Theoretical Pathologies in DESI DR2 Dark Energy Claims
The nature of dark energy remains one of the central problems in cosmology. The simplest explanation—a cosmological constant Λ with equation of state w = -1—has remained consistent with all observations for over two decades, yet lacks theoretical motivation from fundamental physics. The possibility that dark energy evolves with redshift has driven major observational programs aimed at detecting deviations from w = -1.
The Dark Energy Spectroscopic Instrument (DESI) collaboration recently reported their Year 1 baryon acoustic oscillation (BAO) measurements combined with cosmic microwave background (CMB) and Type Ia supernova (SNe) data [1]. Fitting to the Chevallier-Polarski-Linder (CPL) parametrization w(a) = w₀ + wₐ(1-a), they report:
\[w_0 = -0.752 \pm 0.057, \quad w_a = -0.861 \pm 0.222\]
This has been widely interpreted as evidence for “evolving dark energy” at high statistical significance, with some analyses claiming tension with ΛCDM exceeding 5σ.
In this paper, we argue that this interpretation is problematic on three independent grounds:
Statistical: The apparent significance is inflated by parametrization artifacts; at the model-independent pivot redshift, the constraint is only 1.9σ from w = -1.
Systematic: Multiple independent analyses have identified calibration issues, statistical fragility, and parametrization dependence.
Theoretical: The DESI best-fit requires phantom dark energy (w < -1), which cannot be realized by any stable field theory.
The theoretical argument is particularly decisive: even if the statistical and systematic concerns were resolved, the physics required to explain DESI’s claimed w(z) evolution is pathological. This places a strong theoretical prior on w = -1 that must be weighed against any observational claim to the contrary.
We performed an independent analysis of the publicly available DESI MCMC chains from the w₀wₐCDM fit combining BAO, CMB (Planck + ACT), and Type Ia supernovae (DESY5). The dataset comprises 61,168 samples with approximately 37,000 effective samples after accounting for chain weights.
The weighted posterior mean and standard deviation are:
| Parameter | Mean | Std |
|---|---|---|
| w₀ | -0.752 | 0.057 |
| wₐ | -0.861 | 0.222 |
The correlation coefficient between w₀ and wₐ is:
\[\rho(w_0, w_a) = -0.907\]
This extreme anti-correlation is not a feature of the data—it is intrinsic to the CPL parametrization and creates a strong degeneracy that inflates apparent tension with any constant-w model.
The CPL parametrization w(a) = w₀ + wₐ(1-a) creates an inherent degeneracy between w₀ and wₐ. At a specific “pivot” redshift, this degeneracy is minimized and the constraint becomes model-independent.
The variance of w(a) is:
\[\sigma^2[w(a)] = \sigma^2_{w_0} + 2(1-a)\text{Cov}(w_0,w_a) + (1-a)^2\sigma^2_{w_a}\]
Minimizing with respect to a:
\[a_p = 1 + \frac{\text{Cov}(w_0,w_a)}{\sigma^2_{w_a}} = 0.766\]
This corresponds to pivot redshift z_p = 0.31.
At the pivot redshift:
\[w(z_p) = -0.954 \pm 0.024\]
The distance from w = -1:
\[\frac{w(z_p) - (-1)}{\sigma[w(z_p)]} = \frac{0.046}{0.024} = 1.9\sigma\]
The 95% credible interval is [-1.001, -0.906], which includes w = -1.
This is the model-independent constraint. The DESI data are consistent with w = -1 at the 2σ level.
The proper measure of tension between a point prediction and a 2D posterior is the Mahalanobis distance:
\[D^2_M = (\mathbf{x} - \boldsymbol{\mu})^T \Sigma^{-1} (\mathbf{x} - \boldsymbol{\mu})\]
For the ΛCDM point (w₀ = -1, wₐ = 0):
\[D^2_M = 18.71\]
For a χ² distribution with 2 degrees of freedom, this corresponds to p = 8.7 × 10⁻⁵, or equivalently 3.9σ.
