{"id":"51a48bb9-6c03-476b-8371-b229a0875210","arxiv_id":"2411.13312","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A complete Birkhoff normal form theorem on the torus gives polynomial and sub-exponential stability for nonlinear wave and Schrödinger equations with Sobolev index s = O(rτ), improving known bounds.","lead":"This paper proves an abstract Birkhoff normal form theorem for Hamiltonian partial differential equations on the torus, removing all non-resonant terms up to an arbitrary finite order. It applies the theorem to the 1D nonlinear wave equation and to high-dimensional nonlinear Schrödinger equations, yielding longer stability times and lower Sobolev regularity requirements than earlier work.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.6's homological-equation estimate is algebraically false: the norm shift s→s+τ does not absorb the small-divisor denominator, so the iterative bounds (3.14)–(3.16) are unsupported.","rationale":"The reader's conditional verdict identifies the measure estimates in Lemma 4.2 as the weakest point and notes an invalid inequality there. That inequality is indeed a typo (an exponent should be 4τ/(21l) rather than τ/(21l)), but it only makes the displayed bound stronger than the algebraically correct one, and the final summability estimate survives. The truly load-bearing failure is earlier, in Lemma 3.6, which is used at every step of the iterative construction. The proof of (3.12) omits the unavoidable factor (j*_2...j*_l)^{2τ}; this factor is unbounded in the Fourier modes and is not controlled by the paper's Sobolev-shift device. Since the whole theorem, including the applications, depends on this estimate, the central claim is not established by the manuscript as written. My recommendation is therefore UNVERDICTED rather than CONDITIONAL: the flaw is not a minor correction to constants but a false inequality in the engine of the proof. The concrete single-monomial computation is a decisive, low-cost check that would settle whether the estimate can be repaired or the theorem needs a different norm/non-resonance condition.","tokens_in":31078,"tokens_out":35202,"duration_ms":338577,"concrete_test":"Verify Lemma 3.6 analytically on a single monomial: let f(z)=z_a\\bar z_{-a}z_1, with ω_a=√(a^2+m), m∈[1,2], and solve {H0,χ}+Z=f. For the non-resonant mode, compute the ratio of the contribution to |χ|_{s+τ,N} over that to |f|_{s,N}; it is 2^τ⟨a⟩^τ/|√(1+m)|, which is unbounded as |a|→∞. If this computation reproduces the displayed inequality (3.12) as false, the proof of Theorem 2.1 is not valid; then check whether a different non-resonance condition (e.g., with denominator j*_2...j*_l instead of j*_1/(⟨M⟩j*_2...j*_l)) or a modified norm restores the estimate without breaking the measure estimates in Section 4.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The central proof hinges on Lemma 3.6, which claims that if |freq| ≥ γ(j*_1/(⟨M(j)⟩j*_2...j*_l))^τ, then the solution χ of {H0,χ}+Z=f satisfies |χ|_{s+τ,N} ≤ γ^{-1}|f|_{s,N}. This is not justified. For a non-resonant monomial j, |χ_j|=|f_j|/|freq| ≤ γ^{-1}|f_j|(⟨M(j)⟩j*_2...j*_l/j*_1)^τ. The ratio of the weight in |χ|_{s+τ,N} to that in |f|_{s,N} is (j*_1...j*_l)^τ⟨M(j)⟩^{-τ}. Multiplying by the inverse small-divisor bound gives (j*_2...j*_l)^{2τ}, which is ≥1 and typically unbounded in the Fourier modes; it does not cancel. Concretely, take l=3, j=(a,-a,1) with ω_a=√(a^2+m), m∈[1,2]. Then freq=√(1+m) is independent of a and O(1), while j*_1=⟨a⟩, j*_2=⟨a⟩, j*_3=2, so the claimed estimate forces 2^τ⟨a⟩^τ ≤ γ^{-1} for all large a, which is impossible. Thus (3.12) is false as stated. This invalidates the bound |χ_l|_{s0+lτ,N} ≤ C_N/(γR_{*l-1}^{l-1}) used in (3.22), and therefore the iterative Lemma 3.7 and Theorem 2.1 do not follow. The reader's flagged issue in Lemma 4.2 is a minor exponent typo that does not break the measure bound; the real load-bearing gap is this homological-estimate error.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves an abstract Birkhoff normal form theorem (Theorem 2.1) for Hamiltonian PDEs on the torus: under the non-resonance condition (2.18), the Hamiltonian H = H0 + P is conjugate to H0 + Z + R, where Z is a (γ,τ)-normal form of order at most r+2 and R has order at least r+3, with explicit estimates (2.15)-(2.17). Corollary 2.1 derives polynomial-in-time stability of Sobolev norms and of super-actions. The abstract result is then applied to the one-dimensional nonlinear wave equation and to the high-dimensional nonlinear Schrödinger equation, yielding improved Sobolev-index requirements and, by taking r depending on the initial datum size, sub-exponentially long stability