Chicago-2016: r Forecasting

From CMB-S4 wiki
Jump to: navigation, search

Next-generation "r" forecast planning

Moderator: Raphael Flauger


Status Updates

  • David Alonso (8 min) File:SimulatedR DavidAlonso.pdf
    • Simulations with PySM
      • Basics: Lensed CMB, sync, 1-component dust
      • More complicated options: 2-component dust, polarized AME
      • Foregrounds based on templates, but add Gaussian small-scale power
    • Map-based foreground cleaning
      • Fit for foreground spectral indices, amplitudes
    • Power spectrum and parameter estimation
      • Gaussian likelihood, ell > 30
    • Explored non-flat noise, parametrized with ell_knee.
      • Need to keep ell_knee in the [10,100] range.
    • 2-component dust doesn't bias r strongly (when analyzed assuming 1-component)
    • AME polarized at 2% can cause significant bias.
  • Victor Buza (8 min) File:Perfomance based sigma(r) projections.pdf
    • Fisher optimization framework -- spectral-based, parametrized foreground model
      • Dust and synchrotron foreground model, includes dust decorrelation parameter
      • Choose eight observing frequencies for CMB-S4 plus "delensing" band
      • Bandpower covariance is derived from BICEP/Keck performance, includes reality factors from detector yield, atmosphere, etc.
      • Optimization distributes effort between various frequency bands to get best sigma(r)
    • Generated strawman survey definition that has been used for projections by various groups.
      • Heterogenous survey with 500k detectors x 4 years, divided between deep and wide surveys.
      • Deep survey has low-ell component for r and high-ell component for delensing. Delensing effort is about 30% of total.
    • Work in progress
      • Need to vary foreground amplitude with fsky -- currently assumes BICEP-level foregrounds at all fsky.
      • Add foreground residuals to account for imperfect modeling.
    • Map-based mock trials
      • Can provide N_ell, f_sky(ell) curves for survey definition.
      • Validation of component separation frameworks via these maps.
  • Josquin Errard File:Forecast complex fgs.pdf
    • Updated forecasting framework. Now includes semi-analytic estimates of bias due to mismodeling.
  • Blake Sherwin (8min) File:Delensing residuals.pdf
    • Delensing reduces sample variance, but also changes the bias in C_l^BB due to lensing power.
      • What if we don't accurately understand the mean lensing residual after delensing? -> leads to bias in r.
    • Biases in the lensing reconstruction can be multiplicative or additive.
    • Pessimistic case: added 5% biases randomly from ell-to-ell
      • Encouraging result: residuals are small and don't look like r.
    • Would be good to have a parametrization of bias to marginalize over.

Going Forward

  • Cross-checks of frameworks with map-based mocks
    • Model of foregrounds
    • Noise model
    • Systematic effects
  • How do we interface with instrumentation working groups (and are we ready)?
    • Make sure that they are getting latest results.
    • Do we need additional justification for splitting atmospheric windows.
  • How do we coordinate with non-r forecasts?
  • Can we get groups other than BICEP to release noise curves? (Please?)
    • Get agreement on how to specify detector-years. Use actual deployed detectors (not good yield) and actual calendar time (not just data passing cuts).
  • Are there updates on sky models?
    • PSM v2.0 will become public very soon.
    • PySM new version
    • Boulanger/Ghosh dust model: 3D, multi-layer
  • Delensing and foregrounds, masking, etc.
    • Model for how high-ell errors in lensing reconstruction turn into low-ell residuals. Then can better fold delensing into optimization.