This is below the 5σ discovery threshold and substantially less than headline claims of “5.8σ” or similar, which likely arise from incorrectly treating w₀ and wₐ as independent measurements.
| Test | Result | Significance |
|---|---|---|
| Pivot redshift (z = 0.31) | w = -0.954 ± 0.024 | 1.9σ from -1 |
| 95% CI at pivot | [-1.001, -0.906] | Includes -1 |
| 2D Mahalanobis | χ² = 18.7 | 3.9σ |
| Posterior fraction with w₀ < -1 | 0.0% | — |
| Posterior fraction requiring phantom | 99.9% | See Section IV |
The statistical evidence for w ≠ -1 is substantially weaker than headlines suggest. At the model-independent pivot redshift, the 95% credible interval includes w = -1.
Multiple independent analyses have raised serious concerns about the robustness of the DESI dark energy signal.
Efstathiou [2] demonstrated that the preference for evolving dark energy is strongly dependent on the supernova sample used. When DESI BAO data are combined with the Pantheon+ supernova compilation instead of DESY5, the signal for w ≠ -1 largely disappears and results become consistent with ΛCDM.
Efstathiou identifies a 0.04-0.06 magnitude offset between low and high redshift supernovae in DESY5, attributable to calibration systematics rather than new physics. The conclusion is stark: the claimed detection depends on which supernova dataset is used, indicating that systematic errors in SNe calibration may be driving the signal.
Dinda et al. [3] applied a model-independent null test to the DESI data and found “no strong evidence for dynamical dark energy.” More concerning, they demonstrated that excluding the single BAO measurement at z = 0.51 largely eliminates the evolving dark energy signal.
When a claimed detection can be erased by removing one data point out of several, the result is statistically fragile and should not be interpreted as robust evidence for new physics.
Giarè et al. [4] showed that the “phantom crossing” implied by the DESI w₀-wₐ fit—where w transitions from w > -1 at low redshift to w < -1 at high redshift—may be an artifact of the CPL parametrization’s forced linear evolution in scale factor.
They demonstrate that thawing quintessence models, which by construction have w ≥ -1 at all times, fit the DESI data comparably well. The apparent phantom behavior is therefore not required by the data; it is imposed by the parametrization choice.
Three independent groups using different methodologies have identified concerns:
This does not prove DESI is wrong, but it establishes that the claimed detection is not robust and should be treated with appropriate caution.
The most serious problem with DESI’s claimed detection is theoretical. Even setting aside statistical and systematic concerns, the physics required to produce DESI’s w(z) trajectory is pathological.
The CPL best-fit parameters imply an equation of state evolution:
\[w(z) = w_0 + w_a \frac{z}{1+z}\]
With w₀ = -0.752 and wₐ = -0.861:
| Redshift | w(z) | Regime |
|---|---|---|
| z = 0.0 | -0.75 | Quintessence (w > -1) |
| z = 0.4 | -1.00 | Crosses cosmological constant |
| z = 0.5 | -1.04 | Phantom (w < -1) |
| z = 1.0 | -1.18 | Deep phantom |
| z = 2.0 | -1.33 | Deep phantom |
The trajectory crosses w = -1 at z ≈ 0.4. For all z > 0.4, the equation of state is in the “phantom” regime.
This is not a feature of the best-fit alone. We computed w(z=1) = w₀ + 0.5wₐ for each MCMC sample:
99.9% of the posterior requires w < -1 at z = 1.
Phantom behavior is not an edge case—it is demanded by essentially the entire DESI posterior.
For a canonical scalar field with Lagrangian:
\[\mathcal{L} = \frac{1}{2}(\partial\phi)^2 - V(\phi)\]
the equation of state is:
\[w_\phi = \frac{\frac{1}{2}\dot{\phi}^2 - V}{\frac{1}{2}\dot{\phi}^2 + V}\]
Since kinetic energy is positive definite, we have:
\[\rho + p = \dot{\phi}^2 \geq 0\]
This means w ≥ -1 always for any canonical scalar field. This is a consequence of the null energy condition.
To achieve w < -1 requires negative kinetic energy:
\[\mathcal{L}_{\text{phantom}} = -\frac{1}{2}(\partial\phi)^2 - V(\phi)\]
The wrong sign in front of the kinetic term defines a “ghost” field. (See Appendix C for discussion of effective phantom behavior in broader frameworks and why these do not generically evade the stability concern.)
Ghost fields are not merely exotic—they are theoretically pathological:
1. Unbounded Hamiltonian
The Hamiltonian for a ghost field is:
\[H = -\frac{1}{2}\dot{\phi}^2 + \frac{1}{2}(\nabla\phi)^2 + V(\phi)\]
The kinetic contribution is negative, making H unbounded from below. The system has no ground state.
