{"id":"1b1df537-4f35-4602-a777-556e4873e868","arxiv_id":"2607.07018","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":4,"one_line_summary":"Gradient continuity (C¹-regularity) for p-Laplacian obstacle problems is established under Dini mean oscillation conditions on coefficients, source term, and obstacle, weakening classical Dini continuity assumptions.","lead":"The paper proves that solutions to p-Laplacian obstacle problems have continuous gradients when the data (coefficients, source, obstacle) satisfy a Dini mean oscillation condition rather than classical Dini continuity. This extends known regularity theory by weakening the regularity assumptions on input data from pointwise continuity to mean oscillation.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"Unjustified WLOG assumption on q = (2-p)(2+κ)/κ in the I₂ estimate for 1<p<2 may restrict the admissible parameter range","rationale":"The reader correctly identified the 'without loss of generality' assumption as one of the key gaps. This is the most load-bearing concern because it sits directly in the comparison estimate pipeline: Lemma 3.2 → Corollary 3.3 → Lemma 3.4 → the iteration in Section 4. If the I₂ estimate for 1<p<2 does not actually hold for all κ>0, then Theorem 1.1 as stated (for all 1<p<∞) has a gap in the range 1<p<2 when p is near 2. The other concerns raised by the reader — reliance on the preprint [1] for Lemma 2.9, the comparison principle citation, and the 'similar argument' references — are real but are standard practice in PDE papers and typically resolve upon checking. The WLOG assumption is different: it is a mathematical claim that may simply be false for certain parameter combinations, and the authors provide no workaround. The paper's overall structure is sound and the result is a genuine extension, but this specific gap prevents unconditional acceptance. The condition for p>2 being 'slightly stronger than expected' (acknowledged by the authors) is a separate, less critical issue — it means the result may not be sharp but does not threaten correctness.","tokens_in":20919,"tokens_out":914,"duration_ms":495171,"concrete_test":"Re-derive the I₂ estimate for 1<p<2 in Lemma 3.2 without assuming (2−p)(2+κ)/κ > p. Specifically, take p=1.9 and κ=0.2 (so q=(0.1)(2.2)/0.2=1.1 < p=1.9) and check whether the Hölder inequality application and the subsequent use of (2.9) with q=1.1 and t=1/p still yield the bound on I₂ in terms of ω_{data}(r,x₀)². If the estimate fails, determine whether an alternative choice of q in (3.1) or a different splitting of the integrand recovers it.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In the proof of Lemma 3.2, when estimating I₂ for the case 1<p<2, the authors state: 'where we use the inequality 0<2−p<p and assume that (2−p)(2+κ)/κ > p without loss of generality.' This assumption is not, in fact, without loss of generality. The quantity q = (2−p)(2+κ)/κ is already fixed in equation (3.1) as the exponent used in the Calderón–Zygmund estimate (2.9) with t=1/p. The condition q > p translates to (2−p)(2+κ)/κ > p, i.e., κ < 2(2−p)/(2p−2) = 2(2−p)/(2(p−1)). For p close to 2 (say p=1.9), this requires κ < 2(0.1)/(1.8) ≈ 0.111, while for p close to 1 (say p=1.1), κ < 2(0.9)/(0.2) = 9, which is easily satisfied. So the restriction bites when p is near 2. But Condition 2.3 only requires κ > 0 to be arbitrary, and the Dini condition on ω_data must hold. If a user has data satisfying the Dini condition with some κ₀ > 0 but κ₀ ≥ 2(2−p)/(2(p−1)), the proof as written does not apply. The authors do not explain how to handle this case — they simply assert WLOG. If the estimate for I₂ in the regime q ≤ p requires a different Hölder exponents arrangement or a different choice of q in (3.1), the comparison estimate (3.9) and hence Corollary 3.3 may fail, breaking the iteration in Section 4.","agreement_with_reader":"agree"},"referee_report":{"model":"glm-5.2","summary":"This paper establishes C^1-regularity (gradient continuity) for solutions to obstacle problems associated with p-Laplacian type equations with a coefficient matrix A, source term F, and obstacle ψ. The main result (Theorem 1.1) replaces the classical Dini continuity assumption on the data with a Dini mean oscillation (DMO) type condition (Condition 2.3). The proof proceeds