Effect of Template Uncertainties on the WMAP and Planck Measures of the Optical Depth Due To Reionization

TL;DR

This study investigates how template uncertainties and systematics in WMAP and Planck polarization data influence measurements of the reionization optical depth, highlighting biases and the importance of systematic error considerations.

Contribution

It provides a detailed analysis of systematic biases in reionization optical depth measurements due to foreground template choices and data processing methods.

Findings

Evidence of systematic leakage in LFI polarization data.

Bias in optical depth when using LFI 30 GHz as a template.

Systematic uncertainties significantly affect τ estimates.

Abstract

The reionization optical depth is the most poorly determined of the six $\Lambda$CDM parameters fit to CMB anisotropy data. Instrumental noise and systematics have prevented uncertainties from reaching their cosmic variance limit. At present, the datasets providing the most statistical constraining power are the WMAP, Planck LFI, and Planck HFI full-sky polarization maps. As the reprocessed HFI data with reduced systematics are not yet publicly unavailable, we examine determinations of $\tau$ using 9-year WMAP and 2015 Planck LFI data, with an emphasis on characterizing potential systematic bias resulting from foreground template and masking choices. We find evidence for a low-level systematic in the LFI polarization data with a roughly common-mode morphology across the LFI frequencies and a spectrum consistent with leakage of intensity signal into the polarization channels. We demonstrate significant bias in the optical depth derived when using the LFI 30 GHz map as a template to clean synchrotron from WMAP data, and recommend against use of the 2015 LFI 30 GHz polarization data as a foreground template for non-LFI datasets. We find an inconsistency between versions of the 2015 polarized 353 GHz dust templates reconstructed from the Planck likelihood and those from delivered maps, which can affect $\tau$ at the 1$\sigma$ level. The spread in $\tau$ values over the ensemble of data combinations we study suggests that systematic uncertainties still contribute significantly to the current uncertainty in $\tau$, but all values are consistent with the range of $\tau$ = 0.07 +/- 0.02.