CECILIA: The Mass-Metallicity Relation of Low-Mass Galaxies at Cosmic Noon

TL;DR

This work probes the low-mass end of the mass-metallicity relation (MZR) at $z\sim2-3$ by obtaining ultra-deep JWST/NIRSpec spectra for seven continuum-faint CECILIA Faint galaxies, enabling robust metallicity measurements down to $\log(M_*/M_\odot)\approx7.2$. Using the Nakajima22 $O3$-based calibration with $N2$ branch discrimination, the authors derive gas-phase abundances in the range $12+\log(O/H)\approx7.3$–$8.4$ and fit a linear MZR: $12+\log(O/H)=Z_{10}+\gamma\log(M_*/10^{10}\,M_\odot)$, finding $Z_{10}=8.58\pm0.14$ and $\gamma=0.48\pm0.11$ over $\log(M_*/M_\odot)\sim7.2$–$9.4$. The results indicate a steep slope at low masses, consistent with energy-driven outflows or reduced star-formation efficiency, and reveal evolution in normalization from $z\sim0$ to $z\sim2$, while showing little evolution from $z\sim2$ to the Epoch of Reionization. The paper also emphasizes the need for larger samples and direct $T_e$ measurements to calibrate strong-line metallicities and unravel the physical drivers of the MZR across cosmic time.

Abstract

A galaxy's metallicity and its relation to stellar mass encode the history of gas accretion, star formation, and outflows within cosmic ecosystems. We present new constraints on the low-mass end of the mass-metallicity relation (MZR) at $z\sim2-3$ from ultra-deep JWST/NIRSpec spectroscopy of seven continuum-faint galaxies in the Chemical Evolution Constrained using Ionized Lines in Interstellar Aurorae (CECILIA) Faint sample (Raptis et al. 2025). Our sample includes Ly$α$-selected and other low-luminosity star-forming galaxies with stellar masses $\log(M_\star / M_\odot)\sim7.2-9.7$ and moderately faint rest-UV magnitudes ($-20.7 \lesssim M_{\rm UV} \lesssim -17.3$). Gas-phase oxygen abundances, calculated using empirical calibrations of [O III]/H$β$ together with [N II]/H$α$ constraints, span $\sim0.04-0.5$ $Z_\odot$. We measure a steep MZR slope of $γ= 0.48 \pm 0.11$, suggesting a rapid increase in metal retention efficiency with mass, consistent with energy-driven outflows. Comparison with lower- and higher-redshift studies indicates an evolution in normalization from $z\sim0$ to $z\sim2$, reflecting less metal enrichment in early galaxies. We find no significant evolution in the MZR between $z\sim2$ and the Epoch of Reionization, suggesting that our galaxies may serve as useful analogs of reionization-era systems. Expanded samples and direct $T_e$-based abundance measurements will be crucial to fully trace the build-up of metals in low-mass galaxies during the peak epoch of cosmic star formation and to test the reliability of strong-line calibrations in these galaxies.

Figures (7)

