ILC noise for large apertures at CDT noise levels
Mar 16 2018, Raphael Flauger (posted by Clem Pryke)
This is from en email from Raphael to Clem and Tom dated 7 March 2018 with subject "Delensing survey".
"To check things, I've run component separation on the 04 data challenge high resolution simulations to generate a noise curve for an ILC high resolution map. (The philosophy here is that I would clean the high resolution maps and derive a kappa map from the resulting cleaned map.) I'm attaching the noise curve here with white noise of 1uK arcmin shown for comparison. So over the relevant range it's roughly 1uKarcmin as expected (as one could also see without much effort from the noise models). I then ran iterative delensing for this noise curve (at the power spectrum level). I find a residual AL=0.123. It's 20% higher that what we used (and I can check what that does to r), but in any case, it's reasonably close."
Clarification from Raphael: The analysis is based on Fisher calculations that rely on the 04 params.dat file, which is based on the numbers in the CDT report. The 220 and 270 GHz noise levels may be too low. Updated noise levels sent by Tom (10 uK-arcmin at 220 GHz and 17 uK-arcmin at 270 GHz) only lead to a slightly worse delensing efficiency with lensing residual of AL=0.13, but significantly higher noise levels on large scales, warranting more detailed study.
Further clarification from Raphael: These are Fisher forecasts that include a 2-component dust and synchrotron model with a scaling to the foreground level appropriate for the size of the patch according to Planck data. These forecasts do quite well predicting noise in the ILC maps and foreground residuals for the degree scale simulations we have been using. I haven't validated this on high resolution simulations yet but will as soon as I get around to it. I don't know why it shouldn't work there, but that definitely should be checked.