Mathieu ran 20 simulations of 90.92 with r=0, 85% delensing, 36% sky coverage. The results are in the logbook.
We see significant residuals above lensing at \ell<20. Discussing what's going on.
SH asks: is the excess power at low \ell the result of small sky coverage, or of the actual foreground residual. How does the pseudo cl code deal with small sky coverage? Actions:
Mathieu will plot the input CMB for the same mask 36% mask
Mathieu will use a much larger 98%, and much smaller 2.5% mask.
Mathieu will show the masks.
We note that if we believe the results, the foreground residual is white below \ell=200, and happens to peak above the delensed at \ell<10. Is that the limit for GNILC cleaning?
Mathieu says that the code has not been optimized yet.
Graca ran Soumen's r=0 maps. She gets r=0, but the dispersion is 10^{-7}. Far too low.
Suggestion: take a step back; validate code on pure CMB simulations with no foregrounds, but with noise.
without foregrounds, recover r=0 with full lensing, and with 85% delensing.
without foregrounds, recover r=1e-3 with 85% delensing with small and larger noise.