Galaxies Are Making Fewer Stars, but They Still Have Their Atomic Hydrogen

The universe is building stars at well under half the rate it managed 4.5 billion years ago, and the obvious suspect has long been the supply: run a galaxy short of gas and it stops making stars. The trouble is that every time astronomers have gone out and weighed the gas, it has refused to disappear. Earlier surveys reached a few thousand galaxies at a time, leaving more room for any missing hydrogen to hide in the noise.

A measurement published online Sept. 1 in Nature Astronomy closes much of that room. Chuan-Peng Zhang of the National Astronomical Observatories in Beijing, Hong Guo of the Shanghai Astronomical Observatory and colleagues combined radio spectra from China's Five-hundred-meter Aperture Spherical Telescope with distance measurements from the Dark Energy Spectroscopic Instrument. Their sample includes about 2.5 million galaxies across roughly a third of the sky.

What they find is a reservoir that has hardly moved. The density of atomic hydrogen in the universe falls by a factor of 1.35 ± 0.10 over that span in the raw stack, and by only 1.12 ± 0.10 after applying the conservative systematic corrections from the team's forward model. Those two figures are not the ends of a range. The first is the raw measurement, while the second incorporates the forward model's treatment of systematic effects. The corrected result still indicates a much weaker evolution in atomic hydrogen than the decline in the cosmic star-formation rate.

Set that against the stars. Over the same span the rate at which the universe forms stars fell by a factor of about 2.46, based on the established cosmic star-formation history described by Piero Madau and Mark Dickinson rather than a measurement made directly by FAST and DESI. Star formation went down steeply. The atomic hydrogen reservoir did not follow it.

One obvious objection is bookkeeping: perhaps the mix of galaxies being counted shifts with distance in a way that makes a real decline appear flatter. The team's answer is that, at a fixed stellar mass, the average atomic gas fraction changes by less than 0.2 dex, or roughly a factor of 1.6, across the interval. The relative stability therefore belongs to the galaxy population, not merely to the composition of the sample.

Nobody sees these galaxies one at a time

Most of those millions of galaxies are not individually detectable in hydrogen at these distances. Stacking is what makes the measurement possible. DESI supplies a precise distance for each galaxy, the FAST spectra are shifted so that the 21-centimeter hydrogen line from each of them lands in the same place, and then they are added together: the noise averages away and the signal does not. The catch is FAST's beam, about 2.9 arcminutes wide, roughly a tenth of the width of the full Moon, and wide enough to take in neighboring galaxies along with the target.

The team first addressed source confusion by excluding galaxies with close companions, using angular separation and velocity difference as the criteria. That cut removed 33.4% of the sample. Even so, residual confusion remained a major source of systematic uncertainty and was addressed through the team's forward model. This treatment contributes to the difference between the raw and systematically corrected results. The direction of the effect is physically intuitive: uncounted neighboring galaxies add signal, and a fixed angular beam encompasses a larger physical region at greater distances, so an uncorrected measurement can overestimate the atomic hydrogen reservoir at the earlier end of the interval.

The two soft spots, and which way each one pushes

The isolation cut carries an asymmetry of its own. Because it is fixed in angle, it excludes companions over a larger physical radius at the far end of the survey than at the near end. That difference can alter the environmental mix of the distant and nearby samples and, depending on the relationship between environment and gas content, could make the measured evolution appear steeper than it is.

A further methodological step comes next. FAST and DESI measure average gas fractions across the galaxy sample; converting those measurements into a cosmic density requires a model for the abundance of galaxies at different stellar masses. The team integrates a modeled stellar-mass function from UniverseMachine into a mass range extending below the galaxies directly accessible to an optically selected survey at the far end of this sample. Because lower-mass galaxies can make an important contribution to the atomic gas budget, that extrapolation is part of the inference. The reservoir reported here is inferred rather than directly weighed. That is standard practice for this kind of measurement and is worth keeping in mind when interpreting the result as a cosmic inventory.

Atomic hydrogen is not what stars are made from

Stars form primarily from colder, denser molecular gas rather than directly from atomic hydrogen. Atomic hydrogen is part of the broader gas cycle that can supply the molecular phase, and the authors note that the molecular reservoir “is known to evolve more closely with star formation.” What the measurement rules out, in their words, is “rapid depletion of galaxy H I as the primary driver” of the star-formation decline. A decline in the molecular gas available for star formation remains a possible explanation.

The direction of the result is not entirely new. Earlier stacking experiments using the Giant Metrewave Radio Telescope, the VLA and ASKAP have also reported relatively weak evolution in the atomic hydrogen reservoir, although the samples and redshift ranges differ. What this survey adds is scale and statistical precision. The figure data and stacked H I spectra are public through Zenodo, but the DESI DR2 redshift catalog used in the study is not yet public, limiting independent reproduction of the full analysis.

That leaves a shortlist the authors decline to narrow: how much fresh gas flows into galaxies, how efficiently atomic gas converts to molecular gas and how star formation regulates itself. The paper describes the roles of the atomic and molecular gas reservoirs as uncertain and presents the measurement as a stringent benchmark for models rather than a verdict on them. Models predicting a steep decline in atomic hydrogen over the past few billion years can now be tested against a substantially tighter observational constraint.