The Impact of Prior Assumptions on Bayesian Estimates of Inflation Parameters and the Expected Gravitational Waves Signal from Inflation

TL;DR

This paper investigates how prior assumptions influence Bayesian estimates of inflationary parameters and the expected gravitational wave signals, highlighting the sensitivity of results to prior choices and the current data's limitations.

Contribution

It demonstrates the impact of different priors on inflationary parameter estimates and assesses how effectively existing data constrain these parameters.

Findings

Existing data are insufficient for prior-independent inflationary estimates.

Choice of prior significantly affects the inferred tensor-to-scalar ratio r.

Current data only provide an upper bound on r, consistent with zero.

Abstract

There has been much recent discussion, and some confusion, regarding the use of existing observational data to estimate the likelihood that next-generation cosmic microwave background (CMB) polarization experiments might detect a nonzero tensor signal, possibly associated with inflation. We examine this issue in detail here in two different ways: (1) first we explore the effect of choice of different parameter priors on the estimation of the tensor-to-scalar ratio r and other parameters describing inflation, and (2) we examine the Bayesian complexity in order to determine how effectively existing data can constrain inflationary parameters. We demonstrate that existing data are not strong enough to render full inflationary parameter estimates in a parametrization- and prior-independent way and that the predicted tensor signal is particularly sensitive to different priors. For parametrizations where the Bayesian complexity is comparable to the number of free parameters we find that a flat prior on the scale of inflation (which is to be distinguished from a flat prior on the tensor-to-scalar ratio) leads us to infer a larger, and in fact slightly nonzero tensor contribution at 68% confidence level. However, no detection is claimed. Our results demonstrate that all that is statistically relevant at the current time is the (slightly enhanced) upper bound on r, and we stress that the data remain consistent with r = 0.