Friday 16 May 2014

Follow up on BICEP

The BICEP2 collaboration claims the discovery of the primordial B-mode in the CMB at a very high confidence level.  Résonaances recently reported on the chinese whispers that cast doubts about the statistical significance of that result.  They were based in part on the work of Raphael Flauger and Colin Hill, rumors of which were spreading through email and coffee time discussions. Today Raphael gave a public seminar describing this analysis, see the slides and the video.

The familiar number r=0.2 for the CMB tensor-to-scalar ratio is based on the assumption of zero foreground contribution in the region of the sky observed by BICEP. To argue that foregrounds should not be a big effect, the BICEP paper studied several models to estimate the galactic dust emission. Of those, only the data driven models DDM1 and DDM2 were based actual polarization data inadvertently shared by Planck. However, even these models suggest that foregrounds are not completely negligible. For example, subtracting the foregrounds estimated via DDM2 brings the central value of r down to 0.16 or 0.12 depending how the model is used (cross-correlation vs. auto-correlation). If, instead,  the cross-correlated  BICEP2 and Keck Array data are used as an input, the tensor-to-scalar ratio can easily be below 0.1, in agreement with the existing bounds from Planck and WMAP.

Raphael's message is that, according to his analysis, the foreground emissions are larger than estimated by BICEP, and that systematic uncertainties on that estimate (due to incomplete information, modeling uncertainties, and scraping numbers from pdf slides) are also large. If that is true, the statistical significance of the primordial B-mode  detection is much weaker than what is being claimed by BICEP.

In his talk, Raphael described an independent and what is the most complete to date attempt to extract the foregrounds from existing data. Apart from using the same Planck's polarization fraction map as BICEP, he also included the Q and U all-sky map (the letters refer to how polarization is parameterized), and models of polarized dust emission based on  HI maps (21cm hydrogen line emission is supposed to track the galactic dust).  One reason for the discrepancy with the BICEP estimates could be that the effect of the Cosmic Infrared Background - mostly unpolarized emission from faraway galaxies - is non-negligible. The green band in the plot shows the polarized dust emission obtained from the  CIB corrected DDM2 model, and compares it to the original BICEP estimate (blue dashed line).

The analysis then goes on to extract the foregrounds starting from several different premises. All available datasets (polarization reconstructed via HI maps, the information scraped from existing Planck's polarization maps) seem to say a similar story: galactic foregrounds can be large in the region of interest and uncertainties are large.  The money plot is this one:

Recall that the primordial B-mode signal should show up at moderate angular scales with l∼100 (the high-l end is dominated by non-primordial B-modes from gravitational lensing). Given the current uncertainties, the foreground emission may easily account for the entire BICEP2 signal in that region. Again, this does not prove that tensor mode cannot be there. The story may still reach a happy ending, much like the one of  the discovery of accelerated expansion (where serious doubts about systematic uncertainties also were raised after the initial announcement). But the ball is on the BICEP side to convincingly demonstrate that foregrounds are under control.

Until that happens, I think their result does not stand.

17 comments:

Nathaniel said...

I think the DDM2P is not a "CIB-corrected" version of the BICEP map extracted from the Planck polarization fraction plot (what BICEP called DDM2). Rather, in DDM2P Raphael constructs a new polarization fraction of his own from the Planck Q and U polarization maps, and uses T from public Planck data to get the rest of the denominator. Then that polarization fraction is input into a foreground analysis analogous to BICEP's DDM2 (which uses Planck dust map models, etc.). It doesn't have the CIB contamination but it's *not* a CIB-subtracted version of the original Planck polarization fraction plot. (The third plot, also confusingly labeled with Q/U, is an attempt to construct the whole dust foreground from the Q and U maps directly, using a procedure entirely different from BICEP.). As far as I'm aware, there is no actual demonstration that the BICEP foreground line came from the CIB-contaminated Planck polarization fraction, or what would happen if CIB were "subtracted" from that polarization fraction and the analysis repeated. So the commentary is not directly about the BICEP analysis itself.

Jester said...

But the title of that slide (no.38) is "CIB corrected foreground models"... The input that goes into DDM2P is listed on the previous slide. I agree it's definitely not just CIB subtraction of Bernard's plot. But my understanding is that this additional information allows him, in particular, to subtract CIB and get the pure dust polarization fraction (though it may be too naive).

Nathaniel said...

Sure thing, the DDM2P is not CIB contaminated by construction. So it's a bit of a pedantic distinction, but my take-home from this route is that caution is advised since it's sort of an apples-to-quinces comparison -- if it were really a CIB-subtracted version of the BICEP foreground analysis, then more dire conclusions would be warranted. We haven't established that BICEP actually used a CIB-contaminated map for their foreground, or that corrections to that hypothetical map are what's shown in the plots.

