Draft version November 26, 2024

Typeset using LATEX twocolumn style in AASTeX63

A Glimpse of the New Redshift Frontier Through Abell S1063

Vasily Kokorev, Hakim Atek,2 John Chisholm,1 Ryan Endsley,1 Iryna Chemerynska,2 Julian B. Muñoz,1
1

Lukas J. Furtak,3 Richard Pan,4 Danielle Berg,1 Seiji Fujimoto,1, 5 Pascal A. Oesch,6, 5 Andrea Weibel,6
Angela Adamo,7 Jeremy Blaizot,8 Rychard Bouwens,9 Miroslava Dessauges-Zavadsky,6 Gourav Khullar,10
Damien Korber,6 Ilias Goovaerts,11 Michelle Jecmen,1 Ivo Labbé,12 Floriane Leclercq,13 Rui Marques-Chaves,6
Charlotte Mason,5 Kristen B. W. McQuinn,11, 14 Rohan Naidu,15 Priyamvada Natarajan,16, 17 Erica Nelson,18
arXiv:2411.13640v2 [astro-ph.GA] 23 Nov 2024

Joki Rosdahl,8 Alberto Saldana-Lopez,7 Daniel Schaerer,6 Maxime Trebitsch,19 Marta Volonteri,2 and
Adi Zitrin3
1 Department of Astronomy, The University of Texas at Austin, Austin, TX 78712, USA
2 Institut d’Astrophysique de Paris, CNRS, Sorbonne Université, 98bis Boulevard Arago, 75014, Paris, France
3 Department of Physics, Ben-Gurion University of the Negev, P.O. Box 653, Be’er-Sheva 84105, Israel
4 Department of Physics & Astronomy, Tufts University, MA 02155, USA
5 Cosmic Dawn Center (DAWN), Niels Bohr Institute, University of Copenhagen, Jagtvej 128, København N, DK-2200, Denmark
6 Département d’Astronomie, Université de Genève, Chemin Pegasi 51, 1290 Versoix, Switzerland
7 Department of Astronomy, The Oskar Klein Centre, Stockholm University, AlbaNova, SE-10691 Stockholm, Sweden
8 Universite Claude Bernard Lyon 1, CRAL UMR5574, ENS de Lyon, CNRS, Villeurbanne, F-69622, France
9 Leiden Observatory, Leiden University, NL-2300 RA Leiden, Netherlands
10 Department of Physics & Astronomy and PITT PACC, University of Pittsburgh, Pittsburgh, PA 15260, USA
11 Space Telescope Science Institute, 3700 San Martin Dr., Baltimore, MD 21218, USA
12 Centre for Astrophysics and Supercomputing, Swinburne University of Technology, Melbourne, VIC 3122, Australia
13 Univ Lyon, Ens de Lyon, CNRS, Centre de Recherche Astrophysique de Lyon UMR5574, F-69230, Saint-Genis-Laval, France
14 Department of Physics & Astronomy, Rutgers, The State University of New Jersey, Piscataway, NJ 08854, USA
15 MIT Kavli Institute for Astrophysics and Space Research, 70 Vassar Street, Cambridge, MA 02139, USA
16 Department of Astronomy, Yale University, 219 Prospect Street, New Haven, CT 06511, USA
17 Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, MA 02138, USA
18 Department for Astrophysical and Planetary Science, University of Colorado, Boulder, CO 80309, USA
19 LERMA, Sorbonne Université, Observatoire de Paris, PSL Research University, CNRS, 75014 Paris, France

(Received n/a; Revised n/a; Accepted n/a)

Submitted to ApJL

ABSTRACT
We report the discovery of five galaxy candidates at redshifts between 15.9 < z < 18.6 in JWST
observations from the GLIMPSE survey. These robust sources were identified using a combination
of Lyman-break selection and photometric redshift estimates. The ultra-deep NIRCam imaging from
GLIMPSE, combined with the strong gravitational lensing of the Abell S1063 galaxy cluster, allows us
to probe an intrinsically fainter population (down to MUV = −17.5 mag) than previously achievable.
These galaxies have absolute magnitudes ranging from MUV = −17.7 to −18.0 mag, with UV continuum
slopes between β ≃ −2.3 and β ≃ −3.0, consistent with young, dust-free stellar populations. The
number density of these objects, log10 (ϕ/[Mpc−3 mag−1 ])=−3.43+0.28 −0.64 at MUV = −18 is in clear
tension with pre-JWST theoretical predictions, extending the over-abundance of galaxies from z ∼ 10
to z ∼ 18.6. These results, together with the scarcity of brighter galaxies in other public surveys,
suggest a steep decline in the bright-end of the UV luminosity function at z ∼ 17, implying efficient
star formation and possibly a close connection to the halo mass function at these redshifts. Testing a
variety of star formation histories suggests that these sources are plausible progenitors of the unusually
UV-bright galaxies that JWST now routinely uncovers at z = 10 − 14. Overall, our results indicate

Corresponding author: Vasily Kokorev

[email protected]
2 Kokorev et al.

that the luminosity distribution of the earliest star-forming galaxies could be shifting towards fainter
luminosities, implying that future surveys of cosmic dawn will need to explore this faint luminosity
regime.

Keywords: High-redshift galaxies (734), Early universe (435)

1. INTRODUCTION Finkelstein et al. 2023; Harikane et al. 2023; Sun et al.

According to the standard paradigm of structure for- 2023; Li et al. 2024).
mation, the same primordial fluctuations that gave rise The ancestors of these UV-bright galaxies at even
to hot and cold spots in the cosmic microwave back- higher redshift (z > 15) have remained elusive. Many
ground (CMB) will eventually grow, collapse, and form observational campaigns have searched for detections
the first galaxies during cosmic dawn, ushering in the of first light galaxies, but few candidates pass rigorous
epoch of first light (e.g. Loeb & Furlanetto 2013). These scrutiny. Extremely dusty star-forming galaxies (Naidu
first galaxies have remained outside of our observational et al. 2022; Zavala et al. 2023; Arrabal Haro et al. 2023),
reach for decades. That is because they are faint and and spurious detections can masquerade as faint F200W
highly redshifted. Even the deepest Hubble Space Tele- drop-out galaxies. A key issue with detecting first-light
scope (HST ) surveys have fallen short of observing first galaxies is that they are expected to be intrinsically faint
light (Oesch et al. 2016). The ultraviolet (UV) light as they have had little time to assemble significant stellar
from the first galaxies drops precipitously due to ab- mass. The discovery and characterization of these elu-
sorption by foreground neutral gas (the Lyman-α break, sive primordial galaxies provides empirical constraints
see e.g., Steidel et al. 1996), which makes galaxies above on the astrophysics shaping their formation.
z ≳ 10 invisible in the HST IR filters. Here we report on observations from JWST’S
The James Webb Space Telescope (JWST ) was de- GLIMPSE survey (Atek et al., 2025, in prep.).
signed to observe the first galaxies. With its enormous GLIMPSE was specifically aimed to detect galaxies dur-
collecting area and unprecedented near infrared imaging ing the epoch of first light, by pairing ultra-deep NIR-
and spectroscopic capabilities, JWST /NIRCam is sensi- Cam imaging across seven wide and two medium bands
tive to faint light up to 5 µm (Rieke et al. 2023). These with the gravitational lensing of the foreground galaxy
are exactly the wavelengths required to discover galax- cluster Abell S1063. With even modest magnification
ies forming stars during the first few hundred million factors of 2, the 30.6 mag GLIMPSE observations can
years of cosmic history, as their Lyman-α break falls reach absolute magnitudes (MUV ) of −18 at z ∼ 17,
at 2.0 µm and 2.5 µm at redshifts 16 and 20, respec- probing the faint galaxies that likely existed during the
tively. This means that broad imaging bands, such as epoch of first light. We use these observations to identify
the F200W and F277W can be paired to search for con- five robust z ≳ 16 candidates, observed within a single
tinuum “dropouts” up to z ∼ 20, a mere 180 million NIRCam pointing, and assess the validity of these can-
years after the Big Bang. didates via model fitting, morphology and properties of
JWST has successfully used this technique to discover their stellar populations.
dozens of galaxies at 10 < z < 15. These observations Throughout this work we assume a flat ΛCDM cos-
have revealed a surprising over-abundance of UV-bright mology with Ωm,0 = 0.3, ΩΛ,0 = 0.7 and H0 = 70 km
galaxies at z > 10, challenging pre-JWST predictions s−1 Mpc−1 , and a Chabrier (2003) initial mass function
(to name just a few, Adams et al. 2023; Austin et al. (IMF) between 0.1 − 100 M⊙ . All magnitudes are ex-
2023; Atek et al. 2023; Curtis-Lake et al. 2023; Don- pressed in the AB system (Oke 1974).
nan et al. 2023; Finkelstein et al. 2023; Pérez-González
2. OBSERVATIONS AND DATA
et al. 2023; Castellano et al. 2024; Carniani et al. 2024a;
McLeod et al. 2024; Robertson et al. 2024; Chemerynska Detailed descriptions of the observations, data reduc-
et al. 2024). Attempts to explain this over-abundance of tion, cluster light removal, and source extraction will be
bright galaxies at extreme redshifts invoke extremely ef- presented in the GLIMPSE survey (Atek et al., 2025, in
ficient star formation in the early Universe, bursty star prep.); and are briefly summarized below.
formation histories, a lack of dust attenuation, a top-
heavy initial mass function (IMF), more efficient for- 2.1. GLIMPSE Survey
mation of dark matter halos, or modifications to the This work uses the ultra-deep imaging from the pub-
ΛCDM paradigm (e.g., Pacucci et al. 2022; Boylan- lic GLIMPSE survey (PID: 3293, PIs: H. Atek & J.
Kolchin 2023; Dekel et al. 2023; Ferrara et al. 2023; Chisholm). GLIMPSE targets Abell S1063 (AS1063
 GLIMPSE Discovery of z > 16 galaxies 3

