 |
| Party like it's 1899. |
But I shall say no more of that here. On with the science ! Or, if you prefer, a much shorter summary with hardly any jokes is available here.
Introduction : How To Kill A Galaxy
Cast your mind back, dear reader, to the heady days of 2006. Picture the younger me, fresh-faced [beardless] and fancy-free, strutting gaily through the corridors of Cardiff University's Physics and Astronomy Department, unfettered by the latter-day cares of right-wing populism in a World Gone Mad. Carefree Rhys had two main topics in his PhD : finding gas clouds that looked like galaxies but didn't have any stars, and
Long story short, we found some stuff that fits the bill quite well for the first one, which I've written plenty about already. But we never saw much in the way of streams. Considering the target area was the Virgo Cluster, a region so dense galaxies can legitimately be said to be harassing each other, this was a bit odd. It'd be like going into a house party and finding everyone is blind drunk but no-one has thrown up on the sofa. What gives ?
To understand that, it probably helps to have some background. A typical galaxy looks like this :
Note that the gas is a lot more extended than the stars. Being further from the centre makes it easier to remove since it's less tightly held by the galaxy's gravity. The whole thing is embedded in a much larger, more massive dark matter cloud called a halo, though we can largely ignore this here.
A galaxy cluster consists of hundreds or even thousands of these beasties all buzzing around like... well, I usually like to say a swarm of bees, but that's not really accurate. Time for some movies. Here's a beautiful one from the mighty Illustris "Turn On ALL the Physics !!!" simulation :
Very cool, despite a daft choice of music, but quite difficult to understand what we're looking at. A galaxy cluster is more than just a bunch of separate galaxies hanging out : the cluster itself has its own dark matter, gas, and even stars, which makes it horribly complicated. So if we want to get a feel for the sort of stripped gas features we expect to find here, we need to simplify things. Let's start with the orbits. Nice, simple, happy orbits. We think they'd look a bit like this :
 |
| Trajectories of 250 galaxies from a numerical simulation. |
* The house party is probably a bad analogy. A better one would be a heaving nightclub in which almost everyone is very drunk, kinda horny, but somewhat antisocial and super judgemental.
And thirdly, the gas in each galaxy feels a ram pressure from the cluster gas as it moves through it. Stars are too small and too dense to be affected by this, but the gas is low density and very spread out, so it has no choice but to get clobbered. If the ram pressure is greater than the gravity binding the gas to the galaxy, the gas will be pushed out and lost forever. More on ram pressure stripping here.
 |
| A galaxy consisting only of stars (left) smirks at the cluster gas as unimportant. A galaxy with its own gas (right), however, is in for a big surprise. |
To get a feel for this, a very useful rendering comes from the webpage of astronomer Rukmani Vijayaraghavan. This simulation shows only the hot gas, both in the galaxies and the cluster. It shows very clearly how the gas is stripped and swirled and squished by the galaxies silently swarming about.
Ram Pressure Stripping of a Cluster of Galaxies from Rukmani Vijayaraghavan on Vimeo.
Why mention that this shows only the hot gas ? While the gas in the cluster can be described only as "hot and thin" (...insert joke about David Tennant here), galactic gas can be broadly divided into three main components :
- The hot, diffuse gas shown in Vijayaraghavan's simulations. This is very easy to remove by ram pressure, but it doesn't get involved in star formation much because its heat keeps it from collapsing.
- The warm atomic HI gas, i.e. the best, sexiest gas that we study with radio telescopes. Much denser and cooler than the the hot gas, this can also be removed as long as the ram pressure is strong enough. Exactly how (and even if) it relates to star formation is unclear.
- The cold molecular gas, which we now think is the main component directly involved in star formation. This can only be directly stripped if the pressure is extremely high.
All of these play different roles in star formation - if we want to understand the effect of gas stripping on the life of a galaxy, rather than studying the stripped gas for its own sake, then ideally we'd measure all three components. And all three respond to ram pressure and the cluster gas a bit differently, and need different methods to detect them. This means the above animation is useful, but only as a rough guide to what we expect.
