{"id":"2bb696a0-1121-48e2-9618-9bf70e825d98","arxiv_id":"2411.19367","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For divergence-form elliptic equations, the paper proves regularity estimates with constants exponential in a coefficient norm times domain radius, shows these exponents are optimal by explicit examples, and extends Landis-type spherical decay estimates to general operators.","lead":"This paper sharpens the classical estimates in the theory of uniformly elliptic equations, so that their constants grow only like the exponential of a coefficient norm times the domain size, instead of a worse superlinear power. Each estimate comes with explicit counterexamples showing the constant is optimal, and the results feed into eigenvalue bounds and the Landis conjecture on decay of solutions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 5.1's global Harnack-chain step is the load-bearing point, and its constant is imported from [50]/[54] rather than proved; if the iteration constant is worse than e^{C0 D(M+rΩ^{-1})}, the main linear-in-M estimates degrade.","rationale":"The reader's weakest-assumption already identifies Theorem 5.1, and I agree. The central claim of the paper is not merely existence of some exponential bound; it is that the exponent is linear in M and R, which is what the counterexamples in Section 4 match. All the main regularity estimates for positive solutions and the Landis application are built on this single inequality. The proof localizes the operator on r0-balls where constants are bounded, then invokes a chain iteration whose constant is not derived in the paper. Since the chain length N is proportional to D/r0 = D(M + rΩ^{-1}), a constant C^N is acceptable and C^{N^2} is not; the difference is exactly the headline exponent. The paper is otherwise unusually careful and self-critical, and I found no independent contradiction. I also verified the paper's own flags: Remark 2.2 and Remark 4.2 state the limitations of the optimality claims, which reduces the risk that the abstract overstates the body. Therefore the appropriate verdict remains CONDITIONAL: the concern is a specific, checkable gap in the proof of the central constant, not a demonstrated error.","tokens_in":55637,"tokens_out":11231,"duration_ms":102271,"concrete_test":"Have a specialist referee complete Section 12.3: starting from (12.31)–(12.33), derive (12.34) for two chain balls B_k, B_l with an explicit constant C(N) and track how C(N) depends on N from Proposition 12.2. The test is whether C(N) can be bounded by exp(C0 N) with universal C0; if instead C(N) is exp(C0 N^2) or exp(C0 N I), then Theorem 5.1 and the downstream linear-in-M estimates fail. A quicker independent check is to re-derive [54, Theorem 2.1]'s Harnack chain constant for the one-dimensional model u'' + c u = 0 with large c on a chain of N unit balls and compare the propagated constant with e^{C0 N}.","verdict_should_be":"UNCHANGED","load_bearing_attack":"All of the paper's headline estimates for positive solutions—Theorems 2.8, 2.9, 3.3, 3.4—and the duality step in Theorem 3.3(i) pass through the global weak Harnack inequality (5.1)–(5.5). The proof in Section 12.3 is not self-contained: after directly verifying the normalized local estimates (12.31)–(12.33) on the covering balls, it says 'we can repeat the iteration argument on [54, p.12]' and asserts (12.34) with constant C0^N. The length of the Harnack chain is controlled by Proposition 12.2: N ≤ 2 + 6D/r ≤ C D/r0 = C D(M + rΩ^{-1}). If the [54] iteration really gives a factor C0^N, then C0^N ≤ e^{C0 D(M + rΩ^{-1})}, and (5.1) follows. But this is precisely the step that is not written out; if the imported iteration instead produces C0^{N^2}, C0^I, or C0^{NI}, the exponential would become superlinear in D(M + rΩ^{-1}) and every downstream estimate, including the claimed optimal linear-in-M Hopf and log-gradient bounds, would lose its stated form. The paper explicitly lists this as a proof sketch and defers the technical differences, so the constant is a genuine open check rather than a cosmetic gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper revisits classical linear second-order uniformly elliptic equations in divergence form on bounded C^{1,ᾱ} domains, Lu = div(A