{"id":"088e9c9f-320d-4c78-91d9-689e8b51bca8","arxiv_id":"2507.06150","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Almost every level set of the viscosity solution to obstacle-constrained mean curvature flow is a distributional (BV) solution of the obstacle problem.","lead":"This paper builds a level-set version of mean curvature flow that respects obstacles, and proves that almost every level set is a weak solution to the obstacle problem. It is the first construction of generic BV (bounded variation) solutions for mean curvature flow with obstacles.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The L1 bound (18) assumed in Theorem 3.1 is not verified for well-prepared data and is load-bearing; the reader's quadratic counterexample is flawed, but the missing justification remains a genuine gap.","rationale":"The reader correctly identifies the initial L1 bound (18) as the weakest point in the argument: Theorem 3.1 assumes it, Theorems 3.5 and 3.7 use it, and Definition 2.2 does not state it. However, the reader's concrete counterexample g(x)=x^2 is incorrect: in one dimension the regularized curvature is O(1) in L1, not O(1/epsilon). A correct flat example may or may not disprove (18); the manuscript simply never settles the question. Because the central claim of Theorem 1.2 rests on the validity of (18), and because no derivation of (18) from the stated hypotheses is provided, the paper is not fully justified as written. The condition is plausibly automatic for C^2 initial data, in which case the gap is easily fixed by adding a lemma; if not, Definition 2.2 must be strengthened. Either way, the current proof is conditional on an unverified estimate, matching the reader's CONDITIONAL verdict. I do not see a deeper flaw in the energy-convergence, Young-measure, or coarea arguments once (18) is supplied, so REJECT is not warranted.","tokens_in":28519,"tokens_out":44354,"duration_ms":525705,"concrete_test":"Prove or disprove: for every g in C^2(T^d), sup_{0<epsilon<1} || div(grad g / sqrt(epsilon^2+|grad g|^2)) ||_{L1} < infinity. As a decisive special case, take a smooth flat function such as g(x)=exp(-1/|x|^alpha) (extended periodically, with constant obstacles far below/above) and compute the L1 norm of the regularized curvature as epsilon -> 0; if it diverges, Definition 2.2 must be augmented with (18), while if it is bounded, a short lemma should be added to the paper. Also evaluate the quadratic example g=x^2 on a periodic domain to confirm the integral is O(1), which tests the reader's proposed counterexample.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1.2 depends on Theorem 3.1's uniform L1 estimate (19) for -H_epsilon + f_epsilon. That estimate is proved only under hypothesis (18), namely sup_epsilon || -H_epsilon(0) + f_epsilon(0) ||_{L1} < infinity. Definition 2.2 of well-prepared data does not include (18), and the paper never proves that smooth data satisfying Definition 2.2 satisfy it. This is load-bearing because Theorem 3.5's energy convergence and Theorem 3.7's compensated-compactness argument both require the uniform L1 bound on -H_epsilon + f_epsilon; without it, the signed measures appearing in the proof are not controlled and the derivation of (50)-(53), and hence of the level-set BV solution, collapses. The reader's specific example g(x)=x^2 in one dimension is not a valid counterexample: with f_epsilon(0)=0, the regularized curvature is -div(g'/sqrt(epsilon^2+g'^2)) = -2 epsilon^2/(epsilon^2+4x^2)^{3/2}, whose L1 norm is O(1), not O(1/epsilon). However, the underlying concern survives: the authors do not state or prove that (18) is automatic for C^2 data with phi <= g <= psi. The condition may be true, but it is a nontrivial pointwise-a.e. estimate near critical sets of g, and the manuscript does not supply the argument. Until either (18) is derived from the hypotheses or explicitly added to Definition 2.2, the proof of the main theorem has an unverified load-bearing hypothesis.