{"id":"379fb750-7650-4c45-9e23-21477438c7d4","arxiv_id":"2505.06526","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For every c≥1, the 1D nonlinear Klein-Gordon equation with a convolution potential and small cubic nonlinearity admits linearly stable full-dimensional KAM tori with action decay e^{-2r ln^σ |n|}.","lead":"This paper proves that the one-dimensional nonlinear Klein-Gordon equation in the non-relativistic limit, with periodic boundary conditions and a small cubic nonlinearity, has invariant tori of full dimension whose amplitudes decay slowly with the mode number. The result extends a known KAM construction from the Schrödinger equation to the Klein-Gordon equation and gives a concrete step toward a question raised in a 2025 paper on the non-relativistic limit.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Frequency space Π_c in Eq. (6) is not the image of V under λ_n, so Theorem 1.4 as stated is false for frequencies near the upper endpoint.","rationale":"The reader's weakest_assumption identifies precisely the same load-bearing flaw: the frequency space Π_c in Eq. (6) is wider than the image of the parameter space V under λ_n, so the inversion T_n=ω_n^2/c^2-c^2-n^2 can fall outside [0,1]. This is an internal inconsistency in the theorem statement, not a disagreement with consensus. The proof in Section 3.2 explicitly chooses V^*_0=T and relies on eV_0=id, so if T∉V the KAM parameter can never land in the allowed set V, and the assertion 'there exists V∈V' fails. The concrete test with c=1,n=10 shows the failure numerically, settling the concern. Because the intended result is plausibly repairable by replacing Π_c with the actual image (or enlarging V), the conditional verdict is appropriate rather than outright rejection. A secondary issue appears in Lemma 5.3, where the bound c^3/√(c^2+n_1^2)+c^3/√(c^2+n_2^2)≤c/|n_2| is false for c>1; this further indicates that the manuscript needs revision before the proof is complete, but it is not the primary obstruction to the theorem as stated.","tokens_in":24221,"tokens_out":9095,"duration_ms":86103,"concrete_test":"Take c=1 and n=10. The upper endpoint in (6) is U_n=1/(√101+1)≈0.0905, so choose r_n=U_n and ω_n=√101+r_n. Then T_n=ω_n^2-1-10^2 = 2r_n√101+r_n^2 ≈ 1.827 > 1. Since no V_10∈[0,1] satisfies 1·√(1+10^2+V_10)=ω_10, this ω lies in Π_c but is not the image of any V∈V, directly contradicting the assertion after (6) and the existence part of Theorem 1.4.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that every ω∈Π_c be realizable as λ_n(V)=c√(c^2+n^2+V_n) for some V∈V with V_n∈[0,1]. The definition (6) asserts exactly this, but the assertion is false. For fixed c≥1 and large n, the upper endpoint in (6) is U_n=c^3/(√(c^2+n^2)+1)≈c^3/n. The actual range of λ_n(V)-c√(c^2+n^2) as V_n runs over [0,1] is [0, c(√(c^2+n^2+1)-√(c^2+n^2))]=[0, c/(√(c^2+n^2+1)+√(c^2+n^2))]≈[0,c/(2n)]. The ratio is about 2c^2. Consequently the quantity T_n=ω_n^2/c^2-c^2-n^2 used in (58) can be as large as ≈2c^2 at the upper endpoint (for c=1, T_n≈2), not confined to [0,1]. Since Section 3.2 sets V^*_0=T and eV_0=id, the inversion condition (28) then produces V^* outside V. Thus the literal statement of Theorem 1.4, 'there exists V∈V for every ω∈Π_c\\R', collapses for frequencies with r_n near the upper endpoint. This is not a missing estimate; the parameterization map simply does not cover Π_c.