by deriving L^1-comparison estimates between the solution u and solutions to frozen-coefficient problems (Section 3), then using an iteration argument (Section 4) to obtain local boundedness of Du and finally continuity. The framework adapts techniques from [20] (for non-obstacle nonlinear equations) and [32] (for obstacle problems with Dini-continuous data) to the DMO setting.","tokens_in":21827,"tokens_out":1284,"duration_ms":200398,"significance":"The result is a genuine extension of known C^1-regularity theory: it weakens Dini continuity to a mean oscillation condition for obstacle problems, which is the natural borderline for gradient continuity. The adaptation of the sharp L^1-comparison and iteration framework from [20] to the variational inequality setting is non-trivial, as the obstacle introduces additional comparison terms (the I_2 estimate in Lemma 3.2). The result also extends the source-term regularity in [27] for 1<p≤2. The paper is well-motivated and the proof structure is clear.","major_comments":[{"comment":"Lemma 3.2, I_2 estimate for 1<p<2 (p. 10): The authors state 'we use the inequality 0<2-p<p and assume that (2-p)(2+κ)/κ > p without loss of generality.' This assumption is not WLOG. The exponent q=(2-p)(2+κ)/κ is already fixed in (3.1) and used in the Calderón–Zygmund estimate (2.9) with t=1/p. The condition q>p translates to κ < 2(2-p)/(2(p-1)). For p close to 2 (e.g., p=1.9), this requires κ < 0.111, while Condition 2.3 only requires κ>0 to be arbitrary. If data satisfies the Dini condition with some κ_0 ≥ 2(2-p)/(2(p-1)), the proof as written does not apply. The authors need to either (a) justify that κ can always be chosen small enough without loss of generality (e.g., by noting that DMO_{2+κ} implies DMO_{2+κ'} for κ'<κ, so one can always reduce κ), or (b) handle the case q≤p separately. Without this, the comparison estimate (3.9) and Corollary 3.3 may not hold for all admissible κ","section":null}],"minor_comments":[{"comment":"p. 3, line below (1.3): 'Yonung's inequality' should be 'Young's inequality'. The same typo appears in the proof of Lemma 3.1.","section":null},{"comment":"p. 7, proof of Lemma 3.1: 'qusai-triangle' should be 'quasi-triangle'.","section":null},{"comment":"p. 10, I_2 estimate for 1<p<2: The phrase 'without loss generality' is missing 'of'.","section":null},{"comment":"Lemma 2.9 cites [1, Theorem 4.1], which is a very recent preprint. The authors should verify that the result in [1] applies exactly as stated (particularly the L^1 excess decay formulation in (2.12)) and clarify the dependency.","section":null},{"comment":"Condition 2.3 (p. 4): The definition of ω_data uses different exponents for p>2 and 1<p≤2. A brief remark explaining why the exponent p'/2 appears for p>2 but 2+κ for 1<p≤2 would help readers understand the structure of the condition.","section":null},{"comment":"Remark 2.4 (p. 5): The computation showing that θ_p-Dini continuity of Dψ implies p'/2-DMO_{p'} for p>2 is sketched but not fully detailed. Adding one or two lines of explanation would make the comparison with [32] more transparent.","section":null},{"comment":"p. 12, Lemma 3.4: The statement says 'there exists C_3' but the proof uses C_3 in the final estimate without explicitly tracking its dependence. The dependence on n, p, Λ should be stated explicitly.","section":null},{"comment":"Section 4, proof of Theorem 1.1: The 'Claim' on p. 18 and its proof involve a case analysis with four sub-cases (i)-(iv). The logic is correct but dense; a brief roadmap sentence before the case analysis would improve readability.","section":null},{"comment":"p. 17, proof of Proposition 4.3: The definition of M involves δ^{-2n}/(3ε_0^3), and the choice ε=ε_0=2^{-n-2} is made. It would help to explicitly note that with this choice, the conditions (4.2)–(4.3) are satisfied, since this is used in the iteration.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The stress-test concern about the WLOG assumption on q=(2-p)(2+κ)/κ in Lemma 3.2 is valid and is the most important point to address. However, I believe it is likely fixable: since DMO_{2+κ} implies DMO_{2+κ'} for any κ'<κ (smaller κ means weaker oscillation control), one can always choose κ small enough to ensure q>p. The authors should make this argument explicit. The reliance on [1] (a 2026 preprint) for Lemma 2.9 is a minor concern but the result is standard C^{1,α} regularity for frozen-coefficient p-Laplacian equations, so alternative references (e.g., [17, 10]) likely suffice."