DESI and Gravitational Wave Constraints Challenge Quintessential α-Attractor Inflation

Source: arXiv:2605.00735 · Published 2026-05-01 · By Changcheng Jing, George Alestas, Sachiko Kuroyanagi

TL;DR

This paper studies whether a specific class of α-attractor quintessential inflation models can simultaneously fit late-time dark-energy data and survive primordial gravitational-wave constraints. The key physical tension is that quintessential inflation typically produces a long kination phase after inflation, and that stiff epoch amplifies high-frequency primordial gravitational waves (GWs). Those GWs behave like extra radiation at recombination, so they are constrained by ΔN_eff. The authors argue that once this bound is imposed, the reheating/kination history required to keep the model consistent with DESI and ACT pushes the scalar spectral index n_s too low.

The main result is a full numerical pipeline that evolves the scalar field from inflation through reheating and into the dark-energy era, then computes the GW spectrum from the same background. Using recent CMB, BAO, SnIa, DESI, and ACT data, they show that the model parameter region preferred by late-time dynamics is under severe pressure from GW-induced extra radiation. Their strongest conclusion is not merely that GW constraints tighten the fit, but that the model becomes disfavored once the ΔN_eff upper bound is enforced, because the allowed reheating temperatures no longer give an n_s compatible with ACT/Planck-era constraints.

Key findings

• The authors find that imposing the GW contribution to ΔN_eff makes the α-attractor quintessential inflation model disfavored, because the allowed reheating histories push n_s below the observationally allowed range.
• ACT gives n_s = 0.974 ± 0.003 and ΔN_eff < 0.17 (95%), and the model points that satisfy the GW bound fail to remain inside the 2σ n_s region in Fig. 4.
• In Fig. 4, the largest n_s values they obtain are around n_s ≈ 0.967 for T_reh ∼ 10^7 GeV, but those points still lie outside the 2σ P-ACT-LB contour.
• Figure 3 shows that lowering T_reh from 10^10 GeV to 10^6 GeV steepens the high-frequency GW tail and raises Ω_GW enough that the BBN bound Ω_GW h^2 < 1.3 × 10^-6 excludes the lowest-reheating cases.
• For the representative GW spectrum in Fig. 3, the transition frequency scales as f_reh = 3.8 × 10^-8 (g_*,reh/106.75)^1/2 (g_*s,reh/106.75)^-1/3 (T_reh/GeV) Hz, so the kination-radiation break shifts directly with reheating temperature.
• The UV cutoff frequency f_end scales as f_end ∝ T_reh^-1/3 H_end and, using Eq. (12), as f_end ∝ T_reh^-1/3 M exp(-\u221a(3\u03b1)/2), implying smaller α and lower T_reh move the cutoff to higher frequencies.
• Exp-model I cannot simultaneously realize a dynamically evolving dark-energy equation of state and a strong high-frequency GW enhancement; the paper therefore focuses on Exp-model II for the MCMC analysis.
• The authors report that their fully numerical treatment of scalar-field evolution changes the reheating-to-dark-energy mapping relative to their earlier work because backreaction and Hubble friction shift the freeze-out field value φ_f.

Methodology — deep read

The paper’s threat model is really a cosmological consistency problem rather than an adversarial-security one: the “adversary” is observational data, specifically the combined requirements of CMB scalar-amplitude and spectral-index measurements, BAO/SnIa/DESI late-time expansion data, and upper limits on extra radiation from ΔN_eff. The model assumptions are standard for quintessential inflation: one scalar field drives inflation and later acts as quintessence; after inflation there is no oscillatory reheating phase, so the Universe enters a kination epoch; reheating is modeled with instant preheating; and the GW background comes only from the primordial tensor sector, with anisotropic stress neglected. They also assume the Standard Model value N_ν = 3.044 and treat any extra radiation as coming from primordial GWs. A key modeling choice is that the instant-preheating coupling h is fixed to 0.01, while the other parameters are scanned or determined by shooting.

