Missing Baryons: How FRBs Weigh the Cosmic Web

When a galaxy’s matter goes missing, is the error in the Universe or in our inventory?

Welcome to FreeAstroScience. We read both baryon-census papers against each other and kept only the numbers that carry the argument, uncertainties included.

About 76 percent of the Universe’s ordinary matter sits outside galaxies, in gas so thin that a cubic centimeter of it holds fewer than a thousandth of a particle. That share comes from 69 localized fast radio bursts, and it closes an accounting gap that stayed open for roughly two decades. Nothing was ever lost. Diffuse plasma simply does not shine.

Ordinary matter, not the dark kind

Baryons are protons and neutrons, the matter that builds stars and the iron in your blood. Dark matter is something else: a component detected through its gravity, whose particle identity nobody has pinned down. Conflating the two is the commonest error in coverage of this result, and any headline announcing that researchers finally located the missing dark matter has it backwards.

Early-Universe physics pins the baryon budget tightly, and two independent handles agree on it: the cosmic microwave background, and the abundances of light elements forged in the first minutes. Counts of stars and of the gas inside nearby galaxies and clusters recovered far less than that. Astronomers named the shortfall the missing baryon problem, which was always an unfair label for what was really a bookkeeping failure.

Table 1 — what is actually unknown in each case (Macquart et al. 2020; Connor et al. 2025)

Electrons are not baryons. In ionized gas they travel alongside the protons in equal numbers, which turns them into a stand-in: count the free electrons along a path, and you have counted the ordinary matter on it. Every measurement below rests on that substitution, and it repays holding onto, because it is also where the method’s assumptions live.

For the opposite side of the confusion, our report on the third galaxy found without a dark matter halo shows what a genuine dark-matter result looks like, and how differently it has to be measured.

Why thin gas stayed off the books

Spotting a star is easy, because a star is dense and hot. Detecting a medium at fewer than 10⁻³ particles per cubic centimeter is not.

Each older technique samples one physical phase and misses the rest. Hydrogen absorption picks out the small neutral fraction along narrow sightlines. X-ray emission climbs steeply with density, so it lights up around groups and clusters and fades in the filaments between them. The thermal Sunyaev-Zel’dovich effect traces pressure and has found filament gas, though on its own it does not deliver an all-temperature census. Every one of those methods detects real gas. None of them was ever going to add up to the whole.

Faint gas is not absent gas. When most baryons sit in low-density plasma, any method that depends sharply on density or temperature will undercount them however large the total mass turns out to be.

Try holding that density in your hand. One gram of hydrogen is about 6 × 10²³ protons, and at 10⁻³ per cubic centimeter those protons occupy roughly 6 × 10²⁰ cubic meters. Take the cube root and you get a box about 8,450 km on a side, two thirds of Earth’s diameter. An Earth-sized volume of intergalactic gas would weigh around three grams, and the cosmic web still holds three quarters of all ordinary matter, because the amount of space involved is beyond anything intuition handles.

Geography does the rest. Galaxies and clusters sit at the knots of the cosmic web, with filaments running between them. Ionized gas fills most of the room around both. One burst’s sightline can cross dozens of filaments, so a single measurement stacks many faint environments into one number. Stacking like that is a strength and a limitation at the same time. It buys enormous reach along the line of sight, and gives nothing at all across it.

Dispersion turns a radio flash into a scale

A fast radio burst leaves its source as a sharp pulse spread across a wide band of radio frequencies. Free electrons along the way hold back the lower frequencies more than the higher ones, so the pulse arrives smeared in time, with the low end trailing. That smear is the dispersion measure, written DM and quoted in parsecs per cubic centimeter, and it counts the electrons in a one-square-centimeter column running the whole way from source to telescope.

Nothing in that chain asks the gas to emit anything at all. DM barely cares whether the plasma is warm or blisteringly hot, which is exactly the property absorption lines and X-ray emission lack.

What produces the bursts is a separate question, and a live one. Our guide to the origins of fast radio bursts covers the magnetar and binary-merger candidates. When we published that guide in November 2024, the baryon census still rested on five objects. Sixty-nine carry it now, while the engine question has barely moved. For this measurement the engine hardly matters. All that counts is that the pulse left its source sharp.

