Roles of Supernova and Active Galactic Nucleus Feedback in Shaping the Baryonic Content in a Wide Range of Dark Matter Halo Masses

TL;DR

The paper tackles the challenge of reproducing realistic baryon content in halos by enhancing a semi-analytic model (FEGA25) with an AGN-driven hot-gas ejection channel. It compares two implementations, AGNeject1 (continuous, early-on) and AGNeject2 (delayed, late-on), to quantify how SN and AGN feedback eject gas beyond the virial radius while preserving stellar and cold gas components. Calibrated against DM-only simulations and wide redshift SMFs, the model reproduces the stellar-to-halo mass relation up to $z=3$ and reveals that SN feedback dominates in low-mass halos, whereas AGN feedback becomes crucial at intermediate-to-high halo masses, with AGNeject2 naturally producing a late-time baryon cavity consistent with some hydrodynamic simulations. The study highlights the necessity of AGN-driven hot-gas ejection for accurately modeling the baryon cycle and its implications for cluster cosmology and emerging observational probes, while acknowledging current data limitations in decisively distinguishing between the two ejection schemes.

Abstract

We build upon FEGA25 (Contini et al 2025), a previously introduced semi-analytic model for galaxy formation and evolution, focusing on its enhanced treatment of supernova and active galactic nucleus feedback mechanisms. In addition to the traditional AGN feedback modes, negative (suppressing cooling), and the new positive mode (triggering star formation), we introduce two implementations of a third mode: the ejection of hot gas beyond the virial radius, AGNeject1 and AGNeject2. This component addresses a longstanding issue in semi-analytic models and hydrodynamical simulations: the overestimation of hot gas fractions in low and intermediate mass halos. FEGA25 is calibrated via MCMC using a suite of cosmological N-body simulations YS50HR, YS200, and YS300, and a comprehensive set of observed stellar mass functions across a wide redshift range. We find that supernova feedback dominates gas ejection in halos with logM_{halo} < approximately 12, while AGN feedback becomes increasingly important at higher halo masses. The AGNeject2 model, which activates primarily at late times, redshift < 1, reproduces a characteristic cavity, a U shaped feature in the baryon fraction at redshift zero, similar to trends observed in simulations like SIMBA and IllustrisTNG. Conversely, AGNeject1 yields a smoother, redshift independent evolution. Both models preserve the stellar and cold gas components and successfully reproduce the stellar to halo mass relation up to redshift 3. Our results emphasize that a physically motivated AGN driven mechanism capable of selectively removing hot gas is essential to accurately model the baryon cycle, particularly in intermediate halo mass regimes.

Figures (7)

• Figure 1: Baryon fraction normalized to the universal value ($\rm{Y}{200}$) as a function of halo mass for the AGNeject1 (left) and AGNeject2 (right) models. Solid lines indicate different redshifts, compared with predictions from NewCluster (filled circles) and literature data (symbols in the legend). AGNeject1 shows almost no evolution with time, while AGNeject2 predicts a strong late-time decline in baryon content between $11.8 < \log M{\rm{halo}} < 13.4$. On small scales, NewCluster supports the near-constant trend of AGNeject1. On cluster scales, both models agree with chiu2016 and akino2022, but fall below angelinelli2023. The most distinctive feature is the deep $z=0$ cavity in AGNeject2, signaling enhanced baryon depletion at late times.
• Figure 2: Hot gas fraction normalized to the universal baryon fraction as a function of halo mass at different redshifts (separate panels), predicted by AGNeject1 (blue) and AGNeject2 (red). Results are compared with observational constraints and simulations at $z=0$ and $z=1$, and with simulations only at higher redshifts. Predictions from NewCluster are also shown at $z \geq 1$ (magenta circles). Both models broadly agree with observations at $z=0$ and show hints of convergence at $z=1$, though they predict lower fractions than MAGNETICUMangelinelli2023 at all epochs. The main difference arises in Milky Way–like halos: AGNeject1 is more effective until $z=1$, while AGNeject2 becomes dominant in the last $\sim 7$ Gyr. At $z=3$, NewCluster gives slightly lower values but follows the same trends. This explains the deep baryon fraction cavity at $z=0$ in AGNeject2 (Figure \ref{['fig:fig1']}).
• Figure 3: Stellar-to-halo mass relation for central galaxies at different redshifts (panels), as predicted by AGNeject1 (blue) and AGNeject2 (red). At $z=0$, predictions are compared with the SHARK2 SAM lagos2024, the empirical relation of moster2013, and observational estimates from romeo2020, hinshaw2013, gonzalez2013, kravtsov2018, and mancera-pina2025. At $z \geq 1$, results are compared with moster2013 and the NewCluster simulation. Both models agree well with semi-analytic, empirical, and observational constraints at $z=0$. At higher redshifts, NewCluster tends to overpredict stellar masses in low-mass halos. Matching this relation over cosmic time is crucial, especially in low-mass halos where the central galaxy contains most of the stellar mass within the virial radius.
• Figure 4: Efficiency of AGN feedback in ejecting gas beyond the virial radius as a function of halo mass. The efficiency is defined as the total gas mass expelled by AGN up to a given redshift (including all progenitors), normalized by the halo mass at that epoch. Predictions from our two models are shown: AGNeject1 (left) and AGNeject2 (right), with colors indicating different redshifts. In AGNeject1, efficiency gradually rises with time, peaking between $12.5 < \log M_{\rm{halo}} < 13$, depending on redshift, and flattening to $\sim 5\%$ on cluster scales. By contrast, AGNeject2 remains negligible until $z=1$, then increases sharply at $\log M_{\rm{halo}} > 12$, reaching a peak near $\log M_{\rm{halo}} \sim 13$ at $z=0$. This highlights the contrasting timing and strength of AGN-driven gas ejection in the two models.
• Figure 5: SN feedback efficiency as a function of halo mass for the AGNeject1 (left) and AGNeject2 (right) models, shown at different redshifts (color-coded lines as indicated in the legend), following the same format as Figure \ref{['fig:fig4']}. SN feedback is most effective in low-mass halos and progressively declines with increasing halo mass. The two panels are identical, reflecting the fact that the AGN feedback implementations in these models do not influence the SN feedback.