Even if BICEP2 is wrong, inflation is still science

Paul Steinhardt played a major role in developing the theory behind cosmological inflation, but he has since turned into one of the theory’s biggest detractors. Sometimes, theorists get so attached to their theories that they become blind proponents of them, so it’s quite commendable for someone to become a critic of a theory that he pioneered. But of course that doesn’t mean that Steinhardt’s specific criticisms are correct.

He’s got a short and fiery manifesto in Nature (not behind a paywall, I think, but if you can’t get to it, let me know). The title and subheading:

Big Bang blunder bursts the multiverse bubble

Premature hype over gravitational waves highlights gaping holes in models for the origins and evolution of the Universe.

For a column like this (as opposed to a research article), the author isn’t necessarily responsible for the title, but in this case the headlines pretty accurately capture the tone of the piece.

The hook for the piece is the controversy surrounding the BICEP2 claim to have detected the signature of gravitational waves from inflation in the cosmic microwave background (CMB) radiation. Since my last post on this, the reasons for doubt have gotten stronger: two preprints have come out giving detailed arguments that the BICEP team have not made a convincing case against the possibility that their signal is due to dust contamination. The BICEP team continues to say everything is fine, but, as far as I know, they have not provided a detailed rebuttal of the arguments in the preprint.

For what it’s worth, I find the doubts raised in these preprints to be significant. I’m not saying the BICEP2 result is definitely not CMB, but there’s significant doubt in my mind. At this point, I would place an even-odds bet that they have not seen CMB, but I wouldn’t make the bet at 5-1 odds.

So I share Steinhardt’s skepticism about the BICEP2 claim, at least to some extent. But he leaps from this to a bunch of ridiculously overblown statements about the validity of inflation as a scientific theory.

The common view is that [inflation] is a highly predictive theory. If that was the case and the detection of gravitational waves was the ‘smoking gun’ proof of inflation, one would think that non-detection means that the theory fails. Such is the nature of normal science. Yet some proponents of inflation who celebrated the BICEP2 announcement already insist that the theory is equally valid whether or not gravitational waves are detected. How is this possible?

The “smoking gun” is a terribly overused metaphor in this context, but here it’s actually helpful to take it quite seriously. A smoking gun is strong evidence that a crime has been committed, but the absence of a smoking gun doesn’t mean there was no crime. That’s exactly the way it is with inflation, and despite what Steinhardt says, this is perfectly consistent with “normal” science. People searched for the Higgs boson for decades before they found it. When a search failed to find it, that didn’t mean that the Higgs didn’t exist or that the standard model (which predicted the existence of the Higgs) wasn’t “normal science.”

Steinhardt knows this perfectly well, and by pretending otherwise he is behaving shamefully.

Steinhardt goes on to say

The answer given by proponents is alarming: the inflationary paradigm is so flexible that it is immune to experimental and observational tests.

Whenever someone attributes an opinion to unnamed people and provides no citation to back up the claim, you should assume you’re being swindled. I know of no “proponent” of inflation who bases his or her support on this rationale.

There is a true statement underlying this claim: inflation is not a unique theory but rather a family of theories. There are many different versions of inflation, which make different predictions. To put it another way, the theory has adjustable parameters. Again, this is a perfectly well-accepted part of “normal science.” If BICEP2 turns out to be right, they will have measured some of the important parameters of the theory.

It’s certainly not true that inflation is immune to tests. To cite just one obvious example, inflation predicts a spatially flat Universe. If we measured the curvature of the Universe and found it to be significantly different from zero, that would be, essentially, a falsification of inflation. As it turns out, inflation passed this test.

I put the word “essentially” in there because what inflation actually predicts is that the probability of getting a curved Universe is extremely low, not that it’s zero. So a measurement of nonzero curvature wouldn’t constitute a mathematical proof that inflation was false. Once again, that’s science. No matter what Popper says, what we get in science is (probabilistic) evidence for or against theories, not black-and-white proof. We use Bayesian reasoning (or as I like to call it, “reasoning”) to draw conclusions from this evidence. A curved Universe would have been extremely strong evidence against inflation.

Part of Steinhardt’s objection to inflation stem from the fact that inflationary models often predict a multiverse. That is, in these theories there are different patches of the Universe with different properties.

Scanning over all possible bubbles in the multi­verse, every­thing that can physically happen does happen an infinite number of times. No experiment can rule out a theory that allows for all possible outcomes. Hence, the paradigm of inflation is unfalsifiable.

This once again ignores the fact that essentially all scientific tests are probabilistic in nature. Because measurements always have some uncertainty, you pretty much never measure anything that allows you to conclude “X is impossible.” Instead, you get measurements that, by means of [Bayesian] reasoning, lead to the conclusion “X is extremely unlikely.” Even if anything can happen in these bubbles, some things happen much more than others, and hence are much more likely. So by observing whether our patch of the Universe fits in with the likely outcomes of inflation or the unlikely ones, we build up evidence for or against the theory. Normal science.

To be fair, I should say that there are technical issues associated with this paradigm. Because inflation often predicts an infinite number of bubbles, there are nontrivial questions about how to calculate probabilities. The buzzword for this is the “measure problem.” To be as charitable as possible to Steinhardt, I suppose I should allow for the possibility that that’s what he’s referring to here, but I don’t think that that’s the most natural reading of the text, and in any case it’s far from clear that the measure problem is as serious as all that.

One final note. As Steinhardt says, future experiments will shed light on the BICEP2 situation, and these experiments will justifiably face heightened scrutiny:

This time, the teams can be assured that the world will be paying close attention. This time, acceptance will require measurements over a range of frequencies to discriminate from foreground effects, as well as tests to rule out other sources of confusion. And this time, the announcements should be made after submission to journals and vetting by expert referees. If there must be a press conference, hopefully the scientific community and the media will demand that it is accompanied by a complete set of documents, including details of the systematic analysis and sufficient data to enable objective verification.

For what it’s worth, I don’t think that people should necessarily wait until results have been refereed before announcing them publicly. In astrophysics, it’s become standard to release preprints publicly before peer review. I know that lots of scientists disagree about this, but on balance I think that that’s a good thing. The doubts that have been raised about BICEP2 could very easily not have been caught by the journal referees. If they’d waited to announce the result publicly until after peer review, we could easily be having the same argument months later, about something that had undergone peer review. Errors are much likely to be caught when the entire community is scrutinizing the results rather than one or two referees.

I should add that Steinhardt is completely right about the “accompanied by a complete set of documents” part.

Update: Peter Coles has a very nice post (as usual) on this. His views and mine are extremely similar.