Saturday, January 31, 2015

Planck+BICEP2+KeckArray: diluting the discovery

No certainty yet about the validity of the BICEP2's 2014 discovery

Two or three days before the preprint will appear on the arXiv, BICEP2/KeckArray have released their joint analysis with Planck:
Last March, the discovery papers were and still are offered at BICEPkeck.ORG.

The new joint paper is framed as a discovery of the B-modes at 7 standard deviations but these B-modes are presented as "dust and gravitational waves combined". This literally means that all probabilities of their model are obtained as sums (integrals, the so-called marginalization) over models with all possible values of the tensor-to-scalar ratio, \(r\).

When the high-depth, window-only BICEP2 data are mixed with the low-depth, global Planck data – just like if you dilute your favorite whisky in water at a 100-to-1 ratio, it becomes impossible to discover the primordial gravitational waves separately, the paper confirms. Instead, one may impose one of the usual Planck-like upper bounds on \(r\), namely \(r\lt 0.12\) at the 95% confidence level. This is slightly higher – more signal-like – than all the previous upper bounds coming from Planck.

Needless to say, virtually all science inkspillers totally misinterpret the new release when they write (or spill) say that "it disproved the gravitational waves". It didn't disprove a damn thing. It confirmed the discovery of the B-modes but was unable to reproduce the discovery of the primordial gravitational waves themselves.

I must mention Dennis Overbye of the New York Times as a positive exception. He really is a class above almost all of his fellow writers and his article – and the corrected title (with the word "obscures" instead of "rebuts") – is more accurate than others. Still, the article is bad, with weird claims such as "the paper has no evidence of inflation". Wow, how can say such a thing? Even in the absence of a measurement of an exact \(r\), this is a nontrivial paper verifying many features of inflation.

I don't see how this paper differs from the numerous informal "combos" of the BICEP2 and Planck datasets done by the theorists, for example from the paper by Colley and Gott that concluded that the primordial gravitational waves are shown there (in the combination of the experimental datasets) and \(r=0.11\pm 0.04\) because the B-mode signal seems to be about half dust, half gravitational waves.

Your humble correspondent doesn't feel comfortable about this very method of "forced collaboration" proposed by the "stronger team that hasn't discovered anything". I think it's a tricky way for Planck to "link itself" to the discovery when it's done again even though this collaboration has clearly nothing to do with the discovery.

Well, I still think that the primordial gravitational wave signal is there because of the clear "local maximum" of various BB-spectra and cross-correlations near \(\ell\approx 100\) which no existing model without primordial gravitational waves seems to predict. Lensing of dust only gives one monotonic curves in this interval of \(\ell\).

Ron Cowen of the tabloid named Nature will be my exemplary inkspiller of choice. His title is "Gravitational waves discovery now officially dead". That surely sounds impressive or funny, depending on your point of view, but science cannot make discoveries "officially dead".

The last time when someone tried to make a discovery "officially dead" was in 1633 when Galileo Galilei was was forced to participate in a trial that "officially" decided that heliocentrism was wrong. "Eppur si muove," Galileo mentioned (although this proclamation was probably just silent, because he would have been executed if he were loud). And no doubts about it, the Earth continued to revolve, indeed! ;-)

Science obviously doesn't care about "official authorities" in this sense.

So sorry, this is a different paper by a different collaboration using a different method that got (somewhat) different results and it in no way proves that the March 2014 discovery was wrong. And the hypothesis that \(r\approx 0.12\), for example, is compatible with both powers up to approximately 2-sigma deviations and in this sense, none of the two papers "disproves" the other one.

The arrogance of some people with which they make themselves "official" is manifesting itself at other points of Ron Cowen's rant, too. He teams up with an equally arrogant member of Planck:
Planck researcher George Efstathiou of the University of Cambridge, UK, says that the conclusion is no surprise. “I don’t know why people are so excited,”he says. “There’s nothing dramatic here form the science point of view.”