Du + b1 u) + b2·Du + c u, and proves versions of the generalized maximum principle, Lipschitz/C^{1,α} estimates, the Hopf-Oleinik lemma, boundary weak Harnack inequalities, and logarithmic gradient estimates in which the constants depend optimally on a uniformly local, scale-invariant norm M of the coefficients and on the geodesic diameter D. The main estimates have the form e^{C0 M D} (or variants) with linear dependence on M and D, which improves on previously available superlinear dependence for the maximum principle. The paper also proves upper and lower bounds for the first eigenvalue, gives Landis-conjecture-type lower bounds for solutions in balls, and shows by explicit counterexamples that the exponential dependence on M is optimal for the main estimates. The proofs combine rescalings, Harnack chains, duality arguments, and a doubling/contradiction method, with detailed arguments in Sections 6–11 and auxiliary results in Section 12.","tokens_in":36,"tokens_out":15854,"duration_ms":309092,"significance":"If correct, the results settle the sharp dependence of several classical a priori estimates on coefficient norms and domain size. The optimized Stampacchia–Trudinger estimate with exponent linear in M, the new up-to-the-boundary log-gradient bound for equations with right-hand side, and the optimized quantitative Hopf lemma are significant advances over the existing literature, and the counterexamples show that no better growth in M is possible. The paper is also careful about its own limitations: Remark 2.2 leaves an open problem, Remark 4.2 flags that the inhomogeneous log-gradient term is not known to be optimal, and Proposition 2.7 shows that the spectral assumption in the Landis-type result is necessary rather than merely technical. The authors ship detailed proofs, not just sketches, and the constructions in Propositions 4.1–4.4 use the actual quantities M of the operators involved, so no fitted parameter is disguised as a prediction.","major_comments":[{"comment":"The proof of the optimized global weak Harnack inequality, Theorem 5.1, is not fully written out. Section 12.3 verifies the normalized local estimates (12.31)–(12.33) on covering balls and then states \"we can repeat the iteration argument on [54, p. 12]\" and asserts (12.34) with a constant C0^N. Theorem 5.1 underpins Theorems 2.8, 2.9, 3.3 and 3.4, and the claimed linear-in-M form of the exponents in those theorems depends on the chain product being exactly e^{C0 N} = e^{C0 D(M + r_Ω^{-1})}, not e^{C0 N^2} or e^{C0 N I}. Please expand this step: either reproduce the propagation inequality from [54, Theorem 2.1] with its explicit dependence on the number of balls, or give a direct induction from (12.31)–(12.33) in which each step multiplies the L^ε norm by a universal constant and the number of steps is bounded by the N in Proposition 12.2. This is a load-bearing point because any superlinear growth in N would destroy the optimized form of the main estimates.","section":"Section 12.3, Theorem 5.1"}],"minor_comments":[{"comment":"The symbol after \"f\" in the sentence \"(2.18) fails for n ≥ 2 and f /greaterornotdbleql0\" is a typesetting artifact; it should read \"f \\not\\equiv 0\" (or the intended relation).","section":"Remark 2.3"},{"comment":"In the displayed integral after (10.6), the factor \"(1-x)\" should be \"(1-|x|)\" to match the weight d(x) for the unit ball.","section":"Proof of Proposition 4.3"},{"comment":"In the OCR text the expressions \"d(x)1− n/q\" appear without proper superscripts; the intended expressions are d(x)^{1-n/q} in (2.19) and Theorem 3.4(iii).","section":"Theorems 2.9 and 3.4"},{"comment":"In the proof of part (ii), the inequality in the claim (12.19) would be easier to follow if the authors explicitly noted that r ≤ (120 k_Ω)^{-1/α} implies r/52 ≥ k_Ω(3r/2)^{1+α}, which is needed in the displayed chain of inequalities.","section":"Proposition 12.