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies mean curvature flow with obstacles φ ≤ u ≤ ψ. It adds to the Evans–Spruck geometric vanishing-viscosity approximation a quartic obstacle penalty term and proves that the regularized solutions converge to the unique viscosity solution of the obstacle level-set problem. The main theorem asserts that, for well-prepared initial data and for almost every level γ, the superlevel sets {u > γ} are distributional BV solutions of the obstacle problem in the sense of Definition 2.1, with the obstacle contact encoded through inequalities in the motion law. The proof combines a uniform L1 estimate for the signed quantity −Hε + fε with an energy-convergence and compensated-compactness argument, and then uses coarea and layer-cake formulas to pass from the level-set equation to individual level sets.","tokens_in":28881,"tokens_out":25000,"duration_ms":301163,"significance":"If established in the stated generality, the result would provide the first weak solution concept for obstacle mean curvature flow that describes the behavior at contact with the obstacles, going beyond Brakke-type solutions that are only valid away from the obstacles. The paper is clearly written and contains detailed proofs; the use of the signed quantity −Hε + fε and the contact-set sign estimates in Lemma 3.6 are original and natural. However, the key uniform L1 estimate is assumed at time zero rather than derived, and this assumption is not a consequence of the well-preparedness definition. Moreover, the assumption can fail for a C² initial datum that satisfies the paper's stated hypotheses. This makes the main theorem currently unproven for the full class of data announced in Theorem 1.2.","major_comments":[{"comment":"The uniform initial L1 bound (18) is load-bearing and is not implied by Definition 2.2. In fact it fails for some well-prepared data. On T¹, choose 0 < α < 1 and a periodic C² function g with g''(x) = x^α sin(log(1/x)) near x = 0, extended smoothly and periodically away from 0. Such g can be bracketed by C¹ obstacles φ < ψ with only finitely many critical points, so Definition 2.2 is satisfied. The zeros x_n of g' satisfy x_n ≈ e^{−π n} and |g''(x_n)| ≈ x_n^α. Since Hε(0) = −ε² g''/(ε² + (g')²)^{3/2}, each critical point with x_n ≫ ε^{1/(α+1)} contributes approximately 2 to ∫ |Hε(0)| dx, and there are O(log(1/ε)) such points. Hence the left side of (18) grows like (π(α+1))^{-1} log(1/ε), so (18) is false for this admissible initial datum. A simple quadratic profile has O(1) L1 norm and is not a counterexample, but the oscillatory construction above is a genuine one.","section":"Theorem 3.1, Eq. (18)"},{"comment":"Because (18) is false for some data satisfying Definition 2.2, the proofs of Theorems 3.5 and 3.7 do not apply to the full stated class. Theorem 3.5 uses Theorem 3.1 to extract a signed measure µ^t with uniformly bounded variation; without (18) the family (−Hε + fε) dx need not be tight. This signed measure is then used in Eqs. (35)–(46) to prove |∇uε| ⇀ |∇u|, and Section 4 uses the resulting convergences (50)–(53) to define the velocity field V and derive the level-set motion law. Thus the main theorem rests on an unverified hypothesis at a central step. The authors should either prove (18) under a strengthened, well-motivated condition on g, or add (18) to Definition 2.2 and verify that the examples in Remark 2.3 satisfy it.","section":"Theorems 3.5, 3.7 and 1.2"},{"comment":"The mismatch between Definition 2.2 and hypothesis (18) should be resolved in the statement of the main result. The current well-preparedness condition restricts only the critical sets of φ and ψ; it says nothing about the critical set of g or the behavior of ∇g near its zeros. Since (18) concerns exactly the singular behavior of ∇g near its critical points, the theorem's hypotheses and its proof are not aligned. Restricting Theorem 1.2 to data satisfying (18) would make the proof correct, but it would require showing that the restricted class contains geometrically meaningful examples; such a verification is not present in the current version.","section":"Definition 2.2 / Theorem 1.2"}],"minor_comments":[{"comment":"In the definition of Ω_T^+ inside the proof, the inequality should involve uε rather than u; as written, the notation is inconsistent with the surrounding equations and with the intended maximum-principle argument.","section":"Proposition 3.3, proof"},{"comment":"The proof of the relabeling property is omitted. Since Theorem 4.7 uses it for arbitrary smooth increasing F in the layer-cake argument, the paper would be easier to verify if a proof or a precise reference covering the obstacle case were included.","section":"Theorem 5.6"},{"comment":"There are several minor typographical issues: in Remark 2.3, '∂U = {ψ = 0}' should read '{x : ψ(x) = 0} = ∂U'; in Eq. (54), the convergence is on Ω × (0, ∞), not on R^d; and in Definition 2.1, Γ_t is defined only for almost every t, so the essential supremum in (7) should be phrased accordingly.","section":"Remark 2.3 and Eq. (54)"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the unverified hypothesis (18); it is a genuine mathematical gap, not a presentation problem, because (18) can fail for a C² initial datum satisfying the paper's stated assumptions. If the authors re-scope the main theorem to a class for which (18) holds and verify that class, I expect the remaining arguments to be sound. I did not find issues with the contact-set machinery or with the layer-cake passage beyond this load-bearing point. Given the paper's level of detail, a careful revision addressing this gap is feasible."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Tim and Keisuke have written a genuinely useful paper. The main theorem is the first BV/distributional solution concept for MCF with obstacles, extending [UL24] to the constrained setting, and the proof strategy is coherent: singular penalty with exponent 4, L1 estimate for -Hε+fε, compensated compactness, then layer-cake/coarea to pass to generic level sets. I believe the result is new and likely correct.\n\nThe paper also does honest work on the viscosity side: the obstacle-aware comparison principle in the appendix is written out carefully, and the authors correctly attribute the exponent-2 penalty idea to Mercier. The critical-point finiteness condition in Definition 2.2 is not decorative; it is used in Theorem 4.4 to control the contact sets.\n\nNow the soft spot, and it is real. The entire proof leans on Theorem 3.1, which needs the initial L1 bound (18): sup_ε ∫ |−Hε(0)+fε(0)| < ∞. For well-prepared data fε(0)=0, so this is a bound on ∫ |Hε(0)|. The paper never states this as part of well-preparedness and never proves it. Theorem 3.5 uses Theorem 3.1 to get the uniform signed-measure bound that drives the compensated compactness, but Theorem 3.5's statement omits (18) entirely. So as written, the main theorem has an unverified load-bearing hypothesis.\n\nThe reader's specific counterexample g(x)=x² in 1D is wrong: the regularized curvature is O(ε²/(ε²+4x²)^{3/2}), with L1 norm of order 1, not 1/ε. But the underlying concern survives. It might be that (18) is automatic for C² data with φ≤g≤ψ and finite critical sets for φ,ψ — the scaling near an isolated critical point of g suggests O(1) — but the manuscript does not contain the argument, and Theorem 3.5 needs it at every time.\n\nMy recommendation: this deserves a serious referee and, with a modest revision, likely acceptance. The cleanest fix is to add (18) to Definition 2.2 or to prove it in a lemma before Theorem 3.1. Since the rest of the proof is coherent and the gap is local, I would not reject.","headline":"A solid and likely correct extension of Ullrich-Laux to obstacle-constrained MCF, but the main theorem currently rests on an unverified initial L1 bound that the authors need to either state as a hypothesis or prove.","tokens_in":29418,"tokens_out":10564,"would_cite":true,"duration_ms":107640,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["53E10","35K65","35A15","35D30","35D40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that almost every super-level set of the unique viscosity solution to the level-set obstacle problem for mean curvature flow is a distributional solution to the geometric obstacle flow.","keywords":["mean curvature flow","obstacle problem","level set method","viscosity solution","BV solution","compensated compactness","coarea formula","singular perturbation"],"falsifier":"A direct check is to take one-dimensional data $g(x)=x^2$ between constant obstacles and compute $H_\\varepsilon(0)$: the regularized curvature is $-2/\\varepsilon$ at the origin, so the pointwise size of the quantity controlled by (18) diverges. If such data fall under the paper's well-preparedness definition, then the uniform initial $L^1$ bound is not a consequence of that definition and Theorem 3.1 must supply an additional mechanism. A second check is the proof step asserting $|\\nabla(-H_\\varepsilon+f_\\varepsilon)|=0$ almost everywhere on the set $\\{-H_\\varepsilon+f_\\varepsilon=0\\}$, which is not true for generic smooth functions.","tokens_in":28292,"feed_emoji":"🌀","tokens_out":11208,"duration_ms":129474,"temperature":0.7,"pith_summary":"The paper studies a surface moving by its mean curvature