Bridging scales: Modeling suppressed Bondi accretion on black holes and its impact on galaxy growth

TL;DR

The paper demonstrates that magnetically arrested disk (MAD)–driven suppression of Bondi accretion, informed by GRMHD simulations, can be integrated into a population-scale semi-analytic model (Dark Sage) to study SMBH–galaxy co-evolution. By testing four suppression scenarios across hot and cold accretion channels and two AGN feedback modes, the authors find that suppressing sub-Eddington BH growth and rescaling feedback (Case D) reproduces the local SMF and BHMF while preserving rapid black hole growth at high redshift (z>6) required by JWST observations. This work highlights the sensitivity of galaxy assembly to small-scale accretion physics and provides a physically motivated link between horizon-scale simulations and cosmological galaxy formation, suggesting that BH growth regulation must be state-dependent to simultaneously match multiple observables. The results imply that magnetic suppression need not erase early BH growth but must be coupled with calibrated feedback to survive into the present day, offering a pathway to reconcile horizon-scale physics with galaxy-scale phenomena.

Abstract

The accretion and feedback processes governing supermassive black hole (SMBH) growth span an enormous range of spatial scales, from the Event Horizon to the circumgalactic medium. Recent general relativistic magnetohydrodynamic (GRMHD) simulations demonstrate that strong magnetic fields can substantially suppress Bondi accretion by creating magnetically arrested disk (MAD) states, reducing inflow rates by up to two orders of magnitude relative to classical Bondi predictions. We incorporate this magnetic suppression prescription from Cho et al. (2023, 2024) into the Dark Sage semi-analytic model (SAM), which tracks SMBH and galaxy co-evolution within hierarchical merger trees derived from the IllustrisTNG cosmological simulation. Implementing the suppression across different Eddington-ratio regimes, we explore its impact on black hole mass functions (BHMFs), stellar mass functions (SMFs), and AGN luminosity functions. Restricting suppression to sub-Eddington accretors ($f_{\rm Edd} < 3 \times 10^{-3}$) and rescaling AGN feedback efficiencies gives simultaneous agreement with observed $z = 0$ SMFs and BHMFs, as illustrated by Case D in this work. At $z > 6$, super-Eddington growth episodes dominate in the SAM, reproducing JWST-inferred luminous AGN number densities. Our results highlight the critical sensitivity of galaxy assembly to the coupling between small-scale accretion physics and large-scale feedback regulation. Magnetic suppression of hot gas accretion can reconcile low-redshift constraints while preserving the rapid black hole growth required at early cosmic epochs, thereby providing a physically motivated bridge between horizon-scale GRMHD simulations and cosmological galaxy-formation models.

Figures (6)

• Figure 1: The stellar mass function (SMF, Left) and black hole mass function (BHMF, Right) at $z=0$ for fiducial Dark Sage ( Case A, purple), suppressed BH accretion ( Case B, red), sub-Eddington suppressed BH accretion ( Case C, orange), and rescaled AGN suppression ( Case D, green). The grey shaded regions show the observational constrains for the SMF Baldry2008Bernardi2013 and the BHMF PorrasValverde2025MassAssembly in the local Universe. Among our suppression models ( Cases B, C, D), only the rescaling suppression model ( Case D) matches both the SMF and BHMF simultaneously.
• Figure 2: The $M_{\rm BH}-M_{\star}$ (left) and $M_{\rm BH}-M_{\rm bulge}$ (right) relation at z=0 for all models. We include observations from Scott2013THEGALAXIES, ReinesVolonteri2015, Greene2020, and SturmReines2024. When BH growth is entirely suppressed ( Case B), the resulting mean values are systematically reduced, yielding predictions that are completely inconsistent with observations. The other models agree qualitatively with the data.
• Figure 3: The bolometric AGN luminosity function across redshifts $z=0-6$ for the four models under consideration: Case A (fiducial Dark Sage), Case B, Case C, Case D. The comparison includes observational data from Shen2020AGNbol (shown in grey), which span multiple wavebands including rest-frame infrared, B-band, ultraviolet, soft and hard X-ray, along with comprehensive bolometric, dust, and extinction corrections. Shen2020AGNbol gives the Global fit A line, here represented by the dashed olive line. At $z=5-6$, the shaded red region indicates the inferred bolometric luminosity under the assumption that the "little red dots" identified by JWST are AGNs 2024arXiv240610341APacucci-Narayan2024LRD. Additional JWST observations are included from 2023ApJ...959...39H, 2024ApJ...964...39G, 2024arXiv240919205G, 2024AA...691A.145M, 2024ApJ...963..129M, and 2024ApJ...968...38K, each plotted at their corresponding redshift intervals. Most models show better agreement with JWST results at high redshift ($z>3$), transitioning to closer alignment with Shen2020AGNbol at lower redshifts. Among the three tested models, Case D gives the best match to data, performing almost as well as Case A, and Case B performs worst.
• Figure 4: Histogram of the Eddington ratio distributions across redshifts $z=0-6$ for the four models under consideration: Case A (fiducial Dark Sage), Case B, Case C, Case D. All models converge at $z=6$ but show significant differences at lower redshifts.
• Figure 5: The X-ray luminosity in the 0.5-7 keV band as a function of halo mass for the four models under consideration: Case A (fiducial Dark Sage), Case B, Case C, Case D at $z\sim0$. Black circles, squares, and inverted triangles represent observations from Kim2013ApJ, Stanek2006, and ReiprichBohringer2002, respectively. Our models fall within the observational scatter and are consistent with the observed trend. Our models do not extend to halos above $10^{14.5}\,M_\odot$ due to the finite volume of the TNG100-1 simulation box.