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is well within the scope of math.AP, and the novelty is clear. The main concern is the proof sketch of Theorem 5.1, which can be fixed by expanding the iteration argument; the rest of the technical content appears sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a serious paper and the main claims look right, but the proof of Theorem 5.1 is where I would send a referee before believing the headline constants. The new material is real: the duality argument for the Lipschitz bound with exponent linear in M, the first optimized Morel-Oswald/Hopf estimate for general divergence operators, the nonhomogeneous log-gradient estimate, and the counterexamples that match the claimed exponential rates. The authors are also unusually straight about what is open: Remark 2.2 on the unoptimized case, partial optimality of (2.19), the necessity of the spectral positivity assumption.\n\nI checked the one delicate computation the reader flagged, the adjoint-factorization step in Lemma 8.1 Step 3, and it is consistent: the drift coefficient and the identity div(b1hat)+chat = L* phi = 0 check out. The estimates are scale invariant and the counterexamples use the actual M of the constructed operators, so no parameter fitting is hidden in the optimality claims.\n\nThe soft spot is Theorem 5.1, the optimized global boundary weak Harnack inequality. It is the load-bearing result: Theorems 2.8, 2.9, 3.3, and 3.4 all pass through it, and its proof in Section 12.3 is not self-contained. The local estimates (12.31)-(12.33) are verified, but the Harnack-chain iteration is imported from [54] with \"we can repeat the iteration argument on [54, p.12]\". If that iteration gives a C0^N factor, then with the N bound from Proposition 12.2 the stated linear-exponential form follows. If it gave something like C0^{N^2}, the main estimates would lose their claimed linear-in-M form. I have no reason to think the imported constant is wrong, since these are the authors' own published results, but it is a genuine open check, not a cosmetic gap. A referee should work through that chain.\n\nAlso the abstract's optimality claim overstates slightly: the counterexamples cover the b2 and c dependencies, and the nonhomogeneous log-gradient estimate is only partly shown optimal. That is disclosed in the text, so it is a presentation issue, not a substantive flaw.\n\nWho this is for: specialists in elliptic regularity who care about sharp constants, and people working on Landis-type quantitative unique continuation. It deserves a serious referee. I would send it to review, and I would want the referee to focus on Section 12.3 rather than the main theorems.","headline":"Sharp-constant elliptic estimates, mostly solid and honest, with one load-bearing imported Harnack iteration that a referee must verify.","tokens_in":56532,"tokens_out":1775,"would_cite":true,"duration_ms":17179,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35B45","35B50","35B65","35J15","35P15"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves optimized constants for the main elliptic regularity estimates and supplies counterexamples showing none can be improved.","keywords":["optimized constants","elliptic regularity","C^{1,α} estimates","Harnack inequality","Hopf–Oleinik lemma","Landis conjecture","first eigenvalue estimates","Morel–Oswald inequality"],"falsifier":"Compute the ratio in (2.2) for the explicit radial solutions in Propositions 4.1 and 4.3 (for example u = $e^{{λ(1−|x|^2)/2}}$−1 with operators −Δ+c and −Δ+b·∇) as λ grows: if the ratio sup v/d divided by $e^{{C0 M R}}$ $R^{{1−n/q}}$||f||_q is unbounded, the exponential constant is wrong; alternatively, run the Section 12.3 iteration on the half-ball with b=λx and c=0 to check whether the weak Harnack constant grows like $e^{{Cλ}}$ rather than $e^{{Cλ^2}}$.","tokens_in":55259,"feed_emoji":"📐","tokens_out":8099,"duration_ms":70699,"temperature":0.7,"pith_summary":"Classical a priori estimates for uniformly elliptic equations in divergence form carry constants that depend on the coefficients and the size of the domain, but the classical proofs leave that dependence far from sharp. This paper establishes versions of the generalized maximum principle, the $C^{{1,α}}$ estimate, the quantitative Hopf lemma, the boundary weak Harnack inequality, and the logarithmic gradient estimate in which the constant is optimized: the exponential growth is linear in the coefficient measure M and the geodesic diameter D of the domain. It also shows by counterexamples that no slower growth in M, R, or D is possible for