Into Darkness (II) : Attack Of The Flying Snakes

Cosmological Models Are Lousy And I Don't Like Them

Last time we looked at the observational evidence for and against dark galaxies. But that post was boring, why did you bother wasting your time reading it ? This one is much better. Here we'll look at the latest theoretical evidence for galaxies which don't do the whole star-spangled thing because it's "too mainstream".

The standard model of cosmology is that most of the mass in the Universe is dark matter. Simulations predict that this can reproduce the very large structures we see - filaments of galaxies and voids where there's not much of anything - extremely well. But on smaller scales, they're about as successful as any character ever played by Sean Bean.

With the notable exception of Sharpe, obviously.

The problem is that the simulations predict about ten times as many galaxies as are actually observed. Now, because dark matter is so dominant - about ten times as much as the visible matter (gas and stars, normally known as baryonic matter) - the simulations only use dark matter. Clever mathematical recipes are used to work out how much visible matter each dark matter "halo" should contain. Since there's so much more dark matter, the baryonic matter shouldn't be able to affect it very much.

On the other hand, the physics of gas and stars is very much more complicated. Stars emit radiation and winds which pushes gas away and heats it up. Occasionally they explode and spew heavy elements back into the interstellar medium, which can help the gas cool and form more stars. Then there are active galactic nuclei caused by supermassive black holes, which can inject even more energy into the gas. Not to mention magnetic fields, different phases of the gas, etc.

So even though the the baryons probably can't affect the dark matter all that much, the amount of baryons each halo contains could be extremely complicated : i.e. we might have got that wrong. Maybe the reason we don't detect all those galaxies is because they've never gained enough gas to start forming (m)any stars.


Dark Galaxies... Or Just Cosmic Fluff ?

As we saw previously, there are some very intriguing candidate objects for these (nearly) starless galaxies. The major objection is that simulations have shown they could just be "tidal debris" - gas that's been ripped out of galaxies during interactions. In that case these gas clouds wouldn't solve one of the most important problems in contemporary cosmology at all : they'd just be uninteresting bits of hydrogen fluff, floating through the Universe feeling a bit foolish and causing no end of red-faced embarrassment for anyone claiming to have found a dark galaxy.

Which of these explanations if correct ? Given that solving a very important problem is generally defined to be a very important thing, remarkably there are only two studies showing how such hydrogen fluff could form. And neither of them do a very good job, either.

First, here's a reminder of the parameters of the clouds that might be dark galaxies :

• Gas mass of about 30 million times that of the Sun.
• No more than about 55,000 light years diameter (though they could be smaller).
• Line width (how fast they appear to be rotating, even if it's not actual rotation) of 100 - 170 km/s.
• At least 300,000 light years from the nearest other detectable gas.

Both studies I'll describe were prompted by one particular object, VIRGOHI21, which is broadly similar to the AGES clouds. The main difference is that it's part of a much larger hydrogen stream, which these clouds aren't - but see the previous post and also this for details.


Tidal Debris : Attempt No. 1

The first attempt was a simulation by Bekki et al. 2005. This publication was only a letter, which means it's just five pages long so doesn't have much in the way of details.  But what they did was to model two galaxies and bang them together. One galaxy was modelled properly (like this), the other used (essentially) a single particle just to approximate its gravity. Which is reasonable as computation time is a limited resource, but of course it would be better to model both galaxies properly.

The results of Bekki's simulation of two galaxies having an uncomfortably close encounter. Blue shows stars and pink shows gas.

At this point the whole scenario proposed breaks down miserably. They state they tried a "large number" of models, but don't say how many. They show a representative sample of those models which do produce isolated debris, but don't state what fraction of their models produce this debris. So we're told nothing as to how likely is is that tidal encounters can produce these sort of features. They don't quantify what happens to their modelled galaxy, but the figures make it appear to be pretty devastated. And bizarrely, they define "optically dark" to mean "anything that's fainter than the Magellanic clouds"*. The Magellanic Clouds are bright enough to be visible to the naked eye. Ummm.... ?

* They don't phrase it like this, nonetheless, it's what their statement means.

