Generic mean curvature flow with obstacles

Keisuke Takasao; Tim Laux

REVIEW 3 major objections 3 minor 17 references

Generic mean curvature flow with obstacles

T0 review · 3 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read 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.

desk verdict 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. read the letter →

arxiv 2507.06150 v1 pith:BWOEJDBM submitted 2025-07-08 math.AP math.DG

Tim Laux , Keisuke Takasao This is my paper · ORCID

classification math.APmath.DG MSC 53E1035K6535A1535D3035D40

keywords meancurvatureflowobstacleproblemlevelsetmethodviscositysolutionBVcompensatedcompactnesscoareaformulasingularperturbation

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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 $V\le -H$.

Load-bearing premise

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 $1/\varepsilon$.

Editorial extensions

If this is right

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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

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.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the 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.

The 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.

Now 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.

The 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.

My 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.

Referee Report

3 major / 3 minor

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.

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 (3)

[Theorem 3.1, Eq. (18)] 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.
[Theorems 3.5, 3.7 and 1.2] 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.
[Definition 2.2 / Theorem 1.2] 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.

minor comments (3)

[Proposition 3.3, proof] 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.
[Theorem 5.6] 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.
[Remark 2.3 and Eq. (54)] 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.

Circularity Check

0 steps flagged · score 2.0 of 10

No constructional circularity: the obstacle-penalty derivation is self-contained, and the unverified L1 hypothesis (18) is a correctness gap, not a circular reduction.

full rationale

The paper's central derivation is not circular by construction. The penalized Evans–Spruck approximation (16) with the obstacle penalty V_epsilon is new input, and the proof chain (Theorem 3.1 -> Theorem 3.5 -> Theorem 3.7 -> Theorem 4.7 -> Theorem 1.2) does not define its conclusion in terms of its hypotheses. In particular, the key estimate (6) is proved under the explicit hypothesis (18) on the initial data; this hypothesis is not a renamed version of the claimed BV-solution property and is not derived from Definition 2.2. That is a genuine missing justification, but it is a correctness gap rather than a circular step, since the subsequent compensated-compactness argument genuinely proves strict convergence from (18), (A1)-(A4) rather than assuming it. The self-citations [UL24], [Tak21], and [NT25] appear in the introduction as contextual background and for comparison, but the proofs of the main theorems do not invoke them as load-bearing justifications; the technical arguments explicitly extend [ES91] and [ES95], which are independent works. The relabeling property used in Theorem 4.7 is stated as Theorem 5.6 with proof omitted and attributed to [ES91], not to a self-citation. No fitted parameters are involved, and no prediction is renamed from a fit. Accordingly, the score is low: minor non-load-bearing self-references are present, but the central claim does not reduce to its own inputs.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The singular penalty potential Vε is a construction, not a fitted quantity or a new physical entity. The only load-bearing addition not justified from the stated hypotheses is the initial L1 bound (18), which the main theorem needs but does not prove.

assumptions (4)

ad hoc to paper Initial L1 bound (18): sup_{ε∈(0,1)} ∫_Ω |−Hε(x,0)+fε(x,0)| dx < ∞.
Assumed in Theorem 3.1 but not derived from Definition 2.2 in Theorem 1.2; can fail for smooth g with critical points. This is an unproved input to the main theorem.
standard math Comparison principle and relabeling property for viscosity solutions (Theorems 5.4 and 5.6).
Imported from [ES91] and [Mer14]; used to establish uniqueness and the relabeling F∘u, a key step in Theorem 4.7.
domain assumption Well-preparedness: critical sets of φ and ψ are finite and inequalities (13) hold.
Definition 2.2; used in Theorem 4.4 to control neighborhoods of critical points via (74).
standard math Coarea and layer-cake formulas, BV theory, and compensated compactness via Young measures.
Used in Lemma 4.5 and Theorem 3.7; standard results from [AFP00] and [Eva90].

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generic mean curvature flow with obstacles." pith.science (2026). https://pith.science/paper/BWOEJDBM

@misc{pith2026250706150,
  author       = {Pith},
  title        = {Pith review of: Generic mean curvature flow with obstacles},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BWOEJDBM}},
  note         = {Machine review of arXiv:2507.06150}
}

read the original abstract

We study the obstacle problem associated to mean curvature flow. We add to the geometric vanishing-viscosity approximation of Evans and Spruck a singular perturbation that penalizes the violation of the constraint, and pass to the limit. The resulting level set formulation has unique solutions - up to fattening. Extending the work of Evans and Spruck and a work by Ullrich and one of the authors, we show that generic level sets of this flow are distributional solutions of the obstacle problem.

Figures

Figures reproduced from arXiv: 2507.06150 by the authors.