times.","tokens_in":31472,"tokens_out":40221,"duration_ms":381667,"significance":"If the results stand, the paper gives a genuinely self-contained abstract normal-form theorem with concrete constants, and it improves the known regularity requirements for the two applications: s = O(r^4) for 1D NW instead of O(r^8), and s = O(r) for high-dimensional NLS instead of O(r^3). The use of the s,N-norm (2.7) is well adapted to the Poisson bracket, and the constants in the iterative construction are tracked explicitly. I specifically checked the homological-equation estimate in Lemma 3.6 that could appear problematic: it is correct, because the factor ⟨M(j)⟩^{N−s} in the norm makes the shift s → s+τ exactly cancel the small-divisor denominator in (2.18). Thus the main abstract theorem is not undermined by that concern.","major_comments":[{"comment":"The proof of Lemma 4.3 contains an invalid estimate in the smooth (non-analytic) case: it bounds |Pn|_{0,N} = ∑_{b∈Z^d} ⟨b⟩^N |\\hat G_n(−b)| by using the decay |\\hat G_n(b)| ≤ C̃_N (2R′)^{-n} ⟨b⟩^{-N−2} and then treats ∑_{b∈Z^d} ⟨b⟩^{-N−2} as bounded by 3. This is false for d ≥ 2 when N is small; for example when N=0 the series diverges for d=2. The lemma as stated, 'for any N ≥ 0', is therefore not proved and is false as written. The fix is local: either require N > d−2 (which is automatically satisfied in Theorem 4.2, where N = s+1 with s ≥ rτ+(d+1)/2), or replace the decay exponent by ⟨b⟩^{-N−d−1} so that the sum converges for all N ≥ 0. This correction does not affect the abstract Birkhoff theorem or the nonlinear wave application, but it is a load-bearing step in the proof of the NLS application and must be repaired.","section":"§4.2, Lemma 4.3"}],"minor_comments":[{"comment":"The constant ρ appears to be 2^{d+4}3^d, as used in Lemma 3.1 and in the proof of Theorem 4.1 where ρ=96 for d=1; the displayed '2 d+43d' in the text is a typographical corruption and should be corrected.","section":"§1 and §2, formula (2.12)"},{"comment":"In the analytic case of Lemma 4.3, the claimed estimate C_N = C′((N+2)/(eμ))^{N+2} is plausible, but the proof should state which norm of the x-analytic function is used to justify the exponential decay of the Fourier coefficients uniformly in the Taylor degree n.","section":"§4.2, Lemma 4.3"},{"comment":"There are several typographical errors that should be corrected in a revision: 'miantained' and 'ineqaility' in Section 1, 'Moverover' in Lemma 3.6, 'Schr¨odinger' versus 'Schr\"odinger', and inconsistent uses of 'tours' for 'torus'.","section":"Throughout"},{"comment":"The measure estimates in Lemma 4.4 are written for an infinite product space of parameters {v_a}_{a∈Z^d}; the proof implicitly uses finite-dimensional marginal measure. The authors should state this convention explicitly, since it is essential for the meaning of |V′|.","section":"§4.2, Lemma 4.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is not circular and the central normal-form argument is sound; the pre-review concern about Lemma 3.6 does not land once the full norm (2.7) is used. The main substantive defect is the NLS nonlinearity estimate in Lemma 4.3, which is fixable locally but must be corrected before the NLS application can be considered proved. The paper fits the journal's scope and the improvements over [BG06] and [BMP20] are clearly stated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a solid contribution to the Birkhoff normal form literature. The abstract theorem — a complete normal form up to arbitrary finite order on the torus, with explicit constants — is genuinely new, and the applications lower the Sobolev thresholds for 1D NLW and high-dimensional NLS. The new s,N norm, which is preserved under Poisson brackets, is the key technical idea and it works.\n\nThe stress-test's claim that Lemma 3.6 is algebraically false does not survive reading the norm. The norm in (2.7) includes ⟨M(j)⟩ in the denominator: the weight is (j*_1/(⟨M(j)⟩j*_2...j*_l))^s, not (j*_1...j*_l)^s. Under that definition, the small-divisor denominator cancels exactly when shifting from s to s+τ. The concrete example j=(a,-a,1) gives a constant factor 4^τ and then ⟨1⟩^{-τ} brings it down, so the estimate holds with no additional a-dependence. The stress-test's calculation omits the ⟨M(j)⟩ term in the norm.\n\nThe reader's concern about Lemma 4.2 is also a sign error. For γ<1 and l≥3, the exponents satisfy 1/l > 1/(7l) > 1/(7l^2), so γ^{1/l} < γ^{1/(7l)} < γ^{1/(7l^2)}. Bounding the first two terms by γ^{1/(7l^2)} is valid; the inequality is in the correct direction.