Constant-roll in the Palatini-R² models

Short version: the setup is fine and the qualitative claim is probably right, but the numbers in Tables 1–2 come from an expansion that isn't controlled, and the paper's own numbers show it. What's new: the group takes their Palatini R² + quartic model (their refs [37,39], slow roll) and runs it under the constant-roll condition φ̈ = βHφ̇. The effective Lagrangian (2.7–2.8) is standard and correctly assembled, the constant-roll machinery is applied cleanly, and the finding is real: with β ~ 0.02 the tilt lands in the Planck 1σ band at N ≈ 50–60, where the slow-roll version needed N ≈ 75. That's a modest but genuine result, and the paper is upfront that α and λ are fixed to reproduce P_S (following [43,45,48]) while β is scanned — so r and N are the actual outputs. Where it's soft: the φ̇ solution comes from truncating the EOM at O(φ̇³) (Eq. 3.10–3.15, footnote 2). Footnote 2 justifies this by 'the smallness of φ̇,' but the expansion parameter is αφ̇², not φ̇, because A′ and B′ carry α and α². The paper's own C_s² ≈ 0.34 (Eq. 3.6) gives αφ̇² ≈ 33, so the dropped (3/4)B′φ̇⁴ term is ~50× the retained (1/2)A′φ̇² term. The stress-test arithmetic checks out. Worse, the same αφ̇² ≈ 33 makes the H ≈ √(U/3) expansion in Eq. 3.9 marginal at best — the first correction Aφ̇²/(2U) is O(1), and the second is larger still. If that's right, the whole chain from (3.10) to n_s, r, P_S is not under control. The numerical check that φ̇[φ] satisfies (3.8) is only as good as the H used in it; the paper doesn't say which H. This is fixable — solve the quartic (or the full EOM) numerically and quote the shift in n_s and r — but until then the tables are central values with no error bar on the thing that matters. Who should read it: people working on Palatini inflation and constant-roll model building. Worth a referee report, mainly to force the truncation and error analysis.

The central numbers (n_s, r, P_S in Tables 1–2 and Fig. 1) all descend from the algebraic solution φ̇[φ] of Eq. 3.10–3.15, obtained by truncating the exact EOM (Eq. 3.8) at O(φ̇³). Footnote 2 justifies the truncation by 'the smallness of φ̇.' But the relevant dimensionless expansion parameter is not φ̇; it is αφ̇² (equivalently BX/|A|), because the coefficients A′, B′ scale as α and α². Concretely, for minimal coupling (Eq. 4.2), B′ = −αA′, so the dropped term 3B′X² compares to the retained A′X as (3B′X²)/(A′X) = −(3/2)αφ̇². The paper's own C_s² ≈ 0.34 (Eq. 3.6, ξ=0) gives (1+αφ̇²)/(1+3αφ̇²) = 0.34, i.e. αφ̇² ≈ 33. Hence the term dropped as 'O(φ̇⁴)' is numerically ~50× the term retained as 'O(φ̇²)' in the same equation. This is not a small correction: either A′X is negligible (in which case retaining it is harmless but the truncation bookkeeping is wrong) or it matters (in which case the dropped term matters ~50× more). Either way the cubic solution for φ̇ is not under control, and φ̇ feeds directly into ϵ₁, ϵ₄, C_s², n_s, r and P_S. The paper gives no error estimate on n_s or r from this truncation, and Eq. 4.9 (AX/BX² ~ 10⁻² at φ_i, ~1 at φ_f) confirms the quartic kinetic sector is dominant, not perturbative, throughout inflation. This sharpens the reader's weakest_assumption from 'no error estimate' to 'the expansion parameter is O(30), so the truncation is uncontrolled.'

The reader correctly identified the O(φ̇³) truncation as the weakest link, but framed it as a missing error estimate. The sharper and more damaging version is that the truncation is not a controlled expansion at all: the dropped O(φ̇⁴) term is α²-enhanced and, by the paper's own quoted C_s² ≈ 0.34, exceeds a retained term by roughly a factor of 50. This is a concrete, checkable, internal-consistency issue rather than a matter of taste, and it sits directly upstream of every quoted observable. It is nonetheless fixable — the quartic is solvable and the exact EOM is integrable — so the qualitative claim that constant-roll restores n_s into the Planck band at N ≈ 50–60 may well survive. That keeps the verdict at CONDITIONAL rather than REJECT. I did not find a second concern of comparable weight; the tuning of α and λ to P_S and the scan over β are standard practice for this class of model, and the C_s omission in Eq. 3.7 (if any) affects normalization, not the shape of n_s or r. The paper's derivations are otherwise standard and correctly assembled, and the incremental result — constant-roll at conventional e-fold count where slow roll needed N ≈ 75 — is genuine if the truncation holds. The recommended check is a single numerical recomputation, not a re-derivation of any cited result.

If the result holds, it would show that the Palatini R² quartic model is viable at a conventional e-fold count, which is a useful addition to the Palatini inflation literature and a genuine contrast with the slow-roll analysis of the same model. The k-inflation structure with a significant quartic kinetic term is interesting, and the constant-roll treatment is a reasonable extension of the standard slow-roll analysis. The paper is clearly written and the derivation follows standard Palatini R² machinery. However, the central numerical results depend on an uncontrolled truncation and on parameter choices that are partly fixed by the power spectrum, so the significance is conditional on resolving the issues below.

