First-hand information for everyone
The James Webb Space Telescope’s deep infrared images have produced a large crop of candidate galaxies at what would be very early cosmic times, but confirming which of those candidates truly lie at redshifts beyond about 8.5 requires spectroscopy. A new study using spectroscopic follow-up from the CAPERS JWST/NIRSpec PRISM program provides the first large-scale, empirical look at how well photometric selection methods perform at these extreme distances.
The authors started with 183 galaxy candidates that had been selected from JWST/NIRCam imaging and then observed with NIRSpec PRISM to obtain secure spectroscopic redshifts. Of those, 48 sources were spectroscopically confirmed at z ≥ 8.5 — either through detection of emission lines or, in many cases, by identifying the characteristic Lyman-α break in the continuum. The highest-redshift secure confirmation in the CAPERS sample is UDS-111893 at z_spec = 12.98; no objects in the sample were confirmed above z = 13.
The study quantifies which photometric indicators are most predictive of a real high-redshift galaxy. Candidates are more likely to be genuine high-z systems when they have a high signal-to-noise ratio (SNR) in the detection filter, a strong Lyman-α break (the sharp drop in flux shortward of 1216 Å in the rest frame), and a large Δχ2 between the best-fitting low-redshift and high-redshift spectral-energy-distribution (SED) models — the last being a measure of how much better a high-z fit is than any plausible low-z alternative.
Perhaps the most notable result is that including a prior based on the ultraviolet luminosity function (UV LF) — essentially folding in our expectations for how many galaxies of a given brightness exist at each redshift — substantially improves selection efficiency. Using the UV LF prior raised the confirmation success rate from roughly 35% (a typical photometric selection baseline) to about 77%, while retaining more candidates than the very conservative SNR and Δχ2 cuts. In other words, a luminosity-based prior helps pick out high-redshift galaxies more effectively than simple signal-to-noise or fit-quality thresholds alone.
This finding is useful for surveys that must balance purity (the fraction of true high-z objects among candidates) and completeness (the fraction of real high-z objects that are selected). A higher selection efficiency means fewer wasted spectroscopic follow-ups on contaminants, which is important given the limited time available on instruments like NIRSpec.
The authors are careful to point out trade-offs and limitations. Any prior — including one based on the UV LF — introduces model dependence. If the assumed UV LF is inaccurate at the earliest times, using it as a prior could bias the selected sample and downstream measurements that depend on the same data (for example, re-estimating the UV LF). Spectroscopic confirmation itself can be challenging: some genuine high-redshift galaxies have faint or no emission lines and rely on detecting the continuum break, which becomes harder for faint objects or lower SNR spectra.
More broadly, the work demonstrates the importance of spectroscopic follow-up for translating JWST’s rich photometric catalogs into robust constraints on galaxy formation in the first few hundred million years after the Big Bang. The CAPERS results give a quantitative benchmark for photo-z selections at z ≥ 8.5 and suggest practical improvements for prioritizing candidates for spectroscopy.
Looking ahead, expanding spectroscopic samples and refining priors with independent datasets will help tighten the picture. For now, these results offer a pragmatic path: using informed priors can greatly increase the yield of true high-redshift galaxies from JWST imaging, but researchers must remain cautious about the priors’ assumptions and the remaining incompleteness and biases in any selected sample.