First Light And Reionisation Epoch Simulations (FLARES) XX: Comparing semi-analytic models at high-redshift

TL;DR

This study uses the FLARES zoom-simulation framework to compare high-redshift galaxy predictions from the EAGLE hydrodynamical model and four semi-analytic models (GALFORM, L-GALAXIES, SC-SAM, SHARK) in the redshift range $5\le z \le 12$. By running SAMs on the same zoom regions (DMO), and weighting across overdense regions to form a global population, the work assesses stellar mass functions, stellar-to-halo mass relations, star formation rates, and black hole properties, with tests against JWST data. The results show broad agreement for the stellar mass function but large differences in black hole populations and passive galaxy abundances, driven by how each model handles SMBH growth and satellite quenching; AGN feedback dominates passivity in Eagle, while environmental processes govern quenching in the SAMs. Overall, the comparison highlights how satellite treatment and SMBH physics shape high-redshift predictions, underscoring the need for more JWST constraints to refine these models and improve our understanding of galaxy formation in the Epoch of Reionisation.

Abstract

We explore how the choice of galaxy formation model affects the predicted properties of high-redshift galaxies. Using the FLARES zoom resimulation strategy, we compare the EAGLE hydrodynamics model and the GALFORM, L-Galaxies, SC-SAM and SHARK semi-analytic models (SAMs) at $5\leq z \leq 12$. The first part of our analysis examines the stellar mass functions, stellar-to-halo mass relations, star formation rates, and supermassive black hole (SMBH) properties predicted by the different models. Comparisons are made with observations, where relevant. We find general agreement between the range of predicted and observed stellar mass functions. The model predictions differ considerably when it comes to SMBH properties, with GALFORM and SHARK predicting between 1.5-3 dex more massive SMBHs ($M_{\rm BH}>10^6\ {\rm M_\odot}$) than L-Galaxies and SC-SAM, depending on redshift. The second half of our analysis focuses on passive galaxies. We show that in L-Galaxies and SC-SAM, environmental quenching of satellites is the prevalent quenching mechanism, with active galactic nuclei (AGN) feedback having little effect at the redshifts probed. On the other hand, $\sim40\%$ of passive galaxies predicted by GALFORM and SHARK are quenched by AGN feedback at $z=5$. The SAMs are an interesting contrast to the EAGLE model, in which AGN feedback is essential for the formation of passive galaxies, in both satellites and centrals, even at high redshift.

Figures (10)

• Figure 1: Snapshot cadence used in various simulations, from top to bottom: the fiducial Flares suite, the Flares DMO suite (resimulated from the parent simulation with the same cadence as the P-Millennium simulation Baugh_2019_GALFORM), the GUREFT simulation suite Yung_2024_GUREFT, the SURFS simulation suite Elahi_2018_SURFS, and the Millennium-I and II simulations Springel_2005_LHaloTreeBK_2009_MRII. The dashed red lines in the bottom-most row represent the snapshots used when running L-Galaxies on the Flares DMO suite -- these are subsampled from the SAM DMO snapshot cadence (same as P-Millennium).
• Figure 2: Stellar mass function predicted by the four SAMs (solid lines, or dotted to denote bins containing fewer than 10 galaxies) and Eagle (dashed grey line). A stellar mass cut of $M_\star>10^8\ \rm{M_\odot}$ is applied to the Eagle results. As the outputs of the fiducial Flares suite are saved at integer redshift, the Eagle predictions in the $z=10.5$ and $z=12.5$ panels are in fact at $z=10$ and $z=12$ respectively. Observations are plotted as symbols with error bars.
• Figure 3: Each row shows a different quantity predicted by the four SAMs and Eagle across a range of redshifts. From top to bottom: stellar-to-halo mass relation, SFR distribution function, star-forming main sequence. Note that a stellar mass cut of $M_\star>10^8\ \rm{M_\odot}$ is applied to the Eagle results, and that the Eagle predictions at $z=10.5$ are at $z=10$, as outputs of the fiducial Flares suite are saved at integer redshift. For the SAMs, dotted lines denote bins containing fewer than 10 galaxies. Shaded regions denote the 1$\sigma$ range of the SAMs. The dotted horizontal line in the final row marks the sSFR threshold below which galaxies are considered passive (relevant to § \ref{['sec:passive']}).
• Figure 4: Each row shows a different quantity predicted by the four SAMs and Eagle across a range of redshifts. The top row shows the black hole mass function, and the bottom row shows the stellar-to-black hole mass relation. Note that a black hole mass selection of $M_{\rm BH}>10^7\ \rm{M_\odot}$ is applied to the Eagle results, and that the Eagle predictions at $z=10.5$ are at $z=10$, as outputs of the fiducial Flares suite are saved at integer redshift. For the SAMs, dotted lines denote bins containing fewer than 10 galaxies, and shaded regions denote the 1$\sigma$ range.
• Figure 5: Number density of passive galaxies predicted by the four SAMs and Eagle when run on the Flares suite. In the left panel, the passive galaxy population is obtained using an sSFR cut of $\log_{10}(\rm{sSFR}/\rm{Gyr^{-1}})<-1$. For a more appropriate comparison with observations, the right panel uses an sSFR cut of $\log_{10}({\rm sSFR}/{\rm Gyr^{-1}})<0.2/t_{\rm obs}$. A mass cut of $\log_{10}(M_\star/\rm{M_\odot})>9.5$ is applied to facilitate comparison with observations.