So how do you kill a galaxy ? Think of its gas as its fuel for star formation. If you remove only the hot gas, you've essentially skimmed a bit off the top. Sure, it'll run out of fuel eventually, but it's still got plenty in the tank for now. You can't really get to the cold gas much, which is actually flowing from the tank and already in the pipes. But you can still remove the warm gas, i.e. emptying the tank and letting the galaxy sputter its last bits of fuel into the engine before it finally gives up the ghost.
So how do you kill a galaxy ? Think of its gas as its fuel for star formation. If you remove only the hot gas, you've essentially skimmed a bit off the top. Sure, it'll run out of fuel eventually, but it's still got plenty in the tank for now. You can't really get to the cold gas much, which is actually flowing from the tank and already in the pipes. But you can still remove the warm gas, i.e. emptying the tank and letting the galaxy sputter its last bits of fuel into the engine before it finally gives up the ghost.
Where are all the bloody entrails ?
But all that fuel can't just disappear. Or to take the galaxy death analogy much more literally, courtesy of the fabulous Robert Minchin :
This isn't exactly a whodunnit though. We already think we know who did it (the intracluster gas), how they did it (ram pressure stripping), what they did (strip the galaxies), where and when (using our earlier model and observations, we can calculate these), and also why (because it's physically inevitable). What we don't know - what threatens to undermine the whole otherwise elegant hypothesis - is where the hell are all the bloodstains ?
Or to put it another way, about 60% of all the hundreds of galaxies in the cluster appear to have lost significant amounts of gas, but only 3% of them show streams. Now, it's possible that a lot of them had their gas removed ages and ages ago and it's since dispersed and become undetectable. But we also see a lot of gas-poor galaxies in close proximity to those which are still gas-rich - implying that quite a few ought to be losing gas right now. So naively, we expect to see more streams than we actually do. But how many exactly ?
That's what we tried to quantify in this paper, a problem that had me worried since the heady days of my PhD. In fact, my very first presentation at a major conference was all about a new method of searching for very faint streams that hadn't found anything. The audience humoured me. The prevailing wisdom in the room seemed to be that the cluster's hot gas would rapidly heat up any escaping gas, rendering it undetectable. "It's not really a puzzle", they said. The younger me was no way going to debate this in front of a live audience*, so I said something about just wanting to test a new method, which was bollocks and I knew it.
* Or even a dead audience.
The thing is, there are some streams in the cluster. So how do they survive while most of the others are apparently destroyed more quickly than the reputation of British children's entertainers from the 1970s ?
Known features don't show any particular pattern than could explain this. Some are tiny, some are huge. Some are massive, some pathetic. Some are near the violent, fun-filled cluster centre, where ram pressure and evaporation should be strongest, while others are on the outskirts where nothing much at all is going on. So the mechanism that destroys the streams seems weirdly inconsistent and almost magical, like a mad wizard who hates gassy strippers. Don't tell me that's not a puzzle.
In Vijayaraghavan's animation, you can see the tails flaring out as the gas escapes. But that was hot gas, and simulations of the warm (HI) gas show it should remain narrow and confined. Still, over time even this gas ought to disperse, and if its density becomes too low it should no longer be detectable. So we made a simple model to quantify this, accounting for how the viewing angle changes the appearance of a stream in our data. The bottom line is that if the known streams are representative of the general population, then with our survey we ought to detect about 11 streams (we actually detected 2) and another, larger survey should have detected 46 (it actually detected 5). Not great.
Fortunately, this so-called "geometrical dilution" is just one factor that could explain the missing streams. We also need to consider how many galaxies are currently actively losing gas (i.e. expected to have streams right now) and what happens to the gas after it leaves (i.e. how quickly it's dispersed).
That's where our earlier model comes in. One thing I will concede that the torturous refereeing process did improve was our use of our model to predict how many galaxies should be actively losing gas. I would dearly love to say we could use this to make a prediction, but the reality is that we can't - it has just too many uncertainties and the data we need is available for too few galaxies. Booo ?
Well, yes, but we can still use it to do a couple of neat things. We can say which galaxies are more likely to be losing gas right now (just not for the whole sample, unfortunately), and we can estimate how quickly the gas must be dispersing in each case. The model also still has a big advantage over measurements of how much gas has already been lost, in that it models current stripping activity* - it's just not good enough to make a honest-to-goodness prediction. At least not yet.
