{"id":"917d3043-8b11-4811-9cce-19a0fed45525","arxiv_id":"2411.18852","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For the Neumann problem of -Δu + V u = div f above a convex graph, the paper establishes the sharp W^{1,p} estimate for all 1 < p < ∞ when V is a B∞ weight.","lead":"This mathematics paper proves a sharp gradient estimate for solutions of a Schrödinger equation in the unbounded region above a convex graph, for every exponent p between 1 and infinity. The result extends known estimates for bounded convex domains to a class of unbounded domains with nonsmooth boundaries and B-infinity potentials.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reverse Hölder estimate underlying Theorem 1.1 is proved only for C^2 convex boundaries (Lemmas 3.3–3.4), while the theorem asserts Lipschitz convex graphs; the paper gives no approximation or density argument that would pass the estimate to the non-smooth case with uniform constants.","rationale":"The reader's weakest assumption is exactly that Lemma 3.3 and Theorem 3.4 assume C^2 boundary while Theorem 1.1 only assumes a Lipschitz convex graph, with no approximation argument supplied. My review confirms that this is the most load-bearing gap: all later statements, including the duality argument for 1 < p < 2 and the boundary-data estimate, rely on the reverse Hölder inequality whose proof is C^2-specific. I also checked whether there is a stronger objection, such as an internal inconsistency in Lemma 3.3's cutoff, but the displayed identity appears to intend a Euclidean cutoff supported in B(x0,2r) rather than in B(x0,2r)∩Ω_R, and the sign of β is used only through nonnegativity, so the main missing ingredient is the passage from C^2 convex to Lipschitz convex. The constants in Lemma 3.3 do not involve the size of curvature, only the Lipschitz bound, so an approximation argument is likely to succeed; this is why the appropriate verdict is conditional acceptance rather than rejection. I therefore leave the reader's verdict unchanged: the central claim is plausible but not proved at the stated generality until the C^2-to-Lipschitz passage is supplied, and the d = 2 Sobolev issue is also addressed.","tokens_in":12345,"tokens_out":16564,"duration_ms":157083,"concrete_test":"Construct a family of smooth convex graphs φ_ε → φ locally uniformly, with ‖∇φ_ε‖∞ ≤ M and φ_ε = φ outside a large ball, for a non-C^2 convex φ such as φ(x') = |x'|. Re-run the proof of Lemma 3.3 and Theorem 3.4 on Ω_ε, tracking every constant through the integration-by-parts inequality (3.13), the co-area step (3.14), and the Sobolev embedding (3.15). If the constant C0 in (3.4) remains bounded as ε → 0 independently of the curvature of ∂Ω_ε, then prove the limiting passage of (3.4) to Ω; if the constants require curvature bounds or the limiting eigenfunctions fail to converge, then (1.3) is not established for Lipschitz convex graphs. For d = 2, separately check whether replacing 2* by an arbitrary finite Sobolev exponent and iterating in Theorem 3.4 reaches every p > 2; if not, the theorem fails in the claimed two-dimensional case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (1.3) requires the weak reverse Hölder inequality (3.4) for Ω = {t > φ(x')} with φ merely convex and ‖∇φ‖∞ ≤ M. But the proof of (3.4) is carried out only under the standing assumption in Lemma 3.3 and Theorem 3.4 that ∂Ω is C^2. The C^2 assumption enters essentially in Lemma 3.3, where the integration-by-parts identity (3.13) uses the second fundamental form β of ∂Ω and convexity enters as the sign condition β ≥ 0. For a merely Lipschitz convex graph, β is not defined, and no approximation by smooth convex φ_ε is supplied. A mollification argument is plausible because the constants in Lemma 3.3 do not visibly depend on the size of β, only on its sign and on the Lipschitz bound; but the paper does not provide it. Without such a passage, the estimate (1.3) is not proved at the stated level of generality, even though it may be true. This is a proof gap rather than a demonstrated counterexample: the constants in the reverse Hölder argument appear to be uniform in the C^2 regularization, so the gap is likely repairable. The same lack of generality affects the duality proof in Section 4, which invokes Theorem 3.4, and hence the final theorem. A secondary but related technical hole is that Lemma 3.3 uses the Sobolev exponent 2* = 2d/(d−2), which is not available for d = 2, and the case d = 2 is claimed in Theorem 1.1; no separate argument is given there either.