Iterative curved sky s06b nofg lensing templates

From CMB-S4 wiki
Jump to navigationJump to search

Julien Carron, April 10 2020

This pages describes the 100 iterative (maximum a posteriori, 'MAP') lensing map built from the s06b simulations without foreground, and the associated lensing templates.

The lensing B-mode templates should remove about 90% of the lensing power in the central region and are available as alm fits files in

/global/cscratch1/sd/jcarron/share_cmbs4/06b_nofg_carron_100420

The CMB likelihood model used to build the maps include:

  • Curved-sky geometry
  • An inhomogeneous noise model built from the hit map /project/projectdirs/cmbs4/expt_xx/06b/rhits/n2048.fits together with the empirically estimated central noise value
  • Exact deprojection of all B-modes from 2 to 200, to ensure these modes are not used to build the lensing template
  • Used are E-modes from 2 to 3000 and B-modes from 200 to 3000
  • The same mean-field is used for each iteration

All sims used 12 iterations, though a few less would have been enough.

Caveats:

  • The approximation to the mean-field leaves a residual sometimes visible at low-L near the edges of the patch. This is however restricted to low lensing multipoles, and the impact on the B-template quality is expected to be very small
  • The tracer could be improved by going to higher CMB lmax, with maybe ~5% gain at high-L ~ 2000, but again this will have a small impact only on the B-template


A typical cross-correlation to the input lensing looks like this (calculated in a central region of 900 sq. deg. within the s06b patch), where the L-modes relevant to the B-power are centred on L~500. The correlation there goes from ~85% to ~96%

        Cs MAP rho.pdf

The B-templates are built remapping the estimated unlensed E-mode.

Across the central 900 sq. deg. region, I estimate the residual B-power after template substraction to be about 0.1 (for something close to 0.3 for the QE, the starting point)

        Cs resAL.png

Obviously the quality of the tracer and template degrades closer to the edges


        Cs resB.png