COSMOS-Web: does halo mass alone shape the clustering of star-forming and quiescent galaxies?

Source: arXiv:2606.31978 · Published 2026-06-30 · By Louise Paquereau, Clotilde Laigle, Henry Joy McCracken, Olivier Ilbert, Hollis B. Akins, Rafael C. Arango-Togo et al.

TL;DR

This study investigates whether halo mass alone governs the clustering and star-formation activity of galaxies, or whether secondary halo properties and environment contribute significantly. Using the deep, wide COSMOS-Web survey leveraging JWST imaging, the authors measure the auto- and cross-correlation functions of star-forming (SFGs) and quiescent galaxies (QGs) over a wide redshift range (z = 0.1 to 5). To isolate effects beyond halo mass, they introduce a halo mass matching (HMM) technique using the UniverseMachine model to ensure SFG and QG samples have identical halo mass distributions. Despite halo mass control, QGs remain more strongly clustered than SFGs by about 0.5-1 dex at all redshifts. This excess clustering for QGs is especially pronounced for low stellar masses (log(M*/M☉) ≤ 9.5) at z ≤ 2, within the one-halo term regime, suggesting environmental quenching mechanisms such as ram-pressure stripping or suppressed gas accretion are at work. Cross-correlations reveal one-halo conformity up to z ≈ 2, where low-mass QGs cluster more strongly around massive QGs than star-forming centrals of matched halo mass, implying correlated environmental influences on centrals and satellites or secondary halo properties modulate quenching. Beyond z ≈ 2-5, this environmental quenching and conformity vanish. These results challenge the common assumption in galaxy-halo models that halo mass alone sets galaxy clustering and star-formation activity.

Key findings

• Quiescent galaxies are 0.5 to 1 dex more strongly clustered than star-forming galaxies at fixed stellar mass across all redshifts 0.1 ≤ z ≤ 5, even after halo mass matching.
• At z ≤ 2, the relative clustering excess of QGs versus SFGs increases toward lower stellar masses, with clustering ratios (QG/SFG) reaching 5-13 for mass thresholds and 10-30 for mass bins below log(M*/M☉) ~ 9.5.
• One-halo conformity is detected at z ≤ 2, with low-mass or satellite QGs more strongly clustered around massive or central QGs than around star-forming centrals of matched halo mass, at >3σ significance.
• Environmental quenching effects, inferred from the enhanced clustering of low-mass QGs, vanish at z ≳ 2-5, suggesting environment-driven quenching is mostly significant at lower redshifts.
• Halo mass distributions for SFGs and QGs differ substantially without matching: QGs have median halo masses ~10× higher at fixed stellar mass ≥ 10^9 M☉.
• Post halo mass matching, clustering differences between QGs and SFGs remain but amplitude variations with stellar mass reduce, confirming non-halo-mass environmental effects.
• Cross-correlations split by centrals/satellites confirm strong quiescent conformity signals before HMM that are reduced but remain marginally significant post HMM, underscoring secondary environmental factors.
• Clustering measurements extend to z=5 with JWST data, allowing insight into early galaxy environments previously inaccessible.

Methodology — deep read

1. Threat model and assumptions: The study assumes that halo mass primarily governs galaxy properties but tests whether secondary halo or environmental factors shape galaxy clustering and star-formation activity. The 'adversary', so to speak, is the hypothesis that halo mass alone explains observations. They assume accurate photometric redshifts and galaxy classifications using COSMOS-Web data, a large-area JWST survey. Different populations (SFGs, QGs) are selected in stellar mass and redshift bins.

2. Data provenance: The primary dataset is the COSMOS-Web JWST near-IR imaging survey over 0.54 deg^2, combined with COSMOS2025 multiband photometry, yielding ~80,000 galaxies split into star-forming and quiescent classes over z=0.1-5. Photometric redshifts have median Δz/(1+z) ~ 0.04. Stellar masses and SFRs are estimated with SED modeling codes LePHARE and CIGALE. A COSMOS group catalogue up to z~3.7 identifies centrals and satellites.

