Clone of Dissertation Defense: Searching for Be stars using multi-band photometry: Case study of NGC330 using SOAR and S-PLUS data

Due to the complex variety of astronomical classes, each responsible for relevant scientific contributions to Astronomy, multiple observational techniques and object classification methods have been developed throughout history. In the case of Classical Be stars (CBe) – B spectral type objects with high rotation rates (close to the critical limit), non-supergiant and that have or already had a Keplerian disk ejected by the star itself – the pioneering observational technique was spectroscopy, in which the eponymous emission lines became their most notable feature. However, through other techniques such as photometry and polarimetry, different manifestations were detected, such as continuum excess and non-zero linear polarization. One of the still ongoing tasks in the field of CBes is finding accurate ways to classify these objects in a given young stellar population. For example, in the young stellar cluster NGC330, located in the Small Magellanic Cloud, a high fraction of CBes was found through spectroscopic surveys starting in the 1970s. From the 1980s onwards, it was discovered that CBes in a star cluster could be detected using photometry, specifically by means of narrow-band filters centered in Hα. However, there is a strong limitation in the method: only classical Be stars with an active and sufficiently dense disks can be detected with this technique. In order to circumvent this problem, this work uses realistic models of B and CBe stars to produce synthetic clusters and the subsequent synthetic photometry to model NGC330 with the supervised machine learning algorithms k-Nearest Neighbors (k-NN) and Decision Tree (DT). As input data we use photometry from the S-PLUS survey and the SAMI/SOAR imager, both using narrow band Hα filters. The k-NN and DT models trained on the synthetic cluster magnitudes were applied to make unseen predictions, resulting in the classification of the observed sources into three different classes: main-sequence and CBe stars belonging to the cluster and stars not belonging to the main sequence nor members of the cluster, such as foreground objects and/or evolved stars. The obtained result consisted of 44 and 47 stars classified as CBes in the S-PLUS magnitudes set, and a total of 206 and 289 objects classified as CBes in the SOAR set, for k-NN and DT predictions, respectively. In the SOAR ensemble, the DT model estimated a minimum CBe/(B + CBe) fraction of 26%. For both estimators, almost every CBe with Hα emission was classified as a candidate. One of the remarkable results also lies on the prediction of CBe candidates in the "redde" side of the MS, location of inactive CBes or systems with less dense disks (e.g., the beginning of building-up or end of dissipation phases). As the first work to consider every known properties of CBe disks in the formulation of the synthetic cluster, it can be stated that our developed methodology is promising as a preliminary analysis resulted in a higher CBe content than previous estimates, including high resolution spectroscopic surveys, when taking into account the same magnitude depth.

Keywords: Be stars, emission-line stars, Stellar clusters, Young stellar clusters, Machine Learning