Dark Energy Survey Year 3 Results: Cosmology from Cosmic Shear and Robustness to Data Calibration

TL;DR

DES Year 3 provides high-significance cosmic shear measurements over 4143 deg$^2$ with >$10^8$ galaxies, delivering precise $S_8$ constraints and exploring robustness to redshift and shear calibrations. The analysis combines Metacalibration shape measurements with a sophisticated redshift-calibration framework (SOMPZ, clustering redshifts, and shear ratios) and realistic image-simulation-based shear calibration, validated against cosmological simulations. In ΛCDM, DES Y3 finds $S_8=0.759^{+0.023}_{-0.025}$ (Fiducial) and $S_8=0.772^{+0.018}_{-0.017}$ (ΛCDM-Optimized), with a mild ~2σ tension relative to Planck; intrinsic alignments and baryonic physics are identified as the main modeling systematics limiting precision. The companion analysis shows further improvements in handling small-scale information and blending effects, and the work emphasizes the role of joint 3×2pt analyses to break degeneracies and enhance cosmological constraints for future surveys.

Abstract

This work, together with its companion paper, Secco and Samuroff et al. (2021), presents the Dark Energy Survey Year 3 cosmic shear measurements and cosmological constraints based on an analysis of over 100 million source galaxies. With the data spanning 4143 deg$^2$ on the sky, divided into four redshift bins, we produce the highest significance measurement of cosmic shear to date, with a signal-to-noise of 40. We conduct a blind analysis in the context of the $Λ$CDM model and find a 3% constraint of the clustering amplitude, $S_8\equiv σ_8 (Ω_{\rm m}/0.3)^{0.5} = 0.759^{+0.025}_{-0.023}$. A $Λ$CDM-Optimized analysis, which safely includes smaller scale information, yields a 2% precision measurement of $S_8= 0.772^{+0.018}_{-0.017}$ that is consistent with the fiducial case. The two low-redshift measurements are statistically consistent with the Planck Cosmic Microwave Background result, however, both recovered $S_8$ values are lower than the high-redshift prediction by $2.3σ$ and $2.1σ$ ($p$-values of 0.02 and 0.05), respectively. The measurements are shown to be internally consistent across redshift bins, angular scales and correlation functions. The analysis is demonstrated to be robust to calibration systematics, with the $S_8$ posterior consistent when varying the choice of redshift calibration sample, the modeling of redshift uncertainty and methodology. Similarly, we find that the corrections included to account for the blending of galaxies shifts our best-fit $S_8$ by $0.5σ$ without incurring a substantial increase in uncertainty. We examine the limiting factors for the precision of the cosmological constraints and find observational systematics to be subdominant to the modeling of astrophysics. Specifically, we identify the uncertainties in modeling baryonic effects and intrinsic alignments as the limiting systematics.

Figures (19)

• Figure 1: DES Y3 footprint showing the variation in the number density across the sky, as determined with the Metacalibration catalog (left) and the variation in mean redshift of that catalog (right). Overlaid on the left is the red outline of the Y1 footprint and on the right, the locations of the four DES Deep Fields y3-deepfields (the fourth field, COSMOS, is positioned at $\sim150$deg, outside of the DES footprint, but has been rotated here to be shown on the map). The catalog spans a final effective area of 4143 deg$^2$ with an average number density of 5.59 arcmin$^{-2}$ and a mean redshift of 0.63.
• Figure 2: Estimated redshift distributions for the weak lensing catalog, divided into four redshift bins (upper panel). Fainter lines indicate the ensemble of realisations, whose spread represents the uncertainty in their estimation while the darker, solid lines denote the mean of the ensemble. These are derived using the joint SOMPZ and WZ methodology, summarised in Section \ref{['sec:zmethod']} and detailed in y3-sompz. The stair-step appearance is an artifact of using a binned representation for the $n(z)$ and is immaterial to the cosmological results. In the lower panel, the lensing efficiency kernel (defined in equation \ref{['eqn:kernel']}) for each of the source redshift bins, following the same color scheme, demonstrates that the DES Y3 sample is most sensitive between $z=0.1-0.7$.
• Figure 3: Measured tomographic DES Y3 cosmic shear two-point correlation functions: $\xi_{+}(\theta)$ (left) and $\xi_{-}(\theta)$ (right), scaled by the angular separation, $\theta$, to emphasize differences relative to the best-fit model (upper panels). The correlation functions are measured for each redshift bin pair, indicated by the label and the error bar represents the square root of the diagonal of the analytic covariance matrix. The best-fit $\rm \Lambda CDM$ theoretical prediction from the cosmic shear-only tomographic analysis is denoted by a green line. Scales excluded from the analysis, due to their sensitivity to small-scale systematics, are shaded in light blue for the Fiducial analysis and darker blue for the $\rm \Lambda CDM$-Optimized analysis. The signal-to-noise of the measurement is 40 using all angular scales and 27 (31) using the Fiducial ($\rm \Lambda CDM$-Optimized) scale-selection. For comparison, the yellow shaded region shows the Y1 uncertainty, with a factor of $\sim\sqrt{2}$ lower signal-to-noise. The lower panels plot the fractional difference between the measurements and best-fit, $\delta\xi_{\pm}/\xi_{\pm}=(\xi_{\pm}-\xi_{\pm}^{\rm theory})/\xi_{\pm}^{\rm theory}$. We find that the $\chi^2$ per effective d.o.f of the $\rm \Lambda CDM$ model is $237.7/222.2 = 1.07$, and the $p$-value is 0.223.
• Figure 4: Impact of choices in redshift calibration on predicted cosmic shear observables. The fractional difference between the fiducial simulated signal and one with an alternative analysis choice is shown, $\delta\xi_{\pm}/\xi_{\pm}$. Plotted are predicted data vectors (i) with the purely COSMOSC-redshift sample (blue solid line) (ii) with the artificially biased spectroscopic MB-redshift sample (yellow dash-dotted line) (iii) without accounting for the redshift-mixing effects of blending (red dotted line). Fiducial ($\rm \Lambda CDM$-Optimized) scale cuts are shown as (dark) blue shaded regions. The shaded green regions represent the simulated signals corresponding to the full range of hyperrank$n(z)$ realisations described in Section \ref{['sec:hyperrank']}, and the dashed grey lines show the 5th and 95th percentiles of these simulated signals.
• Figure 5: Cosmological constraints on the clustering amplitude, $\sigma_8$, (left) and $S_8$ (right) with the matter density, $\Omega_{\rm m}$ in $\rm \Lambda CDM$. The marginalised posterior contours (inner 68% and outer 95% confidence levels) are shown for the Fiducial DES Y3 analysis in green and Planck 2018 CMB in yellow Planck2018. The black dashed contours represent the $\rm \Lambda CDM$-Optimized analysis, that preserves more small-scale information compared to the Fiducial analysis, as described in Section \ref{['sec:agg']}.