Components of star formation in NGC 253 : Non-negative Matrix Factorization Analysis with the ALCHEMI integrated intensity images

TL;DR

This study applies Non-negative Matrix Factorization to ALMA data of NGC 253, identifying molecular components related to star formation processes and demonstrating NMF's advantages over PCA in interpreting complex molecular line data.

Contribution

The paper introduces NMF as an effective method for decomposing molecular line images into physically meaningful components in starburst galaxies.

Findings

NMF identified five key molecular components linked to star formation phenomena.

NMF components correlate with physical conditions like excitation levels and shocks.

NMF outperforms PCA in isolating single physical components in molecular data.

Abstract

It is essential to examine the physical or chemical properties of molecular gas in starburst galaxies to reveal the underlying mechanisms characterizing starbursts. We used non-negative matrix factorization (NMF) to extract individual molecular or physical components involved in the star formation process in NGC\,253. We used images of 148 transitions from 44 different species of the ALMA large program ALCHEMI. Additionally, we included the continuum images at ALMA Bands 3 and 7 from the same dataset. For the five NMF components (NF1--5), we obtained that their distributions correspond to various basic phenomena related to star formation: i) low-density gas extended through the galactic central molecular zone (NF2), ii) shocks (NF3), iii) starburst regions (NF4), and iv) young star-forming regions (NF5). The other component (NF1) is related to excitation; three components obtained by NMF (NF3, 1, and 5) show a strong dependence upon the upper state energies of transitions, and represent low-, intermediate-, and high-excitation, respectively. We also compared our results using principal component analysis (PCA) previously applied to the same dataset. Molecular components extracted from NMF are similar to the ones obtained from PCA. However, NMF is better at extracting components associated with a single physical component, while a single component in PCA usually contains information on multiple physical components. This is especially true for features with weak intensities like emission from outflows. Our results suggest that NMF can be one of promising methods interpreting molecular line survey data, especially in the upcoming era of wide-band receivers.