A count of 2,884 exploding stars suggests dark energy has not held steady

Each of the 2,884 points in a dataset called Unite marks a star that blew apart. All are likely Type Ia supernovae, the kind astronomers use as standard candles for measuring cosmic distance. A team led by the University of Queensland merged two large catalogues, Pantheon+ and the Dark Energy Survey’s five-year sample, into that single set, and when the data are fitted, they lean toward a dark energy that changes over time rather than holding steady.

AI, explained every day: The Deep View is the go-to daily newsletter for 600,000+ founders, knowledge workers and professionals who want to stay up to date with artificial intelligence.

Get The Deep View →

We may earn a commission if you sign up.

The lean is statistical, and its strength is the whole story: 3.3 sigma by one measure, 3.1 by another, and only weak support by a third.

Two catalogues, one framework

The University of Queensland announced the dataset in a release dated 8 September 2026, describing a global effort to combine 30 years of observations. The analysis went onto arXiv on 4 September 2026 as “Supernovae Unite: Combining Pantheon+ and DES-SN5YR”, revised on 9 September, with PhD candidate Ryan Camilleri as first author and Professor Tamara Davis among the co-authors. No journal reference had been found for it at the time of writing, so it stands as a preprint. Camilleri said the team had rebuilt three decades of astronomical observations into “a single, consistent framework.” Older observations were reanalysed with modern techniques, corrections were applied for cosmic dust, galaxy mass and gravitational lensing, and host-galaxy stellar masses were redetermined for over 98% of the sample.

Unite is not a simple pile of everything. According to the full text of the preprint, the Dark Energy Survey contributes 1,635 candidates, of which 1,499 are photometrically classified as Type Ia, spanning redshifts of 0.05 to 1.13. Pantheon+ supplies 1,701 light curves of 1,550 unique supernovae from 17 sources, with redshifts from 0.001 to 2.26. Where the two overlap, only the 145 events shared between the Dark Energy Survey’s three-year and five-year samples are used. The paper does not break the final 2,884 down by survey.

On its own, with the standard assumption of a flat universe containing a cosmological constant, Unite gives a matter density of 0.310 (+0.012, −0.011).

What 3.3 sigma says, and what it leaves out

The time-varying result needs other data alongside the supernovae. Combining Unite with cosmic microwave background measurements and with baryon acoustic oscillation data from the Dark Energy Survey and DESI, a flat model that lets dark energy evolve returns a present-day equation-of-state parameter w0 of −0.861 (+0.044, −0.042) and an evolution term wa of −0.60. A cosmological constant would sit at w0 of −1 and wa of 0. The same combination has a dark-energy figure of merit of 315, and the abstract reports roughly a 30% smaller confidence region than earlier combinations of Dark Energy Survey Year 6, DESI and microwave background constraints.

The preference over the standard flat ΛCDM model is 3.3 sigma using the maximum a posteriori estimate and 3.1 sigma using maximum likelihood. The abstract adds that the Bayesian evidence indicates only a weak preference for time-evolving dark energy. Sigma counts how many standard deviations a result sits from the baseline expectation; roughly 3 is conventionally interesting, and 5 is the usual threshold for claiming a discovery in physics. Unite alone is only approximately 2 sigma from a cosmological constant in a simpler flat model, according to the preprint. Weak Bayesian support and a roughly three-sigma frequentist preference describe a hint, in Davis’s word, and no discovery is claimed.

The preprint carries its own warnings. In one flat model the authors find strong evidence of tension between the microwave background and the combined baryon-oscillation and supernova data, and they write that the model “cannot provide a good fit to all data simultaneously” and that its best-fit parameters should be treated with caution. They also say it remains important to consider whether the disagreements could arise from systematic errors, and name the cross-calibration of different surveys as one of the largest systematic uncertainties in supernova cosmology.

Two measurements pointing the same way

Davis placed the result in a sequence. The Dark Energy Survey’s supernova data in 2024 first showed hints that dark energy may vary with time, she said, and the 2026 analysis adds that “2 completely independent measurements have found hints of time variation in dark energy.” Camilleri contrasted the result in the release with the standard model of cosmology, which assumes dark energy is fixed and unchanging.

The ScienceDaily account of 22 October 2026 lists the surveys feeding the combined sample, among them the Dark Energy Survey, Foundation, Pan-STARRS1, the Sloan Digital Sky Survey, the Supernova Legacy Survey, the Carnegie Supernova Project and the Harvard-Smithsonian Center for Astrophysics samples. The host-galaxy mass measurements are published separately as a companion preprint, which the release identifies as arXiv 2609.05321.

Next in line is the Dark Energy Bedrock All-Sky Supernova program, DEBASS, which the University of Queensland says will add data on hundreds more supernovae closer to Earth. Whether the 3.3 sigma hardens into something firmer, or fades as the nearby anchor of the distance ladder gets denser, is the question that program is positioned to test.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

AI, explained every day: Get The Deep View →

We may earn a commission if you sign up.