hereafer), one of the highest-magnification regions in the 2.3. Cluster Light Contamination
Hubble Frontier Fields (HFF; Lotz et al. 2017), with Careful handling of the contamination light from the
7 broadband filters (F090W, F115W, F150W, F200W, bCGs and intra-cluster light (ICL) is necessary for our
F277W, F356W, F444W) and 2 medium-band filters science objectives. Improper treatment of the ICL could
(F410M, F480M). The total duration of the survey is potentially lead to inaccurate colors which in return af-
∼ 155h of science time, reaching unprecedented ob- fect the derived photometric redshifts, negatively im-
served depths, down to 30.6 mag uniformly across all pacting the high-z selection in the proximity of the clus-
wide bands. Specifically, the bands used in the z > 15 ter. To model and subtract the bCG and ICL light, we
drop-out selection - F200W and F277W have integration follow the robust methods described in Ferrarese et al.
times of 19 and 23 hours respectively. Furthermore, the (2006), Shipley et al. (2018) and Weaver et al. (2024) for
lensing magnification of Abell S1063 also means that we the Hubble Frontier Fields-Deep Space (HFFDS) and
can probe intrinsically faint sources which would other- UNCOVER/Mega Science photometric catalogs (Suess
wise be invisible in the deepest JWST blank field sur- et al. 2024). After both the bCGs and the ICL have
veys. been modeled and subtracted, we perform an additional
GLIMPSE observations use a MEDIUM8 readout pat- local background subtraction pass in the affected areas,
tern for all exposures, and adopt a 6 position pri- largely following the same methodology as in Section 2.2.
mary dither pattern to cover the short wavelength intra-
module gaps while maximizing the full-depth area. In
addition, we use a subpixel dither with 4 positions to 2.4. Source Extraction
best sample the PSF. This was done to optimize the
We construct empirical point-spread functions (PSFs)
S/N, with dithers large enough to mitigate the fixed
by stacking the available stars in the field, and match
pattern noise, imperative when searching for high-z tar-
all available bCG-subtracted JWST and HST images
gets.
to our lowest resolution PSF - F480M. The sources are
detected using SExtractor (Bertin & Arnouts 1996),
which we run in a single mode on an inverse-variance
2.2. Data Reduction weighted combination of the native PSF F277W, F356W
NIRCam imaging data for all 7 broad and 2 medium and F444W images. We then use photutils (Bradley
band filters in AS1063 are reduced following the proce- et al. 2020) to measure flux densities in circular aper-
dure in Endsley et al. (2024) using the jwst 1293.pmap tures with varying diameters from D = 0.′′ 1 – 1.′′ 2. The
context map. We implement crucial enhancements over photometric errors are determined separately for each
the standard STScI pipeline, including corrections for object and filter, taking into account both the aperture
cosmic rays, stray light, 1/f noise, and detector arti- size and the depth variation across the image. In the
facts (Bradley et al. 2023; Rigby et al. 2023). Given the vicinity of each source, we place 2000 random apertures
depth of the GLIMPSE campaign and the presence of in source-free parts of the image (as determined by the
bright cluster galaxies (bCGs), we model and subtract segmentation map). The standard deviation of the flux
the background on an amplifier basis with sep (Barbary density within empty apertures, plus the Poisson noise
2016), while manually masking out the bright regions. are then used as the final uncertainty. This method
This further improves the image background and allows is generally preferable over just using the weight/error
the mosaics to reach ∼ 0.3 mag deeper across all filters. maps, as it better accounts for effects of correlated noise
Our final data-set achieves 5σ aperture-corrected nomi- (e.g. see Endsley et al. 2023; Weaver et al. 2024). The
nal depths of 30.6 mag across all broad-bands in D=0.′′ 2 aperture corrections are calculated by assuming a point-
apertures. source profile. We use the empirical F480M symmetric
We also process and incorporate the deep HST ACS PSF curve of growth to determine the fraction of the to-
and WFC3 mosaics from the Hubble Frontier Fields tal flux that falls outside of each specified aperture size
(Lotz et al. 2017) and BUFFALO (Steinhardt et al. and then use that as our correction factor.
2020) programs. The HST images are based on Gaia- In order to select high-fidelity high-z candidates, we
aligned mosaics from the CHArGE archive (Kokorev flag all sources that fall on the edge of the mosaic, in-
et al. 2022), which are hosted on the Dawn JWST tersect with diffraction spikes from bright stars, or are
Archive (Valentino et al. 2023). We drizzle our final mo- close (< 2.′′ 0) to the modeled and subtracted bCGs. As
saics onto a 0.′′ 02/pixel grid for the JWST short wave- galaxies at high-z are generally small, we will only use
length (SW) filters, and 0.′′ 04/pixel for JWST long wave- the total fluxes within a D = 0.′′ 2 aperture for the re-
length (LW) and HST. mainder of our work.
4 Kokorev et al.

between the F277W and F200W filters, while simulta-

3 neously excluding objects with red continua in the rest-
frame UV between the F277W and F356W filters. These
color criteria are determined from running synthetic
2
(F200W-F277W)AB

photometry on a set of galaxy templates generated using

BEAGLE (Chevallard & Charlot 2016). We used star-
1 burst templates in the redshift range 15 < z < 20, incor-
porating IGM attenuation following Inoue et al. (2014),
with varying levels of attenuation between AV = 0 and
0 AV = 0.5, based on the SMC extinction law (Gordon
et al. 2003). We also consider low-redshift quiescent
This Work, z 17 galaxies that could mimic a Lyman break, using SEDs
1 Dusty SFG z 4.5 5.5 from the SWIRE template library (Polletta et al. 2007).
z 16 21 selection In addition, we include cool stars and brown dwarfs,
using stellar templates from Chabrier et al. (2000) and
1.0 0.5 0.0 0.5 1.0 1.5 Allard et al. (2001). The color-color tracks for all these
(F277W-F356W)AB simulated sources are shown in Figure 1, along with the
adopted selection window, which is defined by the fol-
Figure 1. High-z color-color selection. Our final high-
lowing criteria:
z sample, which we also present in Figure 2 is shown with
red diamonds. The additional, ∼ 4σ sources from our se-
lection are shown with black open symbols. The color-color M200 − M277 > 1.0
Lyman break selection is indicated with black dashed line. M200 − M277 > 1.2 + 2.0(M277 − M356 )
Solid lines show the tracks followed by starburst galaxies M277 − M356 < 0.5
at z ∼ 17 (Chevallard & Charlot 2016) with varying lev-
els of attenuation AV = 0–0.5 (shown with blue to green). In addition to the color criteria, we require that
We highlight low-redshift quiescent galaxies (Polletta et al. sources are well detected (at the > 3σ level in all three
2007), whose Balmer break could mimic the Lyman break at of the broad-band LW filters), and have a SNR of ≳ 4
higher redshifts, as dashed lines. Extreme dusty starbursts in at least one band, while remaining undetected (at
(Naidu et al. 2022; Arrabal Haro et al. 2023; Zavala et al.
the < 2σ level) in bands blueward of the Lyman break
2023) at z ∼ 4.5 − 5.5 (orange squares) look quite similar
to a high-z galaxy with this selection alone (see Section 3.3 (F090W, F115W, and F150W). This first selection leads
for why these are disfavored). Finally, cool stars and brown to a sample of ∼ 40 z ≳ 16 galaxy candidates.
dwarfs (Chabrier et al. 2000; Allard et al. 2001) are shown
in purple. 3.2. Photometric Redshift
To calculate photometric redshifts (zphot ) for all ob-
jects in the GLIMPSE catalog, we use the Python
3. DATA ANALYSIS version of EAZY (Brammer et al. 2008). We choose
Our goal is to find galaxies above z > 16. Previous the blue sfhz 13 model subset1 that contains redshift-
experience has demonstrated that a single selection tech- dependent SFHs, and dust attenuation values. More
nique, however, is insufficient to provide robust z > 16 specifically, the linear combinations of log-normal SFHs
candidates, and might result in low-z contaminants (e.g. included in the template set are not allowed to exceed
see Naidu et al. 2022). To mitigate that, we combine redshifts that start earlier than the age of the Universe
our selection to use the Lyman Break (Section 3.1) and (for more detail see Blanton & Roweis 2007). These
photometric redshift test (Section 3.2) which stringently models are further complemented by a blue galaxy tem-
tests for low-z contamination (Section 3.3) and uses the plate, derived from a JWST spectrum of a z = 8.50
size estimates (Section 3.5) to further test their robust- galaxy with extreme line equivalent widths (Carnall
ness. et al. 2022).
Our work is concerned only with the F200W dropouts,
3.1. Lyman Break Selection for which the HST data are simply too shallow to be
useful. Therefore, in our analysis we only focus on the
The first selection applied to the photometric catalog
is based on identifying dropouts in the rest-frame UV,
1
using the Lyman-α break technique. We define a se- https://github.com/gbrammer/eazy-photoz/tree/master/
lection window in color-color space to isolate dropouts templates/sfhz
 GLIMPSE Discovery of z > 16 galaxies 5

