Large-Scale Structure and the Acoustic Ruler

Summary

Galaxies are not scattered at random.  They lie in sheets and filaments wrapped around enormous empty voids.  That pattern was a surprise when it was first mapped in 1986, and the discoverers said so in print.¹  Since then the pattern has been measured well enough to carry a ruler.

The central result of this article is the baryon acoustic oscillation scale, and it is the cleanest prediction-then-measurement sequence in observational cosmology.

• Before the first atoms formed, sound waves ran through the hot plasma. When the plasma turned neutral the waves stopped, freezing a preferred separation into the matter. That separation is the sound horizon.
• Its length can be computed from the density of ordinary matter and the density of all matter, both of which the microwave background measures. The answer is 147.05 Mpc, fixed to 0.2 percent, using no galaxy-survey data at all.² About 480 million light years.
• The number was computed first and looked for afterward. In 2005 two teams found it independently, in different data, on different telescopes, with different statistics. Daniel Eisenstein's group found a bump in the clustering of 46,748 luminous red galaxies "at 100h⁻¹ Mpc separation that is an excellent match to the predicted shape and location."³ Shaun Cole's 2dF team found the same feature in the power spectrum, published back to back with it.⁴
• Twenty years later DESI has measured the same ruler in seven tracer samples spanning redshift 0.3 to 2.3, from more than 14 million galaxies and quasars, at sub-percent precision.² Same length. No adjustment.
• This is a hard test, and it passed. Nothing about galaxy formation, dust, star colours or telescope calibration sets the length of a sound wave in a plasma at 380,000 years. Had the bump appeared at 80 Mpc or 200 Mpc, the model would have been dead.

The rest of the structure story is strong, with one honest dependency and one genuine open problem.

• The seeds are visible. The microwave background's temperature ripples are a few parts in 100,000, and gravity turns ripples that small into galaxies given billions of years and a matter density that dominates radiation for most of that time.
• The dependency is dark matter, and it is load-bearing rather than decorative. Ordinary matter cannot start growing until it stops being dragged by radiation, which happens at 380,000 years. Growth after that is too slow, by a large factor, to reach the clumpiness now observed. Cold dark matter is assumed to have been growing already, since it never felt the radiation pressure. That is an assumption imported into the calculation, not a thing anyone has weighed. The particle has never been detected — see Dark Matter.
• Redshift-space distortions test the growth rate rather than the geometry, and they agree. The completed twenty-year SDSS programme reports σ₈ = 0.85 ± 0.03 from clustering and distortions, "consistent with predictions from Planck temperature and polarization data and with General Relativity."⁵
• The S8 tension — weak-lensing surveys finding slightly less clumpy structure than the microwave background predicts — has largely dissolved. It went from about 3σ in 2021 to 0.73σ in the final Kilo-Degree Survey analysis of 2025, and that collaboration blames its own earlier redshift calibration, not new physics.⁶ Apologetics that still cites S8 as a crisis is citing a dead number.

The JWST "impossible early galaxies" claim does not survive contact with the follow-up spectroscopy, and creationist writing that still leans on it is leaning on a preprint that its own authors replaced.

• The July 2022 preprint by Ivo Labbé and colleagues was titled "A very early onset of massive galaxy formation" and reported "seven galaxies with M∗ > 10¹⁰ M⊙ and 7 < z < 11 … including two galaxies with M∗ ∼ 10¹¹."⁷
• The version published in Nature in 2023 is titled "A population of red candidate massive galaxies ~600 Myr after the Big Bang." Seven became six. The redshift range narrowed to 7.4 ≤ z ≤ 9.1. The two 10¹¹ objects became one "possible." A forecast of galaxies out to z ∼ 18 was deleted.⁷
• The paper says of itself: "We note that these masses and redshifts are not definitive and that all galaxies should be considered candidates."⁷
• Every one of the six that has since been observed with a spectrograph came in lower — lower redshift, lower mass, or both. One of them, L23 ID 13050, is a broad-line active galactic nucleus at z = 5.624, not a galaxy at z = 8.13.⁸
• Michael Boylan-Kolchin, whose 2023 paper is the source of the "challenge to ΛCDM" framing, stated the bar conditionally in that same paper: "If none of these explanations holds up and these massive galaxies are spectroscopically confirmed, they will pose a serious challenge."⁹ His threshold was a baryon-to-star conversion efficiency of 0.57. With mid-infrared photometry included, the measured value for the most massive z ∼ 8 galaxies is about 0.3.¹⁰
• Something real does remain, and it should not be buried by the walk-back: a quiescent galaxy at z = 7.3 whose number density exceeds every model by more than a hundredfold, and an abundance of ultraviolet-bright galaxies at z = 10 to 14 that models did not anticipate. These are disagreements about how fast stars form, inside an agreed framework, not evidence of a young universe.

What this article does not claim.  The acoustic ruler measures a geometry.  It does not by itself date anything, and every young-earth cosmology that has survived peer review inside creationist journals reproduces standard cosmological observations by construction — see Young-Universe Cosmologies.  What the growth of structure does constrain is time:  the amplification of a 10⁻⁵ ripple into a galaxy cluster is a slow process, and no published young-earth model performs it in thousands of years.  That is stated here plainly because it is true.

