
CosmoCoffee

Authors:  C.L. Kuo (1,2), P.A.R. Ade (3), J.J. Bock (1,2), J.R. Bond (4), C.R. Contaldi (4,5), M.D. Daub (6), J.H. Goldstein (7), W.L. Holzapfel (6), A.E. Lange (1,2), M. Lueker (6), M. Newcomb (8), J.B. Peterson (9), C. Reichardt (1), J. Ruhl (7), M. 
Abstract:  We report improved measurements of temperature anisotropies in the cosmic
microwave background (CMB) radiation made with the Arcminute Cosmology
Bolometer Array Receiver (ACBAR). In this paper, we use a new analysis
technique and include 30% more data from the 2001 and 2002 observing seasons
than the first release to derive a new set of bandpower measurements with
significantly smaller uncertainties. The planetbased calibration used
previously has been replaced by comparing the flux of RCW38 as measured by
ACBAR and BOOMERANG to transfer the WMAPbased BOOMERANG calibration to ACBAR.
The resulting power spectrum is consistent with the theoretical predictions for
a spatially flat, dark energy dominated LCDM cosmology including the effects of
gravitational lensing. Despite the exponential damping on small angular scales,
the primary CMB fluctuations are detected with a signaltonoise ratio of
greater than 4 up to multipoles of l=2000. This increase in the precision of
the finescale CMB power spectrum leads to only a modest decrease in the
uncertainties on the parameters of the standard cosmological model. At high
angular resolution, secondary anisotropies are predicted to be a significant
contribution to the measured anisotropy. A joint analysis of the ACBAR results
at 150 GHz and the CBI results at 30 GHz in the multipole range 2000 < l < 3000
shows that the power, reported by CBI in excess of the predicted primary
anisotropy, has a frequency spectrum consistent with the thermal
SunyaevZel'dovich effect and inconsistent with primary CMB. The results
reported here are derived from a subset of the total ACBAR data set; the final
ACBAR power spectrum at 150 GHz will include 3.7 times more effective
integration time and 6.5 times more sky coverage than is used here. 

[PDF]
[PS] [BibTex] [Bookmark]

View previous topic :: View next topic 
Author 
Message 
Antony Lewis
Joined: 23 Sep 2004 Posts: 1328 Affiliation: University of Sussex

Posted: November 08 2006 


This paper gives a detailed analysis of 2002 ACBAR data, including a nice treatment of the SZ effect, which they claim is the most likely explanation of the excess power seen by CBI on small scales.
1. What's the status of the rest of the data? (out soon, or another three years?)
2. Parameters constraints seem to have τ = 0.097± 0.014, an error less than half the size of that usually obtained from WMAP3. This doesn't look consistent with Fig 5.
3. Are CosmoMC data files available with recommended settings for a nonSZ analysis? 

Back to top 


Hans Kristian Eriksen
Joined: 25 Sep 2004 Posts: 58 Affiliation: ITA, University of Oslo

Posted: November 08 2006 


Antony Lewis wrote: 
2. Parameters constraints seem to have τ = 0.097± 0.014, an error less than half the size of that usually obtained from WMAP3. This doesn't look consistent with Fig 5.

I may be wrong here, but several parameters look a little funny, I think. First, n_{s} for the updated WMAP likelihood is listed to be 0.966, while Jostein Kristiansen found 0.960 in his test run, I believe. At least a part of this could be due to inclusion of lensing, though.. Second, the marginalized likelihoods in Figure 5 have too many bumps and wiggles on them for my taste.
All in all, I have a slightly uneasy feeling about convergence in these results. The authors say in the text that they run CosmoMC until "the largest eigenvalue of the GR test is less than 0.1". Not sure what eigenvalue means in this respect, but if it's simply the GR statistic as outputted from CosmoMC, then I think 1.1 is a little too liberal. For publications, I generally prefer R < 1.01. But of course, I may be entirely wrong here, and perhaps eigenvalue means something different in this respect.
I also noted that there seems to be a trend for high values in Figure 3 – except for the fifth Acbar data point, which is spot on the theoretical curve, the first nine points are high. Any idea what this could be due to? Statistical fluctuation seems unlikely..? 

Back to top 


Carlo Contaldi
Joined: 19 Nov 2004 Posts: 16 Affiliation: Imperial College

Posted: November 08 2006 


Quote:  I may be wrong here, but several parameters look a little funny, I think. First, ns for the updated WMAP likelihood is listed to be 0.966, while Jostein Kristiansen found 0.960 in his test run, I believe. At least a part of this could be due to inclusion of lensing, though.. Second, the marginalized likelihoods in Figure 5 have too many bumps and wiggles on them for my taste.

