Comparison of dynamical dark energy with ΛCDM in light of DESI DR2

Adam Ormondroyd, Will Handley, M. P. Hobson, A. LasenbyarXiv 2025

TL;DR

This paper re-examines the nature of dark energy using the latest DESI DR2 data, finding that while the overall picture of dark energy's evolution remains consistent, the statistical evidence for complex dynamical models is slightly weakened. However, simpler models like CPL are favored over the standard $\Lambda$CDM, especially when combined with the DES5Y supernova data, suggesting dark energy might be more dynamic than previously thought.

What Problem Does This Paper Address?

Cosmology's standard model, $\Lambda$CDM, has been incredibly successful, but a persistent question is the nature of dark energy, the mysterious component driving the universe's accelerated expansion. The standard model assumes dark energy has a constant equation of state parameter, $w$. However, previous analyses, including the authors' own work, have hinted at a more dynamic $w(a)$ (where $a$ is the scale factor) that deviates from this constant value. This paper addresses whether the improved precision of the new DESI DR2 Baryon Acoustic Oscillation (BAO) data, combined with existing supernova datasets, strengthens or weakens the evidence for this dynamical dark energy, and whether simpler parameterized models are sufficient or if more complex, non-parametric approaches are still necessary.

Methods

The core of this study employs a flexknot reconstruction approach. This is a non-parametric method for modeling the dark energy equation of state, $w(a)$. Instead of assuming a specific functional form for $w(a)$ (like the constant $w$ in $\Lambda$CDM or the parameterized CPL model), flexknots use a flexible linear spline with adjustable "knots" (points where the spline's behavior can change). This allows for the reconstruction of arbitrarily sharp features in $w(a)$ and is particularly useful for exploring deviations from simple models. The authors also utilize the Bayes factor to compare the evidence for different models (flexknot models with varying numbers of knots, CPL, wCDM, and $\Lambda$CDM). Additionally, they employ a tension analysis using the log R statistic to quantify the consistency between different datasets.

Datasets

The primary datasets used in this analysis are:

  • DESI DR2 Baryon Acoustic Oscillation (BAO) data: This is the newly released, high-precision data from the Dark Energy Spectroscopic Instrument.
  • Pantheon+ Type Ia Supernovae data: A compilation of Type Ia supernovae measurements.
  • DES5Y Type Ia Supernovae data: Another set of Type Ia supernovae measurements.

The authors explicitly state that they use these datasets in the same manner as in their previous work (Ormondroyd et al. 2025), with the DESI DR2 data being the key update.

Key Insights and Ideas

  • DESI DR2 data alone weakens evidence for complex dynamics: While the overall shape of $w(a)$ is consistent with previous findings, the statistical support for more complex flexknot models (those with many knots) is reduced with DESI DR2 data compared to DR1. This suggests that the increased precision of DR2, by tightening error bars, can sometimes make it harder to distinguish complex dynamical behavior from a simple constant $w$.
  • Simpler models gain traction: Paradoxically, for DESI DR2 data alone, simpler dynamical models like wCDM ($n=1$ knot) and CPL ($n=2$ knots) show increased evidence relative to $\Lambda$CDM. The CPL model emerges as the preferred dynamical model, aligning with other DESI analyses.
  • Supernova combination matters: The conclusions shift significantly depending on which supernova dataset is combined with DESI DR2 BAO.
  • DESI DR2 + Pantheon+: The results remain broadly consistent with the authors' previous work, showing a preference for simpler models.
  • DESI DR2 + DES5Y: This combination leads to a renewed preference for flexknot models with all numbers of knots, including complex ones. The CPL model is the most favored, with a Bayes factor of ~2.3 over $\Lambda$CDM. This suggests that DES5Y data, when combined with the latest BAO, might be more sensitive to dynamical dark energy.
  • Dynamical dark energy resolves dataset tension: The analysis of dataset tension (log R statistic) reveals that dynamical dark energy models, particularly flexknot models, can help resolve discrepancies between the DESI and DES5Y datasets, whereas $\Lambda$CDM shows a higher tension.

Key Results

  • With DESI DR2 BAO data alone, the statistical support for flexknot models with a higher number of knots decreases compared to DR1.
  • With DESI DR2 BAO data alone, the evidence for wCDM and CPL models increases relative to $\Lambda$CDM, with CPL being the preferred dynamical model.
  • When combining DESI DR2 BAO with Pantheon+ supernovae, the results are similar to previous analyses, favoring simpler models.
  • When combining DESI DR2 BAO with DES5Y supernovae, there is a significant increase in preference for flexknot models (including complex ones) over $\Lambda$CDM, with CPL being the most favored.
  • Dynamical dark energy models help resolve tensions between DESI and DES5Y data, unlike $\Lambda$CDM.

Limitations and Future Work

  • Prior dependence: The authors acknowledge that their Bayesian framework is dependent on prior choices. They demonstrate that using a wider prior for the $w$ parameters in the flexknot reconstruction does not change the overall conclusions, but it does affect the visual appearance of the posteriors at higher redshifts.
  • Statistical significance: While the paper discusses preferences and increased evidence, the statistical significance of the dynamical dark energy models over $\Lambda$CDM can vary depending on the dataset combination and the specific model complexity. The DESI DR2 + DES5Y combination shows the strongest evidence for dynamical models.
  • Future work: The paper implicitly suggests that further investigation with even more precise data, or exploring different parameterizations of dark energy, would be valuable. The differing results between Pantheon+ and DES5Y when combined with DESI DR2 also warrant deeper understanding. The authors also mention a simplification in their marginalization technique for $H_0$, which is a refinement of their methodology.

Summary updated

Abstract

ABSTRACT We present an updated reconstruction of the dark energy equation of state, $w(a)$, using the newly released DESI DR2 Baryon Acoustic Oscillation (BAO) data in combination with Pantheon + and DES5Y Type Ia supernovae measurements, respectively. Building on our previous analysis, which employed a non-parametric flexknot reconstruction approach, we examine whether the evidence for dynamical dark energy persists with the improved precision of the DESI DR2 data set. We find that while the overall qualitative structure of $w(a)$ remains consistent with our earlier findings, the statistical support for dynamical dark energy is reduced when considering DESI DR2 data alone, particularly for more complex flexknot models with higher numbers of knots. However, the evidence for simpler dynamical models, such as wCDM and CPL (which correspond to $n=1$ and $n=2$ knots, respectively), increases relative to $\Lambda$CDM with DESI DR2 alone, with CPL being the preferred dynamical model, consistent with previous DESI analyses. When combined with Pantheon + data, the conclusions remain broadly consistent with our earlier work, but when instead combined with DES5Y supernovae data, there is an increased preference for flexknot models for all values of n considered. This results in all such models being preferred over lambda cold dark matter ($\Lambda$CDM), with the CPL model being the most favoured by a Bayes factor of $\sim 2.3$ relative to $\Lambda$CDM.
BibTeX
@misc{ormondroyd2025comparison,
  author = {Adam Ormondroyd and Will Handley and M. P. Hobson and A. Lasenby},
  title = {{Comparison of dynamical dark energy with ΛCDM in light of DESI DR2}},
  year = {2025},
  eprint = {2503.17342},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2503.17342},
  howpublished = {arXiv},
}
Figure