Telecon Notes June 25 2020

Link to main foregrounds telecons page
Link to logbook

Attendance: Jacques, Mathieu, Ragnhild, Shaul
Regrets:

Notes: Shaul

• Ragnhild ran 7 simulations
  • all 90.92; all smoothed to 60' and nside = 256
  • all have 100 lensing
  • Four with r=0.003, three with r=0.
  • She fits 10 parameters per pixel: cmb, 2xA_{dust}, 2xT_{dust}, 2xbeta_{dust}, A_{synch}, beta_{synch}, curv_{synch} (two more parameters than presented in last week's slides
  • She is now applying the mask and in-painting
  • Next steps: finish inpainting and calculate spectra
• SH: what about r and sigma r estimates? Ragnhild not sure. Action: Ragnhild will check with Ingunn and Hans-Kristian
• Question: what about delensing? We need to do the delensed case, but perhaps also the full lensed, so that we have a baseline. Action: Shaul will review situation with level of delensing
• Mathieu Update: Mathieu shows r posteriors for the foreground residuals presented in the PICO report. The values are consistent with zero, which indicate no bias from foreground residuals after cleaning.
  • He is showing results from 6 realizations.
  • Question: where do we move next? More realizations? removing frequency bands? Perhaps run models with r=3e-3?
  • SH: If we are constraining r, what should the fiducial value be? Is 3e-3 a good value to use?
  • Jacques and MR say r=0 and r=0.003 are two reasonable markers.
  • Action: SH will investigate which value to use with other people
    • SH Follow Up: wrote to Raphael. The response is attached below
    • I think actual results will tell us where to go. Let's check what is sigma_r with r=0.003. Perhaps we'll only barely get a detection?
  • MR: over few weeks I'll dig back into my code and potentially implement improvements.

From Raphael (June 25, 2020)

Hi Shaul,

I think r=0.003 is a good choice, and I think should be one of the examples. In addition, it could be good to have an example that is inaccessible from the ground but could be detected with PICO, e.g. 5e-4. Right now I'm not sure if it would be strategically better to tie that to a model, like the Goncharov-Linde model (which would be around 5e-4), or be more agnostic and just pick a number around there. Would three fiducial values be too many? For 0.003, I think the appeal would be to show that one can really slice things in many different ways and make sure everything is consistent. For 5e-4 the appeal would be that it's something that is out of reach for any other experiment.