QUBIC

Minutes:

- The not realistic noise problem has been solved by Tom: the resolution of the n_sub_out maps built from n_sub_in maps is not trivial, because the n_sub_in maps that are combined have different resolutions.
--> the solution was to save the convolved n_sub_out maps instead of the n_sub_out maps at infinite resolution and then convolve them.

- AI tool: After some questions, ChatGPT gets less relevant in the free version, it could be useful to upgrade to the Team version.
--> Deepseek is a new contender, cheaper and open source: we will try it for now

- Preconditioner for the polychromatic map-making: the current preconditioner only works for monochromatic map-making or if n_sub=n_rec (it hasn't been coded yet)
--> it seems to be easily upgraded, it is not a fundamental problem but it has to be done
Remarks:

• Planck has to be taken into account, with H = vertical(H_QUBIC H_Planck)
• the goal of the preconditioner is to change the basis to the one where the noise covariance is diagonal: it shouldn't affect noiseless maps (right?) --> maybe not, because it may help convergence regardless (residuals might be treated as noise and it helps not getting stuck in local minimum)

- projection operators: Tom has made many tests and needs an external point of vue on the problem
Current notebook on tom-dev branch for peak amplitude (maybe related): qubic/scripts/Forecast_Bmodes/Nsub input & output/ProjectionOperator.ipynb
--> work on the minimal script for which the bug arises
--> dig into the mixing of Q & U during projection

- Moon data analysis: Alexandre adapted the beginning of JC code to create maps from TES TOD. Gaussian smoothing replaces spline/polynomial fitting. Goal would be to filter data in Fourier space.
--> iterative approach?
--> check resolution
--> compare with Eoife's work

• hard to remove Moon if Gaussian fitting

- 1D QUBIC simulation: JC works on it, there is still a problem in the reconstruction. A periodic feature (2 x distance between subpeaks) in the ratio of the average (over random pointings of the same sky) spectra reconstructed/real.

• in the code, a BI with Kmax=0 is just like an imager.

- Leonora has to finish her proceeding.

- Atmosphere:
Tom simulated monochromatic noiseless HEALPix atmosphere maps and reconstructed them.
He also did it with polychromatic atmosphere but this time the residuals are completely correlated with the signal
Wind is beeing implemented as an impact on the scanning strategy to reduce the computation time
--> with monochromatic, reduce the convergence tolerence from 10^{-10} to even less
--> with monochromatic, try to reconstruct atmosphere and CMB at the same time: use two different sets of coordinates (galactic and local)