UMICH-2015:ChallengeForegrounds

From CMB-S4 wiki
Jump to: navigation, search

Return to break-out session page

Foregrounds

What are our latest estimates of foreground levels post-Planck?

Steve, Raphael...

  • Steve Choi/Lyman Page Figures:

Choi.png

  • Stephen Feeney/Josquin/Hiranya/Andrew Jaffe Figure:

Errard.jpg [From Errard*, Feeney*, Peiris and Jaffe (*co-first authors), in prep 2015]

  • Raphael Figure:

Raphael2.png


  • John Ruhl: Can we make a new version of this figure (Steve could make input, Tom Crawford made it originally):

OldFG.png

  • Clem: We have not been able to exactly reproduce fig8 of PIPXXX. Here's a comparison:

Pipxxx fig 8.gifPipxxx fig 8 orig.png

  • A. Rotti and K. Huffenberger : We are using an isotropy violation estimator to trace foregrounds.

Bbpam.png

  • C. Lawrence : Figure 10 from Planck 45.

File:Planck 2015 Fig45.pdf


What simulations and tools do we have for forecasting impact on r? (detailed discussion on Tues)

Jacques, Raphael, Stephen F, Aurelien, Brendan, Jo...

  • Aurelien: Addressing questions about model uncertainty requires a flexible simulation facility, which can be easily modified and run by everyone for a wide range of parameters, and interfaced with current analysis and forecasting pipelines. Ideally would use python and have mechanism for the community to contribute to development and easily make and share own simulations. PSM is great, but how do we make the best use of it as a starting point.
  • Brendan: had a problem using the PSM during the BKP analysis – the dust polarization model in the PSM comes from a filtered version of Planck 353, so includes Planck noise, and there is nowhere in the PSM sky that is as clean as dust as the real Bicep region
  • We still still want ability to do quick and dirty Fisher forecasts for exploring different set-ups, so still want quick ways to inflate Nl for example. There is a useful prescription in upcoming paper, Feeney, Errard, Peiris, Jaffe 2015, that includes foregrounds and delensing [still makes assumptions about FG of course].
  • Feeney/Errard: useful 3 figures of merit and some estimates of residual r:

Fom.png


Noise.png

[From Errard*, Feeney*, Peiris and Jaffe (*co-first authors), in prep 2015] Here - follow the parametric max-likelihood formalism e.g. Errard et al (2011+2012). Not full simulations, but estimate the error bar on dust and synchrotron spectral indices in a Fisher way: error bars are derived using the semi-analytical expression of the second derivative of the likelihood, centered at the true values. Residuals are proportional to the square of these error bars, multiplied by the input foregrounds spectra. Dust temperature is fixed, indices constant over independent ~15deg angular size patches on the sky.


What are still physical unknowns?

Aurelien, (Brandon Hensley, Bruce Draine, Al Kogut), Brendan

  • They include magnetic dust, spinning dust polarization, spatial dispersion of dust temperature/emissivity, correlation of synch and dust, synchrotron index curvature...
  • Brendan: At what point we think that the greybody model for dust emission will break down?
  • Aurelien: worried by the spatial variations of dust properties. Taking these into account in forecasts is difficult, since they represent a failure in our theoretical understanding. How do we get around this? Is there a way to cook up a strategy, in terms of frequency and spatial coverage, that would make our forecasts somewhat immune to these uncertainties?
  • Jo: Don't yet know enough from Planck: dispersion in dust index between individual 300 deg2 (i.e. large!) patches is about +-0.15 and varies between beta = 1.1 and 2.2 (PIPXXII Fig 9b). And this is assuming single greybody model with fixed temp.

Pipxx11 fig9b.png


  • Clem: PIPXXX sec 2.2.1 says 1 sigma dispersion of B_d=0.17 for 10deg radius patches but "This is a conservative choice because this uncertainty includes the effects of noise in the data". Scaling to B2 patch size they say uncertainty of B_d=0.11. In this regard Fig E.1 of PIP-XXX is impressive. Bottom line: although there must be spatial variation of the spectral behavior of dust it's surprisingly small and the Planck data don't have enough s/n to measure it well.

Pipxxx fig e1.png


Do we need new data at <40 GHz and >230-300 GHz? And what about 60GHz?

Clem,John, Aurelien, Jo...

  • The effective polarized noise from cleaning 150 GHz with Planck 353 GHz is about 19 uK-arcmin, i.e. an additional sigma(r) ~0.01 for fsky=0.2
  • Al Kogut has done sims for PIXIE to look at possible residuals given varying dust temperature, showing that broader freq coverage is important
  • John: useful figure from BKP (updated):

Bkp fig13.png

  • Jo: pessimistic/realistic inflation of CMB errors using map-based cleaning if we loosen dust assumptions:

Index.png

  • Aurelien: For 60 GHz, want something there in the long term to understand dust / synchrotron correlations. But in the medium term, how well can we do without it?
  • Clem: for given s/n it's better to constrain the dust at (e.g) 220 than 350GHz - the shorter the lever-arm the less the impact of spatial variation of the frequency spectrum.

How important is large sky area?

  • If we detect B-modes, more sky area will help with cosmic variance
  • Without foregrounds, fsky doesn't help you that much for a detection. But! required if we want detections from different regions of sky.
  • But if r is undetectably low, strongest upper limits come from small f_sky.
    • Is it clear which direction this optimization is pushed by foregrounds? Systematics?

What is the path to a convincing claim that a B-mode signal is not foregrounds?

  • Nulls between frequencies, areas, scales