Efstathiou characterized the 29 January leak as both “unfortunate” and “careless”. “The BICEP people felt that they had not handled the publicity right the first time” and that everyone was eager to make public the joint analysis in a more cautious manner.
The results of the joint analysis are not a surprise because to dilute BICEP2's signal with Planck's non-signal was the very plan of this joint analysis from the beginning. People are still excited by the BICEP2's discovery of the primordial gravitational waves because if it is true, it is one of the greatest experimental discoveries in cosmology for decades.

And what about the "leak" stuff? This is clearly a completed paper that was, according to all the information I can have, finalized for the arXiv submission. Mr Efstathiou is just trying to act as a bully and show his loud voice and "muscles" because that's how he may be more visible than by doing research in cosmology.

BICEP2 + Keck Array are doing the right thing by not being secretive and by avoiding bureaucratic unjustified delays before they provide the scientific public with the results of their work. Mr Efstathiou should learn to do it in this way, too.


  1. If it's true it sounds like a dirty game for science..As you mentioned in a previous post the preferred lenght scale found in the B-modes from bicep2 still remain something to think about..Maybe it's a laymen question but i just don't understand how can they say that the contribution from primordial gravitational waves seem to be the half of what they thought before and then say the signal is apparently all due to dust..Maybe im just misunderstanding something

  2. Dear Andrea, thanks for the comment!

    No paper shows that "everything is due to dust". This is a pure fabrication of some science writers.

    The conclusion in the papers that make no discovery, such as the new joint paper, is that there is *some* contribution from the dust, and if you subtract this approximately estimated dust contribution from the observations, something is left and it is small enough so that because of noise or error margins, you can't be sure that it is nonzero.

    But all the published data in the papers are perfectly compatible with the assumption that 1/2 is from dust, 1/2 is from primordial gravitational waves. It's just that the latter thing was "clearly nonzero" in the March 2014 discovery paper - it was "five or more sigma" different from zero. They saw the waves clearly or they thought they did because they had a rather small estimate for how large the dust contribution was.

    No, the dust contribution is larger so what's left in the B-mode signal for the gravitational waves is smaller but it may be as much as 50% or so of the B-mode signal. It's just not enough to be "sure at 3 or 5 sigma" about the nonzeroness of the gravitational wave contributions.

    Also, I want to emphasize that my point about the "characteristic distance scale in the maps" is exactly the same point as the point about the "local maximum in the spectrum as a function of L" - these two statements are the same because of basic properties of the Fourier transform (or decomposition to spherical harmonics).

    When there is a preferred "periodicity" or length scale in the x-space (think about quantum mechanics), then there is a corresponding peak in the p-space - some frequency or wave number k (the inverse wavelength) dominates.

    Hey, Lubos, here's your chance to win $1,000! I picked up the latest Scientific American today and found an ad in it offering $1,000 to the first person to disprove "AP Theory." The ad doesn't explain, but it names a website, AP Theory, which does explain. There you'll find a theory based on ridiculous misconceptions. Actually, I don't think it's possible to win the $1,000, since the guy could always find that you had violated the contest rules.

  4. According to John Kovac, the leader of the research team, BICEP3 is already operating with much more sensitive instruments and will find the imprints if they are there.


  5. Not on topic:

    Sabine H. now has a music video on YouTube called "I'm a little funny." She may not be a better theoretician but she's got you beat in singing. Maybe.

  6. So, if you don't come up with 5 sigma right away, it is not considered a discovery anymore? IF another team reaches 5 sigma, they get the Nobel but BICEP team gets nothing, except the bad rep because they 'rushed' and 'leaked' under 5 sigma?

  7. Dear Lumo,

    I am working on polarization, but different way.
    Do you have any data that I could use in my research?

    My progress:

    Aim of my work is use algorithms I get for fusion reactor based on polarisation, during this phenomenon photons break down to electron-positron pairs they --> split and produce energy.

