Rob Simcoe seems slightly nervous as he fishes a balloon from a canister of liquid nitrogen. "I haven't done this demonstration before," the astronomer admits, moving the balloon over to a nearby candle flame. The balloon is actually a galaxy. Heated by the candle, it should expand and pop, just as we think galaxies in the early universe produced violent outflows of hot gas when warmed by exploding stars.


At the last second, Simcoe fumbles. The candle burns a hole in the balloon's side and is extinguished by escaping cold air. As the smell of burnt rubber reaches the front row of the lecture hall at the Massachusetts Institute of Technology, laughter spreads. Astronomers, says Simcoe in an elegant save, don't usually do lab demonstrations.

Would that they could. We would love to know more about galaxies, those luminous punctuation marks in a seemingly monotonous cosmos. We know they are the nurseries of stars and, in at least one case, life. We admire their dazzling diversity: big ones, small ones, fiery star factories and quiescent discs, elegant spiral-shaped pinwheels and amorphous blobs. But how did all that diversity arise? Things might be clearer if only astronomers could test out their theories in the lab.

Increasingly, they can. Not with balloons and candles, true, but with the number-crunching power of supercomputers. Together with ever-improving observations of the early universe, grand simulations are beginning to paint a single, unifying picture of why the universe looks as it does. At its heart is an almost invisible scaffold of dark matter and cold gas on which the visible constituents of the universe hang - a structure known as the cosmic web.

Now simulations and observations are in tandem fleshing out this web's vital role. It seems it is far more than just a framework. According to the latest ideas, its filaments are in fact the umbilical cords that provide galaxies with the nourishing gas they need to grow.

That would be an intriguing twist in the cosmic web's story. The concept of the web first arose in the mid-1980s as a by-product of a rather different picture of galaxy formation, known as the hierarchical model, that was proposed in 1978 by Martin Rees and Simon White of the University of Cambridge.

This model started with a universe that expanded breathtakingly fast following the big bang, but not evenly. Tiny irregularities in the distribution of matter created regions of higher density in which "haloes" of invisible dark matter quickly coalesced. Each of these created a deep well of gravity into which gas zinging freely around the cosmos at the time - principally hydrogen - quickly fell. Well and truly trapped, this gas heated up rapidly and ignited into the first stars.

These agglomerations of matter, the seeds of future galaxies, were islands of activity in an otherwise almost empty cosmos - the nodes of the developing cosmic web. They were anything but static, though. Attracted by their mutual gravity, they began to collide and consume each other in violent mergers that triggered convulsive bouts of star formation. Star-forming then subsided until another merger stirred things up, creating another, even larger body. The varied galaxies we see now are the result of that stop-start construction programme.

This picture seemed to have a lot going for it. It fitted very well with observations of the early cosmos. Ever finer measurements of the cosmic microwave background - a snapshot of the universe when it was just 300,000 years old - confirmed the picture of irregularities arising in the spread of matter in the universe. Invoking only the power of gravity, the hierarchical model provided a smooth pathway from those tiny density fluctuations to the mature galaxies we see today.

Then cracks began to appear. One of the biggest showed up in 2006, when astronomer Reinhard Genzel and his colleagues published observations of some of the very earliest star-forming galaxies with the European Southern Observatory's Very Large Telescope at Paranal in Chile. They were exploiting a new technique called integral field spectroscopy, which involves analysing the colour of light from tiny spots on the sky 20,000 times smaller than the full moon to reveal whether they are moving towards or away from the Earth. Used on distant galaxies, it can reveal something of their internal dynamics. "In terms of technology it was really a tremendous breakthrough," says Genzel.

Strange calm

It uncovered something rather odd: a galaxy so far away that when the light now arriving at Earth left it the universe was only 3 billion years old - not even a quarter of its current age. The galaxy was very bright. Indeed, it was forming stars with a total mass of about 100 times our sun's mass every year.

Yet there was no sign of any violent collision. A central conclusion of the hierarchical model is that star formation happens in galaxies that are undergoing mergers. This galaxy, however, was a neatly rotating disc, not the bulky, irregular object you might expect to result from gas plunging chaotically inwards during a merger (Nature, vol 442, p 786).

It was not the only anomaly of its kind. Between 2006 and today, Genzel and his team have collected images of dozens more early galaxies that look placid, but are forming new stars with vigour. For these galaxies at least, the picture of hierarchical mergers is clearly not the whole story. So what is?

The options were limited. The cosmic background observations provided a clear picture of density fluctuations in the very early universe that any new model could not ignore. The idea of dark-matter haloes that grow and merge seemed incontrovertible too: mysterious as dark matter is, its unseen gravitational hand is needed to steady galaxies and stop them flying apart under the force of their own rotation.

The only room for manoeuvre lay with what happened to the gas involved in star formation. "The details of the gas were never on as firm a footing as the idea of the dark matter haloes merging," says Simcoe.

In the early 2000s, Avishai Dekel and Yuval Birnboim of the Hebrew University of Jerusalem in Israel had already been constructing supercomputer simulations to test out alternative scenarios along those lines. Their idea was that, rather than being under the sway of dark matter from the very beginning, gas could be quietly accumulating on the haloes all the time.

The initial results were promising. A continuous stream of cold gas falling onto a dark-matter halo could be compressed and piped efficiently into the developing galaxy's centre, where it would ignite into new stars. The implication was that mergers weren't needed for stars to form.

Smooth operator

Where would this gas come from? In 2005, Dušan Kereš of the Harvard-Smithsonian Center for Astrophysics in Cambridge, Massachusetts, and his colleagues came up with an answer. They simulated the movement of around 2 million "particles" representing bits of independently evolving gas and dark matter making up 1000 early galaxies. Startlingly, they found that streams of cold gas seemed to form naturally, linking one galaxy to the next through the near-void. If the galaxies were the nodes of the cosmic web, these streams were its filaments - feeding tubes sucking up gas from the surrounds that allowed galaxies to form stars steadily and quietly over time.

