Exploring Cosmic Dawn with PANORAMIC II: Cosmic Variance and Galaxy Clustering at $z\sim10$

TL;DR

This study delivers the first direct measurement of cosmic variance for UV-bright galaxies at $z\sim10$ using PANORAMIC along 34 independent sightlines, revealing substantial clustering and a high apparent galaxy bias when interpreted within the halo framework. By combining bootstrapping and Bayesian forward modeling, the authors derive $\sigma_{\rm CV}$ values that, while large, align with UniverseMachine predictions for clustering but show a deficiency in the predicted mean counts, indicating more UV-bright galaxies than some models produce. They test a suite of simple UniverseMachine-based models that modify UV luminosity, star-formation efficiency, and the scatter in the UV–halo mass relation, finding that models with halo-mass–dependent SFE or reduced mass-to-light ratios fit better than global UV boosts or high UV-scatter, though none perfectly match all UV limits yet. The results demonstrate that joint UVLF and clustering constraints can distinguish competing physical mechanisms for early galaxy growth, and they emphasize the importance of future wide-area pure-parallel JWST imaging to tighten these constraints and illuminate the physics of cosmic dawn.

Abstract

Observational campaigns with JWST have revealed a higher-than-expected abundance of UV-bright galaxies at $z\gtrsim10$, with various proposed theoretical explanations. A powerful complementary constraint to break degeneracies between different models is galaxy clustering. In this paper, we combine PANORAMIC pure parallel and legacy imaging along 34 independent sightlines to measure the cosmic variance ($σ_{\rm CV}$) in the number count of Lyman break galaxies at $z\sim10$ which is directly related to their clustering strength. We find $σ_{\rm CV}=0.96^{+0.20}_{-0.18}$, $1.46^{+0.54}_{-0.44}$, and $1.71^{+0.72}_{-0.59}$ per NIRCam pointing ($\sim9.7\,{\rm arcmin}^2$, $\lesssim1.5\,{\rm pMpc}$ at $z\sim10$) for galaxies with M$_{\rm UV}<-19.5$, $-20$, and $-20.5$. Comparing to galaxies in the UniverseMachine, we find that $σ_{\rm CV}$ is consistent with our measurements, but that the number densities are a factor $\gtrsim5$ lower. We implement simple models in the UniverseMachine that represent different physical mechanisms to enhance the number density of UV-bright galaxies. All models decrease $σ_{\rm CV}$ by placing galaxies at fixed M$_{\rm UV}$ in lower mass halos, but they do so to varying degrees. Combined constraints on $σ_{\rm CV}$ and the UVLF tentatively disfavor models that globally increase the star formation efficiency (SFE) or the scatter in the M$_{\rm UV}$-$M_{\rm halo}$ relation, while models that decrease the mass-to-light ratio, or assume a power-law scaling of the SFE with $M_{\rm halo}$ agree better with the data. We show that with sufficient additional independent sightlines, robust discrimination between models is possible, paving the way for powerful constraints on the physics of early galaxy evolution through NIRCam pure parallel imaging.

Figures (7)

