Sound in a hot early universe
When the universe was about a thousand times smaller than it is now, ordinary matter was a hot plasma. Light could not travel far through it before scattering off a free electron. Light and matter were locked together as one fluid, and a fluid carries sound.
Picture one spot that starts out a little denser than its surroundings. The extra pressure pushes the light and the ordinary matter outward as a spherical sound wave. The dark matter does not feel that pressure, so it stays behind at the centre. The wave moves at roughly half the speed of light.
Then the universe cooled enough for electrons to join nuclei. Light stopped scattering, and a little later the ordinary matter stopped being dragged along by it. Cosmologists call that moment the drag epoch. It came at a redshift of about 1060. The sound wave stalled where it was, leaving a thin shell of extra ordinary matter around the original clump of dark matter.
Eisenstein, Seo and White describe this picture in their 2007 paper. Everything that happened later is gravity. Matter falls toward both the central clump and the shell. Galaxies form more often at both places. So a galaxy at the centre has a slightly better than random chance of having another galaxy one shell radius away.
One ruler, many galaxies
The real universe started with countless small bumps, not one. Each launched its own shell, and they all overlap. No single shell can be seen on the sky.
The signal appears only when you count pairs. Take every pair of galaxies in a survey, measure how far apart they are, and compare the count with the number of pairs you would expect if galaxies were scattered at random over the same sky. At most separations the two counts are close. Near 147 megaparsecs there is a small bump. At that separation a typical survey finds only a few tenths of a percent more pairs than random, and the bump is part of that small excess. A bump that small needs hundreds of thousands of galaxies.
The first clear detection came in 2005. Eisenstein and colleagues used 46,748 luminous red galaxies from the Sloan Digital Sky Survey and found the bump at about 100 megaparsecs in units that include the Hubble constant. With today's measured expansion rate that is close to 150 megaparsecs. They reported it at 3.4 standard deviations.
Why a known length tells you a distance
You can judge how far away a person is from how tall they look, because you know roughly how tall people are. The BAO spacing works the same way, in two directions.
Across the sky, the spacing appears as an angle. A known length that looks smaller in angle must be farther away. That angle gives the distance to the galaxies at that redshift.
Along the line of sight, the spacing appears as a difference in redshift. How much redshift a fixed length spans depends on how fast the universe was expanding at that time. That gives the expansion rate.
To turn redshifts into positions, a survey first assumes a distance for each redshift. If the assumption is wrong, the spacing comes out too large or too small. If the assumed distance along the line of sight is wrong in a different way from the assumed distance across the sky, the round shell looks squashed. The survey fits the shell back into shape, and the size of the correction is the measurement.
One detail matters for reading results. The survey compares a distance with the length of the ruler. What it reports is distance divided by the sound-travel distance, which cosmologists write as D/rd. Turning that into a distance in megaparsecs needs the ruler length, which comes from the early-universe physics and the amount of ordinary matter.
What DESI has measured
The Dark Energy Spectroscopic Instrument, DESI, is mapping tens of millions of galaxies and quasars. In its 2024 analysis of the first year of data it used more than 5.7 million redshifts and detected the BAO bump in all six redshift ranges it studied. The closest sample, 300,017 bright galaxies around redshift 0.3, gave the weakest detection, 4.0 standard deviations, and measured the distance there to about 2 percent, in units of the ruler length.
The 2025 analysis of three years of data used more than 14 million galaxies and quasars. Within the standard cosmological model, the parameters DESI prefers sit 2.3 standard deviations away from those preferred by the cosmic microwave background. When DESI is combined with that background and with supernova data, models in which dark energy changes over time fit better, by 2.8 to 4.2 standard deviations depending on the supernova sample. That is a preference in the data, not a settled discovery. The paper guide in this series explains how to read those numbers.
Try it yourself
The Lab builds a synthetic catalogue of galaxies with acoustic shells hidden inside. You can watch the shells grow and freeze, then count pairs and see the bump appear. Drag the corner of the box to assume the wrong distances, and watch the ring in the pair map turn into an ellipse.
If you want the physics step by step, the course starts with how far sound travelled before the drag epoch.
Sources: Eisenstein, Seo and White 2007, the acoustic shell picture, Eisenstein et al. 2005, first detection in SDSS luminous red galaxies, DESI 2024 III, galaxy and quasar BAO, DESI DR2 Results II, BAO and cosmological constraints.
Press play, watch the shells freeze at the drag epoch, then find the bump near 147 Mpc in the pair-count plot.