{"id":"3c7126cc-5280-4a19-b2d7-6eda3d14f275","arxiv_id":"2501.00928","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The Gamma-convergence result is correct, but the abstract monotonicity theorem used to prove the polygonal-boundary result is false.","lead":"This paper studies how well a convex shape can be approximated from inside by a smaller convex shape of fixed area, measuring the gap with L^p distances between support functions. It proves existence and Gamma-convergence to the Hausdorff distance, and claims planar optimal boundaries are polygonal, but a key abstract theorem is false.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reader's Theorem 3 objection does not land; the real gap is Theorem 2's proof for 1<p<2, where m'' is singular and the stated L^2 bound is false.","rationale":"The reader's identified weakest assumption is the strict monotonicity in Theorem 3, supposedly false because Proposition 11 infers monotonicity from absence of local minima. That criticism does not withstand scrutiny: under Hypothesis (D), f cannot have a local minimum at any x<sup I, including boundary points, since any local minimum of f would force a local minimizer of J in C. The proof as written only treats interior points, but the gap is an omitted endpoint argument, not a false theorem. Therefore the rejection based on Theorem 3 is not justified.\n\nA genuine but different concern is the applicability of [20, Theorem 2.9] for 1<p<2. The second derivative of m is singular, and the displayed inequality bounding m'' by a constant times the L^2 norm of the variation is false in that range. The paper does not establish that the hypotheses of [20] hold for these p, although the theorem is stated for all p in [1,infty). This is a concrete, load-bearing gap in the proof of the paper's main structural result. It is conditional rather than a definite refutation because a suitable H^s estimate or a separate argument for 1<p<2 might repair the proof.","tokens_in":17847,"tokens_out":24985,"duration_ms":258786,"concrete_test":"Take Omega the unit disk and a support function h with hOmega-h behaving like |theta| near theta=0, so contact occurs at one point. For 1<p<2, choose v_n(theta)=sqrt(n) on an interval of length 1/n around 0 and 0 elsewhere. Compute int (hOmega-h)^{p-2} v_n^2 dtheta: it diverges like n^{2-p} while ||v_n||^2_{L^2}=1, contradicting the uniform beta bound stated in Section 3.2. Then consult [20, Theorem 2.9] to determine whether only an H^s bound is actually needed; if so, the proof may be repairable, but as written Theorem 2 is unproven for 1<p<2 and should be restricted to p>=2 or given an additional argument.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The reader's rejection hinges on Proposition 11, where the proof passes from absence of local minima of f on the interior of I to strict monotonicity on all of I. That inference is written too quickly, but it is repairable: Hypothesis (D) in Theorem 3 forbids local minima of f at every x < sup I, including endpoints, because a local minimum of f at x0 would make any minimizer Omega0 with F(Omega0)=x0 a local minimizer of J in C. Thus Theorem 3 is not the sound basis for rejecting the paper.\n\nThe real load-bearing gap is in the proof of Theorem 2 for 1 < p < 2. The proof defines m(h)=int (hOmega-h)^p and needs the bound ||m''(h)(v,v)|| <= beta ||v||^2_{L^2} to apply [20, Theorem 2.9]. For 1<p<2, m''(h)(v,v)=p(p-1)/2 int (hOmega-h)^{p-2} v^2 is singular near the contact set where hOmega-h=0: the weight is unbounded. The asserted L^2 bound fails; for v_n supported on an interval of length 1/n near a contact point with ||v_n||_{L^2}=1, the integral grows like n^{2-p}. The paper does not verify whether [20] can instead use an H^s bound, and it does not restrict the statement to p>=2. Since Theorem 2 is claimed for every p in [1,infty), the central structural result is not established for 1<p<2; p=1 is separately treated in Proposition 13.