• Figure 1: Left: H$\alpha$-based SFRs versus stellar mass for our CECILIA Faint sample (pink squares), derived using a metallicity-dependent conversion. For consistency, we recalculate the SFRs of the $z \sim 2.56$ stack from Trainor2016 and of the KBSS LBGs (stro2017; gray circles) using the same method. For comparison, we also show the H$\alpha$-based SFR--$M_\star$ relation for $1.4 < z < 2.7$ galaxies from Clarke2024, as well as individual lensed dwarf galaxies at $z \sim 2$--3 selected as [O iii] emitters in JWST/NIRISS spectroscopy by Li2023. We generally find consistency between the CECILIA Faint sources and these comparison samples in the SFR-M$_*$ plane. Right: The MEx diagram for the CECILIA Faint galaxies and the same comparison samples as in the left panel. The gray dashed contours represent SDSS $z \sim 0$ galaxies. We include the original Juneau2014 star-forming/AGN boundaries (purple dashed) and the redshift-adjusted versions shifted by $\Delta\log M_\ast = 0.75$ dex from Coil2015, appropriate for $z \sim 2$ galaxies. The CECILIA Faint galaxies broadly agree with the Li2023 sample at $M_\ast > 10^{8}\,M_\odot$, but have significantly lower O3 ratios at lower masses.
• Figure 2: Left: O3 vs. $12 + \log(\mathrm{O/H})$ diagram with O3-based metallicity calibration curves, as indicated in the legend. While most calibrations agree at intermediate to high metallicities ($12+\log(\mathrm{O/H}) \gtrsim 7.5$), they show substantial variation in O3 at fixed metallicity in the low-metallicity regime. We include the CECILIA Faint points (pink squares) based on their O3 measurements and inferred metallicities from the Nakajima22 curve. Because the O3–metallicity relation is double-valued, we use N2 (right panel) as a branch discriminator (turnover at $\mathrm{N2} \approx -1.7$). For galaxies with only upper limits in N2 that are consistent with both branches, we take the average of the two inferred metallicities and assign uncertainties that encompass the full possible range. Right: N2-BPT diagram Baldwin1981 showing SDSS $z \sim 0$ galaxies (gray contours), KBSS LBGs stro2017, and individual CECILIA Faint galaxies (pink squares; $2\sigma$ limits). The solid black line marks the AGN classification boundary from Kauffmann2003. Colored curves show metallicity calibrations for which both O3- and N2-based measurements are available at low and high redshift (same as in the left panel, except Jones2015 and Sanders2024). We adopt the Nakajima22 calibration for this study, which best tracks our data in the N2-BPT space.
• Figure 3: MZR for our CECILIA Faint sample, with gas-phase metallicities $12 + \log(\mathrm{O}/\mathrm{H})$ derived using the O3-based calibration of Nakajima22. The solid purple line shows the best-fit linear relation from our MCMC analysis, with the shaded region indicating the $1\sigma$ uncertainty. The slope inferred using the curt2017 calibration is shown in teal for comparison. Our measurements are compared with stacked spectra at $z \sim 2.3$ and $z \sim 3.3$ from Sanders2021, and with the KBSS relation determined using photoionization models Strom2022. Both MOSDEF and KBSS are consistent with our highest-mass ($M_\star > 10^{9}\,M_\odot$) CECILIA Faint galaxies. The two strong-line calibrations curt2017Nakajima22 agree within their $1\sigma$ uncertainties, highlighting that the steeper slope we observe at low masses is not driven by the choice of calibration.
• Figure 4: Comparison of the CECILIA Faint galaxy metallicities (pink squares) to the KBSS MZR at $z\sim2$ (black circles), where metallicities were inferred using the Nakajima22 O3-based calibration for consistency with the CECILIA Faint sample. The solid green line shows the median of 10,000 bootstrap fits to the KBSS sample, and the shaded regions indicate the 1$\sigma$, 2$\sigma$, and 3$\sigma$ prediction regions, from darkest to lightest, derived from the bootstrap ensemble after augmentation with the measured KBSS intrinsic scatter (0.10 dex). The purple line and shaded region show the Nakajima22 fit to our data and its 1$\sigma$ uncertainty. The two lowest-mass CECILIA Faint galaxies that drive the MZR slope are inconsistent with the KBSS MZR.
• Figure 5: A comparison of various low-mass MZR measurements at Cosmic Noon. The MZR for our CECILIA Faint sample (pink squares) is shown with the best-fit relation and scatter (purple line and shaded region). We include other direct-$T_e$ metallicity samples: two CECILIA Faint sources from Rogers2025, a stack of 52 LAEs at $z \sim 2.56$ from Trainor2016, and a stack of gravitationally lensed dwarfs at $z \sim 2.3$ from Gburek2023, which appear generally consistent with the CECILIA Faint sample. We also show the MZR fits at $z \sim 2$ and $z \sim 3$ from Li2023. As discussed in the text, we expect that significant differences in the selection of galaxies between our sample and the Li2023 sample drive the differences in our inferred MZRs.