CosmoStat/Shear-and-PSF-Reading-Group

PSF modelling notes

Opened this issue · 2 comments

The objective of this issue is to gather notes on the various methods for modelling a PSF discussed during the reading group.

@sfarrens notes:

  • PSFEx (for HSC)
    • pixel-based PSF from single expsoures
    • Co-add PSFs
    • Pixels of postage stamp used as a PSF model
    • What are basis vectors for pixel-based PSF?
      • Features of the PSF that you learn
  • LensFIT
    • PSFEx only works on independent images (or CCDs)
    • LensFITS can handle the full field of view
    • Similar basis to PSFEx
    • Weighting of stars
    • Intra-pixel shift
  • PIFF (for DES)
    • Similar to PSFEx
    • Modelling in sky coordinates instead of pixel coordinates
      • because of astrometric distortions
      • smoother variations in sky coordinates?
    • Basis-function polynomial for interpolation of PSF model
    • Delayed solution
      • don't need to solve for the PSF at the position of each star
      • use common information for all single exposures to improve model
    • Tree ring pattern
      • strong for some CCDs, but not for others
    • Gaussian processes not implemented for Y3
      • potential best solution for atmospheric distortion
    • PIFF has a modular implementation
  • MCCD
    • Aim to capture global and local features
    • Denoising terms
    • Radial Basis Function (RBF) interpolation of weights to galaxy positions
  • Discussion
    • Hard to make pixel grid that makes sense for entire focal plane
      • sky coordinate grid potentially a better solution, but this is on a sphere
      • CCD-specific models important for discontinuities

Proposed Subsections

  • Break sections into rough approaches
    • Can mention specific codes under each approach with some method specific details
    • References to codes and papers
  • PIFF image coordinates/world coordinates in separate section (what name?)
  • Interpolation separate or included with approach?
    • Plan A to keep interpolation integrated with the modelling approach