Machine learning unveils the linear matter power spectrum of modified gravity

TL;DR

This paper employs machine learning, specifically genetic algorithms, to develop an analytical formula for the matter power spectrum in modified gravity models, achieving 1-2% accuracy and aiding future cosmological surveys.

Contribution

It introduces a novel machine learning approach to derive an analytical expression for the matter power spectrum under modified gravity, extending beyond the standard model.

Findings

Achieves 1-2% accuracy compared to numerical data.

Provides a parametric function for P(k) in modified gravity.

Facilitates analytical understanding for upcoming surveys.

Abstract

The matter power spectrum $P(k)$ is one of the main quantities connecting observational and theoretical cosmology. Although for a fixed redshift this can be numerically computed very efficiently by Boltzmann solvers, an analytical description is always desirable. However, accurate fitting functions for $P(k)$ are only available for the concordance model. Taking into account that forthcoming surveys will further constrain the parameter space of cosmological models, it is also of interest to have analytical formulations for $P(k)$ when alternative models are considered. Here, we use the genetic algorithms, a machine learning technique, to find a parametric function for $P(k)$ considering several possible effects imprinted by modifications of gravity. Our expression for the $P(k)$ of modified gravity shows a mean accuracy of around 1-2% when compared with numerical data obtained via modified versions of the Boltzmann solver CLASS, and thus it represents a competitive formulation given the target accuracy of forthcoming surveys.