The first GLIMPSE of the faint galaxy population at Cosmic Dawn with JWST: The evolution of the ultraviolet luminosity function across z~9-15

Using ultra-deep JWST NIRCam imaging from the GLIMPSE Survey, enhanced by gravitational lensing of the AbellS1063 cluster, we investigate the faintest galaxies ever observed in the redshift range z9 to z15. We identify 105 galaxy candidates within this range, spanning absolute ultraviolet (UV) magnitudes from M_UV~-18 to M_UV~-13 mag, about three magnitudes fainter, on average, than prior JWST studies. We place strong constraints on the ultra-faint end of the UV luminosity function (UVLF), finding minimal evolution in the faint-end slope, which varies from $α=-2.01\pm0.20 at z=9 to α=-2.10\pm0.19$ at z=13. This behaviour contrasts with the rapid evolution of the faint-end slope observed from z0 to z9. By integrating the UVLF down to M_UV=-16, we derive the cosmic star formation rate density (SFRD)from z=9 to z=13, revealing a best-fit redshift evolution that follows $\propto(1+z)^{-2.94^{+0.06}_{-0.10}}$. This slope is significantly shallower than predictions from most theoretical models. Extending the integration limit to M_UV=-13, we find that galaxies fainter than M_UV=-16 contribute more than 50% of the total cosmic SFR density at z~12. The observed excess in the cosmic SFRD at these high redshifts may suggest an enhancement in the star formation efficiency during the earliest phases of galaxy formation. Alternatively, this could result from other physical mechanisms, such as bursty star formation histories; minimal dust attenuation; or an evolving initial mass function. However, existing models that incorporate these scenarios fail to fully reproduce the observed redshift evolution of SFRD. Finally, we acknowledge the potential impact of low-redshift contamination and cosmic variance, as the small survey volume may not represent the broader galaxy population. Similar observations in different fields and spectroscopic confirmation are required to validate these findings

💡 Research Summary

The paper presents the first ultra‑deep JWST NIRCam study of the faintest galaxies at Cosmic Dawn (z ≈ 9–15) by exploiting the strong gravitational lensing provided by the massive cluster Abell S1063 in the GLIMPSE program. Using seven broad‑band and two medium‑band filters, the authors achieve a 5σ depth of ≈30.8 AB mag, and after careful background subtraction, intra‑cluster light removal, and PSF homogenisation, they construct a high‑quality photometric catalogue. Source detection is performed separately on the short‑wavelength (SW) and long‑wavelength (LW) stacks with SExtractor, then merged to retain all SW detections and any additional LW‑only sources. Aperture photometry (0.2″) is corrected using empirical PSFs, and uncertainties are derived from random aperture placements.

A new parametric strong‑lensing model of Abell S1063 is built on the Zitrin et al. (2015) framework, consisting of two large‑scale pseudo‑isothermal elliptical mass distributions (PIEMDs) and 303 cluster member galaxies modelled as dual‑pseudo‑isothermal ellipsoids (dPIEs). The model is constrained by 75 multiple images from 28 sources (24 with spectroscopic redshifts) and optimized via MCMC, achieving an image‑plane RMS of 0.54″. Magnification factors (μ) and their uncertainties are extracted from the posterior, with an additional conservative 15 % systematic term added.

High‑redshift candidates are selected through classic Lyman‑break colour cuts combined with photometric‑redshift fitting (EAZY). The selection criteria enforce non‑detections in filters blueward of the break and require robust detections in redward bands, minimizing low‑z interlopers. This yields 105 galaxy candidates spanning 9 ≲ z ≲ 15 and absolute UV magnitudes M_UV ≈ ‑13 to ‑18, i.e., roughly three magnitudes fainter than previous JWST surveys.

Completeness is quantified via injection‑recovery simulations that account for magnification‑dependent effective area. The UV luminosity function (UVLF) is derived using a 1/V_max estimator and fitted with a Schechter function. The faint‑end slope α shows little evolution: α = ‑2.01 ± 0.20 at z = 9 and α = ‑2.10 ± 0.19 at z = 13, contrasting with the steepening observed from z ≈ 0 to z ≈ 9. Integrating the UVLF down to M_UV = ‑16 gives a cosmic star‑formation rate density (SFRD) that evolves as ρ_SFR ∝ (1 + z)^{‑2.94^{+0.06}_{‑0.10}}, significantly shallower than most theoretical predictions (which typically predict ≈(1+z)^{‑4} to (1+z)^{‑5}). Extending the integration limit to M_UV = ‑13 reveals that galaxies fainter than ‑16 contribute >50 % of the total SFRD at z ≈ 12, implying that the bulk of early star formation resides in an ultra‑faint population.

The authors discuss several physical interpretations: (i) an enhanced star‑formation efficiency in low‑mass halos at early times; (ii) bursty star‑formation histories that temporarily boost UV output; (iii) negligible dust attenuation in these primitive systems; and (iv) a possible evolution of the stellar initial mass function toward more massive stars. Existing semi‑analytic and hydrodynamic models that incorporate some of these effects still fail to reproduce the observed shallow SFRD decline, suggesting missing physics or incorrect parameter choices.

Potential systematic uncertainties are acknowledged. Low‑redshift contamination, cosmic variance due to the modest survey volume, and residual lens‑model systematics could bias the faint‑end slope and SFRD estimates. The paper emphasizes the need for independent fields, deeper observations, and spectroscopic confirmation (e.g., with JWST NIRSpec) to validate the results.

In summary, this work pushes the frontier of high‑redshift galaxy studies by directly measuring the UVLF down to M_UV ≈ ‑13 at z ≈ 9–15, demonstrating a remarkably stable faint‑end slope and a higher-than‑expected contribution of ultra‑faint galaxies to the cosmic star‑formation budget. These findings provide critical new constraints for models of early galaxy formation, reionization, and the buildup of stellar mass in the first few hundred million years of the Universe.