Inverting Fisher biases for fast systematics exploration

TL;DR

This paper presents a novel application of the Fisher matrix formalism to quickly assess whether unmodelled systematic effects can explain discrepancies between different cosmological measurements.

Contribution

It introduces a method to invert the Fisher matrix process, enabling the investigation of systematic effects' potential to account for observed tensions in data.

Findings

The method accurately estimates small parameter biases caused by systematics.

It is applicable to various systematic effects, demonstrated with galaxy intrinsic alignments and baryonic feedback.

The approach provides a practical tool for bias assessment beyond the linear regime.

Abstract

Upcoming cosmological surveys will achieve increasingly precise constraints in cosmological parameter estimation. To guarantee the robustness of cosmological analyses, it is essential to account for and model systematic effects that can bias cosmological constraints, shifting the best fit parameters away from their fiducial values. It is possible to approximately infer the biases that un-modelled systematic effects might introduce in cosmological parameter estimation by means of the Fisher matrix formalism. In this paper, we introduce a new application of this formalism, where by inverting the process, we investigate whether a specific missing or mis-modelled systematic effect can explain away a given tension between two different probes or experiments. We showcase the proposed methodology by examining two representative systematics: galaxy intrinsic alignments and baryonic feedback. As the method is agnostic to the systematic effect and can be applied to a wider range of scenarios, we discuss more possible future applications. While the proposed approach is accurate in the limit of small offsets in the cosmological parameters, where the likelihood can be considered linear in both the cosmological parameters and the systematics, in practice, the region of validity depends on the systematic effect. In general, even beyond this region, the approach still provides a useful test that helps indicate the magnitude and direction of potential biases from systematic effects in data.