Model selection applied to reconstruction of the Primordial Power Spectrum

TL;DR

This paper uses Bayesian model selection to compare different models of the primordial power spectrum, finding current data favors the Lasenby & Doran model over the standard power-law form.

Contribution

It introduces a Bayesian approach to reconstruct the primordial spectrum with variable complexity and compares multiple models to identify the most supported by data.

Findings

Power-law spectrum is disfavored by data.

Lasenby & Doran model is preferred over power-law.

Reconstruction reveals potential features in the spectrum.

Abstract

The preferred shape for the primordial spectrum of curvature perturbations is determined by performing a Bayesian model selection analysis of cosmological observations. We first reconstruct the spectrum modelled as piecewise linear in \log k between nodes in k-space whose amplitudes and positions are allowed to vary. The number of nodes together with their positions are chosen by the Bayesian evidence, so that we can both determine the complexity supported by the data and locate any features present in the spectrum. In addition to the node-based reconstruction, we consider a set of parameterised models for the primordial spectrum: the standard power-law parameterisation, the spectrum produced from the Lasenby & Doran (LD) model and a simple variant parameterisation. By comparing the Bayesian evidence for different classes of spectra, we find the power-law parameterisation is significantly disfavoured by current cosmological observations, which show a preference for the LD model.