The paper asks whether a scalar field living in quadratic $R+R^2$ gravity, treated in the Palatini formulation where the metric and the connection are independent, can produce a viable inflationary epoch without assuming slow roll. In the Einstein frame the theory becomes a generalized $k$-inflation model with ordinary and quartic kinetic terms whose coefficients depend on the field, and the authors impose the constant-roll condition $ddot{phi} = beta H dot{phi}$ with $beta$ of order $10^{-2}$ rather than near zero. Focussing on a quartic self-interaction potential — the shape relevant to Higgs inflation — they solve for the field velocity algebraically, then compute the spectral tilt, the tensor-to-scalar ratio and the scalar power spectrum for both minimal and non-minimal coupling to gravity. For $alpha$ around $10^{7}$, $lambda$ around $10^{-13}$ and $beta$ around $0.02$, the tilt lands inside the Planck one-$\sigma$ band and the tensor-to-scalar ratio stays well below the current bound, with the power spectrum matching its measured amplitude. The point of the exercise is comparative: the same quartic Palatini model under slow roll only fit the data at an unusually large number of e-folds, and the authors argue that constant roll removes that awkwardness.

What carries the argument. The machinery is a two-step reduction. First, an auxiliary field $chi$ with $chi^2 = R$ turns $R + alpha R^2/4$ plus a non-minimally coupled scalar into a scalar-tensor action, and a Weyl rescaling $bar{g}_{mu nu} = (1 + xi phi^2 + alpha chi^2) g_{mu nu}$ brings it to the Einstein frame. Second, solving the constraint $delta_chi S = 0$ for $chi^2$ and substituting back produces the $k$-inflation Lagrangian $L = A(phi) X + B(phi) X^2 - U(phi)$ with $X = (nabla phi)^2/2$ and $A = -(1 + xi phi^2 + 4 alpha V (1+xi phi^2))^{-1}$, $B = alpha ((1+xi phi^2)^2 + 4 alpha V)^{-1}$, $U = V ((1+xi phi^2)^2 + 4 alpha V)^{-1}$. The constant-roll condition then converts the field equation into an algebraic cubic for $dot{phi}$ in terms of $phi$, solved in closed form, which is fed into the slow-roll parameters, the sound speed $C_s^2 = (1 + xi phi^2 + alpha dot{phi}^2)/(1 + xi phi^2 + 3 alpha dot{phi}^2)$, and the power spectrum $P_S$. This is what lets the authors trade a differential slow-roll analysis for a point-by-point scan over $alpha$, $lambda$, $beta$ and $N$.

Core claim. The central claim is that the Palatini $R+R^2$ model with a quartic potential, analysed under a constant-roll condition rather than slow roll, yields inflationary observables in agreement with current data at a conventional e-fold count. Concretely, for $alpha = 10^{7}$, $lambda = 10^{-13}$ and $beta = 0.021$, the spectral tilt at $N=55$ is $n_s = 0.9681$ and the tensor-to-scalar ratio is $r = 5 times 10^{-3}$ in the minimally coupled case; varying $alpha$ upward raises $n_s$ and lowers $r$, while increasing $beta$ lowers $n_s$ quickly and leaves $r$ nearly unchanged. Adding a non-minimal coupling $xi phi^2 R$ with $xi = 10^{-6}$ shifts the tensor-to-scalar ratio up to $r approx 3 times 10^{-2}$ at $N=50$ and $2 times 10^{-2}$ at $N=60$, still inside the allowed region. The authors emphasise that the quartic kinetic term $B X^2$ is not a spectator: at $alpha = 10^{7}$, $lambda = 10^{-13}$, $beta = 0.021$ the ratio $A X / B X^2$ runs from about $10^{-2}$ at the start of inflation to order one by its end, so the higher-derivative kinetic structure genuinely shapes the dynamics. They contrast this with the slow-roll treatment of the same model, which reached agreement only at $N approx 75$.

The premise it rests on. The field velocity is obtained by truncating the equation of motion at third order in $dot{phi}$ and expanding the Hubble rate to second order in $dot{phi}^2/(2U)$, and the resulting $n_s$ and $r$ are quoted as central values without an error estimate, even though the paper's own Eq. (4.9) shows the quartic kinetic term becoming comparable to the standard one by the end of inflation.

Quartic Palatini inflation with an $R^2$ term becomes viable at $N = 50$–$60$ once constant roll with $beta approx 0.02$ replaces slow roll, removing the need for the unusually large e-fold count the slow-roll version required.

The predicted tensor-to-scalar ratio separates the two coupling cases sharply: about $5 times 10^{-3}$ for minimal coupling and about $2$–$3 times 10^{-2}$ for $xi = 10^{-6}$, both below the current bound of $0.064$.

The quartic kinetic term contributes at the same order as the standard one by the end of inflation, so $k$-inflation corrections cannot be dropped when comparing this class of models to data.

The sound speed of scalar perturbations is $C_s^2 approx 0.34$ throughout inflation in both coupling cases, rising toward one as $phi$ approaches the Planck scale.

The spectral tilt retains a dependence on $alpha$ under constant roll, whereas in the slow-roll limit it is independent of $alpha$ apart from subleading kinetic corrections.

A sound speed of $C_s^2 approx 0.34$ is far from one, so this model should generate a sizeable equilateral-type non-Gaussianity in the bispectrum; that is a sharper discriminator against ordinary slow-roll inflation than the tilt or the tensor ratio alone.

Because the tensor-to-scalar ratio differs by a factor of several between the minimal and non-minimal cases, a next-generation CMB polarization measurement sensitive to $r$ around $10^{-3}$ could separate them, or rule out the $xi = 10^{-6}$ branch.

The constant-roll parameter $beta$ is treated here as a free input; if it could be derived from the shape of the potential or from the same radiative corrections that generate the $R^2$ and $xi phi^2 R$ terms, the parameter scan would collapse to a one-parameter family with a genuine prediction.

The authors' own numerical check of the truncated velocity solution against the field equation is encouraging but is not an error propagation; repeating the $n_s$ and $r$ computation with the quartic term retained would show whether the quoted central values sit inside or outside the Planck band.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review