What a redshift survey actually measures

A galaxy map is not a photograph.  Understanding what the surveys did is the first step in judging what they show, and most of the disagreement about large-scale structure comes from skipping it.

A telescope gives two angles on the sky and a spectrum.  The spectrum gives a redshift.  Turning the redshift into a distance takes four separate steps, and each one can go wrong.

Step one: get the redshift at all.  This means dispersing the light of a faint galaxy until its emission and absorption lines can be identified.  For a magnitude-19 galaxy on a 4-metre telescope that is a serious exposure.  The whole subject is limited by how many spectra can be taken at once, which is why the history below is a history of instruments as much as of ideas.

Step two: convert redshift to distance.  Doing this requires a cosmological model, because the relation between redshift and distance depends on the expansion history.  A survey therefore starts by assuming a fiducial cosmology, builds the map in those coordinates, and then measures how much the assumed cosmology has to be stretched to make the data fit.  This sounds circular and is not, because the stretch is measured along and across the line of sight separately.  If the assumed cosmology is wrong, a spherical feature comes out elliptical.  That asymmetry is a measurement, first proposed by Charles Alcock and Bohdan Paczyński in 1979.¹¹

Step three: correct for the fact that redshift is not distance.  A galaxy has a motion of its own on top of the expansion, typically a few hundred kilometres per second, and that motion adds to its redshift.  Nick Kaiser worked out the consequences in 1987 and they are large.  Inside a cluster, random motions of a thousand kilometres per second smear the cluster along the line of sight into a long spike pointing at the observer — the "fingers of God."  On larger scales, matter falling toward an overdense region makes the region look flattened along the line of sight.  Kaiser's summary states the general point:  "Peculiar velocities distort the clustering pattern in redshift space on all scales."¹²

That distortion is a nuisance for mapping and a measurement in its own right, which is the subject of a later section.

Step four: account for what the survey could not see.  Every survey has a magnitude limit, so it sees bright galaxies further away than faint ones, and the apparent density falls with distance for reasons that have nothing to do with cosmology.  Fibre-fed spectrographs cannot place two fibres closer than a fixed angular separation, so close pairs are systematically under-counted.  Dust in the Milky Way dims patches of sky.  Bright stars punch holes in the catalogue.  All of this is modelled with a "random catalogue" — a synthetic sample with the same selection function and no clustering — and the measurement is the ratio between the two.

The upshot.  A published correlation function is the end of a long chain of corrections.  This is a fair thing for a skeptic to press on.  It is also why the acoustic peak matters so much:  a sharp feature at one particular separation is hard for a smooth, slowly varying selection error to fake.  Selection effects are broad.  The signal is narrow.

Mapping the web, oldest first

1986: the CfA survey finds bubbles, and nobody expected them

Valérie de Lapparent, Margaret Geller and John Huchra measured 584 new redshifts and assembled 1,100 galaxies in a strip six degrees wide and 117 degrees long, running through the Coma cluster.  What they found is worth quoting from their abstract, because the second half of it is usually left out:

The galaxies appear to be on the surfaces of bubble-like structures.  The bubbles have a typical diameter of ~ 25h⁻¹ Mpc. …  All of these features pose serious challenges for current models for the formation of large-scale structure.¹

They meant it.  Their introduction says that the standard cold-dark-matter model of the day "cannot produce voids as large as the one in Bootes,"¹ and their own preferred explanation was Jeremiah Ostriker and Lennox Cowie's explosive galaxy formation, an idea that is now abandoned.

This is the honest history and it should be told this way.  The cosmic web was not predicted in detail and then found.  It was found, and the models were then rebuilt to produce it.  What was predicted in advance is the acoustic scale buried inside it, which is a different claim and a much stronger one.

Three years later Geller and Huchra reported the "Great Wall," a sheet of galaxies running across the survey.¹³  Gott and colleagues later measured its length in comoving coordinates at 240 Mpc.¹⁴

2001-2003: 2dF makes the map statistical

The Two-degree Field Galaxy Redshift Survey used a 400-fibre spectrograph on the 3.9-metre Anglo-Australian Telescope.¹⁵  The final catalogue holds "reliable redshifts for 221 414 galaxies."⁴  That is a two-hundred-fold increase over the CfA slice in fifteen years, and it moved the subject from pictures to statistics.  With that many galaxies the clustering can be measured as a function of separation, with error bars, and compared to a calculation.

2000-2011: the Sloan Digital Sky Survey

SDSS used a dedicated 2.5-metre telescope at Apache Point with plug-plates drilled for 640 fibres per field, later 1,000.¹⁶  Its main galaxy sample and its luminous red galaxy sample are the data in which the acoustic peak was first seen.  Its map contains the "Sloan Great Wall," which Gott and colleagues measured at "1.37 billion light years long, 80% longer than the Great Wall discovered by Geller and Huchra."¹⁴

A caution about superlatives.  Claims of ever-larger structures are a genre with a poor record.  Seshadri Nadathur showed in 2013 that the algorithm used to find the "Huge Large Quasar Group" regularly finds even larger clusters in explicitly homogeneous random simulations.¹⁷  The "Giant Arc" has since had the same treatment, in a 2025 paper titled "The Giant Arc - Filament or Figment?"¹⁸  Treat any new record-breaking structure as unproven until someone has run the finder on a random catalogue.

