From Nonparametric Power Spectra to Inference About Cosmological Parameters: A Random Walk in the Cosmological Parameter Space

TL;DR

This paper uses nonparametric methods to analyze WMAP data, providing a data-driven approach to infer cosmological parameters and exploring their uncertainties without relying on specific models.

Contribution

It introduces a novel sampling technique to explore the confidence set of cosmological parameters derived from nonparametric fits, validating known degeneracies and highlighting data limitations.

Findings

Cosmological parameters are less constrained by WMAP data alone.

Additional priors improve parameter constraints.

Degeneracies in parameter space are correctly identified.

Abstract

What do the data, as distinguished from cosmological models, tell us about cosmological parameters that determined the model of the universe? In this paper, we address this question in the context of the WMAP angular power spectra for the cosmic microwave background radiation. Nonparametric methods are ideally suited for this purpose because they are model-independent by construction, and therefore allow inferences that are as data-driven as possible. Our analysis is based on a nonparametric fit to the WMAP 7-year power spectrum data, with uncertainties characterized in the form of a high-dimensional confidence set centered at this fit. For the purpose of making inferences about cosmological parameters, we have devised a sampling method to explore the projection of this confidence set around the nonparametric fit, into the space of seven cosmological parameters Omega_b, Omega_c, Omega_Lambda, Omega_k, H_0, n_s, tau). Our sampling method is justified by its computational simplicity, and validated by the fact that well-known degeneracies in this cosmological parameter space are correctly reproduced in our results. Our results show that cosmological parameters are not as tightly constrained by these data alone. However, incorporating additional prior information in the analysis (e.g., constraining the values of H_0 or Omega_k) leads to tighter confidence intervals on parameters that are consistent with uncertainties estimated by the mainstream parametric methods.