Mid-infrared diagnostics for identifying main-sequence galaxies in the local Universe

¹ Physics Department, and Institute of Theoretical and Computational Physics, University of Crete, 71003 Heraklion, Greece
² Institute of Astrophysics, Foundation for Research and Technology-Hellas, 71110 Heraklion, Greece
³ Center for Astrophysics | Harvard & Smithsonian, 60 Garden St., Cambridge, MA 02138, USA

^⋆ Corresponding author: cdaoutis@physics.uoc.gr

Received: 29 March 2025
Accepted: 17 September 2025

Abstract

Context. A galaxy’s mid-infrared spectrum captures a significant amount of information about its internal conditions such as the radiation field strength and its star formation activity. It is also intricately connected to dust characteristics, and it contains spectral lines that serve as crucial indicators for galaxy activity diagnostics. Thus, characterizing galaxies’ mid-infrared spectra is highly constraining of their nature.

Aims. This project describes a diagnostic tool for identifying main-sequence star-forming galaxies in the local Universe using infrared dust emission features that are characteristic of galaxy activity.

Methods. A physically motivated sample of mock galaxy spectra has been generated to simulate the infrared emission of star-forming galaxies in the local Universe. Using this sample, we developed a diagnostic tool for identifying main-sequence star-forming galaxies with machine learning methods. Custom photometric bands were defined to target key dust emission features, including polycyclic aromatic hydrocarbons (PAHs) and the dust continuum. Specifically, three bands were selected to capture PAH emission peaks at 6.2 μm, 7.7 and 8.6 μm, and 11.3 μm, along with one band estimating the strength of the radiation field illuminating the dust. This diagnostic was subsequently applied to observed galaxies to evaluate its effectiveness in real-world applications.

Results. Our diagnostic achieves high performance scores, by accurately identifying 90.9% of main-sequence star-forming galaxies in a sample of observed galaxies. Additionally, it demonstrates low contamination, with only 16.2% of active galactic nuclei (AGN) galaxies being misidentified as star-forming based on our test sample.

Conclusions. It is possible to combine observational studies and stellar population synthesis frameworks to generate physically motivated simulated samples of star-forming galaxies that exhibit similar spectral properties as their observed counterparts. By strategically positioning custom photometric bands on the mid-infrared spectrum to target specific dust emission features, our diagnostic can extract valuable information without the need to measure specific emission lines. Although PAHs are sensitive indicators of star formation and interstellar medium radiation hardness, PAH emission alone is insufficient for identifying main-sequence star-forming galaxies. Finally, we developed a physically-motivated spectral library of main sequence star-forming galaxies spanning from ultraviolet to far-infrared wavelengths.