2021 onward: DESI

The Dark Energy Spectroscopic Instrument sits on the 4-metre Mayall telescope at Kitt Peak and carries a robotic focal plane that positions 5,000 fibres per observation.²  It has "measured redshifts for over 30 million galaxies and quasars in just three years of operation," of which about 14 million enter the DR2 acoustic-oscillation analysis.²

That is the instrumental arc:  584 redshifts in 1986, 221,414 in 2003, 30 million by 2025.  The pattern found in the first is still there in the last, measured to a fraction of a percent.

The acoustic ruler

This is the centre of the article.  A length was calculated from one dataset, in advance, and then found in a completely different dataset taken with different instruments by different people.  That is the shape of a real prediction, and it is rarer in cosmology than the popular literature suggests.

What the sound waves were

Before the universe was about 380,000 years old, ordinary matter and light were locked together.  Free electrons scatter photons, so the two behaved as one fluid.  That fluid had pressure, so a lump of extra density did not simply collapse.  It pushed outward as a spherical sound wave.

When the electrons combined with nuclei into neutral atoms, the photons stopped scattering and streamed away.  The pressure vanished.  The sound wave stopped where it was, leaving a shell of extra ordinary matter around each original lump — and the dark matter, which never felt the pressure, still sitting at the centre.  Both then attracted material, so today there should be a slight excess of galaxies at the radius that shell reached.

The radius is called the sound horizon at the drag epoch, written r_d.  The same physics produces the acoustic peaks in the microwave background — that side of the story is told in The Cosmic Microwave Background, and this article does not repeat it.  What matters here is that the two are the same wave, measured at two different times in the life of the universe:  once at 380,000 years, in light, and once billions of years later, in galaxies.

The number was computed first

1970.  Jim Peebles and Jer Yu integrated the equations for a coupled photon-baryon fluid through recombination.  "Starting from an assumed initial power spectrum well before recombination, we obtain a final spectrum of density perturbations after recombination."¹⁹  Their Figures 4 through 7 show the oscillations.  Rashid Sunyaev and Yakov Zel'dovich reached the same conclusion in the same year, working independently in Moscow.²⁰

1998.  Daniel Eisenstein and Wayne Hu published fitting formulae accurate to better than ten percent that separate out every physical effect involved.  They stated the stakes plainly:  "baryonic features in the power spectrum are a fundamental prediction of the gravitational instability paradigm, and their discovery would represent a strong consistency test for the cosmological model."²¹

2003.  Two papers, months apart, proposed using the feature as a ruler.  Chris Blake and Karl Glazebrook:  "the baryonic oscillations expected in the galaxy power spectrum may be used as a 'standard cosmological ruler' … which is fixed by fundamental linear physics in the early Universe and hence is highly model-independent."²²  Hee-Jong Seo and Eisenstein:  "The cosmic microwave background provides the scale of the oscillations as a standard ruler that can be measured in the clustering of galaxies."²³

How the length is fixed.  It depends on two densities and almost nothing else.  Eisenstein's own statement of the scaling is that for models near the concordance cosmology the sound horizon "scales as (Ω_mh²)^−0.25 (Ω_bh²)^−0.08," and, critically, "given Ω_mh² and Ω_bh², the two physical scales are known in absolute units, with no factors of H₀⁻¹."³  Both densities are measured by the microwave background from the heights of its acoustic peaks, which is a different measurement from the peak spacing used to get distances.

The modern number, from DESI's own statement of the method:

the CMB provides tight constraints on ω_b and ω_bc, leading to a 0.2% determination of r_d …  Second, the same physics imprints both the BAO and the acoustic peaks in the CMB power spectrum.²

r_d = 147.05 Mpc, to 0.2 percent, with no galaxy data used.  In the units the 2005 papers used, that is close to 100 h⁻¹ Mpc.

The 2005 measurements

Two teams published within days of each other, using different data and different statistics.

Eisenstein and colleagues, SDSS luminous red galaxies.  They measured the correlation function — the excess probability of finding two galaxies at a given separation — from 46,748 galaxies over 3,816 square degrees:

We find a well-detected peak in the correlation function at 100h⁻¹ Mpc separation that is an excellent match to the predicted shape and location of the imprint of the recombination-epoch acoustic oscillations on the low-redshift clustering of matter.  This detection demonstrates the linear growth of structure by gravitational instability between z ≈ 1000 and the present and confirms a firm prediction of the standard cosmological theory.³

A useful technical detail:  the whole harmonic series of peaks that appears in the power spectrum collapses to a single bump in the correlation function.  So the correlation-function measurement is looking for one feature at one place, which is why a 3.4σ detection is meaningful at all.

And the honest part.  The significance was 3.4σ.  Not five.  The paper says so:  "The baryon signature is therefore detected at 3.4 σ."³  The same section states the independence that makes the result a test rather than a fit:  "no information from the CMB on Ω_mh² has been used in computing χ², and so our constraint on Ω_mh² is separate from that from the CMB."³

Cole and colleagues, the 2dF Galaxy Redshift Survey.  They measured the power spectrum from 221,414 galaxies and found the same physics as wiggles rather than a bump.  Their statement about their own significance is more careful than the headline, and deserves quoting in full because it is the model of how to report a result like this:

Our figure of Ω_b/Ω_m = 0.185 ± 0.046 appears at face value to be a 4-σ detection of baryon features, although this overstates the significance. …  It therefore seems fair to reject the zero-baryon hypothesis at about the 1% level.

