A Bayesian Perspective on Evidence for Evolving Dark Energy

Dily Duan Yi Ong, D. P. Yallup, Will HandleyarXiv 2025 · DOI · arXiv

TL;DR

This paper re-evaluates recent claims from the DESI Collaboration regarding "evolving" dark energy using a rigorous Bayesian framework. While frequentist tests suggested strong evidence for a dynamic dark energy model ($w_0w_a$CDM), this analysis finds that the evidence is significantly weaker when accounting for model complexity, suggesting the preference for a new model is primarily driven by resolving statistical tensions between specific datasets rather than a fundamental shift in our understanding of the universe.

What Problem Does This Paper Address?

In early 2024, the Dark Energy Spectroscopic Instrument (DESI) collaboration released data (DR2) that sent ripples through the cosmology community. Their analysis reported up to a $4.2\sigma$ preference for a model where dark energy changes over time ($w_0w_a$CDM), potentially dethroning the reigning "Cosmological Constant" ($\Lambda$CDM) model.

However, the DESI conclusion relied on frequentist hypothesis testing (likelihood ratios). In complex model selection, frequentist methods can sometimes overstate significance because they don't inherently penalize adding new parameters. The authors of this paper address whether this "discovery" holds up under Bayesian model comparison, which naturally incorporates "Bayesian Occam’s Razor"—the principle that simpler models should be preferred unless the data provides overwhelming evidence for complexity.

Methods

The authors employed a robust Bayesian pipeline to calculate Bayesian Evidence ($Z$), a marginal likelihood that integrates over the entire parameter space.

  • Algorithm: They used the nested sampling algorithm PolyChord, integrated via the Cobaya framework.
  • Model Comparison: They calculated the log-Bayes factor ($\ln B$), which compares the evidence for $w_0w_a$CDM against $\Lambda$CDM.
  • Tension Analysis: To understand why the models differed, they used the unimpeded framework, which employs five different metrics (like "suspiciousness" and "Bayesian dimensionality") to quantify whether different datasets (e.g., CMB vs. Supernovae) actually agree with each other under a given model.

Datasets

The study utilized several "frontier" cosmological probes:

  • DESI DR2 BAO: Measurements of Baryon Acoustic Oscillations (the "standard ruler" of the universe).
  • Planck 2018 CMB: Cosmic Microwave Background data (the "baby picture" of the universe).
  • Type Ia Supernovae: Three different catalogs were compared: Pantheon+, Union3, and the recent DES-Y5 (Dark Energy Survey Year 5).

Key Insights and Ideas

  • The "Evidence" is Dataset-Dependent: The preference for evolving dark energy isn't a universal feature of the new DESI data. When combining DESI with Planck CMB data, the Bayesian evidence actually favors the simpler $\Lambda$CDM model ($\ln B = -0.57$).
  • The DES-Y5 Conflict: The strong signal for evolving dark energy only emerges when the DES-Y5 supernova catalog is added to the mix.
  • Resolution of Tension, Not New Physics: The authors' tension analysis reveals a significant $2.95\sigma$ conflict between DESI and DES-Y5 data when viewed through the lens of the standard $\Lambda$CDM model. The $w_0w_a$CDM model is "preferred" by the statistics not necessarily because it's the truth, but because its extra parameters provide the mathematical flexibility to "soak up" the disagreement between these two specific datasets.
  • Frequentist vs. Bayesian Gap: There is a persistent gap between the significance reported by the DESI collaboration ($4.2\sigma$) and the Bayesian significance ($3.07\sigma$). The authors suggest that frequentist asymptotic formulae may be overestimating significance for these specific data combinations.

Key Results

Dataset Combination Bayesian $\ln B$ Bayesian Significance Frequentist Significance (DESI)
DESI DR2 + CMB -0.57 (Favors $\Lambda$CDM) N/A 3.1$\sigma$
DESI DR2 + DES-Y5 +1.56 2.33$\sigma$ 3.3$\sigma$
DESI + CMB + DES-Y5 +3.32 3.07$\sigma$ 4.2$\sigma$
  • $\ln B < 0$ indicates a preference for the standard $\Lambda$CDM model.
  • While the triplet combination (DESI+CMB+DES-Y5) does show a Bayesian preference for evolving dark energy, it is "modest" rather than "definitive," falling below the typical $5\sigma$ threshold for discovery in physics.

Limitations and Future Work

The authors note that their Bayesian results depend on the choice of priors for the dark energy parameters ($w_0$ and $w_a$). While they used the priors established by the DESI collaboration for consistency, different prior volumes could shift the evidence.

Furthermore, the study highlights a "methodological discrepancy" between Bayesian and frequentist approaches in cosmology. They suggest future work should involve extensive Monte Carlo pseudo-experiments to validate the frequentist asymptotic approximations, which might be unreliable given the relatively small effective size of the current data vectors. Ultimately, the community needs to determine if the DESI/DES-Y5 tension is a hint of new physics or an unidentified systematic error in one of the instruments.

Summary updated

Abstract

The DESI collaboration reports a significant preference for a dynamic dark energy model ($w_0w_a$CDM) over the cosmological constant ($Λ$CDM) when their data are combined with other frontier cosmological probes. We present a direct Bayesian model comparison using nested sampling to compute the Bayesian evidence, revealing a contrasting conclusion: for the key combination of the DESI DR2 BAO and the Planck CMB data, we find the Bayesian evidence modestly favours $Λ$CDM (log-Bayes factor $\ln B = -0.57{\scriptstyle\pm0.26}$), in contrast to the collaboration's 3.1$σ$ frequentist significance in favoring $w_0w_a$CDM. Extending this analysis to also combine with the DES-Y5 supernova catalogue, our Bayesian analysis reaches a significance of $3.07{\scriptstyle\pm0.10}\,σ$ in favour of $w_0w_a$CDM. By performing a comprehensive tension analysis, employing five complementary metrics, we pinpoint the origin: a significant ($\approx 2.95σ$), low-dimensional tension between DESI DR2 and DES-Y5 that is present only within the $Λ$CDM framework. The $w_0w_a$CDM model is preferred precisely because its additional parameters act to resolve this specific dataset conflict. The convergence of our findings with independent geometric analyses suggests that the preference for dynamic dark energy is primarily driven by the resolution of inter-dataset tensions, warranting a cautious interpretation of its statistical significance.
BibTeX
@misc{ong2025bayesian,
  author = {Dily Duan Yi Ong and D. P. Yallup and Will Handley},
  title = {{A Bayesian Perspective on Evidence for Evolving Dark Energy}},
  year = {2025},
  eprint = {2511.10631},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2511.10631},
  howpublished = {arXiv},
}
Figure