GLIMPSE JWST bands. We fit the aperture corrected JWST data have significantly matured, such that the
D = 0.′′ 2 JWST flux densities, including the upper lim- vast majority of ∼ 95 photometrically-identified z > 10
its, using the 0.01 < z < 30 redshift grid. The uncer- galaxies have been spectroscopically confirmed at high-
tainties on the photometric redshifts are computed from redshift (Arrabal Haro et al. 2023; Hainline et al. 2024;
the 16th and 84th percentiles of the redshift probability Harikane et al. 2024). At this moment, these initial
distributions - p(z). The best-fit EAZY SEDs are only misidentifications provide a road map to rigorously scru-
used to validate the color-color high-z selection. tinize the z > 16 candidates, however at the moment no
In order to select z > 16 galaxy candidates with galaxies with z > 14 have been confirmed spectroscopi-
EAZY we then require that the sources must be de- cally.
tected (S/N> 3) in at least 3 bands, similarly to our To address the potential contamination from low-z in-
color-color selection, and have a well constrained p(z) terlopers, we re-fit all of our 15 high-z candidates with
(FWHM< 2.5) without a prominent secondary redshift the EAZY blue sfhz 13 model suite, plus an addi-
solution. Cross-referencing our EAZY high-z sample tional dusty starburst template (Naidu et al. 2022; Arra-
with the color-color selected F200W dropout from Sec- bal Haro et al. 2023), to mimic the low-z interlopers.
tion 3.1 results in 15 candidates at z ≃ 15.5 − 19.2. The redshift grid in this case is forced to be z < 6, to
In addition to EAZY, we use BEAGLE (Chevallard match the redshift range of dusty interlopers. Decid-
& Charlot 2016) and the methods outlined in Ends- ing which model is better relies on the Bayesian Infor-
ley et al. (2024), to independently derive photometric mation Criterion (BIC) test (Schwarz 1978). This test
redshifts. Briefly, we use the BEAGLE models con- computes which template is the most likely fit to the
structed from updated Bruzual & Charlot (2003) stellar observed data, while penalizing models that have too
templates that use the PARSEC isochrones (Bressan et al. many free parameters. In our case a high-z fit is statis-
2012; Chen et al. 2015). These models have been passed tically preferred over forced low-z (z < 6) solutions with
through Cloudy (Ferland et al. 2017) to self-consistently the additional “Schrödinger” template, when a BIC dif-
produce nebular (both line and continuum) emission ference reaches 5 or more (using the criteria defined in
(Gutkin et al. 2016). We explore different parametric Jeffreys 1961). We find that almost half – 7/15 – of our
star formation histories (including constant, burst, and a high-z objects can be fit equally well with a low-z dusty
two-component star formation history), stellar ages (be- starburst template (∆BIC< 5). These galaxies are re-
tween 1 Myr and 30 Gyr), metallicities (0.0063–0.5 Z⊙ ), moved from our final sample, leaving us with 8 z ∼ 17
and attenuation laws (we settle on the SMC law; Pei galaxy candidates. While these are robustly detected
1992). The two codes agree exceptionally well (Figure 2) (∼ 3σ) in 3 bands, we further limit our final sample to
on the high-z solutions. only include galaxy candidates that reach at least 5σ in
one or more bands. This is done to ensure that only
3.3. Low-z Confusion the best high-z candidates end up in our final sample.
This leaves us with 5 objects which we show in Figure 2.
The early days of JWST, initial discoveries saw a slew
We show SEDs of the remaining 3/8 objects with peak
of z > 16 galaxy candidates, for which valid concerns
SNR> 4 in the Appendix A. We also would like to note
have been raised in the literature regarding the ability
that the morphology of our objects (Figure 2), does not
of photometric redshifts to pick out true high-z candi-
change from band-to-band when moving to redder fil-
dates (Arrabal Haro et al. 2023; Donnan et al. 2023;
ters, as it would for dusty galaxies.
Zavala et al. 2023). These works have shown that dusty
extremely star-forming galaxies at z ∼ 5 can masquer-
ade as high-z objects and bias our view of high-redshift 3.4. Gravitational Magnification
galaxies. For example, the so-called “Schrödinger” To take into account the effects from gravitational
galaxy was initially photometrically-identified at z > 16 lensing we use a new strong lensing (SL) model of
(Naidu et al. 2022; Donnan et al. 2023), and then had AS1063 constructed for GLIMPSE with the updated
its redshift revised to a z ∼ 5, via extreme emission version of the Zitrin et al. (2015) parametric method
lines found in NIRSpec spectra, as an interloper (Arra- that was recently used for several JWST SL clusters
bal Haro et al. 2023; Zavala et al. 2023). Despite this (e.g. Pascale et al. 2022; Furtak et al. 2023; Meena et al.
misidentification, these outliers provide invaluable in- 2023). We model the cluster with two smooth dark mat-
sight into the potential pitfalls of our methodology. Sub- ter (DM) halos parametrized as pseudo-isothermal ellip-
sequently, identification methodologies have been im- tical mass distributions (PIEMDs; Kassiola & Kovner
proved, and the techniques used to process (e.g. vari- 1993): one centered on the bCG, and the other on a
ous pipeline improvements and calibrations) and analyze group of galaxies in the north-east of the cluster (e.g.
6 Kokorev et al.

Figure 2. Most secure z ≳ 16 candidates from GLIMPSE. For each source we show 1.′′ 0 cutouts in all broad and medium
band NIRCam filters, a detection (F277W+F356W+F444W) LW stack with a D=0.′′ 2 aperture (green) overlaid. On the SED
plot, we show the best fit high-z vs low-z EAZY models (maroon and blue), as well as the same for BEAGLE (dashed orange
and green). The colored circles represent the integrated flux density for each solution. We explicitly highlight the difference
between a low-z interloper and high-z EAZY solution with a ∆BIC statistic on each panel, as discussed in Section 3.3. In the
inset panel we show the p(z) for both EAZY and BEAGLE fits with a free redshift solution. Detections are shown as black
squares, 3σ upper limits are shown as downward arrows.
 GLIMPSE Discovery of z > 16 galaxies 7

