The Deeper Studies of IC 2488 and IC 2714

Jin-Sheng Qiu, Qingfeng Zhu, Xuzhi Li, Xiao-Hui Xu, Hang ZhengThe Astrophysical Journal 2024 · DOI

Abstract

In this paper, we use two methods, an algorithm of the density-based spatial clustering of applications with noise combined with the Gaussian mixture model, and the maximum likelihood method, to study two open clusters: IC 2488 and IC 2714 with Gaia Data Release 3 data. We compare the two methods in terms of cluster star number, isochrone fitting, cluster mass, and radius. The comparison shows that the selections of cluster stars by the two methods are consistent. The parameters obtained by the two methods are very close. But there are slight differences between the two methods for faint stars in IC 2714. The maximum likelihood method selects more faint stars outside of the main-sequence stripe. We conclude that the maximum likelihood method is more likely affected by field stars.

Introduction

Open Clusters (OCs) are groups of stars formed together from the same molecular cloud, sharing the same age, chemical abundance, and distance. This makes them invaluable cosmic laboratories for studying the structure and evolution of the Milky Way. However, to study an OC, astronomers must first isolate true cluster members from foreground or background "field" stars. Historically, this membership determination has been difficult, especially for faint stars with higher measurement uncertainties. This paper addresses the challenge by comparing two distinct membership determination algorithms on two specific open clusters: IC 2488 and IC 2714, using high-precision Gaia Data Release 3 (DR3) data.

Methodology

The authors compare two distinct algorithms to calculate the probability that a star belongs to a cluster:

  1. The DG Algorithm (DBSCAN + GMM): A two-step machine learning approach. It first uses DBSCAN (Density-Based Spatial Clustering of Applications with Noise) for a preliminary classification based on local stellar density. These results are then refined using a Gaussian Mixture Model (GMM) that calculates precise membership probabilities using proper motions, positions, and parallaxes.
  2. Maximum Likelihood Method (MLM): A traditional statistical approach that calculates membership probability based on the Gaussian distributions of cluster stars versus field stars, relying primarily on stellar proper motions.

The study utilizes Gaia DR3 five-parameter astrometry and photometric data, focusing on sources within a 20-arcminute radius of the cluster centers and filtering out stars fainter than G = 20 mag.

Results and Comparative Analysis

Both algorithms identify bright cluster members with high consistency. However, the MLM struggles significantly with faint stars (G > 18 mag). Because faint stars have higher proper motion errors, the MLM mistakenly flags many background field stars as cluster members.

Astrometric errors vs. G magnitude for IC 2488 and IC 2714.

As shown in the figure above, the measurement errors for parallax ($\sigma_\omega$) and proper motion ($\sigma_\mu$) rise sharply at higher G magnitudes. This increase in noise makes the MLM's reliance on proper motions a liability for faint-star classification. For IC 2714, the DG algorithm found 1,097 members while the MLM found 1,419, with only a 58% overlap. The extra stars selected by the MLM scatter widely off the main-sequence band on a Color-Magnitude Diagram, confirming they are likely field stars.

Cluster Parameters

Despite the differences in membership counts, the derived physical parameters for the clusters remained relatively consistent between methods:

  • IC 2488: Age $\approx$ 112 Myr ($\log(\text{age}) \approx 8.05$), distance $\approx$ 1.36 kpc.
  • IC 2714: Age $\approx$ 316 Myr ($\log(\text{age}) \approx 8.50$), distance $\approx$ 1.35 kpc.

Both clusters exhibit clear mass segregation, where massive stars are concentrated near the cluster centers, indicating that both systems have undergone significant dynamic relaxation.

Conclusion

The DG algorithm's success stems from its multi-parameter approach, incorporating parallax and spatial data alongside proper motions. Furthermore, it requires fewer hardcoded prior assumptions, making it highly suitable for automated, large-scale cluster discovery. While the MLM remains a standard tool, its susceptibility to field star contamination at faint magnitudes highlights the need for robust machine learning pipelines in the era of high-precision astrometry. Future work should cross-validate these findings across multiple theoretical stellar evolution models to ensure the stability of derived ages and masses.

Summary updated

Abstract

Abstract In this paper, we use two methods, an algorithm of the density-based spatial clustering of applications with noise combined with the Gaussian mixture model, and the maximum likelihood method, to study two open clusters: IC 2488 and IC 2714 with Gaia Data Release 3 data. We compare the two methods in terms of cluster star number, isochrone fitting, cluster mass, and radius. The comparison shows that the selections of cluster stars by the two methods are consistent. The parameters obtained by the two methods are very close. But there are slight differences between the two methods for faint stars in IC 2714. The maximum likelihood method selects more faint stars outside of the main-sequence stripe. We conclude that the maximum likelihood method is more likely affected by field stars.
BibTeX
@article{qiu2024deeper,
  author = {Jin-Sheng Qiu and Qingfeng Zhu and Xuzhi Li and Xiao-Hui Xu and Hang Zheng},
  title = {{The Deeper Studies of IC 2488 and IC 2714}},
  journal = {The Astrophysical Journal},
  year = {2024},
  doi = {10.3847/1538-4357/ad3f1a},
}
Figure