Comparing cosmic shear nulling methods for Stage-IV surveys

TL;DR

This paper compares three nulling strategies for cosmic shear data to reduce baryon feedback bias, evaluating their effectiveness in improving cosmological parameter estimates for Stage-IV surveys.

Contribution

It introduces and assesses three nulling methods—BNT, LUnul, and cross-correlation—for bias mitigation in cosmic shear analyses, highlighting their relative strengths and trade-offs.

Findings

All three methods effectively reduce bias in $S_8$ and Dark Energy parameters.

Cross-correlation requires additional galaxy clustering data for bias reduction.

LUnul reduces bias most efficiently but at the cost of decreased precision.

Abstract

We present an analysis comparing nulling strategies for reducing the impact of baryon feedback on cosmic shear measurements. We consider three different approaches which aim to `null' the high-$k$ modes using transformations applied to the data vector: the Bernardeau-Nishimichi-Taruya (BNT) transform which operates on the lensing field, a new implementation of an LU factorisation of the discretized Limber integral (LUnul) which operates on the lensing two-point statistics, and finally a method which uses a correlated LSS tracers to suppress contributions from lower redshifts (cross-correlation). We compare these methods to un-nulled (or standard) cosmic shear at the data vector level and assess whether these methods are able to reduce the bias on cosmological constraints using a Fisher forecast. We find that the nulling techniques considered can have a large impact on reducing the bias on $S_8$ and Dark Energy parameters. The cross-correlation method is effective at reducing biases in $S_8$, but requires additional information from galaxy clustering. The LUnul method is the most aggressive of the methods and hence reduces biases most efficiently as $k_{\rm max}$ is increased, although this improvement in accuracy comes at the cost of precision. The BNT approach preserves more information than LUnul, and has a more rigorous theoretical grounding. We demonstrate that all three of these methods are effective at mitigating bias, and can be readily applied in forthcoming lensing analyses.