# An Application of HEP Track Seeding to Astrophysical Data

**Authors:** Mine Gökçen, Maurice Garcia-Sciveres, Xiangyang Ju  
**arXiv:** [2401.06011](https://arxiv.org/abs/2401.06011)  
**Date:** 2024-01-11  
**Summary page:** https://www.summarizepaper.com/en/paper/2401.06011v2/

## Key points

- Application of particle track reconstruction methods from High Energy Physics (HEP) to search for distinct stellar populations in the Milky Way using Gaia EDR3 data set
- Similarities between 3D space points in HEP detectors and positions of stars in coordinate space
- Track reconstruction process involving seeding as the initial step
- Utilization of machine learning tools from FAISS library, specifically k-nearest neighbors (kNN) search for seeding method
- Use of five positional and kinematic parameters from Gaia EDR3 catalog: galactic longitude, latitude, proper motion in right ascension and declination, and parallax
- Focus on first four parameters due to high uncertainties associated with parallax measurements
- Analysis involves subsetting data to identify differences in proper motion as a signature of distinct stellar populations
- Aim to distinguish unique stellar populations within the galactic disk region automatically using variations in proper motion values

## Summary

In this study, we apply methods of particle track reconstruction from High Energy Physics (HEP) to the search for distinct stellar populations in the Milky Way using the Gaia EDR3 data set. The motivation behind this approach stems from the similarities between 3D space points in HEP detectors and the positions of stars in a coordinate space. Just as collections of space points correspond to particle trajectories in HEP, collections of stars from different populations, such as stellar streams, can resemble tracks. The process of track reconstruction involves multiple steps, with seeding being the initial step. In our research, we focus on describing our implementation and results of the seeding step for identifying distinct stellar populations. We utilize machine learning tools from the FAISS library, specifically employing k-nearest neighbors (kNN) search for our seeding method. The Gaia EDR3 catalog provides observations on approximately two billion stars and their properties. We specifically utilize five positional and kinematic parameters – galactic longitude and latitude, proper motion in right ascension and declination, and parallax. Due to the high uncertainties associated with parallax measurements, we primarily rely on the first four parameters in our seeding method. Our analysis involves subsetting the data to identify differences in proper motion as a signature of distinct stellar populations. The underlying assumption is that most stars have a proper motion consistent with the galaxy average, which we classify as background stars. By focusing on variations in proper motion values, we aim to distinguish unique stellar populations within the galactic disk region automatically. Overall, this research showcases an innovative application of HEP track reconstruction techniques to astrophysical data analysis, offering insights into identifying and characterizing distinct stellar populations within the Milky Way using advanced machine learning tools.