====== Telecon Notes January 07 2021 ======

[[https://spa-zzz-01.spa.umn.edu/!ipsig/foregrounds|Link to main foregrounds telecons page]] \\
[[https://spa-zzz-01.spa.umn.edu/!ipsig/logbook|Link to logbook]]

Attendance:  \\
Regrets:  \\

Agenda:
  * recap: where were we in 2020, what has happened since - Ragnhild and Mathieu 
  * [[https://spa-zzz-01.spa.umn.edu/!ipsig/paperoutline|Paper Outline]] 

Notes: Shaul \\

==== Notes ====

  * Ragnhild revised and posted an {{::model_commander.pdf |updated Commander memo}}. 
  * She provided {{ ::20210107_commander_pico.pdf | slides summarizing her work to date}} 
    * Discussion about slide 3, third and fourth bullet: with input 90.92, fitting 90.91 shows residuals and large chi^2, indicating that 90.91 is not the right model. Some discussion ensues on the reverse case in which the input is 90.91 (6 input parameters), and we are fitting with 90.92 that has 12 fitting parameters. No significant improvement in chi^2 is observed. 
    * To treat the data like it would be in real life we should: have an N-parmaeter model as a input, fit it with M<N parameter and get large residuals. Then we'll add parameters, hopefully progressively showing that reduced chi^2 approaches 1, and then, when the parameters are >N the residuals stop decreasing. 
    * Slide #4 gives Ragnhild's current status: she is analyzing the 20 simulations of 90.91 and will soon analyze the 20 of 90.92. 
    * Next: removing high frequency bands, and a new model. Jacques advocates using his proposed model, but MR has been running other models. 

    * Email communication with SH after the telecon
       * Slide 6: "the reduced chi^2 is consistent with 1" - I suggest plotting a distribution of the reduced chi^{2} including only the masked sky. Relying on visual impression is not very quantitative.
       * Slide 7: at 21 GHz there is obvious residual. Is this because AME is not included in the fit? It would be nice to do a chi^2 distribution (see above) without and with AME. Hopefully the the distribution with would be nicely clustered around 1 ...
         * Response: The patterns seen in the residuals at 21 GHz comes from synchrotron. This shows that we are not picking up all the synchrotron signal, so some of it is left in the residuals instead of being a part of the synchrotron map. I will try to fit synchrotron at a lower resolution (nside 64) to increase the signal-to noise for synchrotron. Hopefully this will improve the result. There are no polarized AME in these simulations. 
       * So far you have only shown a couple of r likelihoods in October, is that correct? This is the Blackwell Rao estimator you've been working on, is that right? 
         * Response: Yes, the r plots I have been showing is from my Blackwell Rao estimator. I’ll hope to show more r plots from the runs I had going over Christmas next telecon. 
       * Take a look at slide three of the sept. 17 presentation: Is this how the CMB *should* look? With these large scale cold and hot stripes?
         * Response: Yes, this is close to how the polarized Stokes Q and U maps should look like. Different realizations will of course look a bit different, but Q (left) will always have these hot and cold stripes horizontally and U (right) will have these stripes in a 45 degree angle. I think it’s quite incredible how exact the Pico frequency range and sensitivities are able to reproduce the input that went into creating the simulations. It’s of course more complicated with real data, but even Hans Kristian and Ingunn were surprised by such good results when they first saw them. 




  * MR begins reviewing his work: 
    * Foreground separation is model independent. There is no need to model the sky with "models other than the input".
    * Most recent update is Nov. 12, {{ ::nilc_cmilc_pico_10jan2021.pdf |but here are these same slides without the 'full delensing' likelihoods.}}