• Figure 1: Cosmic variance $\sigma_{\rm CV}$ in the galaxy number count at $z\sim10$ for a NIRCam pointing sized survey ($9.7\,{\rm arcmin}^2$). We plot values inferred through two different methods: bootstrapping Equation \ref{['eq:cv_bootstrap']} over 34 independent fields (crosses), and MCMC-fitting to the distribution of number counts per field (stars). The secondary y-axis shows the galaxy bias inferred by cosmic variance, as defined in Equation \ref{['eq:bias_cv']}. Measurements are shown for three M$_{\rm UV}$ limits, M$_{\rm UV}<-19.5$, $-20$, and $-20.5$ with different markers being slightly displaced on the x-axis for better visual separation. For comparison, we show values extracted using the same MCMC-fitting method from the UniverseMachine (green dots).
• Figure 2: Combined constraints on the abundance of galaxies at $z\sim10$, quantified as the mean number of galaxies per field (i.e. per NIRCam pointing), $\mu$, and their clustering, quantified by the cosmic variance $\sigma_{\rm CV}$. Each row of panels corresponds to a simple model implemented in the UniverseMachine to represent a class of models invoked to explain the abundance of UV-bright galaxies at $z\sim10$. From left to right, panels correspond to different M$_{\rm UV}$ limits as indicated in the top row of panels. The red stars and gray contours show our measurements, and the 1, and 2$\sigma$ confidence regions obtained through MCMC-fitting. We plot different realizations of each model by gradually increasing the quantity of interest: $\Delta {\rm M_{UV}}$, the boost to the UV-luminosity of all galaxies (in mag); the SFE (defined as $M_*/(f_b\,M_{\rm halo})$); and $\sigma_{\rm UV}$, the scatter in the M$_{\rm UV}$-M$_{\rm halo}$ relation (in mag). This illustrates that different models follow similar tracks on the $\sigma_{\rm CV}$-$\mu$-plane, showing an inverse relationship between $\mu$ and $\sigma_{\rm CV}$. Models can however be distinguished by considering different M$_{\rm UV}$ limits simultaneously (see Section \ref{['sec:model_tension']}).
• Figure 3: Same as Figure \ref{['fig:toy_models1']}, but for three models where the model parameters depend on halo mass. In the first one (top, A2), the UV-luminosity is boosted by $\Delta \rm M_{UV}$ in halos above a threshold mass. If the threshold mass is low enough ($10^{10}\,{\rm M_\odot}$ or lower), this produces results that are nearly identical to a global boost (Figure \ref{['fig:toy_models1']}, A1). If the threshold mass is higher $(>10^{10.5}\rm M_\odot)$, there are not enough halos of that mass to reproduce the measured number density of M$_{\rm UV}<-19.5$ galaxies. The second model (middle, B2) implements two power-law scalings of the SFE with halo mass, assuming two different slopes (0.5 and 0.6), akin to the density modulated SFE scenario from Somerville2025. Both produce models consistent with the data within 1$\sigma$, with the steeper slope implying a higher SFE$_{\rm peak}$. The third model (bottom, C2) shows a scaling of $\sigma_{\rm UV}$ with log($M_{\rm halo}$), following Sun2023. Similar to the global increase in $\sigma_{\rm UV}$, this is inconsistent with the data at $>1\sigma$ for each M$_{\rm UV}$ limit individually.
• Figure 4: Combined model tension, calculated as the equivalent posterior distance in Gaussian units, across three distinct M$_{\rm UV}$ bins ($-20<{\rm M_{UV}}<-19.5$, $-20.5<{\rm M_{UV}}<-20$, and ${\rm M_{UV}}<-20.5$), for the different models presented in Section \ref{['sec:toy_model_comparison']}. The secondary y-axis on the right shows the combined model tension relative to the best-fitting model (sharp DMSFE), indicated as the horizontal dashed line. This shows that models that enhance the UV-scatter $\sigma_{\rm UV}$ (bursty SF) are disfavored at $\sim2\sigma$. The global increase in SFE is disfavored at $>2\sigma$, consistent with our qualitative assessment in Section \ref{['sec:toy_models_global']}. As can be clearly seen from Figure \ref{['fig:toy_models2']}, only boosting M$_{\rm UV}$ in halos with $M_{\rm halo}>10^{10.5}\,{\rm M_\odot}$ cannot reproduce the measured galaxy abundance, resulting in a highly disfavored model ($\Delta\sigma>4$).
• Figure 5: Galaxy bias $b_{g,\,\rm CV}$, as defined in Equation \ref{['eq:bias_cv']}, as a function of the field side length in arcmin for M$_{\rm UV}<-19.5$, -20, and -20.5, measured from the UniverseMachine at $z\sim10$. The secondary x-axis corresponds to the physical scale for the respective field side length in co-moving Mpc at $z=10$. Linear bias values are shown as horizontal dashed lines, and the shaded regions are 1$\sigma$ uncertainties respectively. This illustrates how the measured bias increases for small field sizes and exceeds the linear bias, driven by an increasing contribution from clustering on small, non-linear scales. If approximated by a square, a NIRCam pointing corresponds to a field side length of $\sim3.1\,$arcmin which is shown as a vertical dotted line where the bias deviates strongly from the linear value and shows a clear dependence on M$_{\rm UV}$.