{"id":"7598f409-9f96-4dd6-b5d8-ff781592ab76","arxiv_id":"1908.11316","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Leading-order exponential contributions to higher-order correlators in rapid-turn inflation cancel exactly in the in-in formalism, leaving order-one non-Gaussianity and preserving perturbative control.","lead":"Rapid-turn inflation was thought to fail because its curvature perturbation grows exponentially and seems to generate hyper-large non-Gaussianity. This paper shows that the leading exponential terms cancel inside the in-in correlation functions, leaving mild non-Gaussianity and restoring perturbative control.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The tree-level cancellation is robust, but the central perturbative-control claim is conditional on an unproven property of loop corrections to ζ: no time-dependent relative phase in the mode function, which footnote 2 explicitly concedes.","rationale":"The reader's weakest assumption points to the phase structure of Eq. (3.2) as the key risk. I agree that a time-dependent relative phase would break the exact cancellation, but at tree level this structure is not an extra assumption: the linear mode equations are real, so one can choose real growing and decaying basis solutions and rotate the complex mode function to make the growing coefficient real; the constant complex phase of the decaying coefficient does not affect the cancellation because the real part can be absorbed into a redefined f(τ). The four-point e^{8x} terms cancel exactly even with a nonzero real part in the decaying mode, since the two orderings in Eq. (3.1) give complex-conjugate products for the H2-A contractions and identical products for the rest. The general-n nested-commutator argument is terse but consistent with the tree-level contraction counting. The real soft spot is the paper's extension to loops: footnote 2 admits that time-dependent phases could appear in loop corrections to ζ, and the end of Sec. 3.2 gives only a heuristic power-counting argument for their suppression, not a proof that the phase structure is preserved. Since the tree-level cancellation is exact and relies on the mode-function decomposition, any loop-induced time-dependent phase of non-exponentially-small size would reintroduce an e^{10x} contribution and make gNL exponentially large. The WKB growth formula in Sec. 4 is well supported by numerics, and the tree-level cancellation itself appears sound, but the central perturbative-control claim extends beyond what is demonstrated. The reader's CONDITIONAL verdict is therefore appropriate; my read does not change it.","tokens_in":15567,"tokens_out":44735,"duration_ms":455898,"concrete_test":"Compute the one-loop correction to the ζ mode function in the EFT of [12], or at next-to-leading order in the WKB scheme of Sec. 4 in the full two-field model. Decompose the corrected mode function as u(τ)=A(τ)e^{I(τ)}+B(τ)e^{-I(τ)} with e^{±I} the real WKB basis, and determine whether the ratio B(τ)/A(τ) is time-independent. If it is, evaluate the four-point integrand in Eq. (3.1) with the corrected mode functions and verify that the residual grows as e^{6x}; if it is not, compute the residual and check whether it contains e^{8x} or e^{10x} terms, which would imply gNL∼e^{2x} or e^{4x} and would invalidate the perturbative-control claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The exact cancellation in Sec. 3 is derived for mode functions of the form (3.2), and footnote 2 explicitly limits its validity: time-dependent phases in one of the two terms are 'not considered' and 'could show up in loop corrections to ζ'. At tree level this limitation is harmless, because the quadratic equations (4.1)-(4.2) are real linear ODEs; one can rotate the complex mode function so that the growing-basis coefficient is real, and the decaying coefficient has a constant complex ratio. The real part of the decaying coefficient can be absorbed into f(τ) without changing the scaling argument, so the e^{10x} and e^{8x} cancellations in Eq. (3.1) are exact at leading order. The gap is the claimed extension to loops: a one-loop correction δu to ζ is not guaranteed to preserve this phase structure, and if it introduces a time-dependent relative phase θ(τ), the cancellation of the e^{10x} and e^{8x} terms is no longer exact. Because the uncancelled four-point amplitude would scale as α^6 e^{10x}, even a correction of relative size λ that is not exponentially small leaves gNL∼λ e^{4x}, which for x∼ω∼90 is astronomically large. The paper's loop discussion at the end of Sec. 3.2 is only a power-counting heuristic and does not control this phase issue; the footnote concedes the possibility. Thus the central claim of no loss of perturbative control is conditional on a property that is unproven beyond leading order.