Torsional pseudo-inflation beyond Einstein-Cartan

This paper investigates whether torsion can sustain a quasi-de Sitter inflationary phase in cosmology, revisiting the Einstein–Cartan spin-fluid inflation proposal in a constructive spirit. The authors compute the budget of the minimal mechanism and then explore what torsion would need to sustain inflation, giving a two-branch answer in Riemann–Cartan cosmology with nonminimal couplings.

Budget of the minimal mechanism: For a Weyssenhoff spin fluid with barotropic source, the accelerated window following the nonsingular Einstein–Cartan bounce spans at most:

Ne = ln[4/(1+3w)] / [3(1−w)] ≤ (1/3) ln 4 ≃ 0.46 e-folds for w ∈ [0,1]. Dust gives 0.46; radiation gives (1/2) ln 2 ≃ 0.35; the stiff limit gives 1/4. Positive spatial curvature shrinks the window further. The spin density dilutes as a−6, and the acceleration condition is s2 > (1+3w)ρ, independent of spatial curvature, while the bounce sits at s2 = 4ρ. The authors note: The spin density is a fuel that dilutes faster than anything else in the inventory, so the engine, however sound, carries a small fuel tank. This is two orders of magnitude short of inflationary needs.

Theoretical framework: The paper works in the nonminimally coupled scalar-field torsion cosmology of Ref. [15], with action containing N(ϕ)R, a canonical scalar kinetic term, potential V(ϕ), and U(ϕ) times the Gauss–Bonnet invariant built from the full torsionful connection. The torsion tensor reduces to vector (h) and axial (f) components. The field equations include two torsion constraints: N h = ϕ̇[4U′Z2 − f2 + 12N′] and f[N − 8U′ϕ̇Z] = 0, where Z ≡ H + h. The second constraint is the fork: either f = 0 (vectorial branch) or N = 8U′ϕ̇Z (axial branch).

Vectorial branch: Setting U = 0, the generalized Friedmann equation collapses to 3NZ2 = ϕ̇2/2 + V, with Z = H + Ṅ/2N. The combination a√N is the Einstein-frame scale factor, and in Einstein-frame time the equation becomes a canonical Friedmann equation with kinetic coefficient 1/N and potential V/N2, without the (N′/N)2 kinetic term that metric formulation adds. The authors state: "That absent term is the entire difference between metric and Palatini inflation, and its absence here is not an accident: the vectorial torsion is the conformal compensator that the Palatini formulation smuggles in through the connection." This branch reproduces Palatini inflation, observationally alive, with ns = 1 − 2/N⋆ and tensor-to-scalar ratio suppressed far below the metric-formulation value.

Axial branch: With U ≠ 0 and taking the branch N = 8U′ϕ̇Z, the authors choose minimal hardware: N = 1, U = u1ϕ (linear Gauss–Bonnet coupling), and flat potential V = V0. The system admits an exact de Sitter solution with all torsion ingredients constant: a = e Ht, ϕ̇ = v, f = f0, h = h0, where the constraints lock the constants: Z2 = v2 + 3f02, h0 = (Z2 − f02)/(2Z), H = (Z2 + f02)/(2Z), u1 = 1/(8vZ), V0 = 6f02 + (5/2)v2. The effective torsion fluid evaluates to ρ T = 3H2, p T = −3H2, i.e., a cosmological constant assembled from dynamical parts. The authors emphasize: Nothing slow-rolls; the expansion is held by geometry. This is why we insist on pseudo-inflation: the phenomenology is inflationary, the mechanism is not the inflaton's.

Exit mechanism: Promoting u1 to a slowly varying U′(ϕ), the potential stays exactly flat, so V0 is fixed. Solving the condensate relations at fixed V0 gives f02 = (2V0 − 5v2)/12, Z2 = (2V0 − v2)/4, U′ = 1/[4v√(2V0 − v2)], with U′ strictly decreasing in the roll speed. The condensate reaches f02 = 0 at v e2 = 2V0/5, where the slope meets its floor U′ floor = 5/(16V0). There the fork snaps back to the vectorial branch: the condensate switches itself off, no tunneling and no cliff. The number of e-folds is set by the stretch of field over which U′ stays above the floor: Ne = H dϕ/v.

Attractor and eigenvalue: The homogeneous dynamics collapses to a single flow ṗ = F(p) for p ≡ ϕ̇, with the condensate as fixed point and eigenvalue exactly F′(v) = −3H, independent of every parameter. Homogeneous deviations die as e−3Ht = a−3.

Strong-coupling estimates: The operator U(ϕ) times Gauss–Bonnet carries a naive cutoff Λ = 1/U′ = 8vZ. On the condensate, H/Λ = (v2 + 4f02)/(16v√(v2 + 3f02)) ∈ [1/(16v), 1/(12v)]. Keeping H below Λ requires v ≳ 10−1 in Planck units; a comfortable decade of hierarchy requires v of order one. Demanding H/Λ ≤ 1/10 forces v ≥ 5/8, and since H ≥ v/2, the Hubble rate during the condensate phase is pinned above roughly 0.3 in reduced Planck units. The authors note: This phase operates near the top of the energy ladder or it does not operate, and we record that as a sharp feature of the mechanism rather than hide it.

Worked profile: With V0 = 17/2 and U′(ϕ) = (1/16)[1 + (ϕ/ϕ c)2]−1, the phase starts at the benchmark condensate (v = f0 = 1, H = 5/4) and exits at ϕ e = √(7/10) ϕ c. Integrating gives Ne ≃ 0.81 ϕ c: sixty e-folds cost ϕ c ≃ 75, a field excursion ∆ϕ ≃ 62 in Planck units, with adiabaticity ratio never exceeding 10−2.

