The JWST EXCELS Survey: gas-phase metallicity evolution at 2 < z < 8

TL;DR

This study maps gas-phase metallicity evolution from 2 < z < 8 using 65 star-forming galaxies from JWST/EXCELS, employing homogeneous strong-line calibrations calibrated to EXCELS and direct $T_e$ metallicities for a subset. It finds a persistent, near-constant MZR slope across redshifts with a normalization that declines by ~0.1 dex from z ≈ 3.2 to z ≈ 5.5, indicating rapid early enrichment in the first ~1 Gyr. The analysis shows only tentative evidence for a metallicity–SFR (FMR) coupling at high redshift, with offsets from the local FMR and a breakdown of the local FMR at z ≳ 3, suggesting evolving baryon-cycle physics; comparisons to literature and simulations place IllustrisTNG as the best-matching model for these data. Together, the results support rapid chemical enrichment in the early Universe and emphasize the need for larger, homogeneous high-z samples to constrain the FMR and its evolution.

Abstract

We present an analysis of the gas-phase mass-metallicity relationship (MZR) and fundamental metallicity relationship (FMR) for $65$ star-forming galaxies at $2 < z < 8$ from the JWST/EXCELS survey. We calculate gas-phase metallicities (12 + log(O/H)) using strong-line calibrations explicitly tested against the EXCELS sample, and report direct-method metallicities for $25$ galaxies. Our sample spans $8.1<\log(\rm M_\star/M_\odot)<10.3$ in stellar mass and $0<\log(\rm SFR/M_\odot \, yr^{-1})<2$ in star-formation rate, consistent with typical main-sequence star-forming galaxies at the same redshifts. We find a clear MZR at both $2<z<4$ ($\langle z \rangle = 3.2$) and $4<z<8$ ($\langle z \rangle = 5.5$), with consistent slopes and mild evolution in normalization of $\simeq 0.1 \, \mathrm{dex}$, matching trends from simulations and recent observations. Our results demonstrate rapid gas-phase enrichment in the early Universe, with galaxies at fixed mass reaching $\simeq 50$ per cent of their present-day metallicity by $z \simeq 3$ (within the first $\simeq 15$ per cent of cosmic time). We find tentative evidence for SFR-dependence in the MZR scatter, though the results remain inconclusive and highlight the need for larger high-redshift samples. Comparison with locally derived FMRs reveals a clear offset consistent with other $z > 3$ studies. We discuss potential drivers of this offset, noting that high-redshift samples have significantly different physical properties compared to local samples used to define the $z=0$ FMR. Our results confirm that low-mass, high specific star-formation rate galaxies common at high redshift are inconsistent with the equilibrium conditions underlying the local FMR, and highlight the rapid chemical enrichment at early cosmic epochs.

Figures (15)

• Figure 1: An example rest-frame optical spectrum of one galaxy in our sample at $z=5.225$ (EXCELS-121806). The top and lower panels show the 1D and 2D spectra for the G235M and G395M gratings respectively. The black lines show the 1D spectrum with the corresponding $1\sigma$ uncertainty in grey shading. The combined continuum and emission line model fit is plotted in purple. Vertical black dashed lines highlight an number of nebular emission features across the full wavelength range, with an inset in the top panel showing the H $\gamma$ and [O iii]$\lambda 4363$ features. In the bottom panel we include a false three-colour image of EXCELS-121806 generated from the F115W + F150W, F200W + F277W, and F356W + F444W PRIMER imaging with the position of the NIRSpec MSA slitlets shown in white.
• Figure 2: The [N ii]-BPT diagram baldwin1981 for the EXCELS galaxies in two redshift bins ($2<z<4$ in blue; $4<z<8$ in red), with $2\sigma$ upper limits shown by triangles. All of the galaxies in our sample with coverage of the required lines exhibit [O iii]$\lambda 5007$/H $\beta$ and [N ii]$\lambda 6584$/H $\alpha$ ratios fully consistent within $1\sigma$ with ionization via star-formation activity rather than AGN according to the separation lines from kewley2001 and kauffmann2003. We additionally show galaxies in the selected sample which do not have coverage of [N ii]$\lambda 6584$ and H $\alpha$ as faint points at the right of the plot.
• Figure 3: Sample distributions in $z$, $\log (M_\star / {\rm M_\odot})$ and $\log ({\rm SFR_{\rm SED}} / {\rm M_\odot \, yr^{-1}})$ for the strong-line $2<z<4$ and $4<z<8$ samples (blue and red) versus the parent sample (grey) in panels from left to right. For clarity, the selected sample are shown as raw counts, whilst the parent sample is shown as a percentage distribution. The median and standard deviation of stellar mass and SFR for the $2<z<8$ selected sample are shown at the top their respective panels. Our selected sample is generally representative of the EXCELS parent sample, with similar average masses of $10^{9}$ versus $10^{9.1}\,\rm M_\odot$ and star-formation rates of $5$ and $4\,\rm M_\odot\,yr^{-1}$ respectively.
• Figure 4: In the left panel, we show the agreement between our SED- and H $\alpha$-based SFR probes (with $2<z<4$ galaxies in blue and $4<z<8$ galaxies in red). We find an average offset of $-0.07\pm 0.01\,{\rm dex}$, indicating that our SED modelled parameters are representative of our spectroscopically derived parameters. In the centre and right panels we show the ${\rm SFR_{H\alpha}}$ star-forming main sequence for our sample in two redshift bins: $2 < z < 4$ (blue hexagons) and $4 < z < 8$ (red hexagons). For each redshift bin we plot a series of empirical relations at similar redshifts from speagle2014, popesso2023, clarke2025, and the theoretical prescription from mcclymont2025a. In both the $2 < z < 4$ and $4 < z < 8$ redshift bins $94$ per cent of galaxies fall within at least $2\sigma$ of the star-forming main sequence, indicating these galaxies are broadly representative of the general star-forming galaxy populations at these epochs.
• Figure 5: The strong-line metallicity based MZR for the EXCELS sample (hexagons) using the scholte2025 calibrations for the full sample ($\langle z \rangle = 4.0$, left panel), $2 < z < 4$ ($\langle z \rangle = 3.2$, centre panel) and $4 < z < 8$ ($\langle z \rangle = 5.5$, right panel). Each point is colour coded according to its redshift, with the associated colour bar in the top left of each panel, above which is plotted a histogram showing the redshift distribution of each sample. The best fit MZR and $1\sigma$ uncertainty for the $2<z<4$ and $4<z<8$ subsamples are shown by solid grey lines and a grey shaded region in the centre and right panels respectively, with the alternate redshift MZR shown as a black dotted line for comparison. In the centre and right panels, we additionally show inset panels with the direct-method metallicities plotted against the strong-line MZR in each redshift regime.