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies rapid-turn inflation models in negatively curved field spaces, where the curvature perturbation ζ undergoes a transient exponential growth before horizon crossing. The authors first provide an analytic WKB solution for the mode functions in the full two-field theory, obtaining the growth exponent x = (2−√(3+ξ))πω/2 and showing that it matches existing numerical results and the single-field EFT with imaginary speed of sound. They then address the previously claimed exponential enhancement of higher-order correlators. Using the in-in formalism, they argue that the nested commutator structure causes the naively leading exponentially large contributions to cancel: for the four-point function the e^{10x} and e^{8x} terms drop out (Eqs. 3.1–3.5), giving gNL ∼ 1, and for general n the connected ratio (3.14) is shown to be ∼1. The paper concludes that these models do not lose perturbative control and remain observationally viable.","tokens_in":15829,"tokens_out":4791,"duration_ms":39539,"significance":"If the conclusions hold, the paper resolves an apparent contradiction between rapid-turn inflation and observational bounds on non-Gaussianities, and it provides a useful analytic handle (the WKB growth formula) for a class of models that is otherwise studied numerically. The tree-level four-point cancellation is demonstrated explicitly with concrete mode-function scaling, and the WKB result is benchmarked against independent numerics in Figure 2, with quantitative agreement. The general-n argument is less complete, and the extension to loop corrections is only heuristic; this is the main gap in support of the paper's central perturbative-control claim.","major_comments":[{"comment":"The cancellation proof relies on the mode-function decomposition ζ(τ) = f(τ) e^x + i g(τ) e^{-x} with f, g real and no time-dependent relative phase. Footnote 2 explicitly concedes that time-dependent phases in one of the two terms are not considered and 'could show up in loop corrections to ζ'. The subsequent loop discussion at the end of Sec. 3.2 is only a power-counting heuristic and does not control this possibility. If a one-loop correction to ζ introduces a time-dependent relative phase θ(τ), the leading e^{10x} and e^{8x} cancellations in Eq. (3.1) are no longer exact; the uncancelled four-point amplitude would scale as α^6 e^{10x}, yielding gNL ∼ λ e^{4x}, which for x ∼ ω ∼ 90 is astronomically large. Since the abstract and Sec. 5 assert that there is 'no problem with perturbative control', the central claim is currently conditional on an unproven property of loop corrections. The manuscript should either prove the absence of such phases at loop level or restrict the no-loss-of-control claim to tree level.","section":"Sec. 3.2, Eq. (3.2), footnote 2"},{"comment":"The general-n cancellation is established by an iterative argument rather than a complete derivation. In particular, the claim that every non-zero term in an n-nested commutator must contain contractions across each commutator, and that each such commutator contributes a factor e^{-2x} to the scaling, is stated without a fully rigorous combinatorial treatment when the interaction Hamiltonian contains derivatives (as in Eq. (3.3)) and when multiple operators within the same H_int may be contracted with each other. The four-point example is explicit, but the extension to arbitrary n rests on a schematic argument (Eqs. 3.9–3.14) that would benefit from a complete proof or a clearly stated conjecture with supporting evidence.","section":"Sec. 3.2, after Eq. (3.13)"}],"minor_comments":[{"comment":"The mass term is written as '−H^2ω^2(ξ−1)σ2'; this should presumably be '−H^2ω^2(ξ−1)σ^2' for consistency with the rest of the equation.","section":"Sec. 2.2, Eq. (2.15)"},{"comment":"There is a typo: 'for simplicitly' should be 'for simplicity'.","section":"Sec. 3.1, before Eq. (3.2)"},{"comment":"The notation f^{(n−2)}_{NL} is