Likelihood, Fisher information, and systematics of cosmic microwave background experiments

TL;DR

This paper analyzes how systematic effects like beam asymmetry, noise, and sky coverage impact the likelihood function in CMB experiments, revealing that some effects are manageable while masking complicates data analysis.

Contribution

It provides an analytical study of systematic effects on the likelihood function in CMB experiments, highlighting their influence on information content and data correlations.

Findings

Likelihood can deviate from Gaussian at high multipoles.

Asymmetric beams and correlated noise do not significantly reduce information.

Sky masking introduces complex correlations in the likelihood.

Abstract

Every experiment is affected by systematic effects that hamper the data analysis and have the potential to ultimately degrade its performance. In the case of probes of the cosmic microwave background (CMB) radiation, a minimal set of issues to consider includes asymmetric beam functions, correlated noise, and incomplete sky coverage. Presuming a simplified scanning strategy that allows for an exact analytical treatment of the problem, we study the impact of systematic effects on the likelihood function of the CMB power spectrum. We use the Fisher matrix, a measure of the information content of a data set, for a quantitative comparison of different experimental configurations. In addition, for various power spectrum coefficients, we explore the functional form of the likelihood directly, and obtain the following results: The likelihood function can deviate systematically from a Gaussian distribution up to the highest multipole values considered in our analysis. Treated exactly, realistic levels of asymmetric beam functions and correlated noise do not by themselves decrease the information yield of CMB experiments nor do they induce noticeable coupling between multipoles. Masking large fractions of the sky, on the other hand, results in a considerably more complex correlation structure of the likelihood function. Combining adjacent power spectrum coefficients into bins can partially mitigate these problems.