The JWST EXCELS survey: The ages and abundances of $3<z<5$ massive quiescent galaxies show that downsizing was already in place by $z\simeq4$

TL;DR

This paper utilizes deep, medium-resolution JWST/NIRSpec spectra combined with HST+JWST photometry to study 12 robustly fitting massive quiescent galaxies at 3<z<5, constraining their star-formation histories and metallicities. The authors clearly detect an archaeological downsizing trend: more massive galaxies form their stellar mass earlier, with a slope of about 1.5 Gyr per dex in mass, consistent with lower-redshift results, indicating this pattern was already established by z≈4. They report rapid assembly (τ10-90 ≈ 0.2 Gyr) and generally high metallicities, though a handful of low-metallicity outliers suggest either divergent evolutionary pathways or limitations in current stellar models at young ages. A central finding is that inferring detailed α-element abundances remains highly model- and data-dependent, requiring higher SNR and broader wavelength baselines for robust interpretation. Overall, EXCELS provides critical constraints on early massive galaxy formation and quenching, while highlighting the need for improved abundance-model tools in the high-redshift regime.

Abstract

We present deep, medium-resolution $λ=1-5\,μ$m JWST/NIRSpec spectroscopy for 14 quiescent galaxies at $3<z<5$ with $\log_{10}(M_*/\mathrm{M_\odot}){\,>\,}10$, obtained as part of the EXCELS survey. We perform a complete re-reduction of these data, including a custom optimal-extraction approach to combat the spectral "wiggles" that result from undersampling of the NIRSpec spatial PSF. We constrain the star-formation histories and stellar metallicities of these objects via full-spectral fitting, finding a clear stellar age vs stellar mass correlation, in which more massive galaxies assembled their stellar mass at earlier times. This confirms spectroscopically that the archaeological "downsizing" trend was already in place by $z\simeq4$. The slope of our measured relation ($\simeq1.5$ Gyr per dex in stellar mass) is consistent with literature results at $0 < z < 3$. We do not observe objects with $\log_{10}(M_*/\mathrm{M_\odot})\lesssim10.5$ and ages of more than a few hundred Myr at this epoch, suggesting that recently reported examples of higher-redshift quiescent galaxies at these masses are likely to soon rejuvenate. We measure relatively high stellar metallicities for the majority of our sample, consistent with similar objects at $0 < z < 3$. Finally, we explore evidence for $α$-enhancement in six older and more luminous galaxies within our sample, finding considerable disagreements in the chemical abundances measured using different stellar population models, different fitted rest-frame wavelength ranges, star-formation history models and fitting codes. We therefore conclude that inferring detailed stellar chemical abundances for the earliest quiescent galaxies remains challenging, and higher signal-to-noise spectra are required (SNR per resolution element $>100$ for $R\simeq1000$).

Figures (10)

