Mitigating bias from n(z) misestimation for 3x2pt measurements

Motivation

Reliance on photometric redshifts will be a major source of systematic uncertainty in cosmology from the Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST). In particular, non-representative spectroscopic training samples will make obtaining accurate photometric redshifts of faint galaxies difficult. If the redshift distribution, or n(z), of the galaxy sample is misestimated, this can lead to biased cosmology results. My previous work has laid the fround work for mitigating this bias. In Moskowitz et al. 2024, we successfully show that you can improve the photo-z estimates by augmenting non-representative spectroscopic training samples with simulated galaxies, which will improve the recovered n(z). In Moskowitz et al. 2023, we show that by using wider redshift bins at higher redshift, where the photo-z's are worse, we can extract more cosmological information by limiting poor n(z) reconstruction to only a few bins. This allows for narrower bins at low redshift, allowing us to extract more information in regimes where the n(z) is better known. We also use a neural network classifier, first introduced in Broussard & Gawiser 2021, to successfully select galaxies most likely to have accurate photo-z estimates. This current project combines all three techniques from my previous work to perform a 3x2pt measurement using the DC2 simulation to investigate how well they can mitigate the bias arising from n(z) misestimation.

Preliminary Results

The figure below show the galaxy clustering 2pt correlation function measured in each of 10 bins, where the photo-z's relied on three different training samples: a fully representative, but unrealistic, training sample; a realistically non-representative training sample; and a non-representative training sample that has been augmented. As seen in the figure, the recovered 2pt measurements for the augmented sample are closer to that of the representative sample than the unaugmented results are. This shows promise that the recovered cosmology will also be less biased!