerful diagnostics of source conditions. We introduce a novel bottom-up approach employing Machine Learning (ML) algorithms to directly infer physical conditions from line profiles without using radiative transfer equations. We simulated HCN and HNC emission under representative dense molecular cloud and star-forming region conditions across five rotational transitions (J=1-0 to J=5-4) within 30-500 GHz. The generated data cloud was parameterized using line intensities and widths to infer the physical conditions of the analyzed regions. Three ML algorithms were trained, tested, and compared to unravel the excitation conditions of HCN and HNC and their abundance ratio. ML results obtained with two spectral lines, one for each isomer, were compared with a Local Thermodynamic Equilibrium (LTE) analysis for the cold source R CrA IRS 7B, yielding excitation temperatures and relative abundances in agreement with LTE. The optimized pipeline (training, testing, and prediction) can predict interstellar cloud properties from line profile inputs at lower computational cost than traditional methods. This work represents the first mapping of spectral line profiles to physical conditions by charting isomer abundance ratios and excitation temperatures. Our bottom-up approach, based on simulated and semiempirical spectra, offers a new tool to interpret line observations and estimate interstellar conditions using ML methods.