The Hierarchical Structure of Galactic Haloes: Generalised N-Dimensional Clustering with CluSTAR-ND

TL;DR

CluSTAR-ND is a fast, adaptable hierarchical clustering algorithm for identifying galaxy and halo structures in high-dimensional astrophysical data, outperforming previous methods in speed and robustness.

Contribution

The paper introduces CluSTAR-ND, a novel hierarchical clustering algorithm that is faster, more adaptable, and capable of handling data with any number of features, improving upon Halo-OPTICS.

Findings

CluSTAR-ND is at least 1000 times faster than Halo-OPTICS.

It produces similar robust clustering structures as Halo-OPTICS.

The algorithm's parameters are automatically calibrated for optimal performance.

Abstract

We present CluSTAR-ND, a fast hierarchical galaxy/(sub)halo finder that produces {\bf Clu}stering {\bf S}tructure via {\bf T}ransformative {\bf A}ggregation and {\bf R}ejection in {\bf N}-{\bf D}imensions. It is designed to improve upon Halo-OPTICS -- an algorithm that automatically detects and extracts significant astrophysical clusters from the 3D spatial positions of simulation particles -- by decreasing run-times, possessing the capability for metric adaptivity, and being readily applicable to data with any number of features. We directly compare these algorithms and find that not only does CluSTAR-ND produce a similarly robust clustering structure, it does so in a run-time that is at least $3$ orders of magnitude faster. In optimising CluSTAR-ND's clustering performance, we have also carefully calibrated $4$ of the $7$ CluSTAR-ND parameters which -- unless specified by the user -- will be automatically and optimally chosen based on the input data. We conclude that CluSTAR-ND is a robust astrophysical clustering algorithm that can be leveraged to find stellar satellite groups on large synthetic or observational data sets.