{"id":"6dd66828-5bd5-49d9-b3a6-4ddc7e23a803","arxiv_id":"2501.00974","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"On a strictly mean-convex manifold, the least area of a cut in a fixed homology class equals the greatest flux of a unit-bounded divergence-free vector field across it.","lead":"This paper proves a continuous version of the max flow/min cut theorem on curved spaces with boundary, where the cut must lie in a prescribed topological class. It connects this geometric optimization problem to holographic entanglement and to Thurston's theory of geodesic laminations.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.5's p→1 limit uses the invalid inequality ∫|du_p| ≤ ∫|du_p|^p, blocking the max-flow equality (3.2) and the duality in Theorem 1.2.","rationale":"The reader identified two concerns: the strict mean convexity hypothesis and an unjustified inequality in Lemma 3.5. I focused on the latter because it is a concrete, load-bearing flaw in the proof of the central duality: without (3.2), the equality mass(C*) = max_F ∫_{C*} flux F is unproven. The mean convexity assumption is a geometric restriction that is clearly stated and independently motivated by the failure example in the introduction; it is not an internal inconsistency, and the proof of Theorem 3.3 appears plausible. By contrast, the inequality ∫|du_p| ≤ ∫|du_p|^p is simply false for functions with small gradients, and it appears at a critical junction. This is a correctness risk, not merely a limit of applicability. The theorem may still be true and the proof repairable, so I do not recommend REJECT; the reader's CONDITIONAL verdict remains appropriate. My recommendation is UNCHANGED, with the understanding that the authors should fix the p→1 limiting argument in Lemma 3.5. I agree with the reader's overall assessment but not exactly with the selection of the weakest assumption, hence 'partial' agreement.","tokens_in":14439,"tokens_out":9157,"duration_ms":77136,"concrete_test":"On the unit interval with boundary values 0 and a (0 < a < 1), the p-Laplacian minimizer is u_p(x) = ax, so ∫|du_p| = a and ∫|du_p|^p = a^p < a for p > 1; this directly falsifies the inequality used in Lemma 3.5. Then revisit the proof of Lemma 3.5: check whether the conclusion (3.2) can be recovered by replacing the invalid step with the correct inequality (∫|du_p|)^p ≤ ∫|du_p|^p, and by establishing uniform lower/upper bounds on ∫|du_p|^p as p → 1. If the corrected argument yields liminf ∫|du_p| ≤ ∫ du ∧ γ, the gap is repairable; otherwise the duality statement in Theorem 1.2 lacks a valid proof of the max-flow side.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Lemma 3.5, the proof derives (3.2) through the chain: inf_{u*}∫|du*| ≤ liminf ∫|du_p| ≤ liminf ∫|du_p|^p = liminf ∫ du_p ∧ γ_p. The second inequality is false in general: if |du_p| < 1 pointwise, then |du_p|^p < |du_p| for p > 1, so ∫|du_p|^p can be strictly smaller than ∫|du_p|. The correct inequality from Jensen/Hölder is (∫|du_p|)^p ≤ ∫|du_p|^p, i.e., ∥du_p∥_1 ≤ ∥du_p∥_p, which yields liminf ∫|du_p| ≤ liminf (∫|du_p|^p)^{1/p}, not ≤ liminf ∫|du_p|^p. This matters because the proof needs to identify the liminf of the L^1 energies with the limit of ∫ du_p ∧ γ_p = ∫|du_p|^p. The p-Laplacian minimizers can have |du_p| < 1 on sets of positive measure; for affine boundary data with small slope, |du_p| is uniformly less than 1, making the displayed inequality demonstrably false. Since this step is essential to establish (3.2), and (1.4) relies on (3.2), the central max-flow/min-cut equality is not proven as written. The gap may be repairable by using the L^1 ≤ L^p inequality together with additional energy bounds and compactness, but the current text does not provide that argument.