What was the universe like before there were stars?

About 380,000 years after the beginning, the universe cooled enough for electrons and protons to combine into neutral hydrogen. Light stopped scattering and began travelling freely. That light is still arriving, stretched into microwaves, and it is the oldest thing anyone can see.

It is also the last thing to see for a very long time. After it, the universe was a transparent, cooling expanse of neutral gas, mostly hydrogen, with no sources of light in it at all.

The first stars formed from that gas, and ultraviolet light from them and from the stars and quasars that came after stripped electrons back off the hydrogen. That transition, reionisation, is detectable in the microwave background as a slight smearing, and is largely complete by about a billion years.

So the dark ages are bracketed. What is missing is what happened inside them.

Neutral hydrogen has a faint transition at 21 centimetres. Ultraviolet light from the first stars changes which of its two hyperfine states the gas sits in, and that makes it absorb photons from the microwave background at radio frequencies that stretch with redshift, leaving a dip in the all-sky spectrum.

In 2018 the EDGES experiment reported exactly that: an absorption profile centred at 78 megahertz. The dip was more than twice as deep as the largest standard predictions, which would have meant either colder gas than expected or a hotter radio background.

In 2022 a second experiment, SARAS 3, a radiometer floated on lakes in southern India, looked in the same band and did not find it. It rejects the EDGES best-fitting profile with 95.3 per cent confidence.

The signal being hunted is smaller than the foreground our own galaxy produces by at least four orders of magnitude. Extracting it means modelling that foreground almost perfectly, and reasonable choices about how to do that can create or erase a feature of the size in question: a third group refitted the EDGES data and found that other simple signals, of different shapes, fit it too.

This is what a frontier measurement looks like from inside. The objections in print, from that refit and from SARAS 3, are to the modelling and to how much the data can bear, not to anyone’s good faith.

Simulations put the first generation at redshifts of about 30 and below, which is somewhere after a hundred million years, and have them dominating star formation until redshift 15 or so. They depend on assumptions about how primordial gas cools without the heavy elements that make cooling easy today.

No first-generation star has been confirmed. One candidate, LAP1-B, a lensed system the James Webb Space Telescope found at redshift 6.6, long after the dark ages, is argued by one group to match three key predictions for them. A confirmed one would settle several of these questions at once.

The expectation is that their masses ran heavier than stars forming today. The older reason given for it, that metal-free gas resists breaking up, has not survived the simulations: primordial gas fragments readily, and the first stars are now expected in small multiple systems. What survives is the shape of the distribution, which comes out roughly flat in the logarithm of mass and so top-heavy beside the present-day one, over a range running from below a solar mass to several hundred.

The review that assembles those models calls the understanding behind them still quite limited. The reported mass spectra vary enormously, and how much of that is physics and how much is the resolution a simulation could afford is not settled. Meanwhile the star formation the James Webb Space Telescope finds around redshift 12, now confirmed with spectra, sits above what a constant efficiency predicts, and a top-heavy early mass function is one of the things offered to explain it.

The 21-centimetre signal remains the only practical route, and the work is now about foregrounds and instruments rather than about theory. Arrays rather than single antennas, and for the dark ages proper a telescope off the Earth altogether: one review concludes that only a space-based or lunar instrument, shielded from terrestrial transmissions and from our own ionosphere, can reach them.

If that works, it would be the first direct observation of an era rather than an inference from its edges. It would also be a great deal more of the early universe than anything yet measured: the microwave background carries about a million independent modes of the primordial fluctuation field, galaxy surveys about a hundred million, and the dark ages hold up to a million million.