* Just as with real life, where it's far more important to know who's about to get naked than who already took their clothes off.
Here they are !
Having done all this, it really seemed like there should be more streams present. Plenty of objects seemed to be experiencing enough ram pressure that they should have streams, yet barely any did. Was it possible they'd simply escaped detection ?
Given that I'd spent bloody years staring at these data cubes, that didn't seem likely. I used to joke that I should go on Mastermind with the VC1 cube as my specialist subject. And I did know more about this particular data cube than anyone else alive, because no-one else had done much more than glance at it. But I was foolish to think I knew everything about it.
See, younger me expected that all streams would be spectacular, really obvious features like this one :
Which is the famous VIRGOHI21 as seen with the Westerbork interferometer. But the data I had was from Arecibo, which is more sensitive but has lower resolution. So the kind of tails we should expect to detect with this ought to be faint and appear to be very short. It was fair to say there were no features as spectacular as VIRGOHI21 lurking in the data, but this didn't mean there weren't any detectable streams present at all.
If my first mistake was not understanding what kind of features to look for, my second was over-confidence in my knowledge of how to look for them. Which didn't come from nowhere : having spent several years in creating a better data visualisation tool, I thought I must surely have looked at the data very thoroughly indeed. In fact I had, but in the wrong way.
I have a penchant for volume renders, which show the whole data present in a cube and look cool. I dismissed isosurfaces and contour plots as beingnot cool enough good for analysis but not for finding sources, because they inherently limit the information shown. Whoops. In fact, isosurfaces - basically just contour plots in 3D - are an awesome tool for finding short extensions, as we'll see if you just bear with me a minute longer.
The thing with our survey is that it's so damn sensitive that it makes the galaxies look incredibly bright. Unless you subtract the "glare" from the galaxy, their tails can remain forever invisible. I'd very successfully subtracted the galaxy emission in previous, more distant cases. Those were easy, because very distant objects become point sources, which have a distinct, known profile shape that's given by the telescope geometry. So you can just input the known shape and subtract it, et voila, the galaxy is removed, leaving the stream behind. Kindof like the Chesire Cat, only waaay less creepy.
And for very close (or very large, well-resolved) sources, you don't need to remove anything : you can see by eye where the galaxy disc ends and its extension begins, if it has any. No need to do anything to the data here at all, you can just go ahead and measure it.
But in this case we were in the unhappy middle ground. The Virgo cluster galaxies were not so distant that they could be considered as point sources : trying the usual subtraction procedure left hideous artifacts that were more offputting than Justin Beiber's Lyme disease. But neither were they close enough that we could directly see the extensions and clearly state that this part is the galaxy's disc and this part is the tail. So what to do about these marginally resolved cases ?
That's where contour plots and isosurfaces come in. Our radio data images the sky just like an optical telescope, but it takes thousands of images of the same region. Each of these frequency channels shows how bright the hydrogen is at slightly different velocities. Since galaxies rotate, different parts of galaxies are visible in different channels. So when you go through channel by channel, you see something like this :
The galaxy (optical image here) slowly drifts across the image, with one side coming towards us (lower velocities) and the other side moving away (higher velocities) due to its rotation. This example shows a case where there's no extension present, so you just get nice circular contours that drift a bit. But looking at every single image is a tiresome as watching the Star Wars Holiday Special, so normally we use different display techniques. One, shown on the left below, is to sum up all the flux along the line of sight - a volume render. The alternative, on the right, is to show all the contours at once with a different colour for each velocity (this is called a renzogram).
Ahh. This is not quite so circular any more. It's not a very pronounced effect in this case, but if the centre of the flux were to drift a bit more, we'd get an elliptical appearance even though the galaxy itself is pretty circular.
Trying to spot extensions in these complicated cases just by looking at them is... well, I'd been looking at this data for the best part of a decade and not seen them. With volume renders, the streams tend to be so faint that it's hard to set a display range for the data that shows them clearly without having them get lost in the glare of the galaxy. With individual channel maps, as well as being incredibly tedious, it's hard to know if any channel contains any features more elliptical (i.e. sticky-out extendy bits) than any of the others. But using renzograms, it becomes much much easier to see the extensions even when the centre of the contours drifts from channel to channel.
And we can tweak this for display purposes. We don't have to show contours for every channel, which can be confusing. Instead we can select the channels which show the extensions most clearly. Here's our main figure showing the streams we're most confident about :
How do we select the best range ? Well, we can also plot the contours in 3D, where these "renzograms" become true isosurfaces. Here's VCC 2070 again. You can clearly see the extension is found only on one side (in velocity) of the galaxy.
And here's the entire data cube, with the surface levels chosen for each galaxy to show everything as nicely as possible. Of course, this works much better when you have a fully interactive model to play with.
So, mystery solved, right ?
Well, umm... well... yeah, actually, it is ! Given how much I've warned y'all about science headlines that use this abominable phrase I don't say this lightly. But in this case, it really seems to be true.
Now you might think, looking at the above figures, that there's not much scope for doubt about the streams. You might think that with even more gusto if you know that most of the galaxies in our data have perfectly decent circular contours with no signs of streams at all - if they all showed such features, it might be due to some problem with the data processing.
But only a few show extensions. And the number of streams well-matches our expectation, and the basic predictions of our model stand up well. Galaxies our model says should be stripping have a much stronger preference to have tails, and galaxies which shouldn't be stripping don't have tails. The morphology, mass, length and velocity of the streams all exactly fits what we expect for ram pressure stripping. Everything just works.
And yet we had a terrible time convincing not one but two referees about this. For the life of me, I can't understand why. It's true that most streams are quite faint, but they're not that faint. So we spent an inordinate amount of time proving - yes, actual proper proof in this case - that the number of similar features we expect to see due to the noise in the data is essentially zero. We injected fake galaxies into empty regions of the cube containing nothing but noise, then searched through these to see how many contained what looked like streams. And to avoid fooling ourselves, we also injected fake streams only into a random number of these tests, so that when searching we couldn't be sure if we were looking at something we'd injected or a genuine artifact.
The bottom line is that we don't think any of our ten most confident detections are false. Our 16 other, less sure detections, may well have a significant fraction of false positives, but it's hard to quantify that. So we basically ignored all except the ten best streams in our analysis, to be on the safe side.
When we run the numbers, we find that we can indeed solve the mystery of the lack of bloodstains. Four factors combine to explain why so few streams were visible. First, they do exist, but it takes the right survey and the right visualisation techniques to find them. Second, you have to account for the number of galaxies currently losing gas. Then, third, the geometry of the streams affects their detectability. And finally, using the observational measurements plus our analytic model, we can calculate how fast the gas disperses. The dispersal rate seems to be fast enough that low-mass streams quickly fade from view, but the high mass ones can last much longer. So there's no magic wizard, just gas evaporating quite naturally.
(Or, if you prefer, if you prick someone with a needle, the slowly dribbling blood will quickly evaporate away. Stab 'em in an artery with a knife and gushing torrent will take ages before it disappears.)
The other thing I have to mention is the orientation of the streams. Google Plus survivors may remember I ran a poll to check if I'd identified the direction of the streams correctly. What was odd about this was that they all seemed to be pointing either towards or away from the centre of the cluster, even those that were from galaxies in separate sub-clusters that are just too far away for the main cluster to have any real influence :
Here I will again reluctantly concede to one of the referees that it was probably better that we took this out of the final paper, but I'm not thrilled about it. What we think is going on is just small number statistics. For the galaxies in the main body of the cluster, we expect there to be this sort of alignment so there's no problem. Once you take those out, the numbers in the sub-clusters are pretty small. It would only take a couple of membership misidentifications before you could find completely different patterns - with regards to cluster centre alignment, we're essentially seeing what we wanted to see, not necessarily what's really there. The same data can be interpreted very differently, depending on what you think is significant.
Of course, we've only looked at a small part of the cluster, so it's still possible there are other mysterious things going on elsewhere. In fact we know there are, because as I described previously, certain features just don't make sense. And we've solved the mystery statistically, not for every single individual object. So this claim that everything is done now is strictly limited. But the fundamental problem of why we don't see that many streams - that, I think, is well and truly done.
What's next ? Well, we've got loads of other old data cubes that we could re-analyse, and brand new Virgo data coming in. With more galaxies, we might be able to say something about the orbits of the galaxies in different parts of the cluster, and see where gas loss is most active. It's also a bit of a puzzle how some galaxies are losing gas despite being in the cluster outskirts. But for now, given the years it took to get this far, it's time to do something else completely. After all, there's only so much research into the death-throes of a windy stripper that any self-respecting astronomer can put up with.
Neutral hydrogen is the life blood of galaxies - it enables them to continue forming stars, and galaxies that have lost their hydrogen are frequently described as 'dead'. Our radio telescope can see this hydrogen, and we can use this to find galaxies. In this new survey, we will be looking at the Virgo cluster - our nearest galaxy cluster. This is a 'galaxy city' - lots of galaxies are crowded together and interact with each other, often violently. We act like forensic scientists trying to piece together what happened from the small bits of evidence we can find: 'wounded' galaxies that are in the process of turning into 'dead' elliptical galaxies and dark clouds of hydrogen lying around outside of their original galaxies, like pools of blood at a murder scene. These clues allow us to peer deeper into the violent world of the Virgo cluster and trace the fate of its denizens.I could add that since the galaxies are dying through loss of gas, they're essentially farting themselves to death. Lovely. Fortunately I won't add this, because the blood analogy works much better, as we'll see.
This isn't exactly a whodunnit though. We already think we know who did it (the intracluster gas), how they did it (ram pressure stripping), what they did (strip the galaxies), where and when (using our earlier model and observations, we can calculate these), and also why (because it's physically inevitable). What we don't know - what threatens to undermine the whole otherwise elegant hypothesis - is where the hell are all the bloodstains ?
Or to put it another way, about 60% of all the hundreds of galaxies in the cluster appear to have lost significant amounts of gas, but only 3% of them show streams. Now, it's possible that a lot of them had their gas removed ages and ages ago and it's since dispersed and become undetectable. But we also see a lot of gas-poor galaxies in close proximity to those which are still gas-rich - implying that quite a few ought to be losing gas right now. So naively, we expect to see more streams than we actually do. But how many exactly ?
That's what we tried to quantify in this paper, a problem that had me worried since the heady days of my PhD. In fact, my very first presentation at a major conference was all about a new method of searching for very faint streams that hadn't found anything. The audience humoured me. The prevailing wisdom in the room seemed to be that the cluster's hot gas would rapidly heat up any escaping gas, rendering it undetectable. "It's not really a puzzle", they said. The younger me was no way going to debate this in front of a live audience*, so I said something about just wanting to test a new method, which was bollocks and I knew it.
* Or even a dead audience.
The thing is, there are some streams in the cluster. So how do they survive while most of the others are apparently destroyed more quickly than the reputation of British children's entertainers from the 1970s ?
 |
| Map of the known optically HI (atomic hydrogen, a.k.a. warm neutral gas) features in Virgo. Black arrows show streams to their actual scale. Grey arrows show smaller features and are not to scale. Black diamonds show unresolved clouds. Red, green and blue points show optically bright galaxies, while the big grey rectangle highlights our survey area. |
In Vijayaraghavan's animation, you can see the tails flaring out as the gas escapes. But that was hot gas, and simulations of the warm (HI) gas show it should remain narrow and confined. Still, over time even this gas ought to disperse, and if its density becomes too low it should no longer be detectable. So we made a simple model to quantify this, accounting for how the viewing angle changes the appearance of a stream in our data. The bottom line is that if the known streams are representative of the general population, then with our survey we ought to detect about 11 streams (we actually detected 2) and another, larger survey should have detected 46 (it actually detected 5). Not great.
Fortunately, this so-called "geometrical dilution" is just one factor that could explain the missing streams. We also need to consider how many galaxies are currently actively losing gas (i.e. expected to have streams right now) and what happens to the gas after it leaves (i.e. how quickly it's dispersed).
That's where our earlier model comes in. One thing I will concede that the torturous refereeing process did improve was our use of our model to predict how many galaxies should be actively losing gas. I would dearly love to say we could use this to make a prediction, but the reality is that we can't - it has just too many uncertainties and the data we need is available for too few galaxies. Booo ?
Well, yes, but we can still use it to do a couple of neat things. We can say which galaxies are more likely to be losing gas right now (just not for the whole sample, unfortunately), and we can estimate how quickly the gas must be dispersing in each case. The model also still has a big advantage over measurements of how much gas has already been lost, in that it models current stripping activity* - it's just not good enough to make a honest-to-goodness prediction. At least not yet.
* Just as with real life, where it's far more important to know who's about to get naked than who already took their clothes off.
Here they are !
Having done all this, it really seemed like there should be more streams present. Plenty of objects seemed to be experiencing enough ram pressure that they should have streams, yet barely any did. Was it possible they'd simply escaped detection ?
Given that I'd spent bloody years staring at these data cubes, that didn't seem likely. I used to joke that I should go on Mastermind with the VC1 cube as my specialist subject. And I did know more about this particular data cube than anyone else alive, because no-one else had done much more than glance at it. But I was foolish to think I knew everything about it.
See, younger me expected that all streams would be spectacular, really obvious features like this one :
Which is the famous VIRGOHI21 as seen with the Westerbork interferometer. But the data I had was from Arecibo, which is more sensitive but has lower resolution. So the kind of tails we should expect to detect with this ought to be faint and appear to be very short. It was fair to say there were no features as spectacular as VIRGOHI21 lurking in the data, but this didn't mean there weren't any detectable streams present at all.
If my first mistake was not understanding what kind of features to look for, my second was over-confidence in my knowledge of how to look for them. Which didn't come from nowhere : having spent several years in creating a better data visualisation tool, I thought I must surely have looked at the data very thoroughly indeed. In fact I had, but in the wrong way.
I have a penchant for volume renders, which show the whole data present in a cube and look cool. I dismissed isosurfaces and contour plots as being
The thing with our survey is that it's so damn sensitive that it makes the galaxies look incredibly bright. Unless you subtract the "glare" from the galaxy, their tails can remain forever invisible. I'd very successfully subtracted the galaxy emission in previous, more distant cases. Those were easy, because very distant objects become point sources, which have a distinct, known profile shape that's given by the telescope geometry. So you can just input the known shape and subtract it, et voila, the galaxy is removed, leaving the stream behind. Kindof like the Chesire Cat, only waaay less creepy.
 |
| From an earlier paper. |
But in this case we were in the unhappy middle ground. The Virgo cluster galaxies were not so distant that they could be considered as point sources : trying the usual subtraction procedure left hideous artifacts that were more offputting than Justin Beiber's Lyme disease. But neither were they close enough that we could directly see the extensions and clearly state that this part is the galaxy's disc and this part is the tail. So what to do about these marginally resolved cases ?
That's where contour plots and isosurfaces come in. Our radio data images the sky just like an optical telescope, but it takes thousands of images of the same region. Each of these frequency channels shows how bright the hydrogen is at slightly different velocities. Since galaxies rotate, different parts of galaxies are visible in different channels. So when you go through channel by channel, you see something like this :
The galaxy (optical image here) slowly drifts across the image, with one side coming towards us (lower velocities) and the other side moving away (higher velocities) due to its rotation. This example shows a case where there's no extension present, so you just get nice circular contours that drift a bit. But looking at every single image is a tiresome as watching the Star Wars Holiday Special, so normally we use different display techniques. One, shown on the left below, is to sum up all the flux along the line of sight - a volume render. The alternative, on the right, is to show all the contours at once with a different colour for each velocity (this is called a renzogram).
Ahh. This is not quite so circular any more. It's not a very pronounced effect in this case, but if the centre of the flux were to drift a bit more, we'd get an elliptical appearance even though the galaxy itself is pretty circular.
Trying to spot extensions in these complicated cases just by looking at them is... well, I'd been looking at this data for the best part of a decade and not seen them. With volume renders, the streams tend to be so faint that it's hard to set a display range for the data that shows them clearly without having them get lost in the glare of the galaxy. With individual channel maps, as well as being incredibly tedious, it's hard to know if any channel contains any features more elliptical (i.e. sticky-out extendy bits) than any of the others. But using renzograms, it becomes much much easier to see the extensions even when the centre of the contours drifts from channel to channel.
 |
| The clearest stream we detected was from the galaxy VCC 2070. The orange contours show a tail, whereas the blue contours (at higher velocities) are nicely circular. It's much easier to see the sticky-out extendy bits when you've got the circular bits visible for comparison at the same time. |
How do we select the best range ? Well, we can also plot the contours in 3D, where these "renzograms" become true isosurfaces. Here's VCC 2070 again. You can clearly see the extension is found only on one side (in velocity) of the galaxy.
