{"id":"e491a84a-e7f1-435b-9a3f-b60191d76731","arxiv_id":"2411.14824","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a class of pseudodifferential operators, the distance between spectral edges under a dilation-type perturbation decays like |δ|^ν with ν between 1/2 and 1, depending on the decay rate of the second derivatives of the perturbation field.","lead":"The paper proves quantitative bounds on how the spectrum of a pseudodifferential operator changes when its symbol is shifted by a small vector field. The main result shows that the edges of the spectrum move at a rate between one-half and one in the perturbation parameter, depending on how fast the curvature of the vector field decays at infinity.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed edge rate in Theorem 2.3 depends on an unproved bound for the cutoff term in I1; the decay assumption on F'' alone may only justify a slower rate for μ<1.","rationale":"The reader identified the decay condition on F'' as the weakest assumption, but did not notice that the proof of Proposition 4.2, specifically the bound on I1, may not actually follow from that condition. The central theorem is plausible, and the algebraic slip in the parameter choice is repairable, but the cutoff-derivative issue is a genuine gap: the estimate |I1|≤κ^{-2}θ^{1+μ} is asserted without controlling the terms where derivatives hit χθ. If those terms contribute O(κ^{-2}θ), the proof's optimization cannot reach the stated exponent for μ<1. This does not prove the theorem false; it shows the proof as written is incomplete. Conditional acceptance is therefore appropriate, with the condition that (4.18) be either rigorously established, including the cutoff terms, or replaced by a corrected rate. The concrete test proposed would settle the asymptotic size of I1 in a simple one-dimensional case.","tokens_in":18806,"tokens_out":20197,"duration_ms":204154,"concrete_test":"Take d=1, 0<μ<1, and F(x)=x+g(x) with g'(x)=C−∫_0^x(1+t^2)^{-(1+μ)/2}dt, so g''=(1+x^2)^{-(1+μ)/2} and F'' has the assumed decay while F grows linearly and F' tends to a nonzero constant. Choose an even χ with χ=1 on [−1,1] and χ=0 outside [−2,2], set χθ(x)=χ(θx), take φ(x)=π^{-1/4}e^{-x^2/2}, and Wκ as in (4.2). Compute I1 from (4.17) explicitly, isolating the contribution of the terms −2θχ'(θz)F'(z)−θ^2χ''(θz)F(z) to the z-integral after integrating u against (z−u)^2Wκ(z−u), which equals 2κ^{-2}. If the leading asymptotic is κ^{-2}θ rather than κ^{-2}θ^{1+μ}, then (4.18) fails and the parameter optimization gives at best δ^{2/3}, contradicting the claimed rate for μ<1.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The load-bearing estimate is (4.12) in Proposition 4.2, whose proof rests on the asserted bounds |I1|≤κ^{-2}θ^{1+μ} and |I2|≤κ^{-2} for the outer-region integrals I1, I2 defined near (4.17). The bound on I1 is what converts the decay |∂xj∂xk F|≤C<x>^{-(1+μ)} into the edge exponent (1+μ)/(2+μ). As written, this bound is not justified. The integrand of I1 contains (∇⊗∇)Fθ^⊥, where Fθ^⊥=(1−χθ)F. On the support of 1−χθ one has |z|≳θ^{-1}, so (1−χθ)∂^2F is O(θ^{1+μ}); but Leibniz gives additional terms −2θχ'(θz)∂F − θ^2χ''(θz)F, supported on the annulus |z|∼θ^{-1}. Under the hypothesis only on ∂^2F, these terms are O(θ), not O(θ^{1+μ}); if F grows linearly (which is allowed since F∈C∞_1 permits unbounded F), the term θ^2χ''(θz)F is O(θ). No cancellation or integration by parts is shown to remove them. If these terms survive, (4.12) acquires a contribution δκ^{-2}θ. Optimizing with κ^2=δ^ρ and θ=δ^{1−ρ} gives max(ρ, 2−2ρ)≥2/3, so the proof yields at best δ^{2/3}, violating the claimed δ^{(1+μ)/(2+μ)} for 0<μ<1. The final parameter equation in the paper, 'ρ=(2+μ)(1−ρ) implies ρ=(1+μ)/(2+μ)', is also algebraically wrong, though that slip alone is harmless because δ≤δ^ρ for small δ.