Forecasting Primordial Non-Gaussianity from UNIONS Lyman-Break Galaxies and Planck CMB lensing

TL;DR

This study forecasts constraints on local primordial non-Gaussianity, quantified by $f_{ m NL}^{\rm loc}$, from the cross-correlation between UNIONS-selected high-redshift Lyman-Break Galaxies and Planck CMB lensing. Using an MCMC approach, it jointly fits $f_{ m NL}^{\rm loc}$ and a galaxy bias parameter while modeling LBG properties, redshift distributions, and potential outliers. Key findings show baseline precision around $\sigma(f_{ m NL}^{\rm loc})\approx 34$, with significant gains to $\approx 20$ achievable after DESI spectroscopic follow-up, and similar prospects for early UNIONS samples. However, uncertainties in the clustering-redshift distribution can substantially degrade PNG constraints, underscoring the value of precise redshift calibration and robust bias modeling for fully data-driven, high-redshift PNG tests.

Abstract

Primordial non-Gaussianities (PNGs), characterized by $f_{\rm NL}^{\rm loc}$, provide a powerful window into the physics of inflation. Cross-correlating high-redshift tracer samples with the CMB lensing potential offers a particularly robust probe of PNGs, mitigating imaging systematics that typically affect large-scale measurements from tracer auto-spectra. In this context, UNIONS enables the selection of $u$-dropout high-redshift Lyman-Break Galaxies (LBGs). We perform a MCMC-based forecast to estimate the uncertainties on $f_{\rm NL}^{\rm loc}$ and on a galaxy bias parameter, which captures our uncertainty in the tracer bias. From the angular cross-power spectrum between LBGs and Planck CMB lensing, we forecast $σ(f_{\rm NL}^{\rm loc})=34$ for an idealized photometric sample of $r<24.3$ LBGs selected with a Random Forest classification algorithm from UNIONS-like $ugriz$ imaging, with a resulting surface density of $1,100$ deg$^{-2}$. This precision can be improved to $σ(f_{\rm NL}^{\rm loc})=20$ after spectroscopic follow-up with DESI, during its next phase starting in 2029, DESI-II. We test a more realistic $u$-dropout LBG selection using early UNIONS data, which yields a denser sample of $r<24.2$ objects at $1,400$ deg$^{-2}$. From this sample, covering a larger footprint and expected to have a higher large-scale galaxy bias, we forecast $σ(f_{\rm NL}^{\rm loc})=20$, with similar precision achievable after DESI spectroscopic follow-up. In addition, we perform preliminary validation of the redshift distribution using the clustering-redshift method with DESI DR1 data, confirming the calibration from deep, small-area photometric fields. However, accounting for uncertainties in the clustering-redshift distribution significantly degrades the $f_{\rm NL}^{\rm loc}$ constraining power.

Figures (5)

• Figure 1: Left: Photometric redshift distributions of the photometric LBG samples UNIONSlike_RF in red (the convolution of the UNIONSlike_RFdistribution with the DESI spectroscopic efficiency is shown in dashed and dotted lines, they are not normalized to show the impact at $z\sim 2.5$). The distribution of the samples UNIONSlike_RF_u180s, and LSSTY4like_RF_u2x180s are shown in blue and dashed cyan lines. Right: Photometric redshift distributions of the UNIONS LBG samples UNIONS_colorcut and UNIONS_PZcut in the XMM field. The sample selected with the hybrid method à-la SOM is shown for illustration. We also show the quantity $nb$ (the product between the large-scale bias and the normalized redshift distribution) resulting from our clustering-redshift calibration methods.
• Figure 2: Top panel shows the angular power spectra $C_\ell^{gg}$, $C_\ell^{\kappa g}$, and $C_\ell^{\kappa\kappa}$ (solid lines) for the baseline sample. The cosmological-only contribution is shown as dashed lines, while the noise contribution is shown as dotted lines. Bottom panel shows the signal-to-noise ratio, defined as the ratio of the cosmological-only term to the noise, for $C_\ell^{gg}$ and $C_\ell^{\kappa\kappa}$.
• Figure 3: Left: Posterior distribution of the parameters $f_{\rm NL}^{\rm loc}$ and the large-scale linear bias (expressed in terms of relative error). Right: Forecasted error on $f_{\rm NL}^{\rm loc}$ as a function of the lower multipole cut $\ell_{\rm min}$. Solid lines show constraints using only $C_{\ell}^{\kappa g}(\ell > \ell_{\rm min})$, while dashed lines correspond to the combination of $C_{\ell}^{\kappa g}$ and $C_{\ell}^{gg}(\ell > \ell_{\rm min})$. Different colors indicate different LBG number densities: 800, 1100, and 2000 LBGs per deg$^{2}$.
• Figure 4: Left: Forecasted error on $f_{\rm NL}^{\rm loc}$ as a function of the high-redshift bias prescription $\bar{b}_2(z)$. Solid lines correspond to a full rescaling of the galaxy bias by $b_0$ in Eq. \ref{['eq:two_population_bias_b0_b1b2']}. In contrast, dashed lines correspond to rescaling only the high-redshift population bias by $b_0$ in Eq. \ref{['eq:two_population_bias_b1_b0b2']}. Right: Forecasted error on $f_{\rm NL}^{\rm loc}$ for different choices of $b_{\Phi}(b)$, beyond the prescription given in Eq. \ref{['eq:bphi_b1']}.
• Figure 5: Upper panel: Bias between the mean decoupled power spectra and the theoretical power spectra. Lower panel: Bias between the dispersion of the estimators over 100 simulations and (i) the theoretical variance computed with the full mode-coupling matrix $\mathcal{M}_{\ell\ell'}$, and (ii) the approximation $\mathcal{M}_{\ell\ell'} = \delta^K_{\ell \ell'}$.