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a continuous analogue of the max flow/min cut theorem on compact Riemannian manifolds with strictly mean convex boundary. The main result, Theorem 1.2, asserts that for a relative homology class α and a prescribed boundary cycle S, there is a mass-minimizing d−1-current C* with ∂C*=S and [C*]=α, and that its mass equals the supremum of the flux of all divergence-free vector fields bounded in L∞ by 1. The proof is built from least-gradient functions on the universal cover, a p-Laplacian approximation scheme to construct a dual calibration, and a formalism for currents on manifolds with boundary. The paper also derives Corollary 1.5, asserting that in dimension d≤7 the minimizer is a measured oriented minimal lamination, and discusses an anisotropic generalization and connections to Thurston's asymmetric metric and to holography.","tokens_in":14734,"tokens_out":5893,"duration_ms":57639,"significance":"If the main theorem is correct, it provides the first continuous max flow/min cut theorem that incorporates real-coefficient topology of the domain, going beyond earlier Euclidean and L1-spectral variants. The manuscript is conceptually valuable: it identifies the correct hypotheses (strict mean convexity, finite-mass boundary currents, real coefficients) with concrete counterexamples, and it connects least-gradient theory, p-Laplacian methods, and geometric measure theory. The paper is also honest about its tools: it relies on substantial external results rather than fitted parameters, and the presentation of the current-theoretic boundary formalism is a useful contribution in itself. However, the proof as written contains a load-bearing gap in Lemma 3.5, and the central equality (1.4) is not established without repairing it.","major_comments":[{"comment":"The proof of (3.2) contains an invalid inequality. The displayed chain asserts liminf_{p→1} ∫_M |du_p| dV ≤ liminf_{p→1} ∫_M |du_p|^p dV. For p>1, the inequality |du_p| ≤ |du_p|^p holds only when |du_p| ≥ 1; when |du_p| < 1 pointwise, the reverse inequality holds. The p-Laplacian minimizers considered here can certainly have |du_p| < 1 on sets of positive measure, for instance for affine boundary data with sufficiently small slope. The valid L1-Lp comparison is (∫_M |du_p| dV)^p ≤ ∫_M |du_p|^p dV, which yields only liminf ∫_M |du_p| dV ≤ liminf (∫_M |du_p|^p dV)^{1/p}, not the claimed bound. Since this chain is the step that identifies the limiting L1 energy with the limit of ∫_M du_p ∧ γ_p, the equality (3.2) is not proved as written. This gap is load-bearing: Theorem 1.2 relies on (3.2) for the existence of the dual calibration γ and hence for the max-flow/min-cut equality (1.4). The gap may be repairable by additional uniform energy bounds or a different compactness argument, but such an argument is not present in the manuscript.","section":"Section 3, Lemma 3.5"},{"comment":"The proof of Theorem 3.4 and the construction in Lemma 3.5 both work with the equivariant setting on the universal cover, but the transition from the p-Laplacian minimizers u_p to the least-gradient function u is not fully documented at the level of traces. In Lemma 3.5, after obtaining γ as a weak limit, the integration-by-parts identity uses that u has trace f; this is justified only if the trace of the weak limit of the p-Laplacian solutions is controlled. The manuscript invokes the normal trace theorem for γ_p but does not state a corresponding trace-compactness statement for the sequence u_p. This is a secondary point, but it should be clarified, especially because the p→1 limit is delicate and the same page contains the incorrect energy inequality described above.","section":"Section 3, Theorem 3.4 and Lemma 3.5 interaction"}],"minor_comments":[{"comment":"The sentence 'Since [C] = α and ∂S = C' should be 'Since [C] = α and ∂C = S'; as written it reverses the boundary relation.","section":"Introduction, paragraph after (1.2)"},{"comment":"The name 'Hahn-Banach' is misspelled as 'Hanh-Banach' in both occurrences.","section":"Sections 1.2 and 3, Lemma 3.5"},{"comment":"In the first paragraph of the proof, 'applied to M \\ N rather than M' should read 'applied to N \\ M rather than M'; the current phrase refers to the complement of N in M, which is empty in the intended construction.","section":"Section 3, Theorem 3.4"},{"comment":"The expression ∫_∂M (f−h)γ_p writes a function multiplied by a (d−1)-form without indicating scalar multiplication; for readability, use (f−h)γ_p with an explicit