# Dissecting stellar populations with manifold learning I. Validation of the method on a synthetic Milky Way-like galaxy

**Authors:** A. W. Neitzel, T. L. Campante, D. Bossini, A. Miglio  
**arXiv:** [2501.16294](https://arxiv.org/abs/2501.16294)  
**Date:** 2025-01-27  
**Summary page:** https://www.summarizepaper.com/en/paper/2501.16294v1/

## Key points

- Neitzel et al. study identifying and differentiating stellar populations in the Milky Way using manifold learning.
- Stellar populations have variations in chemical, kinematic, and chronological properties due to diverse physical mechanisms.
- Accurately defining these populations is challenging due to overlaps in their characteristics.
- Researchers aim to develop a methodology exploring nonlinear aspects of multidimensional parameters with manifold learning.
- Synthetic stellar samples resembling red-giant stars are generated for validation from Gaia DR3-like catalogs and TESS data.
- Manifold learning reduces the 5-dimensional parameter space into 2-dimensional embeddings for analysis.
- The results show promising capabilities of manifold learning in discerning stellar populations effectively under observational constraints.

## Summary

In their study, Neitzel et al. explore the task of identifying and differentiating stellar populations within the Milky Way galaxy using manifold learning - a powerful unsupervised machine learning technique. Stellar populations exhibit variations in chemical, kinematic, and chronological properties due to diverse physical mechanisms governing their origin and evolution. However, accurately identifying and defining these populations can be challenging due to significant overlaps in their chrono-chemo-kinematic characteristics. The researchers aim to develop a novel methodology that delves into the nonlinear and nonparametric aspects of multidimensional physical parameters defining stellar populations. By leveraging manifold learning, they seek to uncover hidden manifolds within input data without making strong assumptions about the number or nature of stellar populations present. To validate this approach, synthetic stellar samples resembling red-giant stars from Gaia DR3-like catalogs are generated based on FIRE-2 cosmological simulations and asteroseismic data from TESS. Through manifold learning, the team reduces the complex 5-dimensional chrono-chemo-kinematic parameter space into 2-dimensional latent space embeddings. These embeddings are then analyzed to evaluate their fidelity in representing original data and whether they contain meaningful information for distinguishing between different stellar populations. The results indicate promising capabilities of manifold learning in effectively discerning stellar populations under realistic observational constraints. Overall, this research sheds light on the potential of manifold learning as a valuable tool for dissecting complex datasets and revealing underlying structures that can enhance our understanding of the evolutionary history of galaxies like the Milky Way. , , , , .