DESI: Evolving Dark Energy?
The DESI survey maps 14 million galaxies and hints that dark energy weakens over time (2.8 to 4.2 sigma).
Open in the interactive tree →DESI, on the Mayall telescope in Arizona, measures distances to galaxies and quasars across 11 billion years using baryon acoustic oscillations as a ruler. In the standard cosmological model (LambdaCDM) dark energy is constant. DESI data combined with the cosmic microwave background and supernovae fit a time-varying dark energy better.
As of October 2026
The Data Release 2 BAO results (19 March 2025, over 14 million galaxies and quasars) prefer evolving dark energy at 3.1 sigma when combined with the CMB alone, and at 2.8 (Pantheon+), 3.8 (Union3) or 4.2 sigma (DESY5) when supernova samples are added. No combination is a 5-sigma discovery; the result depends on the choice of supernova sample and on the parametrization, and critics advise caution about the significance. DESI finished its planned five-year survey on 15 April 2026 with more than 47 million galaxies and quasars; the analysis of the full data set is still to come.
Open steps
- Blind analysis of the full survey Medium AI leverageRun the baryon-oscillation and full-shape analysis on all 47 million objects with the blinded pipeline; first results are expected in 2027 and may confirm or erase the hint.
- Supernova calibration and sample choice Medium AI leverageCross-calibrate the Pantheon+, Union3 and DES supernova samples and their host-galaxy corrections; the 2.8 to 4.2 sigma range depends mainly on which sample is used.
- Joint analysis with Euclid and Rubin Medium AI leverageCombine DESI with weak lensing and galaxy clustering from Euclid and Rubin in one pipeline with shared systematics, to test evolving dark energy beyond baryon oscillations.
- Tests without a fixed parametrization High AI leverageCheck evolving dark energy with free-form reconstructions of the expansion history, so the result does not rest on the w0-wa shape alone.
Where AI could help
Medium AI leverage. AI emulators speed up analysis of the full DESI data, but whether the hint reaches 5 sigma depends on data and supernova calibration.
- Neural emulators that make joint fits across many dark-energy models orders of magnitude faster
- Automated cross-checks between supernova samples that disagree
- Simulation-based inference that propagates systematics from the full DESI data set
Shown so far
- A June 2026 preprint (revised September 2026) benchmarked neural simulation-based inference on DESI DR2, Pantheon+ and chronometer data and reported order-of-magnitude shorter runtime, with parameter shifts up to about 1.5 sigma in some cases. source
Prerequisites
- Theory of Relativity1905
- Expansion & Big Bang1929
- Cosmic Microwave Background1964DESI results are combined with CMB data