Cosmological model selection

TL;DR

This paper reviews model selection methods in cosmology, emphasizing Bayesian evidence, and introduces CosmoNest for efficient Bayesian model comparison applied to WMAP data.

Contribution

It presents CosmoNest, the first efficient computational tool for Bayesian model selection in cosmology, and applies it to analyze WMAP data for spectral index deviations.

Findings

Bayesian evidence effectively distinguishes models with different spectral indices.

CosmoNest enables practical model comparison in cosmological data analysis.

Results suggest data may favor a non-scale-invariant spectral index.

Abstract

Model selection aims to determine which theoretical models are most plausible given some data, without necessarily asking about the preferred values of the model parameters. A common model selection question is to ask when new data require introduction of an additional parameter, describing a newly-discovered physical effect. We review several model selection statistics, and then focus on use of the Bayesian evidence, which implements the usual Bayesian analysis framework at the level of models rather than parameters. We describe our CosmoNest code, which is the first computationally-efficient implementation of Bayesian model selection in a cosmological context. We apply it to recent WMAP satellite data, examining the need for a perturbation spectral index differing from the scale-invariant (Harrison-Zel'dovich) case.