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Neumann problem for the Schrödinger equation -Δu+Vu=div f in the unbounded domain Ω={t>φ(x')} above a convex graph, where V is a positive B∞ weight. Theorem 1.1 claims unique W^{1,p} solvability for all 1<p<∞ together with the estimate (1.3), including boundary data in B^{-1/p,p}(∂Ω), with constants independent of V and of the domain diameter. The proof combines a real-variable perturbation theorem, a reduction of W^{1,p} estimates to weak reverse Hölder inequalities for local solutions, proofs of those reverse Hölder inequalities using convexity and Fefferman-Phong-Shen estimates, and a duality argument for 1<p<2.","tokens_in":12648,"tokens_out":15660,"duration_ms":141031,"significance":"If the gaps identified below are repaired, the result would be a sharp full-range W^{1,p} estimate for Schrödinger operators with B∞ potentials in unbounded convex graph domains, extending existing results for bounded convex domains and for Lipschitz graphs. The overall strategy is sound and well matched to the problem: the reduction via the real-variable theorem in Section 3 is standard, the use of the Fefferman-Phong-Shen maximal function is appropriate, and the convexity of the boundary is exploited through the sign of the second fundamental form. The paper also gives explicit duality arguments for p<2 and for Besov boundary data. However, several load-bearing technical steps are not proved at the level of generality asserted in Theorem 1.1.","major_comments":[{"comment":"The reverse Hölder estimate is proved only for Ω with C^2 boundary: Lemma 3.3 assumes a C^2 boundary and uses the nonnegativity of the second fundamental form β in (3.13), and Theorem 3.4 repeats the C^2 assumption. Theorem 1.1, however, states the result for arbitrary convex φ with ‖∇φ‖∞≤M, which need not be C^2. No approximation by smooth convex functions is supplied, and the paper does not show that the constants in (3.12) and (3.4) are preserved under such an approximation. Since Theorem 3.4 is invoked in Lemma 4.1 and in the proof of Theorem 1.1, the estimate (1.3) is not established for the stated class of domains.","section":"§3, Lemmas 3.3 and 3.4; Theorem 1.1"},{"comment":"The proof of Lemma 3.3 uses the Sobolev exponent 2*=2d/(d−2) in Eq. (3.15), which is defined only for d≥3. Theorem 1.1 claims the result for all d≥2, and the paper gives no separate argument for d=2. Consequently the reverse Hölder inequality (3.4) and the final estimate (1.3) are not proved in two dimensions.","section":"§3, Eq. (3.15), Lemma 3.3, Theorem 1.1"},{"comment":"The hypothesis of Theorem 3.2 requires the reverse Hölder inequality (3.4) for every ball centered at x0∈∂Ω_R, which includes the artificial vertical sides of Ω_R. The proof of Lemma 3.3 is an integration-by-parts argument on ∂Ω using g·n=0 on ∂Ω and β≥0 on ∂Ω, so it does not apply to balls centered on the vertical portions of ∂Ω_R. No separate estimate for such balls is provided, and the reduction to Ω_R together with the limit R→∞ in Theorem 3.2 is therefore incomplete.","section":"§3, Theorems 3.2 and 3.4; the approximation domains Ω_R"},{"comment":"The duality proof for boundary data introduces c=⨏_Ω v dx and uses the Poincaré inequality ‖v−c‖_{W^{1,q}(Ω)}≤C‖∇v‖_{L^q(Ω)}. Since Ω is unbounded, the average over Ω is not defined and this global Poincaré inequality is not available without additional decay assumptions or a bounded-domain approximation. The same issue recurs in the proof of (4.8) for the component u2. Hence the estimate (4.5) for g∈B^{-1/p,p}(∂Ω), and with it the boundary-data part of Theorem 1.1, is not proved as written.","section":"§4, Lemma 4.2, Eq. (4.6)"}],"minor_comments":[{"comment":"The statement of Lemma 3.3 is confusing: the hypothesis says u is a weak solution but the boundary condition is written for ∂v/∂n, and the cut-off ϕ is introduced only after the inequality. The proof also switches between Ω and Ω_R; the domain of the cut-off and the domain of integration should be stated consistently.","section":"§3, Lemma 3.3"},{"comment":"The text attributes Theorem 3.1 to [24, Theorem 3.2] but the proof line says 'See [12, Theorem 2.1]'; these references should be aligned.","section":"§3, Theorem 3.1 proof"},{"comment":"The sentence 'For Schrödinger equations, Z. Shen [19] obtained the W^{1,p} estimate...' cites reference [19], which is Righi and Shen (2024), for what appears to be an older result; the intended citation should be checked and corrected.","section":"§1, paragraph on prior Schrödinger results"},{"comment":"Lemma 2.5 is stated for A∞ weights but is applied to B∞ weights; this is harmless because B∞ implies A∞, but the implication should be stated explicitly.","section":"§2, Lemma 2.5"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a serious and plausible strategy, and the main gaps—the C^2-to-Lipschitz passage, the d=2 case, the artificial side boundaries of Ω_R, and the unbounded-domain Poincaré step—appear repairable rather than fatal. I recommend major revision rather than rejection, but the authors should be asked to address all four points explicitly, since each affects a claim needed for Theorem 1.1."