Galactic foreground residual biases in CMB lensing convergence reconstruction and delensing of B-mode maps

TL;DR

This work analyzes how Galactic foreground residuals bias CMB lensing reconstruction and delensing of B-mode maps for a CMB-S4–like experiment. Using realistic simulations with three foreground models and harmonic ILC component separation, it shows that Gaussian foreground residuals dominate reconstruction noise while non-Gaussian residuals induce small biases, which are largely mitigated after component separation; mean-field biases from masking must be corrected, particularly for EB estimators. Delensing with a gradient-order template can remove about $65\%$ of the lensing B-modes, but residual lensing and foreground residuals still limit $r$-constraints, and the study finds that, despite improvements, a robust $3\sigma$ detection of $r\sim10^{-3}$ remains challenging without further delensing advances or external LSS tracers. The results highlight the critical role of accurate foreground modelling and aggressive delensing in achieving unbiased tensor mode measurements with future CMB polarization experiments.

Abstract

Diffused contamination from Galactic foreground emission is one of the main concern for reconstruction of the Cosmic Microwave Background (CMB) lensing potential for next-generation of CMB polarisation experiments. Using realistic simulations we investigate the impact of Galactic foreground residuals from multi-frequency foreground cleaning method on the lensing reconstruction, delensing CMB B-mode maps and constraints of the tensor-to-scalar ratio for CMB-S4-like experiment. We pay special attention to studies of the errors coming from small angular scale non-Gaussianity of the foreground residuals. We show that component separation is essential for the lensing reconstruction reducing Galactic emission contribution to the lensing reconstruction errors by one order of magnitude. The residual foreground contribution is dominated by terms coming from Gaussian components of the residual maps. Errors coming from non-Gaussian components are around three orders of magnitude smaller than the Gaussian one even for recent and the most complex models of the Galactic emission considered in this work. Although the bias in the reconstruction errors due to Gaussian component of the residuals is small, it is comparable to the cosmic variance limit for the lensing power spectrum. For this reason we correct for this bias in delensing of B-mode maps and constraining the tensor-to-scalar ratio. We show also that for the delensed B-mode maps with a simple quadratic estimator, residuals of the Galactic emission after component separation, errors are two orders of magnitude smaller than uncertainties from leftover of the lensing signal. However, for high-sensitivity CMB experiments and more efficient delensing algorithms that remove up to 90% of the lensing signal, the foreground residuals will become one of the main sources of errors.

Figures (15)

• Figure 1: Analytical zeroth-order lensing reconstruction bias, $N_L^{(0)}$ for different estimators given by Eq. (\ref{['eq:nl0_anal_bias']}). The instrumental specification is considered for a CMB-S4-like experiment with beam resolution 2.5 arcminutes and white noise level in temperature $\Delta_T$ = 2 $\mu K$-arcmin and polarisation $\Delta_P$ = 2.8 $\mu K$-arcmin.
• Figure 2: Masks used in our analysis to mimic CMB-S4 sky coverage shown in Ecliptic coordinates with Galactic emission at 145 GHz in the background. The left and middle panels are considered for wide sky survey corresponding to the LATs configuration: the left panel shows mask used for component separation ($f_{sky}=0.64$) and the middle panel shows mask used for lensing reconstruction ($f_{sky}=0.37$). The right panel shows the sky area considered for deep survey ($f_{sky}=0.025$) corresponding to the SATs configuration which is used for delensing study.
• Figure 3: Beam-deconvolved noise power spectra for temperature (solid) and polarisation (dashed) in LAT configuration.
• Figure 4: Angular power spectra for CMB B-mode cleaned maps for the SAT configuration and sky coverage. The grey solid zig-zag lines are BB power spectra of 100 simulations and the average is shown as red stars. The theoretical input power spectra are shown as black solid line. The residual HILC noise and foreground levels are shown as dotted and dashed lines, respectively. The magenta solid line is average delensed B mode spectra over 100 simulations. The dashed blue colored line shows the average power spectra of Galactic foreground B modes at the frequency channels 95 and 145 GHz. The dashed grey line shows tensor B mode level for $r=0.003$.
• Figure 5: Angular power spectra of reconstructed lensing field shown as blue solid line. The mean field effect on EB estimator for MF channels for d11s6 and HILC cleaned maps for d11s6 and d12s7 are shown as red, green and grey solid lines, respectively. The mean-filed effect on EE estimator for HILC cleaned maps of d11s6 is shown as cyan solid line. Theoretical lensing power spectra is shown in black solid line. The dashed magenta line is the total sum of theory power spectra and lensing reconstruction noise terms, $N^{(0)}$ and $N^{(1)}$ shown in shown as dashed (orange) and dash-dotted (purple) lines, respectively.