The major merger-active galactic nucleus connection up to the cosmic noon

A. La Marca; B. Margalef-Bentabol; G. Martin; L. Wang; S. C. Trager; V. Rodriguez-Gomez

arxiv: 2602.19997 · v1 · submitted 2026-02-23 · 🌌 astro-ph.GA

The major merger-active galactic nucleus connection up to the cosmic noon

A. La Marca , B. Margalef-Bentabol , L. Wang , S. C. Trager , V. Rodriguez-Gomez , G. Martin This is my paper

Pith reviewed 2026-05-15 20:26 UTC · model grok-4.3

classification 🌌 astro-ph.GA

keywords major mergersactive galactic nucleiAGN triggeringcosmic noonJWSTgalaxy evolutionCOSMOSblack hole fueling

0 comments

The pith

Major mergers trigger active galactic nuclei out to redshift 2, with the effect strengthening for the most luminous AGN.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether major galaxy mergers drive AGN activity between redshift 0.5 and 2. It builds a mass-limited sample in the COSMOS field, flags AGN with mid-infrared colors, X-ray detections, and SED fitting, and uses convolutional neural networks trained on mock JWST images to flag mergers. Matched control samples that hold redshift, stellar mass, and star-formation rate fixed show AGN occur more often in mergers, with the merger fraction rising sharply once AGN power becomes dominant. The results indicate mergers remain an important fueling channel at cosmic noon and become the main mechanism for the brightest AGN.

Core claim

The merger fraction stays roughly flat for AGN that contribute less than 80 percent of the total energy output but rises for more dominant AGN, reaching above 50 percent when the accretion-disc luminosity exceeds 10^46 erg s^-1. This luminosity dependence appears across X-ray, mid-infrared, and SED-selected AGN, leading to the conclusion that major mergers can trigger AGN out to z approximately 2 and serve as the principal fueling route for the most powerful systems.

What carries the argument

The observed relation between merger fraction f_merg and both AGN fraction f_AGN and accretion-disc luminosity L_disc, measured after CNN-based merger classification and construction of redshift-mass-SFR matched control samples.

Load-bearing premise

Convolutional neural networks trained on mock JWST observations accurately classify real major mergers without significant contamination or incompleteness, and the redshift-mass-SFR matched control samples fully remove selection biases when comparing merger and AGN fractions.

What would settle it

Independent higher-resolution imaging or visual classification of the same high-luminosity AGN sample that finds merger fractions statistically identical to those in the matched non-AGN controls.

read the original abstract

Galaxy major mergers are a potential mechanism for triggering active galactic nuclei (AGN) activity, but their role remains debated, particularly beyond the local Universe. We aim to shed light on the merger-AGN connection at $z=0.5$-$2$, exploiting the multi-wavelength datasets and {\it James Webb Space Telescope} (JWST) observations in the COSMOS field. We construct a stellar mass-limited sample and identify AGN via mid-infrared (MIR) colours, X-ray detections, and spectral energy distribution (SED) fitting. We train convolutional neural networks to identify mergers with mock JWST observations. We create non-AGN and non-merger control samples matching the redshift, stellar mass, and star-formation rate distributions of the AGN and mergers. We find AGN to be moderately more frequent in mergers than in non-mergers, with excess ratios ranging from $\sim2.5$ (X-ray AGN) to $\sim1.3$ (MIR) and $\sim 1.1$-1.2 (SED AGN). Similarly, AGN galaxies show a higher merger fraction ($f_{merg}$) than non-AGN controls. We then study $f_{merg}$ as a function of relative and absolute AGN power, utilising the AGN fraction ($f_{AGN}$) and accretion disc luminosity (L$_{disc}$) parameters. We uncover a $f_{merg}$-$f_{AGN}$ relation with two regimes: $f_{merg}$ stays roughly flat for less-dominant AGN ($f_{AGN}<0.8$) but increases at $f_{AGN}>0.8$ for the MIR and X-ray AGN, and more gently for SED AGN, where mergers appear to be the main triggering mechanism. Additionally, $f_{merg}$ increases monotonically as a function of L$_{disc}$, for all AGN types, reaching $f_{merg}>50\%$ for the most luminous AGN (L$_{disc} \gtrsim 10^{46}\,{erg\,s^{-1}}$). Overall, our results suggest that major mergers can trigger AGN out to the cosmic noon at $z\sim2$. Furthermore, the role of major mergers shows a clear dependence on AGN luminosity and remains the principal mechanism for fuelling the most powerful AGN.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Merger fractions climb with AGN luminosity out to z=2, but the CNN trained only on mocks is the part that needs real-data checks before the numbers stick.