Bergamini et al. 2019; Beauchesne et al. 2024). In ad- mass, from SED fitting codes would be simply unrea-
dition, we model 303 cluster member galaxies as dual sonable. As such, we will only focus on the observable
pseudo-isothermal ellipsoids (Elı́asdóttir et al. 2007). stellar population parameters that can be derived from
The model is constrained with 75 multiple images of the rest-UV photometry alone.
28 sources, 24 of which have spectroscopic redshifts We estimate the absolute UV magnitude for each
(Balestra et al. 2013; Monna et al. 2014; Richard et al. galaxy from the observed F277W band, which samples
2021; Beauchesne et al. 2024; Topping et al. 2024). The λrest ∼ 1500 Å at this redshift. We find that our
model achieves an average image reproduction error of targets cover a very narrow range of derived MUV =
∆RMS = 0.54′′ . We refer the reader to Furtak et al. (in −17.91+0.21
−0.10 , after accounting for lensing magnification.
prep.) for more details on the lens model. A previous We derive the UV-slope β for each object by assuming
(HST -based) version of this model was recently used in fλ ∼ λβ , and fitting it to all the photometric points that
Topping et al. (2024). fall within the λrest in 1260 – 2500 Å range, which effec-
Magnifications are computed analytically at the po- tively traces the observed F277W-F356W color. We find
sition of each galaxy candidate and adopting its photo- that our β values span a range from −3.0 to −2.3, with
metric redshift. The magnifications and their uncertain- a median of −2.81 ± 0.16. We also show this distribu-
ties are listed in Table 1. tion in Figure 3. Finally, we derive the SFRUV directly
from our de-lensed MUV , by following the relation from
3.5. Size Measurements (Kennicutt & Evans 2012). We find that the SFRUV
The cutouts we show in Figure 2 imply that our ranges from 0.7 to 1.0 M⊙ yr−1 . Since our derived β
sources are quite compact, yet appear to be resolved. values imply negligible dust reddening, we do not apply
To measure the effective radii we model each source with a correction for dust when deriving the SFR. All the
GALFIT (Peng et al. 2002, 2010), using a Sérsic (Sérsic de-lensed physical parameters are listed in Table 1.
1963) profile where the source position, brightness, ef-
fective radius, Sérsic index, and axis ratio are allowed to
vary. When fitting, we take into account the effects of 4.2. Number Density
the PSF, which we measure empirically from the bright
We compute the number density of 16 < z < 20
stars in the field. We perform this procedure on our
sources in GLIMPSE by utilizing a 1/Vmax method
brightest band - F277W, to ensure optimal S/N per pixel
(Schmidt 1968), where Vmax corresponds to the maxi-
is achieved to accommodate robust size measurements.
mum volume a galaxy could occupy and still be detected
We find that the on-the-sky sizes of our sources, range
in the appropriate filter for our redshift range. Due to
from Reff ∼ 0.′′ 10 − 0.′′ 24. A source can be considered to
lensing, the effective area, and therefore volume, cov-
be resolved when its effective radius is larger than the
ered by our observations, is smaller than it would be in
empirical PSF half-width at half maximum (HWHM).
a blank field. In addition, an accurate derivation of the
Since the HWHM of F277W PSF is ∼0.′′ 046, we can
UV number density requires a robust estimate of the
consider all our sources to be resolved.
completeness of our survey, for a given selection func-
Taking the redshift and gravitational magnification
tion and depth. To account for the former, we use our
into account, we convert our angular sizes to physical
magnification maps and together with the p(z) for our
effective radii. These range from Reff = 250 − 550 pc.
sources (to account for the redshift range), we derive a
We list the de-lensed sizes in Table 1.
total survey volume of 17055 ± 71 Mpc3 .
4. RESULTS The second step to derive the effective volume con-
sists of computing the survey completeness through the
4.1. Stellar Population Properties lensing cluster. To do that, we use the same approach
After completing the multiple stages of selecting our adopted in Atek et al. (2018) and Chemerynska et al.
final z > 16 sample, we now compute a range of relevant (2024). Briefly, the procedure includes generating a
physical parameters. The uniqueness of our sample lies large set of mock galaxies that are distributed directly
in the unprecedented depth of the GLIMPSE observa- in the source plane, which in turn was generated from
tions, which we further push to their limit with gravi- our lensing model. The properties of the simulated ob-
tational lensing in order to obtain these candidates. On jects span the redshift, size and luminosity range of our
average, our sources have 3 individual band detections, objects. More details regarding this procedure will be
with the rest of the JWST photometry being upper lim- presented in Chemerynska et al. in prep. Combining
its. While this is perfectly adequate to constrain the red- this with our derived volume, we find an effective sur-
shift, deriving any physical parameters, especially stellar vey volume Vmax = 13714 ± 1784 Mpc3 .
 8 Kokorev et al.
.
Table 1. Properties of the GLIMPSE high-z sample† .

ID RA[deg] Dec[deg] zphot ∆BIC1 µ MUV [ABmag] SFR [M⊙ /yr] β Reff [pc]
1121 342.2361 −44.5712 15.9 ± 1.2 5.1 1.25 ± 0.01 −18.0 ± 0.2 1.0 ± 0.2 −2.8 ± 0.2 310 ± 19
1260 342.2358 −44.5708 18.6 ± 1.3 11.1 1.25 ± 0.01 −18.0 ± 0.3 1.0 ± 0.3 −2.8 ± 0.7 248 ± 18
4390 342.2162 −44.5633 17.6 ± 1.1 10.4 1.40 ± 0.02 −18.0 ± 0.2 0.9 ± 0.2 −2.8 ± 0.5 550 ± 297
47861 342.1888 −44.5525 17.6 ± 1.3 9.3 1.96 ± 0.06 −17.7 ± 0.2 0.7 ± 0.2 −2.7 ± 0.1 477 ± 251
48028 342.1946 −44.5528 17.5 ± 1.3 10.0 1.80 ± 0.04 −17.7 ± 0.2 0.7 ± 0.2 −2.3 ± 0.6 491 ± 173
24260∗ 342.2464 −44.5337 17.6 ± 1.4 7.9 1.48 ± 0.02 −17.7 ± 0.3 0.7 ± 0.2 −3.0 ± 0.7 319 ± 174
30698∗ 342.2608 −44.5379 17.6 ± 1.3 9.2 1.34 ± 0.01 −17.9 ± 0.2 0.9 ± 0.2 −2.6 ± 0.3 304 ± 61
39057∗ 342.2726 −44.543 17.6 ± 1.4 8.3 1.26 ± 0.01 −18.0 ± 0.3 1.0 ± 0.3 −3.0 ± 0.2 251 ± 35
† All values are corrected for gravitational magnification.
∗ Less secure candidates where the highest significance detection in a single band is 4σ.
1 BIC(low-z) - BIC(high-z).

1.0 16<z<20 MUV= 18±0.5

This Work, z 17
1.5 Cullen+23, z 8 16 GLIMPSE
This Work
Post-JWST Theory
2.0
Feldmann+24
2.5 Li+24
3.0 Munoz+23
Wilkins+23 Pre-JWST Theory
3.5 Behroozi+19
4.0 20 18 16 Mason+15
MUV Dayal+14
Figure 3. UV continuum slope and de-lensed lumi- 10 5 10 3
nosity. The black dashed line shows the best fitting β-MUV
relation from Cullen et al. (2023) for SFGs at z = 8 − 16.
Number Density [Mpc 3 mag 1]
Grey shaded area corresponds to the 68 % confidence in-
terval. Our secure (> 5σ) sources at z ∼ 17, shown with Figure 4. Theoretical comoving number densities of
red diamonds, appear to be consistent with the predicted z > 16 galaxies in the literature. Our completeness-
trend within 1-sigma, although systematically shifted to corrected densities are shown in red. Black lines are median
bluer slopes. We show the tentative (∼ 4σ) z > 16 galaxy values, shaded areas show 68 % percentiles. Values have been
candidates with open symbols. homogenized in terms of redshift interval (16 < z < 20) and
de-lensed. While high, our predictions generally align with
the upper edge of the theoretical predictions (Mason et al.
2015; Dayal et al. 2014; Behroozi et al. 2019; Muñoz et al.
All of our objects cover nearly the same range of MUV , 2023; Wilkins et al. 2023; Feldmann et al. 2024; Li et al.
so to derive the final number density we will just assume 2024).
a single bin with a width of 1 magnitude – MUV =
18.0 ± 0.5. The final number density for our sample
in the 16 < z < 20 range is therefore equal to log10 5.1. Stellar Populations at High Redshift
(ϕ/[Mpc−3 mag−1 ])=−3.43+0.28−0.64 .

