Little Red Dots on FIRE: The Ability of Bursty Galaxies to Host an Abundant Population of High-Redshift AGN

TL;DR

The paper investigates whether bursty, high-redshift galaxies can host the abundant AGN population inferred from JWST’s little red dots (LRDs) by applying two black hole accretion prescriptions—gravitational torque-driven accretion and a simple free-fall model—to FIRE-2 zoom-in simulations and constructing AGN bolometric luminosity functions via halo-mass weighting. It finds that modest mean accretion (BHs accreting a small fraction of their central gas supply within $R\approx100$ pc per free-fall time) can reproduce the bright end of the LRD luminosity function, but fiducial models overpredict the faint end; including subgrid variability softens this tension yet often remains too high at $L_{\rm Bol}\sim10^{43-44}$ erg s$^{-1}$. A plausible scenario to match both the LRD bolometric and UV luminosity functions requires super-Eddington accretion onto $M_{\rm BH}\gtrsim 2\times10^5\,M_\odot$ black holes hosted in $M_\star\gtrsim 2\times10^7\,M_\odot$ galaxies, with bolometric luminosities capped at $L_{\rm Edd}$ and UV emission dominated by host-galaxy stellar light. This demonstrates that bursty high-$z$ galaxies can sustain an abundant AGN population and supports super-Eddington accretion as a viable component of LRD phenomenology, while highlighting the need for self-consistent BH feedback modeling in future work.

Abstract

The James Webb Space Telescope has unveiled an abundant population of potential active galactic nuclei (AGN) at high redshift ($z\gtrsim4$) known as little red dots (LRDs), which are likely hosted in relatively low-mass galaxies. However, previous theoretical models have highlighted the difficulty in continuously feeding massive black holes in the central regions of bursty, high-redshift galaxies because of repeated gas evacuation by stellar feedback. We analyze galaxies in high-redshift FIRE-2 simulations to understand whether they are capable of hosting the observed abundant population of high-redshift AGN. We use a gravitational torque-driven accretion (GTDA) model and a simple free-fall accretion model to derive black hole accretion rates and construct predicted AGN bolometric luminosity functions for $z=5-7$. The GTDA model and the free-fall model with black holes accreting $\lesssim 1$ percent of their central gas supply ($<100 \rm \ pc$) per free-fall time predict AGN abundances that are more than sufficient to explain the most recent LRD observations. The fiducial models, in fact, overpredict the number of low-luminosity AGN as compared with observations. We explore possible resolutions of this tension. A plausible, though likely not unique, scenario for alleviating the AGN overpredictions and which also provides a good match to the host-galaxy UV luminosity distribution suggests that LRDs are super Eddington-accreting, Eddington luminosity-limited, $M_{\rm BH}\gtrsim 2\times10^5 \ \rm M_\odot$ black holes residing in $M_\star \gtrsim 2\times10^7 \ \rm M_\odot$ galaxies.

Figures (8)