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"The main thing to know: this paper proves C¹-regularity for p-Laplacian obstacle problems under Dini mean oscillation conditions on the data, weakening the Dini continuity assumption from the second author's earlier work [32]. That is a real result — the DMO framework for nonlinear equations was developed in [20] for homogeneous equations, and extending it to variational inequalities with obstacles is nontrivial. The comparison estimates in Section 3 and the iteration in Section 4 are the technical core and are carefully structured. The authors earn credit for honestly flagging that their condition for p>2 is slightly stronger than expected, rather than hiding it. The proof architecture — comparison to frozen-coefficient problems, L¹ excess decay, induction on scales — is sound and follows established patterns adapted competently to the obstacle setting. The self-citations to [20] and [32] are appropriate since those results genuinely feed into the current argument; no circularity issue there. Now the soft spots. The stress-test concern about the WLOG assumption q = (2−p)(2+κ)/κ > p in the I₂ estimate for 1<p<2 is legitimate and is the most concrete gap. The exponent q is fixed in (3.1) and used in the CZ estimate (2.9) with t=1/p. The condition q > p translates to κ < 2(2−p)/(2(p−1)), which bites when p is near 2. The authors assert this is WLOG but do not explain why — if a user's data satisfies the Dini condition with some κ₀ that violates this bound, the proof as written does not cover that case. This needs either a justification or a separate argument for q ≤ p. The reliance on the very recent preprint [1] for the excess decay estimate (Lemma 2.9) is a dependency risk but not a logical flaw — the result is plausible and the citation is standard in form. The comparison principle w ≥ ψ invoked in Lemma 3.2 citing [7, Lemma 3.5] is used without proof, which is fine if that reference is correct but should be verified. Several estimates are deferred to 'by a similar argument,' which is normal for this level of paper but means a referee needs to check them carefully. This is a solid paper that deserves a serious referee. The main theorem is a genuine advance, the proof structure is correct, and the one real gap is localized and likely fixable. Recommend conditional acceptance pending resolution of the q > p issue for 1<p<2.","headline":"Genuine extension of C¹-regularity for p-Laplacian obstacle problems to DMO conditions; one gap in the 1<p<2 argument needs fixing","tokens_in":21741,"tokens_out":604,"would_cite":true,"duration_ms":154633,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"glm-5.2","headline":"Mean oscillation replaces Dini continuity for gradient regularity","keywords":[],"falsifier":"A counterexample would be a p-Laplacian obstacle problem where the coefficient matrix, source, and obstacle satisfy the DMO condition of Condition 2.3 but the gradient Du fails to be continuous — for instance, if Du is locally unbounded at some point despite the Dini mean oscillation integrability. Alternatively, if the cited Calderón–Zygmund estimates or the excess decay estimate for the frozen-coefficient equation fail for obstacle problems with VMO coefficients, the main comparison estimates would not hold.","tokens_in":21059,"feed_emoji":"📐","tokens_out":900,"duration_ms":148002,"temperature":0.7,"pith_summary":"The paper proves that the gradient of a solution to a p-Laplacian obstacle problem is continuous provided the data — the coefficient matrix, the source term, and the obstacle — satisfy a Dini mean oscillation condition rather than the classical Dini continuity condition. The obstacle problem is a variational inequality: one seeks a function u that stays above a given obstacle ψ and satisfies a nonlinear divergence-form equation involving a coefficient matrix A and a source F. The p-Laplacian structure, governed by the exponent 1 < p < ∞, makes the equation nonlinear in the gradient. The authors show that what matters for gradient continuity is not pointwise continuity of the data (the