the drift and potential coefficients b2 and c, and it uses the optimized estimates to prove a Landis-type lower bound on ∫_{∂B_R}|u| and upper and lower bounds for the first eigenvalue. A sympathetic reader should care because these are the constants that matter whenever the estimates are used to prove existence, uniqueness, decay, or spectral statements.","feed_headline":"Elliptic estimates reach their optimal constants","feed_subtitle":"Sharpened constants for five classical elliptic estimates, with counterexamples proving each bound is final.","key_machinery":"The machine is the optimized boundary weak Harnack inequality in Theorem 5.1, combined with a duality argument using the adjoint operator, its positive first eigenfunction, and an integration-by-parts lemma. The quantity M in (1.7) collects uniformly local norms of A, b1, b2, c raised to exponents βq and γq, together with a radius r0 in (3.2) chosen so that after rescaling each local operator has uniformly bounded coefficients; this is what converts classical non-optimized estimates into estimates with linear-in-M exponents. A Harnack chain whose length is controlled by the geodesic diameter D carries the estimates across the domain, and optimality is shown by explicit radial or eigenfunction-based solutions.","core_discovery":"The paper's central claim is that, for operators L satisfying (1.3)-(1.4) on bounded $C^{{1,ᾱ}}$ domains, the sharp a priori estimates take the forms stated in Theorems 2.1–2.9 and 3.1–3.4. In particular, sup_{B_R} v/d ≤ C0 $e^{{C0 M R}}$ $R^{{1−n/q}}$ ||f^+||_{L^q(B_R)} holds under div(b1)+c ≤ 0, with M defined in (1.7) through uniformly local coefficient norms and a coefficient-adapted radius r0; the quantitative Hopf estimate u(x) ≥ $e^{{−C0(1+MR)}}$ $R^{{−n}}$ ||f||_{$L^{1}$_d} d(x) and the log-gradient estimate d|∇u|/u ≤ C0 max{1,Md} also hold with linear-in-M exponents. The paper proves these bounds and, in Propositions 4.1–4.4, constructs sequences of operators with growing b2 or c for which the exponential in M is attained, so the dependence cannot be improved.","pith_inferences":["Beyond the paper: the same coefficient-adapted rescaling that powers the proofs should carry the optimized constants to operators with VMO or merely continuous leading coefficients near the boundary, since only uniformly local norms enter M; the Hölder assumptions in (1.4) feed the iteration but may not be the true threshold.","Beyond the paper: the quantitative Hopf bound with the L^1_d norm is built from a duality with the first eigenfunction of the adjoint, so the same argument should yield Green-function lower bounds with explicit constants for operators with first-order terms, which currently are only known in special cases.","Beyond the paper: the n=1 inhomogeneous log-gradient bound suggests that the failure of the estimate for n≥2 and sign-changing f is a dimension effect; a finite-difference check on the explicit example in (11.16) would test whether a positive-part condition on f suffices in higher dimensions."],"forward_implications":["The Landis lower estimate becomes ∫_{∂B_R}|u| dσ ≥ e^{−C0 K R} ∫_{B_R}|u| for any weak solution of Lu = 0 in R^n with λ1(−L,R^n) ≥ 0, where K is the explicit coefficient data in (2.14).","The first eigenvalue satisfies λ1(−L,B_R) ≤ C0(M+R^{-1})^2, and if λ1(−L,B_{R+κ}) ≥ 0 then λ1(−L,B_R) ≥ e^{−C0(M+κ^{-1})R} R^{−2}.","The quantitative Hopf–Oleinik inequality holds for general divergence operators: u(x) ≥ e^{−C0(1+MR)} R^{−n} ||f||_{L^1_d(B_R)} d(x).","For positive solutions of homogeneous equations, d|∇u|/u ≤ C0 max{1,Md}, and for the inhomogeneous equation with f ≥ 0 an additional term involving ||f||_{L^q}/||f||_{L^1_d} appears.","No estimate with slower growth in the drift or potential coefficients is possible: Propositions 4.1–4.4 exhibit operators attaining the exponential factor in M."],"supporting_citations":[{"why":"Supplies the boundary weak Harnack inequality that Theorem 5.1 adapts; the quantitative Hopf and Harnack results rest on it.","marker":"[50]"},{"why":"Provides the Harnack chain iteration and the Landis-type lower bound that this paper extends to general divergence operators.","marker":"[54]"},{"why":"The classical generalized maximum principle and C^{1,α} estimates