introduced without definition; clarifying the placement of the superscript relative to the NL subscript (e.g., f_{NL}^{(n−2)}) would help the reader track the standard hierarchy of non-Gaussian shapes.","section":"Sec. 2.3, Eq. (2.19)"},{"comment":"The bound (4.13) uses a minimal reheating temperature T_min from BBN but does not explain how H_min is computed from T_min. A sentence with the standard relation H_min ∝ T_min^2/M_Pl and the relevant prefactor would make the estimate reproducible.","section":"Sec. 4, after Eq. (4.13)"}],"recommendation":"major_revision","confidential_remarks":"The tree-level cancellation and the WKB growth formula are valuable and likely correct; the main obstacle to acceptance is the gap between the tree-level proof and the claimed all-order perturbative control, which the authors themselves flag in footnote 2. If the authors can either close that gap or carefully restate the claim, I would be supportive of publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Let me get straight to the point. This paper gives the first analytic handle on the growth of ζ in rapid-turn inflation with ξ<1 and, more importantly, argues that the exponentially large non-Gaussianities found in [12] cancel at leading order because of the nested commutator structure of the in-in correlators. The WKB computation in Section 4 is real work: the growth integral (4.8)-(4.9) is evaluated in closed form, the resulting amplitude x(ω,ξ)=(2−√(3+ξ))πω/2 is new, and it matches the numerics of Mizuno & Mukohyama to better than a percent (lnγ∝0.920ω vs 0.924ω). That alone is worth a cite. The four-point cancellation in Section 3.1 is explicit and clean: the e^{10x} pieces are real symmetric products and the two orderings in the commutator cancel; only e^{6x} survives. The analogy to the occupation-number cancellation in gauge-field inflation is apt and makes the physics believable.\n\nSoft spots, in proportion. The general-n argument in Section 3.2 is iterative and schematic, not a full derivation; the claim that every surviving term contains at least one cross-contraction per commutator is plausible but is asserted more than proven. The bigger gap is loop corrections. The authors restrict the mode function to ζ = f(τ)e^x + i g(τ)e^{-x} with no time-dependent relative phase, and footnote 2 concedes this. The stress-test note is right that this is harmless at tree level — one can rotate the growing-basis coefficient real and absorb the constant ratio into f(τ) — but the same is not shown for loop corrections to ζ. The loop discussion at the end of Section 3.2 is a power-counting heuristic; it does not control a phase that slips in through a one-loop correction. That is a specific, named gap, not a demonstrated failure.\n\nI would send this to a serious referee. The central tree-level claim probably survives, and the WKB result is solid. Ask the referee to focus on the general-n commutator proof and the loop phase issue, and to verify the matched normalisation between α, e^{2x} and P_obs used in Eq. (2.17). The citation pattern is fine: the self-citations to [3,4] are the source of the attractor background, which is appropriate.\n\nWho reads this: anyone working on non-geodesic multi-field inflation, effective field theories with imaginary sound speed, or the validity of perturbative inflation. It deserves a referee rather than a desk rejection.","headline":"A solid, genuinely useful paper: it supplies the first analytic growth formula for rapid-turn inflation and shows the leading exponential non-Gaussianities cancel at tree level, but the loop-level extension is a real, self-admitted gap.","tokens_in":16394,"tokens_out":2008,"would_cite":true,"duration_ms":18906,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Rapid-turn inflation survives: exponentially large non-Gaussianities cancel exactly once nested commutators are accounted for.","keywords":["rapid-turn inflation","hyperinflation","non-Gaussianity","in-in formalism","nested commutators","imaginary speed of sound","trispectrum","perturbative control"],"falsifier":"Compute the tree-level four-point function in the full two-field theory beyond the leading WKB approximation, using the exact numerical mode functions, and check whether $g_{NL}$ scales as $e^{4x}$; if a time-dependent relative phase appears in the mode functions, the leading $e^{10x}$ term survives and the scaling would be