Nongaussian dust in lensing

From CMB-S4 wiki
Revision as of 15:23, 10 February 2017 by Sherwin (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search

Alex writing -- all results are preliminary

One of the potential issues for CMB-S4 is the possibility of non-Gaussian foregrounds on the small scales that enter lensing reconstruction. These could impact not only measures of lensing itself but also delensing.

Here, we describe an ongoing effort to detect / characterize a lensing bias from non-Gaussian dust on small scales.

Colin Hill and Susan Clark have provided a map of small-scale polarization using correlation with HI angles (and for the pol amplitude assuming a constant pol fraction). I (Alex) cut this map out into 10x10-deg chunks.

Fig 1. T map:

QuickPlot T.png

Fig 2. Q map:

QuickPlot Q.png

Fig 3. Power spectra — darker colors are at higher Galactic latitude:


Fig 4. Ratio of EE to BB power in each patch - not exactly 2:


We run these maps through a 4-pt estimator to get the bias to the lensing autospectrum. For each patch we subtract a piece from the disconnected trispectrum.

Fig 5: Bias to lensing power spectrum:


Fig 6: Lensing bias vs. dust power for each patch:


  1. The power is very high in the EE and BB maps; we are working to understand the origin of this. This is likely leading to a significant overestimate of the bias on the lensing which we need to quantify.
  2. We are currently doing a similar analysis on maps from Vansyngel et al
  3. Note that the correlated delensing correction to low-ell B mode power arises from the same <EBEB> trispectrum (though with most weight falling on different quadrilaterals, some of which have zero weight here). We have a strawperson plan for computing the impact on delensing:
    1. From these maps construct phi_dust ~ E_dust B_dust
    2. Perform lensing for B modes, B_template = E_dust . phi_dust (where . represents the lensing operator)
    3. Compute <B_dust B_template>

In addition to this correlated / non-Gaussian contribution to delensing (which may be a worry as it could cause a small, unknown bias), the increase in the reconstruction noise (due to excess foreground power) will also degrade the delensing performance. However, this is probably less of a worry, as it can be understood / absorbed in an increase of A_L more easily. We may have to bear it in mind, however, when considering very high delensing fractions.