J-PAS: forecast on the primordial power spectrum reconstruction

TL;DR

This study forecasts J-PAS's capability to reconstruct the primordial power spectrum with a non-parametric Bayesian method based on linearly interpolated knots in the log k–log P_R(k) plane, tested on a local oscillatory feature template within k in [0.02,0.2] h Mpc^-1. By jointly sampling knot positions with a minimal cosmology parameter set using PolyChord, and employing Bayes factors and a localized hypothesis test, the work assesses detectability across different J-PAS specifications. Results show that combining redshift bins and multiple tracers enables detection of oscillatory features as small as 2% relative to the power-law, while single-bin analyses require larger amplitudes (5% or more). The findings highlight J-PAS's potential to probe non-standard inflationary scenarios through intermediate-scale PPS features, with room for further enhancements via non-linear modeling and expanded tracers.

Abstract

We investigate the capability of the J-PAS survey to constrain the primordial power spectrum using a non-parametric Bayesian method. Specifically, we analyze simulated power spectra generated by a local oscillatory primordial feature template motivated by non-standard inflation. The feature is placed within the range of scales where the signal-to-noise ratio is maximized, and we restrict the analysis to $k \in [0.02,0.2] \text{ h} \text{ Mpc}^{-1}$, set by the expected J-PAS coverage and the onset of non-linear effects. Each primordial power spectrum is reconstructed by linearly interpolating $N$ knots in the $\{\log k, \log P_{\mathcal{R}}(k)\}$ plane, which are sampled jointly with the cosmological parameters $\{H_0,Ω_b h^2, Ω_c h^2\}$ using PolyChord. To test the primordial features, we apply two statistical tools: the Bayes factor and a hypothesis test that localizes the scales where features are detected. We assess the recovery under different J-PAS specifications, including redshift binning, tracer type, survey area, and filter strategy. Our results show that combining redshift bins and tracers allows the detection of oscillatory features as small as 2\%.

Figures (7)

• Figure 1: Galaxy number densities $n$ (left) and median photometric redshift errors $\delta_z$ (right) from the 2024 IDR. The densities are derived with the CBM and the photometric redshift errors with LePhare. The values are shown for different tracers and tray configurations. Note that, for the four lowest $z$-bins of red galaxies, the 3-tray configuration yields higher galaxy densities than the 5-tray setup, due to differences in the galaxy samples used to estimate them.
• Figure 2: Signal-to-noise ratios $S/N$ of the J-PAS galaxy power spectra across redshift bins. Left panels show results for blue galaxies; right panels for red galaxies. Top panels correspond to the T12345 tray strategy; bottom panels to the T125 strategy. All results assume a primordial power-law model and a survey area of $8500\,\mathrm{deg}^2$.
• Figure 3: Top left panel: signal-to-noise ratio $S/N$ for the cosmic variance component (dashed) and the total covariance (solid). Top right panel: comparison of $S/N$ with (solid) and without (dashed) photometric redshift errors. We plot three representative redshift bins corresponding to low ($z = 0.1$), intermediate ($z = 0.4$), and high ($z = 0.9$) redshift. These top panels correspond to blue galaxies with $8500 \deg^2$ under the T12345 strategy. Bottom left panel: effective number density of red galaxies relative to the blue ones. Bottom right panel: $S/N$ comparison of different tracers and tray strategies at the $z = 0.4$ bin.
• Figure 4: Local oscillatory feature template (red), exhibiting an oscillation of 10% amplitude w.r.t. the power law (blue). The values of the feature parameters are given in the table inside the plot (see appendix of MethodologicalPaper).
• Figure 5: Top: contours of the reconstructed primordial power spectrum $P_{\mathcal{R}}(k)$ for blue galaxies at the $z = 0.4$ bin, using the T12345 strategy over $8500\,\text{deg}^2$. The input feature is a 10% local oscillatory feature (LO 10%, black dashed line). The magenta contours show the $N = 2$ power-law reconstruction; the blue contours represent the reconstruction marginalized over $N$. Bottom left: power of the hypothesis test. 72% of scales exceed the 0.5 "hint of detection" threshold, and 32% exceed the 0.9 "detection" threshold. Bottom right: corresponding evidences $Z_N$ as a function of the number of knots $N$. The evidences are normalized to have their maximum value equal to 1.