2. Vacuum Instability
A ghost can decay into positive-energy particles plus ghost quanta (which carry negative energy) while conserving total energy. The rate for this process is infinite when integrated over phase space. The vacuum decays instantaneously.
3. Ostrogradsky Instability
In any interacting theory containing a ghost, Ostrogradsky’s theorem guarantees that the Hamiltonian is linear in at least one canonical momentum. This leads to classical runaway solutions and quantum vacuum decay.
4. Observational Contradiction
If phantom dark energy existed, the universe would have decayed into a bath of high-energy particles and ghost quanta long ago. We would not be here to observe DESI data.
Several theoretical mechanisms have been proposed to evade the phantom problem:
Quintom (two-field) models: These introduce both a quintessence field (w > -1) and a phantom field (w < -1), allowing crossing. But the phantom component remains a ghost.
k-essence / non-canonical kinetics: These can achieve w < -1 in certain limits, but typically at the cost of gradient instabilities or superluminal propagation.
Modified gravity: The “effective w” reconstructed from H(z) can cross -1 without a phantom field, but post-GW170817 constraints (gravitational wave speed = c) severely limit this possibility.
Interacting dark energy: Energy exchange between dark energy and dark matter can produce effective w < -1, but typically requires fine-tuning and faces structure formation constraints.
None of these provides a theoretically satisfactory explanation for DESI’s w(z) trajectory. The phantom crossing at z ≈ 0.4 with deep phantom behavior at z > 0.5 remains problematic.
There is an additional theoretical constraint from the scalar field mass scale.
For a scalar field with mass m in an expanding universe with Hubble parameter H, the equation of state deviation from w = -1 scales as:
\[1 + w \sim \left(\frac{H}{m}\right)^2\]
To achieve |1 + w| ~ 0.25 (as DESI requires at z = 0), we need:
\[\frac{H_0}{m} \sim 0.5 \implies m \sim 2H_0 \sim 3 \times 10^{-33} \text{ eV}\]
This is an extraordinarily light mass—comparable to the Hubble scale itself. Such ultra-light scalars face severe constraints from fifth-force experiments and equivalence principle tests unless they are screened or couple only gravitationally.
For comparison, typical particle physics scales: - Neutrino mass: ~
10⁻² eV - Axion (QCD): ~ 10⁻⁵ eV
- Required for DESI: ~ 10⁻³³ eV
A 30 order-of-magnitude gap from even the lightest known particles.
Any theoretically healthy dark energy model with scalar mass m >> H predicts:
\[w = -1 + \mathcal{O}\left(\frac{H^2}{m^2}\right) \approx -1\]
The deviation is suppressed by the ratio (H/m)². For any m >> H₀:
| Scalar mass | (H₀/m)² | Predicted w |
|---|---|---|
| m ~ 10⁻²³ eV | 10⁻²⁰ | -1 + 10⁻²⁰ |
| m ~ 10⁻¹⁵ eV | 10⁻³⁶ | -1 + 10⁻³⁶ |
| m ~ 1 eV | 10⁻⁶⁶ | -1 + 10⁻⁶⁶ |
All healthy scalar field theories with m >> H predict w = -1 to extraordinary precision.
The only way to achieve observable w ≠ -1 is: 1. m ~ H (ultra-light, fine-tuned, constrained), or 2. Ghost (pathological)
This places a strong theoretical prior on w = -1.
Concrete realization — STF framework: The Selective Transient Field (DHOST Class Ia, GW170817-compatible) provides a specific zero-parameter derivation of w(z=0) = −1 rather than a prior. The T² coupling integral α(θ) = ∫₀^θ cos²θ’dθ’ has a third-order tangency at the current epoch (θ = π/2): dα/dθ|_{π/2} = cos²(π/2) = 0. The rate of change of the coupling vanishes exactly at z = 0, forcing Λ̇_eff = 0 and therefore w(z=0) = −1 independent of all free parameters. The field mass is m_s = 3.94 × 10⁻²³ eV >> H₀ ~ 10⁻³³ eV, placing it in the m >> H regime of the table above (predicted deviation |1+w| ~ 10⁻²⁰, consistent). The STF trajectory — w = −1 at z = 0, monotonically phantom at higher z from T² coupling accumulation, no crossing from above — is categorically different from the DESI CPL best-fit (w₀ = −0.752, crossing at z ≈ 0.4). A theory that derives w₀ = −1 exactly from topology, predicts a specific phantom history without a ghost, and whose trajectory shape disagrees with DESI’s claimed trajectory — is precisely the kind of stable microphysical alternative that the theoretical prior argument is pointing toward. See Appendix C.2 and STF Energy V0.3 §VIII for the complete derivation.
