Density modulated star formation efficiency: implications for the observed abundance of ultra-violet luminous galaxies at z>10

TL;DR

This study addresses the unexpectedly high abundance and slow redshift decline of UV-luminous galaxies at z>10 observed by JWST. It introduces Density Modulated Star Formation Efficiency (DMSFE), linking cloud-scale SFE to gas surface density and embedding it within a Santa Cruz semi-analytic model calibrated to high-z sizes, while incorporating dust attenuation and bursty star formation in post-processing. The main finding is that DMSFE can boost high-z star formation by orders of magnitude depending on the dense-gas fraction f_dense, and that evolving dust attenuation (via an sSFR threshold) and halo-mass-dependent bursts further shape the UV luminosity function, producing a shallower evolution that better matches observations in concert with a realistic cosmic SFR density. The results suggest that early galaxies formed stars efficiently in dense, GMC-like environments, and that a combination of evolving density, attenuation, and stochasticity is needed to interpret JWST-era galaxy statistics; future work should explore redshift-dependent f_dense and incorporate nebular emission and IMF evolution.

Abstract

The number density of UV luminous galaxies discovered by the James Webb Space Telescope at ultra high redshift ($z \gtrsim 10$) is higher, and declines much more slowly with increasing redshift, than expected from extrapolations of lower redshift observations or pre-launch physics-based models. Most of these models assume star formation efficiencies (SFE) of only a few percent, motivated by observations of nearby galaxies. In this work, we incorporate a scaling of SFE with gas surface density (which we refer to as Density Modulated SFE; DMSFE), motivated by cloud-scale simulations and theory, into a semi-analytic cosmological model (SAM) of galaxy formation which is calibrated to match the observed rest-UV sizes of high redshift galaxies. We also model the impact of dust and bursty star formation on the SAM-predicted properties of observed galaxies. We show that with plausible values of the main parameters, such as the fraction of gas in dense clouds $f_{\rm dense}$, our new models easily reproduce or even exceed the observed galaxy number densities at $z\sim 6$-17. While no single value of $f_{\rm dense}$ is able to reproduce the very shallow observed decline of the galaxy number density at $z\gtrsim 12$, it is plausible and even expected for $f_{\rm dense}$ to have some effective dependence on cosmic time, which could bring these models into closer agreement with the data. We show that the combined effects of DMSFE, decreasing dust attenuation, and increasingly bursty star formation at earlier cosmic epochs could conspire to reproduce the observed evolution.

Figures (19)

• Figure 1: Star formation efficiency per star forming cloud (left) and cloud lifetimes (in units of the free fall time (middle), and in Myr (right)) as a function of cloud surface density. The grey shaded regions represent the star formation efficiency (left panel), lifetimes (middle), and surface density range of GMCs in local universe star forming galaxies Chevance2023. Symbols show cloud-scale star formation efficiencies (integrated over the cloud lifetime; left) and lifetimes (middle) from cloud-scale simulations by Lancaster2021, Menonfesc2024, and Kimjg2018. The solid blue lines (left and middle) show the analytic scalings for $\epsilon_{\rm *, cl}$ and cloud lifetime that are used in the SAM (see text).
• Figure 2: The (physical) half-light (or half-mass) radius of ultra-high redshift galaxies as a function of their rest-UV magnitude. Symbols show observed UV half-light radii for $9\lesssim z \lesssim 13$ galaxies from several recent JWST surveys, as indicated by the figure legend. The cyan and red lines show the median half-mass radii of galaxies in our KS and cloud-fd=0.1 models at $z=11$, and shaded areas show the 16th and 84th percentiles. The solid grey line shows the fit to the median observed galaxy UV half-light radii at $z\sim 11$ from Morishita2024, and the grey dashed lines show their quoted 1$\sigma$ dispersion around the median. The horizontal dashed and dotted lines show the physical size of one NIRCam pixel at $z=9$ and $z=13$, respectively.
• Figure 3: The gas surface density as a function of redshift for a simple toy example of a halo with mass $M_{\rm h}=10^{11}$ M$_\odot$ at the plotted redshift, with a gas radius that is 0.06 times the halo virial radius, and a gas fraction $f_{\rm gas}=0.2$, 0.5, or 1.0, where $f_{\rm gas}=m_{\rm gas}/(f_b M_{\rm h})$ is the fraction of the halo baryon budget in the cold ISM. This illustrates that the gas surface density may have been up to two orders of magnitude higher in galaxies at ultra-high redshifts, and that galaxies may cross the critical gas surface density of $\Sigma_{\rm gas, critical} \sim 2000$ M$_\odot$ pc$^{-2}$, where stellar feedback begins to become ineffective, at around $z\sim 10$.
• Figure 4: The average gas surface density of galaxies in our KS model as a function of halo mass, for several redshifts as marked on the panels. The solid line shows the median and the shaded region shows the 16th and 84th percentiles. The dashed line shows the relation at $z=6$, repeated in each panel to guide the eye. This figure shows that although the gas is predicted to be denser in early galaxies than in the local universe, the evolution is not quite as strong as that predicted by the simple toy model shown in Fig. \ref{['fig:gasdensityz']}. This is because the gas fraction is not a constant in the SAM, but is determined by star formation and gas inflows and outflows, and shows trends with both halo mass and redshift.
• Figure 5: Baseline models (no dust or enhanced bursts): Rest-UV luminosity functions at redshift $6 < z < 17$. Symbols show a compilation of observational luminosity function estimates, as specified in the figure legend. Also shown are the predictions from the KS, Y25, and KS-nowind models (without the inclusion of dust or enhanced bursts) as labeled in the figure legend (see text for details). As shown by previous works, the baseline models (KS and Y25) reproduce the observations well from $6 \lesssim z \lesssim 10$, but underproduce the observed UVLF estimates at $z\gtrsim 12$. The excess in the model predictions at bright UV luminosities at $z \sim 6$--7 is plausibly due to dust attenuation, which has not been accounted for in the model predictions shown here. The KS-nowind model, which has supernova driven winds switched off, better reproduces luminous galaxies at $z\gtrsim 12$, but overproduces lower-luminosity galaxies.