5. DISCUSSION
 GLIMPSE Discovery of z > 16 galaxies 9

1
16<z<20
2
log10 ( /[Mpc 3 mag 1])

3
4
5
6 Our Work ( > 31.5 mag)
Feldmann+24 (FIREbox)
7 Halo Mass Function at z 17
22 18 14 10
MUV
Figure 5. Observed abundance of bright galaxies at
z > 16. We report the approximate magnitude limits for a
number of deep extragalactic surveys corresponding to the Figure 6. The abundance of bright sources at cosmic
2σ depth of the dropout filter - F200W, which is required for dawn. We trace the possible evolutionary paths of our ob-
the Lyman break identification. In varying shades of gray, jects from z ∼ 17 to some of the brightest spectroscopically
we show CEERS (Bagley et al. 2023; Finkelstein et al. 2023) confirmed high-z objects at z ≃ 11–14 (Arrabal Haro et al.
and PRIMER (both COSMOS and UDS Donnan et al. 2023) 2023; Bunker et al. 2023; Curtis-Lake et al. 2023; Carniani
combined, JADES (Origins Field, Eisenstein et al. 2023) et al. 2024a; Castellano et al. 2024). Shown are the tracks of
and NGDEEP (Leung et al. 2023; Bagley et al. 2024). The constant star-formation at 1 M⊙ /yr (purple), and a ΛCDM-
dashed line is a scaled version of the Halo Mass Function limited maximal possible accretion scenario (pink line and
that assumes a 30% star formation efficiency of the gas, no black shaded region, Dekel & Krumholz 2013), where all gas
dust, and a continuous star formation history. The dashed is converted to stars. The overabundance of bright galaxies
line is simply a toy model to illustrate how the UVLF at observed at high-z so far is fully consistent with our observa-
these redshifts could steeply decline. tions. We only show the MUV uncertainties that are larger
than the marker size.

In this section, we reflect on the implications of our

suggests that the UV emission is not dominated by un-
derived parameters if the selected sample truly resides
obscured AGN, unlike other high-z galaxies (Maiolino
at z ∼ 17.
et al. 2023; Harikane et al. 2023).
First, we compare our calculated Reff with empiri-
Using our derived SFRUV and Reff , we calculate the
cal results for spectroscopically confirmed galaxies at
UV-based star formation rate surface density, finding a
z ∼ 13 − 14. Both Carniani et al. (2024a) and Curtis-
median value of ΣSFR,UV = 0.93+0.92−0.66 M⊙ yr
−1
kpc−2 .
Lake et al. (2023) report UV sizes in the range of
This is 5 − 10 times lower than values reported for spec-
∼ 100 − 300 pc, consistent with our findings. While
troscopically confirmed objects at z ∼ 14 by Carniani
high-redshift size predictions from simulations are lim-
et al. (2024a) and is similarly lower when extrapolat-
ited, the TNG50 simulations have accurately reproduced
ing the trend in Calabrò et al. (2024). However, higher
galaxy morphologies across a wide redshift range (e.g.,
ΣSFR,UV found at z < 15 likely reflect observational
see Tacchella et al. 2019). The latest high-z size evo-
biases and small number statistics at high-z, with the
lution analyses in Costantin et al. (2023) and Mor-
bulk of the population likely containing less intense star-
ishita & Stiavelli (2023) suggests a scaling relation of
formation. Moreover, if the accretion rate changes ex-
∼ 8.66 × (1 + z)−1.15 kpc. Extrapolating this to z ∼ 17
ponentially with redshift for galaxies such as ours, even
predicts Reff ∼ 300 pc, aligning with our measurements
small changes in redshift will likely lead to significantly
within 1σ. Generally, the relatively large sizes of our
more accretion. We will explore this further in Sec-
objects may indicate that UV light from these galax-
tion 5.3.
ies arises from extended stellar populations, as observed
The UV slopes in our sample also do not exhibit un-
in Carniani et al. (2024a). Furthermore, the spatially
usual characteristics. We find a median β of −2.81, con-
resolved nature of our sources, combined with an ab-
sistent with a largely dust-free stellar plus nebular con-
sence of a PSF-like, centrally concentrated component,
10 Kokorev et al.

tinuum. This is reasonable, as our objects are observed Over the past two years, JWST has identified dozens
just ∼ 200 Myr post-Big Bang, with limited time for of bright MUV < −18 high-z galaxy candidates at
substantial dust production. Within the uncertainties of z = 10−14, challenging pre-JWST expectations (Adams
our β values, our sources align with the β-MUV luminos- et al. 2023; Austin et al. 2023; Atek et al. 2023; Curtis-
ity relation from the surveys used in Cullen et al. (2023), Lake et al. 2023; Donnan et al. 2023; Finkelstein et al.
which are based on recent JWST data for z = 8 − 16 2023; Castellano et al. 2024; Carniani et al. 2024a;
star-forming galaxies. Finally, we do not observe ex- Robertson et al. 2024; Chemerynska et al. 2024; McLeod
tremely negative beta slopes that have been postulated et al. 2024; Leung et al. 2023; Pérez-González et al.
for metal-free Population III stars (e.g. see Jaacks et al. 2023). Surprisingly, despite the large areas and depths
2018). of various surveys, none have detected bright galax-
ies beyond z > 16. Given the detection limits of
∼ 30 − 30.5 mag in the F200W/F277W filters (used
5.2. Where are the Bright High-z Galaxies? for Lyman break dropout selection at z > 15), bright
galaxies with MUV < −19 would likely have been iden-
The GLIMPSE imaging data are contained within a
tified by these surveys if they indeed existed at these red-
single NIRCam pointing, with the effective area fur-
shifts. Selecting these galaxies at z > 16 requires deep
ther limited by lensing effects. Despite the relatively
F200W (drop-out band) and F277W photometry, as the
small volume surveyed, we identify 5 robust high-z can-
break has to be unambiguous (> 1 mag). To compute
didates, all exhibiting characteristics consistent with a
the expected number densities from various extragalac-
star-forming population at this epoch. We now discuss
tic surveys, based on non-detections (Gehrels 1986) we
the number density derived from our sample, compar-
do the following. Using the quoted F200W 2σ depths
ing it with theoretical predictions from both pre- and
for CEERS (Bagley et al. 2023; Finkelstein et al. 2023),
post-JWST launch models (see Figure 4).
PRIMER (both COSMOS and UDS), JADES Origins
Regardless of the simulation type—whether semi-
Field (Eisenstein et al. 2023; Robertson et al. 2024),
analytic (Dayal et al. 2014; Mason et al. 2015; Behroozi
and NGDEEP (Leung et al. 2023; Bagley et al. 2024),
et al. 2019) or hydrodynamical (Rosdahl et al. 2022;
we assume M200 −M277 ≥ 1 mag and then compute the
Wilkins et al. 2023)—we find that our derived number
expected MUV and the corresponding number density
density at 16 < z < 20 at MUV = −18.0 ± 0.5 signif-
(from Vmax ) for each survey using the reported survey
icantly exceeds the pre-JWST theoretical expectations,
area. These are shown in Figure 5 as gray upper limits,
with a discrepancy of over 3 − 5σ. This mirrors an ob-
alongside the GLIMPSE number density at MUV = −18
served discrepancy in the UVLF suggested by spectro-
and an upper limit at MUV = −19. The discrepancy
scopically confirmed z ∼ 12 − 14 objects (Donnan et al.
between the limits at MUV < −19 and our observed
2023; Finkelstein et al. 2023; Casey et al. 2024; Carni-
number densities at MUV = −18 is quite jarring. The
ani et al. 2024a; Chemerynska et al. 2024; Harikane et al.
upper limits reach ∼ 10−6 Mpc−3 mag−1 (for CEERS
2024; Robertson et al. 2024). In contrast, post-JWST
plus PRIMER) at MUV ∼ −19, in a stark contrast to
calibrated simulations, including semi-analytic models
GLIMPSE where we predict there are ≳ 2000× more
(Muñoz et al. 2023), analytic predictions that include
galaxies per volume at MUV = −18. If MUV < −18
the feedback-free burst scenario (Li et al. 2024), and
galaxies existed at this epoch, they surely should have
FIREbox hydrodynamical simulations (Feldmann et al.
been detected by now.
2024), show better, generally within 1 − 2σ, agreement.
So where are they? One possibility is that because
We find the best agreement with the FIREbox simu-
of cosmic variance, the current deep fields have missed
lations presented in Feldmann et al. (2024). This might
some of the bright z > 15 galaxies. After all, the depths
suggest that the abundance of UV-luminous galaxies at
of GLIMPSE are comparable to JADES and NGDEEP,
z ≳ 11 (Arrabal Haro et al. 2023; Maiolino et al. 2023;
as such we would expect these surveys to find a z > 15
Curtis-Lake et al. 2023; Castellano et al. 2024; Carni-
source. In this scenario, wider-field surveys that reach
ani et al. 2024a) could be due to a relatively constant
MUV ∼ −20 mag would discover bright z > 15 galaxies.
star-formation efficiency (SFE) that is largely indepen-
For instance, objects brighter than MUV ∼ −18 are too
dent of Mhalo (Chemerynska in prep.). We will explore
rare to appear in the FIREbox simulated volume at this
the connection of our sample to these lower-z UV-bright
redshift (see Figure 5).
galaxies in the next section. Encouragingly, our sam-
Furthermore, this difference can be potentially due to
ple, while showing somewhat higher number densities
the bright end of the UVLF at z > 16 lying exactly at
at this absolute magnitude, remains broadly consistent
the GLIMPSE detection limit of MUV ∼ −18, explain-
with many post-JWST theoretical predictions.
 GLIMPSE Discovery of z > 16 galaxies 11