It should be emphasised that the above statements depend on the theoretical framework of the ΛCDM model. …  What is impressive in our data is not simply that the results suggest departures from a smooth curve, but that these deviations occur in the locations expected from theory.  It is this prior knowledge that gives the extra statistical power needed in order to reject a zero-baryon model with confidence.⁴

That cuts both ways and this article prints it for that reason.  The signal is significant because the location was predicted first.  A skeptic is entitled to say that a 3.4σ bump found where you were told to look is weaker evidence than a 3.4σ bump found by accident.  The reply is that the location was published years earlier by people who had no galaxy data, and that the prediction has since been tested at far higher precision.

Twenty years later

DESI's DR2 analysis measures the ruler in seven tracer samples with effective redshifts running from 0.295 to 2.330, the highest of them the Lyman-α forest seen in quasar spectra, drawn from more than 14 million galaxies and quasars.²  The distance measurements are sub-percent.  The ruler is the same length everywhere it is measured, once the expansion history is folded in.

The twenty-year arc is worth stating as one sentence.  In 2005 the feature was a 3.4σ bump in 46,748 galaxies.  In 2025 it is a sub-percent distance measurement in seven tracer samples from 14 million objects, and the length has not moved.

What would falsify it, and what nearly does

A test that cannot fail is not a test, so here is what could have gone wrong and did not.

• Wrong length. Had the bump appeared at 80 or 200 Mpc, the model would have been finished. Nothing about galaxy formation sets that length.
• Wrong shape. The predicted feature is broad and asymmetric because of Silk damping and nonlinear smearing. A sharp spike or a step would have failed.
• Drifting length. The comoving length must be the same at z = 0.295 and z = 2.330. A ruler that changed size with redshift would have killed it. DESI tests exactly this.

Three honest weaknesses on the other side.

• The peak is not perfectly stable. Nonlinear gravitational evolution broadens the bump and shifts its centre by a few tenths of a percent. Surveys correct for this by "reconstruction," which uses the observed density field to estimate and undo the bulk flows. That correction is model-dependent at the sub-percent level, which is now the level of the measurement.
• Galaxies are biased tracers. Galaxies do not sample mass uniformly. Bias is fitted, not derived. It affects the amplitude far more than the peak position, which is the reason the acoustic scale is used and the amplitude largely is not.
• DESI and the microwave background do not perfectly agree. DESI DR2 reports that its preferred parameters are "in mild, 2.3σ tension" with the microwave background's, and that the tension eases if dark energy evolves.² That is a live dispute about dark energy, discussed in Dark Energy, not a dispute about the ruler. The ruler itself is what makes the tension measurable.

How the structure grew, and where dark matter is load-bearing

The seeds are visible

The microwave background is smooth to about one part in 100,000.  Those tiny temperature ripples are the same ripples in density that later became galaxies, seen at 380,000 years before gravity had done much to them.  This is not an inference — it is a photograph of the initial conditions.

Gravity amplifies them.  A region slightly denser than average pulls harder, so it grows denser still.  While the universe is matter-dominated and the fluctuations are small, the growth is simple:  the density contrast grows in direct proportion to the scale factor.  From recombination to now, that is a factor of about a thousand.

The problem with ordinary matter, stated honestly

Take the ripples at face value and grow them with ordinary matter alone.  Start at 10⁻⁵ at recombination, multiply by a thousand, and finish at 10⁻².  That is a hundred times too small.  Galaxies and clusters have density contrasts far above one.  The universe today is not one percent lumpy;  it is completely lumpy on the scales of galaxies and clusters.

The reason ordinary matter starts so late is the same physics that makes the acoustic ruler.  Before recombination, baryons are locked to photons, and radiation pressure prevents them from collapsing.  They oscillate instead of growing.  So the baryon clock does not start until 380,000 years.

Cold dark matter fixes this by not being there for the collision.  Because it does not scatter photons, it feels no radiation pressure.  It has been growing since matter began to dominate radiation, at a redshift of 3387 ± 21.²⁴  By recombination its density contrast is already well above the baryons'.  When the photons let go, the baryons fall into wells that are already deep, and catch up quickly.

Say plainly what this is

This is an assumption imported into the calculation, not a measurement of a substance.  No experiment has weighed a dark matter particle, and forty years of direct searches have found nothing — the full accounting is in Dark Matter.  The structure-growth argument does not detect dark matter.  It shows that if structure grew by gravity from the observed microwave-background ripples over roughly thirteen billion years, then something that clumps and does not interact with light must supply about five times the mass of the ordinary matter.

Two things keep this from being circular, and both are worth stating.

• The microwave background measures the same ratio independently, from a different feature. The relative heights of the acoustic peaks are set by how much of the oscillating fluid was pressure-supported baryons and how much was pressureless matter riding along. That measurement gives a cold-dark-matter density of 0.1200 ± 0.0012 against a baryon density of 0.02237 ± 0.00015, a factor of 5.4.²⁴ It comes from peak heights, not from the growth calculation.
• The Bullet Cluster measures a mass-light offset with no cosmology at all — an 8σ separation between where the gas is and where the gravity is.²⁵

So the honest statement is:  the growth argument does not stand on its own, and it does not need to.  It agrees with two other lines of evidence that are independent of it.  What none of the three does is produce a particle.

