Model selection as a science driver for dark energy surveys

Abstract

A key science goal of upcoming dark energy surveys is to seek time evolution of the dark energy. This problem is one of {\em model selection}, where the aim is to differentiate between cosmological models with different numbers of parameters. However, the power of these surveys is traditionally assessed by estimating their ability to constrain parameters, which is a different statistical problem. In this paper we use Bayesian model selection techniques, specifically forecasting of the Bayes factors, to compare the abilities of different proposed surveys in discovering dark energy evolution. We consider six experiments -- supernova luminosity measurements by the Supernova Legacy Survey, SNAP, JEDI, and ALPACA, and baryon acoustic oscillation measurements by WFMOS and JEDI -- and use Bayes factor plots to compare their statistical constraining power. The concept of Bayes factor forecasting has much broader applicability than dark energy surveys.