while it is forbidden to enter two fixed regions, the obstacles. It approximates the flow by a smooth level-set equation with a singular penalty that makes violation of the constraint very expensive, and then lets the penalty vanish. The central claim is that for almost every height $\\gamma$, the evolving super-level set $\\{u(\\cdot,t)>\\gamma\\}$ of the limiting level-set function is a weak BV solution of the obstacle flow: it has uniformly bounded perimeter, moves with a square-integrable normal velocity, dissipates energy, and obeys the correct inequalities where it touches the lower or upper obstacle. If this is right, it supplies the first general weak-solution framework that describes the motion at contact with the obstacles, not only away from them.","feed_headline":"Generic level sets solve mean curvature flow with obstacles","feed_subtitle":"Almost every super-level set of the viscosity solution satisfies the flow's weak equations, including at contact.","key_machinery":"The engine is the modified geometric vanishing-viscosity equation $$\\partial_t u_\\varepsilon=\\left(\\nabla\\cdot\\frac{\\nabla u_\\varepsilon}{|\\nabla u_\\varepsilon|_\\varepsilon}-V_\\varepsilon'(u_\\varepsilon)\\right)|\\nabla u_\\varepsilon|_\\varepsilon,\\qquad |p|_\\varepsilon=\\sqrt{\\$varepsilon^{2}$+|p|^2},$$ with quartic penalty $V_\\varepsilon(u)=\\frac{1}{\\varepsilon}(\\phi-u)_+^4+\\frac{1}{\\varepsilon}(u-\\psi)_+^4$. The load-bearing estimate is the uniform $L^1$ bound $\\sup_{\\varepsilon,t}\\int_\\Omega|{-H_\\varepsilon+f_\\varepsilon}|\\,dx<\\infty$, where $H_\\varepsilon=\\operatorname{div}(\\nabla u_\\varepsilon/|\\nabla u_\\varepsilon|_\\varepsilon)$ is the regularized mean curvature and $f_\\varepsilon=-V_\\varepsilon'(u_\\varepsilon)$; because this bound has no area factor $|\\nabla u_\\varepsilon|_\\varepsilon$, it survives at the contact sets. Together with convexity of the energy and its $\\Gamma$-convergence to the total variation with constraints, the bound feeds a compensated-compactness argument that yields the strict convergence $|\\nabla u_\\varepsilon|\\rightharpoonup|\\nabla u|$ and $-\\nabla u_\\varepsilon/|\\nabla u_\\varepsilon|_\\varepsilon\\to -\\nabla u/|\\nabla u|$ away from flat regions. Coarea and layer-cake formulas then transfer this averaged equation to almost every individual level set.","core_discovery":"On the paper's own terms, the discovery is Theorem 1.2: for well-prepared initial data $g$ and $C^1$ obstacles $\\phi<\\psi$, the unique viscosity solution $u$ of the constrained level-set equation has the property that for almost every $\\gamma\\in\\mathbb{R}$, the super-level set $U(t)=\\{u(\\cdot,t)>\\gamma\\}$ is a distributional solution of mean curvature flow with obstacles $\\Phi=\\{\\phi>\\gamma\\}$ and $\\Psi=\\{\\psi>\\gamma\\}$ in the sense of Definition 2.1. That means the set has finite perimeter uniformly in time and a normal velocity $V$ with $\\int V^2\\,d\\mathcal{H}^{d-1}dt<\\infty$; it satisfies the transport equation $\\int_0^\\infty\\int_{U_t}\\partial_t\\zeta\\,dx\\,dt=-\\int_0^\\infty\\int_{\\Gamma_t}\\zeta V\\,d\\mathcal{H}^{d-1}dt-\\int_{U_0}\\zeta(\\cdot,0)\\,dx$ and the energy dissipation inequality, and it satisfies the motion-law inequality against vector fields that are admissible at the obstacle boundaries. Away from obstacles this inequality reduces to the weak form of normal velocity equals minus mean curvature; at the lower obstacle it encodes $V\\ge -H$ and at the upper obstacle $V\\le -H$.","pith_inferences":["The method is silent on the measure-zero set of levels where the viscosity solution fattens; at those levels the super-level set may not be a well-defined geometric evolution, and the theorem deliberately excludes them.","The quartic exponent in the penalty is a regularity device; the same scheme with a quadratic penalty was proposed earlier, and the proof suggests that any strictly convex penalty strong enough to force the constraint in the $\\Gamma$-limit could play the same role, though the $L^1$ estimate would need rechecking.","A natural testable extension is to obstacles with flat boundary portions: the proof uses lower bounds on $|\\nabla\\phi|$ and $|\\nabla\\psi|$ near the contact sets, so flat obstacle boundaries would require a different argument and may change the contact condition.","The construction gives a candidate notion of normal velocity on the contact set itself, where the flow is not