**Figure 1.** A mean curvature flow with obstacles (left) and a corresponding level set flow (right; for simplicity, one spatial dimension is omitted here). Away from the contact sets {u = ϕ} and {u = ψ}, this formulation should encode the classical level set mean curvature flow (2) ∂tu = ∇ · ∇u |∇u| |∇u|, while on the contact sets, we need to encode the inequalities (3) ∂tu ≥ ∇ · ∇u |∇u| |∇u| on {u = ϕ} and ∂tu ≤ ∇ · ∇… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 15 canonical work pages

[1]

Almeida, A

L. Almeida, A. Chambolle, and M. Novaga. Mean curvature flow with obstacles. Ann. Inst. H. Poincar\'e C Anal. Non Lin\'eaire , 29(5):667--681, 2012

work page 2012
[2]

Functions of bounded variation and free discontinuity problems

Luigi Ambrosio, Nicola Fusco, and Diego Pallara. Functions of bounded variation and free discontinuity problems . Oxford Mathematical Monographs. The Clarendon Press, Oxford University Press, New York, 2000

2000
[3]

Taylor, and Lihe Wang

Fred Almgren, Jean E. Taylor, and Lihe Wang. Curvature-driven flows: a variational approach. SIAM J. Control Optim. , 31(2):387--438, 1993

work page 1993
[4]

An algorithm for mean curvature motion

Antonin Chambolle. An algorithm for mean curvature motion. Interfaces Free Bound. , 6(2):195--218, 2004

work page 2004
[5]

Mean curvature evolution of entire graphs

Klaus Ecker and Gerhard Huisken. Mean curvature evolution of entire graphs. Ann. of Math. (2) , 130(3):453--471, 1989

work page 1989
[6]

L. C. Evans and J. Spruck. Motion of level sets by mean curvature. I . J. Differential Geom. , 33(3):635--681, 1991

1991
[7]

Evans and Joel Spruck

Lawrence C. Evans and Joel Spruck. Motion of level sets by mean curvature. IV . J. Geom. Anal. , 5(1):77--114, 1995

work page 1995
[8]

Lawrence C. Evans. Weak convergence methods for nonlinear partial differential equations , volume 74 of CBMS Regional Conference Series in Mathematics . Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 1990

work page 1990

Show all 17 references

[9]

Ishii, H

K. Ishii, H. Kamata, and S. Koike. Remarks on viscosity solutions for mean curvature flow with obstacles. In Mathematics for nonlinear phenomena---analysis and computation , volume 215 of Springer Proc. Math. Stat. , pages 83--103. Springer, Cham, 2017

2017
[10]

An obstacle problem for mean curvature flow

Philippe Logaritsch. An obstacle problem for mean curvature flow . Phd thesis, Universit\"at Leipzig, 2016

2016
[11]

Implicit time discretization for the mean curvature flow equation

Stephan Luckhaus and Thomas Sturzenhecker. Implicit time discretization for the mean curvature flow equation. Calc. Var. Partial Differential Equations , 3(2):253--271, 1995

1995
[12]

Mean curvature flow with obstacles: a viscosity approach

Gwenael Mercier. Mean curvature flow with obstacles: a viscosity approach. arXiv:1409.7657 [math.AP] , 2014

2014 arXiv
[13]

Mean curvature flow with obstacles: existence, uniqueness and regularity of solutions

Gwena\"el Mercier and Matteo Novaga. Mean curvature flow with obstacles: existence, uniqueness and regularity of solutions. Interfaces Free Bound. , 17(3):399--426, 2015

2015
[14]

On an obstacle problem for the B rakke flow with a generalized right-angle boundary condition

Katerina Nik and Keisuke Takasao. On an obstacle problem for the B rakke flow with a generalized right-angle boundary condition. SIAM J. Math. Anal. , 57(1):452--494, 2025

2025
[15]

Mean-convex sets and minimal barriers

Emanuele Spadaro. Mean-convex sets and minimal barriers. Matematiche (Catania) , 75(1):353--375, 2020

2020
[16]

On obstacle problem for B rakke's mean curvature flow

Keisuke Takasao. On obstacle problem for B rakke's mean curvature flow. SIAM J. Math. Anal. , 53(6):6355--6369, 2021

2021
[17]

Generic level sets in mean curvature flow are BV solutions

Anton Ullrich and Tim Laux. Generic level sets in mean curvature flow are BV solutions. J. Geom. Anal. , 34(12):Paper No. 375, 18, 2024

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Almeida, A

L. Almeida, A. Chambolle, and M. Novaga. Mean curvature flow with obstacles. Ann. Inst. H. Poincar\'e C Anal. Non Lin\'eaire , 29(5):667--681, 2012

work page 2012

[2] [2]

Functions of bounded variation and free discontinuity problems

Luigi Ambrosio, Nicola Fusco, and Diego Pallara. Functions of bounded variation and free discontinuity problems . Oxford Mathematical Monographs. The Clarendon Press, Oxford University Press, New York, 2000

2000

[3] [3]

Taylor, and Lihe Wang

Fred Almgren, Jean E. Taylor, and Lihe Wang. Curvature-driven flows: a variational approach. SIAM J. Control Optim. , 31(2):387--438, 1993

work page 1993

[4] [4]

An algorithm for mean curvature motion

Antonin Chambolle. An algorithm for mean curvature motion. Interfaces Free Bound. , 6(2):195--218, 2004

work page 2004

[5] [5]

Mean curvature evolution of entire graphs

Klaus Ecker and Gerhard Huisken. Mean curvature evolution of entire graphs. Ann. of Math. (2) , 130(3):453--471, 1989

work page 1989

[6] [6]

L. C. Evans and J. Spruck. Motion of level sets by mean curvature. I . J. Differential Geom. , 33(3):635--681, 1991

1991

[7] [7]

Evans and Joel Spruck

Lawrence C. Evans and Joel Spruck. Motion of level sets by mean curvature. IV . J. Geom. Anal. , 5(1):77--114, 1995

work page 1995

[8] [8]

Lawrence C. Evans. Weak convergence methods for nonlinear partial differential equations , volume 74 of CBMS Regional Conference Series in Mathematics . Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 1990

work page 1990