ing why we are seeing these objects for the first time Since we want to show the absolute limit possible, given
(though see a candidate in NGDEEP found by both Le- our observations, we start to evolve this SFH at the
ung et al. (2023) and Austin et al. (2023) at zphot ∼ 15.6, maximum redshift of our sample - z = 18.6, instead of
with MUV of ∼ −19.2, which implies a log10 (ϕ/[Mpc−3 the median. Both scenarios begin at MUV ∼ −18. To
mag−1 ])∼ −4.65). The steepness of the UVLF implied evolve these SFHs we have used the toolkit within BAG-
by the GLIMPSE detections against the brighter lim- PIPES (Carnall et al. 2019), which outputs MUV in a
its can be fit if the UVLF closely tracks the halo mass pre-defined grid of lookback-time. We show the results
function, as we show in Figure 5 (translating halo mass of our modeling, alongside the bright spectroscopically
to SFR with a toy model given by the cosmic baryon confirmed galaxies in Figure 6.
fraction, zero dust, and 30% efficiency of gas into stars We find that if our galaxies followed a constant SFH
over a fixed 100 Myr timescale). We note that the most and continued forming stars at their current rate, they
plausible way to reproduce the observed number density can easily reproduce the objects observed in Curtis-
is with a constant and relatively high SFE. Lake et al. (2023) and the fainter one of the two galax-
The lensing magnification present in AS1063, as ies presented in Carniani et al. (2024a). Reproducing
opposed to deep blank fields such as JADES and some of the brightest (MUV < −20) known galaxies at
NGDEEP, could also play a role in this discrepancy. We z ≥ 10 would, however, require very (nearly 100%) ef-
speculate that even the modest lensing magnification of ficient star-formation episodes. Following the maximal
µ ∼ 1.5, which will double the exposure time on source exponential accretion scenario, we find that galaxies in
(adding ∼ 0.5 mag in depth), will provide the additional our sample can, in fact, become as bright as JADES-
sensitivity needed to uncover these early galaxies. There GS-z14-0 (MUV = −20.81 at z = 14.17933; Carniani
simply was no feasible way to see these relatively rare et al. 2024a,b; Schouws et al. 2024), or GNz11 (although
objects prior to GLIMPSE. the stellar contribution to the MUV could be lower, due
to the potential AGN; Oesch et al. 2016; Bunker et al.
5.3. Abundance of Bright Galaxies at z ∼ 12 − 14 2023; Maiolino et al. 2023), without a need to invoke
any exotic cosmological framework. Curiously, the SFH
Our observations indicate that one possible explana-
for JADES-GS-z14-0 presented in Carniani et al. pre-
tion for the lack of MUV < −19 galaxies at z ∼ 17 is that
dicts SFR ∼ 1 M⊙ yr−1 at z ∼ 17 − 18, which matches
brighter galaxies are extremely rare at high-z, so only
our sample if it were a potential progenitor. This fur-
“faint” galaxies would be abundant enough to be found
ther reemphasizes that these z ∼ 17 candidates could
in the volume probed so far at this epoch. Therefore one
plausibly grow to become the extremely bright z ∼ 14
can imagine that the over-abundance of bright galaxies
galaxies without extraordinarily new cosmologies.
observed at z = 12–14 have likely evolved from earlier
(fainter) galaxies, such as the ones presented here, that
are surprisingly abundant for their relatively faint UV
luminosities. In return, this suggests that the discrep- 6. CONCLUSIONS
ancy between theoretical predictions and observations In this study, we leverage ultra-deep GLIMPSE NIR-
may begin as early as z = 18, potentially during the Cam imaging covering the AS1063 lensing cluster to
epoch when the first galaxies formed. identify a unique sample of high-redshift (z > 16) galaxy
To test that, we would like to see how different SFHs candidates. Using a combination of observed colors
starting at z ∼ 17 reproduce the observed bright galax- alongside an extensive suite of SED modeling routines,
ies at z < 14. To do this we consider two cases. First, we we select five robust candidates at z ≃ 15.9−18.6, which
consider a constant SFH which begins forming stars at all fit into our single NIRCam pointing.
the median redshift of our sample (z ∼ 17.6) at a rate of We examine the stellar population properties of our
∼ 1 M⊙ yr−1 (to match our derived SFRs). As the sec- sample, such as their delensed, absolute UV-brightness,
ond case, we would consider the maximum growth rate β slopes and sizes, finding that these high-z galaxies
allowable for these galaxies by forming stars at 100% of fit very well into the picture of galaxy evolution being
the accretion rate set by the Extended Press-Schechter established at z ∼ 10 − 14. Crucially, our findings re-
approximation in a ΛCDM Universe (e.g., see eq. 7 in veal a significant excess in number density of galaxies at
Dekel & Krumholz 2013, with a 0.3 dex scatter as in Ren MUV = −18 of log10 (ϕ/[Mpc−3 mag−1 ])=−3.43+0.28 −0.64 ,
et al. 2018 and Mirocha et al. 2021). This in turn cre- compared to theoretical predictions from both semi-
ates an exponentially increasing star formation history analytic and hydrodynamical simulations calibrated be-
that puts a maximum limit on the luminosity growth ex- fore the JWST launch, mirroring similar discussions for
perienced by these galaxies observed at these redshifts. bright galaxies at lower z ∼ 10−14 (e.g. see Oesch et al.
12 Kokorev et al.

2016; Carniani et al. 2024a; Finkelstein et al. 2023; Arra- ACKNOWLEDGMENTS

bal Haro et al. 2023).
Surprisingly, none of the current extragalactic JWST We thank Steven Finkelstein and Volker Bromm for
surveys have managed to successfully identify brighter helpful discussions which helped improve the quality of
objects beyond z > 16. No bright (MUV < −19) galaxy this manuscript. VK acknowledges support from the
candidates have been identified by substantially larger University of Texas at Austin Cosmic Frontier Cen-
areas covered by CEERS and PRIMER (Bagley et al. ter. HA and IC acknowledge support from CNES, fo-
2023; Finkelstein et al. 2023; Donnan et al. 2023). The cused on the JWST mission, and the Programme Na-
same is the case for moderately large and deep observa- tional Cosmology and Galaxies (PNCG) of CNRS/INSU
tions from JADES (Eisenstein et al. 2023) and NGDEEP with INP and IN2P3, co-funded by CEA and CNES.
(Leung et al. 2023; Bagley et al. 2024). What our ob- The BGU lensing group acknowledges support by grant
servations seem to indicate is that this dearth of bright No. 2020750 from the United States-Israel Binational
galaxies at high-z is driven by the shape of the UVLF Science Foundation (BSF) and grant No. 2109066 from
itself, which can be accommodated by very efficient star the United States National Science Foundation (NSF),
formation on top of the steep halo mass function, requir- and by the Israel Science Foundation Grant No. 864/23.
ing significantly fainter intrinsic observations to discover JBM acknowledges support through NSF Grants AST-
z > 16 galaxies in reasonable volumes. It is unsurpris- 2307354 and AST-2408637. ASL acknowledges support
ing, therefore, that the unique sensitivity of GLIMPSE – from Knut and Alice Wallenberg Foundation. AA ac-
achieved through a combination of lensing magnification knowledges support by the Swedish research council
and ultra-deep imaging – has managed to potentially re- Vetenskapsradet (2021-05559). This work has received
veal some of the earliest galaxies in the Universe. funding from the Swiss State Secretariat for Educa-
Finally, by evaluating various star-formation scenar- tion, Research and Innovation (SERI) under contract
ios, we demonstrate that MUV ∼ −18 galaxies at z ∼ 17 number MB22.00072, as well as from the Swiss Na-
can easily grow into the observed UV-bright galaxies tional Science Foundation (SNSF) through project grant
seen by JWST at z ∼ 10 − 14, without violating the 200020 207349. This work is based on observations
standard cosmological framework. The extreme SFE, made with the NASA/ESA/CSA James Webb Space
coupled with a steep HMF are both sufficient to explain Telescope. The data were obtained from the Mikulski
the observed trends. This connection further emphasizes Archive for Space Telescopes at the Space Telescope
the idea that faint, yet numerous, high-z galaxies play Science Institute, which is operated by the Association
a crucial role in the stellar mass assembly of galaxies in of Universities for Research in Astronomy, Inc., under
the early Universe. NASA contract NAS 5-03127 for JWST. Some of the
data products presented herein were retrieved from the
Dawn JWST Archive (DJA). DJA is an initiative of the
Cosmic Dawn Center, which is funded by the Danish
National Research Foundation under grant No. 140.

