Kinetic Mixing and the Phantom Illusion: Axion-Dilaton Quintessence in Light of DESI DR2

TL;DR

The paper addresses the DESI hints of dynamical dark energy with phantom-like evolution by proposing KMIX, a string-m motivated two-field quintessence with kinetic mixing between axion and dilaton fields. It develops a fast inference framework using normalizing flows to map KMIX onto the CPL parameterization via w(a)=w0+w_a(1-a) and to invert that mapping, enabling efficient KMIX constraints from existing CPL MCMC chains. Results show Planck+DESI BAO favor KMIX at about 2.5σ relative to ΛCDM, with SN data shifting the significance depending on the SN sample, while full-shape data reduce the deviation to ~1.7–2.0σ; KMIX predicts distinctive perturbation effects such as a small suppression of power on large scales and moderate enhancement on small scales, distinguishing it from CPL. The methodology demonstrates a general approach to testing multi-field dark energy theories against phenomenological fits, with KMIX offering a consistent explanation for phantom-like signatures if DESI deviations persist in future data.

Abstract

Recent results from DESI BAO analyses suggest that dark energy may not be a cosmological constant and is in fact dynamical. Furthermore, the data suggest that the equation of state may have been in the phantom regime in the distant past, recently undergoing a phantom crossing. In this work, we investigate whether this preference can be realized within a kinetically mixed axion-dilaton (KMIX) quintessence model, a string-motivated system in which an axion-like field couples exponentially to a dilaton-like (moduli) field. Crucially, KMIX can appear phantom in a standard Chevallier-Polarski-Linder (CPL) based analysis. To confront the model with data, we develop a fast pipeline based on normalizing flows that (i) learns a theory-informed prior on $(w_0,w_a)$ from KMIX realizations and (ii) provides an inverse mapping from CPL parameters back to the physical KMIX parameters. By importance-sampling pre-computed CPL chains using this framework, we effectively transform generic phenomenological constraints into direct, computationally efficient constraints on the underlying KMIX theory, avoiding the prohibitive cost of full parameter space exploration. Applied to Planck+DESI DR2 BAO measurements, our framework finds support for KMIX at $2.5σ$ compared to the base CPL fit at $3.1σ$, demonstrating that KMIX may account for the DESI preference without invoking true phantom behavior. When additionally including Type Ia supernovae data, we find that the preference remains above $3σ$ for Union3 and DES Y5, but drops to $2.1σ$ with Pantheon+. The latter, combined with the DESI full-shape power spectrum and bispectrum data, further reduces the preference to $1.7σ$. Ultimately, should the DESI deviation persist with future data, KMIX may offer a theoretically well-motivated explanation for the phantom-like signatures inferred from phenomenological fits.

Figures (8)

• Figure 1: Example background dynamics for the KMIX model with $\lambda=-1.25$, $\alpha=1.7$, $\log_{10}\chi_i=-4$, $\log_{10}m = -31.8$, $\log_{10}f_\phi = 26.3$, and $\theta_i=3.12$. Top panel: Field evolution in units of the reduced Planck mass, with $\phi$ (red) and $\chi$ (blue) shown relative to the black dashed line marking $\phi, \chi = 0$. Both fields remain frozen at early times and become dynamical once $H \sim m_\phi$. Middle panel: Evolution of the fractional density for KMIX scalars $\chi$ (blue) and $\phi$ (red) as well as for all of matter (including the scalar contribution) (solid purple). Additionally, we show the evolution of the early-Universe evolved dark matter density (dashed purple) which makes clear the impact this model has on the inferred dark matter density at late times. Bottom panel: Evolution of the equation of state for the $\phi$ (red) and $\chi$ (blue), together with the effective dark-energy equation of state $w_{\chi,{\rm eff}}$ (green), as defined in Eq. \ref{['chi_eff']}. The horizontal black dashed lines mark $w=-1$ and $w=1$. For $f(\chi)=e^{\lambda\chi}<1$, consistent with the choice of $\lambda$ used here, the kinetic coupling transfers energy from the axion ($\phi$) to the dilaton ($\chi$), effectively increasing $\rho_\chi$ over time. This energy flow produces an apparent phantom phase (seen here starting at $z\approx0.5$ where $w_{\chi,{\rm eff}}<-1$) even though the total system satisfies the Null Energy Condition as evidenced from the combined equation of state for $\phi + \chi$ (orange).
• Figure 2: Theory-informed prior distributions for the KMIX model, derived through normalizing flow (NF) training using samples from the parameter ranges listed in Table \ref{['tab:kmix_priors']}. Left panel: Raw training samples drawn from broad flat priors on the KMIX theory parameters, shown here for representative quantities $\chi_i$, and $\lambda$, demonstrating full coverage of the prior volume. Right panel: Triangle plot of the corresponding mapped $(w_0, w_a)$ distributions obtained through the CPL-matching procedure. Light red points show the raw training data, while increasingly darker contours indicate the 99%, 95%, and 68% confidence regions of the resulting NF prior. Black curves denote the NF-learned priors: dashed lines for the 1D marginalized distributions and solid contours for the 2D joint distribution.
• Figure 3: Two-dimensional posterior distributions in the $(w_0,\,w_a)$-plane at the 68% and 95% confidence levels for the kinetically mixed axion--dilaton (KMIX) model. The analysis combines Planck PR4 CMB anisotropies (with lensing) and DESI DR2 BAO, shown in purple, and separately includes Type Ia supernova datasets from Pantheon+ (blue), Union 3 (orange), and DES Y5 (green). The gray dashed lines denote the $\Lambda$CDM limit, $(w_0,\,w_a)=(-1,\,0)$.
• Figure 4: Comparison of best-fit predictions for DESI DR2 BAO distance scales relative to the best-fit Planck$\Lambda$CDM baseline. The panels show the volume-averaged distance $D_V/r_d$ (left), the transverse comoving distance $D_M/r_d$ (center), and the Hubble distance $D_H/r_d$ (right). Black points indicate DESI DR2 measurements with $1\sigma$ errors. The red line corresponds to the best-fit KMIX model, plotted using the corresponding $(w_{0}^{\rm (eff)}, w_{\rm a}^{\rm (eff)})$, while the black dot-dashed line represents the best-fit standard CPL value.
• Figure 5: Two-dimensional posterior distributions in the projected$(w_0,\,w_a)$-plane at the 68% and 95% confidence levels from the analysis Chudaykin:2025auxChudaykin:2025auz based on DESI DR2 BAO, Planck CMB, Pantheon+ SNe, and DESI DR1 full-shape data. Results are shown for the KMIX (red) and $w_0w_a$CDM (blue) models. The gray dashed lines indicate the $\Lambda$CDM limit, $(w_0,\,w_a)=(-1,\,0)$. While both models suggest consistency (with $\Lambda$CDM within $\sim\!2\sigma$) KMIX exhibits a slightly weaker apparent deviation.