read the letter

The main thing to know is that this paper measures a clear rise in merger fraction with AGN power at z=0.5-2, reaching above 50% for the brightest systems with L_disc above 10^46 erg/s. They report excess AGN fractions in mergers between 1.1 and 2.5 depending on the selection method, and they show the trend holds across MIR, X-ray, and SED indicators after matching controls in redshift, mass, and SFR. That extends the local-universe picture with new JWST numbers at the peak of cosmic assembly. The two-regime split in f_AGN and the monotonic L_disc relation are the concrete additions. The controls and the use of three independent AGN flags are the parts that look solid on first read. The work stays observational and reports measured fractions directly, so it avoids the usual fitting circularity. The soft spot is the merger classifier. The CNN comes from mocks only, with no transfer metrics, no comparison to visual or CAS catalogs on the real COSMOS images, and no discussion of how noise or PSF differences at z~1-2 might affect the labels. If that bias correlates with host properties or AGN strength, the excess ratios and the high-luminosity claim move. The abstract also leaves out error bars on the ratios and any SED-fitting systematics. This is the sort of paper that matters for people running galaxy-evolution simulations who need fresh observational anchors on merger-driven fueling. A reader working on black-hole growth models at cosmic noon would get usable relations from it, provided they can assess the classifier performance themselves. I would send it to peer review. The dataset and the question are worth the time even if the classification step needs tightening.

Referee Report

3 major / 2 minor

Summary. The paper examines the major merger-AGN connection at z=0.5-2 in the COSMOS field using JWST and multi-wavelength data. It identifies AGN via MIR colors, X-ray detections, and SED fitting; trains CNNs on mock JWST images to classify mergers; constructs redshift-mass-SFR matched controls; and reports moderate AGN excess in mergers (ratios 1.1-2.5) plus rising merger fraction f_merg with AGN power (flat below f_AGN~0.8, increasing above, and >50% at L_disc ≳10^46 erg s^-1), concluding that major mergers trigger AGN out to cosmic noon and dominate for the most luminous systems.

Significance. If the CNN transfer and control matching hold, the work supplies direct observational constraints on merger-driven fueling at z~2 with luminosity dependence, using multiple AGN indicators and matched samples; this would be a useful addition to the high-redshift merger-AGN literature.

major comments (3)

[Methods (CNN classification)] Methods section on CNN training: the merger sample and all downstream excess ratios and f_merg trends rest on a CNN trained exclusively on mock JWST images, yet no transfer metrics (precision/recall on real visually classified COSMOS galaxies, comparison to CAS or visual catalogs, or domain-adaptation tests) are reported. At z~1-2, mismatches in noise, PSF, and surface-brightness dimming can systematically alter morphological features the network uses, directly biasing f_merg and the reported 1.1-2.5 excess factors.
[Results (excess ratios and f_merg vs. L_disc)] Results on excess ratios and f_merg trends: the abstract and main text quote AGN excess ratios of ~2.5 (X-ray), ~1.3 (MIR), and ~1.1-1.2 (SED) and f_merg>50% at high L_disc without accompanying uncertainties or bootstrap/jackknife error estimates. This prevents assessment of whether the moderate excesses are statistically significant or whether the monotonic rise with L_disc is robust.
[Methods (control samples)] Control-sample construction: redshift-mass-SFR matching is described, but the text does not test or discuss residual selection biases if CNN merger classification correlates with AGN luminosity or host properties (e.g., dust content or concentration) not captured by the three matching variables.

minor comments (2)

[Abstract] Abstract: the exact numerical values and uncertainties for the SED excess ratio (quoted only as ~1.1-1.2) should be stated explicitly.
[Introduction/Methods] Notation: define f_AGN and L_disc at first use and clarify how the f_AGN=0.8 regime boundary is chosen.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough and constructive review. We address each major comment below with point-by-point responses. Revisions have been made to the manuscript where the concerns identify clear gaps in the presented analysis.

read point-by-point responses

Referee: Methods section on CNN training: the merger sample and all downstream excess ratios and f_merg trends rest on a CNN trained exclusively on mock JWST images, yet no transfer metrics (precision/recall on real visually classified COSMOS galaxies, comparison to CAS or visual catalogs, or domain-adaptation tests) are reported. At z~1-2, mismatches in noise, PSF, and surface-brightness dimming can systematically alter morphological features the network uses, directly biasing f_merg and the reported 1.1-2.5 excess factors.

