Methods for robustly measuring the minimum spanning tree and other field level statistics from galaxy surveys

TL;DR

This paper introduces a novel Monte Carlo subsampling and jittering method to mitigate survey systematic effects in galaxy surveys, enabling more efficient and direct measurement of field level statistics like the MST for cosmological parameter inference.

Contribution

The authors propose a new inversion technique that reduces reliance on computationally intensive forward modelling by addressing survey systematics through Monte Carlo subsampling and jittering.

Findings

Jittering effectively masks small scale uncertainties and survey systematics.

Monte Carlo subsampling removes survey selection effects.

Method simplifies measurement of field level statistics from galaxy surveys.

Abstract

Field level statistics, such as the minimum spanning tree (MST), have been shown to be a promising tool for parameter inference in cosmology. However, applications to real galaxy surveys are challenging, due to the presence of small scale systematic effects and non-trivial survey selection functions. Since many field level statistics are 'hard-wired', the common practice is to forward model survey systematic effects to synthetic galaxy catalogues. However, this can be computationally demanding and produces results that are a product of cosmology and systematic effects, making it difficult to directly compare results from different experiments. We introduce a method for inverting survey systematic effects through a Monte Carlo subsampling technique where galaxies are assigned probabilities based on their galaxy weight and survey selection functions. Small scale systematic effects are mitigated through the addition of a point-process smoothing technique called jittering. The inversion technique removes the requirement for a computational and labour intensive forward modelling pipeline for parameter inference. We demonstrate that jittering can mask small scale theoretical uncertainties and survey systematic effects like fibre collisions and we show that Monte Carlo subsampling can remove the effects of survey selection functions. We outline how to measure field level statistics from future surveys.