Simulations, and what they do and do not prove

Cold dark matter structure formation is usually shown as a simulation — the Millennium Run of 2005 and its successors.  The Millennium team stated the logic clearly:  "Together with the theory of cosmic inflation, this model makes a clear prediction for the initial conditions for structure formation and predicts that structures grow hierarchically through gravitational instability."²⁶

A simulation is not evidence by itself.  The gravity-only part is honest:  once the cosmological parameters and the initial power spectrum are fixed, the dark-matter skeleton follows with no free parameters.  The visible part is not:  star formation, supernova feedback and black-hole feedback are inserted as tunable recipes, and different groups tune them differently to match the same data.  When a simulation reproduces the galaxy population, that is partly a fit.  When it reproduces the shape of the cosmic web at large separations, that is closer to a prediction, because the recipes barely touch scales of tens of megaparsecs.

Redshift-space distortions and the growth rate

The acoustic ruler measures geometry — how far away things are.  Redshift-space distortions measure dynamics — how fast structure is growing.  These are different tests, and the second is the one that checks whether gravity behaves as general relativity says on hundred-megaparsec scales.

The idea is Kaiser's from 1987.  Galaxies fall toward overdense regions.  On the near side of a growing structure they fall away from us, adding redshift;  on the far side they fall toward us, subtracting it.  The structure therefore looks squashed along the line of sight.  How much it is squashed depends on how fast material is falling in, which depends on how fast structure is growing.

The measured quantity is written fσ₈ — the growth rate times the clumpiness.  Its predicted value follows from general relativity acting on the ΛCDM expansion history, with no freedom to adjust it.

The result.  The completed twenty-year SDSS programme combined eight acoustic-oscillation measurements and six growth-rate measurements:

The combined BAO and RSD measurements indicate σ₈ = 0.85 ± 0.03, implying a growth rate that is consistent with predictions from Planck temperature and polarization data and with General Relativity.⁵

The same paper reports a second result that is easy to miss and hard to dismiss:  "the BAO data alone are able to rule out dark-energy-free models at more than eight standard deviations."⁵  That is the acoustic ruler, on its own and without supernovae, ruling out a universe with no dark energy.

Where this test is weak.  Redshift-space distortion measurements need a model of nonlinear clustering and of how galaxies trace mass, and different modelling choices shift fσ₈ by an amount comparable to the error bars.  Individual measurements at the 5-10 percent level are not as clean as the acoustic peak position.  The strength of the result comes from six of them agreeing across a wide redshift range.

A second, independent ruler: the power-spectrum turnover

The acoustic scale is not the only length frozen into the matter distribution.  There is also the turnover — the peak of the matter power spectrum, set by the size of the horizon when matter first came to dominate radiation.  It is a completely separate scale from the sound horizon, fixed by different physics.

DESI reported the first detection of it in the galaxy auto-power spectrum in 2025, and did the analysis blinded to avoid confirmation bias.²⁷

But report the precision honestly.  The turnover gives Ω_mh² = 0.139, with errors of +0.036 and −0.046.  That is roughly a 30 percent measurement.  The microwave background gives 0.1430 ± 0.0011, which is 0.8 percent.²⁷  The turnover confirms the equality scale;  it does not compete with the microwave background, and anyone citing it as a precision result is overselling it.  Its value is that it is model-independent and blinded.

The S8 tension, and how it dissolved

For about five years, weak gravitational lensing surveys found the low-redshift universe slightly less clumpy than the microwave background predicted.  The parameter is S₈, defined as σ₈ times the square root of Ω_m/0.3.  This was widely reported as a crisis for ΛCDM.  It is now essentially gone, and the way it went is instructive.

Two things stand out.  First, DES never had a strong tension:  its Year 3 analysis said it agreed with Planck in 2022, while headlines were still describing S8 as a crisis.  Second, when KiDS moved, its own collaboration attributed the shift to its own methods:  "the increase seen in S₈ primarily results from our improved redshift distribution estimation and calibration, as well as a new survey area and improved image reduction."⁶

The lesson is worth more than the number.  Weak lensing measures cosmic shear of a few parts in a thousand, and the dominant error is not the shear — it is knowing the redshift distribution of the source galaxies from broadband colours alone.  A small bias in that distribution shifts S8.  When the calibration improved, the tension went away.  Any apologetic argument built on S8 has now expired, and citing the 2021 number in 2026 is citing a superseded measurement.

JWST and the "impossible early galaxies"

This is the one live structure-formation claim from the last four years, and it is the one apologetics reaches for most often.  It does not hold up, and the reason it does not hold up is that the paper everyone quotes was replaced by its own authors before it was published.

That said, something real does remain at the end of this section, and it is not the thing usually cited.