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies almost-periodic solutions of the one-dimensional nonlinear Klein-Gordon equation (2) on [-π,π] with periodic boundary conditions in the non-relativistic limit c≥1. The main theorem (Theorem 1.4) claims that for every 2<σ≤3, r>1, γ′>0 and c≥1, outside a small-measure set R⊂Π_c of frequencies there is an invariant torus for (2) with prescribed frequencies ω, amplitudes comparable to e^{-2r ln^σ⌊n⌋}, and linear stability. The proof follows Bourgain's full-dimensional KAM scheme: a weighted norm for Hamiltonians (§2), an abstract iterative KAM lemma (§3), application to NLKG via the [FM23] transformation (§4), and measure estimates for the nonresonance conditions (§5). Several key lemmas are adapted from the author's earlier papers [Con24] and [CY21].","tokens_in":24560,"tokens_out":17809,"duration_ms":157208,"significance":"If correct, the result would be a substantial extension of the full-dimensional KAM literature: it unifies the NLS and NLW regimes of NLKG in the singular limit c→∞, handles the double eigenvalues of periodic boundary conditions via the momentum condition, and achieves subexponential decay σ≤3 for every c≥1. The paper is largely self-contained in its KAM iteration structure and gives quantitative bounds (e.g., estimates (32)-(36)). However, the correctness of the stated theorem depends on two issues in the printed manuscript: the frequency box Π_c in (6) is not the image of the parameter space V, and the coefficient bound (68) appears to have the wrong scaling in c. Both are fixable locally, but as printed the proof does not establish Theorem 1.4.","major_comments":[{"comment":"The frequency space Π_c defined in (6) is not the image of V={V_n∈[0,1]} under λ_n(V)=c√(c²+n²+V_n). For each n, the reachable range of λ_n-c√(c²+n²) has width c(√(c²+n²+1)-√(c²+n²)) ~ c/(2n), while the interval in (6) has width c³/(√(c²+n²)+1) ~ c³/n; the ratio is about 2c². Hence the sentence after (6) asserting the existence of V for every ω∈Π_c is false. In particular, for ω near the upper endpoint, T_n=ω_n²/c²-c²-n² in (58) exceeds 1 (for c=1 and large n it is about 2), so the initialization V*_0=T in §3.2 is not admissible, and the inverse-function theorem argument in (59)-(60) can produce a limit V* outside V. The literal statement of Theorem 1.4 therefore collapses for frequencies in the unused upper part of Π_c. The fix is to replace the upper endpoint in (6) by c(√(c²+n²+1)-√(c²+n²)) (or a fixed universal multiple of it) and to re-check the measure estimates in §5, which currently use the oversized endpoint, e.g., in (71) and (85). Since the corrected interval is smaller, the measure estimates should improve rather than fail.","section":"§1.2, Eq. (6); §3.2, Eq. (58)"},{"comment":"As printed, the coefficient bound (68) reads |R_{n1...n4}^{σ1...σ4}| ≤ Cε (c√(c²+n_1²)···c√(c²+n_4²))^{1/2}. This grows with n and c; for the monomial with n_1=...=n_4=0 the bound is ~ε c⁴, so the claim ∥R∥_{ρ0}≤Cε at the start of §4.2 fails for large c, contrary to the text's assertion in §2 that the coefficients decay with c. The intended expression is presumably (c/√(c²+n_i²))^{1/2}, which is the eigenvalue of D^{-1/2} with D as defined in (64), and which makes the product with the factor ⟨n/c⟩^{1/2} in the norm (14) exactly 1. The definition of d_n just above (68) has the same missing-fraction ambiguity. Please correct the displayed formulas and verify that the final Hamiltonian norm is indeed uniformly bounded in c.","section":"§4.1, Eq. (68); §4.2"}],"minor_comments":[{"comment":"The word 'samll' appears in Lemmas 5.1, 5.2, and 5.3; it should be 'small'.","section":"§5"},{"comment":"Theorem 1.4 states meas R=O(γ′), while Lemma 5.1 proves meas R=O(γ^{1/3}); the relation between γ and γ′ should be stated explicitly.","section":"§1.2 / §5"},{"comment":"The symbol V is used both for the convolution potential in (2) and for the parameter space in (5), which is confusing in (63)-(67); a different letter for one of them would help.","section":"§4.1"},{"comment":"The iteration parameters η_s and λ_s are introduced with an ambiguous formula (η_{s+1}=1/(20λ_s)η_s versus η_{s+1}=(1/20)λ_sη_s), and the displayed product after (56) should be checked for consistency with the chosen definition.","section":"§3.1"},{"comment":"The proof of