Perhaps worst of all, the debris they produce is huge - at least five times larger and more massive than our observed clouds. In fairness, at the time of that study the observations of VIRGOHI21 only had low spatial resolution, so it could have been much larger. Subsequent observations showed that it's very much smaller. So the Bekki scenario is now decisively ruled out. There are other problems with the paper but they're not worth mentioning.

Tidal Debris : Attempt No. 2

The second main effort was by Duc & Bournaud 2008. This is a much more detailed paper which attempts to reproduce VIRGOHI21 with great precision. By this time the high resolution observations of the object had been released, so they knew the size of the object. Their procedure was similar to that of Bekki - hurling two galaxies past each other, one with gas and one without.

Contrary to popular opinion, the results of Duc and Bekki are actually extremely similar. Which is not all that surprising considering they did essentially the same thing. While they do form a gas blob in the tail that has the right mass and is the right size, its velocity gradient is about three times smaller than VIRGOHI21 or the other clouds. There are some other questionable points too : their progenitor galaxy is incredibly gas rich with a very extended gas disc, meaning it has lots of gas that can be easily stripped, and it's significantly less massive than the real galaxy. It's at the very limit of what's permitted by the observations.

Or in other words they gave it the best chance possible, and still it failed.

Now, since VIRGOHI21 was, at the time, thought to be a really exceptional object, those latter points might be acceptable. A weird galaxy that produces a weird object. OK, fair enough. But now we know there are more such clouds, ones which don't have streams at all. Worse though is the failure of the Duc model to reproduce the velocity gradient - they very sneakily adjust the scales on their figures 2 and 6 (observation and simulation respectively), making it look as though they made something much more similar to the real VIRGOHI21 than they actually did.

The observation (left) and simulation (right) data as they appear in the Duc paper, on separate pages. Let's face it, the sharpness of the "kink" in the stream at VIROGHI21 isn't well-reproduced by this simulation even in this figure.

After correcting the figures to have the same scale, it's clear that the simulated VIRGOHI21 is even worse. There's absolutely nothing in the simulation that anyone in their right mind would mistake for a dark galaxy, which was the whole aim of the paper. Tsk tsk !

On the positive side, the Duc model does also produce a lot of gas north of VIRGOHI21. Very sensitive ALFALFA observations had detected this, which a third, far less well-known model of the system had failed to predict. This approach, by my PhD supervisor, had VIRGOHI21 as dark galaxy that came along and pulled out a long gas tail. This one was never published in a refereed journal, which is why I'm not going to go into any details.

So both major studies have failed to reproduce one of the most well-known dark galaxy candidates. The Duc model got pretty close, but that velocity width is a real show-stopper. You may not think that's such a big deal, but, as I shall show, you'd be wrong. It's a massive problem.

Which is not to say the Duc paper isn't important, because it is. It showed that the large-scale properties of the system could be reproduced in quite a simple way that didn't require a dark galaxy. It's completely understandable that people would assume the sharp kink is a mere detail. Understandable, but wrong.


I'd Like To Hear From Fictional Mathematician Ian "I Nearly Got Eaten By A Tyrannosaurus" Malcom At This Point

Not quite, Ian. Actually the weakness of the previous studies was that the authors never stopped to think if they would. That is, they came up with possible formation scenarios, but they never investigated how likely they were to really occur. Which was reasonable at the time (with just one weird object, it's perfectly fine to invoke a weird explanation), but with more such objects now known that needs to be addressed.

So what we did in the latest paper was to model the entire cluster, using an existing simulation. Not just two galaxies any more, but 400. Of course we couldn't model the gas in each galaxy because that would be far too computationally expensive, but now we could model the gravity far more accurately than the previous studies. Rather than dropping a gas-rich galaxy through the cluster, we dropped a gas stream.

"WHAT ?!?!" I hear you cry. "THAT'S F*£!@ING RIDICULOUS, YOU DOLT !"

I know, I know. If we wanted to do a proper comparison of the previous studies, we should have used a galaxy. Our scenario was even more ambitious though, because there's another mystery in the Virgo cluster : there aren't many long hydrogen streams there. Only four, in fact, plus a few much shorter ones. Simulations predict that such features ought to be very common - but they're actually extremely rare.

