Impact on the tensor-to-scalar ratio of incorrect Galactic foreground modelling

TL;DR

This paper investigates how incorrect modeling of Galactic foregrounds can bias the estimation of the tensor-to-scalar ratio in CMB polarization measurements, emphasizing the need for accurate foreground models to detect primordial gravitational waves.

Contribution

It demonstrates the impact of mismatched foreground models on biasing the tensor-to-scalar ratio and highlights the importance of balanced complexity in foreground modeling for CMB experiments.

Findings

Incorrect foreground modeling can bias r by up to 3-sigma.

Allowing too much freedom in models can cause significant bias.

Adding more data reduces the bias in r estimation.

Abstract

A key goal of many Cosmic Microwave Background experiments is the detection of gravitational waves, through their B-mode polarization signal at large scales. To extract such a signal requires modelling contamination from the Galaxy. Using the Planck experiment as an example, we investigate the impact of incorrectly modelling foregrounds on estimates of the polarized CMB, quantified by the bias in tensor-to-scalar ratio r, and optical depth tau. We use a Bayesian parameter estimation method to estimate the CMB, synchrotron, and thermal dust components from simulated observations spanning 30-353 GHz, starting from a model that fits the simulated data, returning r<0.03 at 95% confidence for an r=0 model, and r=0.09+-0.03 for an r=0.1 model. We then introduce a set of mismatches between the simulated data and assumed model. Including a curvature of the synchrotron spectral index with frequency, but assuming a power-law model, can bias r high by ~1-sigma (delta r ~ 0.03). A similar bias is seen for thermal dust with a modified black-body frequency dependence, incorrectly modelled as a power-law. If too much freedom is allowed in the model, for example fitting for spectral indices in 3 degree pixels over the sky with physically reasonable priors, we find r can be biased up to ~3-sigma high by effectively setting the indices to the wrong values. Increasing the signal-to-noise ratio by reducing parameters, or adding additional foreground data, reduces the bias. We also find that neglecting a 1% polarized free-free or spinning dust component has a negligible effect on r. These tests highlight the importance of modelling the foregrounds in a way that allows for sufficient complexity, while minimizing the number of free parameters.