# MapBasedR

## Summary

We checked the case in Victor's forecasts using our map-based component separation code on PySM simulations. Bullet-point results (details below):

• Assuming no delensing:
• Assuming a 0.25 delensing factor:

These number roughly agree with Victor's forecast , considering our potentially different assumptions about noise power spectra and delensing efficiency.

## Simulations

We used PySM to generate full-sky simulations including:

• Power-law synchrotron (spatially-varying spectral index).
• Single-component thermal dust (spatially-varying temperature and spectral index).
• , partially de-lensed CMB.
• Noise levels compatible with Victor's case.

The maps are cut using a mask defined by selecting the cleanest 4000 sq-deg of the southern sky in polarization.

## Foreground removal

We run a map-based Bayesian component-separation code (Alonso et al. in prep.) on the simulations. The code samples the fully-resolved amplitudes of the three different components as well as spectral parameters ( and ). The latter are assumed constant on larger pixels, with HEALPix resolution (corresponding to ~4 deg). Fig. below shows the B-mode map at 145 GHz (left) and the mean CMB-only B-mode map output by the code (right).

## Estimating r

We compute the B-mode power spectrum for each simulation and fit a primordial + lensing template with amplitudes for both components (the first one being ). For this we only use multipoles . See result below for the no-delensing and 0.25-delensing cases.