3. Novel Halo Mass Matching (HMM) technique: They use the UniverseMachine semi-empirical model, tuned to reproduce observed stellar mass and halo mass distributions, to derive matched halo mass samples of SFGs and QGs. The procedure involves iteratively matching stellar mass distributions in model and data, then adjusting the observed stellar mass distributions to align the halo mass distributions of SFG and QG subsamples, ensuring clustering differences beyond halo mass can be probed. This is repeated 20 times for stability.

4. Clustering measurement: The angular two-point correlation function w(θ) is computed using the Landy & Szalay estimator, with random catalogs matched to survey area/masks. The TreeCorr algorithm is used, with jackknife resampling (20 patches) to estimate covariance. Integral constraint corrections are not applied, limiting accuracy at large scales (θ > 0.02 deg).

5. Sample definitions: Stellar mass completeness limits are applied to ensure unbiased clustering (Pozzetti et al. 2010 formula). Galaxies are split into bins by redshift and stellar mass thresholds or ranges. Star-forming/quiescent classification uses an evolving specific SFR cutoff (sSFR ≤ 0.2 / t_Hubble(z)).

6. Cross-correlation analysis: To probe conformity, cross-correlations are measured between high- and low-mass galaxies and between centrals and satellites (identified from group catalogs). Differences in cross-correlation amplitudes between quiescent and star-forming samples are used to detect one-halo conformity.

7. Evaluation: Clustering amplitudes are fit with double power laws in one-halo and two-halo regimes. Significance of conformity detections is assessed via S/N ratios and chi-squared tests. The impact of halo mass matching is evaluated by comparing results with and without HMM.

8. Reproducibility: The COSMOS2025 catalog and group catalogs referenced are publicly available; the UniverseMachine model is publicly documented, but exact code or frozen weights for this analysis are not explicitly stated. Random seeds and iteration counts for HMM are specified (20 runs). The cosmology is Planck 2020.

Concrete example: For a low-mass, low-z bin (e.g., log M* = 8.5-9.5 at z=0.6-1.0), after applying HMM to equalize halo masses between SFG and QG subsamples, the angular auto-correlation functions w(θ) still show QGs clustering ~5-10× higher than SFGs at one-halo scales (r < 1 Mpc), implying environmental effects beyond halo mass. Cross-correlation with high-mass centrals further shows enhanced clustering of low-mass QGs around QG centrals, a signal interpreted as one-halo conformity.

Technical innovations

• Introduction of a halo mass matching (HMM) method utilizing the UniverseMachine semi-empirical model to construct star-forming and quiescent galaxy samples with identical halo mass distributions, enabling isolation of environmental effects beyond halo mass.
• Application of deep JWST COSMOS-Web imaging to measure galaxy clustering up to unprecedented redshifts (z=5), extending environmental quenching and conformity analysis well beyond prior z<3 studies.
• Joint use of auto- and cross-correlation functions between stellar mass- and star-formation-classified populations, combined with halo mass controls, to robustly detect one-halo conformity effects up to z ~ 2.
• Demonstration that low-mass quiescent galaxies exhibit stronger clustering than higher mass ones at z ≤ 2, opposite to expectations from hierarchical halo mass growth models, indicating environment-driven quenching mechanisms.

Datasets

• COSMOS-Web survey — 0.54 deg2 area, ~80,000 galaxies from z=0.1 to 5 — JWST imaging plus COSMOS multiwavelength data
• COSMOS2025 catalogue — >700,000 galaxies with photometry and derived properties — publicly available at https://cosmos2025.iap.fr/catalog.html
• COSMOS group catalogue — 1678 candidate groups up to z ≈ 3.7 identified with AMICO algorithm

Baselines vs proposed

• Without halo mass matching: QG clustering amplitude exceeds SFG by 0.5 to >1 dex across all redshift bins at fixed stellar mass.
• With halo mass matching: QG clustering remains 0.2 to 1 dex higher than SFG across 0.1 < z < 5, confirming environmental effects.
• Cross-correlation [low-mass QGs × high-mass QGs] versus [low-mass QGs × high-mass SFGs] at z ≤ 1.5 shows >3σ conformity signal; no significant difference for SFG cross-correlations.
• One-halo conformity detection significance drops from >6σ without HMM to ~1σ with HMM for satellite-central samples at 0.6 ≤ z < 1.0.
• QGs with log(M*/M☉) ≤ 9.5 at z ≤ 2 have clustering amplitude ratios to SFGs of up to 10-30 in mass bin analyses.

