Dark Energy Survey Year 6 Results: Galaxy-galaxy lensing

TL;DR

DES Year 6 delivers the most precise galaxy–galaxy lensing measurement to date, achieving a total S/N of $S/N=173$ across six lens and four source tomographic bins within the $4031\, ext{deg}^2$ footprint. Leveraging the MagLim++ lens sample and Metadetection sources, the analysis employs robust scale cuts, point-mass marginalization, and improved redshift calibration (SOMPZ+WZ+BL) to mitigate small-scale and redshift-systematic uncertainties, while enabling an independent shear-ratio validation of the lens-redshift distributions. The modeling integrates linear and nonlinear galaxy bias, HMCode2020 for nonlinear matter power, and IA/Magnification contributions (TATT; GI, II, and magnification terms), with a CosmoCov analytic covariance validated against jackknife estimates. The shear-ratio results provide geometric constraints on redshift and shear calibration, offering internal consistency checks at small scales and reinforcing the reliability of the redshift distributions used in the 3×2pt cosmology. Overall, this DES Y6 GGL measurement establishes a milestone in wide-field photometric lensing, delivering a robust, scalable framework for upcoming Stage IV surveys like LSST, Euclid, and Roman while showcasing precise control of systematics and the power of the 3×2pt approach.

Abstract

We present galaxy--galaxy lensing (GGL) measurements from the full six years of data from the Dark Energy Survey (DES Y6), covering $4031\,\mathrm{deg}^2$ and used in the DES Y6 $3\times2$pt cosmological analysis. We use the MagLim++ lens sample, containing $\sim 9$ million galaxies divided into six redshift bins, and the Metadetection source catalog, including $\sim 140$ million galaxies divided into four redshift bins. The mean tangential shear signal achieves a total signal-to-noise ratio (S/N) of $173$, corresponding to a $17\%$ improvement over DES Y3. After applying the scale cuts used in the cosmological analysis, with $R_{\min}=6\,\mathrm{Mpc}/h$ ($4\,\mathrm{Mpc}/h$) for the linear (nonlinear) galaxy-bias model, the S/N is reduced to $75$ (90). A comprehensive suite of validation tests demonstrates that the measurement is robust against observational and astrophysical systematics at the statistical precision required for the DES Y6 analysis. Although not used in the main cosmological analysis, we extract high--signal-to-noise geometric shear-ratio measurements from the galaxy--galaxy lensing signal on small angular scales. These measurements provide an internal consistency check on the photometric redshift distributions and shear calibration used in the $3\times2$pt analysis.

Figures (16)

• Figure 1: Redshift distributions for the source (top) and lens (bottom) samples, with the corresponding lensing efficiency shown in the middle panel. In all panels, solid lines indicate the mean, while shaded regions denote the standard deviation across 100 realizations of reconstructed $n(z)$, sampled from the posterior of our fiducial $3\times2$pt linear bias analysis in which lens bin 2 is excluded. For lens bin 2 (shown with unfilled lines), the $n(z)$ realizations are instead drawn from a $3\times2$pt linear bias analysis including all bins.
• Figure 2: Contributions to the tangential shear signal from the different components of the fiducial model, evaluated at the best-fit parameters of the DES Y6 3$\times$2pt analysis, shown for each lens–source bin pair. Solid curves correspond to the prediction adopting linear galaxy bias, while dashed curves show the prediction with nonlinear bias. The black curve denotes the total model prediction, and the individual contributions from intrinsic alignments, lens magnification, and their cross term are shown separately. The black points (with error bars) represent the measured tangential shear. Gray shaded regions mark the angular scales removed from the analysis: darker gray for the linear bias scale cuts and lighter gray for the nonlinear bias cuts. Lens bin 2 is fully shaded because it is excluded from the fiducial data vector. Note that this figure does not illustrate the point-mass marginalization, which is applied during likelihood evaluation through the inverse covariance matrix and therefore cannot be represented in terms of the covariance elements shown here.
• Figure 3: Tangential shear measurement for the 6 tomographic bins of MagLim++ and the 4 tomographic bins of METADETECT. We overlay the best fit model for the linear bias case (solid lines) and for the nonlinear bias case (dashed lines). Gray shaded regions mark the angular scales removed from the analysis: darker gray for the linear bias scale cuts and lighter gray for the nonlinear bias cuts. Lens bin 2 is fully shaded because it is excluded from the fiducial data vector. Note that this figure does not illustrate the point-mass marginalization, which is applied during likelihood evaluation through the inverse covariance matrix and therefore cannot be represented in terms of the covariance elements shown here.
• Figure 4: Boost factors with their uncertainty for the 6 tomographic bin of MagLim++ and the 4 bins of METADETECT. Gray shaded regions mark the angular scales removed from the analysis: darker gray for the linear bias scale cuts and lighter gray for the nonlinear bias cuts. Lens bin 2 is fully shaded because it is excluded from the fiducial data vector. Note that this figure does not illustrate the point-mass marginalization, which is applied during likelihood evaluation through the inverse covariance matrix and therefore cannot be represented in terms of the covariance elements shown here.
• Figure 5: The cross-component of shear, $\gamma_\times(\theta)$, for each lens bin across six lens bins. The error bars are computed from the Jackknife covariance. Each row represents a different source bin combination, showing the behavior of $\gamma_\times(\theta)$ as a function of angular separation $\theta$ (arcmin). Gray shaded regions mark the angular scales removed from the analysis: darker gray for the linear bias scale cuts and lighter gray for the nonlinear bias cuts. Lens bin 2 is fully shaded because it is excluded from the fiducial data vector.