[0712.1148] Detection of primordial nonGaussianity (fNL) in the WMAP 3year data at above 99.5% confidence
Authors:  Amit P. S. Yadav, Benjamin D. Wandelt 
Abstract:  We present evidence for the detection of primordial nonGaussianity of the local type (fNL), using the temperature information of the Cosmic Microwave Background (CMB) from the WMAP 3year data. We employ the bispectrum estimator of nonGaussianity described in Yadav et al. 2007b which allows us to analyze the entirety of the WMAP data without an arbitrary cutoff in angular scale. Using the combined information from WMAP's two main science channels up to l_{max}=750 and the conservative Kp0 foreground mask we find 26.9 < fNL < 146.7 at 95% C.L., with a central value of fNL=86.8. This corresponds to a rejection of fNL=0 at more than 99.5% significance. We find that this detection is robust to variations in l_{max}, frequency and masks. We conclude that the WMAP 3year data disfavors single field slowroll inflation. 
[PDF] [PS] [BibTex] [Bookmark] 

 Posts: 26
 Joined: November 14 2006
 Affiliation: ITP Heidelberg
 Contact:
[0712.1148] Detection of primordial nonGaussianity (fNL) i
Another notquite3sigma detection!!
Actually, this one looks as if the authors have taken some care to address the main potential problems: in particular foreground subtraction.
The main point in the analysis seems to be the twist of adding the 'linear term of the estimator' (unfortunately not defined or explained in the paper) to improve the statistical behaviour of the result.
The result only becomes significant when l's of up to 500 or so are included. (Nor does the paper explain how the choice of l_max is implemented.) So it is appaently not the case that most of the nonGaussianity is associated with largerscale CMB glitches or spots.
I guess the main scientific point is to prepare the way for an application of the method on future experimental data, rather than actually prove something about inflation.
Actually, this one looks as if the authors have taken some care to address the main potential problems: in particular foreground subtraction.
The main point in the analysis seems to be the twist of adding the 'linear term of the estimator' (unfortunately not defined or explained in the paper) to improve the statistical behaviour of the result.
The result only becomes significant when l's of up to 500 or so are included. (Nor does the paper explain how the choice of l_max is implemented.) So it is appaently not the case that most of the nonGaussianity is associated with largerscale CMB glitches or spots.
I guess the main scientific point is to prepare the way for an application of the method on future experimental data, rather than actually prove something about inflation.

 Posts: 183
 Joined: September 24 2004
 Affiliation: Brookhaven National Laboratory
 Contact:
[0712.1148] Detection of primordial nonGaussianity (fNL)
It seems to me that they are using many aposteriori claims. Why using V+W and not Q+V+W? If Q is contaminated, then why doesn't it show a very strong lmax dependence? And if not, why not using it? (ok, point sources will drive the bispectrum down, but SZ could drive it up, but it seems to be consistent with no more contamination than V+W).
Also, as one of my eminent colleagues pointed out, one cannot convert 2.89 sigma to significance level(i.e. one should pick the significance interval in advance, say 3 sigma and then count any >3sigma as true at 3 sigma, even though some will be 3.5 sigma), so 99.6% is aposteriori as well. If one does simply Q+V+W then the result is 2 sigma and so perfectly consistent with null. And as Andreas Albrecht famously paraphrased Linda Evangelista: "I am not going out of bed for less than 4sigma".
Also, as one of my eminent colleagues pointed out, one cannot convert 2.89 sigma to significance level(i.e. one should pick the significance interval in advance, say 3 sigma and then count any >3sigma as true at 3 sigma, even though some will be 3.5 sigma), so 99.6% is aposteriori as well. If one does simply Q+V+W then the result is 2 sigma and so perfectly consistent with null. And as Andreas Albrecht famously paraphrased Linda Evangelista: "I am not going out of bed for less than 4sigma".
Re: [0712.1148]
I think they believe that Q is contaminated, not by point sources or SZ, but by diffuse foreground emission. So the contamination is already there in very low multipoles. I think this is consistent with how the Q+V+W [tex]f_{NL}[/tex] continues to rise with more aggressive masks, whereas the V+W [tex]f_{NL}[/tex] value seems to be leveling off at around the Kp0 sky cut.

 Posts: 59
 Joined: September 25 2004
 Affiliation: ITA, University of Oslo
 Contact:
[0712.1148] Detection of primordial nonGaussianity (fNL)
The first thing that struck me when reading this paper was the somewhat unusual set of cosmological parameters adopted for the reference model:
Parameter (Y and W) (WMAP; Table 2 of Spergel et al.)
Omega_b 0.042 0.041
Omega_c 0.239 0.237
h 0.73 0.732
tau 0.09 0.091
n 1 0.954
sigma_8 ? 0.756
That is, all parameters adopted by Yadav and Wandelt are quite similar to WMAP's bestfit, except n, which is quite significantly different. And the effect on the power spectrum from this difference is fairly dramatic. Unfortunately, the amplitude isn't listed by Yadav and Wandelt, so it's difficult to know if these differences show mostly up at high or low l's.
So, a question that keeps bothering me when reading this paper, is whether the observed effect is the result of an actual f_nl, or of a power spectrum mismatch. I know that the bispectrum should normalize out the power spectrum to some extent, but it's not clear to me how efficient this normalization is..
Anyway, if I were the referee for this paper, I'd probably ask the authors to redo the analysis, this time adopting the actual bestfit WMAP parameters. Then this wouldn't be an issue at all anymore.
Parameter (Y and W) (WMAP; Table 2 of Spergel et al.)
Omega_b 0.042 0.041
Omega_c 0.239 0.237
h 0.73 0.732
tau 0.09 0.091
n 1 0.954
sigma_8 ? 0.756
That is, all parameters adopted by Yadav and Wandelt are quite similar to WMAP's bestfit, except n, which is quite significantly different. And the effect on the power spectrum from this difference is fairly dramatic. Unfortunately, the amplitude isn't listed by Yadav and Wandelt, so it's difficult to know if these differences show mostly up at high or low l's.
So, a question that keeps bothering me when reading this paper, is whether the observed effect is the result of an actual f_nl, or of a power spectrum mismatch. I know that the bispectrum should normalize out the power spectrum to some extent, but it's not clear to me how efficient this normalization is..
Anyway, if I were the referee for this paper, I'd probably ask the authors to redo the analysis, this time adopting the actual bestfit WMAP parameters. Then this wouldn't be an issue at all anymore.