Sim map sets to demonstrate "real delensing" (02.00 and 02.09)
Feb 7 2018, Clem Pryke
The current plan for CMB-S4 calls for a deep (fsky order 3%) survey with both small and large apertures hitting the same field. The small apertures will deliver very low noise at larger angular scales and the large apertures will provide the lensing reconstruction needed to "delens" the small aperture maps. For the CDT report appendix A simulations we "faked" delensing by simply turning down the strength of the lensing effect in the small aperture maps. The next step is to simulate coordinated small and large aperture maps and do a re-analysis including "real delensing". This post introduces two sets of such maps. This is built on the previous Experiment Definitions 02 set so the frequencies, resolutions, and noise levels do not exactly match the CDT report strawman configuration. However, they should be good enough to test lensing reconstruction in the presence of apodization, noise and (one form of) non-Gaussian foreground.
Two sets of maps are provided on NERSC in directories /project/projectdirs/cmbs4/data_xx.yy/02.YY where YY=00 is a simple Gaussian foreground model and YY=09 is the non-Gaussian model kindly provided by Flavien Vansyngel - see Vansyngel Model.
The plot below shows the Q maps for realization 0001. The left three columns are the low res (small aperture) maps for 95/145/220GHz bands and the right three columns are the high res (large aperture) versions in the same bands. (There are additional low res bands not shown.) The top row of plots is a realization of lensed-LCDM at the appropriate resolution. The second row is the Vansyngel foreground model at each resolution (mostly dust). The third row is the experimental noise which blows up around the edge in an appopriate manner. It is hard to see but the edge taper is actually less for the high res due to smaller instantaneous field of view. The noise extends up to the pixel scale which is nside=512 for the low res maps and nside=2048 for the high res, so at the pixelization of this rendering the high res map is heavily aliased and the noise level looks higher than for the low res - in fact the white noise level is the same for e.g. 95GHz low and high res. The bottom row shows the combination of the above rows which is what one actually gets to measure.
A link to a pdf file is provided to the right of the plot above where you can zoom in and overcome the screen aliasing. Alternatively the plot below is pre-zoomed to the central region of the maps. In the top row one can now clearly see the loss of lensing signal in the low res maps, while the high res look nearly the same because the lensing signal is adequately resolved in all three bands. In the middle row at 220GHz we can see that the dust foreground model contains small scale structure which the low res telescope is unable to see. In the third row for the low res bands we can see the white noise cutting off at the Healpix pixel scale (the white noise level actually being the same between high and low res at each given frequency).
The overall goal is to jointly analyze the full set of low and high res maps for each realization and extract the maximum likelihood value of r, and then to take statistics over the available realizations to investigate bias and uncertainty for a given re-analysis technique. However, the task can be split. One way to proceed is to take the high res maps and analyze them alone to extract the lensing potential map phi. That can then be used to form a lensing template and fed forward to the next step in the re-analysis.
Some more info on the sims is available at Data Challenge Map Sets 04.YY.
For a very first step one can take the high res masked lensed-LCDM single frequency map like data_xx.yy/02.00/cmbs4_02_llcdm_f145_b04_ellmin30_map_2048_mc_0000.fits and run phi reconstruction. The appropriate "relative hits" mask is /project/projectdirs/cmbs4/expt_xx/02/rhits/n2048.fits.