• Figure 1: Predicted black hole accretion rates using the gravitational torque-driven accretion (GTDA) model (solid lines) and the free-fall accretion model with $\varepsilon_{\rm ff}=0.001$ (dashed lines) for three example host galaxies in our sample with $M_{\star}(z=5) \sim 10^6$ (purple), $10^8$ (green), and $10^{10} \ (\rm orange) \ \rm M_\odot$ for $z=5-7$. The horizontal axis shows cosmic time. Bursty stellar feedback in FIRE-2 (and in many other high-resolution simulations) disrupts the central gas supply of galaxies and prevents black holes from continuously accreting at high redshift. This effect potentially drives tension between theoretical models and the observed abundance of AGN at high redshift.
• Figure 2: The AGN bolometric luminosity function predicted at $z \sim 6$ using the GTDA model (blue with stars) and the simple free-fall accretion model (dark red with circles) with $\bar{\varepsilon}_{\rm ff} = 1$ (solid), $0.1$ (dashed), $0.01$ (dot-dashed), and $0.001$ (dotted). Curves show the median luminosity function predicted from our 1000 bootstrapped samples; shaded regions enclose the 16th–84th percentile range. Here, our fiducial aperture of $R=100 \ \rm pc$ is used and we introduce our fiducial amount of variance to the normalization/accretion efficiency ($\sigma_{\log(\epsilon_T)} = \sigma_{\log(\varepsilon_{\rm ff})}=0.5$). The inferred accretion rates (and therefore the predicted AGN bolometric luminosity function) for the GTDA model and the simple free-fall accretion model with $\bar{\varepsilon}_{\rm ff}\sim0.001$ are in close agreement. Green circles with error bars show the LRD bolometric luminosity function inferred from observations by Greene_2025 for $z=4-6$ and black crosses show the pre-JWST luminosity function based on X-ray and UV observations from Shen_2020 (their "Global Fit A") for $z=4-6$. Our simple free-fall accretion model with an accretion efficiency of less than one percent is able to reproduce the abundance of objects in the highest observed luminosity bin ($L_{\rm Bol} \sim 10^{45} \ \rm erg/s$) presented by Greene_2025 but overpredicts the number of AGN in lower-luminosity bins.
• Figure 3: The predicted $z \sim 6$ AGN bolometric luminosity function from FIRE-2 simulations using the GTDA model (left panel) and the simple free-fall accretion model (right panel). To model unresolved time variability, we apply log-normally distributed values of the normalization ($\bar{\epsilon}_T=5$) for the GTDA model and the accretion efficiency ($\bar{\varepsilon}_{\rm ff}$=0.01) for the simple free-fall model with $\sigma_{\log(\epsilon)} = 0$ (dark blue), $0.25$ (blue-gray), $0.5$ (gray), $0.75$ (gold), and $1$ (yellow). Shaded regions represent the range between the 16th and 84th percentile luminosity functions predicted from our bootstrapped samples. Here, our fiducial aperture of $R = 100 \ \rm pc$ is used. Green circles with error bars show the LRD bolometric luminosity function inferred from observations by Greene_2025 for $z=4-6$ which goes through the pre-JWST luminosity function based on X-ray and UV observations from Shen_2020 (their "Global Fit A") represented by black crosses. Introducing scatter to the normalizations/accretion efficiencies upscatters some low-luminosity sources into higher-luminosity bins, thereby making the slope of the predicted luminosity function shallower and in better agreement with observations in the higher luminosity bins. In all cases, however, our models overpredict the number of faint AGN as compared with observations.
• Figure 4: The predicted $z \sim 6$ AGN bolometric luminosity function (top) and UVLF for galaxies hosting an AGN in the LRD luminosity bins ($L_{\rm Bol} \sim 10^{43-45} \ \rm erg/s$; bottom) using the GTDA model (left) and the simple free-fall accretion model (right). Galaxies only contribute to the UVLFs plotted in the bottom panels if they are predicted (by the model plotted) to host an AGN in the range $L_{\rm Bol} = 10^{42.5-45.5} \ \rm erg/s$. Since not all AGN are LRDs, we make different modifications to our fiducial model in an attempt to better match observations of the LRD bolometric luminosity function from Greene_2025 (red circles) and the $z=4.5-6.5$ LRD UVLF from Kokorev_2024 (red diamonds). In these models, the UV luminosity is powered by star formation in the host galaxy only. We present the cases where galaxies of any mass are permitted to host AGN (dark purple) and where the contributions of AGN hosted within galaxies with $M_\star<10^{6} \ \rm M_{\odot}$ (blue), $M_\star<2\times10^{7} \ \rm M_{\odot}$ (green), and $M_\star<10^{8} \ \rm M_{\odot}$ (yellow) are excluded. With these host galaxy stellar mass cuts, we simultaneously explore the cases where the luminosities of AGN are capped at their predicted $L_{\rm Edd}$ (dashed) and where we only include super-Eddington accretors whose luminosities are still capped at $L_{\rm Edd}$ (dotted). The model where we include only super-Eddington-accreting, Eddington luminosity-limited black holes hosted within $M_\star>2\times10^{7} \ \rm M_{\odot}$ galaxies (our "Plausible LRD Scenario"; dotted green line with shaded region representing the range between the 16th and 84th percentile luminosity functions from our bootstrapped samples) is an example of a plausible scenario that comes close to simultaneously reproducing the observed LRD bolometric and LRD UV luminosity functions.
• Figure 5: The stellar mass distribution of FIRE-2 galaxies hosting AGN in luminosity bins centered at $L_{\rm Bol} \sim 10^{43} \ (\rm purple), \ 10^{44} \ (\rm blue), \ 10^{45} \ (\rm green), \ and \ 10^{46} \ (\rm yellow) \ \rm erg/s$ for our fiducial models (including unmodified $L_{\rm Bol}$ values for AGN in all galaxies; top) and our "Plausible LRD Scenario" (as in Figure \ref{['fig:LowL_Tension']}, including only super Eddington-accreting, Eddington luminosity-limited black holes hosted within $M_\star>2\times10^{7} \ \rm M_{\odot}$ galaxies; bottom) for the GTDA model (left) and the free-fall accretion model (right) at $z \sim 6$. While more luminous AGN are typically hosted in more massive galaxies, our fiducial models predict AGN in each luminosity bin to have a wide range ($3-4$ orders of magnitude) in host stellar mass. Our "Plausible LRD Scenario", however, predicts a much narrower range in host stellar masses of LRDs ($\sim 1$ order of magnitude) in a given luminosity bin, although the range is still wide ($\sim 3$ orders of magnitude)) when integrated over the entire LRD population. In either case, the majority of AGN in the luminosity range of observed LRDs are hosted by low-mass ($\log(M_\star/\rm M_\odot) \lesssim 9$) galaxies whose central black holes accrete intermittently (see Figure \ref{['fig:Accretion_Time_Series']} for examples).