Scale helps here. For a burst with DM near 500 pc cm⁻³, the pulse recorded at 1.2 GHz lands roughly 0.52 seconds behind the same pulse at 1.5 GHz. Half a second, from material too faint to photograph.

Telescopes never receive the cosmic term on its own. What arrives is one summed quantity with four contributions inside it: the Milky Way’s disk, the Milky Way’s halo, the intergalactic path, and plasma in the host galaxy where the burst went off. Prising out the intergalactic part means modeling the other three, and that step is where most of the argument in both papers actually happens.

From burst to baryon

1. Catch the pulse.
2. Localize it to a single host galaxy rather than a patch of sky, using interferometry.
3. Measure that host’s redshift with optical spectroscopy, which converts a sky position into a cosmic path length.
4. Model and subtract the Galactic disk, the Galactic halo, the host, and any intervening halos.
5. Repeat dozens of times, then fit how DM grows with redshift.

Why one burst is never enough

Any large DM has three innocent explanations, and the burst itself will not tell you which: a long path, a dirty host, or a foreground crowded with halos. Localization plus redshift kills the first ambiguity outright. Everything after that is handled by sample size, which is why this field’s story reads as a history of counting. Five bursts gave a wide answer, 69 gave a narrow one, and neither sample changed the underlying physics by a single step.

Did the 2020 census actually find them?

Yes, inside uncertainties wide enough to be honest about. J.-P. Macquart and colleagues worked from five arcsecond-localized ASKAP bursts and found the mean cosmic DM climbing with redshift at the rate diffuse ionized gas predicts. Their baryon density came out at Ω_b = 0.051, with a 95 percent interval of plus 0.021 and minus 0.025 after scaling to h₇₀⁻¹.

Read those error bars before reading the headline. A range running from 0.026 to 0.072 spans nearly a factor of three, and it is lopsided, which is roughly what a five-object sample buys you. Their result still mattered, because it agreed with the cosmic microwave background and with Big Bang nucleosynthesis while using a completely independent late-Universe method.

Reach mattered as much as precision. FRB 20220610A, whose light spent eight billion years in transit, proved the technique survives at cosmological distance. Somewhere between that detection and the 69-burst sample, a promising idea turned into a measurement.

The 2025 partition

Liam Connor and colleagues pushed the sample to 69 localized bursts, 39 of them from the Deep Synoptic Array at Owens Valley and 30 already in the literature, then asked a harder question than 2020 had asked. Where does the ordinary matter actually sit?

Their answer was Ω_bh₇₀ = 0.051 ± 0.006. That is the same central value as five years earlier, and roughly four times tighter.

Table 2 — where cosmic baryons reside today, rounded (Connor et al. 2025)

Those three numbers get reproduced everywhere as a clean pie chart, and they should not be. The intergalactic share carries plus 10 and minus 11 percentage points, the halo figure about 10, and the stellar slice is simply what remains once the other two are subtracted. Push the halo estimate to the top of its range and the intergalactic figure slides toward 66 percent. What survives every reshuffle is the ranking, which is the part worth quoting: most ordinary matter lies between galaxies rather than inside them, and by a wide margin.

One thing we could not settle from the published figures is how much of that 15 percent halo share is data and how much is prior. Splitting a single measured column into bound and unbound pieces leans on simulations and on halo gas prescriptions, and neither paper hands you a clean way to see how far the answer would travel under a different prescription.

Feedback is an interpretation, not a measurement

Gas that once sat inside galactic halos has been driven out into the wider intergalactic medium. Supernovae and the outflows launched near feeding supermassive black holes can heat gas or eject it before it ever cools into stars. Had baryons tracked dark matter closely, far more of them would still be locked inside collapsed halos than the 2025 partition allows.

Be careful with the logic here. Fast radio bursts measure electron columns. Feedback is the physical story that best fits the resulting distribution and the behavior of cosmological simulations, which makes it a strong inference rather than a sighting. Nobody watched the gas leave.

