Radio AGN feedback sustains quiescence only in a minority of massive galaxies

TL;DR

This work addresses whether maintenance-mode radio AGN feedback can sustain quiescence across the population of massive galaxies. It introduces an optical-imaging AI classifier trained with noisy labels to separate radio-feedback-effective (RFE) from ineffective (RFI) galaxies, uncovering that RFE comprises a small, mass-dependent fraction residing in dynamically hot halos with typically little cold gas. By combining MaNGA, xGASS, LOTSS, and weak-lensing data, the study shows RFE halos are 0.6–1 dex more massive than RFI-Q halos at fixed stellar mass, while many RFI-Q galaxies retain substantial HI reservoirs, implying alternative pathways to quiescence and that current models may overestimate the pervasiveness of radio feedback. The results imply radio AGN feedback can sustain quenching over ~hundreds of activity cycles, but only for a minority of massive galaxies, necessitating revisions to galaxy formation models and insights into the long-term impact of AGN feedback on the CGM and gas accretion.

Abstract

Radio active galactic nuclei (AGNs) eject a huge amount of energy into the surrounding medium and are thought to potentially prevent gas cooling and maintain the quiescence of massive galaxies. The short-lived, sporadic, and anisotropic nature of radio activities, coupled with the detection of abundant cold gas around some massive quiescent galaxies, raise questions about the efficiency of radio feedback in massive galaxies. Here we present an innovative method rooted in artificial intelligence to separate galaxies in which radio feedback is effective (RFE), regardless of current radio emission, from those in which radio feedback is ineffective (RFI), according to their optical images. Galaxies categorized as RFE are all dynamically hot, whereas quiescent RFI (RFI-Q) galaxies usually have extended cold-disk components. At given stellar mass, dark matter halos hosting RFE galaxies are between four to ten times more massive than those of RFI-Q galaxies. We find, for the first time, that almost all RFE galaxies have scant cold gas, irrespective of AGN activity. In contrast, many RFI-Q galaxies are surrounded by substantial amounts of condensed atomic gas, indicating a different evolutionary path from RFE galaxies. Our finding provides direct and compelling evidence that a radio AGN has gone through about 300 on-off cycles and that radio feedback can prevent gas cooling over a timescale much longer than that of radio activity. Contrary to general belief, our analysis shows that only a small fraction of massive galaxies are influenced by strong radio AGNs, suggesting that current galaxy formation models need serious revision.

Figures (9)

• Figure 1: AGN-SF diagram and AI pipeline. The observed radio luminosity ($L_{\rm 1.4GHz}$) scaled by the predicted radio luminosity ($L_{\rm SFR,1.4GHz}$) from the star-formation rate(SFR) versus the specific SFR (sSFR). The red dashed contours exhibit Q-LERGs, a representative sample of RFE-on galaxies. The cyan and yellow contours show the RFI and RFE (cRFE) populations, which are the outcome of the AI analysis. The contour lines enclose 10%, 50% and 90% of the galaxies with $\log M_\star/M_\odot>10.5$. The vertical dashed line indicates the demarcation between quiescent and star-forming galaxies. RFE galaxies may move vertically as they enter the duty cycle of AGN activity on a short timescale, while RFI galaxies can move horizontally as their SFRs evolve (usually with a longer timescale). We show the images of randomly selected galaxies in the background. Most star-forming galaxies are blue and exhibit spiral arms; most galaxies with large $L_{\rm 1.4GHz}/L_{\rm SFR,1.4GHz}$ are red and appear to be elliptical galaxies; both elliptical and disk-like galaxies exist in the lower-left section of the diagram. The RFE classifier, which is trained by Q-LERG labels only, can classify RFE and RFI based on SDSS galaxy images. This is achieved by combining both training the model to classify Q-LERG (RFE-on) versus non-Q-LERG (RFE-off + RFI) and deliberately encouraging the model to misclassify RFE-offs into the Q-LERG category through a noise-label learning scheme. Since RFE-off and Q-LERG galaxies share similar optical appearance in the image domain, we successfully generated an RFE classifier that distinguishes RFE (RFE-on + RFE-off) from RFI. See Table \ref{['tab_sample']} for the introduction of the terminologies and samples used in this paper.
• Figure 2: Distributions of $\hat{y}_1$ and $\hat{y}_2$.A, B: the number count histograms of $\hat{y}_1$ and $\hat{y}_2$ for Q-LERGs(orange) and non-Q-LERGs (blue). The dashed lines show the threshold of 0.9. C, D: the 2D histograms of galaxy number. The parameter $\hat{y}_1$ and $\hat{y}_2$ are calculated by our AI technique based on the optical SDSS image with and without data augmentation, respectively. The parameters are used to evaluate the similarity between the images of unclassified galaxies and RFE-on galaxies. Large $\hat{y}$ means that the galaxies are similar to the RFE-on galaxies.
• Figure 3: Optical images for different types of galaxies.A to D: the randomly selected images for RFE-on galaxies in four mass bins. E to H: pRFE galaxies. I to L: cRFE galaxies. M to P: RFI-Q galaxies. Q to T: RFI-SF galaxies. The shown galaxies are selected from those at $z<0.04$ and with $\log M_\star/M_\odot>10.5$. The stellar masses, sSFR, $\hat{y}_1$ and $\hat{y}_2$ are shown in each panel.
• Figure 4: Completeness and purity.A: the lower limit of purity as a function of $M_\star$. The red lines show the results for pRFE-off sample, and the orange lines show the results for cRFE-off sample. The solid lines show the results using all galaxies with $z<0.2$, dotted line for galaxies with $z<0.0906$ and dashed lines for galaxies with $z<0.05$. B: the completeness as a function of $M_\star$. The completeness, estimated using a standard 90/10 train-test split of the data. There is only 34 Q-LERGs (at $z<0.2$) in the lowest $M_\star$ bin, leading to large uncertainties in completeness. Purity refers to the fraction of selected RFE-off galaxies that are observationally confirmed in LOTSS DR2. Error bars represent Poisson 1$\sigma$ uncertainties.
• Figure 5: Distribution of LOTSS DR2 galaxies in L-SFR plane. The background shows a 2D histogram (hexbin) of galaxy number as indicated by the color bar. The solid black line indicates the star-forming main sequence ($L_{\rm{SFR,144MHz}}$-SFR relation, Jin2025), while the dashed red line shows the relation plus 3$\sigma$, where $\sigma=0.2$ dex is the scatter of the relation.