Lemma 5.1 uses the notation n*_i(ℓ) for a decreasing rearrangement of a multiset; a formal definition of the multiset (including the multiplicity encoded by ℓ_n) would improve readability.","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"The two major issues are concrete and local: one is an incorrect frequency/parameter interval in the statement of the theorem, and the other is a scaling error (likely typographical) in the Hamiltonian application. Both are fixable without changing the overall strategy. I would ask for a corrected version with the interval in (6) replaced by the actual image of V and with the d_n/(68) formulas checked, and for a re-verification that the measure estimates in §5 go through with the corrected interval. The heavy dependence on [Con24] and [CY21] is appropriate since those are published papers, but the manuscript should make the adaptations explicit in the deferred proofs."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The paper claims the first construction of full-dimensional KAM tori with subexponential decay for the 1D NLKG equation in the non-relativistic limit, for every c≥1. That is a worthwhile target, and the extension of Cong's NLS scheme to NLKG is nontrivial: the frequencies grow linearly rather than quadratically, so the separation property is lost, and periodic boundary conditions make every eigenvalue double. Using the [FM23] momentum condition to kill the double-degenerate resonant terms is a genuine idea. The KAM iteration lemma is laid out in enough detail that the overall structure is credible.\n\nHowever, there is a load-bearing problem. The frequency space Π_c in (6) is not the image of the parameter space V. For large |n|, the upper endpoint in (6) is about c³/n, while the actual reachable interval for λ_n(V)-c√(c²+n²) with V_n∈[0,1] is about c/(2n). The ratio is roughly 2c², so a large part of Π_c — in particular all frequencies with r_n near the upper endpoint — cannot be written as λ_n(V) for any V∈V. The text right after (6) says the definition ensures this, but it does not. Concretely, T_n=ω_n²/c²-c²-n² in (58) can exceed 1, and Section 3.2 initializes V*_0=T. If T_n>1, V*_0 is outside the box V, and the inverse function step produces a parameter outside the allowed set. As stated, Theorem 1.4 is false.\n\nIs this the end of the paper? No. It is a definitional mismatch, not a missing estimate. Shrinking Π_c to the true image of V would fix the statement, and the measure estimates in Section 5 would only get easier on a smaller set. The proof structure seems sound; the main burden is that several key lemmas (2.4, 2.10, Claim 3.2, parts of Section 5) are delegated to [Con24] and [CY21]. Those are published and checkable, so the division of labor is fair, but a referee should verify the c-dependence in the adaptations. The heavy self-citation is not a red flag here; those results are used as tools.\n\nBottom line: this paper deserves a serious referee, but the referee should be sent in with a specific request to fix the definition of Π_c first. If that is corrected, the result is a solid, publishable extension. I would not cite Theorem 1.4 in its current form.","headline":"First NLKG full-dimensional KAM tori claim, but the frequency space Π_c in (6) is not the image of the parameter space, so Theorem 1.4 as stated is false.","tokens_in":25102,"tokens_out":4074,"would_cite":false,"duration_ms":36706,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37K55","35B15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the one-dimensional nonlinear Klein-Gordon equation in the non-relativistic limit admits full-dimensional, linearly stable KAM tori with subexponential amplitude decay for every speed of light c≥1.","keywords":["full-dimensional KAM tori","almost-periodic solutions","nonlinear Klein-Gordon equation","non-relativistic limit","subexponential