Does Early Dark Energy Absorb the DESI Late-Time Dynamics Signal? A Combined Analysis

Abstract: While the recent Baryon Acoustic Oscillation (BAO) measurements from the Dark Energy Spectroscopic Instrument (DESI) collaboration are largely consistent with a flat $Λ$CDM cosmology, the preferred parameters are in mild tension with those determined from the cosmic microwave background (CMB). A late-time dynamical dark energy (DDE) solution has been proposed by the DESI collaboration to address this tension. In this work, we investigate whether the statistical preference for DDE is a genuine late-time phenomenon or an artifact of unresolved early-universe physics. To do so, we simultaneously allow for both early- and late-time modifications to the expansion history by combining the Early Dark Energy (EDE) framework with the Chevallier-Polarski-Linder (CPL) parametrization. Excluding the DESI BAO measurements, our joint analysis of the CMB+Pantheon+ datasets demonstrates that within an EDE-extended framework, the CPL parameters remain statistically consistent with the standard $Λ$CDM model. This supports the hypothesis that a DDE signal at low redshifts can be effectively accounted for by an EDE component within the $Λ$CDM background. However, upon the inclusion of the DESI BAO measurements in the joint analysis, a statistically significant deviation from a cosmological constant emerges. Within this combined framework, the best-fit CPL parameters robustly indicate a departure from the standard $Λ$CDM model, favoring a phantom-to-quintessence transition in the DE equation of state. This demonstrates that the DESI preference for the late-time DDE is a robust signature that cannot be absorbed by modifying the physics of the early Universe.

• The paper analyzes whether early dark energy (EDE) modifications can explain the late-time dynamical dark energy (DDE) signal from DESI data.
• Using compressed CMB + DESI + Pantheon+ datasets, EDE-CPL models simultaneously achieve high H0 values and retain a phantom-to-quintessence dynamics, suggesting robustness of DESI DDE signal.
• Including DESI BAO reverses findings from without, delaying $\bum{DDE preference persists regardless of early modifications; is not an artifact of unresolved pre-recombination physics.

Motivation and scope

The DESI DR2 BAO measurements, when combined with Planck CMB data and Type Ia supernova compilations, mildly favor a dynamical dark energy (DDE) equation of state within the Chevallier–Polarski–Linder (CPL) parametrization, with $w_0 > -1$ and $w_a < 0$—a phantom-to-quintessence transition (Collaboration et al., 18 Mar 2025). Separately, early dark energy (EDE) remains the leading pre-recombination solution to the $H_0$ tension, and recent work has argued that EDE fits current data combinations better than late-time CPL modifications alone. Malekjani and Pourojaghi address a natural question at the intersection of these two developments: is the DESI preference for late-time dynamics an artifact of unresolved early-universe physics that an EDE component could absorb? To answer this, they construct a joint framework in which both modifications are active simultaneously—the phenomenological Acoustic EDE fluid of Lin et al. combined with the CPL parametrization—and compare four models: flat $\Lambda$CDM, CPL, EDE-$\Lambda$CDM, and EDE-CPL.

Methodology

The analysis uses compressed Planck CMB distance priors $(R, \ell_A, \Omega_b h^2, n_s)$, Pantheon+ SNe Ia (with and without the 77 Cepheid-calibrated SH0ES subsample treated as a separate likelihood), and DESI DR2 BAO (anisotropic $D_M/r_d$ and $D_H/r_d$ pairs at six effective redshifts plus one isotropic $D_V/r_d$ point). Importantly, $r_d$ is computed self-consistently from the sound-horizon integral within each model rather than fixed to a Planck value, so BAO observables are calibrated internally—an essential feature when testing models that modify the pre-recombination expansion rate. Sampling uses the affine-invariant MCMC sampler emcee; model comparison employs $w_a < 0$0, the Akaike Information Criterion, and Gaussian-equivalent significance estimates from the likelihood-ratio improvement.

Results without DESI BAO

With only CMB + Pantheon+ (no SH0ES), no model extension is statistically preferred: CPL parameters remain consistent with a cosmological constant within $w_a < 0$1, and EDE extensions are penalized by AIC ($w_a < 0$2 for EDE-$w_a < 0$3CDM, $w_a < 0$4 for EDE-CPL). The EDE component nonetheless drives $w_a < 0$5 to $w_a < 0$6 km s$w_a < 0$7 Mpc$w_a < 0$8 and pulls the absolute magnitude toward the SH0ES value despite SH0ES being excluded, confirming the expected $w_a < 0$9–$H_0$0 degeneracy behavior.

