Difference between revisions of "ForecastingStep1"
From CMB-S4 wiki
Jump to navigationJump to search (→Notes) |
(→Notes) |
||
Line 192: | Line 192: | ||
*From Alessandro: I can reproduce the scatter in our results by slightly changing my parameter steps or going from a 5 point to a 3 points formula for the derivatives. It seems intrinsic in the Fisher technique, At least for me the matrix inversion is very sensitive to small changes in its parameters in particular for those with strong degeneracies. | *From Alessandro: I can reproduce the scatter in our results by slightly changing my parameter steps or going from a 5 point to a 3 points formula for the derivatives. It seems intrinsic in the Fisher technique, At least for me the matrix inversion is very sensitive to small changes in its parameters in particular for those with strong degeneracies. | ||
*From Mat/Neelima: The choice of fiducials also seems to matter a bit. For example, a small shift in fiducials gives about a ~10-15% change in the mnu error. | *From Mat/Neelima: The choice of fiducials also seems to matter a bit. For example, a small shift in fiducials gives about a ~10-15% change in the mnu error. | ||
+ | *From Joel & Alex: Is it appropriate to do lmax(TE) = 5000 if we have lmax(TT) = 3000? We have done this, but weren't sure of the reasoning here. |
Revision as of 19:07, 26 April 2016
Checking forecasting outputs
This should be run for the test case for S4+Planck. This is T/E/B/kappa. No clusters or BAO at this point (can be posted in other entries though). Settings described here and here:
- S4 = single channel (e.g. 150 GHz) at 1 uK/amin in T and 1.4uK/amin in P, 3 arcmin resolution.
- White noise, no FG inflation
- S4 T/E/B/k over 40% of sky, 30<ell < lmax
- Planck TT/TE/EE from 30<l<2500 over additional 20% of sky. Can use these 'Planck-pol' specs for noise:File:Planck pol.pdf
- Planck TT at l<30 over 80% of sky
- Tau prior 0.06+-0.01
- lmax(TT)=3000 unless explicit foreground cleaning is done in code for kSZ etc
- lmax(TE,EE)=5000 unless explicit foreground cleaning done in code
- kk reconstructed from 30<l<lmax using MV estimate
- quadratic estimator for lensing, ideally with iterative delensing
- Gaussian likelihood neglecting T/E/k covariance is ok, but non-Gaussian better
- non-linear power spectrum for kk, e.g. ok to use halofit in CAMB
- if easy to do, use pivot k=0.05
- don't need to use same fiducial, but useful to note what you have used.
obh2 | och2 | 100 theta | 10^9 As | ns | tau | hubble | |
Stephen/Josquin | 0.0222± 0.000022 | 0.1197±0.00053 | -- | 2.20±0.020 | 0.9655±0.0017 | 0.06±0.0052 | 67.74±0.20 |
Alex/Joel/Dan | 0.022200±0.000029 | 0.1197±0.00059 | -- | 2.196±0.021 | 0.9655±0.0019 | 0.0600±0.0056 | 67.50±0.22 |
Mat/Neelima/Nam | 0.0222±0.00003 | 0.1197±0.00058 | -- | 2.20±0.021 | 0.9655±0.0019 | 0.06±0.0056 | 67.31±0.22 |
Erminia/Jo/Danielle | 0.0222±0.00003 | 0.1197±0.00062 | 1.0459±0.00009 | 2.20±0.021 | 0.9655±0.0020 | 0.06±0.0052 | |
Cora/Kimmy | 0.02225±0.00003 | 0.1198±0.00055 | -- | 2.207±0.019 | 0.9645±0.0018 | 0.06±0.0052 | 67.27±0.17 |
Renee/Doddy/Dan | 0.02222±0.00003 | 0.1197±0.00064 | -- | 2.20±0.021 | 0.9655±0.0020 | 0.06±0.0058 | 69.0±0.25 |
Alessandro | 0.02222±0.000028 | 0.1197±0.00057 | -- | 2.196±0.020 | 0.9655±0.0018 | 0.06±0.0055 | 67.5±0.21 |
obh2 | och2 | 100 theta | 10^9 As | ns | tau | mnu (meV) | hubble | |
Stephen/Josquin | 0.0222± 0.000022 | 0.1197±0.00060 | -- | 2.20±0.035 | 0.9655±0.0019 | 0.06±0.0083 | 60±62 | 67.74±0.71 |
Alex/Joel/Dan | 0.022200±0.000030 | 0.1197±0.00071 | -- | 2.196±0.039 | 0.9655±0.0022 | 0.0600±0.0089 | 60±75 | 67.50±0.88 |
Mat/Neelima/Nam | 0.0222±0.00003 | 0.1197±0.00071 | -- | 2.20±0.039 | 0.9655±0.0020 | 0.06±0.0089 | 60±70 | 67.31±0.84 |
Erminia/Jo/Danielle | 0.0222± 0.00003 | 0.1197±0.00077 | 1.046±0.00011 | 2.20±0.036 | 0.9655±0.0022 | 0.06±0.0084 | 60±68 | |
Cora/Kimmy | 0.02225±0.00003 | 0.1198±0.00065 | -- | 2.207±0.037 | 0.9645±0.0019 | 0.06±0.0086 | 58±64 | 67.27±0.73 |
Renee/Doddy/Dan | 0.0222± 0.00003 | 0.1197±0.00077 | -- | 2.20±0.037 | 0.9655±0.0022 | 0.06±0.0086 | 60±71 | 69.0±0.85 |
Alessandro | 0.0222± 0.000029 | 0.1197±0.0007 | -- | 2.196±0.036 | 0.9655±0.002 | 0.06±0.0086 | 60±64 | 67.5±0.78 |
Notes
- From Jo: got these errors for Planck-alone for LCDM: 0.00017, 0.0014, 0.00047, 0.039, 0.004, 0.01.
- From Alessandro: I can reproduce the scatter in our results by slightly changing my parameter steps or going from a 5 point to a 3 points formula for the derivatives. It seems intrinsic in the Fisher technique, At least for me the matrix inversion is very sensitive to small changes in its parameters in particular for those with strong degeneracies.
- From Mat/Neelima: The choice of fiducials also seems to matter a bit. For example, a small shift in fiducials gives about a ~10-15% change in the mnu error.
- From Joel & Alex: Is it appropriate to do lmax(TE) = 5000 if we have lmax(TT) = 3000? We have done this, but weren't sure of the reasoning here.