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the spectral stability of Weyl pseudodifferential operators with symbols a(x + δF(x), ξ), where a ∈ S^0_{0,0}(R^d × R^d) is real and F ∈ C^∞_1(R^d; R^d) has bounded derivatives of positive order. Theorem 2.1 establishes that the Hausdorff distance between the spectra of the perturbed and unperturbed operators is O(√|δ|). Theorem 2.3 claims that if the second derivatives of F decay as ⟨x⟩^{-(1+μ)} for some μ>0, then the spectral edges E_±(δ) satisfy |E_±(δ) − E_±(0)| ≤ C(a,F)|δ|^{(1+μ)/(2+μ)} for small δ. The proof of Theorem 2.1 uses a quasiresolvent built from a lattice of translated resolvents and commutator estimates. The proof of Theorem 2.3 uses quadratic forms, a Gaussian weight W_κ, and two cutoffs (one on the kernel variable v, one on the field F). The central estimate is Proposition 4.2, which bounds the difference between the quadratic forms of the original and weighted operators by terms δ/θ, δκ^{-2}θ^{1+μ}, and δ^2κ^{-2}; the optimal choice of θ and κ then yields the claimed exponent.","tokens_in":19178,"tokens_out":8674,"duration_ms":77686,"significance":"If the result of Theorem 2.3 is correct, it gives a quantitative spectral edge regularity that goes beyond the general square-root Hausdorff bound and explicitly ties the edge rate to the spatial decay of the perturbation's curvature. This is a meaningful refinement of the authors' earlier work on slowly varying perturbations and is of interest to the spectral theory community. The proof of Theorem 2.1 is coherent and appears sound: the quasiresolvent construction, the use of a quadratic partition of unity, and the balancing of the two commutator estimates at κ=1/2 are natural and well executed. The paper is self-contained, reproducing a needed lemma from the authors' previous paper, and does not rely on any ad hoc or circular assumptions. However, the main new result, Theorem 2.3, rests on an unproved estimate for the outer-region integral I1 in Proposition 4.2. As written, that estimate is not justified under the stated hypotheses, so the claimed exponent is not established. This gap is load-bearing and requires a substantial repair or a revision of the theorem's statement.","major_comments":[{"comment":"The bound |I1[φ]| ≤ κ^{-2}θ^{1+μ} is not justified by the preceding text. In the definition of I1 (eq. (4.17)), the term involving (∇⊗∇)F^⊥_θ is understood pointwise, but F^⊥_θ = (1 − χ_θ)F is a product, so its Hessian contains the cutoff derivatives −2∇χ_θ ⊗ ∇F − (Δχ_θ)F in addition to (1−χ_θ)∇⊗∇F. These extra terms are supported on the annulus |z| ∼ θ^{-1}. Under the hypotheses, ∇F is bounded and F may grow linearly, so these terms are only O(θ), not O(θ^{1+μ}); specifically, F∇χ_θ is O(θ · θ^{-1}) = O(1) at the level of first derivatives and the Hessian terms are O(θ) due to the θ^2 factors from Δχ_θ combined with F ∼ θ^{-1}. No cancellation or integration by parts is shown to eliminate these contributions. If |I1| is only bounded by Cκ^{-2}θ, then the term δ|I1| becomes δ^{2−2ρ} after the parameter choices κ^2 = δ^ρ, θ = δ^{1−ρ}, and the optimization yields the rate δ^{2/3} rather than the claimed δ^{(1+μ)/(2+μ)} for 0 < μ < 1 (and also for μ > 1, where the claimed exponent exceeds 2/3). The proof must either provide a rigorous estimate for I1 that accounts for the cutoff derivatives or the statement of Theorem 2.3 must be weakened accordingly. The current text simply asserts (4.18) after defining I1 and I2, which is insufficient for a result of this precision.","section":"4, Proposition 4.2, eq. (4.18)"}],"minor_comments":[{"comment":"The line 'ρ = (2+μ)(1−ρ) implies ρ = (1+μ)/(2+μ)' is algebraically wrong: the solution is ρ = (2+μ)/(3+μ). Since this correct value gives a larger exponent than the one stated (δ^{(2+μ)/(3+μ)} < δ^{(1+μ)/(2+μ)} for 0<δ<1), the claimed estimate still follows, but the derivation should be corrected.","section":"Section 4, parameter choices"},{"comment":"The abstract says F has 'all its derivatives globally bounded', which could be misread to include F