Software: BAGPIPES (Carnall et al. 2019), BEA-

GLE (Chevallard & Charlot 2016), EAZY (Brammer
et al. 2008), GALFIT (Peng et al. 2002), grizli (Bram-
mer 2023), sep (Barbary 2016), SExtractor (Bertin &
Arnouts 1996)
Facilities: JWST, HST
 GLIMPSE Discovery of z > 16 galaxies 13

APPENDIX

A. REMAINING HIGH-Z CANDIDATES

In this section, we show an additional three z > 16 candidates which did not end up in our final sample due to not
having at least one SN> 5 detection. We show these objects in Figure 7.

Figure 7. Additional z ≳ 16 candidates from GLIMPSE. Colors and symbols are the same as in Figure 2.

REFERENCES
Adams, N. J., Conselice, C. J., Ferreira, L., et al. 2023, Barbary, K. 2016, Journal of Open Source Software, 1, 58,
MNRAS, 518, 4755, doi: 10.1093/mnras/stac3347 doi: 10.21105/joss.00058
Allard, F., Hauschildt, P. H., Alexander, D. R., Tamanai, Beauchesne, B., Clément, B., Hibon, P., et al. 2024,
A., & Schweitzer, A. 2001, ApJ, 556, 357, MNRAS, 527, 3246, doi: 10.1093/mnras/stad3308
doi: 10.1086/321547 Behroozi, P., Wechsler, R. H., Hearin, A. P., & Conroy, C.
Arrabal Haro, P., Dickinson, M., Finkelstein, S. L., et al. 2019, MNRAS, 488, 3143, doi: 10.1093/mnras/stz1182
2023, Nature, 622, 707, doi: 10.1038/s41586-023-06521-7 Bergamini, P., Rosati, P., Mercurio, A., et al. 2019, A&A,
Atek, H., Richard, J., Kneib, J.-P., & Schaerer, D. 2018, 631, A130, doi: 10.1051/0004-6361/201935974
MNRAS, 479, 5184, doi: 10.1093/mnras/sty1820 Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393,
Atek, H., Chemerynska, I., Wang, B., et al. 2023, MNRAS, doi: 10.1051/aas:1996164
524, 5486, doi: 10.1093/mnras/stad1998 Blanton, M. R., & Roweis, S. 2007, AJ, 133, 734,
Austin, D., Adams, N., Conselice, C. J., et al. 2023, ApJL, doi: 10.1086/510127
952, L7, doi: 10.3847/2041-8213/ace18d Boylan-Kolchin, M. 2023, Nature Astronomy, 7, 731,
Bagley, M. B., Finkelstein, S. L., Koekemoer, A. M., et al. doi: 10.1038/s41550-023-01937-7
2023, ApJL, 946, L12, doi: 10.3847/2041-8213/acbb08 Bradley, L., Sipőcz, B., Robitaille, T., et al. 2020,
Bagley, M. B., Pirzkal, N., Finkelstein, S. L., et al. 2024, astropy/photutils: 1.0.0, 1.0.0, Zenodo,
ApJL, 965, L6, doi: 10.3847/2041-8213/ad2f31 doi: 10.5281/zenodo.4044744
Balestra, I., Vanzella, E., Rosati, P., et al. 2013, A&A, 559, Bradley, L. D., Coe, D., Brammer, G., et al. 2023, ApJ,
L9, doi: 10.1051/0004-6361/201322620 955, 13, doi: 10.3847/1538-4357/acecfe
14 Kokorev et al.

Brammer, G. 2023, grizli, 1.8.2, Zenodo, Zenodo, Donnan, C. T., McLeod, D. J., Dunlop, J. S., et al. 2023,
doi: 10.5281/zenodo.7712834 MNRAS, 518, 6011, doi: 10.1093/mnras/stac3472
Brammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, Eisenstein, D. J., Willott, C., Alberts, S., et al. 2023, arXiv
ApJ, 686, 1503, doi: 10.1086/591786 e-prints, arXiv:2306.02465,
Bressan, A., Marigo, P., Girardi, L., et al. 2012, MNRAS, doi: 10.48550/arXiv.2306.02465
427, 127, doi: 10.1111/j.1365-2966.2012.21948.x Elı́asdóttir, Á., Limousin, M., Richard, J., et al. 2007,
Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000, ArXiv e-prints. https://arxiv.org/abs/0710.5636
doi: 10.1046/j.1365-8711.2003.06897.x Endsley, R., Stark, D. P., Lyu, J., et al. 2023, MNRAS,
Bunker, A. J., Saxena, A., Cameron, A. J., et al. 2023, 520, 4609, doi: 10.1093/mnras/stad266
A&A, 677, A88, doi: 10.1051/0004-6361/202346159 Endsley, R., Stark, D. P., Whitler, L., et al. 2024, MNRAS,
Calabrò, A., Pentericci, L., Santini, P., et al. 2024, A&A, 533, 1111, doi: 10.1093/mnras/stae1857
690, A290, doi: 10.1051/0004-6361/202449768 Feldmann, R., Boylan-Kolchin, M., Bullock, J. S., et al.
Carnall, A. C., McLure, R. J., Dunlop, J. S., et al. 2019, 2024, arXiv e-prints, arXiv:2407.02674,
MNRAS, 490, 417, doi: 10.1093/mnras/stz2544 doi: 10.48550/arXiv.2407.02674
Carnall, A. C., Begley, R., McLeod, D. J., et al. 2022, Ferland, G. J., Chatzikos, M., Guzmán, F., et al. 2017,
arXiv e-prints, arXiv:2207.08778. RMxAA, 53, 385, doi: 10.48550/arXiv.1705.10877
https://arxiv.org/abs/2207.08778 Ferrara, A., Pallottini, A., & Dayal, P. 2023, MNRAS, 522,
Carniani, S., Hainline, K., D’Eugenio, F., et al. 2024a, 3986, doi: 10.1093/mnras/stad1095
Nature, 633, 318, doi: 10.1038/s41586-024-07860-9 Ferrarese, L., Côté, P., Jordán, A., et al. 2006, ApJS, 164,
Carniani, S., D’Eugenio, F., Ji, X., et al. 2024b, arXiv 334, doi: 10.1086/501350
e-prints, arXiv:2409.20533, Finkelstein, S. L., Leung, G. C. K., Bagley, M. B., et al.
doi: 10.48550/arXiv.2409.20533 2023, arXiv e-prints, arXiv:2311.04279,
Casey, C. M., Akins, H. B., Shuntov, M., et al. 2024, ApJ, doi: 10.48550/arXiv.2311.04279
965, 98, doi: 10.3847/1538-4357/ad2075 Furtak, L. J., Zitrin, A., Weaver, J. R., et al. 2023,
Castellano, M., Napolitano, L., Fontana, A., et al. 2024, MNRAS, 523, 4568, doi: 10.1093/mnras/stad1627
ApJ, 972, 143, doi: 10.3847/1538-4357/ad5f88 Gehrels, N. 1986, ApJ, 303, 336, doi: 10.1086/164079
Chabrier, G. 2003, PASP, 115, 763, doi: 10.1086/376392 Gordon, K. D., Clayton, G. C., Misselt, K. A., Landolt,
Chabrier, G., Baraffe, I., Allard, F., & Hauschildt, P. 2000, A. U., & Wolff, M. J. 2003, ApJ, 594, 279,
ApJ, 542, 464, doi: 10.1086/309513 doi: 10.1086/376774
Chemerynska, I., Atek, H., Furtak, L. J., et al. 2024, Gutkin, J., Charlot, S., & Bruzual, G. 2016, MNRAS, 462,
MNRAS, 531, 2615, doi: 10.1093/mnras/stae1260 1757, doi: 10.1093/mnras/stw1716
Chen, Y., Bressan, A., Girardi, L., et al. 2015, MNRAS, Hainline, K. N., Johnson, B. D., Robertson, B., et al. 2024,
452, 1068, doi: 10.1093/mnras/stv1281 ApJ, 964, 71, doi: 10.3847/1538-4357/ad1ee4
Chevallard, J., & Charlot, S. 2016, MNRAS, 462, 1415, Harikane, Y., Zhang, Y., Nakajima, K., et al. 2023, arXiv
doi: 10.1093/mnras/stw1756 e-prints, arXiv:2303.11946,
Costantin, L., Pérez-González, P. G., Vega-Ferrero, J., doi: 10.48550/arXiv.2303.11946
et al. 2023, ApJ, 946, 71, doi: 10.3847/1538-4357/acb926 Harikane, Y., Inoue, A. K., Ellis, R. S., et al. 2024, arXiv
Cullen, F., McLure, R. J., McLeod, D. J., et al. 2023, e-prints, arXiv:2406.18352,
MNRAS, 520, 14, doi: 10.1093/mnras/stad073 doi: 10.48550/arXiv.2406.18352
Curtis-Lake, E., Carniani, S., Cameron, A., et al. 2023, Inoue, A. K., Shimizu, I., Iwata, I., & Tanaka, M. 2014,
Nature Astronomy, 7, 622, MNRAS, 442, 1805, doi: 10.1093/mnras/stu936
doi: 10.1038/s41550-023-01918-w Jaacks, J., Finkelstein, S. L., & Bromm, V. 2018, MNRAS,
Dayal, P., Ferrara, A., Dunlop, J. S., & Pacucci, F. 2014, 475, 3883, doi: 10.1093/mnras/sty049
MNRAS, 445, 2545, doi: 10.1093/mnras/stu1848 Jeffreys, H. 1961, Theory of Probability (Oxford, England:
Dekel, A., & Krumholz, M. R. 2013, MNRAS, 432, 455, (3rd edn.; Oxford Univ. Press))
doi: 10.1093/mnras/stt480 Kassiola, A., & Kovner, I. 1993, ApJ, 417, 450,
Dekel, A., Sarkar, K. C., Birnboim, Y., Mandelker, N., & doi: 10.1086/173325
Li, Z. 2023, MNRAS, 523, 3201, Kennicutt, R. C., & Evans, N. J. 2012, ARA&A, 50, 531,
doi: 10.1093/mnras/stad1557 doi: 10.1146/annurev-astro-081811-125610
 GLIMPSE Discovery of z > 16 galaxies 15

