For nearly three decades, standard cosmology has treated dark energy as a constant. Now a unified catalogue of 2,884 exploding stars, combined with maps of galaxies and the Big Bang’s afterglow, has added another significant hint that dark energy changes over time—and may be weakening as the universe ages.

Nearly three decades after Type Ia supernovae revealed that cosmic expansion is accelerating, the simplest explanation remains a cosmological constant: dark energy with a density that does not change as space expands. That constant, written as Lambda, sits at the centre of the standard cosmological model.

A new catalogue of 2,884 likely Type Ia supernovae has made the assumption harder to fit alongside several other major measurements. The catalogue is called Unite. Ryan Camilleri and colleagues combined the Pantheon+ compilation with the Dark Energy Survey’s five-year supernova sample, then reanalysed the observations through a more consistent pipeline. Their 26-page preprint was submitted on 4 September 2026 and revised on 9 September.

This is one study, not settled consensus. The work has not completed peer review, and the supernova catalogue by itself remains consistent with no time evolution in dark energy. The stronger hint appears only after Unite is combined with galaxy-based distance measurements and the cosmic microwave background. That boundary is central to understanding the claim.

A catalogue rebuilt rather than merely stacked

Pantheon+ assembled 1,701 light curves from 1,550 unique, spectroscopically confirmed supernovae observed by many telescopes, reaching redshift 2.26. DES-SN5YR supplied a large, relatively homogeneous high-redshift sample from the Dark Energy Survey. Some objects and low-redshift surveys appeared in both, so adding their headline counts would have double-counted data.

Unite resolves the overlaps and applies common choices for light-curve modelling, sample selection and bias correction. It also adopts an updated cross-calibration between instruments and requires Dark Energy Survey candidates to have a Type Ia probability above 80 per cent. The result is not 2,884 newly observed explosions, but 30 years of observations reconstructed as one Hubble diagram.

The team recalculated host-galaxy stellar masses for more than 98 per cent of the sample. That matters because a Type Ia supernova’s standardised brightness retains a small correlation with the mass of its host. The companion host-mass analysis found that redshift-dependent differences of only about 0.01 magnitude can shift cosmological constraints appreciably.

Supernovae measure expansion, not dark energy directly

Type Ia supernovae are thermonuclear explosions involving white dwarfs. Their peak luminosities are not identical, but astronomers can standardise them using features including the width and colour of each light curve. Comparing the calibrated luminosity with observed brightness gives a distance. The host galaxy’s redshift indicates how much the universe expanded during the light’s journey.

Thousands of such measurements produce a Hubble diagram: distance plotted against redshift. Its shape records how the expansion rate changed across cosmic history. The University of Queensland’s account describes Unite as a reconstruction of three decades of observations rather than a simple merger of two spreadsheets.

Dust can dim and redden a supernova. Surveys preferentially find brighter events near their detection limits. Intrinsic colour, host properties and gravitational lensing can all alter the inferred distance. When the sought-after signal is measured in hundredths of a magnitude, consistent treatment can matter as much as adding objects.

Lambda is the standard model’s deliberately simple term

The accelerating universe was reported in 1998 in one of the foundational supernova papers. The prevailing Lambda cold dark matter model combines ordinary matter, cold dark matter and a cosmological constant. Lambda has constant energy density even while matter thins as the volume of space increases.

Cosmologists describe dark energy’s pressure-to-density ratio with a parameter called w. A cosmological constant has w equal to minus one at every epoch. This is not a detailed theory of what dark energy is. It is the mathematically simplest term capable of reproducing accelerated expansion within general relativity.

A widely used extension, w0waCDM, gives w two adjustable parameters. The value w0 describes the present and wa controls a particular smooth evolution with the universe’s scale factor. That flexibility can reveal a trend hidden by Lambda, but it also gives the model extra freedom to fit statistical fluctuations or mismatches among datasets.

The hint appears only when three cosmic rulers meet

Unite alone does not require time evolution. Its supernova-only w0-wa fit is consistent with wa equal to zero, although its preferred w0 remains about two sigma from minus one. The headline result comes from asking whether one cosmological history can also satisfy baryon acoustic oscillations and the cosmic microwave background.

Baryon acoustic oscillations are a fossil spacing in the distribution of galaxies. DES and the Dark Energy Spectroscopic Instrument measure that standard ruler across multiple redshifts. The microwave background, mapped by Planck, the Atacama Cosmology Telescope and the South Pole Telescope, anchors conditions when the universe was only about 380,000 years old.