whose non-optimized constants are the baseline being improved.","marker":"[22]"},{"why":"Proves a restricted exponential C^1 bound for A=Id, b1=0, b2,c bounded; Theorem 2.1(ii) generalizes and sharpens it.","marker":"[4]"},{"why":"Establishes the Morel–Oswald inequality for the Laplacian, extended here to general operators with optimized constants.","marker":"[8]"},{"why":"Gives the two-dimensional log-gradient bound with linear dependence on the potential that Theorem 2.9 extends to general operators and dimensions.","marker":"[30]"},{"why":"Supplies the doubling lemma used in the second proof of the log-gradient estimate.","marker":"[49]"},{"why":"Provides the W^{1,q} regularity and C^{1,α} estimates used in Theorem 3.2 and Lemma 8.1.","marker":"[44]"}],"fun_headline_variants":["Elliptic estimates reach optimal constants","Sharp elliptic estimates with proven optimality","Elliptic regularity: constants are best possible","Optimal constants for elliptic estimates shown","Elliptic bounds proven best via counterexamples"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the optimized boundary weak Harnack inequality (Theorem 5.1), whose proof imports an iteration argument from earlier results rather than re-deriving it in full; if that imported constant were larger than claimed, the linear-in-M exponents in the main estimates would degrade.","fun_headline_variants_meta":{"raw":{"variants":["Elliptic estimates reach optimal constants","Sharp elliptic estimates with proven optimality","Elliptic regularity: constants are best possible","Optimal constants for elliptic estimates shown","Elliptic bounds proven best via counterexamples"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000748,"raw_usage":{"total_tokens":3290,"prompt_tokens":861,"completion_tokens":2429,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":477,"completion_tokens_details":{"reasoning_tokens":2381}},"tokens_in":477,"tokens_out":2429,"duration_ms":17304,"temperature":1.0,"reasoning_tokens":2381,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:16:38.137299+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the ratio in (2.2) for the explicit radial solutions in Propositions 4.1 and 4.3 (for example u = $e^{{λ(1−|x|^2)/2}}$−1 with operators −Δ+c and −Δ+b·∇) as λ grows: if the ratio sup v/d divided by $e^{{C0 M R}}$ $R^{{1−n/q}}$||f||_q is unbounded, the exponential constant is wrong; alternatively, run the Section 12.3 iteration on the half-ball with b=λx and c=0 to check whether the weak Harnack constant grows like $e^{{Cλ}}$ rather than $e^{{Cλ^2}}$.","supporting_citations":[{"cited_title":"Rend´ on, B","cited_arxiv_id":null,"evidence_quote":"Supplies the boundary weak Harnack inequality that Theorem 5.1 adapts; the quantitative Hopf and Harnack results rest on it."},{"cited_title":"Sirakov, Ph","cited_arxiv_id":null,"evidence_quote":"Provides the Harnack chain iteration and the Landis-type lower bound that this paper extends to general divergence operators."},{"cited_title":"Gilbarg, N","cited_arxiv_id":null,"evidence_quote":"The classical generalized maximum principle and C^{1,α} estimates whose non-optimized constants are the baseline being improved."},{"cited_title":"Le Balc’h, Exponential bounds for gradient of solutio ns to linear elliptic and parabolic equations, J","cited_arxiv_id":null,"evidence_quote":"Proves a restricted exponential C^1 bound for A=Id, b1=0, b2,c bounded; Theorem 2.1(ii) generalizes and sharpens it."},{"cited_title":"Brezis, X","cited_arxiv_id":null,"evidence_quote":"Establishes the Morel–Oswald inequality for the Laplacian, extended here to general operators with optimized constants."},{"cited_title":"Kenig, L","cited_arxiv_id":null,"evidence_quote":"Gives the two-dimensional log-gradient bound with linear dependence on the potential that Theorem 2.9 extends to general operators and dimensions."},{"cited_title":"Pol´ aˇ cik, P","cited_arxiv_id":null,"evidence_quote":"Supplies the doubling lemma used in the second proof of the log-gradient estimate."},{"cited_title":"Morrey, Multiple integrals in the calculus of vari ations, Classics Math","cited_arxiv_id":null,"evidence_quote":"Provides the W^{1,q} regularity and C^{1,α} estimates used in Theorem 3.2 and Lemma 8.1."}],"review_version":1}