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the shape optimization problem of finding a convex subset ω of a fixed convex container Ω, with prescribed measure, that minimizes the L^p distance between the support functions of ω and Ω. The main claims are: existence of minimizers and Γ-convergence to the Hausdorff-distance problem as p→∞ (Theorem 1); in the plane, the free part of any optimal boundary is a union of polygonal lines (Theorem 2); and an abstract equivalence theorem (Theorem 3) that is used to replace the measure constraint by a distance constraint in the proof of Theorem 2. The paper also proposes a numerical scheme combining Fourier parametrization with a convexity-preserving discretization.","tokens_in":18163,"tokens_out":16772,"duration_ms":158532,"significance":"The Γ-convergence result and the numerical method are useful contributions, and the polygonal-regularity statement is natural and would be significant if proven. However, the central structural result (Theorem 2) rests on an abstract theorem (Theorem 3) that is false as stated, and on a second-order estimate that fails for 1<p<2. These are load-bearing gaps, not presentation issues, so the paper's main theoretical claims are not established.","major_comments":[{"comment":"Theorem 3 is false as stated. The proof of Proposition 11 only shows that f has no local minimum in the interior of I, and then concludes that f is strictly decreasing. This inference is invalid: a continuous function on an interval with no interior local minima may have local maxima or be increasing. A concrete counterexample satisfying hypotheses (A)-(D) is C = {[0,t] : t∈[0,1]} with the Hausdorff distance, F([0,t])=t, J([0,t])=t, and Ψ_{[0,t]}(x)=[0,x]. Then I=[0,1], all hypotheses hold, and f(t)=t is strictly increasing, contradicting the asserted conclusion. Consequently, Proposition 6, which relies on Theorem 3, is not proven, and the derivation of the equivalent formulation used in the proof of Theorem 2 is unsupported.","section":"Section 3.3, Proposition 11"},{"comment":"The estimate ||m''(h)(v,v)||_{L^2} ≤ β||v||^2_{L^2} is false for 1<p<2. With m(h)=∫(hΩ-h)^p dθ, the second derivative is m''(h)(v,v)=p(p-1)∫(hΩ-h)^{p-2}v^2 dθ. When p<2, the weight (hΩ-h)^{p-2} is unbounded near the contact set {h=hΩ}. For a sequence v_n supported on an interval of length 1/n near a contact point with ||v_n||_{L^2}=1, the integral grows like n^{2-p}, so the L^2 bound fails. Since Theorem 2 is claimed for all p∈[1,∞), the application of [20, Theorem 2.9] is not justified for 1<p<2. The separate treatment of p=1 in Proposition 13 invokes Theorem 2 and therefore inherits this gap.","section":"Section 3.2, proof of Theorem 7"}],"minor_comments":[{"comment":"The constraint in problem (8) is written as ∫(hΩ-h)^p dθ = f(c), but since J_p is defined as the p-th root of the integral, the correct constraint should involve f(c)^p (or a renamed constant). This is a notational inconsistency that should be fixed.","section":"Section 3.2, equation (8)"},{"comment":"In the inferior-limit part of the continuity proof, the line 'F(Ω∗)=x' should read 'F(Ω∗)=x0', since the limit point is x0.","section":"Section 3.3, proof of Proposition 11"},{"comment":"The angle notation '[CAB ≥ [CBA' is garbled; it should be written as ∠CAB ≥ ∠CBA.","section":"Section 4, Conjecture 1"},{"comment":"There is a typo 'on an other note' which should be 'on another note'.","section":"Section 1, introduction"}],"recommendation":"reject","confidential_remarks":"The Γ-convergence part (Theorem 1) is correct and the numerical experiments are interesting, but the main structural theorem (Theorem 2) depends on a false abstract theorem and on an invalid second-order estimate for a substantial range of p. These are not local fixes; they require either a substantial new proof for Theorem 2 or a significant restriction of the claims. The paper is not ready for publication in its current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"My take: the paper's core Gamma-convergence result is correct and clean, and the p=1 reformulation as a reverse isoperimetric problem is a nice observation. The numerical comparison between Fourier-only optimization and Bogosel's discretization is also informative, though it stops short of a real benchmark.