Detection of Quadruple Structure Near the ASCC 32 Region via Machine Learning Methods

TL;DR

This study addresses whether the ASCC 32 region hosts multiple coeval stellar structures by applying Gaia DR3 data with a two-stage unsupervised approach (DBSCAN followed by Gaussian Mixture Models). The method identifies a filament containing ASCC 32, OC 0395, and HSC 1865, then resolves four distinct groups—ASCC 32-1, ASCC 32-2, OC 0395, and HSC 1865—sharing a common CMD and radial velocity distribution, but differing in spatial distribution. Compared with prior works, the DG workflow yields more members and reveals two ASCC 32 substructures, while suggesting OC 0401 and HSC 1894 may be substructures rather than independent clusters. These results demonstrate a robust framework for untangling complex stellar group morphologies in large astrometric surveys and provide improved membership catalogs for the ASCC 32 region.

Abstract

Multiple structures within stellar groups are an intriguing subject for theoretical and observational studies of stellar formation. With the accuracy and completeness of data from Gaia Data Release 3, we now have new opportunities to detect reliable members of stellar groups across a larger field of view than in previous studies. In this work, using machine learning methods and high-accuracy data, we investigate the possibility of detecting multiple structures within 500 arcmin of ASCC 32. We first applied DBSCAN to proper motion and parallax, as multiple structures tend to share similar values for these parameters. Next, we applied GMM to position, proper motion, and parallax for the members detected by DBSCAN. This approach allowed us to identify a filamentary structure among the DBSCAN-detected members. This structure contains all stellar groups previously identified in this region. Subsequently, based on the BIC score, we applied GMM to this filamentary structure. Since multiple structures exhibit distinct positional distributions, GMM was able to effectively separate all groups within the filament. Our methods successfully identified ASCC 32, OC 0395, and HSC 1865 within a 500 arcmin radius. Additionally, we found two distinct substructures within ASCC 32. These four groups exhibit a single main-sequence distribution in the CMD, with proper motion values within three times the standard deviation and slightly differing parallax values, despite having distinct spatial structures. Furthermore, these four groups share the same radial velocity distribution. We provide documentation demonstrating the formation of these stellar groups as a multiple structure, with improved membership identification compared to previous studies.

Figures (12)

• Figure 1: The DBSCAN algorithm selected candidate members based on Eps$=0.08$ and MinPts$=250$. The top-left panel illustrates the candidates' distribution in the position diagram. The top-right panel displays the distribution in the proper motion diagram, the bottom-left panel presents the parallax diagram, and the bottom-right panel shows the CMD. As observed, DBSCAN distinguishes two clusters, neither of which exhibit features typically indicative of star clusters, such as dense regions in proper motion or approximated color-magnitude diagram distribution.
• Figure 2: The DBSCAN algorithm selected candidate members based on $\mathrm{Eps}=0.08$ and $\mathrm{MinPts}=280$. With these parameters, DBSCAN identified three clusters, one of which contains our cluster members (C 3). The top-left panel illustrates the candidates' distribution in the position diagram, while the top-right panel presents their distribution in the proper motion diagram. The bottom-left panel displays the parallax diagram, and the bottom-right panel shows the CMD. As observed, the candidate cluster members are expected to lie within C 3, based on the range of proper motion and parallax values. For C 3, the CMD clearly reveals two distinct cluster distributions and indicates contamination from field stars.
• Figure 3: KDE for DBSCAN selected candidate members. This figure reveals three dense regions in the coordinates of ASCC 32, OC 0395 and HSC 18.
• Figure 4: Applying the GMM algorithm to cluster members detected by DG methods with a hyperparameter value of 3, one filamentary structure emerges that contains ASCC 32, OC 0395, and HSC 1865. The top-left panel shows the spatial distribution, while the top-right panel illustrates the proper motion distribution. The bottom-left panel presents the parallax diagram, and the bottom-right panel displays the CMD, which demonstrates a clear main sequence without contamination.
• Figure 5: The BIC score determines the optimal detection parameters for GMM. At a value of 4, GMM successfully identifies reliable members for each cluster. Our analysis confirms that this value is appropriate.