Self-Interacting Neutrinos in NEDE and DRMD Cosmologies: Posteriors, profiles and evidence using CLiENT

arXiv:2611.13756 · astro-ph.CO · Submitted 2026-11-10 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: We're looking at the paper: "Self-Interacting Neutrinos in NEDE and DRMD Cosmologies".

Paper summary: Vera: So, to wrap up the 'Self-Interacting Neutrinos in NEDE and DRMD Cosmologies: Posteriors, profiles and evidence using CLiENT', this paper is focused on showing that combining these two specific physical extensions resolves the Hubble tension better than either model alone <ref:2611.13756#pg1>.

Jocelyn: They use the CLiENT likelihood emulator to test twelve cosmological parameters, and their main finding is that this combined SI plus DRMD model gives a posterior H zero of seventy point ninety-four with an error of plus or minus zero point sixty km s-one Mpc-one <ref:2611.13756#pg2>.

Subrahmanyan: The authors argue that this suggests that modifications to physics around the epoch of matter-radiation equality, like those found in SI neutrinos and early dark energy decoupling, are viable solutions to the tension <ref:2611.13756#pg3>.

Vera: The title itself points out that they are looking at a combination of these models, which is what makes this study distinct from just testing SI neutrinos or DRMD in isolation <ref:2611.13756#pg2>.

Jocelyn: Ultimately, the paper demonstrates that when you look at the most aggressive data combination available, the SI plus DRMD model provides a much better fit than either component by itself <ref:2611.13756#pg2>.

Subrahmanyan: It suggests that future work needs to focus on observational tests of these specific self-interaction and decoupling mechanisms to see if they can be confirmed or ruled out, perhaps through precision measurements of structure formation <ref:2611.13756#pg3>.

Conclusion: Vera: So we've seen how putting self-interacting neutrinos together with dark radiation and early dark energy decoupling gives us a pretty solid handle on the Hubble tension <ref:2611.13756#pg4>.

Jocelyn: Yeah, it really boils down to this combination of physics beyond standard stuff being the most promising way to explain why we see such a discrepancy in how fast the universe is expanding near us <ref:2611.13756#pg4>.

Subrahmanyan: The authors are using a tool called CLiENT, which is an emulator that learns how the data likelihood works as it changes all those model parameters and nuisance parameters <ref:2611.13756#pg4>.

Vera: Exactly. And when you look at the final result, the SI plus DRMD model gives a posterior for H zero of seventy point nine four with an error of zero point six km s-one Mpc-one <ref:2611.13756#pg4>.

Jocelyn: That’s a specific number that they get from combining the most aggressive data set, which is the CMB data and SH0ES measurements <ref:2611.13756#pg4>.

Subrahmanyan: It's interesting because they show this combined model fits much better than either self-interacting neutrinos or dark radiation matter decoupling on their own <ref:2611.13756#pg4>.

Vera: And the improvement over the standard lambda cdm model is bigger than just adding up the improvements from testing those two models separately <ref:2611.13756#pg4>.

Jocelyn: So, what does this mean for us as listeners who don't deal with cosmology every day? It means that if we are right about these specific interactions in the neutrino sector and the dark sector, then our standard model of cosmology needs some new physics added to it <ref:2611.13756#pg4>.

Subrahmanyan: And the next thing we need is observational tests, like looking at structure formation or perhaps even gravitational waves from these interactions, to see if we can actually find evidence for them <ref:2611.13756#pg4>.

Magnus Uggerhöj, Tobias Johannessen, Jeppe Pagaard Thybo, Steen Hannestad, Thomas Tram

Department of Physics and Astronomy, Aarhus University

astro-ph.CO

Submitted: 2026-11-10

Updated: 2026-11-10

Comments: 24 pages + appendices and references, 10 figures, scripts for reproducing results available at https://github.com/AarhusCosmology/client-f/tree/papers/2026-sidr-drmd

Code: https://github.com/AarhusCosmology/client-f

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 70/100

The gist: A large number of extensions of the ΛCDM model have been proposed as solutions to the Hubble tension…

Key concepts

Self-Interacting Neutrinos (SIν)
This model proposes that neutrinos interact with each other via a new force mediated by a massive scalar or pseudo-scalar particle. This interaction modifies how neutrinos behave in the early universe, offering an alternative mechanism to explain cosmological observations.
Dark Radiation Matter Decoupling (DRMD)
This concept describes an interacting dark sector involving dark radiation and dark matter. The model incorporates self-interactions within the dark radiation sector and interactions between this radiation and a portion of the dark matter, adding complexity beyond standard cosmology.
CLiENT
CLiENT is a likelihood emulator that learns how well different cosmological models fit data by using a neural network trained on many likelihood evaluations. It efficiently explores the parameter space to find the best-fitting model parameters.
Hubble Tension
The Hubble tension refers to the discrepancy between the value of the Hubble constant (the current expansion rate of the universe) measured using different cosmological methods, such as early-universe data (like CMB) and late-universe measurements (like SH0ES).

