What can Machine Learning tell us about the background expansion of the Universe?

TL;DR

This paper employs genetic algorithms to nonparametrically reconstruct the Universe's background expansion from cosmological data, detecting acceleration and exploring deviations from the standard Lambda-CDM model.

Contribution

It introduces a model-independent, nonparametric approach using genetic algorithms to analyze cosmological expansion data without assuming dark energy or flatness.

Findings

Detected accelerated expansion at ~4.5 sigma significance.

Estimated the transition redshift of acceleration as 0.662 ± 0.027.

Found hints of deviations from Lambda-CDM at high redshifts within errors.

Abstract

Machine learning (ML) algorithms have revolutionized the way we interpret data in astronomy, particle physics, biology and even economics, since they can remove biases due to a priori chosen models. Here we apply a particular ML method, the genetic algorithms (GA), to cosmological data that describes the background expansion of the Universe, namely the Pantheon Type Ia supernovae and the Hubble expansion history $H(z)$ datasets. We obtain model independent and nonparametric reconstructions of the luminosity distance $d_L(z)$ and Hubble parameter $H(z)$ without assuming any dark energy model or a flat Universe. We then estimate the deceleration parameter $q(z)$, a measure of the acceleration of the Universe, and we make a $\sim4.5\sigma$ model independent detection of the accelerated expansion, but we also place constraints on the transition redshift of the acceleration phase $(z_{\textrm{tr}}=0.662\pm0.027)$. We also find a deviation from $\Lambda$CDM at high redshifts, albeit within the errors, hinting toward the recently alleged tension between the SnIa/quasar data and the cosmological constant $\Lambda$CDM model at high redshifts $(z\gtrsim1.5)$. Finally, we show the GA can be used in complementary null tests of the $\Lambda$CDM via reconstructions of the Hubble parameter and the luminosity distance.