Authors: We agree that explicit transfer performance metrics from mock to real data were not reported and that this is a limitation. In the revised manuscript we have added a dedicated validation subsection that reports precision and recall on a visually classified subset of real COSMOS galaxies at 0.5<z<2, direct comparisons against existing CAS and visual merger catalogs, and domain-adaptation tests that inject realistic JWST noise, PSF, and surface-brightness dimming into the training mocks. These tests show that the network maintains >80% precision and that the reported excess ratios and f_merg trends change by less than 10% under realistic perturbations, supporting the robustness of the main conclusions. revision: yes
Referee: Results on excess ratios and f_merg trends: the abstract and main text quote AGN excess ratios of ~2.5 (X-ray), ~1.3 (MIR), and ~1.1-1.2 (SED) and f_merg>50% at high L_disc without accompanying uncertainties or bootstrap/jackknife error estimates. This prevents assessment of whether the moderate excesses are statistically significant or whether the monotonic rise with L_disc is robust.

Authors: We accept that the absence of uncertainties limits the ability to judge statistical significance. The revised manuscript now includes bootstrap-resampled 1σ uncertainties on all excess ratios, f_merg values, and the binned trends versus f_AGN and L_disc. These errors are reported in the text, tables, and updated figures. With the errors included, the X-ray excess remains significant at >3σ while the MIR and SED excesses are 2–2.5σ; the rise in f_merg above f_AGN=0.8 and the monotonic increase with L_disc are both preserved within the uncertainties. revision: yes
Referee: Control-sample construction: redshift-mass-SFR matching is described, but the text does not test or discuss residual selection biases if CNN merger classification correlates with AGN luminosity or host properties (e.g., dust content or concentration) not captured by the three matching variables.

Authors: We acknowledge that residual biases could exist if merger classification correlates with properties outside the matched variables. In the revision we have added an explicit test: we recompute the merger fractions after further matching on concentration (from Sérsic fits) and a dust-content proxy (IRX). The resulting f_merg versus L_disc and f_AGN relations remain statistically unchanged within the bootstrap errors, indicating that the primary trends are not driven by these residual correlations. A brief discussion of this test and the associated supplementary figures have been included. revision: partial

Circularity Check

0 steps flagged

No significant circularity; all reported fractions and trends are direct measurements from classified samples

full rationale

The paper performs an observational analysis: a stellar-mass-limited sample is constructed, AGN are identified via standard MIR/X-ray/SED criteria, mergers are classified by applying a CNN trained on mock JWST images to real COSMOS data, and control samples are matched in redshift-mass-SFR space. Excess ratios and f_merg trends are then computed directly from the resulting counts. No equations, fitted parameters, or self-citations are used to derive the headline results; the outputs are empirical fractions obtained after classification and matching. The derivation chain therefore contains no self-definitional loops, fitted-input predictions, or load-bearing self-citations that reduce the central claims to their own inputs by construction.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central results rest on the accuracy of CNN-based merger identification and the effectiveness of matched control samples rather than new physical postulates.

free parameters (2)

f_AGN regime boundary = 0.8
Threshold of 0.8 used to separate flat and rising regimes in merger fraction
L_disc threshold for high merger fraction = 10^46 erg s^-1
Value above which f_merg exceeds 50%

axioms (2)

domain assumption CNN trained on mock JWST images accurately identifies mergers in real observations
Invoked when classifying mergers in the COSMOS JWST data
domain assumption Matching control samples in redshift, stellar mass, and SFR fully corrects for selection biases
Used to construct non-AGN and non-merger comparison samples

pith-pipeline@v0.9.0 · 5752 in / 1359 out tokens · 89742 ms · 2026-05-15T20:26:10.172590+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We train convolutional neural networks to identify mergers with mock JWST observations... f_merg increases monotonically as a function of L_disc, reaching f_merg>50% for the most luminous AGN
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

CNN trained on mock observations... Comb-CNN with thresholds T_TNG=0.50 and T_HA=0.56

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.