Out of oxygen: Extremely metal-poor galaxy candidates identified at $2.5 < z < 6.5$ with deep JADES medium-band imaging

Abstract

JWST is beginning to uncover a population of extremely metal-poor galaxies (EMPGs, $Z < 1\%~\mathrm{Z}_\odot$) at $z > 3$, mostly through serendipitous NIRSpec discoveries and blind slitless spectroscopy. To accelerate our understanding of pristine star formation, we further develop a methodology to identify EMPG candidates from photometry, using the extensive deep medium-band imaging from JADES. Our EMPG candidates at $2.5 < z < 6.5$ exhibit strong photometric boosts by H$α$, yet correspondingly weak boosts by [O III] + H$β$, likely indicating extremely low metallicity to explain their lack of [O III] emission. We further demand our EMPG candidates to have strong Balmer jumps, as revealed by medium-band imaging, to ensure that they are young starbursts, as opposed to broad-line AGN/LRDs, though contamination by dusty/dense-gas starbursts and highly-obscured AGN remains a concern. SED-fitting with near-pristine models (${\sim}0.1$-$1\%~\mathrm{Z}_\odot$) indicates that our 22 EMPG candidates are low-mass (median $M_* \approx 10^{6.7}~\mathrm{M}_\odot$), faint dwarf galaxies ($M_\mathrm{UV} \approx -16.6$), with high ionizing photon production efficiencies ($\log\, (ξ_\mathrm{ion, obs}/\mathrm{(Hz\ erg^{-1})}) \approx 26.0$). Hence these are plausible sites of near-pristine star formation, comprising ${\sim}0.04$-$0.6\%$ of $2.5 < z < 6.5$ galaxies at $-19 < M_\mathrm{UV} < -16$. We discuss this extremely metal-poor extension to the mass-metallicity relation. We forecast that deep (${\sim}28$ h) NIRCam slitless spectroscopy can identify bright EMPGs through strong H$β$ but lack of [O III] emission, or secure the redshifts of fainter systems through H$α$ detections. Highly-multiplexed NIRSpec spectroscopy offers an alternate route to discovering the faintest pristine galaxies out to $z=10$, without requiring deep medium-band/MIRI imaging to identify secure candidates.

Figures (14)