Terminology

Summary

A large number of extensions of the ΛCDM model have been proposed as solutions to the Hubble tension, but they are almost always tested separately Here we study a combination of two models which have each been proposed as a solution: self-interacting neutrinos (SIν) and the dark radiation matter decoupling (DRMD) of hot new early dark energy. p. 0 The SIν+DRMD model provides a much better fit than either SIν or DRMD on its own, and its posterior gives H0 = 70. p. 094 ± 0.60 km s−1 Mpc−1

Model Framework

The cosmological model investigated contains two distinct pieces of physics beyond the standard ΛCDM cosmology: self-interactions in the neutrino sector, and a new interacting dark sector. The self-interacting neutrinos are modeled where all neutrinos are charged under a new interaction mediated by a massive scalar or pseudo-scalar The hot new early dark energy model describes an interacting dark sector consisting of dark radiation and dark matter, with self-interactions in the dark radiation sector as well as interactions between the dark radiation and a fraction of the dark matter. p. 6

How the training signal is built

The likelihood emulator CLiENT learns the likelihood of the data directly as a function of all model and nuisance parameters using a neural network trained by active learning The emulators are trained on about 3 × 105 likelihood evaluations per dataset. The training process involves drawing the first 6000 points from a Latin hypercube in the prior volume, followed by iterative training and sampling at a temperature T = 7.

Where the gains come from

The SIν+DRMD model provides a much better fit than either SIν or DRMD on its own, and its posterior gives H0 = 70.94 ± 0.60 km s−1 Mpc−1. p. 0 p. 0 With the most aggressive data combination, i. p. 0e., with the small-scale CMB data and SH0ES, the SIν+DRMD model provides a much better fit than either SIν or DRMD on its own, and its posterior gives H0 = 70. p. 094 ± 0.60 km s−1 Mpc−1. The improvement over ΛCDM exceeds the sum of the improvements of SIν and DRMD on their own

Results and Comparison

The best-fit χ2 of the SIν+DRMD model, of DRMD, of SIν and of ΛCDM are listed in Table 3 for the four datasets. With the most aggressive data combination, H0WC+SH0ES, the SIν+DRMD model fits the data much better than either SIν or DRMD on its own. p. 0 Its χ2 is 17.7 lower than for DRMD, at the cost of one additional parameter (∆AIC = −15. p. 27), and its improvement over ΛCDM even exceeds the sum of the improvements of SIν and DRMD on their own. p. 0

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Integration of likelihood emulators into parameter inference allows for detailed investigation of complex models like SInu+DRMD which have a bimodal posterior in the neutrino self-interaction and several flat directions in parameter space. This enables the calculation of posteriors, best fits and profile likelihoods for the model and for the models nested within it using tools like CLiENT.

  2. Development of likelihood emulation techniques reduces computational cost significantly, as a single evaluation can take up to 70 core seconds for H0WC, whereas nested sampling would require about 76 core-years of evaluations of the true likelihood. This makes it feasible to study models with twelve cosmological parameters and many parameter degeneracies.

  3. Profile likelihoods are utilized as a complement to posterior distributions, particularly where prior volume effects are important, allowing researchers to compare models by showing that they prefer a non-zero interacting fraction, fidm approximately 0.01–0.04, whereas the marginalised posteriors give only upper limits.

  4. The system can perform systematic model comparisons by using the Akaike information criterion (AIC) to assess whether the improvement from adding parameters justifies the additional parameters based on ∆AIC = χ2 model − χ2 ΛCDM + 2(Nmodel − NΛCDM).

  5. The AI system can identify which data combination yields better results, as demonstrated by finding that with the most aggressive data combination, H0WC+SH0ES, the SInu+DRMD model fits the data much better than either SInu or DRMD on its own.

Sources

Related papers