WHAT LIES BENEATH: USING p ( z ) TO REDUCE SYSTEMATIC PHOTOMETRIC REDSHIFT ERRORS

We use simulations to demonstrate that photometric redshift "errors" can be greatly reduced by using the photometric redshift probability distribution p ( z ) rather than a one-point estimate such as the most likely redshift. In principle, this involves tracking a large array of numbers rather than a single number for each galaxy. We introduce a very simple estimator that requires tracking only a single number for each galaxy, while retaining the systematic-error-reducing properties of using the full p ( z ) and requiring only very minor modifications to existing photometric redshift codes. We find that using this redshift estimator (or using the full p ( z )) can substantially reduce systematics in dark energy parameter estimation from weak lensing, at no cost to the survey.