We have presented three independent arguments against the DESI dark energy detection:
| Argument | Strength | Conclusion |
|---|---|---|
| Statistical | Moderate | 1.9σ at pivot, 95% CI includes -1 |
| Systematic | Strong | Signal depends on SNe sample, single points, parametrization |
| Theoretical | Decisive | Required physics is pathological (ghost) |
These arguments are independent. Even if the statistical analysis were somehow flawed, and even if all systematic concerns were resolved, the theoretical argument remains: DESI’s w(z) requires phantom dark energy, which cannot exist in any stable field theory.
Observational claims must be weighted against theoretical priors. The history of physics provides guidance:
When an observation requires physics that violates fundamental principles (in this case, vacuum stability), the appropriate response is skepticism pending extraordinary confirmation.
Definitive tests of dark energy evolution require:
If future data confirm w ≠ -1 at high significance across multiple independent probes with resolved systematics, the theoretical community will face a profound challenge. But that threshold has not been met.
A fair summary of current evidence:
The data are consistent with w = -1. The claimed detection of evolution is not robust.
The DESI DR2 claim of evolving dark energy faces three independent challenges:
Statistical: At the model-independent pivot redshift z ≈ 0.31, the constraint w = -0.954 ± 0.024 is only 1.9σ from w = -1. The 95% credible interval includes the cosmological constant value. The apparent high significance arises from the w₀-wₐ correlation (-0.91) inherent to the CPL parametrization.
Systematic: The signal disappears with alternative supernova samples (Efstathiou 2025), vanishes upon excluding single data points (Dinda et al. 2024), and may be a parametrization artifact (Giarè et al. 2024). The detection is not robust.
Theoretical: The DESI best-fit requires phantom dark energy (w < -1) at z > 0.4 in 99.9% of the posterior. Phantom dark energy requires ghost degrees of freedom with negative kinetic energy, leading to unbounded Hamiltonians and instantaneous vacuum decay. No stable field theory can produce DESI’s w(z) trajectory.
The theoretical argument is decisive. Even if statistical and systematic concerns were fully addressed, the physics required by DESI’s best-fit is pathological. The theoretical prior strongly favors w = -1.
We conclude that DESI has not detected evolving dark energy. The claimed signal is either a statistical/systematic artifact or would require physics that violates fundamental stability requirements. The cosmological constant remains the theoretically preferred—and observationally consistent—explanation for cosmic acceleration.
[1] DESI Collaboration, “DESI 2024 VI: Cosmological Constraints from the Measurements of Baryon Acoustic Oscillations,” arXiv:2404.03002 (2024).
[2] Efstathiou, G., “Evolving dark energy or supernovae systematics?” MNRAS 538, 875 (2025).
[3] Dinda, B.R., “A new diagnostic for the null test of dynamical dark energy in light of DESI 2024 and other BAO data,” JCAP 09, 062 (2024).
[4] Giarè, W., et al., “Interpreting DESI 2024 BAO: late-time dynamical dark energy or a local effect?” JCAP 10, 035 (2024).
[5] Chevallier, M. & Polarski, D., “Accelerating Universes with Scaling Dark Matter,” Int. J. Mod. Phys. D 10, 213 (2001).
[6] Linder, E.V., “Exploring the Expansion History of the Universe,” Phys. Rev. Lett. 90, 091301 (2003).
[7] Carroll, S.M., Hoffman, M. & Trodden, M., “Can the dark energy equation-of-state parameter w be less than -1?” Phys. Rev. D 68, 023509 (2003).
[8] Cline, J.M., Jeon, S. & Moore, G.D., “The phantom menaced: Constraints on low-energy effective ghosts,” Phys. Rev. D 70, 043543 (2004).
[9] Woodard, R.P., “Avoiding Dark Energy with 1/R Modifications of Gravity,” Lect. Notes Phys. 720, 403 (2007).
[10] Planck Collaboration, “Planck 2018 results. VI. Cosmological parameters,” A&A 641, A6 (2020).
