Extracting intrinsic alignments in the Dark Energy Survey's year 1 data, using the self-calibration method and LSST-DESC tools

TL;DR

This work delivers the first implementation of intrinsic-alignment self-calibration within the LSST-DESC TXPipe framework and applies it to DES Year 1 data. By combining two-point weak-lensing observables with photometric redshift information, the authors isolate the Ig signal and connect it to the IG term through a scaling relation that depends on the photo-z quality and redshift distributions. The study finds indications of IA in higher-redshift DES Y1 bins, but outcomes are strongly influenced by how redshift PDFs are reconstructed and by uncertainties in galaxy bias and photo-z, limiting robust conclusions. The results demonstrate the feasibility and challenges of SCIA in current photometric surveys and chart a path for applying this approach to upcoming LSST data, where per-galaxy PDFs and better bias constraints will be crucial for reliable intrinsic-alignment mitigation.

Abstract

We present the implementation of a Self-Calibration of Intrinsic Alignments of galaxies as an extension to the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) Dark Energy Science Collaboration (DESC)'s weak lensing 3x2pt pipeline (TXPipe). As a demonstration, we have run this pipeline on the Dark Energy Survey (DES) year one data set. We find indications of a non-zero intrinsic alignment signal in the higher redshift bins, while in the lower bins our results look more uncertain. We believe this is caused by known issues with the individual galaxies photo-z estimation. This effect is particularly harmful for the self-calibration method, since it has high requirements for reliable estimation of the photo-$z$s, and the need for individual galaxy point estimates and tomographic binning to match. We show how different methods of recreating the redshift probability distribution can affect the detection of intrinsic alignment.

Figures (11)

• Figure 1: Examples of how the "True" redshift distribution changes, depending on the choices of how we recreate the PDF. the solid lines are the unweighted recreation using $z_\text{mean}$ and $z_{mc}$, while the dash-dotted line is the same but weighed by the MetaCal response. The dashed lines are the Gaussian stacked method, and the dotted line is the Gaussian stacked method, but each galaxy weighted by the MetaCal response. The lines colored purple, marron, pink, and grey are the Gaussian estimates, but centered on the $z_\text{mode}$ rather than the means.
• Figure 2: We illustrate here the values of $\eta$ for each of our four bins. We used a Monte-Carlo integrator, and the errors reflects this. We estimated $\eta$ using 4 different PDFs. The blue is the Gaussianized method, the orange is the Metacal weighted one, the green is the unweighted PDF, and the red is using the Gaussianized method centered on the $z_\text{mode}$ . The grey band highlights the redshift span of the tomographic bin that the $\eta$ corresponds to.
• Figure 3: We here show each of the 4 bins and the results of the different $Q$'s in each bin. The blue lines are the $Q$'s for the "raw" redshift distribution. the Orange lines are the "Gaussian" distribution. The green lines are using the "Gaussian" $\eta_i$ and then assuming we have a $n_i^p$ as our redshift distribution. The full lines are the $Q\left(\theta\right)$, while the dashed lines are $Q\left(\ell\right)$.
• Figure 4: The resultant $\gamma_T$ (purple) and $\gamma_{TS}$ (cyan) corresponding to the left-hand side of Equations \ref{['eq:wgammag']} and \ref{['eq:wgammagS']} for all 4 bins. The error bars are the jackknife-estimated errors from TreeCorr. Each bin is calculated with separation $(\theta$ along the x-axis so it can be compared to the angular separations used for cosmic shear in LSSTDarkEnergyScience:2022amt. The greyed out areas are the scale cuts suggested by DESy1-3x2ptLSSTDarkEnergyScience:2022amt.
• Figure 5: (Left) The jackknife-estimated covariance and cross covariance corresponding to the correlations $\gamma_T$ and $\gamma_{TS}$. (Right) The shot-noise covariance for the $\gamma_\times$ and $\gamma_{\times S}$ terms. The upper left quadrants of each graph are the covariances for the $\gamma_{Ts / \times s}$ signals, while the lower right quadrants are the covariances for the $\gamma_{T / \times}$ signal.