determined by curvature alone; this could be compared with the long-time limit of the flow toward the mean-convex hull of the obstacle."],"forward_implications":["Almost every super-level set of the viscosity solution has uniformly bounded $(d-1)$-dimensional perimeter over time, so the obstacle flow produces a well-defined family of finite-perimeter sets.","Each such level set carries a square-integrable normal velocity $V$ satisfying the weak transport equation; in particular the evolution of the sets is encoded by an $L^2$ velocity rather than only by a comparison principle.","The energy inequality $\\mathcal{H}^{d-1}(\\Gamma_{T'})+\\int_0^{T'}\\int_{\\Gamma_t}V^2\\,d\\mathcal{H}^{d-1}dt\\le \\mathcal{H}^{d-1}(\\Gamma_0)$ holds for almost every level, giving the obstacle analogue of the standard BV-solution dissipation relation.","On the contact sets the motion-law inequality produces one-sided inequalities $V\\ge -H$ at the lower obstacle and $V\\le -H$ at the upper obstacle, which is exactly the information missing in earlier obstacle-flow constructions that only track the motion away from obstacles.","Because the viscosity solution is unique up to fattening, the vanishing-penalty limit selects one distinguished weak evolution for almost every level."],"supporting_citations":[{"why":"Supplies the level-set viscosity solution framework and the geometric vanishing-viscosity approximation that the obstacle construction modifies.","marker":"[ES91]"},{"why":"Supplies the BV/distributional solution machinery, including the convergence-of-normals argument extended here to the obstacle setting.","marker":"[ES95]"},{"why":"Proposed the penalized level-set equation for obstacles and provides the viscosity-solution existence used in Theorem 1.1.","marker":"[Mer14]"},{"why":"Provides comparison and uniqueness for viscosity solutions of the obstacle problem without a graph assumption.","marker":"[IKK17]"},{"why":"Defines the BV-solution notion that Definition 2.1 adapts to obstacles.","marker":"[LS95]"},{"why":"Establishes the generic-level-set-BV-solution result for mean curvature flow without obstacles, which this paper extends.","marker":"[UL24]"},{"why":"Supplies the Young-measure and weak-compactness tools used in Theorem 3.7.","marker":"[Eva90]"},{"why":"Supplies the coarea and rectifiability facts used to pass from level-set equations to individual superlevel sets.","marker":"[AFP00]"}],"fun_headline_variants":["Generic level sets satisfy obstacle MCF","Almost every level set obeys MCF with obstacles","Unique viscosity solutions for obstacle flow","Generic super-level sets meet obstacle constraints"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, for all admissible initial data, the combination of the regularized curvature and the penalty force stays uniformly integrable in space and time as the approximation parameter goes to zero (assumption (18)); the paper assumes this rather than deriving it from its well-preparedness conditions, and smooth initial data with a critical point make the pointwise size of this quantity blow up like $1/\\varepsilon$.","fun_headline_variants_meta":{"raw":{"variants":["Generic level sets satisfy obstacle MCF","Almost every level set obeys MCF with obstacles","Unique viscosity solutions for obstacle flow","Generic super-level sets meet obstacle constraints"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000209,"raw_usage":{"total_tokens":1377,"prompt_tokens":883,"completion_tokens":494,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":441}},"tokens_in":499,"tokens_out":494,"duration_ms":6039,"temperature":1.0,"reasoning_tokens":441,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:10:29.814134+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct check is to take one-dimensional data $g(x)=x^2$ between constant obstacles and compute $H_\\varepsilon(0)$: the regularized curvature is $-2/\\varepsilon$ at the origin, so the pointwise size of the quantity controlled by (18) diverges. If such data fall under the paper's well-preparedness definition, then the uniform initial $L^1$ bound is not a consequence of that definition and Theorem 3.1 must supply an additional mechanism. A second check is the proof step asserting $|\\nabla(-H_\\varepsilon+f_\\varepsilon)|=0$ almost everywhere on the set $\\{-H_\\varepsilon+f_\\varepsilon=0\\}$, which is not true for generic smooth functions.","supporting_citations":[],"review_version":1}