\n\nThe reader's rejection targets the wrong spot. The supposed counterexample to Theorem 3 does not hold up: a continuous value function on an interval that has a local maximum but no local minimum anywhere in [inf I, sup I) cannot exist, since the left endpoint or the turn-around point would be a local min. Hypothesis (D) forbids local minima of J in C, and any local min of f at x<sup I would make the corresponding minimizer a local min of J. So the proof of Proposition 11 is incomplete—it only handles interior local minima—but the gap is repairable. Theorem 3 is likely true as stated.\n\nThe real problem is Theorem 2 for 1<p<2. The proof needs ||m''(h)(v,v)|| ≤ β||v||^2_{L^2}, where m''(h)(v,v)=p(p-1)/2∫(h_Ω-h)^{p-2}v^2. For p<2, the weight is singular near the contact set. A sequence v_n with ||v_n||_{L^2}=1 supported on an interval of length 1/n near a contact point makes the integral grow like n^{2-p}, so the bound fails. The paper supplies no H^s-based substitute and does not restrict the statement to p≥2. For p=1, the second derivative is not defined, and Proposition 13 leans on Theorem 2, so the p=1 case is unsupported too.\n\nThe numerics are suggestive but not strong evidence: no code, no error analysis, just a few energy comparisons.\n\nOverall: Theorem 1 is a modest, correct contribution. The main advertised structural result (Theorem 2) is not established for a substantial range of p. A serious referee should engage because the gap may be repairable—either by a sharper second-order estimate or by restricting the theorem to p≥2—but I would not cite the paper for Theorem 2 yet.","headline":"Gamma-convergence part is solid, but the proof of the main regularity theorem fails for 1<p<2 due to a singular second derivative; the reader's Theorem 3 objection is mis-targeted.","tokens_in":18675,"tokens_out":13240,"would_cite":false,"duration_ms":119731,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q10","52A20","49J45"],"pacs":[],"model":"deepseek-v4-flash","headline":"Optimal convex subsets in the plane have polygonal free boundaries, and the $L^p$ problem converges to Hausdorff as $p$ grows.","keywords":["shape optimization","convex bodies","support functions","Lp approximation","Gamma-convergence","Hausdorff distance","polygonal boundary","reverse isoperimetric problem"],"falsifier":"Compute $f(c)$ numerically on a fine grid for a simple container such as a triangle or disk, with a fixed $p$; if $f$ shows a strict local minimum at any interior value of $c$, the monotonicity behind Proposition 6 fails and the proof of the polygonal-boundary theorem loses its foundation.","tokens_in":17638,"feed_emoji":"📐","tokens_out":14495,"duration_ms":121071,"temperature":0.7,"pith_summary":"This paper studies a shape optimization problem: given a convex container $\\Omega\\subset\\mathbb{R}^n$, find the convex subset $\\omega$ of a prescribed measure $c$ that best approximates $\\Omega$ in the $L^p$ distance between support functions, $\\mathcal{J}_p(\\omega)=\\|h_\\Omega-h_\\omega\\|_{L^p(\\mathbb{S}^{n-1})}$, for $1\\le p<\\infty$. The authors prove that minimizers exist in every dimension, that $\\mathcal{J}_p$ $\\Gamma$-converges to the Hausdorff distance $\\mathcal{J}_\\infty$ as $p\\to\\infty$, and that in the plane every optimal shape has a free boundary made of polygonal lines. The planar structural result is the paper's central claim, because it says that even smooth-looking $L^p$ approximations develop flat edges; the paper then uses that prediction to design a numerical scheme that resolves the straight segments. The contribution is the reduction of the geometric problem to an analytic one and the transfer of a known planar regularity theory to this setting.","feed_headline":"Best Lp convex subsets grow straight edges","feed_subtitle":"Proof that minimizing Lp distance to a convex container forces flat boundary segments, plus a scheme that captures them.","key_machinery":"The support function $h_\\omega(\\theta)=\\sup_{y\\in\\omega}\\langle \\theta,y\\rangle$ is the central object; it encodes a convex set analytically, turns inclusion into pointwise