A sharper ruler for cosmic expansion
The Dark Energy Spectroscopic Instrument collaboration released its three-year baryon acoustic oscillation results on 19 March 2025. The central analysis uses spectra from more than 14 million galaxies and quasars to measure how a characteristic clustering scale appears at different cosmic times. It is the most precise BAO dataset assembled so far and more than doubles the sample used in DESI's first-year analysis.
Baryon acoustic oscillations are the fossil imprint of pressure waves that travelled through the hot early universe. They left a preferred separation between concentrations of matter. Because the physical size of that feature can be calibrated from early-universe physics, astronomers use its apparent size as a standard ruler. Measuring the ruler across redshift reconstructs how the universe expanded.
DESI does this with several populations of galaxies and quasars, each sampling a different period. At redshifts above two, absorption by intergalactic hydrogen in quasar spectra creates the Lyman-alpha forest, extending the expansion history farther back. The collaboration analysed the galaxy, quasar and forest measurements with separate validation studies and synthetic catalogues designed to reveal bias.
What the data say about the standard model
On their own, the new BAO distances are well described by a spatially flat Lambda cold-dark-matter model, or Lambda-CDM. They also agree with DESI's first-year measurements, earlier Sloan Digital Sky Survey results and supernova distances over their shared redshift range. That agreement is an essential part of the result.
The interesting tension appears when DESI is combined with cosmic microwave background data and observations of Type Ia supernovae. In Lambda-CDM, dark energy is represented by a cosmological constant whose density does not evolve. The collaboration also tested a two-parameter model in which the dark-energy equation of state changes with time. Depending on which supernova compilation was included, the evolving model was preferred over Lambda-CDM at a statistical significance ranging from 2.8 to 4.2 standard deviations. DESI plus the microwave background alone gave a 3.1-sigma preference.
Those values are stronger than the first-year hints, but the range itself is a warning. Different supernova samples, calibrations and modelling choices change the answer. A high statistical significance inside an extended model is not automatically proof that the physical model is correct.
Why combination matters
No single probe measures dark energy directly. BAO constrains relative distances, the microwave background anchors early-universe conditions, and supernovae trace luminosity distance at lower redshift. Weak gravitational lensing can add information about how structure grows. Combining them narrows the allowed parameter space, but it can also expose small systematic differences between datasets.
The collaboration reports extensive consistency tests and supporting papers on target selection, mock catalogues and the Lyman-alpha analysis. These checks reduce known sources of bias. They cannot guarantee that every calibration error, astrophysical selection effect or modelling assumption has been found.
A hint, not a discovery of changing dark energy
The result does not show dark energy varying in a laboratory sense, and it does not invalidate Lambda-CDM. Particle physics normally demands evidence around five sigma, together with convincing control of systematic errors, before calling a result a discovery. Here the preference changes with the chosen supernova dataset and comes from fitting an extra degree of freedom.
The responsible conclusion is that the simplest constant-dark-energy model is under meaningful pressure from a particular combination of precise measurements. An evolving equation of state is one possible explanation. Unrecognised systematics or a different extension to cosmology could also account for the pattern.
DESI is continuing its survey, which will enlarge the map and improve its internal tests. Independent supernova calibrations, lensing surveys and future observations from Euclid and the Vera C. Rubin Observatory can test the same expansion history. The 2025 release matters because it converts an intriguing first-year discrepancy into a sharper, reproducible target for the next generation of cosmological measurements.