What was claimed, and what was published

On 25 July 2022, Ivo Labbé and nine colleagues posted a preprint titled "A very early onset of massive galaxy formation."  Its abstract said:

We find seven galaxies with M∗ > 10¹⁰ M⊙ and 7 < z < 11 in the survey area, including two galaxies with M∗ ∼ 10¹¹ M⊙. …  The presence of these galaxies at z ∼ 10 suggests that galaxies with masses M∗ ∼ 5 × 10⁹ M⊙ may be found out to redshifts as high as z ∼ 18.⁷

That version is what the news coverage was written from, and it is what creationist articles still quote.

The version that appeared in Nature on 22 February 2023 is titled "A population of red candidate massive galaxies ~600 Myr after the Big Bang."  Its abstract says:

In the survey area, we find six candidate massive galaxies (stellar mass > 10¹⁰ solar masses) at 7.4 ≤ z ≤ 9.1, 500–700 Myr after the Big Bang, including one galaxy with a possible stellar mass of ~10¹¹ solar masses.  If verified with spectroscopy, the stellar mass density in massive galaxies would be much higher than anticipated.⁷

Line up the changes.  "Very early onset of massive galaxy formation" became "a population of red candidate massive galaxies."  Seven galaxies became six candidates.  The redshift range 7 < z < 11 narrowed to 7.4 ≤ z ≤ 9.1.  Two objects at 10¹¹ became one "possible" one.  The forecast of galaxies out to z ∼ 18 was deleted entirely.  A conditional — "if verified with spectroscopy" — was inserted.

The authors were not shy about the uncertainty.  Three statements from the published paper:

We note that these masses and redshifts are not definitive and that all galaxies should be considered candidates.⁷

Together, these aspects can produce lower redshifts and lower masses by up to factors of 100 in ways that are not reflected by the random uncertainties.⁷

Additionally, we do not consider contributions from exotic emission line species nor include AGN templates in the fits.  All objects in this paper should be considered candidate massive galaxies, to be confirmed with spectroscopy.⁷

That last sentence matters more than any other in the story.  Not fitting active-galactic-nucleus templates is exactly the assumption that later broke.

The spectroscopic scorecard

Photometric redshifts and masses come from fitting a handful of broadband colours.  Spectroscopy resolves the actual emission lines and settles the redshift.  Here is what happened to each of the six candidates above 10¹⁰ solar masses.  The photometric values are from the authors' own published catalogue;⁷  the cross-identifications were checked by sky position, not by assertion.

The pattern is one-directional.  Every spectroscopic redshift came in lower than the photometric estimate.  Not one of the six is confirmed above 10¹⁰ solar masses under a conservative model.  One is not a galaxy at high redshift at all;  it is an obscured accreting black hole at z = 5.624, which Dale Kocevski's team identified after it was "previously identified as a candidate massive galaxy at z = 8.13 by Labbe et al. (2022)."⁸

Another object from the same catalogue of thirteen candidates, ID 39575, which sat below 10¹⁰ and was therefore not part of the headline claim, was confirmed at z = 7.993.³²  So the photometry was not worthless.  It was systematically biased high in exactly the range that produced the headline.

Little red dots, and why the masses were too high

The mechanism behind the bias has a name.  The candidates are compact, red, and point-like.  Jorryt Matthee and colleagues named the class "little red dots" in a June 2023 preprint.³³  They are now known to be mostly accreting black holes, not stellar populations.

• Kocevski and colleagues assembled 341 of them across 2 < z < 11 in 2025: "The confirmed AGN fraction of our sample is 71% for sources with F444W<26.5."³⁴
• Hollis Akins and colleagues, working in COSMOS-Web, found that mid-infrared coverage is essential: "without it, we derive a mass function inconsistent with ΛCDM."³⁵

Why this inflates masses.  A stellar-population fit assumes all the light comes from stars.  Light from an accretion disc, reddened by dust, mimics the spectrum of an old, massive, dusty stellar population.  The fitter converts that light into stellar mass.  There is nothing wrong with the arithmetic;  the input assumption was wrong.

The quantitative fix.  Tao Wang and colleagues re-measured massive galaxies at z ∼ 3-8 with JWST's mid-infrared instrument included:

excluding MIRI overestimates M∗ by ∼ 0.4 dex on average for M∗ > 10¹⁰ M⊙ galaxies …  MIRI inclusion also reduces "Little Red Dot" (LRD) contamination in massive galaxy samples, lowering the LRD fraction from ∼ 32% to ∼ 13%.¹⁰

0.4 dex is a factor of 2.5.  That is most of the discrepancy on its own.

Does any of this break the standard model?

The "challenge to ΛCDM" framing comes from one paper, Michael Boylan-Kolchin's of 2023, and that paper stated its own condition clearly.

His argument is a hard upper bound and worth understanding, because it is the right kind of argument.  In the standard model, only a known fraction of the mass of the universe is ordinary matter, and only a known amount of that has collapsed into structures by a given redshift.  Multiply the two and you get a ceiling on how much stellar mass can exist, no matter how efficient star formation is.  Call the efficiency ε — the fraction of available baryons turned into stars.

His numbers:  the Labbé masses required ε ≈ 0.99 at z ≈ 9 taken at face value, or ε ≥ 0.57 allowing for the 1σ errors.⁹  In the nearby universe ε is under 0.2.  So the claim was extreme.

