Probing Dynamical Dark Energy with Late-Time Data: Evidence, Tensions, and the Limits of the $w_0w_a$CDM Framework

Technical Summary: Probing Dynamical Dark Energy with Late-Time Data

Problem Statement
The standard $\Lambda$CDM model, while empirically successful, faces internal consistency challenges, most notably the "Hubble tension"—a $\simeq 7.1\sigma$ discrepancy between local distance-ladder determinations of the Hubble constant ($H_0 \approx 73.5$ km s$^{-1}$ Mpc$^{-1}$) and values inferred from Cosmic Microwave Background (CMB) data under the assumption of $\Lambda$CDM ($H_0 \approx 67.2$ km s$^{-1}$ Mpc$^{-1}$). A primary motivation for exploring extensions is to determine if late-time dynamics in the dark sector can resolve this tension without introducing new systematics. The Chevallier–Polarski–Linder (CPL) parametrization ($w(a) = w_0 + w_a(1-a)$) is a standard phenomenological framework for testing dynamical dark energy (DDE). However, recent analyses involving the Dark Energy Spectroscopic Instrument (DESI) Data Release 2 (DR2) have yielded conflicting results regarding the preference for DDE and its efficacy in alleviating the Hubble tension. This paper investigates whether the inferred late-time dynamics and the apparent resolution of the Hubble tension are robust or merely artifacts of specific dataset choices, particularly focusing on the consistency between different Baryon Acoustic Oscillation (BAO) compilations.

Methodology
The authors perform a Bayesian parameter estimation and model comparison using the Cobaya framework with the PolyChord nested sampling algorithm. They test two models: the standard spatially flat $\Lambda$CDM and the dynamical $w_0w_a$CDM (CPL) model. The analysis combines early-universe data (Planck 2018 CMB temperature/polarization spectra and Planck+ACT CMB lensing) with a suite of late-universe distance probes:

1. DESI DR2 BAO: High-precision 3D BAO measurements across $z \in [0.295, 2.33]$.
2. SDSS-IV/eBOSS BAO: The completed pre-DESI consensus compilation (14 measurements).
3. BAOtr: A compilation of 15 angular/transverse BAO measurements ($\theta_{BAO} = r_d/D_M$) derived from SDSS data, offering reduced dependence on fiducial cosmology assumptions.
4. PP&SH0ES: The PantheonPlus Type Ia supernova sample calibrated by SH0ES Cepheid distance anchors.

The study evaluates six distinct dataset combinations to assess the stability of CPL inferences. Key diagnostics include:

• Parameter Constraints: Marginalized posteriors for $(w_0, w_a, H_0, \Omega_m)$.
• Expansion History: Reconstruction of the conformal Hubble rate $H(z)/(1+z)$ and deceleration parameter $q(z)$.
• Bayesian Evidence: Calculation of the log-evidence difference $\Delta \ln Z$ to compare CPL against $\Lambda$CDM, accounting for the Occam penalty of additional parameters.
• Data-Level Consistency: Direct comparison of BAO distance ratios ($D_M/r_d$, $D_H/r_d$) between BAOtr and 3D BAO (DESI/SDSS) at the data level, without relying on cosmological fits, to identify intrinsic tensions.

Key Contributions and Results

1. Strong Dataset Dependence of CPL Inferences:
   The paper demonstrates that CPL constraints are highly contingent on the specific low-redshift datasets included.

   • CMB + DESI/SDSS: When combined with standard 3D BAO (DESI or SDSS), the CPL model favors a low $H_0$ solution ($H_0 \approx 63.9\text{--}68.5$ km s$^{-1}$ Mpc$^{-1}$) and a weakly accelerating or nearly coasting present-day universe ($q_0 \simeq 0$). In these cases, the Hubble tension remains severe ($>4\sigma$), and Bayesian evidence is inconclusive or moderately favors $\Lambda$CDM.
   • CMB + PP&SH0ES + BAOtr: Conversely, combinations including the Cepheid-calibrated supernovae (PP&SH0ES) and/or the angular BAO (BAOtr) shift the posterior toward higher $H_0$ ($H_0 \approx 70.8\text{--}71.3$ km s$^{-1}$ Mpc$^{-1}$) and a moderately accelerating expansion ($-1 < q_0 < 0$). These combinations substantially reduce the Hubble tension (to $\sim 2\sigma$) and yield strong-to-very-strong Bayesian evidence favoring CPL over $\Lambda$CDM ($\Delta \ln Z \approx 8\text{--}16$).

2. Identification of BAO Systematics:
   A critical finding is a direct, model-independent mismatch between BAOtr and standard 3D BAO (DESI/SDSS) at low redshift ($z \lesssim 0.5$).

   • At $z=0.510$, the BAOtr angular scale measurement ($\theta_{BAO} = 4.81 \pm 0.17^\circ$) is in $3.34\sigma$ tension with the equivalent angular scale derived from the DESI LRG1 transverse distance ratio ($\theta_{3D} = 4.22 \pm 0.05^\circ$).
   • This discrepancy propagates into the CPL reconstruction: BAOtr prefers slightly lower $D_M/r_d$, while DESI/SDSS prefer higher values. Because the CPL framework allows freedom in the late-time expansion history $E(z)$, these conflicting distance ratios drive the reconstructed $H(z)$ and $q(z)$ into divergent regimes depending on which BAO dataset is used.

3. Nature of the "Super-Acceleration" Tail:
   In CMB-only CPL analyses, the posterior allows for a phantom-like region ($w_0 < -1$) leading to $q_0 < -1$ (super-acceleration, $\dot{H} > 0$). The authors argue this is not a robust physical inference but a degeneracy-driven extrapolation artifact arising from the lack of low-redshift distance anchors. Once low-z data (like DESI) are included, this tail disappears, and the reconstruction shifts to $q_0 \simeq 0$.

4. Sound Horizon Stability:
   Within the CPL framework (where pre-recombination physics is fixed), the sound horizon $r_d$ remains nearly constant across all dataset combinations (varying by $<0.5\%$). Consequently, shifts in $H_0$ are absorbed entirely by changes in the late-time expansion history ($r_d H_0$ and $E(z)$) rather than by a modification of the early-universe sound horizon. This distinguishes the results from Early Dark Energy (EDE) solutions, which alter $r_d$.

Significance and Claims
The paper concludes that the apparent support for dynamical dark energy and its ability to alleviate the Hubble tension are not universal properties of the CPL framework but are strictly dependent on the inclusion of specific low-redshift datasets (specifically PP&SH0ES and BAOtr). The authors claim that:

• The CPL two-parameter form may be too restrictive to simultaneously accommodate all current low-redshift distance probes (CMB, PP&SH0ES, BAOtr, and DESI/SDSS) without internal inconsistency.
• The tension between BAOtr and 3D BAO datasets is a primary driver of the divergent CPL reconstructions.
• The "phantom-like" asymptote ($w(z \to \infty) < -1$) favored by the data is an extrapolated descriptor of low-redshift posterior correlations rather than a direct constraint on early-universe physics.
• Future progress requires either more flexible late-time models (e.g., sign-changing dark energy, modified gravity) or a rigorous re-examination of BAO systematics and cross-calibration among low-redshift probes.

The work does not claim to definitively solve the Hubble tension but rather highlights that the solution is dataset-contingent and that the current "evidence" for DDE is fragile, driven by the specific combination of low-redshift data used.