Multi-tasking the growth of cosmological structures

TL;DR

This paper demonstrates that multi-task Gaussian processes can effectively reconstruct cosmological growth functions from simulated future survey data, outperforming single-task methods and enhancing the detection of deviations from standard models.

Contribution

It introduces a multi-task Gaussian process approach to simultaneously reconstruct multiple cosmological functions, improving accuracy over traditional single-task methods.

Findings

Multi-task approach outperforms single-task in future survey simulations.

Enhanced ability to detect deviations from the standard cosmological model.

Current data limitations hinder agnostic reconstruction due to parameter tuning challenges.

Abstract

Next-generation large-scale structure surveys will deliver a significant increase in the precision of growth data, allowing us to use `agnostic' methods to study the evolution of perturbations without the assumption of a cosmological model. We focus on a particular machine learning tool, Gaussian processes, to reconstruct the growth rate $f$, the root mean square of matter fluctuations $\sigma_8$, and their product $f\sigma_8$. We apply this method to simulated data, representing the precision of upcoming Stage IV galaxy surveys. We extend the standard single-task approach to a multi-task approach that reconstructs the three functions simultaneously, thereby taking into account their inter-dependence. We find that this multi-task approach outperforms the single-task approach for future surveys and will allow us to detect departures from the standard model with higher significance. By contrast, the limited sensitivity of current data severely hinders the use of agnostic methods, since the Gaussian processes parameters need to be fine tuned in order to obtain robust reconstructions.