\n\nThe measure-theoretic parts are largely imported from Bambusi with added detail, and the high-dimensional NLS measure argument on the infinite product space needs a careful check of the finite-dimensional approximation. But these are standard concerns in this area, not fatal flaws. The analyticity assumptions for sub-exponential times are clearly stated.\n\nOverall: the central theorem is proved carefully, the constants are tracked, and the applications deliver the promised improvements. This deserves serious peer review. I would cite it if I worked on long-time stability of Hamiltonian PDEs.","headline":"The stress-test's central objection to Lemma 3.6 does not hold up once you read the norm definition; the paper's core theorem and applications look sound.","tokens_in":32052,"tokens_out":11304,"would_cite":true,"duration_ms":94256,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37K55","35B35","35L05","35Q55"],"pacs":[],"model":"deepseek-v4-flash","headline":"A complete Birkhoff normal form, up to arbitrary finite order, is established for Hamiltonian partial differential equations on the torus.","keywords":["Birkhoff normal form","Hamiltonian PDEs","long time stability","nonlinear wave equation","nonlinear Schrödinger equation","torus","Sobolev spaces","non-resonant conditions"],"falsifier":"A concrete observation that would settle the claim: compute, for a fixed frequency family such as ω_a = √(a² + m) with m ∈ [1,2], the quantity |δ1ω_a1 + ⋯ + δlω_al| for some l ≤ r and a tuple of modes j outside J_l; if for a set of m of positive measure this quantity falls below γ(j*_1/(⟨M(j)⟩ j*_2 ⋯ j*_l))^τ with the paper's choice of γ and τ, the homological equation estimate in Lemma 3.6 loses its bound and Theorem 2.1 fails for those parameters. More directly, one can numerically integrate the measure estimate in Lemma 4.2 for l = 3 over m ∈ [1,2] and compare the observed length of the resonant set with the claimed $e^{{12·3}}$$γ^{{1/(7·9)}}$.","tokens_in":30831,"feed_emoji":"🌊","tokens_out":6903,"duration_ms":61089,"temperature":0.7,"pith_summary":"This paper establishes an abstract Birkhoff normal form theorem for Hamiltonian partial differential equations on the torus: under a non-resonant condition on the frequencies, every Hamiltonian H0 + P can be transformed by a canonical change of variables into H0 + Z + R, where Z is a resonant polynomial of order at most r + 2 and R has order at least r + 3, with explicit constants in all estimates. This makes the normal form complete up to arbitrary finite order, without the truncation in Fourier space that earlier torus results required. The theorem is applied to the one-dimensional nonlinear wave equation and to high-dimensional nonlinear Schrödinger equations, giving long-time stability in Sobolev spaces with considerably smaller index s than previously known, and, when the nonlinearity is analytic, sub-exponentially long stability times.","feed_headline":"Full Birkhoff normal form for PDEs on the torus","feed_subtitle":"Complete normal form up to any order yields sharper Sobolev indices and sub-exponential stability.","key_machinery":"The key object is the s,N-norm (2.7) for homogeneous polynomials, defined by summing over momentum b with weight ⟨b⟩^{N-s} the supremum of |(j*_1...j*_l)^s &#771;f_j|. Its main property is that the Poisson bracket is bounded by a constant factor times the product of the two norms with the same s and N (Lemma 3.3, inequality (3.8)), so the norm is preserved through the Lie transformations that implement the normal form. The second component is the non-resonant condition (2.18), which bounds the small divisors by γ times the rational weight (j*_1/(⟨M(j)⟩ j*_2 ... j*_l))^τ; this weight, together with the norm, lets every non-resonant monomial of order at most r + 2 be eliminated without introducing a truncation parameter N, while keeping all constants explicit in r.","core_discovery":"The paper's central claim is Theorem 2.1: if the Hamiltonian perturbation P satisfies the growth estimate (2.11) in the s,N-norm introduced in (2.7), and the frequencies satisfy the non-resonant condition (2.18) up to order r + 2, then there is a canonical transformation φ with the estimate (2.15) such that H ◦ φ = H0 + Z + R. Here Z is a (γ,τ)-normal form polynomial of order at most r + 2, R has order at least r + 3, and both satisfy the quantitative bounds (2.16) and (2.17). The regularity requirement is s ≥ s0 + rτ + (d+1)/2, which improves on the earlier O($r^{2}$ α) bound from the truncation approach by removing the dependence on a large Fourier truncation parameter. As corollaries, the normal form implies that the Sobolev norm of a small solution stays bounded for times of order $ε^{{-(r+1)}}$, and that the super actions Ja change by at most $ε^{{3-ν}}$ over that time.","pith_inferences":["The same abstract