In January this year, Dekel and his colleagues removed a last objection to this picture: that such understated flows could not possibly supply galaxies with all the gas they need. They published the results of a supercomputer simulation involving a billion gas particles and 4 billion dark matter particles that modelled a junction of three filaments in the cosmic web. They found that gas equivalent to 200 times the mass of the sun could flow into the node every year, enough to support the rate of star formation observed in young galaxies (Nature, vol 457, p 451).

"The secret of cold streams is that they bring in gas smoothly," says Dekel. Without the violence of mergers, far less gas and energy go to waste, leaving more to be ploughed into making stars. So smooth is the process that neat, disc-like galaxies could form early in the universe's history, just as Genzel and his team had seen. To cap that, in April astrophysicist Bruce Elmegreen of IBM, together with Frederic Bournaud of the French Atomic Energy Commission's Service for Astrophysics in Saclay, France, showed how clumps of matter created by gravitational instabilities in discs fed by cold streams could evolve into the spiral arms characteristic of many mature galaxies such as our own Milky Way (The Astrophysical Journal, vol 694, p L158).

The new picture doesn't overthrow the merger idea completely. Merging clumps of dark matter are thought to be important both for the formation of dark matter haloes in the first place, and probably also for the formation of galaxies bigger than the Milky Way. More than 1 in 10 galaxies are massive, elliptical bodies lacking internal features like spiral arms. These galaxies may have formed when smaller systems that had grown by accreting gas from cold streams themselves merged.

For that reason, Dekel describes the cosmic web's new role as a generalisation rather than a refutation of the merger picture. The material falling into galaxies along filaments is sometimes a smooth flow, forming smaller disc-like galaxies, but can sometimes include large chunks of matter, turning the process into something akin to a merger.

According to the models, the accretion of cold gas could be continuing today. In the early universe, the filaments feeding galaxies were around 10,000 light years across, 10 times smaller than a galaxy like the Milky Way. But as the universe has expanded, so the feeding tubes have become wider and less well defined. That makes cold accretion less efficient than it once was, slowing the rate of star formation in mature galaxies such as our own. Cold streams today are less a deluge, and more a gentle rain of gas falling on galaxies from all sides.

That is one good reason why we have never seen any direct evidence for cold streams in our immediate neighbourhood. There are others. The signature of the cold accretion process would be ultraviolet rays known as Lyman-alpha radiation, emitted as gas falls into galaxies and warms up. Unfortunately that radiation is almost completely absorbed by dust in the host galaxy and in Earth's atmosphere, making nearby cold flows impossible to see from Earth.

For that reason, Genzel warns against jumping on a bandwagon. "We have to be careful not to have a paradigmatic effect where everyone thinks they see the same thing," he says. "Nobody has seen a cold flow of gas yet."

Paradoxically, cold flows might be easier to see in faraway galaxies. Light from distant objects is shifted to the red end of the spectrum because of the universe's expansion, meaning originally ultraviolet Lyman-alpha emissions from galaxies in the early universe can be shifted to visible wavelengths that penetrate through to Earth.

Abraham Loeb and Mark Dijkstra of the Harvard-Smithsonian Center for Astrophysics think we have already seen such a thing. They point to a discovery made in 1998 by Chuck Steidel at the California Institute of Technology in Pasadena and his colleagues. These researchers found two huge blobs of gas emitting Lyman-alpha radiation at a time when the universe was just 2 billion years old. The gas spanned two young galaxies, all the way from their bright internal discs and out through their dark matter haloes.

Similar blobs have since been seen in many other early galaxies. What they are, or even if the gas in them is moving into or out of the galaxies, remains something of a mystery. Loeb and Dijkstra think they are filaments of the cosmic web being drawn into galaxies. "Wherever you make a massive galaxy you should see these blobs," predicts Loeb.

Steidel counters that some blobs are known to be clouds of gas around quasars - point-like energetic objects that emit X-rays and ultraviolet light - and that the quasars may be the source of the gas, rather than the other way around. He remains to be persuaded that the blobs are filaments of the cosmic web. "It's a romantic notion but I haven't seen anything that's convinced me it's true," he says.

Late starters

Close-up images of young, star-forming galaxies to compare with supercomputer simulations would put the cosmic web's new role on a firmer footing, and these could be on the way. John Salzer of Indiana University at Bloomington and colleagues have recently found 15 nearby galaxies that have masses comparable to the Milky Way's yet appear to have started forming stars only 3 to 4 billion years ago (The Astrophysical Journal, vol 695, p L67). "This tells you that the standard idea that all galaxies started forming stars early in the universe is wrong," says Salzer.

If all galaxies were formed by mergers, then there should be evidence of violent collisions in these galaxies. If, on the other hand, galaxy formation can be driven by cold streams, the galaxies should look like discs and seem to be tapping a supply of primordial hydrogen gas - a filament of the cosmic web. In that case, Salzer speculates, these galaxies began to form in a void where filaments were relatively scarce, meaning they only took shape late.

As yet, Salzer's images don't have the resolution needed to tease out the internal dynamics of the galaxies. But now we know they are there, observations from a high-powered source such as the Hubble Space Telescope could give us a rare opportunity to see galaxy formation in action close by.

For most astronomers, it's already clear that the overall picture of galaxy formation needs revision. "In the last 4 or 5 years the picture has changed significantly," says Kereš. "Now people are seeing all galaxies as sitting in a filamentary structure that feeds them." Bolstered by cosmic simulations, the cosmic web is real and vital, and here to stay.

View a gallery of the different types of galaxies in the universe