Quenching pathways in the green valley at low redshift: confronting SDSS AGN hosts with IllustrisTNG and EAGLE

TL;DR

This study compares low-redshift SDSS DR7 pure AGN hosts to green-valley analogue centrals in IllustrisTNG-100 and EAGLE, using dataset-specific percentile-based definitions of the green valley to minimize cross-dataset systematics. It finds that IllustrisTNG's kinetic feedback drives nearly binary quenching, yielding green-valley analogues with extremely low sSFR and substantial mass offsets from SDSS AGN hosts. In contrast, EAGLE's thermal, stochastic feedback produces a broad, SDSS-like distribution of specific star formation rates through the green valley, aligning more closely with observations. The forward-modeling and robust statistical tests support the interpretation that real galaxies traverse the green valley more gradually than TNG predicts, highlighting differences in quenching pathways across feedback prescriptions and underscoring the value of openly released data and reproducible workflows.

Abstract

We compare low-redshift ($z<0.1$) BPT-selected pure optical AGN hosts in SDSS DR7 to colour-selected "green-valley" analogue central galaxies in IllustrisTNG100 and EAGLE Ref-L0100N1504. To reduce cross-dataset systematics, we define the green valley internally using $(g-r)$ percentiles: for galaxies with $\log_{10}(M_\star/M_\odot)>10$, we select the 75th-95th percentiles (SDSS observed-frame fibre colours; simulations rest-frame synthetic colours within 30 kpc). SDSS hosts are linked to the MPA-JHU catalogue for stellar masses and aperture-corrected total SFRs. TNG green-valley centrals are almost entirely quenched, with a sharp pile-up at the imposed SFR floor and median $\log_{10}\mathrm{sSFR}\simeq-14.85$ ($\sim$3.5 dex below SDSS). EAGLE instead produces a broad, continuous distribution with median $\log_{10}\mathrm{sSFR}\simeq-11.71$ and substantial overlap with SDSS, robust to varying the lower percentile between 60 and 90. At fixed mass, TNG yields higher green-valley occupancy fractions (reaching $\gtrsim60$ per cent near $M_\star\sim10^{11}M_\odot$) than EAGLE (20-40 per cent). A simple forward model of nebular line ratios places EAGLE analogues across the star-forming and composite loci in the BPT plane, while TNG analogues concentrate in a LINER-like, low-sSFR regime. We infer that TNG's kinetic mode drives an efficient, near-binary shutdown of star formation, whereas EAGLE's stochastic thermal feedback supports a slower decline more consistent with local AGN hosts. All catalogues and analysis scripts are publicly released.

Figures (12)

• Figure 1: Schematic workflow of the analysis pipeline applied to SDSS, TNG100, and EAGLE. The three parallel streams illustrate how observational SDSS DR7 galaxies ($z<0.1$) and simulated galaxies from TNG100 ($z=0$) and EAGLE ($z=0$) are processed, from catalogue ingestion through sample selection and mock-observation/visualisation steps to the extraction of host-galaxy properties. The forward-modelled BPT step is used only as an illustrative mapping for qualitative phase-space comparisons (Section \ref{['subsec:methods_bpt']}); simulated galaxies are not BPT-selected as AGN. All streams converge in the right-hand block, where the statistical comparisons (Sections \ref{['sec:results']} and \ref{['sec:discussion']}) are performed.
• Figure 2: Host-galaxy demographics of SDSS AGN compared to green-valley analogues in IllustrisTNG and EAGLE. Left:$(g-r)$ colour versus stellar mass for SDSS DR7 galaxies (blue contours) and BPT-selected AGN hosts (orange points); for SDSS the colour is the observed-frame fibre value, whereas for the simulations it is rest-frame synthetic photometry within 30 kpc. The shaded bands indicate the 75th--95th percentile "green-valley" windows for TNG (light green) and EAGLE (hatched), defined separately for massive galaxies with $\log_{10}M_\star/\mathrm{M_\odot}>10$. Overplotted symbols show TNG (blue circles) and EAGLE (green diamonds) centrals lying in their respective green-valley windows. The thick black curve shows the running median colour of SDSS AGN hosts. Right: Normalised KDEs of $\log_{10}\mathrm{sSFR}$ for SDSS AGN (orange), TNG green-valley centrals (blue dashed), and EAGLE green-valley centrals (green dotted). Vertical lines mark median values $\tilde{s}\mathrm{SFR}$ with 68 per cent bootstrap uncertainties. TNG analogues pile up at the imposed low-sSFR pile-up corresponding to formally zero instantaneous SFR, whereas EAGLE analogues retain a broad distribution much closer to SDSS.
• Figure 3: Specific star-formation rate versus stellar mass. Left: SDSS BPT-selected AGN hosts at $z<0.1$ shown as orange points (down-sampled for clarity) over a hexagonal density map with overlaid KDE contours. Right: TNG100 (blue circles) and EAGLE (green diamonds) green-valley centrals selected via simulation-specific colour percentiles. Contours show the core of each distribution. SDSS hosts occupy a broad, partially quenched locus. TNG analogues collapse onto a low-sSFR pile-up at $\log_{10}\mathrm{sSFR}\approx -15.5$, whereas EAGLE analogues span a wide range of intermediate sSFRs that more closely resembles SDSS.
• Figure 4: Cumulative distributions of stellar mass (left) and sSFR (right) for massive ($\log_{10}M_\star>10$) SDSS AGN hosts (orange), TNG green-valley centrals (blue dashed), and EAGLE green-valley centrals (green dotted). All quantities are measured within 30 kpc apertures in the simulations and derived from aperture-corrected totals in SDSS. TNG green-valley analogues are both more massive and far more quenched than SDSS AGN hosts, while EAGLE analogues are closer in both mass and sSFR distribution.
• Figure 5: Green-valley occupancy fraction as a function of stellar mass. The fraction of galaxies in the percentile-defined green valley is shown for SDSS AGN hosts (orange squares), a matched SDSS star-forming control sample (blue circles), all TNG100 centrals (grey triangles), and all EAGLE centrals (green diamonds). Error bars are binomial $1\sigma$ uncertainties; bins with fewer than 10 galaxies are omitted. AGN hosts show an enhanced green-valley fraction relative to the control sample, peaking near $\log_{10}M_\star\simeq 11$, while TNG produces an even higher occupancy fraction dominated by already-quenched systems. EAGLE yields a lower and more gradually varying occupancy.