Is the Axis of Evil still a valid feature in the CMB data?

[Garth]

An eprint paper published today on the ArXiv and submitted to MNRAS: The Axis of Evil revisited by Land and Magueijo.

The analysis of the CMB data is statistical by nature and the existence or non-existence of features such as the 'Axis of Evil' is a matter of evaluating its statistical likelihood. Their highly technical paper re-examines the methods used and concludes

We have highlighted weaknesses with the original AOE statistic (2) that probed m-preference for ℓ = 2−5. These are
primarily:
1) lack of robustness: small changes in the data produce very different results, i.e., the statistics are discontinuous;
2) variations with data-set: it is hard to connect varying results to imperfections in the data or the statistic;
3) need simulations to assess significance: no way of penalizing for extra parameters or comparing competing theories, e.g., planarity V’s general m-preference.

We have found an improved formalism by employing a Bayesian approach, which cures the instabilities. These were due to the existence of multiple solutions for a given multipole. But bringing in a penalization related to the number of parameters of the model enforces “Occams Razor” and selects solutions where parameters are common between the multipoles.

The Bayesian approach also allows a better assessment of the relative evidence for planarity (correlation between ℓ = 2, 3, m′ = ℓ modes) and m-preference (a correlation between ℓ = 2 − 5, m′ not restricted). This extends the work of Magueijo & Sorkin (2006) where the low-ℓ low-power evidence was assessed, as well as planarity for some data-sets.

Comparing Table 2 and Table 5 we see that overall there is more evidence for m-preference than for planarity (the exceptions being the BIC for LILC1 and TOHc3), with slightly stronger evidence from the first-year data. We find that although the Bayesian method is more robust, the various information criteria reach different conclusions. The AIC generally provides strong evidence for planarity and decisive evidence for m-preference. The BIC provides varying results. This is to be compared to the evidence for scalar spectral index n_S not equal to 1, where conclusions also differ between IC methodology (Magueijo & Sorkin 2006).

To put it more simply:

Abstract
We find strong (and sometimes decisive) evidence for the “Axis of Evil” in almost all renditions of the WMAP data. However some significant differences between data-sets remain, and the quantitative aspects of the result depend on the particular information criteria used.

Garth

 

[cesiumfrog]

What is this "axis of evil" referring to?

 

[Garth]

'The Axis of Evil': refers to the fact that the low-l anisotropies in the CMB data, revealed both by COBE and now WMAP, appear to be in alignment with each other more than is to be expected from a purely random distribution. Moreover the axis thus obtained may also be aligned with local geometry, that is, of the solar system and our galaxy.

If you click on the link The Axis of Evil revisited and download the pdf document you will see whole sky diagrams showing these anisotropies.

There has been a lot of discussion in this Forum about whether the 'Axis' is a real artefact or just a statistical fluke, the present paper seems to suggest that, after rechecking very carefully, it may be real.

If it is real it may mean that the existing signal at low-l modes is partly or entirely due to local effects, in which case there is a significant loss of signal in the CMB anisotropy power spectrum from the largest primordial hotspots'.

If the Axis actually exists it leaves two questions:

1. What causes the alignment, Local Pancake Defeats Axis of Evil perhaps?

and

2. What causes the loss of signal at the largest angles, perhaps the universe is topologically a soccer ball? Apparently not: Extending the WMAP Bound on the Size of the Universe.

Garth

 

[Chronos]

Indeed, Garth. It is an interesting paper, but a bit more reserved than the original 'Axis of Evil' paper. The authors acknowledge that selection effects remain a possibility. Foreground noise is a very real problem when attempting measurements as fine as the CMB anisotropy. LISA is another example that comes to mind and I suspect it too will exhibit similar symptoms. But that may well be the deeper point the authors are trying to make - that the data may be contaminated and some of the conclusions drawn thus far may not be as secure as we think.

 

[Garth]

If the Axis actually exists it leaves two questions:

1. What causes the alignment, Local Pancake Defeats Axis of Evil perhaps?

and

2. What causes the loss of signal at the largest angles, perhaps the universe is topologically a soccer ball? Apparently not: Extending the WMAP Bound on the Size of the Universe.

Garth

Further confirmation of the latter point: Dodecahedral topology fails to explain quadrupole-octupole alignment.

The CMB quadrupole and octupole, as well as being weaker than expected, align suspiciously well with each other. Non-trivial spatial topology can explain the weakness. Might it also explain the alignment? The answer, at least in the case of the Poincare dodecahedral space, is a resounding no.

Garth

 

[Nereid]

Any thoughts on how well http://www.esa.int/esaSC/120398_index_0_m.html" will be able to constrain this?

And how closely do the polarisation results (from WMAP) align with the axis of evil (I just don't recall what the WMAP team reported, and am too lazy - now - to go look for myself)?

 

[Parlyne]

I attended a colloquium today (err, yesterday) given by one of the primary WMAP authors (David Spergel); and he addressed the question of the Axis of Evil. His claim on the matter is that it most likely arises from artifacts of the method used to remove microwave sources in the galactic plane from the data. If you look at an image of the WMAP data (such as this one), you'll notice that there's a particularly dark spot near the center of the image and a particularly bright spot a bit to its right. On these maps the horizontal is the galactic plane; so both of these regions are subject to whatever uncertainties there were in removing the galaxy. What Spergel pointed out was that if those two regions are removed from the analysis, the axis of evil disappears.

 

[Garth]

Would not removing the largest bright spots in the data make the low-l mode power spectrum even more deficient?

Garth

 

[Parlyne]

Would not removing the largest bright spots in the data make the low-l mode power spectrum even more deficient?

Garth

Honestly, I don't know what exactly it does to the low-l modes other than eliminating the alignment of the dipole, quadrupole and octopole moments. Really, though, if he's right that what's seen in those areas are artifacts of the method used to remove galactic emission, it's reasonable to suggest that it's not valid to draw conclusions from analyses that incorporate data from the galactic plane. As I understand it, the WMAP group only uses data from outside the galactic plane in their analyses specifically to avoid such problems.