The impact of intrinsic alignment on current and future cosmic shear surveys

TL;DR

This work forecasts the impact of intrinsic alignments on current and future cosmic shear surveys by integrating realistic, non-Gaussian covariances into simulated likelihood analyses across multiple IA models (LA, NLA, FR, TA) and survey configurations (DES, Euclid, LSST, WFIRST). It develops a detailed IA amplitude scaling with luminosity and redshift and implements several mitigation strategies based on a nuisance-parameter template (NLA Halofit) to remove biases in nonlinear IA scenarios, while examining the sensitivity to luminosity-function and blue/red galaxy composition. The study finds that Euclid is most susceptible to IA biases due to its depth, while LSST and WFIRST benefit from deeper observations; DES shows smaller biases due to larger statistical errors. A key takeaway is that removing a modest red-galaxy fraction can substantially control IA biases with only modest information loss, and that low-z spectroscopic calibration and joint analyses could further enhance IA self-calibration in future surveys.

Abstract

Intrinsic alignment (IA) of source galaxies is one of the major astrophysical systematics for ongoing and future weak lensing surveys. This paper presents the first forecasts of the impact of IA on cosmic shear measurements for current and future surveys (DES, Euclid, LSST, WFIRST) using simulated likelihood analyses and realistic covariances that include higher-order moments of the density field in the computation. We consider a range of possible IA scenarios and test mitigation schemes, which parameterize IA by the fraction of red galaxies, normalization, luminosity and redshift dependence of the IA signal (for a subset we consider joint IA and photo-z uncertainties). Compared to previous studies we find smaller biases in time-dependent dark energy models if IA is ignored in the analysis; the amplitude and significance of these biases vary as a function of survey properties (depth, statistical uncertainties), luminosity function, and IA scenario: Due to its small statistical errors and relatively shallow observing strategy Euclid is significantly impacted by IA. LSST and WFIRST benefit from their increased survey depth, while the larger statistical errors for DES decrease IA's relative impact on cosmological parameters. The proposed IA mitigation scheme removes parameter biases due to IA for DES, LSST, and WFIRST even if the shape of the IA power spectrum is only poorly known; successful IA mitigation for Euclid requires more prior information. We explore several alternative IA mitigation strategies for Euclid; in the absence of alignment of blue galaxies we recommend the exclusion of red (IA contaminated) galaxies in cosmic shear analyses. We find that even a reduction of 20% in the number density of galaxies only leads to a 4-10% loss in cosmological constraining power.

Figures (8)

• Figure 1: Normalized redshift distributions for the four different surveys. Top panel: DES and Euclid. Bottom panel: LSST and WFIRST.
• Figure 2: Fraction of red galaxies computed from GAMA and DEEP2 luminosity function, respectively. We consider a deep and a shallow survey with limiting magnitude of 27.5 (LSST/WFIRST) and 24.5 (Euclid/DES), respectively.
• Figure 3: The impact of IA on WL constraints ($68\%$ confidence region) from LSST assuming the NLA Halofit scenario. We consider different luminosity functions i.e. GAMA (red/dashed) and DEEP2 (green/long-dashed) and for the GAMA LF we also consider the case for which blue galaxies have a mild NLA IA contribution (blue/dot-dashed). The LSST statistical errors are shown in black/solid. Orange/dot-long-dashed contours show results when using the most extreme of these cases, i.e. the data vector corresponding to the blue contours, as input and including the CosmoLike IA mitigation module in the analysis. The marginalized likelihood is obtained by integrating over a 11-dimensional nuisance parameter space (see text for details).
• Figure 4: Top: Marginalized WL constrains ($68\%$ confidence region) from LSST when marginalizing over Gaussian photo-z uncertainties (red/dashed) and joint uncertainties of photo-z's and the fiducial IA NLA Halofit model. We assume two different levels of photo-z errors resembling optimistic and pessimistic LSST photo-z errors. The black/solid lines again show the LSST statistical errors for comparison. Bottom: The marginalized one-dimensional posterior probabilities for the 12 nuisance parameters used in the optimistic and pessimistic joint photo-z and IA analysis.
• Figure 5: Top: WL constraints ($68\%$ confidence region) from LSST when assuming different models in the data vector, but assuming the fiducial IA NLA Halofit model in the marginalization. The black/solid lines show results from a data vector contaminated by the NLA Coyote Universe model, which is very close to the fiducial NLA Halofit scenario. Red/dashed corresponds to the linear power spectrum model, blue/dot-dashed to the freeze-in model, green/long-dashed to the tidal alignment model. All models assume the GAMA luminosity function. Bottom: The marginalized one-dimensional posterior probabilities for the ten nuisance parameters describing the IA uncertainty.