visible. Equivalently, a direct numerical in-in evaluation of the trispectrum for a representative rapid-turn hyperinflation model would settle whether the cancellation is exact.","tokens_in":15339,"feed_emoji":"🌌","tokens_out":11351,"duration_ms":93730,"temperature":0.7,"pith_summary":"This paper defends a class of inflationary models---rapid-turn inflation in negatively curved field spaces---against a recent argument that they produce 'hyper-large' non-Gaussianities and lose perturbative control. The authors claim that the apparently dominant, exponentially large contributions to the trispectrum and higher-point correlators cancel exactly once the nested commutators of the in-in formalism are handled correctly. After the cancellation, the connected $n$-point functions satisfy $\\langle\\zeta^n\\rangle_c/\\langle\\zeta^2\\rangle^{n-1}\\sim 1$, so $g_{NL}\\sim 1$ rather than $e^{4x}$ with the growth factor $x$, and perturbative control is maintained. The paper also supplies the first analytic WKB solution for the growth of perturbations in two-field rapid-turn models, which matches numerical results and identifies the exponent $x=(2-\\sqrt{3+\\xi})\\pi\\omega/2$. If correct, rapid-turn inflation remains an observationally viable candidate theory of the early universe.","feed_headline":"Rapid-turn inflation is safe: exponential non-Gaussianities cancel","feed_subtitle":"Nested commutators remove the exponential growth, leaving gNL ~ 1 and full perturbative control.","key_machinery":"The carrying mechanism is the nested-commutator structure of the in-in (closed-time-path) formalism for correlators, together with a two-component mode-function ansatz. In the in-in expression for an $n$-point function, any nonzero term must have at least one operator on each commutator's left side Wick-contracted with an operator on its right side; terms without such cross-contractions vanish. A cross-contraction brings in the imaginary part of a product of mode functions, and because the mode functions are written as $\\zeta_i(\\tau)=f_i(\\tau)e^x+i g_i(\\tau)e^{-x}$ with real $f_i,g_i$, each imaginary part carries one power of the decaying component, $\\sim e^{-x}$, instead of the growing $e^x$. The leading $e^{(4n-6)x}$ pieces in the two orderings of each commutator are therefore equal real quantities and cancel exactly, leaving the suppressed imaginary pieces and yielding $\\langle\\zeta^n\\rangle_c/\\langle\\zeta^2\\rangle^{n-1}\\sim 1$. A secondary piece of machinery is the WKB evaluation of the two-field mode-function integral, which yields the closed-form growth exponent used to fix the normalisation and bound the turn rate.","core_discovery":"The central discovery is that in rapid-turn inflation with $\\xi<1$---the regime where the entropic mass lies below its critical value---the curvature perturbation $\\zeta$ undergoes transient exponential growth near horizon crossing, characterised by a large parameter $x$, yet the higher-order correlators are not exponentially enhanced. Writing the mode functions as $\\zeta_i(\\tau)=f_i(\\tau)e^x+i g_i(\\tau)e^{-x}$ with real $f_i,g_i$, the naively leading $e^{10x}$ term in the four-point function from two cubic-interaction insertions is the real part of identical products of mode functions in the two orderings of each commutator, and it cancels exactly. Each nested commutator forces at least one Wick contraction between its left and right sides, producing a factor of the imaginary part of a product of mode functions, which scales as $e^{-x}$; with $n-2$ insertions this turns the naive $\\alpha^{2n-2}e^{(4n-6)x}$ into $\\alpha^{2n-2}e^{(2n-2)x}$, matching the denominator $\\langle\\zeta^2\\rangle^{n-1}$ so that the ratio is of order one. The paper states this explicitly for the four-point function ($g_{NL}\\sim 1$) and for the general $n$-point correlator. In addition, a WKB computation of the two-field linear system gives the analytic growth $\\ln\\gamma^2\\approx(2-\\sqrt{3+\\xi})\\pi\\omega$, in good agreement with numerical results and consistent with the imaginary-speed-of-sound effective field theory.","pith_inferences":["The cancellation mechanism is generic: any inflationary model whose mode functions are a growing-plus-decaying sum with fixed relative phase should show the same suppression, so estimates of non-Gaussianity that ignore commutator nesting will systematically overestimate the signal in