inequality $h_\\omega\\le h_\\Omega$, expresses area as $\\frac12\\int_0^{2\\pi}(h^2-h'^2)\\,d\\theta$, and converts the geometric approximation problem into minimizing $\\int_{\\mathbb{S}^{n-1}}|h_\\Omega-h_\\omega|^p\\,d\\mathcal{H}^{n-1}$ subject to convexity $h_\\omega''+h_\\omega\\ge0$. The load-bearing mechanism is the reduction captured by Theorem 3: it shows that, under the stated hypotheses, minimizing $J$ at fixed $F=x$, minimizing $J$ under $F\\le x$, and minimizing $F$ at fixed $J=f(x)$ are all equivalent. That equivalence turns the measure constraint into a distance constraint and brings the planar problem into reach of a regularity theorem for optimal convex shapes, which forces the free boundary to consist of straight segments.","core_discovery":"The central claim is Theorem 2: in two dimensions, if $\\omega^*$ solves $(P_p)$, then the free part of its boundary, $\\partial\\omega^*\\setminus\\partial\\Omega$, is a union of polygonal lines; in particular, a polygonal container forces a polygonal optimizer. The proof route is Theorem 3, a general statement about two continuous shape functionals $J$ and $F$ on a compact class: under local perturbation hypotheses and a no-local-minimum condition, the value function $f(x)=\\min\\{J(\\omega):F(\\omega)=x\\}$ is continuous and strictly decreasing, and the fixed-constraint problem is equivalent to the one-sided constraint problem. Applied with $F=|\\cdot|$ and $J=\\mathcal{J}_p$, this replaces the area constraint by a constraint on $\\mathcal{J}_p$, an equivalence that lets the authors invoke a known planar regularity theory for optimal convex shapes and conclude that the free boundary is polygonal. The paper also establishes existence of minimizers and $\\Gamma$-convergence to the Hausdorff problem in any dimension.","pith_inferences":["A direct numerical test of the planar theorem would run the proposed scheme on a disk container and check whether every converged optimizer contains at least one straight boundary segment; the theory predicts flat pieces for every finite $p$.","If the general reduction theorem is correct, the same equivalence pattern should hold for other pairs of geometric functionals, such as perimeter and area, giving a reusable template for turning constrained shape problems into one-sided ones.","The $\\Gamma$-convergence result suggests the polygonal-boundary phenomenon is stable as $p$ grows; one could track the number and length of flat segments along a sequence of optimizers for increasing $p$.","In higher dimensions the natural analogue would be polyhedral free boundaries, but the paper explicitly notes the needed regularity theory is unavailable there, leaving open whether flat facets appear in $\\mathbb{R}^3$ or new singularities emerge."],"forward_implications":["Every problem $(P_p)$ has a solution: for any convex body $\\Omega\\subset\\mathbb{R}^n$, any $p\\in[1,\\infty)$, and any $c\\in[0,|\\Omega|]$, there is a convex $\\omega\\subset\\Omega$ with $|\\omega|=c$ attaining the infimum of $\\mathcal{J}_p$.","As $p\\to\\infty$, the minimal values $\\sigma_p$ converge to $\\sigma_\\infty$, and every Hausdorff accumulation point of $L^p$-optimal shapes solves the Hausdorff-distance problem; the $L^p$ problems are thus approximations of the limiting problem.","In the plane, the free part of any optimal boundary is a union of polygonal lines, and if the container is a polygon the optimizer is a polygon too.","At $p=1$, the problem is exactly a reverse isoperimetric problem: maximizing perimeter among convex subsets of $\\Omega$ with area $c$, and every optimizer touches $\\partial\\Omega$ in at least two points and has straight free boundary components.","The numerical experiments show that the Fourier-coefficient method alone misses segments and overshoots the energy, while combining it with the discrete convexity parametrization yields lower-energy shapes with flat edges."],"supporting_citations":[{"why":"Supplies the planar regularity theorem that, once the problem is reformulated, forces free boundary parts to be polygonal lines.","marker":"[20]"},{"why":"Defines the Hausdorff-distance