ML searches with LT on 06 sims - results on models with non-Gaussian foregrounds - v2

 (C. Umiltà)

ML search for DC06

In this posting I update previous results and show the result of ML searches for all parameters using different lensing templates, obtained from sims with different foreground models. A previous study that only had LT with Gaussian foregrounds can be found here here. The final pager with these LT can be found here.

The iterative lensing templates are obtained by Julien Carron from the 95 GHz maps.

Figure 1:
ML searches results.

Note 1: Only "06.00 Ideal template", "06.07" and "06.09" have the option without the lensing template.
Note 2: For 07 and 09 foregorund models, when running ML search over the full bin range with the LT, we observe much larger values of the likelihood. For these cases, we also see ~10 realizations at low likelihood values (~200) for model 07, outside plot scale.
Note 3: For some models, in particular for foreground models 07 and 09, I had to extend the range for the parameters \(A_{d}\) and \(A_{s}\) to \([-2,30]\), while the usual standard is \([-2,15]\).

Comparison to DC04

Previous postings by Ben on DC04 presents a good point of comparison to this analysis. When reading them, one should keep in mind that they discuss many more foregrounds models, include decorrelation in th likelihood model and fit of \(A_{L}\). We only analyze foreground models 00, 07, 09, do not have any parameter for decorrelation and keep \(A_{L}\) fixed.
Also, the masks used have slightly changed as well as naming conventions. This DC06 analysis only uses 06b simulations, also known as "Pole deep". See here for the mask pattern and here for a comparison of simulations with different masks. Finally, DC04 analysis uses more realizations (up to 500, removing faulty ones), while DC06 only has 50 per each case and does not discard any realization.
The DC04 relevant posting are:

The latter is probably closest to the Pole deep mask used for DC06.

In both postings, Figure 1 compares to this analysis when clicking on "free \(\beta \), fixed \(A_{L}\)" option and looking at the left (blue) side of the plot, which has no decorrelation in the model parameterization.
The sky models to compare are 0, 7, 9.
DC04 analysis does not use a lensing template. Instead it artificially scales down the value of "\(A_{L}\)". Figure 2 and Figure 3 below reproduce part of the results of these postings and compares them to the current DC06 results.

Figure 2:
Summary of ML searches results for different values of \(A_{L}\). Red point show the \(r\) results for r=0.003 of DC04 on Pole mask 04c, while green ones show results for r=0. Yellow and blue points show the results for DC06 for r=0.003 and r=0, when using the full bin range or just the first 5 bins. These runs all have no prior on \(\beta\), fied \(A_{L}\) and no decorrelation.
Figure 3:
Summary of ML searches results for different values of \(A_{L}\). Red point show the \(\sigma(r)\) results for r=0.003 of DC04 on Pole mask 04c, while green ones show results for r=0. Values of the fit are taken from Figure 2 in this posting. Error bars are an estimate of the error on the std, estimated as \(\sigma / \sqrt{2 \cdot N_{sims}}\). Yellow and blue points show the results for DC06 for r=0.003 and r=0, when using the full bin range or just the first 5 bins. These runs all have no prior on \(\beta\), fied \(A_{L}\) and no decorrelation.