Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Understanding the Milky Way's Matter Field and Dynamical Accretion History based on AI-GS3 Hunter

Hai-Feng Wang, Guan-Yu Wang, Giovanni Carraro, Yuan-Sen Ting, Thor Tepper-Garcia, Joss Bland-Hawthorn, Jeffrey Carlin, Yang-Ping Luo

TL;DR

Understanding the Milky Way's halo assembly history from stellar kinematics is challenging due to overlapping substructures. The paper introduces GS3 Hunter, a pipeline that fuses Siamese Neural Networks with K-means clustering to detect both cold and hot streams in Gaia EDR3, GALAH DR3, and DESI data. It finds that Gaia-Sausage-Enceladus (GSE) comprises four components (GSE-GSH1–GSE-GSH4), reveals numerous local- and inner-halo substructures, and validates detections with FIRE simulations that recover eight true progenitors among 33 groups. This approach provides a scalable, data-driven framework for constraining the Milky Way's halo assembly and tidal history.

Abstract

We present GS3 Hunter (Galactic-Seismology Substructures and Streams Hunter), a novel deep-learning method that combines Siamese Neural Networks and K-means clustering to identify substructures and streams in stellar kinematic data. Applied to Gaia EDR3 and GALAH DR3, it recovers known groups (e.g., Thamnos, Helmi, GSE, Sequoia) and, with DESI dataset, reveals that GSE consists of four distinct components (GSH-GSH1 through GSE-GSH4), implying a multi-event accretion origin. Tests on LAMOST K-giants recover Sagittarius, Hercules-Aquila, and Virgo Overdensity, while also uncovering new substructures. Validation with FIRE simulations shows good agreement with previous results. GS3 Hunter thus offers a powerful tool to understand the Milky Way's halo assembly and tidal history.

Towards Understanding the Milky Way's Matter Field and Dynamical Accretion History based on AI-GS3 Hunter

TL;DR

Understanding the Milky Way's halo assembly history from stellar kinematics is challenging due to overlapping substructures. The paper introduces GS3 Hunter, a pipeline that fuses Siamese Neural Networks with K-means clustering to detect both cold and hot streams in Gaia EDR3, GALAH DR3, and DESI data. It finds that Gaia-Sausage-Enceladus (GSE) comprises four components (GSE-GSH1–GSE-GSH4), reveals numerous local- and inner-halo substructures, and validates detections with FIRE simulations that recover eight true progenitors among 33 groups. This approach provides a scalable, data-driven framework for constraining the Milky Way's halo assembly and tidal history.

Abstract

We present GS3 Hunter (Galactic-Seismology Substructures and Streams Hunter), a novel deep-learning method that combines Siamese Neural Networks and K-means clustering to identify substructures and streams in stellar kinematic data. Applied to Gaia EDR3 and GALAH DR3, it recovers known groups (e.g., Thamnos, Helmi, GSE, Sequoia) and, with DESI dataset, reveals that GSE consists of four distinct components (GSH-GSH1 through GSE-GSH4), implying a multi-event accretion origin. Tests on LAMOST K-giants recover Sagittarius, Hercules-Aquila, and Virgo Overdensity, while also uncovering new substructures. Validation with FIRE simulations shows good agreement with previous results. GS3 Hunter thus offers a powerful tool to understand the Milky Way's halo assembly and tidal history.

Paper Structure

This paper contains 6 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Data and Methods
  3. Results
  4. Local Halo Results
  5. Inner Halo Results and Simulation results
  6. Summary

Figures (2)

  • Figure 1: The left panel shows the six main groups in the E–L$_z$ plane from 2022AA...665A..57L. The right panel presents our results, with colored labels: solid lines mark our $3\sigma$ substructure regions, while dashed lines show those from 2022AA...665A..57L. Our regions are generally more extended, especially for the Hot Thick Disk. Differences reflect sample coverage and methodology, though several structures overlap.
  • Figure 2: This figure presents the KDE distributions of the four GSE-related structures in chemical abundance space. A colorbar on the right indicates the density levels. In many of the chemical abundance planes, multiple clumps are visible, which may correspond to distinct chemical evolution pathways.