Understanding Interstellar Metals during Reionization with Radiative SPH Simulation: Metallicity and Emission Lines from the ISM at $10 \geq z \geq 5$

TL;DR

This paper addresses how the ISM in high-redshift galaxies is metal-enriched and ionized during the Epoch of Reionization by leveraging Technicolor Dawn radiative SPH simulations. It post-processes simulated galaxies with Cloudy to generate synthetic H II-region spectra, enabling direct comparisons to JWST-era measurements through the UVLF, the $M_{ m UV}-M_*$ relation, and emission-line diagnostics. The authors find no evolution in the mass–metallicity relation but an overabundance of oxygen by ~0.5 dex relative to observations, and they show O32-based metallicities are biased high by ~1 dex, implying calibration caveats and possible IMF or yield issues. Emission lines are generally weaker in the simulations compared to some observations, suggesting higher ionization parameters within the simulated galactic population and highlighting the need for improved ISM modeling, dust treatment, and resolution in high-redshift studies. Overall, the work provides a critical link between cosmological simulations and nebular diagnostics, with direct implications for interpreting reionization-era galaxy spectra and for refining models of metal enrichment and feedback.

Abstract

We compare the \texttt{Technicolor Dawn} cosmological simulations with recent observations of galactic nebular line emission during the Epoch of Reionization, providing stringent tests of the predicted ionization and metal enrichment levels. We validate the simulated population with the UVLF and $M_{\mathrm{UV}}-M_*$ relation and see that the simulated results are consistent with observations at lower masses. We extract local gas volumetric grids of density and mass-weighted metallicity, then we use \texttt{Cloudy} to produce synthetic emission spectra of \species{H}{ii} regions. The mass-metallicity relation does not evolve, which is also consistent with observations. The predicted oxygen abundance exceeds observational inferences by about 0.5 dex, suggesting either overly efficient enrichment or weak feedback. However, applying the O32 diagnostic directly to our synthetic spectra shows an offset of 1 dex from the correct outputted gas-phase metallicity. This suggests that O32 is biased high at a level that is more than sufficient to account for the simulation-observation offset. The simulated galaxies' line diagnostics show mostly weaker [\species{O}{iii}] lines and lower diagnostic values of O3 and Ne3O2 compared to observations. This suggests higher ionization parameters within the simulated galactic population in general.

Figures (8)

• Figure 1: (Top:) UV luminosity function calculated from the simulation both with dust (orange dashed) and without dust (blue solid). The scattered points are from Harikane:2023. Our simulations probe the low-mass galaxies that have not been observed yet. Agreement is good at $z=5$, but the model underproduces observations by an amount that increases with redshift at $z>5$. (Bottom:) Comparison of UV magnitude and stellar mass within our simulation, with estimated fits from the resolved simulated sample (black dashed) alongside observed median bins and 68% confidence at $z=10$ to $6$ from Stefanon:2021 after converting the IMF following Madau:2014. The gray shaded region is our resolution cut-off of 32 star particles. Simulated galaxies are considered unresolved around magnitudes $M_{\mathrm{UV}} =$ -14 to -13. Extrapolating the simulated mass-to-light ratios yields reasonable agreement with observations.
• Figure 2: Example spectrum outputted from our Cloudy runs. Plotted is the total spectrum (gas and star) and the emission spectrum (gas). The flux density is calculated at the "edge" of the galaxy, where the end of the gridding is located.
• Figure 3: Mass weighted oxygen abundance vs. stellar mass of the galactic population in the simulation. Lines are running median of the population in each redshift with the shaded regions the $1\sigma$ scatter. The oxygen abundance was calculated for each galaxy using the grids, but only counting cells that fit the criteria $n_{\mathrm{H}} \geq 1$ H atom/cm$^{3}$. The gray shaded region denotes data points below the simulation resolution of 32 mean star particles. The scatter points with errorbars are observational data from Trump:2023 (red squares, $5.27 \lesssim z \lesssim 8.49$), Tang:2023 (orange circles, $6.92 \lesssim z \lesssim 8.99$), and Atek:2024 (violet diamonds, $6.88 \lesssim z \lesssim 7.70$) for comparison. Although potentially overlapping considering uncertainties, the simulation tends to a higher bias of oxygen abundance.
• Figure 4: (Left) Comparison of oxygen abundance between the grid calculations and Cloudy line ratios using the empirical relation found in Perez-Montero:2021. Black, dashed line is a relation of unity. (Right) Plot of grid oxygen abundance compared to line ratio O32 from Cloudy spectra with linear fit (red, dashed) alongside equation form (red text). Single Cloudy model outputs are presented with dust-free (red circle) and attentuated with $A_V = 1.1, R_V = 3.1$ (red square) for comparison; calibration errors are at edge of markers. We see large offset of the inferred oxygen abundance, where the empirical calibration is biased higher than our simulation.
• Figure 5: Continuation of Figure \ref{['fig:OH12_gridvcal1']} but with a subset of galactic spectra without dust attenuation. These dust-free runs also show a significant higher bias in the line calibration estimate.