itself, but Notation 1.2 defines C^∞_1 as bounded derivatives of strictly positive order. This discrepancy should be resolved, e.g., by saying 'all derivatives of positive order' in the abstract as well.","section":"Abstract and Notation 1.2"},{"comment":"The manuscript contains numerous typos and OCR-like artifacts, such as 'evidengtly' in the proof of Proposition 3.3, misaligned parentheses in (3.1), and inconsistent use of slashes in formulas. A careful proofreading pass is needed before publication.","section":"Throughout"},{"comment":"In Lemma 4.3 and Lemma 4.4, the bound ν_{n,m}(a'_{δ,θ}) ≤ M F θ^{-1} sup_{0≤s≤1} ν_{n+1,m}(a[F]_s) uses the sup of |F_θ|. While this is correct, it would be clearer to explicitly state that sup|F_θ| ≤ Cθ^{-1} because F grows at most linearly on the support of χ_θ.","section":"Section 4, proof of Lemma 4.3/4.4"}],"recommendation":"major_revision","confidential_remarks":"The gap in the proof of Proposition 4.2 is substantial and directly affects the main theorem. If the authors can supply a correct bound for I1 (or modify the claim) without changing the overall framework, the paper may become publishable. The proof of Theorem 2.1 and the quadratic-form strategy are promising, but as it stands the central result is not established. I also note that the self-citation to [2] is used appropriately, and no circularity is apparent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: Thm 2.1 is good, Thm 2.3 is the interesting new claim but its proof has a load-bearing gap in Prop 4.2, and the abstract oversells the edge behavior.\n\nWhat's actually new: the symbol class a(x+δF(x),ξ) with general bounded-derivative F, removing the slow-variation assumption of [2]. The O(√δ) Hausdorff distance (Thm 2.1) is a clean, self-contained argument using quasi-resolvents and localization; I see no issue there. The edge exponent (1+μ)/(2+μ) under the decay of F'' is a genuinely new target.\n\nWhere it breaks: (4.18) asserts |I1| ≤ κ^{-2}θ^{1+μ} without proof. I1 contains (∇⊗∇)Fθ^⊥ after the first-order term is removed by symmetry of Wκ. On the support of 1−χθ, (1−χθ)∂²F is O(θ^{1+μ}), but differentiating the cutoff gives −θχ'(θz)∂F and −θ²χ''(θz)F, supported on |z|∼θ^{-1}, and these are O(θ) because F may grow linearly. No cancellation is shown. If they survive, the middle term in (4.12) becomes δ^{2−2ρ} rather than δ^{(2+μ)(1−ρ)}. For μ>1 the stated exponent is then not reachable by this argument; for μ<1 the argument can still produce a rate at least 2/3, which is stronger than the claim, so the theorem may survive in that range, but the paper's derivation of the exact exponent is wrong. The parameter line \"ρ=(2+μ)(1−ρ) implies ρ=(1+μ)/(2+μ)\" is also algebraically misleading as printed; likely a typo, but it should be corrected. Finally, Theorem 2.3 only bounds sup and inf of the spectrum, while the abstract and introduction talk about spectral edges more broadly, and 'behaves like' implies a two-sided statement when only an upper bound is proven.\n\nBottom line: the problem is meaningful and the strategy is serious. I would not desk-reject it, but it needs a substantive revision: prove or repair (4.18), weaken the statement if necessary, and align the abstract with what is actually shown.","headline":"The square-root Hausdorff bound is solid; the edge-regularity theorem is a genuine but under-supported claim, with the key estimate (4.18) unproved and the abstract overstating the result.","tokens_in":19760,"tokens_out":15035,"would_cite":false,"duration_ms":128844,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35S05","47G30","47A10"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the spectral edges of a Weyl-quantized pseudodifferential operator move under dilation-type perturbations with rate (1+μ)/(2+μ), improving the universal square-root bound.","keywords":["pseudodifferential operators","Weyl quantization","spectral stability","symbol class S^0_{0,0}","dilation