Limitations

• Halo mass estimates rely on semi-empirical models (UniverseMachine) and indirect group catalogs rather than direct measurements, introducing modeling uncertainties.
• Clustering at large scales (two-halo term, θ > 0.02 deg) may be underestimated due to no integral constraint correction applied.
• Lower number statistics for quiescent galaxies in narrow stellar mass and redshift bins, especially at high redshifts, limits precision and prevents HMM application at z ≥ 2 in some cases.
• Environmental quenching interpretations are indirect, inferred from clustering differences; direct gas and star-formation property measurements would strengthen conclusions.
• Disentangling effects of secondary halo properties, assembly bias, and environment remains challenging; the current methodology cannot fully separate these contributions.
• Spectroscopic redshifts are limited; use of photometric redshifts introduces uncertainty especially in group membership and central/satellite classification.

Open questions / follow-ons

• What specific physical processes drive the enhanced clustering of low-mass quiescent galaxies at z ≤ 2? Can hydrodynamical simulations reproduce these trends?
• To what extent do secondary halo properties (e.g., concentration, assembly history) versus local environment independently influence galaxy quenching and clustering?
• How does conformity evolve beyond z ~ 2, and what mechanisms suppress environmental quenching at early cosmic times?
• Can improved direct measurements of halo masses and gas properties clarify the interplay of mass- and environment-driven quenching?

Why it matters for bot defense

While this paper is firmly rooted in astrophysics and galaxy evolution, it presents a sophisticated approach to disentangling correlated effects in large, noisy datasets by carefully matching confounding distributions (halo mass) to isolate secondary relationships (environmental effects). Bot-defense practitioners can draw an analogy to controlling for dominant confounders (e.g., user device or location) when seeking subtle signals of bot behavior influenced by secondary factors. The halo mass matching method is a notable data modeling technique for reducing confounding in observational data, which could inspire analogous approaches in behavioral correlation studies within security and bot detection contexts. Furthermore, the emphasis on multi-scale correlation analysis (auto- and cross-correlations) and consideration of spatial/environmental clustering may resonate with studies examining clustered automated traffic or coordinated malicious actors. However, direct technical applicability is limited as this work does not address behavioral or cybersecurity data; it instead offers methodological inspiration for carefully controlling confounding effects and interpreting residual clustering signals beyond dominant factors.

Cite

@article{arxiv2606_31978,
  title={ COSMOS-Web: does halo mass alone shape the clustering of star-forming and quiescent galaxies? },
  author={ Louise Paquereau and Clotilde Laigle and Henry Joy McCracken and Olivier Ilbert and Hollis B. Akins and Rafael C. Arango-Togo and Nguyen Binh and Caitlin M. Casey and Yohan Dubois and Maximilien Franco and Ghassem Gozaliasl and Santosh Harish and Michaela Hirschmann and Baptiste Jego and Aidan Kaminsky and Jeyhan S. Kartaltepe and Anton Koekemoer and Damien Le Borgne and Joseph S. W. Lewis and Daizhong Liu and Georgios Magdis and Jed McKinney and Wilfried Mercier and Lauro Moscardini and Thibaud Moutard and Jason D. Rhodes and Brant E. Robertson and Sogol Sanjaripour and Marko Shuntov and Greta Toni and Maxime Trebitsch and Laurence Tresse },
  journal={arXiv preprint arXiv:2606.31978},
  year={ 2026 },
  url={https://arxiv.org/abs/2606.31978}
}

Read the full paper

• arXiv:2606.31978 (abstract)
• arXiv:2606.31978 (PDF)