And his conclusion was conditional, in the same paragraph as the escape routes:

The results presented here could also be explained if the stellar IMF differs substantially from the assumed Salpeter form, the flux calibration of NIRCam changes from the latest post-flight determinations, or the volumes currently surveyed turn out to be highly atypical.  If none of these explanations holds up and these massive galaxies are spectroscopically confirmed, they will pose a serious challenge for ΛCDM structure formation.⁹

They were not spectroscopically confirmed.  The measured efficiency, once mid-infrared data are included, is about 0.3 for the most massive z ∼ 8 galaxies — above the local value, below Boylan-Kolchin's threshold.¹⁰

Independent analyses agree.  Katherine Chworowsky and colleagues, working the full CEERS sample:  the excess abundances "can be explained by modest changes to star formation physics and/or the efficiencies with which star formation occurs in massive dark matter halos, and are not in tension with modern cosmology."³⁶  Francisco Prada and colleagues in 2026, running the Uchuu simulation through UniverseMachine:  "earlier claims of insufficient stellar masses at z ∼ 8 are inconsistent with our model and are likely driven by systematic uncertainties, including AGN contamination, dust attenuation, and the lack of JWST/MIRI constraints."³⁷

Boylan-Kolchin did not retract, and did not need to.  He set a conditional bar, the condition failed, and he later published a mechanism inside ΛCDM for making early galaxies form efficiently.³⁸  That is how a falsifiable claim is supposed to behave.

What genuinely remains anomalous

Two results should not be swept up in the walk-back, because they are spectroscopically confirmed and they are hard.

• A quiescent galaxy at z = 7.29. RUBIES-UDS-QG-z7 has a stellar mass of 10^10.23 solar masses, stopped forming stars 50 to 100 million years before it was observed, and is only 209 parsecs across. Its discoverers state the problem: "the estimated number density of quiescent galaxies at z ∼ 7 is > 100× larger than predicted from any model to date."³⁹ A galaxy that has already finished forming stars 700 million years after the Big Bang is a much sharper problem than a galaxy that is merely massive.
• Ultraviolet-bright galaxies at z ≈ 14. MoM-z14 is spectroscopically confirmed at z = 14.44, 280 million years after the Big Bang. Its survey implies a number density "> 100× larger (182⁺³²⁹₋₁₀₅×) than pre-JWST consensus models."⁴⁰

But read what these are.  Both are failures of galaxy formation models — recipes for how gas turns into stars — inside an agreed cosmology.  Neither is a failure of the expansion history, the acoustic ruler, or the age.  And the trend at the very highest redshifts runs the other way:  D. J. McLeod and colleagues, searching more than 0.6 square degrees of JWST imaging in 2026, found the ultraviolet luminosity function falling off a cliff beyond z = 12, with results "entirely consistent with a relatively simple galaxy evolution model with no such adjustments."⁴¹

How this has been used, and where the use goes wrong

Two errors are common enough to name.

First, quoting the preprint.  Jason Lisle's "Creation Cosmology Confirmed!" of 9 September 2022 quotes the Labbé v1 title and v1 abstract.⁴²  Publishing on 9 September 2022 and quoting a July preprint is fair at the time.  Repeating it after February 2023, when the authors themselves cut the claim, is not.

Second, the "three orders of magnitude" gloss.  The v1 abstract says the stellar mass density in massive galaxies is higher than expected by "a factor of 10 − 30 at z ∼ 8 and more than three orders of magnitude at z ∼ 10."⁷  That is a statement about the total mass per unit volume in a rare population — a number dominated by how many such objects there are, not by how big each one is.  Lisle's gloss is that "Three orders of magnitude means that these galaxies are more than a thousand times more massive than the secular predictions."⁴²  That is a category error.  A skeptic will spot it immediately, and it discredits the surrounding argument.

One thing on the creationist side that should be scored fairly.  Lisle published a prediction on 21 January 2022, six months before the first JWST deep field, that the telescope would find "fully-formed (fully-designed) galaxies at unprecedented distances," and guessed at headlines reading "Webb discovers that galaxies formed much earlier than previously thought."⁴³  That was made in public, in advance, and headlines of broadly that form did appear through July and August 2022, the coverage of the Labbé preprint among them.  Score it a hit.  It is also soft and non-quantitative, and it does not distinguish his cosmology from any other, since "galaxies look mature early" is compatible with a great many models.  The related dispute over JWST galaxy sizes — where Luke Barnes replied to Lisle in a young-earth journal — is treated in The Doppler Model of Galaxy Redshifts.

What a young-earth reader should take from this

This article does not end with a young universe, and pretending otherwise would be dishonest.  Large-scale structure is one of the places where the evidence runs hardest against a recent creation, and the reasons are worth stating without softening.

The acoustic ruler is geometry, not a clock.  It measures ratios of distances.  A young-universe model that reproduces the standard distance-redshift relation reproduces the acoustic measurements too, because it is the same geometry described differently.  Jason Lisle's anisotropic-synchrony model and Danny Faulkner's boundary model both do this — they are built so that what is observed is what the standard model predicts.  So the ruler is not, by itself, an argument against a young universe.  What it is is a demonstration that the physics used to compute it is right, which matters because the same physics is used everywhere else.

