GW231123: an unearthly signal

A report on some actual science

Today the LIGO-Virgo-KAGRA (LVK) Collaboration posted a paper on the preprint arXiv, announcing a new gravitational-wave detection. I led the team that wrote the paper, so let me tell you about it¹.

The paper's title says it all: “GW231123: a Binary Black Hole Merger with Total Mass 190-265 solar masses”. The gravitational-wave signal is called GW231123 because it was observed on November 23rd, 2023. The signal was from the merger of two black holes. So far, so good: the LVK has previously published the detection of almost 100 signals from pairs of merging black holes. The two black holes orbit each other, and the orbit loses energy due to the very gravitational waves that we eventually observe billions of years later, and as a result the black holes spiral towards each other and collide and merge. Nothing can escape a black hole, not even another black hole, so what's left is: a bigger black hole.

Now we get to the two things that make this observation especially interesting.

The masses of these two black holes add up to somewhere between 190 and 265 times the mass of our sun. The individual black holes that go into the collision are roughly 100 and 140 times the mass of our sun, respectively.

Here's the interesting thing: black holes aren't supposed to be able to form with those masses! By “aren't supposed to”, I mean, we have theories of how black holes form when stars die, and those theories are fine for black holes that are five times the mass of our sun, or ten times, or even fifty times, but once you get to about 60 times the mass of the sun, some funky nuclear/quantum/whatever processes come into play, and the star blasts aways lots of its mass, and you can't form a really massive black hole. That carries on until you get to really massive stars, and then nothing can stop the collapse, and so you can still have black holes with masses above about 130 times the mass of the sun. So: between about 60 and 130 solar masses, there's supposed to be a gap.

And guess what? The two black holes that merged to produce GW231123 are slap bang in the middle of that gap. (With some caveats — we'll get into that in a minute.)

So how did they form? There are many exciting possibilities. One is that our theories are wrong. We don't like that idea so much. Not because we're emotionally attached to our theories (although of course we are), but because scientists are very sensible and don't just go around throwing their theories onto the fire at the first sign of trouble. There are plenty of other options.

The most popular explanation is that the black holes formed from previous black hole mergers. Two perfectly normal-sized black holes, perhaps each a reasonable 30 solar masses, met across the empty wastes of space, became gravitationally bound, which is our romantic term for pairing off to orbit together, and eventually merged to form a 60-solar-mass black hole. Ok, that's not quite 100 solar masses. Fine: maybe that black hole met another 40-solar-mass black hole. You get the idea.

It could also be that the two black holes formed by standard stellar collapse, and happen to lie on either side of the mass gap. That's possible within the uncertainties of our measurements.

And that brings us to the other especially interesting thing, which is the uncertainties. I'm serious. The uncertainties are fascinating. Read this and then I dare you to tell me otherwise.

When two black holes collide their properties are encoded in the gravitational-wave signal they produce. If we can decode the signal, then we can find out how massive the black holes are and how fast they spin. We decode the signal by comparing it with theoretical models; our models tell us what the signal would look like for all possible combinations of black-hole masses and spins, and we identify which ones agree best with the detector data.

Producing these models is a thrilling business, which happens to be my main research activity. I could rave about it for days on end, and I often do. For now, though, the main thing you need to know is that, although we've produced lots of great models that have been good enough for all those previous hundred observations, we know that for some extreme (and therefore extremely exciting) binaries, our models may not be good enough. Plus, as detectors become more sensitive, we will be able to extract ever finer information from observations, but to do that the accuracy of our models will have to keep up. That's right, we are in a race against the Universe. That's how thrilling it is.

Now it looks like the Universe caught up with us. When we decoded GW231123 with our most accurate available models, they gave different answers. How different? They all say that the black holes that went into GW231123 are really massive, and they all say that the black holes are rapidly spinning. After that it gets messy. One model sees black holes with 130 and 110 solar masses, another 145 and 55. Yes, that's right: one model puts them in that mass gap, and another puts them either side.

We have never seen anything like that before. Sometimes the models give slightly different values, but if you take into account all the uncertainties that enter into the analysis, they're always pretty consistent. Not this time! This time they are definitively agreeing to disagree.

We don't like that. We want clear answers! Dammit!