The value of n_{s} is due to lensing as you mentioned. We get 0.960 without lensing for the WMAP3 (updated likelihood) only case.
Quote:  Second, the marginalized likelihoods in Figure 5 have too many bumps and wiggles on them for my taste...All in all, I have a slightly uneasy feeling about convergence in these results. The authors say in the text that they run CosmoMC until "the largest eigenvalue of the GR test is less than 0.1". 
The largest GR eigenvalue should read 0.01 in the text, this was our criterion for convergence. The current version's 0.1 is a typo. e.g. WMAP3 only has 0.008.
Quote:  I also noted that there seems to be a trend for high values in Figure 3 – except for the fifth Acbar data point, which is spot on the theoretical curve, the first nine points are high. Any idea what this could be due to? Statistical fluctuation seems unlikely..? 
In power the calibration error is 12% so you should bear this in mind when doing a χ^{2}byeye since we haven't rescaled the power spectrum in any of the plots. 

Back to top 


Carlo Contaldi
Joined: 19 Nov 2004 Posts: 16 Affiliation: Imperial College

Posted: November 08 2006 


Antony Lewis wrote: 
2. Parameters constraints seem to have τ = 0.097± 0.014, an error less than half the size of that usually obtained from WMAP3. This doesn't look consistent with Fig 5.

Thanks for spotting that Antony! The errors in the table for τ are a typo, they should be around 0.03 as you said. Figure 5 (and median values of τ) are correct. τ line of table 4 should read
.
All other numbers are unaffected. 

Back to top 


William Holzapfel
Joined: 08 Nov 2006 Posts: 1 Affiliation: University of California, Berkeley

Posted: November 09 2006 


Antony Lewis wrote:  This paper gives a detailed analysis of 2002 ACBAR data, including a nice treatment of the SZ effect, which they claim is the most likely explanation of the excess power seen by CBI on small scales.
1. What's the status of the rest of the data? (out soon, or another three years?) 
I suppose there is no good excuse for taking three years to publish this. As it turned out, we spent a long time sorting out the details of the calibration. In the complete data set, there is sufficient sky coverage to calibrate directly from WMAP>ACBAR and we anticipate that this will go more smoothly. It seems reasonable that we will have the final power spectrum out in ~ six months. 

Back to top 


Hans Kristian Eriksen
Joined: 25 Sep 2004 Posts: 58 Affiliation: ITA, University of Oslo

Posted: November 09 2006 


Carlo Contaldi wrote:  Quote:  I may be wrong here, but several parameters look a little funny, I think. First, ns for the updated WMAP likelihood is listed to be 0.966, while Jostein Kristiansen found 0.960 in his test run, I believe. At least a part of this could be due to inclusion of lensing, though.. Second, the marginalized likelihoods in Figure 5 have too many bumps and wiggles on them for my taste.

The value of n_{s} is due to lensing as you mentioned. We get 0.960 without lensing for the WMAP3 (updated likelihood) only case.
Quote:  Second, the marginalized likelihoods in Figure 5 have too many bumps and wiggles on them for my taste...All in all, I have a slightly uneasy feeling about convergence in these results. The authors say in the text that they run CosmoMC until "the largest eigenvalue of the GR test is less than 0.1". 
The largest GR eigenvalue should read 0.01 in the text, this was our criterion for convergence. The current version's 0.1 is a typo. e.g. WMAP3 only has 0.008.

The marginalized distributions for WMAPonly still looked funny to me (take a look at the bumps at high values for tau and A, and the funny shape of n_{s}), so I started a run including lensing last night. The parameters I got were these:
Omega_b h^{2} 0.0224 +/ 0.0007
Omega_c h^{2} 0.107 +/ 0.008
theta 1.04 +/ 0.004
tau 0.0923 +/ 0.03
n_{s} 0.963 +/ 0.016
log(10^{10} A_{s}) 3.04 +/ 0.07
That is, the shift in n_{s} is smaller than what's listed in Kuo et al., where the value is 0.966, and tau is also smaller. Also, the distributions I get all look smooth and nice, without any bumps and wiggles. For reference, R−1 is 0.011 for this run.
Then again, I don't have a lot of experience with lensing, so I may have forgotten some switch – all I did, was to set CMB_lensing=T in params.ini.
Did you use the WMAP likelihood out of the box, so to speak, or did you make some other changes as well? 

