Wednesday, September 24, 2014

Breaking Down a Big Bang Breakthrough

I like clean, simple arguments. I like inescapable logical conclusions. And when those things appear in a paper telling us something new about the very early universe – even better.

Today, David Parkinson, a research fellow at the University of Queensland, came to Melbourne to give a talk on the paper he just posted to the arxiv today with his colleagues Marina Cortes and Andrew Liddle. It was incredibly (and not coincidentally) well timed. Just two days ago, I spend most of the day grappling with the implications of the latest cosmic microwave background results from the Planck satellite, and now David was going to come and give a new perspective on it.

Marina Cortes
Marina Cortes
Andrew Liddle
Andrew Liddle
David Parkinson
David Parkinson

But let me back up. In March, researchers with a telescope at the South Pole – the BICEP2 collaboration – announced an astonishing new result. They were studying the cosmic microwave background (CMB, the afterglow of the Big Bang), and they presented what they said was the first evidence of primordial gravitational waves – ripples in the fabric of spacetime, from the first tiny fraction of a second after the Big Bang. The announcement made a HUGE splash. There were articles in all the major newspapers and magazines hailing the discovery as the start of a new era in cosmology, and in some cases even as proof of cosmic inflation. This is a scenario in which the early universe, in the first billionth of a billionth of a billionth of a billionth of a second after the moment of creation, expanded extremely rapidly and grew several orders of magnitude in size. The theory of inflation has been around since the 1980s, first proposed by Alan Guth, and developed by Andre Linde, Paul Steinhardt, and others. One of the key predictions of inflation is that it would produce gravitational waves, and in principle these could be seen as little swirls in the pattern of polarization of the CMB. The CMB is one of the strongest pieces of evidence that the Big Bang happened at all, and by studying its light we can learn a huge amount about what the early universe looked like. Some of that light is polarized, meaning it is preferentially oriented one way or another when it reaches the detector. Patterns in the polarization can show us traces of those early spacetime ripples. Although many experiments had been looking for these patterns, BICEP2’s signal was unexpectedly strong, and it was in some tension with previous tensor measurements by the Planck satellite, among others. A viral video went around showing a flabbergasted Andre Linde receiving the news, and even he, a vocal supporter of inflation theory, looked shocked.

Characteristic swirls in cosmic microwave background polarization, found by BICEP2. Image credit: BICEP2 Collaboration.

It wasn’t long after BICEP2’s announcement, though, that problems appeared. Rumors went around saying that the BICEP2 team had made a mistake in their calculations. The problem was interstellar stardust. It turns out that dust can create polarization too, and although the BICEP2 team considered a few different possibilities for the level of contribution of dust to their signal, several cosmologists argued that the estimates were way too low. Two papers came out showing that the BICEP2 signal – the one that was supposed to be a beautiful picture of gravitational waves – could have been entirely due to dust in our Galaxy mimicking the primordial signal.

More articles appeared, now announcing that the “Big Bang result” has “turned to dust” (among other clever puns). Paul Steinhardt, who has spent the last several years developing alternatives to inflation theory, wrote an article proclaiming that inflation was never a good idea to begin with, and the dust problems go to show that the hype was all for nothing. Most of the cosmology community, however, took the attitude that we should probably just wait and see. There were several other experiments taking data to confirm or rule out BICEP2’s discovery, and the Planck satellite – the current flagship in the CMB detection game – would be producing maps of interstellar dust really soon. That should clear everything up.

Two days ago, Planck released their dust polarization results. They specifically addressed the BICEP2 study, and while they were very measured in their statements (pointing to an upcoming joint analysis), the upshot of the work was that the dust polarization signal was so high that it could easily account for everything BICEP2 saw. Maybe the gravitational waves are there, but if Planck is right about the amount of dust in the way, there’s really no way to say that BICEP2 actually discovered them. In physics, a discovery means you’ve shown something to be the case beyond any reasonable (statistical) doubt. Usually that comes in the form of a statement of how incredibly unlikely it is that chance or some spurious signal could have given you the same result. A signal that could just as easily be all dust is definitely not a discovery.
Comparison of original BICEP2 result (left) and Planck dust polarization result (right). The circled region in each shows where the primordial gravitational wave signal is expected to show up for the model supported by BICEP2's result. The colored lines in the BICEP2 figure are their dust models, all well below the signal. The blue boxes in the Planck figure are their estimates for the dust amplitude, and the solid line is where the gravitational wave signal should occur. You can see that the dust amplitude is comparable to the expected gravitational wave signal amplitude, suggesting the two could not be distinguished. Image credits: BICEP2 Collaboration, Planck Collaboration.

This all brings us to David’s paper. The details are technical, but David and his colleagues basically go back to the drawing board to determine how we can analyze data to get the best, most unbiased estimate of the gravitational wave signal. They re-analyze the BICEP2 polarization signal, under a couple of different assumptions, using Planck's previous limits on the gravitational wave contribution as a starting point. First, they assume there was no dust contamination at all. Then they look at an “optimistic” dust model, where dust contamination is there but not bad enough to drown out the signal, and a “pessimistic” dust model, where dust can account for everything. They look at not just the level of primordial gravitational waves – also known as tensor modes – but also the “tilt” of the tensor mode spectrum, an important parameter in inflationary models.