All known hydrogen streams in the Virgo cluster. Only the big ones are easily visible and labelled here - there are in fact four large streams in the cluster (HI1225+01 is probably a bit further away) plus another half-dozen or so very short streams.

So our idea was simple : maybe the long gas streams get broken up by interactions with all the galaxies. Maybe tearing up the stream produces fragments like the clouds that we see. Then we'd kill two birds with one stone and everyone would shout, "hurrah !".

Except the birds, obviously.

OK, big problem. The simulations which predict the existence of those streams are about something called ram pressure stripping. Now I have to disappoint you here, because that's nothing to do with forced bestiality at all.

Sorry. Anyway, galaxy clusters also contain hot, diffuse gas of their own. As galaxies move through it they should create a "ram pressure" which is strong enough to strip their own gas into long streams. Our simulations don't include this hot gas, which is not good. We really would like to include it, but it's much harder from a technical standpoint. Better to start simple and build in the more complicated physics gradually. And neither Bekki nor Duc included the hot gas, so there.

Simple simulation showing gas being ram-pressure stripped

from a galaxy.

But while in some ways our simulations weren't any better than the previous ones, in one very important respect we made a huge improvement. We didn't just drop one stream into the cluster, oh no. We dropped them in batches of 27, with each one at the corner or midpoint of a cube. Like this :

The grey spheres represent the other galaxies in the cluster. If you really want to you can watch a movie of this here, but it's quite dull, so don't.

We varied the starting mass, temperature, and distance of each batch of streams from the cluster centre. All told we ran something like 200 simulations or more. Then (unlike Duc) we converted the data into what we would detect with Arecibo so we could directly compare the simulations and observations.

To be clear, we didn't drop all 27 in at once. We dropped them in one at a time. The point is not to find out if there's some particular path through the cluster which produces fake dark galaxies, but to see how likely such objects are to occur by chance. And although we used a realistic model for the cluster gravity, we're still missing a lot of important physics - not only the intracluster gas but also heating and cooling, star formation... and of course the galaxy from which the gas stream originated. I suspect that most of these won't change the end result that much, though you'll have to read the paper for more details.

The most satisfying result from the whole shebang was that a lot of the streams look quite remarkably like snakes. The second most satisfying result was that the referee of the paper didn't object to the title.

ATTACK OF THE FLYING SNAKES !
... although I suppose that should really be falling snakes.

Life Finds A Way ?

Despite the many, many limitations, the results were remarkably decisive. The "tidal debris" idea fails miserably!

Yes, really ! We produced features just like those seen in Duc and Bekki easily - long features with shallow velocity gradients occur pretty nearly all the time. But clouds like the ones we observe in the real cluster ?

They just don't happen. Strictly speaking they happen 0.2% of the time, which means that "tidal debris" is a patently ridiculous mechanism to form all the clouds we see in the real cluster*. Don't believe me ? Watch the movies for yourself. Like Bekki, we made synthetic observations so we could accurately measure what we'd actually detect.

* Of course we're not claiming that tidal debris is never a good explanation - it's fine for larger clouds or ones with smaller velocity widths - just that it doesn't work for features like the ones we found.

White shows the particle data, red shows what we'd actually observe with Arecibo, and green shows the very isolated clouds similar to the weird ones we were trying to explain. Too difficult to see ? Have a look here.

There's more. We also tested the alternative idea that the clouds could be dark galaxies by dropping in galaxies containing gas but no stars. And again, you can watch the results for yourself.

See the full set of discs here.

Quite unlike the streams, nothing of any consequence happens to the discs. Even after being harassed by 400 galaxies for 5 billion years, they generally lose no more than half of their gas. To an AGES-class survey, they'd appear to do... well, nothing much. The gas density stays low too, so they wouldn't even form stars.