Back to top 


Carlo Contaldi
Joined: 19 Nov 2004 Posts: 16 Affiliation: Imperial College

Posted: November 09 2006 


Hans Kristian Eriksen wrote:  Did you use the WMAP likelihood out of the box, so to speak, or did you make some other changes as well? 
The version we used may have some small differences from the lambda version as it comes directly from the development of the updated likelihood code that was done a couple of weeks ago at CITA by Mike Nolta and Jon Sievers. I will check against the official version to see if there are any differences.
There are some lensing accuracy tweaks that you can set in the CAMB params but I think that should not affect results for WMAP's relatively low l limit. 

Back to top 


Håvard Alnes
Joined: 20 Jan 2005 Posts: 5 Affiliation: University of Oslo

Posted: November 10 2006 


A stuipd question (probably): Why is the errorbar for the WMAP quadrupole extending below zero in figure 3? 

Back to top 


Kazuhide Ichikawa
Joined: 25 Mar 2006 Posts: 3 Affiliation: Institute for Cosmic Ray Research, University of Tokyo

Posted: November 12 2006 


I read the paper with interest but I could not follow some of the details.
1) When they say "WMAP3+ACBAR" in sec 6 and 7, do they use all the bands in table 3 ? I thought highest (4?) bands should not be used when SZ is not included. Or does it make no difference?
2) In sec 7.3 (or table 5), I infer "CMBall" does not include the highest band of CBI but "CMBall+BIMA" does. Am I correct?
In connection with this, I would like to know whether the constraint on α^{SZ} for "CMBall+BIMA" changes when BIMA is not used. 

Back to top 


Massimiliano Lattanzi
Joined: 04 Apr 2006 Posts: 2 Affiliation: Instituto de Fisica Corpuscolar  Valencia

Posted: November 14 2006 


Hi,
I was willing to run some cosmoMC chains using the new ACBAR data, and so I was trying to organize the data in a "cosmoMCfriendly" form, i.e. a form similar to the one of the datafiles included in the cosmoMc distribution like ACBAR_lge800_.dataset. However I have some questions on this:
1. According to the paper the calibration uncertainity is 6%, and it is said that this value is nearly unchanged from the 2002 data release. However the 2002 ACBAR data file in cosmoMC reads
calib_uncertainity = 0.2
so am I missing something or this is just a more conservative estimate?
2. I cannot find, neither in the website nor in the paper, the values of the beam uncertainities. Is this ok (i.e., we have to wait the collaboration to make them available) or again am I missing something? If the values are for the moment unavailable, can I simply put
beam_uncertainity = F
in the .dataset file?
Thanks
Massimiliano 

Back to top 


ChaoLin Kuo
Joined: 28 May 2005 Posts: 2 Affiliation: Stanford, KIPAC

Posted: November 16 2006 


Hi Massimiliano,
The central value didn't change much, but we improve its uncertainty.
The calibration uncertainty was 10% in the previous release,
in temperature unit. The new calibration uncertainty (6%) would
make calib_uncertainity = 0.12.
The beam uncertainty is 3%, as stated in the first paper.
Thank you.
ChaoLin 

Back to top 


Massimiliano Lattanzi
Joined: 04 Apr 2006 Posts: 2 Affiliation: Instituto de Fisica Corpuscolar  Valencia

Posted: November 16 2006 


Hi ChaoLin
Thank you very much for your answer. Things are clearer now.
However I still cannot exactly figure out where the numbers in the fourth column (beam uncertainity) of the ACBAR_lge800_.dataset file provided with cosmomc, come from. They all amount to a few percent of the measured value, but some are large as 7−8%.
Probably this will not make so large a difference in the final MCMC results, but I was wondering anyway. Can you or someone else clarify on this?
Thanks
Massimiliano 

Back to top 


Jason Dick
Joined: 08 Nov 2005 Posts: 11 Affiliation: SISSA

Posted: November 21 2006 


Håvard Alnes wrote:  A stuipd question (probably): Why is the errorbar for the WMAP quadrupole extending below zero in figure 3? 
It looks like those errors are symmetric about the data point, i.e. they are Gaussian errors. A more careful analysis would lead to a lower bound on the quadrupole that is closer to the data point than the upper bound, and above zero. 

Back to top 


Joe Zuntz
Joined: 22 Oct 2004 Posts: 9 Affiliation: UCL

Posted: November 21 2006 


Quote:  A more careful analysis would lead to a lower bound on the quadrupole that is closer to the data point than the upper bound, 
Yes.
Ideally, but not necessarily. Some methods can have error bars that legitimately extend below zero in high noise situations (essentially where you're finding the difference between two large numbers, data and noise, where you haven't perfectly estimated the noise). Bayesian approaches can deal with these situations with a sensible prior. 

Back to top 




You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum

