Twenty Years In Astronomy

Twenty years ago this month marks the start of my career in astronomy. So, hooray, I'm old !

Truly, a life where weekly Bingo is the biggest attraction makes a strong case for assisted dying.
Yes, I'll regret that joke in another twenty years, but I'm not there yet.

This gives the perfect opportunity for the blog equivalent of a nostalgia fest a.k.a. clip show. If we as scientists are actually doing our jobs, we ought to have solved at least some of the problems that were plaguing astronomy in the mid-2000s. So how've we done ? How does astronomy today, both in terms of the science itself and the way it's carried out, differ from the halcyon days of 2006, and where are we going next ? 

Let's find out.

Part One : Science Marches On

I hasten to point out that this is going to be a completely arbitrary list of things I personally find most professionally interesting. No attempt is made at collating all the interesting stuff astronomy has done for the last two decades, because that would be a stupid thing to do.

1) Lie-onisation

I want to start with something small that was once treated as common knowledge but now seems very unlikely. At least, the way I was taught this was as a sort of "everyone knows" claim, that being that HI gas (neutral atomic hydrogen) would always be ionised at low densities. Specifically, below about 10¹⁹ atoms cm^-2, it should, so everyone told me, be entirely ionised by the cosmic UV background. Maybe not instantly, but quickly enough that we wouldn't detect anything.

This claim didn't come from nothing. There were in fact observations showing that the edge of the gaseous disc of galaxies was truncated at about this density... but the claim that "everything is always ionised" is so absolute that it can be disproved with a single counter-example. And findings over the last two decades have left this claim as little more than a rather dull piece of astronomy folklore*.

* If you follow Decoherency, you'll know that my idea of entertaining folklore involves a great deal of sex and violence. The precise value at which gas is ionised doesn't really hold a candle to the Polish tale of a psychotic bear with Tardis ears who attempts to kill a random girl by throwing a bunch of keys at her for no good reason, let alone the Bulgarian story in which God embraces cannibalism or the Czech one with the mysterious hot lesbian fairy.

In this rather crude (because I'm recoding my visualisation tool again) figure of M33, the white contour shows the limit of the gas at the 10¹⁹ limit. Everything beyond this should not exist.

Much later, I learned that the star formation community never expected anything like such a simple density limit, knowing (as they do) a great deal more about the microphysics than us extragalactic astronomers. And we now have considerable evidence that the rumoured ionisation simply doesn't happen, or at the very least, it certainly isn't any sort of universal truth. In the profile of M33, shown above, we see gas directly detected at densities almost a hundred times below the supposed threshold. Others have used stacking to combine the results from less sensitive observations on larger samples, and they also find no clear break in the density profiles. It's pretty much inescapable that this ionisation threshold was simply wrong.

Why would anyone care ? I think it's important to acknowledge when things that are taken as mundane facts quietly change without any big announcements. It might very well be the case that some galaxies do indeed have sharp edges to their gas discs*, but we can say categorically that this not true in all (or even most) cases.

* An interesting caveat is that there might still be a much lower threshold below which everything really is ionised. In that case, we ought to be able to detect the ionised gas directly, given sufficiently sensitive observations.

2) Big Disc Energy

A nice corollary to the above concerns the extended UV discs which enjoyed a brief moment of fame and glory exactly when I was starting my PhD. This was the discovery that about a third of spiral galaxies were found to have low-level UV emission from well beyond their main stellar discs. To find that there might be some stars at greater distances wasn't really surprising, but UV specifically traces the hot, massive, young stars that don't live for very long. That's a bit odd.

GALEX UV image of M83. The bright central disc is perfectly normal. The huge faint extensions came as quite the surprise. But if you're wondering about the first point, this UV emission isn't enough to ionise the outer gas.

Well, their short lifespans meant that there simply wasn't time for these stars to have formed in the main disc and migrated outwards, because they should have died before they got there. But if they were formed in situ, where were the older stars ? For every massive, UV-bright star that forms, you expect to find an additional, larger number of smaller, dimmer, much longer-lived stars as well. And they seemed to be weirdly missing. In fact it was even more oddly specific than this : UV emission actually traces the second-tier massive stars, not the most massive of all... and those too were missing.  

So what the hell was going on ? Was this some sort of cosmic Logan's Run, only worse ?

The solution to this terrific puzzle is another example of something that didn't really get a clear, flashy announcement. There was no classic "mystery solved, say scientists" press release. Which is ironic : this is one of the few cases where such an over-used declaration would actually have been valid, and it didn't happen.

The main answer here seems to have been mainly a selection effect. The GALEX UV satellite was disproportionately sensitive to UV in comparison to the optical data of the day, so the older stars were there, they were just harder to spot. 

But there were some other factors at work too. These UV-emitting stars live a lot longer than the brighter H-alpha sources, so those giant stars likely do exist as well... just more intermittently. And XUV discs were discovered at the same time that the received wisdom of a sharp edge to the gas disc was beginning to fade; people were already finding gas at much further distances than they'd previously suspected. True, overall it was lower density than you typically see in star-forming regions, but locally it might well reach the critical density from time to time. 

The outermost regions of a galaxy, then, are just dense enough to keep star formation ticking over, without ever reaching the blazing levels found deeper in its innards. Occasional flickers of star formation are more than enough to maintain the vast ghostly tendrils seen at just the right wavelengths.

3) Dark Galaxies Are Dead, Long Live Dark Galaxies

My very first conference, in 2007, was an extremely lively affair in which there were some fantastic arguments and one occasion where people nearly came to physical blows (not hyperbole) over whose model of dark galaxies was better. You rarely seem to get that any more, which is probably for the best... although sometimes I miss the "tell us how you really feel" atmosphere of those earlier days.

Anyway, back then the idea of galaxies with no stars at all – just gas and dark matter – really was intensely controversial. The problem was, as I've written about ad nauseum, that cosmological models predicted far more dwarf galaxies than we actually see. With hindsight those early models were hopelessly simplified, but the basic premise is now essentially the mainstream consensus : the smallest dark matter clumps, we think, really do form, it's just that most of them never accumulate enough gas to ever form any stars (or at least so few as to be all but undetectable).

The Via Lactea simulation produces a Milky Way-mass central object but which is surrounded by thousands of smaller sub-haloes. We actually see more like sixty or so. Not sixty thousand, just sixty.

This too is a case of a silent shift. What was once seen as an almost silly idea – though it was never quite fringe science – is now the accepted best explanation. And there's some pretty decent evidence for this, both theoretical and observational. On the theory side, modern simulations are now massively more sophisticated than those of 2006, and they don't have a missing satellite problem. Observationally, populations of so-called "Ultra-Compact High Velocity Clouds" have been found which seem to fit the predictions quite well.

But there are two major caveats. First, the kind of much more massive dark galaxy candidates people were proposing back then are now all but dead as idea; I myself have spent most of my career investigating such objects and concluded that our best candidates are something altogether different. Those objects are still sometimes treated with a level of disdain (but for caveats on this caveat, see below). Second, securely identifying any individual object as a dark galaxy* remains extremely challenging. Statistically the evidence is good, but I personally would be wary of declaring "mystery solved" for the missing satellite problem until we can robustly say we've found a dark galaxy beyond reasonable doubt. We've made huge progress, but we're not there yet.

 * Unfortunately the terminology has proliferated. "Dark galaxy" is now often used to refer only to the larger objects, and one group in particular insists on calling everything "almost dark" which I find immensely irritating. Smaller objects are often referred to as "minihalos" and "RELHICS", but functionally these are identical to dark galaxies.

4) Ultra Diffuse Galaxies

Here's where things get fun. From 2015, people started detecting large numbers of very low surface brightness galaxies. Such objects had long been known about as exotica, but now they started turning up in droves. Claims that some of these had in fact been detected years before are technically true (some were known since at least the 1980s) but missing the point : what once seemed like a few extremes were revealed to be part of entire population.

This in itself wasn't terribly surprising. What was surprising was just about everything else about them, and this continues to be the case to the present day. For starters, these weren't boring little tiddly smudges, but great big Milky Way-sized monsters. Nobody expected that*. All our problems seemed to be at the small end, and to find out we might have messed up with the high-mass end as well came as quite the shock.

* Actually Mike Disney did predict a population of low surface brightness galaxies back in 1976, but he subsequently claimed that UDGs weren't what he predicted, so this (rather confusingly) doesn't count.

From one of the very first discovery papers. It's still an emblematic image, and rightly so : this was one of those "whoa" moments rather than the more usual incremental findings that take more effort to get excited about.

My favourite aspect to this is their probable dark matter content. The debate is by no means settled, but the emerging picture looks like the following. Many, perhaps most, UDGs appear to be huge dwarves. That is, their stars are unusually spread out, but their total dark matter content isn't especially high. But a significant minority may well be the "failed giants" that models never predicted, not only as large as the Milky Way but also just as massive. Probably not enough to give us a real cause of concern about the fundamentals, but this is yet to fully play out. And nobody predicted UDGs as a population anyway, regardless of total mass.

Where it gets really fun is at the opposite end. Two headline galaxies have captured attention as being apparently entirely free of dark matter (literally, their missing matter is missing, yo dawg...). Weird though they are, I don't find these two actually that interesting : the explanations that have been put forth are convincing enough that we can explain these two quite nicely, I think. That is, the investigations have involved a lot of careful work and the explanations turn out to be extremely clever and unexpected*. I don't mean to say that these objects are "boring", because they certainly aren't ! Rather, what I mean is that these are special cases : worthy of study, interesting in their own right, but lacking clear implications for the wider population.

* In brief : dark matter can be preferentially removed because dark matter particles have orbits which take them much further out than the stars and gas, where they can be more susceptible to external gravitational forces. 

Not so the UDGs found in isolation. About three hundred of these have been detected with HI, and these as a population show a tendency to have less dark matter than expected : much less, albeit not often to the level of being completely deficient. You can't blame external disturbances for these. How these form remains a genuine mystery, one absolutely nobody in 2006 would ever have predicted.

5) Blue Blobs

Around the same time as UDGs, another class of object started turning up that's had far less attention but might, just might, help to tie some of this together. So far they've only been detected in Virgo, which means we should be extremely cautious about inferring what they mean more generally. But fuck it, this is a blog post, not a paper. To hell with caution !

Anyway, the so-called "Blue Blobs" have the creative naming potential that would make whoever came up with the "Very Large Telescope" blush. Like UDGs, they're of very low surface brightness, but they tend to be more structured and can be extremely gas rich. Like, crazy gassy, as in "seek medical attention immediately". Whereas your typical galaxy might have at most about as much gas as stars, these objects can have (in extreme cases) thousands of times more gas than stars.

A curse of being interested in faint stuff is that it's bloody hard to find any pretty pictures to impress people. This example comes from the Dey et al. 2025 catalogue paper.

The latest development on this front suggests that these are ram pressure dwarves. We've known about tidal dwarves for ages, in which interacting galaxies tear bits off each other and those bits go on to survive and live independently. Here it looks as though ram pressure stripping is removing the gas from the parent galaxies : as they move though the much thinner, much hotter gas in the cluster, the pressure can displace their own gas into long tails and eventually remove it completely. Parts of those tails, it seems, can condense and survive long after their parent has fled the scene.

This explanation works well for the chemistry of the gas, the extraordinary gas fractions of the BBs, and their lack of obvious parents. What's less clear is why parts of the stripped gas survive at all, given that most of the stripped material really does appear, in this case, to be rapidly ionised after its removal.

I find these objects especially interesting for two reasons. First, some of them are associated with the optically "dark" HI clouds I spent so long investigating, and that changes how we understand those objects substantially. Rather than ascribing their peculiar kinematics to being possible dark galaxies, it now looks like turbulent mixing with the cluster gas might be responsible*. It now looks as though these are indeed a weird class of object after all, just nothing like what we were suspecting.

* And more importantly, our dark clouds are actually a diverse population of objects. Explaining one or two as tidal debris was never a problem, it was explaining all of them as these incredibly transient objects that we showed, definitively, doesn't work.

And secondly, they hint – and just hint mind you – at a more unified picture of what's going on. Our classical view of galaxies has long been :

Nice and simple. Stars, gas, dark matter. Bam, done.

But now it looks like we might have more of a sequence :

Almost certainly reality is more complex than this, and really this diagram should have more axes. But again, this isn't a paper.

With the extremely interesting bit being that both UDGs and BBs appear to lack dark matter. The possibility of a direct connection hasn't yet been explored much, but it's awfully tempting to join the dots.

So galaxian parameter space appears to be fully sampled, from objects of almost entirely dark matter to those consisting entirely of gas and/or stars. The issue is whether this really represents an evolutionary sequence. Do Blue Blobs relate directly to UDGs or are they just coincidentally similar ? Could they actually form from UDGs which lose their gas through ram pressure, rather than from the stripped material itself ? Why don't we see BBs elsewhere, and how does their gas survive ? And how the hell do dark matter deficient objects apparently form in isolation ?

If there's one lesson that the last twenty years have taught me, it's that the answers can be found... but they're only rarely what you expect. And that's part of the appeal of the job.

Part Two : A Parade Of Paraphernalia 

I want to end with a brief look at how the practice of astronomy has changed and some thoughts on the future. In many ways, the core process of doing astronomy in 2026 remains very much like it did in 2006, but there have been some interesting changes... and there are certainly far more on the horizon.

The more things change...

One surprising bit of similarity is that dark galaxies are back in vogue. In the last few weeks there have been maybe a half-dozen papers on the subject, making this an ideal time to write this post as I haven't had time to read a single one of them yet. Will the 40th anniversary post in 2046 look back on this as another key development, or will it instead be like last time, in which dark galaxies are trendy for a little while only to become discretely buried ?

One huge change here is in the sheer volume of publication rates. Submissions to arXiv have approximately doubled in the last 20 years, and this isn't entirely a good thing. Journal standards haven't shifted to keep pace, resulting in a deluge of literature which is increasingly difficult to read just due to sheer time constraints. A new approach is needed here with some urgency, as reading excessive papers can certainly constitute negative work. We need clearer standards of what papers are for, the role of authors, reviewers and editors, and new formats to allow readers to get to the crucial points without having to wade through copious amounts of text that is of no use or relevance to them.

Software, too, has progressed enormously. In particular, when I began, there was no widely-accepted source-finding tool for HI work and most work was still visual. I myself still cling to this approach, having quantified at great effort that this is in fact still the gold standard. But I have to admit, the SoFiA algorithm is tremendously powerful, far better – far better – than anything conceivable in 2006. Even as I would still warmly encourage people to look at their own fucking data, I would welcome SoFiA as not just a tool for the larger data sets we now have available, but something genuinely beneficial rather than a necessary evil as some of the early algorithms were.

More broadly, expectations of software are switching, but slowly. It's now much more common to talk about user interface and experience without this being dismissed as somehow lesser than scientific problems; I for one fervently believe that this stuff really matters. Not only should we damn well have software which is actually nice to use (provide a god damn GUI, FFS people !), but it changes the end result. If something is easy to use, you can use it routinely, and instead of only checking special cases, you can apply it everywhere. That kind of change can have direct scientific impacts. The interface shouldn't be seen as an optional extra, and this is somewhere where AI has the very real prospect of bringing about a genuine sea change in scientific practice. Adoption of the principle that "things should be nice to use" remains stubbornly slow, however, and more work is needed to bring about something which really should be absolutely common bloody sense.

The wider impact of AI remains much harder to gauge. I periodically (but irregularly) test ChatGPT on scientific questions and performance and it's shown massive improvements. I wouldn't trust it to do actual research at this stage, but it would be foolish in the extreme not to consider this as entirely plausible even in the short term. In certain domains it's already better than some experts; those who are pretending that no further improvement will occur are almost certainly deluding themselves.

But with AI development being both rapid and unpredictable, I don't want to go any further down that particular rabbit hole. Rather, I want to first note the tremendous technological advances happening in scientific infrastructure, a.k.a. telescopes. FAST and SKA pathfinders are already making the HI studies of 20 years ago obsolete in terms of sheer catalogue size. Euclid and Roman are opening up the low surface brightness universe in a way that would have caused earlier astronomers to have heart palpitations. Not everything is rosy here (the UK's decision to cut funding to Jodrell Bank, though this may be reversed, is utterly baffling), but the overall trajectory is one that is steeply rising.

...the more they stay the same

Which brings me back to the first and final topic : science. I've already mentioned the unanswered questions on the nature of "blue blobs" and the dark matter content of UDGs. We've also now firmly established that we don't know where galaxies are getting their gas from. Star formation rates appear to have been stable for quite some time, which implies that galaxies must be maintaining their gas content. But star formation itself should have caused a significant reduction in gas content... which means they must be being resupplied from somewhere. Perhaps with the next generation of technology, we'll finally be able to see the signatures of gas accretion directly.

One thing we can certainly look forward to is probing the atomic gas out to much, much greater distances. Star formation in the nearby universe is a pale shadow of what it was at so-called "cosmic noon", about 10 billion years ago, but we can only directly detect HI in individual galaxies out to maybe 3 billion years ago. The SKA should allow us to go much further, and measure what the gas was really doing when all the exciting stuff was going on. We can theorise to the nth degree, but there's no substitute for observations.

And of course, we can expect the unexpected. We'll now doubt continue to go down the occasional blind alley which will see half the community periodically obsessed with things that turn out to be totally pointless, but that's the nature of research. Sometimes major results only turn up through painstaking and careful work, sometimes they come at you full in the face and suddenly you have to chuck out half your findings. That is as it should be.

---

My hope for 2046 is that we'll be in much the same situation as we are now : we'll have solved most of the problems that seem difficult today (I'd especially like to know what dark matter really is and if the Hubble "tension" is real) but have whole new ones to tackle instead. I'd like us to have moved beyond a "publish or perish" culture into one where we publish more but much smaller and more accessible papers. I'd like to think that we'll have figured out a good way to use AI as a valuable assistant to enhance our own abilities rather than letting it do all our thinking for us. 

Above all, what I really hope is that we get some bloody good surprises. I'd love to see a true paradigm shift like the one Einstein instigated, preferably to an idea not yet even conceived. To have the rug pulled from under our feet and enter a whole new world akin to the discovery, a century ago now, that galaxies themselves were distant beyond all previous imaginings. I think that would be something quite  terrifying and wonderful.