Improved Tomographic Binning of 3 × 2 pt Lens Samples: Neural Network Classifiers and Optimal Bin Assignments

Abstract Large imaging surveys, such as the Legacy Survey of Space and Time, rely on photometric redshifts tomographic binning for 3 × 2 pt analyses that combine galaxy clustering weak lensing. In this paper, we propose a method optimizing choice lens sample galaxies. We divide CosmoDC2 Buzzard simulated catalogs into training set an application set, where is nonrepresentative in realistic way, then estimate sets. The galaxies are sorted redshift bins covering equal intervals or comoving distance, with number each bin, consider generalized extension these approaches. find distance produce highest dark energy figure merit initial choices, but bin edges can be further optimized. train neural network classifier to identify either highly likely have accurate estimates correct bin. used remove poor from sample, results compared case when none removed. classifiers able improve by ∼13% recover ∼25% loss occurs used.