Counting faint galaxies at the universe’s star-making noon

A galaxy census is not simply a matter of counting everything in a telescope image. Bright objects are easy to find; faint ones slip through the detection process. Understanding those missing galaxies matters because their collective light helps astronomers trace how galaxies evolved.

Lei Sun and colleagues tackle that problem using the UVCANDELS survey. Their paper is available as an arXiv preprint, with its metadata reporting acceptance for publication in The Astrophysical Journal. It measures the ultraviolet luminosity function: the number of galaxies per unit cosmic volume at different ultraviolet brightnesses.

Looking at ultraviolet light through a blue filter

The researchers used Hubble’s F435W imaging across approximately 289 square arcminutes in the COSMOS and EGS survey fields. Although this filter observes blue optical light, cosmic expansion has stretched the galaxies’ originally ultraviolet emission into its wavelength range.

For galaxies at redshifts between 1.6 and 2.2, these observations sample emission around a rest-frame wavelength of 1500 angstroms. The paper places this interval within cosmic noon, broadly the peak era of cosmic star formation.

The team selected galaxies using photometric redshifts, estimates derived from their measured brightnesses across ultraviolet-to-infrared observations rather than individual spectra. They removed identified X-ray active galactic nuclei and objects meeting their stellar-selection criteria, and visually inspected bright candidates for spurious detections. That produced an initial sample of 6,254 galaxies.

But selection was only half the task. The team also needed to estimate what it had missed.

Adding fake galaxies to count real ones

The researchers inserted artificial galaxies into the actual telescope images and ran them through the same detection and measurement pipeline as the real sources. The fraction recovered provided a completeness estimate for galaxies of different brightnesses and redshifts.

These artificial objects used model galaxy spectra, an observed size–luminosity relation and an assumed distribution of light within each galaxy. This approach tests detectability under realistic image conditions, but its answer still depends on how well the artificial population represents real galaxies.

To avoid relying on the largest corrections, the analysis retained only sources above a 30% completeness threshold. The final sample contained 5,212 galaxies and reached an ultraviolet absolute magnitude of approximately −16.8. On this magnitude scale, more negative numbers mean brighter objects.

The team fitted a Schechter function, a standard description combining a rising abundance toward faint galaxies with a sharp decline at the bright end. Their statistical analysis accounted for survey completeness and magnitude uncertainties, and used Markov chain Monte Carlo sampling to estimate parameter uncertainties.

A steeper census, with an important distinction

The central result is a faint-end slope of −1.586 ± 0.042 at a median redshift of about 1.9. This parameter describes how rapidly galaxy abundance rises toward lower luminosities. A more negative slope means a steeper increase in the numbers of faint galaxies relative to brighter ones; it does not, by itself, establish that faint galaxies supply most of the ultraviolet light.

Splitting the observations into two redshift groups yielded slopes of −1.488 and −1.604. The earlier group therefore had a steeper best-fitting slope, but the difference was only about 1.4 standard deviations—not statistically significant. The characteristic magnitude also shifted toward brighter values at higher redshift, with a more suggestive but still limited significance of about 2.3 standard deviations.

A broader comparison gave stronger evidence. Combining these measurements with earlier UVCANDELS results, the authors fitted four slope measurements across redshifts 0.6–2.2. They report approximately four-standard-deviation evidence for steepening toward higher redshift. That conclusion comes from the combined analysis, not from the two new groups alone.

The light that escapes—and the light it misses

Integrating the fitted luminosity function down to magnitude −15 gave an ultraviolet luminosity density of approximately 1.82 × 10²⁶ erg s⁻¹ Hz⁻¹ Mpc⁻³. This estimate extrapolates beyond the final sample’s faint limit; it is not a direct sum of detected light down to −15.

Combining their results with published measurements, the authors modelled ultraviolet luminosity density across redshifts near zero to ten. Their fitted curve peaks at redshift 2.79 ± 0.21, earlier than the dust-corrected star-formation peak used for comparison in the paper.

That distinction matters: these ultraviolet measurements are not corrected for dust attenuation. The offset suggests an increasing importance of obscured star formation toward cosmic noon, rather than directly measuring the universe’s total star-formation history.

Other limitations remain. Photometric redshifts have errors, rare bright galaxies produce noisy counts, and completeness depends on assumed galaxy structure. Changing the artificial galaxies’ light profiles altered recovery fractions near the faint boundary, with reported relative differences reaching about 21% for one tested profile. The census sharpens the picture, but its faintest population still requires modelling.