• Figure 1: A comparison of different optimal extraction methods applied to our NIRSpec data (see Section \ref{['sec:wiggles']}). Our custom wavelength-varying 1D optimal extraction is shown in blue; the more-conventional fixed-kernel method is shown in black. The top panel shows the 1D extracted spectrum from the G395M grating for PRIMER-EXCELS-55410 (ZF-UDS-7329), extracted using the two methods. The second panel shows the ratio between the extracted spectrum and the spectrum obtained from simply summing the central 5 rows of the 2D spectrum (roughly the projected size of one shutter). The third panel shows the level 3 pipeline output 2D spectrum (after drizzle resampling), and the bottom panel shows the level 2 2D spectrum for one of the three nod positions (before drizzle resampling). The pixel containing the object centroid is marked with a blue line in the bottom panel. In the top two panels we also colour with alternating grey and white bands wavelength ranges for which the centroid of the trace falls on the same pixel row in the bottom panel. Periodic fluctuations at a $\simeq5$ per cent level can be seen in the black spectrum, with the pattern repeating every time the centroid shifts to the next pixel down in the bottom panel. Our custom optimal extraction method mitigates these fluctuations by using a wavelength-varying extraction kernel.
• Figure 2: The spectral energy distributions (SEDs) and spectra for the 14 massive $z>3$ quiescent galaxies observed as part of EXCELS. The calibrated EXCELS NIRSpec observations within the rest-frame wavelength range $3540-7350$ Å used in our primary fitting methodology are shown in blue. PRIMER NIRCam photometry within this wavelength range are shown as red dots. The posterior median model spectra fitted using Bagpipes are shown in black. Vertical blue bars mark the regions masked during fitting.
• Figure 3: Detailed Bagpipes full-spectral fitting of PRIMER-EXCELS-34495, shown as an example of the process described in Section \ref{['sec:fitting']}. Top left: A PRIMER F277W cutout image of the galaxy. The overlaid rectangles mark the MSA slit positions for the first of the three G235M nod positions, while the cross marks the extraction centroid (see Section \ref{['sec:wiggles']}) Top right: Observed 2D spectra for the three medium resolution gratings. Central right: Our 1D spectroscopic extraction within the rest-frame wavelength range $3540{-}7350$ Å (blue) and HST+JWST observed photometry (red points). Plotted on top are the fitted posterior median model spectrum (black line), the physical model spectrum (magenta line: posterior median model with the posterior median GP component subtracted) and the fitted AGN component (orange line). Lower right: Residuals (black) between our fitted model and data, along with the input observational uncertainties (dotted blue lines) scaled by our $s$ parameter. Bottom right: The fitted additive GP noise component in black, along with the same scaled input observational uncertainties. The y-axes of both the residual and noise panels have the same units as the central panel, but an expanded scale. In the central, residual and noise panels, the vertical shaded bars mark the regions masked during fitting. Bottom left: The full observed multi-band photometry (red points), described in Section \ref{['sec:photometry']}, and the corresponding model posteriors (orange patches). The orange curve shows the posterior median model spectrum and the shaded region marks the rest-frame wavelength range from $3540{-}7350$ Å.
• Figure 4: The stellar masses of our sample of 12 massive quiescent galaxies plotted against their SFRs (averaged over 100 Myr), with both quantities measured from full spectral fitting. For data points with posterior median $\log_{10}(\mathrm{SFR/M_\odot\,yr^{-1}})<-2$, we show $3\sigma$ upper limits. We also set a minimum SFR upper limit at $\log_{10}(\mathrm{SFR/M_\odot\,yr^{-1}})<-2$ (grey dashed lines) to limit the vertical dynamic range of the figure. The symbols are coloured according to their observed redshifts. We show the SFMS from Leja2022 at $z\sim2$ and $z\sim3$ with solid lines and shaded regions. We show the SFMS from Speagle2014 at $z\sim3$, $z\sim4$ and $z\sim5$ with dashed lines. We also mark the threshold $\mathrm{sSFR}=0.2/t_\mathrm{H}$ at $z\sim3$, $z\sim4$ and $z\sim5$ with dotted lines. All galaxies in our sample have SFRs at least 0.5 dex below the SFMS at the corresponding redshift, and are below the sSFR threshold in all but one case. The final object, 34495, falls $\simeq2\sigma$ above this limit.
• Figure 5: Relationships between stellar population properties for the 12 EXCELS $z>3$ massive quiescent galaxies for which we obtain robust results. Left: Age of the Universe at which 50 per cent of the stellar mass in each galaxy had formed ($t_\mathrm{form}$) as a function of stellar mass. Each galaxy is coloured according to its observed redshift ($z_\mathrm{spec}$). We fit a straight line with intrinsic scatter to the tight correlation visible, which is represented by the light blue line and shaded region (Equation \ref{['eq:age_vs_mass']}). The dotted light blue lines correspond to the $\pm1\sigma$ intrinsic scatter we measure for the best-fit relation. Right: Stellar metallicity as a function of stellar mass, coloured in the same way as the left panel. The grey star shows the inverse-variance-weighted mean of our sample, with error bars showing the standard error on the mean. The cyan star shows the inverse-variance-weighted mean excluding the 3 very low metallicity objects (see Section \ref{['sec:metallicity']}). For comparison, we also show the stellar mass-metallicity relations of passive/quiescent galaxies from Peng2015 at $z\sim0$ (light-weighted, black line) and Beverage2025 at $z\sim0$ (red line), $z\sim0.7$ (pink line) and $1<z<3$ (blue line).