Link to main foregrounds telecons page
Link to logbook

Attendance: Ragnhild, Mathieu, Charles, Kris, Graca, Shaul
Regrets: Jacques

Notes: Shaul

• Both Ragnhild and Mathieu have more results. Start with Ragnhild.
• Ragnhild reports on a new set of results
  • After success with 90.91 she is working with 90.92 (SH in a follow up e-mail: how do we define success quantitatively?)
  • Slide 3: With 90.92 she is fitting 12 parameters (SH in a follow up e-mail: write a note defining parameters mathematically)
  • Slide 4(?): “high residuals”. (SH: are there higher residuals here compared to 90.91 earlier?)
    • Kris: Please improve color bars; increase font size; add units
  • Slide 6: Reconstructed AME has a lot of small scale structure. That should not be the case. Indicates that the model is not fit properly. Similar for synchrotron. Suggestive of issues at low frequencies. Or that more tweaking of parameters is required.
  • Slide 8: dust is fit very well (note: no small scale structure!)
  • Slide 9: increased chi^2 when 90.91 skies are run with 90.92 underlying model! Explanation: degeneracy between foreground components creates increased chi^2.
  • CL: 2% AME polarization is 'extreme'.
  • Kris: are we using AME temperature or just polarization? Ragnhild: just polarization.

• Nov. 3 e-mail with SH:
  • Regarding the chi^2:
    • One sign of a «good» chi^2 is that there is no big scale patterns in the maps, but rather white noise fluctuations. There will almost always be big scale patterns along the galaxy plane, so this is why we mask this out. I usually smooth the chi^2 maps to easier see if there are any large scale patterns.
    • When there are big scale patterns in the chi^2 maps, I try to recognize which foreground might have caused this pattern to see which foreground that needs to be fitted better. In the case of the real data from Pico, we would start by fitting the known foregrounds and then look at the chi^2 and the residuals if there might be signals that we’re not taken into account.
    • We have 21 input maps, one pr frequency. When I’m fitting for 6 parameters (synchrotron amplitude and beta, dust amplitude, beta and temperature and cmb) as for model 90.91, I will have 21-6=15 degrees of freedom and thus expect the chi^2 for a well fitted model to have an average value of 15. When I’m fitting for 12 parameters (synchrotron amplitude, curvature and beta, dust amplitude, beta and temperature for two dust models, AME amplitude and peak frequency and cmb) as for model 90.92, I will have 21-12=9 degrees of freedom and expect the average chi^2 value to be 9. However, counting degrees of freedom is complicated when informative priors are used since a free parameter then doesn't really count as a full degree of freedom. But it's an OK rule-of-thumb. The chi^2 value for high latitudes in model 90.91 is approximately 15, but the level is not as low as 9 for 90.92, so this can not be used as the only criterion
    • The numerical value for the chi^2 is difficult to see in the unsmoothed chi^2 maps, I will try to make this clearer in future presentations.

• SH: Slide 3: I notice that the masking now is 40%. There was less masking for 90.91. Shouldn't we use the same level of uniform masking? It's ok not to use the same level, but then we'll need to explain what is the criterion for masking.
  • Response: I have been testing different masks to check for biases. I generate the masks based on the chi^2 maps, and before I am happy with my results I will not be able to have the final mask. I think it’s best to use the same mask for both models and all final results published in a paper.