GraphShed: A Data-derived Graph-based waterShed Group Finder
Abstract
This study introduces GraphShed, a data-derived group-finder that applies top-down watershed segmentation to Voronoi-induced graphs, identifying galaxy systems directly from the density field via a GraphShed-derived linking length, with no density thresholds or tunable parameters. A GraphShed galaxy group catalog compared to a Friends-of-Friends (FoF) catalog built from the IllustrisTNG100-1 simulation with a box size of ∼ 75 Mpc/h yielding $1067$ and $987$ systems, respectively. The GraphShed-derived linking length adapts to the global clustering power of the sample, varying by ∼ 18 % relative to FoF across different sample selections; thus, catalog differences originate in the watershed segmentation. Cross-catalog comparison shows that ∼ 58 % of FoF systems are identically reproduced by GraphShed, ∼ 5 % are completely removed, and the remainder are partially shaved and/or split into multiple systems. While the M200 distributions of the two catalogs are statistically consistent, other structural properties including, R200, sphericity, compactness, spin, and centroid shift, differ significantly in at least some richness ranges. The unweighted two-point correlation function of GraphShed systems exhibits a higher amplitude on small scales, r ≲ 1 Mpc/h, and agrees with FoF for r≳ 5 Mpc/h. A velocity-based classification reveals that GraphShed resolves more interacting pairs than FoF, demonstrating its ability to distinguish nearby overdense structures that position-only methods merge into single systems. These results demonstrate that GraphShed preserves cosmological statistics while providing a more resolved detection of galaxy systems and their dynamical interactions.
Related
- A new group of low-spin 50M⊙–70M⊙ black holes and the high pair-instability mass cutoff
- A MUSE study of stellar and ionized-gas kinematics in central group galaxies
- Variable stars in Local Group galaxies. VIII. A new look into the dSph galaxy Canes Venatici I
- SIGNN++: Robust Graph Neural Star Identification on Earth
- Graph Learning with Multiplicative Aggregation for Solar Filament Fragment Association Modeling in Full-disk Hα Images
BibTeX
@misc{ghafour2026graphshed,
title = {{GraphShed: A Data-derived Graph-based waterShed Group Finder}},
author = {P. Ghafour and S. Ansarifard and M. H. Jalali Kanafi and S. M. S. Movahed},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.03790},
eprint = {2606.03790},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.03790},
}
RIS
TY - UNPB
TI - GraphShed: A Data-derived Graph-based waterShed Group Finder
AU - P. Ghafour
AU - S. Ansarifard
AU - M. H. Jalali Kanafi
AU - S. M. S. Movahed
PY - 2026
DA - 2026/09/17
PB - arXiv
DO - 10.48550/arxiv.2606.03790
UR - https://arxiv.org/abs/2606.03790
AB - This study introduces GraphShed, a data-derived group-finder that applies top-down watershed segmentation to Voronoi-induced graphs, identifying galaxy systems directly from the density field via a GraphShed-derived linking length, with no density thresholds or tunable parameters. A GraphShed galaxy group catalog compared to a Friends-of-Friends (FoF) catalog built from the IllustrisTNG100-1 simulation with a box size of ∼ 75 Mpc/h yielding $1067$ and $987$ systems, respectively. The GraphShed-derived linking length adapts to the global clustering power of the sample, varying by ∼ 18 % relative to FoF across different sample selections; thus, catalog differences originate in the watershed segmentation. Cross-catalog comparison shows that ∼ 58 % of FoF systems are identically reproduced by GraphShed, ∼ 5 % are completely removed, and the remainder are partially shaved and/or split into multiple systems. While the M_200 distributions of the two catalogs are statistically consistent, other structural properties including, R_200, sphericity, compactness, spin, and centroid shift, differ significantly in at least some richness ranges. The unweighted two-point correlation function of GraphShed systems exhibits a higher amplitude on small scales, r ≲ 1 Mpc/h, and agrees with FoF for r≳ 5 Mpc/h. A velocity-based classification reveals that GraphShed resolves more interacting pairs than FoF, demonstrating its ability to distinguish nearby overdense structures that position-only methods merge into single systems. These results demonstrate that GraphShed preserves cosmological statistics while providing a more resolved detection of galaxy systems and their dynamical interactions.
ER -