ARCHITECTS I: Impact of subgrid physics on the simulated properties of the circumgalactic medium

TL;DR

This study probes how unresolved subgrid physics alter circumgalactic medium (CGM) properties by running three high-resolution cosmological zoom-in simulations of the same halo at $z\sim1$, each with a distinct feedback implementation (ME: mechanical, MT: hybrid, DC: delayed cooling) but calibrated to the same stellar mass. Using RAMSES-RT, the authors reveal that despite similar stellar masses, the three feedback schemes drive markedly different CGM states, gas phase distributions, inflow/outflow rates, and metal budgets: DC yields strong ejective feedback with a metal-rich CGM and reduced halo baryons, while ME and MT produce predominantly preventive feedback with more baryons retained and less CGM metal enrichment. The results highlight that subgrid prescriptions fundamentally shape CGM observables and galaxy growth, indicating that matching stellar masses alone is insufficient to constrain feedback physics. The work argues for incorporating CGM measurements, such as quasar absorption lines, to break degeneracies among subgrid models and to improve our understanding of feedback regulation across scales.

Abstract

Galaxy evolution is shaped by star formation and stellar feedback at scales unresolved by current high-resolution cosmological simulations. Precise subgrid models are thus necessary, and different approaches have been developed. However, they are degenerate and often primarily calibrated to reproduce stellar masses from observations. To explore these degeneracies, we perform three cosmological zoom-in radiation-hydrodynamics simulations of the same galaxy within a $5\times10^{11}\rm\ M_\odot$ dark matter halo at $z\sim1$, each with a different subgrid model: mechanical feedback, a combination of mechanical feedback and thermal feedback, and delayed cooling. We calibrate the simulations to match in stellar mass, isolating the effect of the models on the circumgalactic medium (CGM). Our findings demonstrate that despite producing galaxies with comparable stellar masses, the three models lead to distinct feedback modes, resulting in notable variations in the CGM properties. The delayed cooling run is dominated by ejective feedback and exhibits high burstiness, whereas mechanical and the hybrid models primarily feature preventive feedback, respectively acting at the galaxy and halo scales. Delayed cooling reduces the baryon mass to half the universal baryon fraction while mechanical feedback retains most baryons, with the hybrid model standing in between. Delayed cooling also ejects significantly more metals into the CGM than both other models. While for delayed cooling and mechanical feedback metals are almost evenly distributed in the CGM, they are concentrated around satellites in the hybrid model. These discrepancies emphasize the need to design an appropriate subgrid model to understand how stellar feedback regulates galaxy growth.

Figures (11)

• Figure 1: Top: Projection of the cells width for one of the ARCHITECT simulations (ME) at redshift $z=1$ in parsec. We trace two circles in dashed and dotted lines, with a respective radius of $0.1 \,R_{\rm 200}$ and $R_{\rm 200}$. Bottom: cell width as a function of radius in units of $R_{\rm 200}$ for the three ARCHITECT simulations (ME, MT, and DC, presented hereafter). The solid line shows the median resolution over $1\rm\ Gyr$ (from $z=1.3$ to $z=1$) and the shaded area corresponds to the 15.9 and 84.1 percentiles. We compare our simulation's resolution to Suresh2019 and Ramesh2024c.
• Figure 2: Star formation histories (top) for mechanical feedback (ME, blue), thermal feedback combined with mechanical feedback (MT, orange) and delayed cooling (DC, green) and burstiness (bottom). The star formation history is computed using a $\rm100\ Myr$ moving average for clarity, and the burstiness is quantified through a $\rm1\ Myr$ moving average to highlight short-term fluctuations. Following Eq. \ref{['eq:burstiness']}, we compute the burstiness from $\rm2\ Gyr$ to the end of the simulation (markers) over moving periods of $\rm30\ Myr$ and $\rm500\ Myr$ (left and right side of the violins). The violins show the probability density function (PDF) of the burstiness, and the horizontal bar corresponds to the median.
• Figure 3: Stellar mass-to-halo mass relation. Each marker represents a different simulation, as indicated in the legend. The colour code shows the redshift, which goes down to $z=1$ (blue). The solid blue line corresponds to the empirical model from Behroozi2019, with the blue shaded area showing the $16^\mathrm{th}-84^\mathrm{th}$ percentiles range. The grey shaded area indicates the broad contour of predictions of Moster2021 through reinforced learning. Both predictions are at redshift $z=1$.
• Figure 4: Phase diagrams of the gas contained within $R_{200}$ for each simulation. The result consists in 100 snapshots stacked over $1\rm\ Gyr$ from $z=1.3$ to $z=1$. From top to bottom are ME, MT, and DC with a colour code corresponding to the hydrogen mass normalised probability density function. The solid line contour encompasses 90% of the mass within $R_{200}$, and the outline in the orange dotted line is that of ME. We show the mass-weighted temperature and density probability distribution functions on the left and lower sides of each panel.
• Figure 5: Spherical instantaneous outflow rates as a function of radius in the CGM of ME, MT, and DC. We show total outflow rates (top) and split it between hot (T$>10^{4.5}\rm\ K$), warm ($\rm 10^{3.5}<T<10^{4.5}\rm\ K$), and cold (T$<10^{3.5}\rm\ K$) outflow phases (from top to bottom). The solid lines correspond to the median of each shell over $1\rm\ Gyr$ from $z=1.3$ to $z=1$, and the shaded area corresponds to the 15.9 and 84.1 percentiles. We split the figure between the ISM and the CGM by placing a vertical grey line at $0.1\ R_{200}$.