We analyzed DESI MCMC chains (chain_1.txt through chain_4.txt) containing 61,168 total samples. All statistics are computed using chain weights to properly represent the posterior distribution.
For the CPL parametrization w(a) = w₀ + wₐ(1-a), the variance of w at scale factor a is:
\[\sigma^2[w(a)] = \sigma^2_{w_0} + 2(1-a)\text{Cov}(w_0,w_a) + (1-a)^2\sigma^2_{w_a}\]
Taking the derivative and setting to zero:
\[\frac{d\sigma^2}{da} = -2\text{Cov}(w_0,w_a) - 2(1-a)\sigma^2_{w_a} = 0\]
Solving:
\[a_p = 1 + \frac{\text{Cov}(w_0,w_a)}{\sigma^2_{w_a}}\]
With the DESI posterior values Cov(w₀,wₐ) = -0.01154 and σ²(wₐ) = 0.04936:
\[a_p = 1 + \frac{-0.01154}{0.04936} = 0.766\]
Corresponding to z_p = 1/a_p - 1 = 0.31.
The squared Mahalanobis distance from the posterior mean to ΛCDM (w₀ = -1, wₐ = 0) is:
\[D^2_M = \begin{pmatrix} -1 - (-0.752) \\ 0 - (-0.861) \end{pmatrix}^T \begin{pmatrix} 0.00328 & -0.01154 \\ -0.01154 & 0.04936 \end{pmatrix}^{-1} \begin{pmatrix} -0.248 \\ 0.861 \end{pmatrix} = 18.71\]
For χ²(2), this gives p = 8.67 × 10⁻⁵, equivalent to 3.9σ.
For a canonical scalar with Lagrangian ℒ = ½(∂φ)² - V(φ), the stress-energy tensor gives:
\[\rho = \frac{1}{2}\dot{\phi}^2 + V, \quad p = \frac{1}{2}\dot{\phi}^2 - V\]
Therefore:
\[\rho + p = \dot{\phi}^2 \geq 0\]
This is the null energy condition. Since ρ > 0 (assuming V > 0 or small enough kinetic energy), we have:
\[w = \frac{p}{\rho} = \frac{\dot{\phi}^2/2 - V}{\dot{\phi}^2/2 + V} \geq -1\]
with equality only when φ̇ = 0 (pure cosmological constant).
To achieve w < -1 requires ρ + p < 0, which requires negative kinetic energy:
\[\mathcal{L}_{\text{phantom}} = -\frac{1}{2}(\partial\phi)^2 - V(\phi)\]
The Hamiltonian is:
\[H = \pi\dot{\phi} - \mathcal{L} = -\frac{1}{2}\dot{\phi}^2 + \frac{1}{2}(\nabla\phi)^2 + V\]
where π = ∂ℒ/∂φ̇ = -φ̇.
The kinetic contribution -½φ̇² is negative and unbounded. The Hamiltonian has no minimum—the vacuum is unstable.
In an interacting theory, a ghost can pair-produce with ordinary particles:
\[\text{vacuum} \to \text{ghost} + \text{particle} + \text{particle}\]
Energy is conserved because the ghost carries negative energy. The rate per unit volume, integrated over final-state phase space, diverges. The vacuum decays instantaneously.
This is not a technical issue that might be resolved with better understanding—it is a fundamental pathology that renders phantom field theory physically meaningless.
The analysis in Section IV establishes that the DESI CPL best-fit implies w < -1 at z > 0.4 for 99.9% of the posterior. We argued this constitutes “phantom” behavior requiring ghost degrees of freedom. Here we clarify the scope of this argument.
Cosmological observations of H(z), D_M(z), and D_H(z) constrain the background expansion history, from which an effective equation of state w(z) can be inferred. However, this effective w(z) is a background-level quantity only. It does not uniquely determine:
Effective w_eff < -1 can arise without a fundamental ghost in specific constructions:
The constructions above are not generic escapes:
The point is not that phantom-like w(z) is absolutely impossible in all frameworks, but that:
The theoretical prior therefore remains: observational claims of w < -1 should be treated with skepticism until a concrete, stable microphysical model matching the claimed w(z) shape is demonstrated. The existence of ghost-free effective phantom models (such as STF) confirms that w < -1 is not inherently pathological — but those models predict a trajectory that is incompatible with DESI’s CPL best-fit.