Adding SH0ES changes the picture substantially. EDE-$H_0$1CDM achieves $H_0$2 relative to $H_0$3CDM ($H_0$4, roughly $H_0$5), outperforming pure CPL by $H_0$6. Crucially, in the combined EDE-CPL model, the late-time parameters relax back to $H_0$7 and $H_0$8, fully consistent with a cosmological constant, whereas standalone CPL shows a $H_0$9 deviation on this dataset. This supports the hypothesis that, absent DESI BAO, an early-universe modification can absorb the apparent late-time signal—which here originates primarily from the SH0ES calibration rather than from intermediate-redshift distance data.

Results with DESI BAO

Including DESI DR2 reverses this conclusion. With CMB + DESI + Pantheon+ (no SH0ES), standalone CPL yields $\Lambda$0 and $\Lambda$1, a $\Lambda$2 deviation with $\Lambda$3—the only statistically preferred extension. In the EDE-CPL model, the CPL parameters remain essentially unchanged ($\Lambda$4, $\Lambda$5), demonstrating that the DESI-driven DDE preference persists independently of the early-time modification. A caveat arises here: without SH0ES anchoring the background, the EDE fraction inflates to $\Lambda$6, pushing $\Lambda$7 to $\Lambda$8 km s$\Lambda$9 Mpc$\Lambda$0—well above local measurements—and the EDE-extended models lose their statistical advantage over $\Lambda$1CDM by AIC.

With the full dataset (CMB + DESI + Pantheon+SH0ES), EDE-CPL provides the best overall fit ($\Lambda$2, $\Lambda$3, $\Lambda$4 versus $\Lambda$5CDM), simultaneously achieving $\Lambda$6 km s$\Lambda$7 Mpc$\Lambda$8 and retaining the phantom-to-quintessence CPL behavior ($\Lambda$9, $(R, \ell_A, \Omega_b h^2, n_s)$0). The paper's central claim follows directly: the DESI late-time dynamics signal is robust against early-universe modifications and cannot be absorbed by EDE.

Dataset	Best model	Key statistic
CMB + Pantheon+	$(R, \ell_A, \Omega_b h^2, n_s)$1CDM	No extension preferred
CMB + Pantheon+SH0ES	EDE-$(R, \ell_A, \Omega_b h^2, n_s)$2CDM	$(R, \ell_A, \Omega_b h^2, n_s)$3; CPL absorbed
CMB + DESI + Pantheon+	CPL	$(R, \ell_A, \Omega_b h^2, n_s)$4; only preferred extension
CMB + DESI + Pantheon+SH0ES	EDE-CPL	$(R, \ell_A, \Omega_b h^2, n_s)$5; DDE persists

Limitations and open questions

Several caveats bear on these conclusions. First, the analysis operates entirely at the level of background expansion: CMB information enters through compressed distance priors, which do not capture perturbation-level effects of EDE or of a varying dark energy EoS on lensing, ISW, or growth observables. Second, the EDE component is modeled as a phenomenological fluid rather than a specific scalar-field potential, and the Gaussian prior on $(R, \ell_A, \Omega_b h^2, n_s)$6 centered near matter-radiation equality constrains where the transition may occur. Third, the significance estimates rely on asymptotic likelihood-ratio approximations applied to nested-model comparisons whose parameter spaces differ nontrivially, and the authors themselves note that the statistical significance and robustness of the DESI DDE signal remain actively debated in the literature. Fourth, the unphysically large EDE fractions inferred without SH0ES indicate that the EDE parameter space is poorly constrained when the local distance ladder is absent—a degeneracy that future independent $(R, \ell_A, \Omega_b h^2, n_s)$7 anchors could break. Finally, whether the phantom-to-quintessence crossing survives more flexible EoS parametrizations beyond CPL, and whether it persists in full Boltzmann-code analyses including perturbations, remain open questions raised but not resolved by this work.

Conclusion

By allowing early- and late-time modifications to act simultaneously, this study disentangles two effects that individual analyses conflate. Without DESI BAO, EDE absorbs the apparent late-time signal driven by the SH0ES calibration, rendering CPL parameters consistent with a cosmological constant. Once DESI DR2 BAO enters, the phantom-to-quintessence CPL behavior persists essentially unchanged regardless of the EDE sector, and the combined EDE-CPL model delivers the best global fit across all dataset combinations. The evidence therefore indicates that the DESI preference for late-time dynamical dark energy is a genuine low-redshift phenomenon, not a projection of unresolved pre-recombination physics.