Principal Component Analysis of PSF Variation in Weak Lensing Surveys

We introduce a new algorithm for interpolating measurements of the point-spread function (PSF) using stars from many exposures. The principal components of the variation in the PSF pattern from multiple exposures are used to solve for better fits for each individual exposure. These improved fits are then used to correct the weak lensing shapes. Since we expect some degree of correlation in the PSF anisotropy across exposures, using information from stars in all exposures leads to a significant gain in the accuracy of the estimated PSF. It means that in general, the accuracy of PSF reconstruction is limited not by the number density of stars per exposure, but by the stacked number density across all exposures in a given survey. This technique is applied to the 75 square degree CTIO lensing survey, and we find that the PSF variation is well described by a small number of components. There is a significant improvement in the lensing measurements: the residual stellar PSF correlations are reduced by several orders of magnitude, and the measured B-mode in the two-point correlation is consistent with zero down to 1 arcminute. We also discuss the applications of the PCA technique to future surveys. Subject headings: gravitational lensing; cosmology; large-scale structure