Benchmarking the cosmological master equations

TL;DR

This paper critically evaluates the use of master equations in cosmology by comparing their predictions to an exactly solvable model, revealing limitations due to initial-condition-dependent terms but also their potential for late-time resummation.

Contribution

It demonstrates how to assess and improve the accuracy of master equations in cosmology by identifying and removing spurious terms caused by dynamical backgrounds.

Findings

Master equations show initial-condition-dependent spurious terms.

Removing these terms improves the accuracy of power spectra predictions.

Master equations can effectively perform late-time resummation in cosmological models.

Abstract

Master equations are commonly employed in cosmology to model the effect of additional degrees of freedom, treated as an "environment", onto a given "system". However, they rely on assumptions that are not necessarily satisfied in cosmology, where the environment may be out of equilibrium and the background is dynamical. In this work, we apply the master-equation program to a model that is exactly solvable, and which consists of two linearly coupled scalar fields evolving on a cosmological background. The light field plays the role of the system and the heavy field is the environment. By comparing the exact solution to the output of the master equation, we can critically assess its performance. We find that the master equation exhibits a set of "spurious" terms that explicitly depend on the initial conditions, and which arise as a consequence of working on a dynamical background. Although they cancel out in the perturbative limit of the theory (i.e. at leading orders in the interaction strength), they spoil resummation. However, when those terms are removed, the master equation performs impressively well to reproduce the power spectra and the amount of the decoherence of the light field, even in the strongly decohered regime. We conclude that master equations are able to perform late-time resummation, even though the system is far from the Markovian limit, provided spurious contributions are suppressed.