theorem should apply to other Hamiltonian PDEs on the torus whose nonlinearity is smooth in x and analytic in u, provided a non-resonant condition of the form (2.18) can be verified with the rational weight; the weight is tailored to the convolution structure of the torus and may be more restrictive than necessary for equations with a different geometry.","The explicit dependence of all constants on r opens the possibility of optimizing the iterative step count to push the stability time beyond sub-exponential, toward exponential, for nonlinearities with stronger decay in Fourier space.","The norm (2.7) compresses the information of high Sobolev indices into a single weight, so the framework could be tested numerically on finite truncations: the predicted estimates (2.15)–(2.17) are quantitative enough to be checked by computing the normal form coefficients for small r."],"forward_implications":["For the 1D nonlinear wave equation utt − uxx + mu + f(x,u) = 0 with m in a set of full measure, solutions with initial Sobolev size ε stay bounded by 2ε for |t| ≤ c1 ε^{-r} in H^s with s = O(r^4), improving the earlier s = O(r^8).","For the d-dimensional nonlinear Schrödinger equation i∂tu = −Δu + V*u + g(x,u,&#773;) with d ≥ 2 and generic potentials V, the same stability holds with s = O(r) instead of O(r^3).","With analytic nonlinearity, choosing r depending on ε gives sub-exponential stability times e^{(1/2)(ln 1/ε)^{1+λ}} for the wave equation and ρ^{(logρ 1/ε)^2/(24τ logρ logρ 1/ε)} for NLS, with the latter improving to ρ^{(1/46)(logρ 1/ε)^2} when the nonlinearity is independent of x.","The super actions (mode energies) are almost conserved: ⟨a⟩^{2s}|Ja(t) − Ja(0)| ≤ c2 ε^{3−ν} for times of order ε^{-(r+ν)}."],"supporting_citations":[{"why":"Supplies the previous abstract Birkhoff normal form theorem with truncation that this paper extends; its bounds s* = O(r^2 α) are the baseline being improved.","marker":"[BG06]"},{"why":"Provides the measure-estimation scheme and the imitated arguments (Lemmas 5 and 7 of the appendix) used to prove Lemma 4.2 for the wave equation.","marker":"[Bam03]"},{"why":"Gives the method of taking iterative steps r depending on ε to obtain sub-exponential stability times, adapted here in Corollaries 4.1–4.3.","marker":"[BMP20]"},{"why":"Introduces the coefficient structure (j*_2⋯j*_l)/j*_1 that the s,N-norm is designed to keep under Poisson bracket.","marker":"[Liu22]"},{"why":"Supplies Lemma 6 of the appendix, a diophantine measure estimate used in bounding the resonant sets.","marker":"[Bam99b]"}],"fun_headline_variants":["Sharper Sobolev indices from complete Birkhoff normal form","Sub-exponential stability from full normal form on torus","Complete normal form for PDEs: better stability","Torus PDEs get normal form up to any order","Birkhoff normal form improves PDE stability times"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the non-resonant condition (2.18): for every Fourier mode and every order up to r + 2, the frequency combination either vanishes exactly (the resonant set J_l) or is bounded below by γ times the stated rational weight; in the applications this is only proved to hold outside a small exceptional set of parameters m or V.","fun_headline_variants_meta":{"raw":{"variants":["Sharper Sobolev indices from complete Birkhoff normal form","Sub-exponential stability from full normal form on torus","Complete normal form for PDEs: better stability","Torus PDEs get normal form up to any order","Birkhoff normal form improves PDE stability times"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000749,"raw_usage":{"total_tokens":3306,"prompt_tokens":887,"completion_tokens":2419,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":503,"completion_tokens_details":{"reasoning_tokens":2341}},"tokens_in":503,"tokens_out":2419,"duration_ms":17651,"temperature":1.0,"reasoning_tokens":2341,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:35:03.862771+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete observation that would settle the claim: compute, for a fixed frequency family such as ω_a = √(a² + m) with m ∈ [1,2], the quantity |δ1ω_a1 + ⋯ + δlω_al| for some l ≤ r and a tuple of modes j outside J_l; if for a set of m of positive measure this quantity falls below γ(j*_1/(⟨M(j)⟩ j*_2 ⋯ j*_l))^τ with the paper's choice of γ and τ, the homological equation estimate in Lemma 3.6 loses its bound and Theorem 2.1 fails for those parameters. More directly, one can numerically integrate the measure estimate in Lemma 4.2 for l = 3 over m ∈ [1,2] and compare the observed length of the resonant set with the claimed $e^{{12·3}}$$γ^{{1/(7·9)}}$.","supporting_citations":[],"review_version":1}