• Figure 1: SEDs and associated NIRCam photometry for 1 Myr old instantaneous starbursts at $z=4.25$, generated using Bagpipes. The [OIII] + H$\beta$ and H$\alpha$ emission lines (denoted) boost the bandpass-averaged flux densities in the F277W and F356W filters above the rest-frame optical continuum level. For typical galaxies, with moderate metallicities (e.g. $Z = 0.2~\mathrm{Z}_\odot$, blue), this results in blue $m_\mathrm{[O\ III]}$ - $m_\mathrm{H\alpha}$ colours in the wide-band filters tracing these emission lines. With decreasing metallicity, the [OIII] boost wanes, resulting in level photometry at low metallicity ($0.02~\mathrm{Z}_\odot$, green), and characteristically red $m_\mathrm{[O\ III]}$ - $m_\mathrm{H\alpha}$ colours for extremely metal-poor galaxies (e.g. $0.002~\mathrm{Z}_\odot$, red), reflecting strong boosts by H$\alpha$, but correspondingly weak boosts by [OIII]. Medium bands disentangle continuum and line emission, enabling EMPG candidates to be identified with greater confidence and with tighter metallicity constraints. Medium bands can directly trace the rest-frame optical (F300M) and NIR (F410M) continuum levels, also providing independent measurements of line boosts (F335M) which can be combined with wide-band measurements to obtain indirect continuum constraints. These continuum measurements enable empirical estimates of emission line equivalent widths from the line boosts, as well as a determination of the Balmer jump, a free--bound hydrogen recombination feature (at 3646 Å rest-frame, roughly at 1.9 µ m observed-frame) associated with young starbursts, through the deficit in rest-frame optical/NIR continuum level compared to the UV.
• Figure 2: Empirical selection of EMPG candidates from their strong H$\alpha$, but weak [OIII] emission, via equivalent width constraints from medium-band imaging. Most Balmer-jump galaxies (with $\Delta m_\mathrm{jump} < -0.15$ and $\mathrm{SNR_{Opt/NIR}} > 4$, green dots, of which there are 7956 in total from the JADES DR5 footprint) have high [OIII] + H$\beta$ rest-frame equivalent widths compared to H$\alpha$, generally being well above the 1:1 relation (black dotted line). Galaxies with low [OIII] + H$\beta$ equivalent widths relative to H$\alpha$ are rare. Our strict (dashed) and looser (dash-dotted) EMPG selection criteria, demanding strong H$\alpha$ emission ($\mathrm{EW_{H\alpha}} > 1500, 1000$ Å, red lines), but correspondingly weak [OIII] emission ($\mathrm{EW_{[O\ III] + H\beta}}/\mathrm{EW_{H\alpha}} < 0.55, 0.75$, orange) are shown. Our final 22 EMPG candidates with $Z < 0.01~\mathrm{Z}_\odot$ that strongly favour close-to-pristine SED model fits are displayed (light blue stars). For reference, the 45 EMPG candidates with $0.01 < Z/\mathrm{Z}_\odot < 0.02$ are also displayed (we do not consider these further, dark blue stars). Balmer-jump galaxies located in the EMPG selection region that are not considered EMPGs (green dots) typically have insufficient evidence to strongly support the very metal-poor scenario (i.e. $\Delta \chi^2$ between low and intermediate metallicity Bagpipes model fits is small), sometimes are inferred to have relatively weak H$\alpha$ equivalent widths (below 1500 Å), or in some cases are inferred to have higher photometric redshifts during Bagpipes fitting than as estimated via EAZY, with what was once considered strong H$\alpha$ emission now attributed to [OIII] + H$\beta$.
• Figure 3: Example SEDs for 4 of our most pristine EMPG candidates. NIRCam + HST/ACS photometry are displayed (black datapoints), with wide-band filters covering [OIII] and H$\alpha$ highlighted in orange and red, respectively. The SNR in the 0.2 arcsec diameter apertures are also shown. Low metallicity Bagpipes models (blue) provide the best possible description of the data, with much lower $\chi^2$ values ($\Delta \chi^2 > 9$) compared to intermediate metallicity models. The median inferred H$\alpha$ and [OIII] + H$\beta$ rest-frame equivalent widths from Bagpipes are displayed, together with the very red observed $m_\mathrm{[O\ III]} - m_\mathrm{H\alpha}$ colour, as well as the median inferred oxygen abundance $12 + \log (\mathrm{O/H})$. $2 \times 2$ arcsec RGB (F444W, F200W, F115W) cutouts are shown, also displaying the galaxy ID and photometric or spectroscopic (for 189323, 1008821) redshift. The EMPG candidates exhibit strong photometric boosts by H$\alpha$, yet correspondingly weak boosts by [OIII], also displaying prominent Balmer jumps, as required by our selection procedure. Potential photometric boosts by strong Ly$\alpha$ emission are visible in the dropout filters (purple) of IDs 51246 and 189323.
• Figure 4: Gas-phase metallicity traced by the oxygen abundance $12 + \log(\mathrm{O/H})$ against redshift. The photometrically-inferred metallicities of our final 22 EMPG candidates are displayed (light blue stars), spanning $2.83 < z < 6.31$. The light blue horizontal dashed line indicates our EMPG selection criterion: $Z < 0.01~\mathrm{Z}_\odot$. Spectroscopically-confirmed EMPGs are also shown, with gravitationally-lensed UNCOVER galaxies Chemerynska2024, a serendipitously-discovered CAPERS galaxy Cai2025, a galaxy with extremely blue UV slope Cullen2025, SAPPHIRES-EDR galaxies Hsiao2025, the extremely metal-poor host galaxy of a lensed LRD Maiolino2025b, the most pristine galaxy currently known Morishita2025, a highly-magnified strong Ly$\alpha$ emitter Nakajima2025, a highly-magnified UNCOVER galaxy Vanzella2025 and a highly-magnified CANUCS galaxy Willott2025. Such extremely metal-poor galaxies are typically missing (due to their rarity and faintness) from general high-redshift galaxy surveys (CEERS, ERO, GLASS and JADES, collectively, shown in grey), with metallicity measurements from Nakajima2023 (triangles), Curti2024 (squares) and Sarkar2025 (pentagons), which typically probe galaxies with much higher metallicities. Hence targeted NIRSpec follow-up of promising EMPG candidates is needed to accelerate our understanding of close-to-pristine star formation at high redshift.
• Figure 5: Dusty starbursts or starbursts radiating into dense gas can potentially masquerade as EMPG candidates in photometry, mimicking the strong H$\alpha$, but weak [OIII] + H$\beta$ emission of extremely metal-poor galaxies. Top panel: NIRSpec PRISM spectrum (black, $\pm 1\sigma$ errors in grey) of EMPG candidate ID 123650 in the vicinity of H$\alpha$ (Gaussian fit highlighted in red) and [OIII] + H$\beta$ (highlighted in orange). Owing to the non-detection of the [OIII] $\lambda5007$ and H$\beta$ lines we cannot make conclusive statements about the gas-phase metallicity of ID 123650. However, the large Balmer decrement $\mathrm{H}\alpha/\mathrm{H}\beta > 5.33$ indicates dust reddening and/or dense gas effects are likely affecting its spectrum and photometry. Bottom-left panel: The photometry of EMPG candidate ID 123650 is well-described ($\chi^2 = 15.4$) by extremely metal-poor models. Bottom-right panel: Models incorporating separate dust attenuation for the stellar (blue, weakly attenuated) and nebular (dark red, strongly attenuated) components of a 1 Myr old instantaneous starburst (green, total) from Bagpipes can also provide a good fit to the photometry ($\chi^2 = 20.8)$. However, this requires an extremely steep dust attenuation law for the nebular component, with power-law index $\delta_\mathrm{neb}=-1.5$ modifying the Calzetti2000 law, as well as a still relatively low metallicity ($Z_\mathrm{gas} \approx 0.03~\mathrm{Z}_\odot$). This extreme nebular dust attenuation is at odds with the detection of Ly$\alpha$ in the 141 h MUSE eXtremely Deep Field spectrum Bacon2023 of ID 123650. Hence we favour the dense-gas scenario (with electron density $n_\mathrm{e} \gtrsim 7 \times 10^5$ cm$^{-3}$) for ID 123650, where the weak [OIII] emission and weak H$\beta$ emission are attributable to collisional de-excitation and resonant scattering, respectively.