QUBIC

Minutes

• AI tool:
  - ChatGPT thinks too now (like DeepSeek)
  - the thinking process might not be useful for coding?
  - Copilot pro version from VS Code?
  --> if someone feels limited with free version of AI, we will discuss it again

• Preconditioner (Tom):
  - Tom implemented an idea from JC, still to be discussed with Leonora (the idea is to use one H per sub-bands)

• Projection operators (Leonora and Tom):
  - at first, the peaks are sorted by order and then, they get sorted by amplitude (after peak_angles)
  - peak_angles has a correction in 1/nu**2 to account for the solid angle of each subpeak but the amplitude scales with 1/nu, so this creates a pattern that might be normal
  --> JC asks Tom to try to compute TOD with all peak amplitudes but one to zero (start with central peak, then try other subpeaks) and see if TOD converge with Nsub increasing
  --> Tom keeps 99.8% of the power but doesn't choose which peaks to use: JS asks to choose always use the same subpeaks (for example, CMB fraction = 1)
  --> plot TOD for one peak at one frequency (Nsub values) and see if it converges or not

• Remarks on the code (Steve):
  - QubicNoise hard codes a dictionnary in the init function: bad practice
  - Qacquisition (and maybe others) classes don't have the same order for arguments

• 1D QUBIC:
  --> Tom proposes to try with infinite resolution (no convolution)