decay","linear stability","KAM theory","zero momentum condition"],"falsifier":"Compare the reachable width max_{v∈[0,1]} c√($c^{2}$+$n^{2}$+v) - c√($c^{2}$+$n^{2}$) ≈ c/(2√($c^{2}$+$n^{2}$)) with the window width in (6), $c^{3}$/(√($c^{2}$+$n^{2}$)+1). For c=10, n=1000, the reachable width is about 0.005 while the window width is about 1; checking whether T_n = $ω_n^{2}$/$c^{2}$ - $c^{2}$ - $n^{2}$ stays in [0,1] for the constructed V^* would settle whether the claimed parameter V actually lies in V.","tokens_in":24038,"feed_emoji":"🌀","tokens_out":8318,"duration_ms":76635,"temperature":0.7,"pith_summary":"The paper claims that the one-dimensional nonlinear Klein-Gordon equation, in the non-relativistic limit and with periodic boundary conditions, possesses full-dimensional invariant tori whose Fourier amplitudes decay subexponentially, for every speed of light c≥1. The tori survive for a large set of frequency parameters and are linearly stable. If correct, this establishes the existence of almost-periodic solutions with slow decay for the NLKG equation across the whole non-relativistic range, answering a question raised in [BBG25]. The proof works by adapting the full-dimensional KAM scheme of [Bou05b] to the linearly growing frequencies of NLKG, using a weighted norm that tames the dependence on c and a coordinate transformation that enforces the zero-momentum condition.","feed_headline":"Full-dimensional KAM tori exist for 1D Klein-Gordon for all c","feed_subtitle":"The proof excites every Fourier mode with subexponential decay and keeps the tori linearly stable in the non-relativistic limit.","key_machinery":"The central mechanism is a KAM iteration lemma (Lemma 3.1) that, at each step, eliminates the non-normal-form terms R0 and R1 by solving a homological equation, then controls the new error R2. Three objects carry the argument. First, the frequency window Π_c, defined by ω_n - c√($c^{2}$+$n^{2}$) ∈ [0, $c^{3}$/(√($c^{2}$+$n^{2}$)+1)], is the parameter space for nonresonance conditions whose right-hand side depends only on the three largest active indices. Second, the weighted ℓ^∞ Hamiltonian norms include the factor ⟨n/c⟩ = √(1+$n^{2}$/$c^{2}$), which makes the frequency shift small in the non-relativistic limit and compensates for the lack of a 1/n saving. Third, the zero-momentum condition (13), inherited from the transformation in [FM23], eliminates resonant monomials q_n\\bar q_{-n} that would otherwise break linear stability under periodic boundary conditions.","core_discovery":"On its own terms, the paper establishes Theorem 1.4: for any 2<σ≤3, r>1, γ'>0, and any c≥1, there is a set R⊂Π_c of measure O(γ') such that for every frequency ω∈Π_c\\R there is a parameter V∈V and a small nonlinearity strength ϵ for which the NLKG equation has an invariant torus E whose amplitudes satisfy (1/4)$e^{{-2r\\ln^\\sigma\\lfloor n\\rfloor}}$ ≤ |I_n| ≤ $4e^{{-2r\\ln^\\sigma\\lfloor n\\rfloor}}$, whose frequencies are exactly ω, and which is linearly stable. The torus is full-dimensional in the sense that every Fourier mode is excited, not just finitely many; the decay is subexponential, slower than super-exponential but faster than any polynomial.","pith_inferences":["Beyond the paper: the same KAM scheme should produce a Cantor family of full-dimensional tori parameterized by the residual frequencies, since the iteration is uniform in the parameter V.","Beyond the paper: the c-uniform measure estimates suggest the result persists under small analytic perturbations f with a zero of order at least three, provided the zero-momentum structure is preserved.","Beyond the paper: replacing the convolution potential by a different finite-rank perturbation may break the ⟨n/c⟩ frequency-shift compensation, so the method is tuned to the non-relativistic scaling rather than generic potentials.","Beyond the