What they find is striking. In the “optimistic” and dust-free models, they find tensor modes, just as BICEP2 did, but they also find a tilt that is utterly incompatible with standard models of inflation. Basically, if BICEP2 and Planck’s previous measurements are correct, and the dust is at a manageable level, BICEP2 not only doesn’t prove inflation – it just about rules it out! The only other option is to use the “pessimistic” dust model, in which case BICEP2 discovered nothing. As it happens, Planck’s new measurements fit the pessimistic dust model best.

Results from Cortes, Liddle & Parkinson paper. The data points are the BICEP2 results, and the circled region can be compared with the regions in the figure above. The black lines are the expected level of the signal for dust plus tensor modes plus the contribution from gravitational lensing. The green line is the tensor contribution -- this can be directly compared to the dashed red line in the BICEP2 figure above. You can see the tilt in the spectrum in the way the green tensor line extends up and to the right in the figure. Image credit: Cortes, Liddle & Parkinson 2014.

In any case, the implication is clear, and somewhat unsettling. It presents us with three items – Planck’s previous tensor limits, BICEP2’s gravitational wave signal, and the inflationary model – and it says we can pick two. At least one has to be incorrect.

That’s a bold statement, and a big deal if it holds up. I love the irony in the suggestion that keeping the “inflation-proving” result requires disproving inflation. But it also illustrates the danger of jumping the gun in these kinds of complicated data analyses. It’s widely believed that BICEP2 made too strong a statement in their original paper and press-release, both in their optimism about dust foregrounds and in their statement of confidence in the signal. Now it appears that their analysis may also have introduced a bias that hid the implications for the tensor tilt.

To know with any degree of certainty what the BICEP2 result really means, we’ll have to wait for a joint analysis being carried out by the BICEP2 and Planck teams in collaboration, and we’ll have to see what the other experiments find. But it’s certainly an exciting time, and, as always, it’s fascinating to see the scientific process in action.

Footnote: My PhD thesis was partially based on a study with a similar sort of gist – that you can have two of three theories, but not all of them together. In my case, the theories were axion dark matter, string theory, and inflation. If you’re really curious, you can find the paper here.


  1. The results of that paper depend on assumptions made about the scalar power spectrum. If it is assumed to be a power-law, the conclusions are reached. If it is allowed to not be, then they aren't.

    It's still true that most inflation models do predict a power-law for the scalar power spectrum, but certainly not all, or "almost all". There are many models that would be able to produce a drop in scalar power at the largest scales, which would allow tensors to exist without them requiring a blue tilt.

    I'm only writing this because you said in your first sentence "inescapable logical conclusions".

  2. Thanks for your comment, Shaun! It's true that since there are a virtual infinity of inflationary theories, it's hard to say anything "inescapable" about them. When I referred to "standard" inflationary models, I had in mind slow-roll, single-field models, but I wasn't thinking about the possibility of a wonky scalar spectrum. I'll have to check out some models like that and see how they could change the conclusions, but I think killing the power-law scalar spectral index would also be a pretty striking result.

    1. "killing the power-law scalar spectral index would also be a pretty striking result."

      Yeah, definitely. This is something that people have been writing papers about since the week after BICEP2 came out. Essentially all of these results come from the fact that PGWs would enhance temperature fluctuations on the largest scales and Planck hasn't measured that (hence why they have an r~<0.1 bound).

      That tension needs to be alleviated by something. One could postulate changes in the scalar power spectrum, or the tensor spectrum. I think the prior expectation would be that it is far more likely that there is a dip in the scalar spectrum, than that the tensors are blue tilted. There isn't enough data to tell the difference (though if there really are tensors then in time there could be).

    2. Hi Shaun,

      I think the point is that it is not very likely that BICEP2 has both measured tensors AND required us to extend the inflation model in order to break the power-law form of scalar perturbations.

      Most plausibly, BICEP2 has measured dust emission and is telling us nothing about tensors or non-power-law scalars.

      Next most plausibly, BICEP2 is only partly dust and there are tensors lurking in there below the current detection threshold. Unless the tensor spectrum is blue (which we can all agree is rather unlikely), such a signal will not be strong enough with current data to imply anything about non-power-law scalars. Even if it is blue, it won't necessarily do so as the blueness will itself resolve the tension.

      Less plausibly again, BICEP2 turns out to have convincingly detected non-blue tensors and then indeed something other than dust has to alleviate the tension.

      all best,

      Andrew (Liddle)

  3. I believe this is called "cooking your data". Also known as "massaging". Is there an hypothesis prior to the experiment and does the observation validate or disprove the hypothesis?

  4. Bruce: What are you referring to, exactly?

  5. Hi ! If Planck mapped the dust is there a way to substract it and see what remains behind the galactic wall of dust?