Open problems: (1) Perturbations on the axial branch remain open—the torsion is algebraic so it can be integrated out, yielding Horndeski-type operators, but the spectrum and stability are not yet computed. The authors flag an expectation: the condensate background is parity-odd, f0 ≠ 0, so nothing obliges the two graviton helicities to propagate identically through it, potentially imprinting net chirality on the tensor spectrum. (2) The handover at the snap and reheating history need numerical treatment. (3) Quantum-level health of the axial sector: parity, coupling to fermion spin, ultraviolet fate of the near-Planckian operating point, and radiative stability of the exactly flat potential.

Demarcation: The pseudoscalaron models make the axial mode propagate and appoint it inflaton; the Einstein–Cartan Higgs program scans couplings linear in curvature plus topological terms but does not include the Gauss–Bonnet invariant of the torsionful connection; designer torsion histories imposed by hand have no Lagrangian origin. The corner occupied here—a non-propagating axial condensate sourced by U(ϕ), with an exit written into the constraint structure—is claimed as a new result, with nearest ancestor the scalar-Euler coupling of Toloza and Zanelli.

The paper closes a torsional trilogy: a no-go for diluting torsion at late times, a companion paper showing coupling-sourced torsion can impersonate the late dark sector, and this paper showing the same couplings hold up a quasi-de Sitter phase at early times.

Based on this paper, here are specific improvements to AI systems and what the improved systems can do:

1. Constraint-Aware Physics Simulation

• Improvement: Train the AI to recognize that physical mechanisms have hard mathematical limits (e.g., the ≤0.46 e-folds bound for spin-fluid torsion) and to automatically flag proposals that violate such bounds before full simulation.

• Capability: An AI cosmology assistant that instantly rejects inflationary models exceeding known fuel-budget ceilings, saving compute and preventing false-positive searches.

2. Branch-Selection Reasoning

• Improvement: Implement a decision-tree module that, given a field-theory action, automatically identifies algebraic constraint forks (like the f=0 vs. N=8U′ϕ̇Z branches) and explores each branch's observational consequences separately.

• Capability: An automated theory-classifier that maps any torsion/scalar action to its Palatini-like (vectorial) or condensate-like (axial) phenomenology, predicting which branch survives CMB constraints without manual derivation.

3. Exact-Solution Discovery from Constraint Locking

• Improvement: Train the AI to search for condensate solutions where algebraic constraints lock all dynamical variables to constants (as in the de Sitter solution), rather than assuming slow-roll or perturbative expansions.

• Capability: A solver that finds non-perturbative fixed points in modified-gravity systems, potentially discovering new exact cosmological solutions in other torsion or nonmetricity theories.

4. Exit-Mechanism Detection

• Improvement: Add a pattern-recognition layer that identifies when a coupling parameter (like U′) has a floor value at which a branch condition (f02=0) automatically terminates a phase, enabling self-switching-off without tunneling.

• Capability: An AI that can design self-limiting inflationary models with built-in graceful exits, useful for generating candidate theories with natural reheating triggers.

5. Strong-Coupling Safety Checker

• Improvement: Integrate the H/Λ hierarchy criterion (H/Λ ≤ 1/10 requiring v ≥ 5/8) as a pre-filter for any proposed high-energy physics model.

• Capability: An automated vetting tool that rejects models operating above their cutoff, flagging near-Planckian regimes as sharp features rather than hiding them, improving transparency in model reporting.

6. Parity-Violation Predictor

• Improvement: Train the AI to recognize that parity-odd backgrounds (f0≠0) imply chiral gravitational wave propagation, and to automatically compute the resulting tensor chirality for any such background.

• Capability: A prediction engine that generates testable signatures (net chirality in B-modes) from any torsionful or parity-violating inflationary model, enabling falsifiable forecasts.

7. E-Fold Budget Optimizer

• Improvement: Implement the relation Ne = H dϕ/v as a cost function, allowing the AI to optimize field excursions and coupling profiles to hit a target (e.g., 60 e-folds) with minimal super-Planckian displacement.

• Capability: A design tool that produces minimal-excursion inflationary potentials with specified e-fold counts, useful for embedding in string-theory compactifications with limited field ranges.

8. Cross-Theory Translation

• Improvement: Use the paper's demarcation (pseudoscalaron vs. Einstein–Cartan vs. designer torsion) to train the AI to map equivalent mechanisms across different formalisms, identifying same physics, different language cases.

• Capability: An AI that can translate a mechanism from Palatini to metric to torsion formulations, revealing hidden equivalences and preventing redundant rediscovery.

9. Radiative-Stability Forecaster

• Improvement: Add a module that predicts whether an exactly flat potential (V=V0) survives quantum corrections, using the paper's open problem on radiative stability as a training case.

• Capability: An AI that flags fine-tuned potentials and suggests symmetry or topological protections (like the Gauss–Bonnet coupling) that preserve flatness, improving model robustness.

10. Torsional-Trilogy Integrator

• Improvement: Train the AI to combine results across a series of papers (no-go, late-time impersonation, early-time quasi-de Sitter) into a unified phase-space diagram of torsion's cosmological role.

• Capability: A synthesis AI that produces a complete torsion timeline from bounce to inflation to dark energy, enabling holistic tests against multi-epoch observations.