Intrinsic galaxy alignments constitute the major astrophysical systematic of forthcoming weak gravitational lensing surveys but also yield unique insights into galaxy formation and evolution. We build analytic models for the distribution of galaxy shapes based on halo properties extracted from the Millennium Simulation, differentiating between early- and late-type galaxies as well as central galaxies and satellites. The resulting ellipticity correlations are investigated for their physical properties and compared to a suite of current observations. The best-faring model is then used to predict the intrinsic alignment contamination of planned weak lensing surveys. We find that late-type galaxy models generally have weak intrinsic ellipticity correlations, marginally increasing towards smaller galaxy separation and higher redshift. The signal for early-type models at fixed halo mass strongly increases by three orders of magnitude over two decades in galaxy separation, and by one order of magnitude from z=0 to z=2. The intrinsic alignment strength also depends strongly on halo mass, but not on galaxy luminosity at fixed mass, or galaxy number density in the environment. We identify models that are in good agreement with all observational data, except that all models over-predict alignments of faint early-type galaxies. The best model yields an intrinsic alignment contamination of a Euclid-like survey between 0.5-10% at z>0.6 and on angular scales larger than a few arcminutes. Cutting 20% of red foreground galaxies using observer-frame colours can suppress this contamination by up to a factor of two.
Download