The measurement could hardly have been made with ordinary instruments. The 21-centimeter radio emission line of neutral atomic hydrogen, written H I by astronomers, is intrinsically faint, and individual galaxies beyond the local universe are simply too dim to detect directly in this line. The team therefore combined two of the most powerful survey machines ever built. FAST, the largest single-dish radio telescope on Earth, provided exquisitely sensitive H I spectra drawn largely from the FAST All Sky H I Survey, or FASHI. DESI, mounted on the Nicholas U. Mayall Telescope at Kitt Peak in Arizona, supplied optical spectroscopy for approximately 2.5 million galaxies covering roughly 12,000 square degrees of sky. By cross-matching the two datasets and statistically stacking the faint radio signals of thousands of galaxies at once, the researchers could recover an average H I signal even for galaxies whose individual emission would be buried in the noise forever.

The technique, known as spectral stacking, works on a simple principle: while random noise averages down as observations are added, a real signal consistently aligns with each galaxy’s known redshift from the DESI catalogue and therefore builds up coherently. Stacking galaxies in bins of redshift and stellar mass allowed the team to measure the average atomic gas fraction as a function of stellar mass out to redshift 0.41, which corresponds to a lookback time of about 4.5 billion years. Converting these stacked signals into a cosmic H I density, denoted Omega_HI, required careful calibration of every systematic effect that could masquerade as cosmic evolution, from radio frequency interference and baseline ripples in the FAST spectra to source confusion, luminosity bias and the detailed selection function of the DESI Bright Galaxy Survey.

The headline result is a raw decline in Omega_HI by a factor of 1.35 plus or minus 0.10 over the past 4.5 billion years. After the team applied conservative systematic corrections through their forward model, the inferred decline shrank further, to just a factor of 1.12 plus or minus 0.10. That number may sound technical, but its implications are profound. Over the same interval, the cosmic star formation rate density has collapsed by a factor of roughly 2.46, more than twice as steep a drop. In other words, while star formation across the universe has been falling off a cliff, the raw atomic hydrogen fuel that feeds star formation has barely diminished at all.

This mismatch strikes at the heart of a decades-old debate. In the standard picture of galaxy evolution, stars form from cold molecular gas, but that molecular gas must be assembled from the more diffuse atomic hydrogen that dominates the cold gas reservoir of most galaxies, particularly in their extended outer discs. If galaxies had been rapidly draining their atomic reservoirs over cosmic time, that depletion could plausibly explain the dying star formation rate. The new measurement rules out this scenario. Atomic hydrogen, it turns out, has been remarkably stable, evolving at most modestly while the star formation it supposedly fuels has collapsed. The study instead finds, consistent with prior work, that the molecular gas density evolves far more closely in step with star formation, pointing the finger at the conversion of atomic gas into molecular gas, and at the regulation of star formation within molecular clouds, as the real bottleneck.

The weak evolution is not confined to a particular class of galaxy. At fixed stellar mass, the team found that the average H I gas fraction of galaxies evolves by less than 0.2 dex, an astronomer’s shorthand for a change of less than about 60 percent, across the full 4.5-billion-year window. This means the flatness of the cosmic H I density is not an artifact of the galaxy population shifting its composition over time, for example through the growth of massive, gas-poor galaxies. Rather, the stability of atomic hydrogen reservoirs appears to be a property of the galaxy population as a whole, from moderate spirals to the most massive systems in the DESI sample. The result aligns with, and substantially tightens, earlier stacking measurements from instruments such as the Giant Metrewave Radio Telescope, the Arecibo Observatory, MeerKAT and the Very Large Array, which individually traced smaller redshift ranges with larger uncertainties.

Reaching this precision demanded an unusually rigorous treatment of systematic uncertainties. The authors built a forward model that simulates how observational effects distort the true underlying signal, testing their measurements against mock catalogues built from cosmological simulations such as IllustrisTNG and SIMBA. They quantified how spectral confusion, in which multiple faint galaxies blend into a single radio feature, inflates the apparent gas fraction at the low-mass end, and corrected for the luminosity bias that arises because brighter galaxies are preferentially detected in the optical parent sample. They also addressed the frequency-dependent environment of FAST, masking regions of strong radio frequency interference and verifying that stacked noise scaled with the expected inverse square root of the number of stacked spectra. The consistency of three independent correction methods for luminosity bias gave the team confidence that the measured trend reflects the universe rather than the instrument.

The findings provide a stringent benchmark for theoretical models of how galaxies acquire and process gas. Simulations must now reproduce a universe in which the atomic hydrogen density varies by only tens of percent over nearly half the age of the Earth, even as star formation plummets. That constraint favors scenarios in which galaxies continuously replenish their atomic reservoirs through accretion of fresh gas from the cosmic web and the circumgalactic medium, while the conversion of atomic gas into molecules and the feedback processes that heat or expel gas become progressively less efficient. Models that allow atomic hydrogen to be rapidly depleted, or that tie star formation directly to the atomic gas supply, will need substantial revision. The result also connects to broader questions about the baryon cycle, the circulation of ordinary matter between galaxies and their surrounding halos that underpins all of galaxy evolution.

There is more to come. FAST continues to expand the FASHI catalogue toward its ultimate goal of a complete all-sky H I census, and DESI is amassing ever larger spectroscopic samples as its five-year survey proceeds. Together, these surveys promise to push H I stacking measurements to higher redshifts, potentially bridging toward the epoch around redshift one where earlier measurements hinted at more substantial evolution, when the universe was roughly half its current age. For now, the message of this study is clear and slightly humbling: the universe’s dimming is not for lack of fuel. Cold atomic hydrogen, the raw material from which galaxies knit their stars, has been there all along, patiently waiting, while the machinery that turns gas into stars has been quietly grinding to a halt.

Subject of Research: Weak evolution of the cosmic atomic hydrogen density over the past 4.5 billion years measured with FAST and DESI.

Article Title: Weak evolution of cosmic atomic hydrogen over the past 4.5 billion years

Article References: Weak evolution of cosmic atomic hydrogen over the past 4.5 billion years. (n.d.). https://doi.org/10.1038/s41550-026-02965-9

Image Credits: AI Generated

DOI: 10.1038/s41550-026-02965-9

Keywords: cosmic atomic hydrogen, H I gas density, FAST telescope, DESI survey, spectral stacking, star formation decline, galaxy evolution, gas accretion, molecular gas, neutral hydrogen, redshift surveys, cosmology