paper: a direct comparison between the width of Π_c and the image of the parameter cube V would settle whether Theorem 1.4 needs a smaller frequency window or a larger parameter range."],"forward_implications":["For every c≥1 there is a set of frequencies of positive measure for which equation (2) admits full-dimensional, linearly stable invariant tori with the prescribed frequencies.","The amplitude decay rate e^{-2r\\ln^\\sigma\\lfloor n\\rfloor}, with 2<σ≤3, is slower than any super-exponential decay, so the result gives maximal tori in Gevrey-type spaces rather than only analytic or Sobolev tori.","The iterative construction is uniform in c, so the same KAM scheme works across the whole non-relativistic range from c=1 to c=∞.","The exceptional frequency set R has measure O(γ'), so the surviving frequencies form a large Cantor-like set inside Π_c.","Linear stability follows from the fact that the final normal form is of order two around the torus, making the tori elliptic."],"supporting_citations":[{"why":"Supplies the full-dimensional KAM method for 1D NLS: exciting all modes, the nonresonance conditions, and the approximate-invariant-manifold construction that this paper adapts to NLKG.","marker":"[Bou05b]"},{"why":"Provides the subexponential-decay KAM scheme for NLS that is extended here; the weighted norms, Poisson-bracket estimates, and combinatorial lemmas are taken from it.","marker":"[Con24]"},{"why":"Gives the KAM framework for the 1D nonlinear wave equation with linearly growing frequencies, including norm properties and the measure-estimate subcases reused in Section 5.","marker":"[CY21]"},{"why":"Establishes quasi-periodic KAM tori for NLKG in the non-relativistic limit and poses the almost-periodic extension problem that Theorem 1.4 addresses.","marker":"[BBG25]"},{"why":"Provides the coordinate transformation that rewrites the NLKG Hamiltonian so that the perturbation satisfies the zero-momentum condition (13), removing the double-root obstruction.","marker":"[FM23]"},{"why":"Documents the difficulty that all frequencies are double under periodic boundary conditions, the obstruction that this paper's momentum-condition argument is designed to bypass.","marker":"[CY00]"}],"fun_headline_variants":["Full-dim KAM tori exist for all c in 1D NLKG","Subexponential full-dim tori: stable for all c","Every mode excited: full-dim KAM tori for all c","Full-dim tori, subexponential decay, for all c"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof collapses if the frequency window in (6) is not exactly the image of the parameter cube V_n∈[0,1], since the window as written is about $c^{3}$/n wide while the parameters only reach about c/(2n).","fun_headline_variants_meta":{"raw":{"variants":["Full-dim KAM tori exist for all c in 1D NLKG","Subexponential full-dim tori: stable for all c","Every mode excited: full-dim KAM tori for all c","Full-dim tori, subexponential decay, for all c"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000932,"raw_usage":{"total_tokens":3908,"prompt_tokens":783,"completion_tokens":3125,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":399,"completion_tokens_details":{"reasoning_tokens":3048}},"tokens_in":399,"tokens_out":3125,"duration_ms":21392,"temperature":1.0,"reasoning_tokens":3048,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:42:19.370066+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the reachable width max_{v∈[0,1]} c√($c^{2}$+$n^{2}$+v) - c√($c^{2}$+$n^{2}$) ≈ c/(2√($c^{2}$+$n^{2}$)) with the window width in (6), $c^{3}$/(√($c^{2}$+$n^{2}$)+1). For c=10, n=1000, the reachable width is about 0.005 while the window width is about 1; checking whether T_n = $ω_n^{2}$/$c^{2}$ - $c^{2}$ - $n^{2}$ stays in [0,1] for the constructed V^* would settle whether the claimed parameter V actually lies in V.","supporting_citations":[],"review_version":1}