perturbations","spectral edges","Hausdorff distance"],"falsifier":"Take F(x)=(1+|x|^2)^{-μ/2}x so that ∂^2F decays like ⟨x⟩^{-(1+μ)}, choose a real symbol a in $S^{0}$_{0,0} whose Weyl quantization has a non-degenerate spectral edge, and compute E_+(δ)-E_+(0) numerically for δ→0; an observed scaling with exponent strictly smaller than (1+μ)/(2+μ) (meaning a slower approach to zero than the theorem allows) would contradict Theorem 2.3.","tokens_in":18570,"feed_emoji":"📐","tokens_out":13757,"duration_ms":119141,"temperature":0.7,"pith_summary":"When a smooth vector field F with all derivatives bounded is used to dilate the position variable of a real pseudodifferential symbol, x → x+δF(x), the spectrum of the corresponding Weyl-quantized operator is shown to stay within a square-root distance of the unperturbed spectrum. The paper's main new result is an edge regularity statement: if the second derivatives of F decay at infinity like a power 1+μ, then the top and bottom of the spectrum move with the faster rate |δ|^{(1+μ)/(2+μ)}, an exponent between 1/2 and 1. This removes the 'slow variation' hypothesis of the earlier paper, replacing it by a purely analytic decay condition on F. The argument works for all real symbols of the class $S^{0}$_{0,0} and yields explicit constants depending only on the symbol seminorms and F.","feed_headline":"Spectral edges beat the square-root bound under a decay condition","feed_subtitle":"A decay condition on the perturbing field upgrades the generic √δ edge bound to a sharper Hölder rate.","key_machinery":"The proof rests on two constructions. For the Hausdorff bound, a quasi-resolvent T(z;δ)=Σ_γ G_γ τ_{−z_γ}(K_0−z)^{-1}τ_{z_γ}G_γ is assembled from translations τ_{z_γ} at lattice points z_γ=δ^κ γ and a quadratic partition of unity G_γ = multiplication by g($δ^{{1−κ}}$F(x)−γ). A Newton–Leibniz expansion of the kernel difference shows the error is O(δ^κ)+O($δ^{{1−κ}}$), optimised at κ=1/2 to give √δ. For the edge estimate the paper changes tactic and works with quadratic forms, replacing the dilation x→x+δF(x) by the translation x→x+δF(u) and using a Gaussian weight identity to localise the difference. A two-zone cutoff (inner region with χ_θ and outer region with 1−χ_θ) separates the regime where the field is effectively constant from the regime where the second derivatives' decay can be traded against a growing δ-dependent weight; balancing the parameters θ and κ yields the exponent (1+μ)/(2+μ).","core_discovery":"The central discovery is Theorem 2.3: for a real symbol a of the class $S^{0}$_{0,0} and a smooth dilation field F with all derivatives bounded, if |∂_{x_j}∂_{x_k}F(x)| ≤ C⟨x⟩^{-(1+μ)} for some μ>0, then the spectral edges E_+(δ)=sup σ(Op^w(a(x+δF(x),ξ))) and E_-(δ)=inf σ(Op^w(a(x+δF(x),ξ))) satisfy |E_±(δ)−E_±(0)| ≤ C(a,F)|δ|^{(1+μ)/(2+μ)} for |δ|≤δ_0. Together with Theorem 2.1, which bounds the Hausdorff distance of the full spectra by C(a,F)√|δ|, this shows that the interior of the spectrum can move at the slower square-root scale while the edges enjoy a strictly better Hölder regularity controlled by the decay of F's second derivatives.","pith_inferences":["A natural test case is F(x)=(1+|x|^2)^{-μ/2}x, for which the decay condition is exactly satisfied; computing E_+(δ) numerically for a simple symbol would show whether the exponent (1+μ)/(2+μ) is sharp or merely an upper bound.","The same Gaussian-weight localisation could be applied to more general perturbations x↦x+δG(x) that are not pure dilations but share the decay of second derivatives, suggesting a wider class of 'asymptotically affine' perturbations with edge regularity.","The paper proves the improved rate for the global spectral edges only; whether internal gaps open or close with the same rate is left open by the argument, since the quadratic-form estimates are global rather than localised to a gap."],"forward_implications":["The Hausdorff distance between σ(K_δ) and σ(K_0) is at most