Kokorev, V., Brammer, G., Fujimoto, S., et al. 2022, ApJS, Ren, K., Trenti, M., & Mutch, S. J. 2018, ApJ, 856, 81,
263, 38, doi: 10.3847/1538-4365/ac9909 doi: 10.3847/1538-4357/aab094
Leung, G. C. K., Bagley, M. B., Finkelstein, S. L., et al. Richard, J., Claeyssens, A., Lagattuta, D., et al. 2021,
2023, ApJL, 954, L46, doi: 10.3847/2041-8213/acf365 A&A, 646, A83, doi: 10.1051/0004-6361/202039462
Li, Z., Dekel, A., Sarkar, K. C., et al. 2024, A&A, 690, Rieke, M. J., Kelly, D. M., Misselt, K., et al. 2023, PASP,
A108, doi: 10.1051/0004-6361/202348727 135, 028001, doi: 10.1088/1538-3873/acac53
Rigby, J., Perrin, M., McElwain, M., et al. 2023, PASP,
Loeb, A., & Furlanetto, S. R. 2013, The First Galaxies in
135, 048001, doi: 10.1088/1538-3873/acb293
the Universe
Robertson, B., Johnson, B. D., Tacchella, S., et al. 2024,
Lotz, J. M., Koekemoer, A., Coe, D., et al. 2017, ApJ, 837,
ApJ, 970, 31, doi: 10.3847/1538-4357/ad463d
97, doi: 10.3847/1538-4357/837/1/97
Rosdahl, J., Blaizot, J., Katz, H., et al. 2022, MNRAS, 515,
Maiolino, R., Scholtz, J., Witstok, J., et al. 2023, arXiv 2386, doi: 10.1093/mnras/stac1942
e-prints, arXiv:2305.12492, Schmidt, M. 1968, ApJ, 151, 393, doi: 10.1086/149446
doi: 10.48550/arXiv.2305.12492 Schouws, S., Bouwens, R. J., Ormerod, K., et al. 2024,
Mason, C. A., Trenti, M., & Treu, T. 2015, ApJ, 813, 21, arXiv e-prints, arXiv:2409.20549,
doi: 10.1088/0004-637X/813/1/21 doi: 10.48550/arXiv.2409.20549
McLeod, D. J., Donnan, C. T., McLure, R. J., et al. 2024, Schwarz, G. 1978, Annals of Statistics, 6, 461
MNRAS, 527, 5004, doi: 10.1093/mnras/stad3471 Sérsic, J. L. 1963, Boletin de la Asociacion Argentina de
Meena, A. K., Zitrin, A., Jiménez-Teja, Y., et al. 2023, Astronomia La Plata Argentina, 6, 41
ApJL, 944, L6, doi: 10.3847/2041-8213/acb645 Shipley, H. V., Lange-Vagle, D., Marchesini, D., et al. 2018,
ApJS, 235, 14, doi: 10.3847/1538-4365/aaacce
Mirocha, J., La Plante, P., & Liu, A. 2021, Mon. Not. Roy.
Steidel, C. C., Giavalisco, M., Pettini, M., Dickinson, M., &
Astron. Soc., 507, 3872, doi: 10.1093/mnras/stab1871
Adelberger, K. L. 1996, ApJL, 462, L17,
Monna, A., Seitz, S., Greisel, N., et al. 2014, MNRAS, 438,
doi: 10.1086/310029
1417, doi: 10.1093/mnras/stt2284
Steinhardt, C. L., Jauzac, M., Acebron, A., et al. 2020,
Morishita, T., & Stiavelli, M. 2023, ApJL, 946, L35, ApJS, 247, 64, doi: 10.3847/1538-4365/ab75ed
doi: 10.3847/2041-8213/acbf50 Suess, K. A., Weaver, J. R., Price, S. H., et al. 2024, arXiv
Muñoz, J. B., Mirocha, J., Furlanetto, S., & Sabti, N. 2023, e-prints, arXiv:2404.13132,
MNRAS, 526, L47, doi: 10.1093/mnrasl/slad115 doi: 10.48550/arXiv.2404.13132
Naidu, R. P., Oesch, P. A., Setton, D. J., et al. 2022, arXiv Sun, F., Helton, J. M., Egami, E., et al. 2023, arXiv
e-prints, arXiv:2208.02794, e-prints, arXiv:2309.04529,
doi: 10.48550/arXiv.2208.02794 doi: 10.48550/arXiv.2309.04529
Oesch, P. A., Brammer, G., van Dokkum, P. G., et al. 2016, Tacchella, S., Diemer, B., Hernquist, L., et al. 2019,
ApJ, 819, 129, doi: 10.3847/0004-637X/819/2/129 MNRAS, 487, 5416, doi: 10.1093/mnras/stz1657
Topping, M. W., Stark, D. P., Senchyna, P., et al. 2024,
Oke, J. B. 1974, ApJS, 27, 21, doi: 10.1086/190287
MNRAS, 529, 3301, doi: 10.1093/mnras/stae682
Pacucci, F., Dayal, P., Harikane, Y., Inoue, A. K., & Loeb,
Valentino, F., Brammer, G., Gould, K. M. L., et al. 2023,
A. 2022, MNRAS, 514, L6, doi: 10.1093/mnrasl/slac035
arXiv e-prints, arXiv:2302.10936,
Pascale, M., Frye, B. L., Diego, J., et al. 2022, ApJL, 938, doi: 10.48550/arXiv.2302.10936
L6, doi: 10.3847/2041-8213/ac9316 Weaver, J. R., Cutler, S. E., Pan, R., et al. 2024, ApJS,
Pei, Y. C. 1992, ApJ, 395, 130, doi: 10.1086/171637 270, 7, doi: 10.3847/1538-4365/ad07e0
Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, Wilkins, S. M., Lovell, C. C., Irodotou, D., et al. 2023,
AJ, 124, 266, doi: 10.1086/340952 arXiv e-prints, arXiv:2305.18175,
—. 2010, AJ, 139, 2097, doi: 10.1088/0004-6256/139/6/2097 doi: 10.48550/arXiv.2305.18175
Pérez-González, P. G., Costantin, L., Langeroodi, D., et al. Zavala, J. A., Buat, V., Casey, C. M., et al. 2023, ApJL,
2023, ApJL, 951, L1, doi: 10.3847/2041-8213/acd9d0 943, L9, doi: 10.3847/2041-8213/acacfe
Zitrin, A., Fabris, A., Merten, J., et al. 2015, ApJ, 801, 44,
Polletta, M., Tajer, M., Maraschi, L., et al. 2007, ApJ, 663,
doi: 10.1088/0004-637X/801/1/44
81, doi: 10.1086/518113