On this page... (hide)

  1.   1.  Surveys reuslts bias test :
    1.   1.1  LiB :
    2.   1.2  Biais ?
  2.   2.  NaMaster dependence over r :
  3.   3.  Pure Tests :
  4.   4.  Estimation and Variance
  5.   5.  Vairance ratio :

1.  Surveys reuslts bias test :

1.1  LiB :

  • Residual computed from {$ R = (\hat C_\ell - C_\ell^{th}) / std(\hat C_\ell) \cdot \sqrt{n_{MC}} = (\hat C_\ell - C_\ell^{th}) / std(\langle \hat C_\ell \rangle )$}
LiBBICEP
  • with naked eye, we already see that LiB seems unbiased, while Bicep is biais mainly for the EE spectrum.

1.2  Biais ?

  • Computing the number of point at 1, 2 or 3 sigma, that is to say {$ (|R|< \sigma)/n_{bins} $} with {$ \sigma = 1, 2, 3 $} : expecting the 68-95-99.7 rule.
  • 100000 simulations

LiB :

 noise1 sigma2 sigma3 sigma
R (EE)0.10 muK1s 0.8042s 0.9783s 1.000
R (EE)1.00 muK1s 0.8262s 0.9783s 1.000
R (EE)10.00 muK1s 0.8042s 0.9783s 1.000
R (BB)0.10 muK1s 0.8482s 0.9783s 1.000
R (BB)1.00 muK1s 0.7832s 0.9783s 1.000
R (BB)10.00 muK1s 0.7832s 1.0003s 1.000
  • conclusion : No bias found. The 68-95-99.7 rule is pretty well verified.

BICEP :

R (EE)0.10 muK1s 0.5562s 0.5563s 0.593
R (EE)1.00 muK1s 0.5562s 0.5563s 0.556
R (EE)10.00 muK1s 0.5562s 0.5563s 0.593
R (BB)0.10 muK1s 0.8152s 0.8523s 0.889
R (BB)1.00 muK1s 0.8152s 0.8523s 0.889
R (BB)10.00 muK1s 0.8522s 0.9263s 0.926
  • conclusion : bias found fo EE. This si due to the theoretical binned spectrum approximation {$ C_b = P_{b \ell} \cdot C_\ell $} and the fact that the EE spectrum is not flat enought.
  • conclusion : less bias found fo BB. This si due to the theoretical binned spectrum approximation {$ C_b = P_{b \ell} \cdot C_\ell $}, but the BB spectrum is sufficiently flat.
  • mode per mode :
 Bins number01234567891011121314151617181920212223242526
 Bins val8.522.536.550.564.578.592.5106.5120.5134.5148.5162.5176.5190.5204.5218.5232.5246.5260.5274.5288.5302.5316.5330.5344.5358.5372.5
R (EE)0.10 muK9967.879.9-11.0-0.8-7.1-8.2-7.6-6.5-6.4-4.30.13.314.625.332.532.627.115.59.33.0-1.2-5.0-7.9-7.9-24.641.8-511.8
R (EE)1.00 muK9822.479.8-11.2-0.9-6.8-8.2-7.5-6.2-6.4-4.00.13.314.525.231.932.326.715.59.53.6-0.7-3.8-6.8-5.6-26.966.5-575.9
R (EE)10.00 muK7827.676.8-12.2-0.6-5.8-7.5-7.1-5.5-5.6-3.50.02.712.620.625.326.620.913.78.23.8-0.9-2.0-4.7-2.2-24.771.8-407.8
R (BB)0.10 muK37.83.9-2.3-0.30.50.60.12.1-0.2-1.11.80.0-1.5-2.1-0.0-2.0-0.9-3.2-2.3-3.3-3.8-6.2-8.4-2.9-33.191.4-585.5
R (BB)1.00 muK31.14.8-2.7-0.10.50.9-0.32.30.1-1.11.90.3-1.3-1.70.4-2.70.6-3.6-1.2-1.8-2.7-5.6-4.6-2.5-23.041.7-211.3
R (BB)10.00 muK-46.928.6-10.33.5-0.4-0.10.41.30.50.2-2.11.10.70.60.6-1.8-0.4-0.90.50.0-0.3-3.8-1.0-2.6-4.6-1.7-17.2
  • first and last modes have to be neglected when computing a likelihood on 'r'.

2.  NaMaster dependence over r :

  • namaster behaves badly when r-> 0 :
  • Blue : ns=8
  • Green : ns=16
  • Red : ns=32
  • dashed black : BB model
  • y-label is wrong, it should be {$ C_\ell $}
r=0.1r=0.01r=0.001r=0.0
 
input EE = TE = 0
  • problem solved : mask aposisation incorrect.

3.  Pure Tests :

  • LiB, _Ns16, Nbins46, Slmax=3, fwhmdeg0.5, r0.001, fsky0.7
  • Color solid lines are the error of the spectrum std(Cl)
  • Dashed black line is the spectrum model
0.1 muK
1.0 muK
5.0 muK
50.0 muK

4.  Estimation and Variance

  • Leakage=False means ClEE =0
SurveyEEBB
LiB, fsky=0.5
LiB, fsky=0.6
BIC, "fsky=0.0159"

5.  Vairance ratio :

 fsky=0.5fsky=0.6
LiB EE
LiB BB