Maybe our models are confused because it's not two black holes colliding after all? Maybe it's something else? Or if they are black holes, maybe they're colliding in a different way? We expect them to spiral inwards as they orbit, but maybe instead they just went crashing straight into each other?

Ok, sure, the chances of two black holes smacking directly into each other across the vastness of space is pretty unlikely, but we're getting desperate here!

Not that that would help. Our models are all seeing the same signal, so even if they're getting it wrong, they should still agree on their wrong answers.

Fine, it must be the models. Some of them aren't accurate enough. No problem! We'll work out which ones don't make the grade, and we'll throw them out, and then we'll be left with the right answer.

No luck there, either. We compared our models against computer calculations of signals for a series of different high-spin binaries, and although one model was generally more accurate than the others, it wasn't always accurate enough to completely rule out the possibility of wrong answers and, besides, for some cases other models were more accurate.

Undeterred, we tried using some of those computer-calculated gravitational waves as fake signals, and looked to see how well our models decoded their properties. We hoped that it would become clear which model or models were fine, and which were not.

“We hoped it would become clear” is the epitaph of many a failed scientific project.

Most of the time, the results looked nothing like what we saw for GW231123. All of the models were wonderful. They agreed perfectly, and they got perfectly correct answers.

What the hell was going on here!?

We kept banging our heads against the wall trying more fake signals.

And then, amazingly, we found a few where the models got the answers wrong.

Remember when Rutherford fired alpha particles at gold foil, and he did it again and again, and every single time they went right through just like everyone thought they would, but then every once in a while one miraculously bounced back, and he discovered the atomic nucleus?

This wasn't like that at all. All we discovered was that we hadn't been going completely mad.

It turned out that sometimes the models could get the answers wrong — but all of the models. We'd always said that for highly spinning binaries our models needed to be more accurate, and now we had concrete examples. That wasn't such great news, though, because instead of identifying one model that we could trust to give us the right answers, we had found that we couldn't completely trust any of the models.

Hence the final results, where we just had to be good honest scientists and admit that we couldn't get as accurate an answer as we'd like, and the best we could do was combine together the results of all of our models, and that's why the final results have large uncertainties. That's why we can't say for sure whether the masses are in the mass gap or not. We'd like to quietly ignore some of our models and provide more accurate answers, but this is science, and you can't do that.

The story isn't over. People will keep looking at this data. They'll try to discriminate better between the models we have, and they'll work to make new models, and they'll also test for all those other not-orbiting-black-holes possibilities. That's for the future.

For now, though, I think it might be time for another holiday.

For your further edification, here’s a link again to the paper, and also a cool infographic, produced by Simona Miller at Caltech:

Postscript 1.

Regular readers will be aware of many veiled references over the last few months to a Very Important Paper that I've been working on. This, of course, is that paper. You may wonder if it is also the paper that's inspired a series of articles in which I'm cynical and sarcastic and often quite frankly rude about the many frustrations involved in writing a scientific paper? Not quite — writing a paper for a huge collaboration is in an entirely different universe to any other paper I've ever worked on. But it has had its frustrations, and at certain moments stirred up my most deeply buried memories of past frustrations, and writing about them has been wonderfully therapeutic. There are two more pieces in that series still to come. Here’s what we have so far:

1. How to get your name on a paper.

2. An artist in peril.

3. The 2025 vacation life hack.

4. A collision of co-authors.
   

Postscript 2.

GW231123 was detected on November 23, 2023. As a veteran Doctor Who fan, I was pleased to note that that date was exactly 60 years after the transmission of the first ever Doctor Who episode, An Unearthly Child. (Hence the rather clumsy tribute in the title.) The episode screened a few minutes late, because the evening news ran overtime due to coverage of the assassination of JFK the day before. You've got to wonder what kind of shocked and bemused state the audience must have been in when that otherworldly electronic thrumming theme tune began...

1. Officially, and for the avoidance of all confusion: hundreds of people were involved in producing results and checking results. The main paper writing was done by a “paper writing team” that consisted of 14 people, each covering a particular aspect of the analysis and interpretation, or a particular section of the paper. There was also an “editorial chair” and a “paper manager”. You can think of the editor and manager as being like the Director and Producer of a film. And if you have never understood the distinction between the Director and Producer, it's quite simple: the Director is like an editor, and the Producer is like a manager. Anyway: I was the “editorial chair”.