minimization problem that $(P_p)$ approximates and provides the $p=\\infty$ equivalence used for comparison.","marker":"[17]"},{"why":"Provides the support-function toolkit: the convexity criterion $h''+h\\ge0$, inclusion as pointwise inequality, and area and perimeter formulas.","marker":"[31]"},{"why":"Introduces the discrete convexity parametrization that the paper combines with Fourier coefficients to capture straight segments numerically.","marker":"[4]"},{"why":"Gives the fundamental theorem of $\\Gamma$-convergence used to pass from convergence of functionals to convergence of minimizers and values.","marker":"[6]"},{"why":"Establishes that the $L^p$ support-function distances are metrics on convex bodies, grounding the choice of $\\mathcal{J}_p$.","marker":"[34]"},{"why":"Solves the reverse isoperimetric problem in a ball, the comparison case for the paper's $p=1$ section and conjecture.","marker":"[5]"}],"fun_headline_variants":["Lp-optimal convex subsets get polygonal free edges","Optimal Lp rounding of convex sets yields straight segments","Planar proof: free boundary of Lp minimizers is polygonal","Gamma-limit: Lp approximation turns into Hausdorff for large p","New Fourier-scheme combo spots flat edges in Lp convex fits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the minimal approximation error $f(c)=\\min\\{\\mathcal{J}_p(\\omega):|\\omega|=c\\}$ is strictly decreasing in $c$; the proof of Theorem 3 derives this monotonicity from the absence of local minima, an inference that is not generally valid, and Proposition 6 uses the monotonicity to replace the measure constraint with a distance constraint.","fun_headline_variants_meta":{"raw":{"variants":["Lp-optimal convex subsets get polygonal free edges","Optimal Lp rounding of convex sets yields straight segments","Planar proof: free boundary of Lp minimizers is polygonal","Gamma-limit: Lp approximation turns into Hausdorff for large p","New Fourier-scheme combo spots flat edges in Lp convex fits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000664,"raw_usage":{"total_tokens":3081,"prompt_tokens":1042,"completion_tokens":2039,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":658,"completion_tokens_details":{"reasoning_tokens":1948}},"tokens_in":658,"tokens_out":2039,"duration_ms":12184,"temperature":1.0,"reasoning_tokens":1948,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:40:52.371371+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $f(c)$ numerically on a fine grid for a simple container such as a triangle or disk, with a fixed $p$; if $f$ shows a strict local minimum at any interior value of $c$, the monotonicity behind Proposition 6 fails and the proof of the polygonal-boundary theorem loses its foundation.","supporting_citations":[{"cited_title":"Lamboley, A","cited_arxiv_id":null,"evidence_quote":"Supplies the planar regularity theorem that, once the problem is reformulated, forces free boundary parts to be polygonal lines."},{"cited_title":"Ftouhi and E","cited_arxiv_id":null,"evidence_quote":"Defines the Hausdorff-distance minimization problem that $(P_p)$ approximates and provides the $p=\\infty$ equivalence used for comparison."},{"cited_title":"Schneider","cited_arxiv_id":null,"evidence_quote":"Provides the support-function toolkit: the convexity criterion $h''+h\\ge0$, inclusion as pointwise inequality, and area and perimeter formulas."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the discrete convexity parametrization that the paper combines with Fourier coefficients to capture straight segments numerically."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the fundamental theorem of $\\Gamma$-convergence used to pass from convergence of functionals to convergence of minimizers and values."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes that the $L^p$ support-function distances are metrics on convex bodies, grounding the choice of $\\mathcal{J}_p$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Solves the reverse isoperimetric problem in a ball, the comparison case for the paper's $p=1$ section and conjecture."}],"review_version":1}