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this paper has a real theorem inside, but it is not exactly the theorem it states. The author proves the full-range W^{1,p} estimate for the Neumann problem with B∞ potential above C^2 convex graphs, and the proof is a credible adaptation of Geng's real-variable method and Shen's Fefferman-Phong-Shen machinery. The claimed passage from C^2 to Lipschitz convex graphs is missing, and the introduction cites a forthcoming paper [14] that appears to already cover the Lipschitz case. Those two issues are the whole story.\n\nWhat is genuinely good: the C^2 argument is substantive. Lemma 3.3 is the heart, a reverse Hölder inequality for gradients of solutions, using convexity through the sign of the second fundamental form. The duality section for 1<p<2 uses Neumann function estimates in a standard way. If the C^2 result is correct, it is a legitimate step toward the Lipschitz case. The author also correctly notes that the range 1<p<∞ is sharp even for the Laplacian.\n\nThe main gap is real: Theorem 1.1 assumes only convex Lipschitz φ, but Lemmas 3.3 and 3.4 assume C^2 boundary. The integration-by-parts identity (3.13) uses the second fundamental form, which is not defined for Lipschitz boundary. No approximation by smooth convex graphs is supplied, and it is not obvious that the constants are uniform under approximation, though they may be. The same gap carries into Section 4 via Theorem 3.4. Also, Lemma 3.3 uses the Sobolev exponent 2* = 2d/(d−2), which does not exist for d=2; the theorem claims d≥2, and no separate argument is given for d=2.\n\nNovelty is a serious question. The introduction says [14] (submitted, by the same group) establishes W^{1,p} solvability above Lipschitz graphs. If so, every convex graph is a Lipschitz graph, and Theorem 1.1 is subsumed. The author does not state how this paper differs from [14]. The referee needs to see [14] and a clarification.\n\nMinor issues: the constant in (1.3) is said to be independent of V, but the proof uses the B∞ constant; that should be tracked explicitly. There are typos, and the citation to [19] for Shen's Lipschitz result looks like a mis-reference, probably [20] or [21].\n\nWho this is for: people working on Lp estimates for Schrödinger operators in rough domains. It deserves a serious referee because the C^2 result is substantial and the gap is likely repairable. I would not cite it as the Lipschitz result until the approximation is written down, but I would send it out with instructions to demand the approximation argument and a direct comparison with [14].","headline":"A credible C^2 convex-graph result that does not yet prove the stated Lipschitz theorem, and whose novelty is undermined by the paper's own citation of a forthcoming Lipschitz-graph paper.","tokens_in":13240,"tokens_out":2495,"would_cite":false,"duration_ms":19833,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35J10","35J25","35B45"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the Neumann Schrödinger problem above a convex graph is W^{1,p}-solvable for every 1<p<∞ under B∞ potentials.","keywords":["Schrödinger operator","Neumann problem","W^{1,p} estimates","convex graph domains","B∞ weights","Fefferman-Phong-Shen maximal function","reverse Hölder inequality","Besov boundary data"],"falsifier":"Take the cone domain $\\Omega=\\{t>|x'|\\}$ with $V\\equiv1$ and run the asserted estimate for smooth compactly supported $f$ whose support touches the ridge; alternatively, smooth the cone through convex $C^2$ graphs and watch the constants in (3.17). If the constant blows up as the approximations converge, or if (1.3) fails for some $1<p<\\infty$ on the cone, the theorem as stated is false. The minimal test is whether the uniform bound survives the $C^2$-to-Lipschitz passage.","tokens_in":12067,"feed_emoji":"📐","tokens_out":16025,"duration_ms":125422,"temperature":0.7,"pith_summary":"This paper proves an a priori $W^{1,p}$ estimate for