Consequences reach into precision cosmology. Gas expelled from halos changes how ordinary matter clumps on small and intermediate scales, suppressing the matter power spectrum that weak lensing surveys measure. Anyone comparing late-Universe structure against cosmic microwave background predictions has to model that suppression, and the 2025 partition is now one of very few direct constraints on its size. That same late-versus-early comparison drives the expansion-rate dispute behind our piece on NGC 5584 and the Hubble tension.

Where the method still hurts

The host galaxy is a guess

A burst going off inside a dense star-forming region picks up local DM that has nothing to do with the cosmic web. Connor’s team inferred a median host contribution near 130 pc cm⁻³ and treated the population statistically instead of assigning every host the same value. For any single burst that estimate could be badly wrong. Across 69 of them, the error is meant to average away.

Our own Galaxy sits in every measurement

Every pulse passes through the Milky Way first. Electron-density models handle the disk reasonably well. Our hot Galactic halo is harder, and a systematic error there would shove every cosmic DM estimate in the same direction at once, which is precisely the kind of error a large sample cannot fix.

Detection is not uniform

Radio arrays and the optical follow-up behind them do not find every burst with equal probability. Faint hosts drop out. Until selection functions are characterized properly, the observed sample is not guaranteed to represent the underlying population, and the partition inherits whatever bias remains.

We are leaving out the X-ray absorption work and the Sunyaev-Zel’dovich stacking results that reached comparable conclusions by other routes. Those deserve an article of their own, and folding them in here would blur the single question this piece exists to answer, which is what a dispersion measure can and cannot establish on its own.

What comes after a census?

Attention is shifting from one global number toward mapping where the gas sits relative to galaxies and to the lensing mass around them. Doing that needs two things at once: precisely localized bursts with measured redshifts, and very large samples with coarser positions.

An April 2026 preprint pushed hard on the second requirement. Kritti Sharma and colleagues took dispersion measures from 3,455 unique CHIME sources and cross-correlated them against ten independent tracers of large-scale structure, reporting correlations between 2.6 and 5 standard deviations, with higher DM along sightlines threading denser regions. As of late July 2026 that work has not cleared journal review, so treat its detailed numbers as provisional.

Five bursts answered whether the total was right, and 69 showed how it divides. The next question is stricter than either, because it asks whether the baryons sit where our models put them.

• The total was never wrong.
• Baryons are ordinary matter, and dark matter remains a separate problem entirely.
• Roughly three quarters of that ordinary matter sits in intergalactic plasma, with error bars near ten percentage points either way.
• Dispersion measures a column and not a location, so the partition rests on models as much as on data.

Updated 28 July 2026: restructured throughout. The dispersion section now carries the full path from burst to baryon, the gas-density arithmetic is shown rather than asserted, the width of the 2020 error bars is spelled out, and the accordion FAQ has been removed because the body already answered every question in it.

The faint majority

Visibility and abundance are different properties, and the Universe was never obliged to make them line up. Everything we can photograph amounts to the bright minority of ordinary matter. Whatever remains is a thin ionized fog stretched through the filaments, weighed by radio flashes lasting a thousandth of a second.

We wrote this at FreeAstroScience for the instrument as much as for the answer. A technique that measures what it cannot see, and states plainly how much of its result came from a model, deserves understanding on its own terms. Keep asking which part of any number came from the data.

Gerd Dani

Sources

1. Macquart, J.-P., Prochaska, J. X., McQuinn, M., and collaborators. A census of baryons in the Universe from localized fast radio bursts. Nature, volume 581, pages 391–395, 2020.
2. Connor, L., Ravi, V., Sharma, K., and collaborators. A gas-rich cosmic web revealed by the partitioning of the missing baryons. Nature Astronomy, volume 9, pages 1226–1239, 2025.
3. McQuinn, M. Locating the missing baryons with extragalactic dispersion measure estimates. The Astrophysical Journal Letters, volume 780, article L33, 2014.
4. Planck Collaboration. Planck 2018 results. VI. Cosmological parameters. Astronomy and Astrophysics, volume 641, article A6, 2020.
5. Sharma, K., Krause, E., Ravi, V., and collaborators. Backlighting the cosmic web with fast radio bursts. arXiv preprint, 2026.