And here's the entire data cube, with the surface levels chosen for each galaxy to show everything as nicely as possible. Of course, this works much better when you have a fully interactive model to play with.
So, mystery solved, right ?
Well, umm... well... yeah, actually, it is ! Given how much I've warned y'all about science headlines that use this abominable phrase I don't say this lightly. But in this case, it really seems to be true.
 |
| Finding the streams one expects to find is very nearly as surprising as finding a tiny hedgehog in one's tea. |
But only a few show extensions. And the number of streams well-matches our expectation, and the basic predictions of our model stand up well. Galaxies our model says should be stripping have a much stronger preference to have tails, and galaxies which shouldn't be stripping don't have tails. The morphology, mass, length and velocity of the streams all exactly fits what we expect for ram pressure stripping. Everything just works.
And yet we had a terrible time convincing not one but two referees about this. For the life of me, I can't understand why. It's true that most streams are quite faint, but they're not that faint. So we spent an inordinate amount of time proving - yes, actual proper proof in this case - that the number of similar features we expect to see due to the noise in the data is essentially zero. We injected fake galaxies into empty regions of the cube containing nothing but noise, then searched through these to see how many contained what looked like streams. And to avoid fooling ourselves, we also injected fake streams only into a random number of these tests, so that when searching we couldn't be sure if we were looking at something we'd injected or a genuine artifact.
 |
| Small selection of our fake galaxies, some with fake streams and some without. In other cases we also tried varying the direction, but here they all run from the centre of the galaxy to the right. Although in some cases a fake stream was injected but isn't clearly visible, hardly any show streams which are actually just artifacts due to the noise. |
When we run the numbers, we find that we can indeed solve the mystery of the lack of bloodstains. Four factors combine to explain why so few streams were visible. First, they do exist, but it takes the right survey and the right visualisation techniques to find them. Second, you have to account for the number of galaxies currently losing gas. Then, third, the geometry of the streams affects their detectability. And finally, using the observational measurements plus our analytic model, we can calculate how fast the gas disperses. The dispersal rate seems to be fast enough that low-mass streams quickly fade from view, but the high mass ones can last much longer. So there's no magic wizard, just gas evaporating quite naturally.
(Or, if you prefer, if you prick someone with a needle, the slowly dribbling blood will quickly evaporate away. Stab 'em in an artery with a knife and gushing torrent will take ages before it disappears.)
The other thing I have to mention is the orientation of the streams. Google Plus survivors may remember I ran a poll to check if I'd identified the direction of the streams correctly. What was odd about this was that they all seemed to be pointing either towards or away from the centre of the cluster, even those that were from galaxies in separate sub-clusters that are just too far away for the main cluster to have any real influence :
 |
| Streams in our survey area. The red arrows show the actual stream direction, while green arrows point towards the cluster centre. Grey and black outlines highlight galaxies which are in more distant sub-clusters where there shouldn't be any coherent alignment towards the main cluster. |
 |
| If we give ourselves a little slack, we can select regions where the streams appear to be aligned in very different ways. So the large-scale alignment is probably just a coincidence. |
What's next ? Well, we've got loads of other old data cubes that we could re-analyse, and brand new Virgo data coming in. With more galaxies, we might be able to say something about the orbits of the galaxies in different parts of the cluster, and see where gas loss is most active. It's also a bit of a puzzle how some galaxies are losing gas despite being in the cluster outskirts. But for now, given the years it took to get this far, it's time to do something else completely. After all, there's only so much research into the death-throes of a windy stripper that any self-respecting astronomer can put up with.
No comments:
Post a Comment
Due to a small but consistent influx of spam, comments will now be checked before publishing. Only egregious spam/illegal/racist crap will be disapproved, everything else will be published.