C(a,F)√|δ| for every real symbol in S^0_{0,0} and every F with globally bounded derivatives, and counterexamples show this scale can be attained by spectral gaps.","If the second derivatives of F decay like ⟨x⟩^{-(1+μ)}, then the supremum and infimum of the spectrum satisfy |E_±(δ)-E_±(0)| ≤ C(a,F)|δ|^{(1+μ)/(2+μ)}.","The exponent ν=(1+μ)/(2+μ) interpolates between 1/2 at μ→0 and 1 at μ→∞, so the edge motion ranges from barely better than square-root to almost Lipschitz.","The constants depend only on finitely many symbol seminorms and on the derivative bounds of F, so the estimates are uniform in the perturbation parameter δ."],"supporting_citations":[{"why":"Supplies the L^2 boundedness criterion for pseudodifferential operators used to control every kernel norm estimate.","marker":"[1]"},{"why":"Preceding paper by the same authors under a slow-variation assumption; its quasi-resolvent and Gaussian-weight constructions are adapted here.","marker":"[2]"},{"why":"Establishes Lipschitz continuity of spectra for affine perturbations, the motivating baseline that the dilation family extends.","marker":"[4]"},{"why":"Defines the symbol class S^0_{0,0} and the Weyl quantization formalism in which the problem is posed.","marker":"[5]"},{"why":"Together with [1] justifies the uniform operator-norm bound for pseudodifferential operators used throughout.","marker":"[7]"}],"fun_headline_variants":["Spectral edges beat √δ bound under decay condition","Holder edge rates from decay of second derivatives","Perturbed symbols: edges converge faster than √δ","Decay in Hessian sharpens spectral edge Holder exponent"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the second derivatives of the perturbing field F decay at infinity at least as fast as the reciprocal of distance raised to the power 1+μ.","fun_headline_variants_meta":{"raw":{"variants":["Spectral edges beat √δ bound under decay condition","Holder edge rates from decay of second derivatives","Perturbed symbols: edges converge faster than √δ","Decay in Hessian sharpens spectral edge Holder exponent"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000449,"raw_usage":{"total_tokens":2250,"prompt_tokens":918,"completion_tokens":1332,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":1269}},"tokens_in":534,"tokens_out":1332,"duration_ms":11402,"temperature":1.0,"reasoning_tokens":1269,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:50:09.708861+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take F(x)=(1+|x|^2)^{-μ/2}x so that ∂^2F decays like ⟨x⟩^{-(1+μ)}, choose a real symbol a in $S^{0}$_{0,0} whose Weyl quantization has a non-degenerate spectral edge, and compute E_+(δ)-E_+(0) numerically for δ→0; an observed scaling with exponent strictly smaller than (1+μ)/(2+μ) (meaning a slower approach to zero than the theorem allows) would contradict Theorem 2.3.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the L^2 boundedness criterion for pseudodifferential operators used to control every kernel norm estimate."},{"cited_title":"Journal of Spectral Theory","cited_arxiv_id":null,"evidence_quote":"Preceding paper by the same authors under a slow-variation assumption; its quasi-resolvent and Gaussian-weight constructions are adapted here."},{"cited_title":"Journal of Spectral Theory","cited_arxiv_id":null,"evidence_quote":"Establishes Lipschitz continuity of spectra for affine perturbations, the motivating baseline that the dilation family extends."},{"cited_title":"H¨ ormander: The Analysis of Linear Partial Differential Operators III: Ps eudo-Differential Operators","cited_arxiv_id":null,"evidence_quote":"Defines the symbol class S^0_{0,0} and the Weyl quantization formalism in which the problem is posed."},{"cited_title":"Taylor: Pseudodifferential operators (PMS-34)","cited_arxiv_id":null,"evidence_quote":"Together with [1] justifies the uniform operator-norm bound for pseudodifferential operators used throughout."}],"review_version":1}