    Preliminary CALCULATIONS ↓ based on Photoelectric effect

    Mapping of Photons v.01 comparison to NASA photo of cluster from which X-Ray possible produced by sterile neutrinos were observed

    Electron magnetic field (looks like that of bar magnet 8) two different methods and same result

    Based on new kind fractal that I developed RTDG n-dimensional enlargement, 3d CAD rendering script (simulation for computation physics of light near perfect realism)
    (not presentation for scientific purpose yet !!)

    DETAIL ZOOM HERE use green frame and click !!!
    (this graph is beautiful art so I sell it as art)

  8. Well, hundreds of thousands of YouTube users surely disagree with you heavily.

    She is completely out of tune. She is impossibly bad. I am actually a trained musician.

  9. Hi Kashyap, there are various models - sometimes using data from other frequencies etc. - that allow you to estimate how strong the B-modes from the dust are.

    Let's call this D.

    Then you measure the total amount of B-modes by BICEP2, let's call this total intensity B.

    Then BICEP2 was using models of dust that implied that

    D / B = 0.2 +/- 0.1

    or so which implies that most of the "B" obviously can't be "D" (dust).

    In the datasets which care about what Planck actively says about the dust, you have something like

    D / B = 0.5 +/- 0.3

    So the 50% is just the mean value. The observed signal looks slightly - like twice - stronger than the expectation from dust. But the factor of "2" is really "2 plus minus almost one" or so, so it can't be excluded with any sufficient certainty that B is actually equal to B and nothing is left for the gravitational waves.

    But these two pictures only differ quantitatively, by shifting the mean value and the error margin a little bit. It's still true that the observed signal doesn't "beautifully" agree with the expectations from the dust. It does look higher although this difference isn't enough for a reliable claim according to this joint analysis, for example.

  10. On it also says:

    "In the joint analysis, the researchers overlaid data recorded by the BICEP2 telescope at a frequency of 150 gigahertz with data recorded from the same patch of sky by Planck at 353 Ghz, a frequency at which virtually all the polarized light comes from dust. (Planck also records polarization signals at lower frequencies). The two data sets proved a match — the region in which BICEP2 found its strongest signal is the same place that the Planck dust signal is strongest, indicating that the BICEP2 signal is due almost entirely to dust."

    And then we have Marc Kamionkowski of the Johns Hopkins University in Baltimore, Maryland, who is not a member of either team:

    "“I no longer believe that BICEP2 detected the signal of’ gravitational waves,”"

    That's not to say there isn't a signal there, the consensus seems to be that if there is one, the BICEP2 team didn't discover it.

  11. kashyap vasavadaJan 31, 2015, 5:27:00 PM

    Dear Lubos: It is not clear to me how they arrived at 50% figure. What is the error in the dust contribution calculation? Also, is the theory on dust contribution rigorous?

  12. You are funny Lubos.

  13. LOL, I realize now that in reality this so called Galileo’s trial was just a collaboration with Holy Inquisition; like saying Galileo et al.

  14. Thanks for the answer Lubos.You just confirmed what i was thinking about..I sure that both teams in the project worked very hard and nobody did nothing wrong with their respective measurements. it's just that someone might think also that nobody wants to be put aside from a big discovery at any cost..maybe in the future..maybe not..The door remains open..Anyway i think that as always the problem are the exaggerated titles and descriptions in hundreds of webzine's articles and blog posts.

  15. well, you know, I used to work at a recording studio, so i am terribly spoiled.
    The both of you are missing the mass quantities of beer that made AC/DC go !

  16. Lubos
    Come clean and admit that you pretend to have answers even when you don't. This most certainly happened here. Best wishes but don't spend your life pretending. Thanks

  17. Looking at the figures on tells us how noisy the Planck data is compared to BK:

    It seems we will have to wait for more Keck data:

  18. Amazingly clear graph and a huge difference, thanks.

  19. IMHO, the BICEP2
    -E and B-Modes could be the result of so called Birkeland –Alfven electric current circuits produced between hypothetical paring
    black holes as the origin of primordial Herbig Haro objects, which could be he origin of the unexpected large B-Modes found in the BICEP2 measurements.