Cosmological back-reaction of baryons on dark matter in the CAMELS simulations

TL;DR

Baryonic physics reshapes dark matter through cooling and feedback, modifying halo masses, inner densities, and large-scale clustering. Using thousands of CAMELS hydrodynamic simulations across four galaxy formation models and diverse parameter variations, the study quantifies back-reaction by comparing to matched N-body runs, revealing substantial model-dependent suppression of power and halo mass changes. Central densification from cooling competes with outer-region expansion from feedback, with the strength and radial extent of these effects tied to cosmology and subgrid parameters. The findings stress the necessity to marginalize baryonic physics in cosmological analyses and demonstrate CAMELS as a powerful platform to calibrate feedback models and interpret weak lensing measurements.

Abstract

Baryonic processes such as radiative cooling and feedback from massive stars and active galactic nuclei (AGN) directly redistribute baryons in the Universe but also indirectly redistribute dark matter due to changes in the gravitational potential. In this work, we investigate this "back-reaction" of baryons on dark matter using thousands of cosmological hydrodynamic simulations from the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project, including parameter variations in the SIMBA, IllustrisTNG, ASTRID, and Swift-EAGLE galaxy formation models. Matching haloes to corresponding N-body (dark matter-only) simulations, we find that virial masses decrease owing to the ejection of baryons by feedback. Relative to N-body simulations, halo profiles show an increased dark matter density in the center (due to radiative cooling) and a decrease in density farther out (due to feedback), with both effects being strongest in SIMBA (> 450% increase at r < 0.01 Rvir). The clustering of dark matter strongly responds to changes in baryonic physics, with dark matter power spectra in some simulations from each model showing as much as 20% suppression or increase in power at k ~ 10 h/Mpc relative to N-body simulations. We find that the dark matter back-reaction depends intrinsically on cosmology (Omega_m and sigma_8) at fixed baryonic physics, and varies strongly with the details of the feedback implementation. These results emphasize the need for marginalizing over uncertainties in baryonic physics to extract cosmological information from weak lensing surveys as well as their potential to constrain feedback models in galaxy evolution.

Figures (15)

• Figure 1: Ratio of matched hydrodynamic to $N$-body halo total masses at $z=0$ (solid lines), $z=1$ (dashed lines), and $z=2$ (dotted lines) as a function of $N$-body halo mass for all matched haloes in the CV set for SIMBA, IllustrisTNG, ASTRID, and Swift-EAGLE (from left to right). Lines represent the median value in each mass bin, while the shaded region represents the $25^{\rm th}$ to $75^{\rm th}$ percentile range at $z=0$. Haloes are matched across hydrodynamic and $N$-body simulations with the same initial conditions, and are selected and binned by the $N$-body halo mass. SIMBA and Swift-EAGLE show the greatest reduction in total halo mass relative to $N$-body simulations, followed by IllustrisTNG and ASTRID. Baryonic effects on halo mass are generally greater at later times.
• Figure 2: Same as Figure \ref{['fig:total_halo_mass']} but for the mass of the dark matter component of haloes. Dark matter particle masses in hydrodynamic simulations are scaled up to be the same as in the $N$-body run, for comparison. The dark matter content of haloes in hydrodynamic simulations follows the same general trend as total halo mass relative to $N$-body simulations, but with a lower magnitude of variation. Back-reaction effects generally reduce dark matter halo masses, with stronger impact at later times.
• Figure 3: Halo baryon fraction relative to the cosmic mean at $z=0$ as a function of $N$-body halo mass for all matched haloes in the CV sets of SIMBA, IllustrisTNG, ASTRID, and Swift-EAGLE (from left to right). The solid line represents the median value in each mass bin, while the shaded region represents the $25^{\rm th}$ to $75^{\rm th}$ percentile range. The dashed and dotted lines represent the gas and stellar components, respectively. Overall, SIMBA, IllustrisTNG, and ASTRID show qualitatively similar trends (albeit with different magnitudes) as a function of halo mass while Swift-EAGLE shows significantly lower baryon fractions in lower mass haloes that increase in proportion to halo mass.
• Figure 4: Ratio of hydrodynamic to $N$-body halo total masses at $z=0$ as a function of $N$-body halo mass for all matched haloes in the six-parameter 1P set for SIMBA, IllustrisTNG, ASTRID, and Swift-EAGLE (same as Figure \ref{['fig:total_halo_mass']} but for cosmological and feedback parameter variations). Each row corresponds to a different simulation suite, and each column corresponds to variations of a different parameter, as indicated in the first row and column. Lines of different colors in each panel show the median halo mass ratio for simulations varying only the corresponding parameter, as indicated by the color bar. Increasing $\Omega_{\rm m}$ (at fixed $\Omega_{\rm b}$) generally increases the mass of hydrodynamic haloes relative to $N$-body simulations (suggesting weaker feedback efficiency with lower baryon to dark matter content), while increasing $\sigma_{8}$ has contrasting effects depending on feedback model. Increasing AGN feedback efficiency generally decreases halo mass at $M_{\rm halo} > 10^{12}\rm{M_{\odot}}$ (but see $A_{\rm{AGN2}}$ in ASTRID). Increasing SNe feedback efficiency generally decreases halo mass at $M_{\rm halo} \lesssim 10^{11} \rm{M_{\odot}}$ but can have the opposite effect in higher mass haloes due to non-linear interaction of stellar and AGN feedback, with significant variation between models.
• Figure 5: Total matter (solid lines) and dark matter (dashed lines) density profiles for matched haloes within different mass ranges (increasing from left to right) at $z=0$ in the SIMBA (orange), IllustrisTNG (green), ASTRID (blue), Swift-EAGLE (red) and $N$-body (black) CV simulations (upper panels) and relative difference between each hydrodynamic profile to the $N$-body profile (lower panels) as a function of radial distance from the center of the halo (relative to the virial radius $r_{200}$ of the matched halo in the $N$-body simulation). Density profiles are multiplied by the square of the radial distance in each bin in units of $r_{200}$ to visually enhance the difference between profiles. Haloes are selected based on mass in $N$-body simulations and hydrodynamic haloes are the corresponding matched counterparts of these $N$-body haloes. When calculating dark matter density profiles in hydrodynamic simulations, dark matter particle masses are scaled up to be the same as in the $N$-body run such that the relative difference (bottom panels) for dark matter profiles (dashed lines) would be zero at all radii in the absence of back-reaction effects. Generally, both total and dark matter densities in hydrodynamic haloes are greatly increased at halo centers ($r<0.05\,r_{200}$) relative to $N$-body simulations, while densities are decreased between $r \sim 0.1$--$1\,r_{200}$, with significant differences across galaxy formation models.