How do supernova remnants cool?

II. Machine learning analysis of supernova remnant simulations

¹ I. Physikalisches Institut, Universität zu Köln, Zülpicher Str. 77, 50937 Köln, Germany
² Institute of Astronomy and Astrophysics, Academia Sinica, No. 1, Sec. 4, Roosevelt Rd., Taipei 10617, Taiwan
³ Stony Brook University, 100 Nicolls Rd, Stony Brook, NY 11794, USA
⁴ Center for Data & Simulation science, Universität zu Köln, Albertus-Magnus-Platz, 50923 Köln, Germany

^★ Corresponding authors; smirnova@ph1.uni-koeln.de, makare nko@ph1.uni-koeln.de

Received: 13 June 2024
Accepted: 16 November 2024

Abstract

Context. About 15%-60% of all supernova remnants are estimated to interact with dense molecular clouds. In these high-density environments, radiative losses are significant. The cooling radiation can be observed in forbidden lines at optical wavelengths.

Aims. We aim to determine whether supernovae at different positions within a molecular cloud (with or without magnetic fields) can be distinguished based on their optical emission (e.g. Hα (λ 6563), Hβ (λ 4861), [O III] (λ 5007), [S II] (λ 6717, 6731), and [N II] (λ 6583)) using machine learning (e.g. principle component analysis and k-means clustering).

Methods. We have conducted a statistical analysis of the optical line emission of simulated supernovae interacting with molecular clouds that formed from the multi-phase interstellar medium modelled in the SILCC-Zoom simulations with and without magnetic fields. This work is based on the post-processing of simulations that have been carried out with the 3D (magneto)hydrodynamic code FLASH. Our dataset consists of 22 simulations. The supernovae were placed at a distance of either 25 pc or 50 pc from the molecular cloud’s centre of mass. First, we calculated optical synthetic emission maps (taking into account dust attenuation within the simulation sub-cube) with a post-processing code based on MAPPINGS V cooling tables. Second, we analysed the dataset of synthetic observations using principle component analysis to identify clusters with the k-means algorithm. In addition, we made use of BPT diagrams as a diagnostic of shock-dominated regions.

Results. We find that the presence or absence of magnetic fields has no statistically significant effect on the optical line emission. However, the ambient density distribution at the site of the supernova changes the entire evolution and morphology of the supernova remnant. Due to the different ambient densities in the 25 pc and 50 pc simulations, we are able to distinguish them in a statistically significant manner. Although, optical line attenuation within the supernova remnant can mimic this result depending on the attenuation model that is used. That is why, multi-dimensional analysis of optical emission line ratios in this work does not give extra information about the environmental conditions (ambient density and ambient magnetic field) of supernova remnant.