Structure growth is a clock, and this is the hard part.  Turning a 10⁻⁵ density ripple into a galaxy cluster is a gravitational process with a rate.  It is slow.  No published young-earth cosmology performs that amplification in thousands of years, and none has tried in a quantitative way.  The usual creationist answer is that galaxies and clusters were created already assembled — mature, as Adam was created adult.  That is a coherent theological position.  It is not a scientific prediction, because it forecasts nothing in advance and forbids nothing.  A model that would look identical whatever the data showed is not tested by the data.

Compare the two sides on the specific point of prediction.  The acoustic scale was computed in 1970, refined in 1998, proposed as a ruler in 2003, and found in 2005 at the predicted place.  It has since been measured at sub-percent precision across a factor of eight in redshift.  Nothing on the young-earth side is comparable.  Lisle's January 2022 JWST prediction was made in advance and was in the right direction, which is worth something, but it was qualitative and it would have been satisfied by many outcomes.

Where a skeptic should concede ground.  Two things in this article are genuinely awkward for the standard picture, and neither is what apologists usually cite.  A galaxy that had already stopped forming stars 700 million years after the Big Bang, at a number density more than a hundred times any model, is a real problem for galaxy formation.  So is a luminous galaxy at z = 14.44 with a number density 182 times pre-JWST expectations.  These are not evidence for a young earth.  They are evidence that the field does not understand early star formation, and honesty about that is what separates a case file from advocacy.

The full treatment of what young-earth cosmologies can and cannot do is in Young-Universe Cosmologies.

Objections and responses

"The acoustic peak was found because they knew where to look.  That is confirmation bias."

The 2dF team raised the point themselves, in print, about their own result:  "What is impressive in our data is not simply that the results suggest departures from a smooth curve, but that these deviations occur in the locations expected from theory."⁴  Knowing where to look is what makes a prediction a prediction.  The location was published years earlier by people working from microwave-background data with no galaxy survey in hand.  If the bump had appeared 30 percent away, no amount of prior expectation would have produced it.  And the test has been repeated at far higher precision since:  DESI measures the same scale in seven tracer samples with sub-percent errors.

"Turning redshifts into a map assumes the expansion.  The whole thing is circular."

It assumes a fiducial cosmology and then measures how much that cosmology must be stretched.  The stretch is measured separately along and across the line of sight, and a wrong assumption shows up as a spherical feature coming out elliptical.  That is the Alcock-Paczyński test, and it is a check on the assumption, not a hiding place for it.  What is genuinely assumed is that the redshift-distance relation is monotonic — that more distant means more redshifted.  Every cosmology on the table, including the young-earth ones, assumes that.

"Dark matter is invented to make structure formation work."

Half true, and worth conceding.  The growth calculation does need something that clumps and does not couple to radiation, and no such particle has been detected in forty years of searching.  What keeps this from being pure invention is that the same ratio falls out of two other measurements that do not use the growth calculation:  the relative heights of the microwave-background acoustic peaks, and the mass-versus-gas offset in colliding clusters.  Three independent routes to one number is not the signature of a fudge factor.  It is also not a particle.  See Dark Matter.

"JWST found galaxies that cannot exist in the Big Bang model."

It found candidates, in a preprint, that the same authors downgraded before publication.  Six candidates went to spectroscopy;  every one came back lower;  one turned out to be an accreting black hole at z = 5.6 rather than a galaxy at z = 8.  The threshold that would have constituted a real crisis, a baryon-to-star efficiency above 0.57, is not met — the measured value is about 0.3.  Anyone still citing the July 2022 preprint title should read the February 2023 paper it became.

"The cosmic web is too structured to have formed by chance."

The web is not a chance pattern.  It is what gravity does to a nearly uniform gas with small ripples, and it comes out of simulations that contain only gravity and the measured initial conditions.  The filaments, sheets and voids are a consequence, not an input.  The place to press is not the pattern but the timescale, which is the argument made above.

"S8 shows the standard model fails on structure growth."

That was arguable in 2021 and is not arguable now.  KiDS-1000 reported about 3σ;  DES Year 3 reported agreement with Planck in 2022;  KiDS-Legacy reports 0.73σ in 2025, and attributes the change to its own redshift calibration.⁶  The number moved because the measurement improved.

"Big Bang cosmology never predicted the cosmic web."

Correct, and this article says so.  The bubble structure of 1986 surprised everyone, and de Lapparent, Geller and Huchra wrote that their findings "pose serious challenges for current models."¹  The models were then rebuilt.  That is a fair criticism of how the field talks about its own history.  It is not a criticism of the acoustic ruler, which was computed before it was measured and is a different kind of claim entirely.

Sources

The primary papers are all open-access preprints on arXiv unless noted.  Full citations and verbatim quotations with their locators are in the footnotes below.  Every source worked up for this article, including the ones not used, is listed in the literature.

Related articles in this encyclopedia:

• Big Bang Cosmology — the standard model in six layers, and which layers are supported by what
• The Cosmic Microwave Background — the acoustic peaks themselves, and what they measure
• Dark Matter — the gravitational case and the forty-year non-detection
• Dark Energy — the DESI evolving-dark-energy result and what it rests on
• The Doppler Model of Galaxy Redshifts — Lisle's static cosmology and Barnes's reply, including JWST galaxy sizes
• Young-Universe Cosmologies — what the published young-earth models can and cannot do