The DESI explanation of these three probes is useful: the microwave background provides an early snapshot, galaxy clustering supplies standard rulers at later times, and supernovae trace relative distances. Their parameter degeneracies run in different directions, so combining them can break ambiguities that each leaves open.

With Unite, DES and DESI baryon acoustic oscillations, and the microwave-background data combined, the evolving flat model gives a matter density of 0.305, w0 of -0.861 and wa of -0.60. The corresponding Lambda values are w0 of -1 and wa of 0. Allowing evolution relieves the tension among the datasets.

“Weakening” is an inference inside that model

The central fitted values imply that w was below -1 at earlier epochs and is above -1 today. Under the two-parameter equation used in the paper, the crossing occurs around redshift 0.3. The inferred dark-energy density rises toward that point and then falls toward the present. That recent decline is the basis for saying dark energy may be weakening as the universe ages.

No instrument watched a dark-energy field fade. The observations measured brightness, redshift, galaxy spacing and microwave-background patterns. Weakening describes the behaviour of an inferred density inside one fitted parameterisation. A different model could describe the same departures in terms of modified gravity, interactions in the dark sector, spatial curvature or an unrecognised offset between datasets.

Nor does weakening mean that cosmic acceleration has already stopped. Expansion history depends on the combined densities and pressures of all components. Extrapolating the fitted curve far into the future would add assumptions well beyond the redshift range of the observations.

Three sigma is serious but not decisive

The latest preprint revision reports a 3.3-sigma preference for evolving dark energy when the comparison uses the maximum a posteriori point. A maximum-likelihood calculation gives 3.1 sigma. Those values are large enough to warrant close attention and remain below physics’ conventional five-sigma threshold for discovery.

The paper’s Bayesian model comparison is less impressed. It finds only weak evidence for time evolution. Bayesian evidence averages across the allowed parameter space and penalises the extra room available to the two-parameter model differently from the frequentist tests. Neither framework is automatically the uniquely correct summary.

A sigma value is not the probability that Lambda CDM is wrong. It describes how much the fitted likelihood improves under specified assumptions. It cannot guarantee that every calibration error, selection effect or analysis choice has been captured. The disagreement between statistical summaries is itself a reason to use the word hint.

Systematics are now part of the signal

Unite is the first major supernova cosmology dataset to include a line-of-sight lensing correction by default. Foreground structures can magnify or demagnify an explosion. Correcting with galaxy-density information reduces the Hubble-diagram scatter by about 5 per cent above redshift 0.5, though the cosmological parameters move by less than 0.2 sigma.

The more persistent vulnerability is the low-redshift sample. Nearby supernovae provide the long lever arm that makes time variation measurable, and they come from the era when dark energy is most influential. Yet that sample was assembled from surveys with different telescopes, filters, calibrations and selection functions. Removing objects below redshift 0.1 noticeably shifts the best fit.

The authors also note a non-linear, hook-shaped degeneracy in the w0-wa parameter space. Projection effects can influence where the preferred point appears, and the team plans to test a different time-varying parameterisation. That is a reminder that a tight contour can still inherit the geometry of the model used to draw it.

A new entry in a longer, unresolved case

The DESI collaboration’s March 2025 result already found that combining its galaxy map with different supernova compilations produced preferences ranging from 2.8 to 4.2 sigma. A previous SpaceDaily analysis followed both that hint and later measurements that moved some evidence back toward Lambda CDM.

Unite is valuable because it removes the choice between Pantheon+ and DES-SN5YR from one part of the comparison. Its combined constraint narrows the allowed w0-wa area by about 30 per cent relative to a previous benchmark. Precision, however, is not confirmation, especially when nearby supernovae and model geometry still exert visible leverage.

What would make the case convincing

First, the preprint must survive peer review and reproduction. Independent pipelines should recover the result from the released Hubble diagram and likelihood. The authors say the key data, code and likelihoods will become public when the paper is accepted, so the most complete external audit lies ahead.

Second, the nearby anchor needs greater uniformity. The Dark Energy Bedrock All-Sky Supernova programme has already measured more than 400 supernovae below redshift 0.08 with the same telescope and filters used for the Dark Energy Survey’s distant sample. That offers a direct way to test whether cross-survey calibration is creating part of the apparent evolution.

Finally, future supernova, galaxy-clustering and lensing surveys must converge on the same history. The result does not establish a new field, a change to gravity or a particular fate for the universe. It establishes a more demanding comparison. Dark energy’s apparent weakening is now a coherent possibility produced by several probes together, not a property that any one of them has measured on its own.