Dark Energy May Not Be Alone: A 3.4-Sigma Hint of an Energy Exchange

Somewhere in the last few billion years, the universe shifted from slowing down to speeding up. The standard explanation is dark energy—a smooth, repulsive influence represented by a cosmological constant, Λ. The standard model, ΛCDM, pairs it with cold dark matter: invisible mass that clumps under gravity. That picture has survived decades of tests. But a new analysis asks whether the dark sectors are truly silent partners.

Dark energy and dark matter are usually treated as separate. If they exchange energy through a non-gravitational interaction, the expansion history of the universe and the growth of cosmic structure would differ from ΛCDM. A preprint posted to arXiv on 13 October 2025 by Tian-Nuo Li and colleagues tests that idea using some of the newest cosmological data available.

A Cosmic Ruler, an Ancient Light, and Exploding Stars

To test interaction, the team combined three kinds of cosmic evidence. The first is baryon acoustic oscillation (BAO) data from DESI DR2. BAO are sound waves frozen into the distribution of galaxies; they act as a ruler that measures how the universe expanded over billions of years. DESI DR2 includes measurements from more than 14 million extragalactic sources across redshift 0.1 to 4.2, where redshift is a measure of how much expansion has stretched light.

The second is the cosmic microwave background (CMB), the relic light from the early universe. The team used CMB measurements from Planck, WMAP, ACT, and SPT. ACT and SPT specialize in smaller angular scales (higher multipoles) with high precision; WMAP mainly covers large scales, and Planck spans a broad range including large scales. The third is type Ia supernovae, exploding white dwarfs whose predictable brightness helps chart cosmic distances. They used two samples: DESY5 and PantheonPlus.

Then they fit four interacting dark energy models. In each, a coupling parameter β controls the energy exchange. If β is zero, there is no interaction and the model reduces to ΛCDM.

One Version of the Interaction Pulls Ahead

The strongest result appears in a model where the interaction is proportional to the dark energy density and the present-day expansion rate, written as Q = β H0 ρ_de. With SPT, DESI DR2, and DESY5, the team finds β = -0.4170 ± 0.1220. That is a 3.4σ deviation from zero. In everyday terms, zero means no interaction; 3.4σ is a notable hint, but not the 5σ standard often used to claim a discovery.

Other forms of the interaction fare worse. Models tied to the dark matter density, Q ∝ ρ_c, give β values very close to zero or are disfavored by the data. A second model tied to dark energy density, Q = β H ρ_de, also gains some support, but less than the H0 version.

Change the Supernovae, and the Hint Softens

Data choices matter. When the team swapped DESY5 for PantheonPlus, the preference weakened. For the same SPT+DESI data, the H0 model gives β = -0.2770 ± 0.1330, a 2.1σ deviation. That is still interesting, but less compelling. The paper notes that this pattern holds regardless of which CMB experiment is used.

CMB choices also shift the evidence. In many cases, SPT strengthened the case for interaction compared with Planck. ACT produced results similar to Planck. Adding WMAP at large scales did not change the picture much. Replacing WMAP with a reduced version of Planck weakened the signal.

Bayes Factors: A Preference, Not a Proof

The team also used Bayesian evidence to compare models. A Bayes factor weighs how well a model fits the data while penalizing extra complexity. For the H0 interaction model, the logarithmic Bayes factors were positive in every dataset combination. With SPT-including data, they ranged from 2.13 to 6.14, which the paper characterizes as a moderate-to-strong preference over ΛCDM overall; on the paper's Jeffreys scale, 2.13 would be weak, while the larger values are moderate to strong. For the Q = β H ρ_c model tied to dark matter density, the Bayes factors were negative in all dataset combinations, indicating that ΛCDM was favored instead. The other dark-matter-density variant, Q = β H0 ρ_c, was negative in all but one combination.

Taken together, the results point in one direction: if dark energy and dark matter interact, the data prefer an interaction that follows dark energy density, especially the form involving H0. But this is a preference among models, not a direct detection of the interaction.

Why the Result Is Not Yet Settled

The study is a preprint and has not completed peer review. The interaction is phenomenological: there is no fundamental theory fixing its exact form, so the four Q models are plausible guesses rather than derived laws. The strongest hint reaches 3.4σ, below the threshold that would make most cosmologists confident. And the signal is sensitive to the supernova sample. DESY5 has been examined for possible calibration issues, and earlier work has debated whether some preferences for new physics come from those data.

Nor is the result independent of CMB assumptions. Different experiments overlap in different ways and cover different angular scales, so combining them requires careful likelihood choices. The paper is transparent about these dependencies, and its own conclusion urges caution.

What Would Turn a Hint Into a Discovery

Future data should help. CMB-S4, the Euclid mission, the Large Synoptic Survey Telescope, and later DESI releases will provide tighter and more independent measurements of expansion and structure growth. If an interaction is real, the evidence should survive changes in supernova samples, CMB experiments, and model assumptions.

For now, the dark sectors may be more connected than ΛCDM assumes—or the hint may fade as measurements improve. Either answer matters. The universe's expansion is the longest-running experiment we can observe, and every new dataset lets us read its results a little more clearly.