classical Dini condition) but rather that the mean oscillation of the data over small balls, raised to an appropriate power, is integrable in a Dini sense. The proof proceeds by comparing the solution u to auxiliary functions v solving frozen-coefficient equations on nested balls, deriving sharp L¹ comparison estimates that track the mean oscillation of the data, and then running an iteration that controls the excess decay of Du across scales. The key mechanism is that the DMO condition, combined with Calderón–Zygmund L^q estimates for q > p, provides enough control at each scale to close the iteration and force the gradient excess to zero.","feed_headline":"Mean oscillation, not pointwise continuity, suffices for smooth gradients","feed_subtitle":"Solutions to p-Laplacian obstacle problems have continuous gradients when the data satisfy a Dini mean oscillation condition — a strictly-we","key_machinery":"The argument uses a two-step comparison: first compare the solution u to an intermediate function w solving a frozen-coefficient obstacle problem, then compare w to a function v solving a homogeneous frozen-coefficient equation. The comparison estimates (Lemma 3.2, Corollary 3.3, Lemma 3.4) bound the L¹ difference of Du and Dv in terms of ω_data(r, x₀), the local mean oscillation of the data. The C^{1,α} excess decay estimate for the frozen-coefficient equation (Lemma 2.9) provides decay at each scale. An iteration (Lemma 4.2) then controls the excess E_j across dyadic scales, and the Dini condition on ω_data ensures the accumulated oscillation errors are summable, forcing Du to be locally L","core_discovery":"The gradient Du of a solution to the p-Laplacian obstacle problem is continuous whenever the coefficient matrix A is DMO_{2+κ} for some κ > 0, and the nonlinear quantities |Dψ|^{p-2}Dψ and F satisfy a (p'/2)-DMO_{p'} condition (for p > 2) or DMO_{2+κ} condition (for 1 < p ≤ 2), with the combined mean oscillation function ω_data satisfying a Dini integrability condition. This replaces the classical assumption of Dini continuity of the data with a strictly weaker mean oscillation condition, extending the linear theory (p = 2) to the nonlinear p-Laplacian setting with obstacles.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Dini mean oscillation replaces Dini continuity for p-Laplacian obstacle gradients","Continuous gradients for p-Laplacian obstacle problems under weaker oscillation data","DMO conditions suffice for C¹ regularity in nonlinear p-Laplacian obstacle problems","Mean oscillation framework extends gradient regularity to p-Laplacian obstacle setting"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The proof relies on Calderón–Zygmund L^q estimates for the variational inequality with VMO coefficients and on a recent C^{1,α} excess decay estimate for the frozen-coefficient p-Laplacian, both cited from external results. If either of these estimates does not hold under the stated conditions, the comparison and iteration arguments would not close.","fun_headline_variants_meta":{"raw":{"variants":["Dini mean oscillation replaces Dini continuity for p-Laplacian obstacle gradients","Continuous gradients for p-Laplacian obstacle problems under weaker oscillation data","DMO conditions suffice for C¹ regularity in nonlinear p-Laplacian obstacle problems","Mean oscillation framework extends gradient regularity to p-Laplacian obstacle setting"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":558,"prompt_tokens":470,"completion_tokens":88,"prompt_tokens_details":null},"tokens_in":470,"tokens_out":88,"duration_ms":30348,"temperature":1.0,"reasoning_tokens":null,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-09T21:33:16.980814+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"A counterexample would be a p-Laplacian obstacle problem where the coefficient matrix, source, and obstacle satisfy the DMO condition of Condition 2.3 but the gradient Du fails to be continuous — for instance, if Du is locally unbounded at some point despite the Dini mean oscillation integrability. Alternatively, if the cited Calderón–Zygmund estimates or the excess decay estimate for the frozen-coefficient equation fail for obstacle problems with VMO coefficients, the main comparison estimates would not hold.","supporting_citations":[],"review_version":1}