the Neumann problem of $-\\Delta + V$ in the unbounded region above a convex graph, for potentials $V$ in the $B_\\infty$ class (a reverse Hölder condition). For every $1<p<\\infty$, the gradient of the solution and $V^{1/2}u$ are controlled in $L^p(\\Omega)$ by the divergence datum and the Besov boundary datum, with a constant that does not depend on the size of $V$ or on the diameter of the domain. This is the same full exponent range known for the Laplacian in bounded convex domains, so the range is sharp. The proof reduces the global estimate to a local weak reverse Hölder inequality for solutions of the homogeneous Neumann problem, and obtains $1<p<2$ by duality.","feed_headline":"Schrödinger Neumann bound proved for every p between 1 and infinity","feed_subtitle":"The gradient and V^{1/2}u stay in every L^p, with constants independent of the potential and domain size.","key_machinery":"The central machinery is a two-step reduction. A real-variable perturbation theorem (Theorem 3.1) says the global $W^{1,p}$ bound follows once every local solution of the homogeneous Neumann problem satisfies a weak reverse Hölder inequality of the form (1.5)/(3.4). The paper proves that local inequality by combining three ingredients: the Fefferman–Phong–Shen maximal function $m(x,V)$ (the scale below which $V$ is small on average), the improved Fefferman–Phong inequality, and the nonnegativity of the second fundamental form on a convex boundary, which makes an integration-by-parts identity coercive. For $1<p<2$, the argument switches to duality, using estimates for the Neumann function $N(x,y)$ and dyadic annuli in the metric of $m(x,V)$.","core_discovery":"On its own terms, the paper establishes the following. Let $\\Omega = \\{t>\\varphi(x')\\}$ with $\\varphi$ convex and $\\|\\nabla\\varphi\\|_\\infty\\le M$, and let $0<V\\in B_\\infty$ satisfy (1.2). Then the Neumann problem (1.1) is uniquely solvable in $W^{1,p}(\\Omega)$ for all $1<p<\\infty$, and the solution satisfies $\\|\\nabla u\\|_{L^p(\\Omega)}+\\|V^{1/2}u\\|_{L^p(\\Omega)}\\le C(\\|f\\|_{L^p(\\Omega)}+\\|g\\|_{B^{-1/p,p}(\\partial\\Omega)})$, with $C$ depending only on $d$, $p$ and the Lipschitz character of $\\Omega$, not on $V$ or the diameter. The exponent range is asserted to be sharp even for the Laplacian. The route is: a refined real-variable perturbation criterion (Theorem 3.1) turns a weak reverse Hölder inequality for local null solutions into the global $W^{1,p}$ estimate; the local inequality is proved using the improved Fefferman–Phong inequality, the Fefferman–Phong–Shen maximal function $m(x,V)$, and convexity of the boundary; and the range $1<p<2$ is completed by duality through estimates for the Neumann function.","pith_inferences":["Testable extension: smooth the cone $\\varphi(x')=|x'|$ by convex $C^2$ graphs and track the constant in (3.17); a uniform bound would confirm the full Lipschitz-level claim, while a blow-up would locate the missing approximation step.","The proof's ingredients are scale-invariant, so finite unions of convex graph domains or domains with bounded geometry should satisfy the same estimate by localization.","The same $m(x,V)$-based argument plausibly gives weighted versions with $V\\,dx$ or $V^\\alpha\\,dx$ and Dirichlet analogues, since the Fefferman–Phong inequality and Neumann function bounds are the only potential-dependent inputs.","The Neumann function bounds used for the duality step suggest that related kernel estimates, for example for the heat semigroup or Riesz transforms of $-\\Delta+V$ under $B_\\infty$ potentials on convex epigraphs, may follow along the same lines."],"forward_implications":["The full range $1<p<\\infty$ is obtained for unbounded convex graph domains, matching the sharp range for the Laplacian in bounded convex domains.","The constant in the estimate is uniform in the potential $V$ and in the domain's diameter, so the result is available for limit and homogenization arguments.","Potentials of the form $|x|^a$ with $a\\ge0$ are covered, so the estimate holds for unbounded potentials, not just bounded ones.","With zero Neumann data the gradient and $V^{1/2}u$ are controlled by the $L^p$ norm of $f$ alone; with $f=0$ the gradient is controlled by the Besov norm of $g$.","Uniqueness and existence follow for every $1<p<\\infty$, so the Neumann problem is well posed throughout the full range."],"supporting_citations":[{"why":"supplies the $B_\\infty$ weight framework, the Fefferman–Phong–Shen maximal function estimates, the boundary $L^\\infty$ estimate, and the Neumann