The data are observational, not a training set. For late-time cosmology they use CMB, BAO, and SnIa datasets, explicitly including DESI DR1/DR2-era BAO information and ACT data; the excerpt names the combined “P-ACT-LB” analysis, which mixes Planck, ACT, ACT lensing, and DESI DR1. For the dark-energy sector they compare against the DESI-preferred dynamical-dark-energy preference and the ACT constraints on n_s and ΔN_eff. No synthetic labels or supervised learning targets are involved; instead the inferred model parameters are compared against published likelihood contours. The paper states that the scalar amplitude A_s is constrained to lie within the 1σ P-ACT-LB range when generating Fig. 4, and that ln(10^10 A_s) is kept fixed within that observational band to suppress degeneracy with the inflationary scale M. The excerpt does not give the full MCMC chain length, burn-in, split, or random seed strategy in the visible text, so those details are unclear from the provided source.

The theoretical model starts from a non-canonical α-attractor action with a pole at φ = ±\u221a(6α). The authors field-redefine to a canonical field φ = \u221a(6α) tanh(\u03c6/\u221a(6α)), which maps the poles to infinity and yields the potential V(φ) = M^2 exp[\u03b3(tanh(φ/\u221a(6α)) - 1)] + V_0. They then derive slow-roll expressions for ε, η, n_s = 1 - 2/N_, and r = 12α (N_ + \u221a(3α)/2)^-2, and fix M from the measured scalar amplitude A_s. Instant preheating is introduced with L_m ⊊ -1/2 g^2 φ^2 χ^2 - h χ ψ ̄ψ, where χ is produced near the enhanced symmetry point φ ≈ 0 and later decays into light fermions ψ. The number density n_χ = (g φ̇)^{3/2}/(8π^3) is injected at φ = 0, and the inflaton equation of motion includes a backreaction term proportional to g n_χ sech^2(φ/\u221a(6α)) | φ | / φ. They rewrite the cosmological system using dimensionless variables x_φ, y_φ, z, n, and h_p, and evolve the coupled ODEs in e-fold time N = ln a. One concrete end-to-end example is: choose α and T_reh, solve the background ODEs from inflation through preheating to today, use the resulting H(a) and a(τ) to evolve tensor modes h_k through Eq. (27), extract |β_k|^2 from the numerical mode function, integrate to get Ω_GW(f), and then map the integrated GW energy density into ΔN_eff via Eq. (41) to check consistency with ACT.

The GW computation is also fully numerical. They solve the tensor-mode equation h_k'' + 2a'/a h_k' + k^2 h_k = 0 in conformal time, impose the Bunch-Davies vacuum for subhorizon modes, and extract the Bogoliubov coefficient |β_k|^2 from the mode function after reentry. The GW spectrum is then Ω_GW(k) = k^4 |β_k|^2 / (3π^2 M_Pl^2 H^2 a^4). This is done on top of the numerically generated background rather than with a piecewise analytic approximation, which matters because the high-frequency modes probe the end-of-inflation-to-kination transition where slow-roll breaks down. Their evaluation explicitly includes the smooth transition in H(t) and the UV cutoff near f_end, rather than inserting an abrupt cutoff by hand. For evaluation they compare against the BBN bound Ω_GW h^2 < 1.3 × 10^-6 above f > 2 × 10^-11 Hz, and against the ACT/P-ACT-LB constraints on n_s and ΔN_eff. The excerpt indicates they also compute Bayesian evidence relative to ΛCDM and modify CAMB for CMB/matter power spectra, but the visible text truncates before the exact likelihood terms, priors, sampler settings, and numerical convergence diagnostics are specified. Reproducibility is therefore partial from the provided excerpt: the equations and pipeline are explicit, but the exact chain configuration and released code/datasets are not shown in the supplied text.

Technical innovations

• They replace approximate reheating-era estimates with a fully numerical evolution from inflation through dark-energy domination, allowing the same solution to determine both late-time w_DE(z) and the primordial GW spectrum.
• They compute the GW-induced ΔN_eff using the numerically derived background rather than an abrupt cutoff approximation at the end-of-inflation Hubble scale.
• They connect DESI/ACT late-time constraints to the reheating duration and then to the GW abundance, showing that the allowed reheating range becomes incompatible with n_s once ΔN_eff is enforced.
• They explicitly model instant preheating with backreaction in the scalar-field equation, which shifts the freeze-out field φ_f and therefore changes the dark-energy trajectory relative to simpler treatments.

