The Lumina Project: The Demographics of Active Galactic Nuclei from Quasars to Little Red Dots at $z\geq 3$

Xuejian Shen, Oliver Zier, Aaron Smith, Rongrong Liu, Rahul Kannan, Teodora-Elena Bulichi, Sonja M. Koehler, Volker Springel, Mark Vogelsberger, Lars Hernquist, Rohan P. Naidu, Anna de Graaff, Elia Pizzati, David M. Alexander, Luis C. Ho, Vasily Kokorev, Gene Leung, Anna-Christina Eilers, Ryan C. Hickox

First listed 2026-05-26 | Last updated 2026-05-22

Abstract

High-redshift active galactic nuclei (AGN) serve as powerful probes of early black-hole growth, galaxy formation, and the evolving intergalactic medium (IGM). In this work, we use Lumina, a cosmological radiation-hydrodynamic simulation spanning the epochs of hydrogen and helium reionization, which combines a large $(500\,{\rm cMpc})^3$ volume with $2\times 6000^3$ resolution elements, to explore high-redshift AGN. The simulation self-consistently follows hundreds of millions of galaxies and supermassive black holes (SMBHs), together with their impact on the ionization and thermal state of the IGM. We exploit this uniquely large dynamic range to predict multi-band AGN luminosity functions (LFs) at $z \geq 3$, from hard X-rays to the mid-infrared. These predictions encompass both moderately luminous quasars and the faint ``Little Red Dots'' (LRDs) uncovered by JWST. We develop an empirical model that maps simulated SMBHs onto observed AGN using bolometric and extinction/absorption corrections for canonical AGN and LRDs, and in which SMBHs with $M_{\rm BH}\leq 10\,M_{\rm seed} \sim 10^{7}\,{\rm M}_{\odot}$ stay in the LRD phase with a duty cycle of $30\%$. This simple framework reproduces the observed LFs and clustering of LRDs. Meanwhile, the pre-JWST quasar LF constraints are recovered, although we find that a $\sim 0.3$ dex log-normal scatter in bolometric luminosity is required to reproduce the bright end. We place the simulated AGN population in the cosmological context by quantifying the redshift evolution of AGN and LRD number densities, and their contributions to the integrated BH mass densities. The same AGN population is the dominant driver for the HeII reionization modelled self-consistently in Lumina. This empirical AGN model paves the way for general population-synthesis models of high-redshift AGN, including LRDs, in a unified cosmological framework.

Short digest

Using the 500 cMpc Lumina radiation-hydrodynamic simulation, this paper builds a unified empirical mapping from simulated SMBHs to observed high-redshift AGN, spanning classical quasars and JWST Little Red Dots across multi-band luminosity functions at z >= 3. The main result is that a simple prescription in which BHs with M_BH <= 10^7 Msun remain in an LRD phase with a 30% duty cycle reproduces current LRD luminosity functions and clustering, while the pre-JWST quasar luminosity function is also recovered once about 0.3 dex log-normal bolometric scatter is added to populate the bright end. That puts LRDs and quasars into a single demographic framework for number densities, BH accretion, and mass buildup rather than treating them as disconnected populations. In Lumina, the same AGN population that matches these observables also dominates the simulated He II-ionizing emissivity, tying early black-hole growth directly to helium reionization.

Key figures to inspect

• Figure 6. This is the clearest one-figure summary of the paper's empirical framework: intrinsic Lumina BH luminosities receive unresolved variability scatter, then split into canonical AGN or an LRD phase based on BH mass, with separate bolometric corrections and an LRD duty cycle. It matters because the central claim of the paper is not only that Lumina has the needed dynamic range, but that this compact three-parameter mapping can connect simulated SMBHs to both quasars and LRDs in observed bands.
• Figure 8. The bolometric luminosity function at z = 4 is the core validation plot for the unified demographics claim, showing Lumina, TNG100, and Mtng against pre-JWST quasar constraints, LRD measurements, and obscured AGN from MEOW. The separate LRD contribution makes it easy to see where LRDs dominate the faint end and where canonical AGN take over, which is exactly the transition the paper is trying to model.
• Figure 14. This figure tests the model against environment rather than abundance by comparing the simulated LRD-galaxy cross-correlation to observed clustering at z = 4 and z = 6. It is especially important because the abstract explicitly claims success on LRD clustering, and the figure also translates that agreement into host-halo scale and bias information instead of stopping at luminosity-function matching.
• Figure 18. This is the cosmological payoff figure: the He II-ionizing photon production rate is decomposed by BH mass across redshift, with the He II reionization window marked directly on the plot. It shows that the same AGN population used to match quasar and LRD demographics is also the dominant driver of He II reionization in Lumina, which makes the paper's population model astrophysically consequential rather than purely phenomenological.
• Figure 19. This diagnostic directly supports one of the paper's sharpest quantitative conclusions, namely that about 0.3 dex bolometric scatter is needed to reproduce the bright end of the quasar luminosity function. Because it isolates how the scatter parameter changes the bright-end counts while leaving the faint end much less affected, it is the most efficient figure for understanding why extra unresolved variability is required in the model.