[Paper Review] Effects of multi-channel AGN feedback in FIRE cosmological simulations of massive galaxies

Lindsey Byrne, Claude‐André Faucher‐Giguère+7 more|arXiv (Cornell University)|Oct 24, 2023

Galaxies: Formation, Evolution, Phenomena4 citations

TL;DR

This study presents the first FIRE-3 cosmological zoom-in simulations with multi-channel AGN feedback—radiative, mechanical, and cosmic ray—applied to massive galaxies (10¹²–10¹³ M☉). It demonstrates that such feedback successfully quenches star formation, reproduces key observed scaling relations (e.g., M*–Mhalo, M_BH–σ), and produces realistic structural properties, including size-mass and Faber-Jackson relations, whereas simulations without AGN feedback overproduce stars and fail to match observations.

ABSTRACT

Feedback from supermassive black holes is believed to be a critical driver of the observed color bimodality of galaxies above the Milky Way mass scale. AGN feedback has been modeled in many galaxy formation simulations, but most implementations have involved simplified prescriptions or a coarse-grained interstellar medium (ISM). We present the first set of FIRE-3 cosmological zoom-in simulations with AGN feedback evolved to $z\sim0$, examining the impact of AGN feedback on a set of galaxies with halos in the mass range $10^{12}-10^{13} M_{\odot}$. These simulations combine detailed stellar and ISM physics with multi-channel AGN feedback including radiative feedback, mechanical outflows, and in some simulations, cosmic rays (CRs). We find that massive (>L*) galaxies in these simulations can match local scaling relations including the stellar mass-halo mass relation and the $M_{ m BH}$-$σ$ relation; in the stronger model with CRs, they also match the size-mass relation and the Faber-Jackson relation. Many of the massive galaxies in the simulations with AGN feedback have quenched star formation and elliptical morphologies, in qualitative agreement with observations. In contrast, simulations at the massive end without AGN feedback produce galaxies that are too massive and form stars too rapidly, are order-of-magnitude too compact, and have velocity dispersions well above Faber-Jackson. Despite these successes, the AGN models analyzed do not produce uniformly realistic galaxies when the feedback parameters are held constant: while the stronger model produces the most realistic massive galaxies, it tends to over-quench the lower-mass galaxies. This indicates that further refinements of the AGN modeling are needed.

Motivation & Objective

• To investigate whether multi-channel AGN feedback in high-resolution cosmological simulations can reproduce the quenching of massive galaxies observed in the local universe.
• To test if including radiative, mechanical, and cosmic ray feedback channels improves agreement with observed scaling relations such as M*–Mhalo and M_BH–σ.
• To assess whether AGN feedback can produce realistic galaxy structural properties, including half-mass radii, velocity dispersions, and central surface densities.
• To compare the impact of different feedback models (e.g., BH-only vs. BH+cosmic rays) on galaxy evolution and quenching efficiency.
• To evaluate whether mass-invariant feedback prescriptions can reproduce differential quenching across the massive galaxy population, despite high spatial and mass resolution.

Proposed method

• Conducting cosmological zoom-in simulations using the FIRE-3 code with adaptive mesh refinement and particle-based hydrodynamics at high resolution (stellar and gas mass resolution ≤ 100 M☉).
• Implementing multi-channel AGN feedback: radiative heating, mechanical outflows, and cosmic ray transport and diffusion, all self-consistently coupled to black hole accretion.
• Using subgrid models for AGN feedback that depend on black hole accretion rate and Eddington ratio, with feedback efficiency tuned to match observed AGN luminosities.
• Evolved simulations to z ≈ 0 to compare final galaxy properties with local observations, including stellar mass, halo mass, velocity dispersion, and size-mass relations.
• Analyzing star formation histories, orbital circularities, and structural parameters (e.g., half-mass radius, central surface density) to assess quenching and morphological realism.
• Comparing results across models: NoBH (no black hole), BH (mechanical + radiative), and BH+CR (with cosmic rays) to isolate feedback channel contributions.

Figure 1: Edge-on mock Hubble images for two example Milky Way-mass simulated galaxies (top two rows) and two example m13 massive galaxies (bottom two rows) at $z\sim 0$ . For each set of initial conditions, we compare a model with no BHs or CRs (left) to a model with BHs but no CRs (middle) and to

Experimental results

Research questions

• RQ1 Can multi-channel AGN feedback in FIRE-3 simulations reproduce the quenching of massive galaxies observed in the local universe?
• RQ2 To what extent does including cosmic rays in AGN feedback improve agreement with observed scaling relations such as M*–Mhalo and M_BH–σ?
• RQ3 How do different AGN feedback models (BH-only vs. BH+cosmic rays) affect the structural properties of massive galaxies, including size, velocity dispersion, and central surface density?
• RQ4 Why do simulations without AGN feedback fail to reproduce observed massive galaxy properties, even at high resolution?
• RQ5 Do mass-invariant feedback prescriptions produce realistic quenching across the full halo mass range (10¹²–10¹³ M☉), or is a mass-dependent feedback efficiency required?

Key findings

• Simulations with AGN feedback successfully quench star formation in massive galaxies, producing quenched, elliptical-like systems consistent with observations, while simulations without AGN feedback produce overmassive, star-forming galaxies.
• Galaxies with AGN feedback match the observed stellar mass-halo mass relation, with stellar masses within a factor of ∼2–3 of observed values, whereas simulations without AGN feedback overpredict stellar masses by a factor of 3–5.
• The inclusion of AGN feedback reduces stellar velocity dispersions to within observed Faber-Jackson relation bounds, while simulations without feedback produce velocity dispersions that are too high by an order of magnitude.
• AGN feedback increases galaxy half-mass radii and reduces central surface densities, bringing them into agreement with the observed size-mass relation, whereas simulations without feedback produce galaxies that are too compact and dense.
• The BH+cosmic ray feedback model produces lower late-time star formation rates than the BH-only model, indicating that cosmic rays enhance quenching efficiency.
• Despite success in reproducing key scaling relations, the feedback models do not uniformly produce realistic galaxies across the full mass range when parameters are held constant, suggesting the need for mass-dependent feedback prescriptions.

Figure 2: Stellar mass-halo mass (SMHM) relation (left) and $M_{\rm BH}-\sigma$ relation (right) for $z\sim 0$ . We compare the NoBH (black, unfilled), BH (green), and BH+CR (purple) models. Left panel: The SMHM relation from (Behroozi et al., 2019 ) and its intrinsic scatter. Right panel: The obser