Assembly bias and local Primordial non-Gaussianity from DESI DR1 Quasars

TL;DR

This work tackles the challenge of constraining local Primordial non-Gaussianity from DESI quasar clustering by addressing the degeneracy between $f_{ m NL}$ and the PNG bias $b_{\phi}$, which hinges on the assembly-bias parameter $p$. By calibrating a physically motivated prior $\Pi(p)$ from IllustrisTNG and CAMELS simulations for DESI-like QSOs, the authors marginalize over $p$ to obtain a single, robust $f_{ m NL}$ constraint from DESI DR1, reporting $f_{ m NL}=-3.3\pm9.2$. The prior is shown to be resilient to selection effects and subgrid physics, with a mean $p$ around $1.4$ at $z\sim1.5$ and only weak redshift dependence. This methodology demonstrates how assembly-bias-informed priors can improve PNG inferences and will be increasingly valuable for future DESI data analyses.

Abstract

The analysis of the large-scale clustering of quasars (QSO) observed by the Dark Energy Spectroscopic Instrument (DESI) represents a promising avenue for constraining local Primordial non-Gaussianity (PNG), parameterized by $f_{\rm NL}$. The signal to be constrained is the scale-dependent bias induced in the 2-point clustering of the considered tracer sample. The resulting constraints on $f_{\rm NL}$, however, are fully degenerate with the local PNG bias parameter $b_φ$, dependent on the assembly bias parameter $p$. Using IllustrisTNG hydrodynamical simulations, we select a QSO sample reflecting the selection criteria and properties of DESI QSOs, and provide a robust prior for $p$, and thus for $b_φ$, building on the findings of Fondi et al. 2024. We find a distribution with mean $\bar{p}\simeq1.4$ with weak redshift dependence, stable to selection noise and consistent with the expected recent merger history typical of quasar-hosting halos. By comparing with the CAMELS simulations we demonstrate that this prior is robust to astrophysical assumptions and cosmic variance. Finally, applying this prior to the DESI DR1 dataset, we derive updated constraints on local PNG, obtaining $f_{\rm NL}=-3.3\pm9.2$.

Figures (6)

• Figure 1: Distributions of host halo mass (top panel) and the $p$ parameter (bottom panel) for the QSO (in blue) and control (in red) samples. Solid curves show kernel density estimates of the normalized histograms and vertical dotted lines indicate the mean. While the two samples have (by construction) very similar halo mass distributions, the quasar hosts are skewed toward larger values of $p$, with $\bar{p}=1.41$, while $\bar{p}=0.99$ for the control sample.
• Figure 2: Left: Effect of selection noise on the $p$ distribution of the QSO sample. Faint histograms show the empirical densities and solid curves the KDE; dashed vertical lines mark the mean of each distribution. Different curves correspond to different values of $r$, the fraction of quasars replaced due to selection effects. The mean $p$ value remains stable despite introducing noise into the selection process. Right: scatter of $p$ versus Eddington ratio $\chi$. Grey points show all halos, while colored points highlight the Eddington-selected subsamples ($\chi>0.01$, $\chi>0.1$ -- threshold used in the literature to define the quasar mode -- and the $N_q$ highest $\chi$).
• Figure 3: Empirical distributions of the parameter $p$ for DESI-like QSO selections in simulations. Filled histograms show the stacked samples; smooth curves are KDE estimates. Blue dashed: TNG-300; green: CAMELS-CV (cosmic-variance set); orange: CAMELS-1P (astrophysics-variation set). The vertical dotted lines mark the mean of the distributions.
• Figure 4: Left: Prior distributions of the assembly–bias parameter $p$ built from the DESI–like QSO selections in IllustrisTNG+CAMELS (combined) at the two OQE effective redshifts. The curves show KDE–smoothed distributions for $z_{\rm eff}\simeq1.9$ (orange) and $z_{\rm eff}\simeq2.1$ (blue); vertical dotted lines mark the corresponding means. Right: One–dimensional posterior for $f_{\rm NL}$ from DESI DR1 QSOs $[b_1,b_{\phi}f_{\rm NL}]$ chains with $p_w=1.6$ OQE weights. Three choices are shown: marginalizing over the $z\simeq2.1$ prior on $p$ (solid blue), fixing $p$ to the mean of that prior (dashed red), and marginalizing over the $z\simeq1.9$ prior (dotted gray).
• Figure 5: Corner plot comparing different chains and weight choices for DESI DR1 QSOs $f_{\rm NL}$ inference. Shaded/solid contours correspond to analyses run with $p_w=1.6$; while dashed contours use $p_w=1$. Blue (solid) and red (dashed) curves show chains in which we sample $[b_\phi f_{\rm NL},\,b_1]$ and obtain $f_{\rm NL}$ constraints by fixing $p$ to the prior mean within these samples. Green (solid) and purple (dashed) show chains in which $p=1.4$ is fixed during inference and $[f_{\rm NL},\,b_1]$ are sampled.