function bound used throughout Sections 2 and 4.","marker":"[20]"},{"why":"supplies the self-improving comparison properties of $m(x,V)$ and the $L^p$ estimates for Schrödinger operators that the duality step relies on.","marker":"[21]"},{"why":"supplies the refined Fefferman–Phong inequality (Lemma 2.5) used to control local $L^2$ norms of $u$ by gradient and potential terms.","marker":"[1]"},{"why":"supplies the real-variable perturbation template and the sufficient-condition formulation for $W^{1,p}$ estimates under Neumann boundary conditions.","marker":"[11]"},{"why":"supplies the convex-domain reverse Hölder analysis that the proof adapts to the Schrödinger setting with potentials.","marker":"[13]"},{"why":"supplies Theorem 3.1, the real-variable perturbation lemma that upgrades local reverse Hölder estimates to the global $L^p$ bound.","marker":"[12]"},{"why":"is quoted as the source of the refined real-variable theorem used as Theorem 3.1.","marker":"[24]"}],"fun_headline_variants":["All-p Schrödinger gradient bounds above convex graphs","Sharp W^{1,p} estimates for Schrödinger above convex graphs","Neumann Schrödinger: all Lp gradient control with V weight","Schrödinger convex-domain Neumann: sharp Lp bounds for all p"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the key local estimate uses a $C^2$ boundary so that the curvature term in an integration-by-parts identity is nonnegative, but the theorem only assumes a Lipschitz convex graph; the paper does not provide the approximation step that would carry the estimate across that gap with uniform constants.","fun_headline_variants_meta":{"raw":{"variants":["All-p Schrödinger gradient bounds above convex graphs","Sharp W^{1,p} estimates for Schrödinger above convex graphs","Neumann Schrödinger: all Lp gradient control with V weight","Schrödinger convex-domain Neumann: sharp Lp bounds for all p"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000971,"raw_usage":{"total_tokens":4145,"prompt_tokens":975,"completion_tokens":3170,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":591,"completion_tokens_details":{"reasoning_tokens":3098}},"tokens_in":591,"tokens_out":3170,"duration_ms":23259,"temperature":1.0,"reasoning_tokens":3098,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:50:00.840493+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the cone domain $\\Omega=\\{t>|x'|\\}$ with $V\\equiv1$ and run the asserted estimate for smooth compactly supported $f$ whose support touches the ridge; alternatively, smooth the cone through convex $C^2$ graphs and watch the constants in (3.17). If the constant blows up as the approximations converge, or if (1.3) fails for some $1<p<\\infty$ on the cone, the theorem as stated is false. The minimal test is whether the uniform bound survives the $C^2$-to-Lipschitz passage.","supporting_citations":[{"cited_title":"Shen, On the Neumann problem for Schr ¨odinger operators in Lipschitz domains , Indiana Univ","cited_arxiv_id":null,"evidence_quote":"supplies the $B_\\infty$ weight framework, the Fefferman–Phong–Shen maximal function estimates, the boundary $L^\\infty$ estimate, and the Neumann function bound used throughout Sections 2 and 4."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the self-improving comparison properties of $m(x,V)$ and the $L^p$ estimates for Schrödinger operators that the duality step relies on."},{"cited_title":"Auscher, On necessary and sufficient conditions for Lp-estimates of Riesz transforms associated to elliptic operators on Rn and related estimates , Mem","cited_arxiv_id":null,"evidence_quote":"supplies the refined Fefferman–Phong inequality (Lemma 2.5) used to control local $L^2$ norms of $u$ by gradient and potential terms."},{"cited_title":"Geng, W 1,p estimates for elliptic problems with Neumann boundary cond itions in Lipschitz domains , Adv","cited_arxiv_id":null,"evidence_quote":"supplies the real-variable perturbation template and the sufficient-condition formulation for $W^{1,p}$ estimates under Neumann boundary conditions."},{"cited_title":"Geng and Z","cited_arxiv_id":null,"evidence_quote":"supplies the convex-domain reverse Hölder analysis that the proof adapts to the Schrödinger setting with potentials."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies Theorem 3.1, the real-variable perturbation lemma that upgrades local reverse Hölder estimates to the global $L^p$ bound."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"is quoted as the source of the refined real-variable theorem used as Theorem 3.1."}],"review_version":1}