transient-instability models.","A natural next step is a numerical in-in computation of the tree-level trispectrum in the full two-field theory, without the effective single-field description, to verify that $g_{NL}$ does not grow as $e^{4x}$; this would test the mode-function ansatz beyond leading WKB order.","If the cancellation persists at loop level, as the paper's power counting suggests, then the practical constraint on rapid-turn models shifts from non-Gaussianity to the power-spectrum amplitude and reheating, which already gives $\\omega\\lesssim 96$ in hyperinflation."],"forward_implications":["The connected four-point function is not exponentially amplified: $g_{NL}\\sim 1$, far below current constraints $g_{NL}\\lesssim 10^4\\text{--}10^6$.","For every $n$, the ratio $\\langle\\zeta^n\\rangle_c/\\langle\\zeta^2\\rangle^{n-1}$ is of order one, so the perturbative expansion in $\\zeta$ is under control despite the exponential amplification of the power-spectrum normalisation factor.","Each insertion of the cubic interaction contributes a factor of order $\\alpha e^x\\simeq \\sqrt{P_\\zeta}\\ll 1$, so loop corrections are not expected to reintroduce exponential enhancement.","The analytic WKB solution gives the first closed-form expression for the perturbation growth in the two-field rapid-turn class and identifies the parameter $x=(2-\\sqrt{3+\\xi})\\pi\\omega/2$.","Combining the growth formula with power-spectrum normalisation and reheating requirements bounds the turn rate, e.g. $\\omega\\lesssim 96$ for hyperinflation with $\\xi=-1$, leaving a large viable parameter space."],"supporting_citations":[{"why":"supplies the single-field EFT estimates of exponentially large trispectrum and higher-order non-Gaussianities that this paper corrects.","marker":"[12]"},{"why":"derives the imaginary-speed-of-sound effective theory and the mode-function form used in the commutator argument.","marker":"[10]"},{"why":"established the analogous exact cancellation for axially coupled gauge fields during inflation, motivating the analysis.","marker":"[36]"},{"why":"provides the numerical growth results that the analytic WKB formula is checked against.","marker":"[2]"},{"why":"introduced hyperinflation and the exponential growth of perturbations in hyperbolic field space.","marker":"[1]"},{"why":"defines the general rapid-turn attractor class that the perturbation theory applies to.","marker":"[4]"}],"fun_headline_variants":["Rapid-turn inflation: exponential non-Gaussianities cancel exactly","Nested commutators cancel exponential growth in rapid-turn inflation","Steep potentials ok: rapid-turn inflation passes non-Gaussian test","Exponential growth cancels: rapid-turn inflation remains viable"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that each mode function is exactly a growing real piece plus a decaying imaginary piece with no time-dependent relative phase between them; if that relative phase rotates in time, the leading exponential terms in the two commutator orderings would no longer coincide and the exact cancellation would fail.","fun_headline_variants_meta":{"raw":{"variants":["Rapid-turn inflation: exponential non-Gaussianities cancel exactly","Nested commutators cancel exponential growth in rapid-turn inflation","Steep potentials ok: rapid-turn inflation passes non-Gaussian test","Exponential growth cancels: rapid-turn inflation remains viable"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000385,"raw_usage":{"total_tokens":2132,"prompt_tokens":1137,"completion_tokens":995,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":753,"completion_tokens_details":{"reasoning_tokens":925}},"tokens_in":753,"tokens_out":995,"duration_ms":8546,"temperature":1.0,"reasoning_tokens":925,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:18:24.993427+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the tree-level four-point function in the full two-field theory beyond the leading WKB approximation, using the exact numerical mode functions, and check whether $g_{NL}$ scales as $e^{4x}$; if a time-dependent relative phase appears in the mode functions, the leading $e^{10x}$ term survives and the scaling would be visible. Equivalently, a direct numerical in-in evaluation of the trispectrum for a representative rapid-turn hyperinflation model would settle whether the cancellation is exact.","supporting_citations":[],"review_version":1}