Datasets

• P-ACT-LB — not specified in excerpt — combined Planck + ACT + ACT lensing + DESI DR1 constraints
• DESI DR1/DR2 BAO — not specified in excerpt — public DESI observations
• ACT CMB/Neff constraints — not specified in excerpt — public ACT collaboration data
• SnIa compilation — not specified in excerpt — public supernova compilation used in the MCMC

Baselines vs proposed

• BBN bound on Ω_GW h^2: < 1.3 × 10^-6 for f > 2 × 10^-11 Hz vs proposed low-T_reh spectra: excluded for T_reh ≲ 10^6 GeV (Fig. 3)
• ACT P-ACT-LB: n_s = 0.974 ± 0.003 vs proposed model at T_reh ∼ 10^7 GeV: n_s ≈ 0.967 (Fig. 4), outside the 2σ region
• ACT P-ACT-LB: ΔN_eff < 0.17 vs proposed GW contribution: upper edge of allowed model space saturates this bound (Fig. 4), forcing reheating temperatures high enough to lower n_s

Limitations

• The excerpt does not provide the full MCMC configuration, so sampler details, convergence checks, priors, and chain lengths are not reproducible from the visible text.
• The model choice is narrowed to instant preheating with a fixed h = 0.01; the dependence on other reheating mechanisms is not explored in the provided text.
• The GW spectrum is computed assuming no significant anisotropic stress from free-streaming species beyond the Standard Model, which could matter in extensions of the model.
• Exp-model I is effectively ruled out for the combined dark-energy/GW target, but the paper does not explore alternative potential shapes beyond the two exponential variants described.
• The analysis leans on current ACT/Planck/DESI constraints; if those posteriors shift, the allowed reheating window could change materially.

Open questions / follow-ons

• How robust is the disfavoring of quintessential α-attractor inflation under alternative reheating channels such as gravitational reheating or curvaton reheating, where the kination duration may differ substantially?
• Would including additional relativistic species or non-standard neutrino physics materially change the ΔN_eff mapping from the GW spectrum and reopen viable parameter space?
• Can a modified quintessential potential preserve the late-time dynamical-dark-energy behavior while reducing the high-frequency GW enhancement enough to satisfy ACT and BBN simultaneously?
• How sensitive are the conclusions to the precise DESI/ACT likelihood combination and to future shifts in n_s and ΔN_eff posteriors?

Why it matters for bot defense

For bot-defense practitioners, the main transferable lesson is methodological: the paper shows how a hidden intermediate process can couple two observables that are usually analyzed separately, and how a constraint from one domain can collapse parameter space in another. In CAPTCHA and bot-detection work, that translates to treating user behavior, challenge design, and backend telemetry as a coupled system rather than optimizing each in isolation. A change that improves one metric locally can create a detectable signature elsewhere, so you need end-to-end evaluation under the actual deployment pipeline.

More concretely, the paper is a reminder to model the full state evolution and not just the endpoint classifier. If you are designing bot defenses, you would want to validate against distribution shifts in traffic mix, challenge frequency, latency, and adversarial adaptation, because the effective security posture often depends on the entire temporal chain. The authors’ use of a fully numerical pipeline is analogous to evaluating a defense under the real request lifecycle instead of with a static offline benchmark.

Cite

@article{arxiv2605_00735,
  title={ DESI and Gravitational Wave Constraints Challenge Quintessential α-Attractor Inflation },
  author={ Changcheng Jing and George Alestas and Sachiko Kuroyanagi },
  journal={arXiv preprint arXiv:2605.00735},
  year={ 2026 },
  url={https://arxiv.org/abs/2605.00735}
}

Read the full paper

• arXiv:2605.00735 (abstract)
• arXiv:2605.00735 (PDF)