Anonymous said...

The difference with \Lambda is that once we learned \Lambda >0, many things fell into place. Here, as much as we'd LOVE to have observed 'r~0.2', this brings more trouble than solutions. For once, it does not explain the issues at low ell in WMAP/Planck, rather it exacerbates it. This is a funny entangled state between \Lambda and Pamela, not sure with which relative fraction :)

Anonymous said...

I'm not an expert, but to me the fact that the signal drops for l<100 somewhat points to an interpretation in terms of primordial B-modes.

Sesh Nadathur said...

I thought your rumours a few days ago were a bit over the top; having seen these slides I guess I'm now a bit closer to your opinion.

Regarding the CIB subtraction, my understanding is that Flauger has created his own version of the Bernard map using the Q, U and T data also taken from digitized slides, but with CIB subtracted, and then used this to get the DDM2P model, but that the conclusions from this don't match the conclusions from the DDM2 model used by BICEP. The indirect inference is that BICEP didn't do the CIB subtraction correctly - though perhaps there are other reasons for the difference?

Jester said...

Yes, my interpretation is similar. Raphael said in the talk that a significant CIB component is needed to make these various maps consistent. As far as i know, BICEP didn't do the CIB subtraction period, but then i was wrong about them admitting, so I may be wrong again...

Anonymous said...

A revised value of (the tensor to scalar ratio) r that is a factor ~2-4 below 0.16 would mean significant deviation(s)from the textbook quadratic inflation model.
Assuming the scalar spectral index remains close to 0.96, a value preferred by WMAP and Planck, we could hope in this case to gain some insights concerning the inflaton coupling to other fields.

Anonymous said...

Raphael did a wonderful job describing the issues. People wanted to focus on the distance of the points from the foreground uncertainty band, but in my mind what throws the BICEP2 results to the garbage is that the shape of measurements and that of the foreground are impressively correlated. I hope there is some accountability for what these guys did. It is a disfavor to the scientific community working away from huge press releases. But many have probably negotiated big positions at other universities. Sad state of affairs.

Anonymous said...

"but in my mind what throws the BICEP2 results to the garbage is that the shape of measurements and that of the foreground are impressively correlated."

What do you mean? All the cross-correlations that have been calculated between the B2 results and different foreground models, including DDM2 (irrespective of CIB subtraction) have shown very small correlation. You can argue about whether that is significant, but if anything the cross-correlations are in B2's favor.

Kasuha said...

I'm not an expert in the field so I may pretty much be wrong about it. But if I understood the talk correctly - and am reading the plot correctly - the foreground may lie anywhere in the shaded region, but cannot have any shape within the shaded region. For any given zero level it is just a line. And for any selected zero level, BICEP2 measurements still deviate significantly from that shape.

Anonymous said...

For reference, one should say that Slava Mukhanov is quoted at 29.4 by a german newspaper what he thinks on Bicep2:

http://www.sueddeutsche.de/wissen/entstehung-des-universums-risse-in-der-urknall-theorie-1.1945672

Whilst the headline of this article is stupid and wrong, Mukhanov probably stands behind these quotes of him, which reflect the usual teaching of him that he inheritec from is phd advisor Vitaly Ginzburg, who came from the Landau school:

“Complete nonsense” argues Mukhanov. "The journals are full of this but it remains nonsense." Mukhanov says that should leave the spectral index alone if you do not want to start a fight with him."

“They should not have made so much noise. It does not fit together. There are serious inconsistencies”…

So much for the russian position on the Biceps2 results.

Rastus Odinga Odinga said...

Game over?

http://arxiv.org/pdf/1405.5857.pdf

Anonymous said...

No. It's a surprisingly disappointing paper from people who should know better. They do a simple LCDM+r model with dust as the only additional parameter, never mind any extra parameter to relieve tension. Then they set a pessimistically flat prior that overweights low-ell TT in the analysis, so all of BICEP2 BB is dust more or less by assumption.

Anonymous said...

A string theorist talking about CMB polarized dust foreground! i hope the gamble is worth the risk Raphael! tenure positions are hard to come by...

RBS said...

BICEP2 saw something in certain patch of sky. Planck's can't tell how much there is dust and how much instrumental error. Nobody knows with any certainty how much there is dust to compare with BICEP2. Until somebody knows how much there is dust, nobody will know for certain whether BICEP2 is right or wrong. Am I missing anything?

Marni Dee Sheppeard said...

Rastus, thanks, that looks interesting for the r=0 fans.