EDGE-INFERNO: How chemical enrichment assumptions impact the individual stars of a simulated ultra-faint dwarf galaxy

TL;DR

This study tackles how uncertainties in chemical enrichment modeling affect the stellar abundances in ultra-faint dwarf galaxies by running high-resolution, star-by-star cosmological zoom-in simulations of a single dwarf while systematically varying massive-star yields, SNe Ia timing, and stochastic sampling. The results show SNe Ia timing has the largest impact on mean abundances and [Fe/H], even in reionization-limited systems, while variations in massive-star yields mainly reshape abundance trends and potential bimodalities; stochastic IMF sampling introduces substantial galaxy-to-galaxy scatter that can obscure model differences in single-object comparisons. The work demonstrates that mean dwarf abundances and the luminosity–[Fe/H] relation are robust to many enrichment choices, but detailed abundance patterns and MDFs retain sensitivity, especially to SNe Ia and rotation-influenced yields, underscoring the need to average over populations to robustly constrain low-metallicity chemical enrichment. By connecting star-by-star modeling with resolved-star observations, the study highlights both systematic and statistical uncertainties in interpreting chemical observables in UFDs and argues for ensemble analyses to extract reliable constraints on nucleosynthesis and feedback in the low-mass regime.

Abstract

The chemical abundances of stars in galaxies are a fossil record of the star formation and stellar evolution processes that regulate galaxy formation, including the stellar initial mass function, the fraction and timing of type Ia supernovae (SNeIa), and nucleosynthesis inside massive stars. In this paper, we systematically explore uncertainties associated with modeling chemical enrichment in dwarf galaxies. We repeatedly simulate a single EDGE-INFERNO dwarf ($M_{\star} \approx 10^5 \, M_{\odot}$), varying the chemical yields of massive stars, the timing and yields of SNeIa, and the intrinsic stochasticity that arises from sampling individual stars and galaxy formation chaoticity. All simulations are high-resolution (3.6 pc), cosmological zoom-in hydrodynamical simulations that track the stellar evolution of all individual stars with masses $>0.5\,{\rm M}_{\odot}$. We find that variations in SNIa assumptions make the largest difference in mean abundance ratios and [Fe/H], highlighting the importance of detailed SNIa modeling even in such low-mass reionization-limited galaxies. In contrast, different massive star yields, accounting (or not) for stellar rotation, result in mean abundances comparable to those arising from stochasticity. Nonetheless, they significantly affect the shape of abundance trends with [Fe/H], for example, through the existence (or not) of a bimodality in the [X/Fe] - [Fe/H] planes, particularly in [Al/Fe]. Finally, we find that the variance arising from random sampling severely limits the interpretation of single galaxies. Our analysis showcases the power of star-by-star cosmological models to unpick how both systematic uncertainties (e.g., assumptions in low-metallicity chemical enrichment) and statistical uncertainties (e.g., averaging over enough galaxies and stars within a galaxy) affect the interpretation of chemical observables in ultra-faint dwarf galaxies.

Figures (11)

• Figure 1: The average cumulative mass of different elements (and total metallicity $Z$) injected by a mono-age population with mass $500\,{\rm M}_{\odot}$. Colors indicate different sources as labeled in the legend of the upper left panel. Except for SNIa, filled lines show the yields from NuGrid, while dashed lines show those from LimongiChieffi2018 with different models ($0,150,300\,{\rm km\,s}^{-1}$) shown with thicker lines for increasing rotation velocity. For SNIa, the filled (dashed) line shows the result with a delay time of $38\,{\rm Myr}$ ($100\,{\rm Myr}$).
• Figure 2: Median total mass loss as a function of time, with colored regions showing the $68$ percentiles for $1000$ samples of $500\,{\rm M_{\odot}}$ drawn from the IMF. This example uses the yields from NuGrid, and type Ia delay time of $38\,{\rm Myr}$, but the scatter in injected mass at a given time is similar regardless of model. Note that the large scatter in SNIa arises because of forcing discrete events for a small amount of stellar mass.
• Figure 3: Cumulative stellar mass as a function of stellar age (bottom axis) and redshift (top axis). Simulations are divided into the same groups as in Table \ref{['tab:models']}, emphasizing the assumption that is varied in each panel (each panel is labeled in the top left corner). The top panel shows three simulations using different lines but the same color for the two different models.
• Figure 4: Mean iron abundance $\langle[{\rm Fe/H}]\rangle$ as a function of total V-band magnitude $M_{\rm V}$ for all simulations presented in this work, divided into four groups (see Table \ref{['tab:models']}) for easier comparison. Top left: exploring the role of stochasticity by fixing the subgrid model to be NuGrid (delay) with $t_{\rm Ia}=38\,{\rm Myr}$ (red) or LimongiChieffi2018 with $t_{\rm Ia}=100\,{\rm Myr}$ (blue) varying only the random number seed. This panel also include the simulations from Andersson+2025 in black color. Top right: exploring the rotation models from LimongiChieffi2018 (teal, green, blue for $0$, $150$, and $300\,{\rm km\,s}^{-1}$, respectively), including one example with the rotation distribution from Prantzos+2018 (purple). All these models use $t_{\rm Ia}=38\,{\rm Myr}$. Lower left: testing different SNIa models, including $38$ (green) and $100\,{\rm Myr}$ (blue) for $t_{\rm Ia}$, one example without SNIa (purple), and one with metallicity-dependent SNIa yields (teal). All models use yields from LimongiChieffi2018 for massive stars. Bottom right: comparing the NuGrid models with the delay (red) and rapid (blue) explosion triggers for CCSN, both with $t_{\rm Ia}=38\,{\rm Myr}$. These simulations use a model identical to NuGrid (delay), $t_{\rm Ia}=38\,{\rm Myr}$, but from different initial conditions. The error bars on the vertical axis indicate the dispersion in [Fe/H], calculated by taking the standard deviation of all stars in each galaxy. Gray error bars show values from dwarf galaxies in the Local Group Volume, taken from the database by Pace2024. For these points, $M_{\rm V}$ has error bars indicating upper and lower errors, while error bars on [Fe/H] indicate the dispersion. The gray dashed line shows the dwarf galaxy luminosity-[Fe/H] fit from Kirby+2013.
• Figure 5: MDF of [Fe/H] of 6 simulations used to estimate the stochastic variance, each displayed showing the fraction of stars in bins with size 0.2 dex, comparable to the typical errors on observed estimates of these quantities. The colored lines show individual simulations, each distinguished by different lines, while the black lines show the combined MDF of all simulations using the same model. The top row shows results from the NuGrid (delay) model with SNIa delay time of $38\,{\rm Myr}$, and the bottom row shows results from the LimongiChieffi2018 model with $150\,{\rm km\,s}^{-1}$ rotation and a delay time of $100\,{\rm Myr}$ for SNIa. Each row includes three simulations (different line styles) executed with a different seed for random number sampling.