In fact it's even more of a dramatic win for dark galaxies. Last time we saw how one of the weirdest properties of the real clouds is not that they appear to be rotating rather fast, but that they're rotating faster than we'd expect based on their mass. It turns out that the dark galaxy models can explain this pretty well, whereas the tidal debris scenario just can't. In fact, if you consider clouds which not only have the correct line width but also the correct mass, that 0.2% drops to 0.0% ! Even the bits of tidal debris which do occasionally come close to matching the deviation only do so for about 50 million years, and then they typically disperse and become undetectable, whereas the dark galaxies remain with the correct high line width and detectable for the whole 5,000 million years of the simulation.

Though there are a couple of points to bear in mind. Firstly, the simulated debris at least moves in the right direction - it also has a higher line width than expected, just nowhere near high enough. Second the simulated dark galaxies don't agree perfectly with the real clouds - however, that's almost certainly because setting up a stable disc is hard. If we spent longer tweaking the initial conditions, we could get the simulated dark galaxies to be in much better agreement with the real clouds.

That, then, is that rarest of things : a sweeping and decisive victory for dark galaxies and a total defeat of the idea of tidal debris. But let's not get too cocky. Remember, our simulations don't include the intracluster gas, the gas in the streams is constrained to have the same temperature, none of the harassing galaxies have gas, and for our streams we didn't include the progenitor galaxies. These are all problems that do need to be addressed. However, the previous studies didn't include the intracluster gas, they kept its temperature constant, and their harassing galaxies were also gas free.

Since our results are often similar to the previous studies, that indicates that the lack of a parent galaxy probably isn't significant. For the first time we can quantify how likely the tidal debris theory is for clouds with high velocity widths : it isn't. That factor of three turns out to be a huge problem for the Duc and Bekki models - not the minor detail everyone seemed to think it was.


Mystery Solved ?

You might wonder why, if we'd got to the stage of dropping dark galaxies into the cluster, we didn't drop normal ones as well. That's surely the next logical step. The paper was 26 pages long at this point though - dangerously long, to the extent that people might not read it. And setting up a stable disc is not so easy, it isn't a matter of just tweaking the numbers of the discs we already used. So it's better to leave that to a second paper. It's tough to see it dramatically affecting the results though.

It's difficult to say how the other parameters would affect the results. The intracluster medium should, in my opinion, make the cloud's velocity widths if anything smaller. Any expansion velocity would be met with pressure acting to prevent it from expanding, slowing it down (though fluid effects are complicated and it's not always a good idea to guess what they'll do). Cooling of the gas has been shown to make a relatively small difference compared to harassment, so that probably won't change anything much either.

Smug as I am with this unexpectedly exciting result, the message I want to end on is rather different. Yes, we got a neat result. But we also know there are problems, and we should improve things accordingly. We're going to solve this one, dammit. But we're going to do it properly - not with hand-waving explanations that don't stand up to scrutiny. So far, the poo-poohed idea of dark galaxies is doing far better than the much more popular idea of "tidal debris". It's too important an issue to dismiss this with models that don't actually work and haven't been repeated. And we should always, always, always bear in mind that a model which works is not the same as a model which is correct.

Because humans are made of squishy neurons rather than silicon chips, belief occurs for both rational and irrational reasons. Scientists have irrational beliefs just as much as anyone else does. In this case, they've become so enamoured of the idea that the clouds could be tidal debris - which I think we've shown isn't even tenable anyway - that they've assumed that that's what they probably actually are. Yet the two situations are very different. Yes, I can run around naked hurling my own bodily fluids at people, but this doesn't mean I'm actually going to do it.

By the same token though, these results aren't enough to make me "believe" in dark galaxies. In fact if some masochist decided to whip me until I was forced to confess my preferred explanation (I dunno what kind of messed-up universe it would take for that to happen, but I'd like to avoid it), I'd probably still say "tidal debris". I couldn't give you a rational reason, but something about the idea of dark galaxies just doesn't smell right.

All we've really shown is that dark galaxies are more likely than was previously thought - nothing more, nothing less. It's a horrid cliché to end on, but often more research genuinely is needed. Given the dismissal of the alternative ideas thus far, if I can convince people that cheering, "tidal debriiiiiiis !" isn't a good enough explanation, I'll be happy.