wedge or scalar product notation, since γ_p is a form.","section":"Section 3, Lemma 3.5"},{"comment":"The arrow in the displayed diagram for the boundary homomorphism is unlabelled; labelling it ∂ would help the reader connect it to the subsequent discussion.","section":"Section 2, Construction 2.2"}],"recommendation":"major_revision","confidential_remarks":"The main theorem is likely true and the architecture is sound, but the gap in Lemma 3.5 is exactly where the proof's central equality is established. I would encourage the editor to seek a revision in which the author either proves the missing L1 convergence of the p-Laplacian minimizers or replaces the p-Laplacian step with a different duality argument. The paper's scope and ambition are appropriate for the journal, and the presentation is otherwise strong."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. The main theorem is genuinely new: a continuous max flow/min cut statement on compact, strictly mean-convex Riemannian manifolds where the cut is a real current in a fixed relative homology class. That picks up the topological content the previous Strang/Nozawa results left out, and it directly addresses the Freedman–Headrick and Thurston setups. The paper also recovers the S^1 case of Thurston's theorem. The rational-chain special case is correctly flagged as already known; the real-coefficient lamination upgrade uses the author's own published [Bac24a] regularity theorem, which is not circular.\n\nThe proof, though, has a genuine gap in Lemma 3.5. The stress-test note is right: the chain\n\ninf ∫|du_*| ≤ liminf ∫|du_p| ≤ liminf ∫|du_p|^p\n\nuses an inequality that does not hold in general. For p>1, on the set where |du_p|<1, raising to the p-th power decreases the integrand. The correct Hölder bound is (∫|du_p|)^p ≤ ∫|du_p|^p, so the best you get is liminf ∫|du_p| ≤ liminf (∫|du_p|^p)^{1/p}, which does not obviously converge to ∫ du ∧ γ. This step is load-bearing: (3.2) is exactly what later gives the max-flow equality (1.4). I don't see the repair in the current text. It may be fixable — standard p-harmonic convergence plus energy bounds — but it needs to be written.\n\nWhat else? The current formalism in §2, especially the boundary operator in Construction 2.2, is careful and interesting. The examples (RP^2 torsion, the Cantor-set trace, the irrational foliation on the torus) do real explanatory work, and they justify why currents with real coefficients and finite mass are the right category. The author also states plainly which parts are conjectural (the Freedman–Headrick discretization) and which are borrowed (JMN18, Górny's theorem). The exposition is clear throughout.\n\nThe strict mean convexity assumption is essential, and the author admits the duality can fail without it (Figure 1, right). That is a real restriction, not a flaw. The only other soft spot is the heavy reliance on [JMN18] and [Gó24] for Theorem 3.4, but that is the standard framework for least-gradient functions and the adaptation to nonzero cohomology class seems plausible.\n\nMy recommendation: send this to a serious geometric measure theory referee. The gap is real but likely repairable, and the theorem statement is important enough to warrant referee time even if the current version needs revision. I wouldn't cite the theorem in my own work until the proof is fixed. But I'd bring it to the reading group.","headline":"A genuinely new topological max flow/min cut theorem, with a real but likely repairable gap in the p-Laplacian duality lemma.","tokens_in":15320,"tokens_out":4306,"would_cite":false,"duration_ms":37887,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q05","35J92"],"pacs":[],"model":"deepseek-v4-flash","headline":"On a compact Riemannian manifold with strictly mean convex boundary, the paper proves that the least-mass current in a prescribed relative homology class has mass exactly equal to the maximum flux of a divergence-free unit vector field…","keywords":["max flow/min cut theorem","currents","functions of least gradient","measured oriented laminations","mean curvature barrier","Plateau's problem","calibrations","holography"],"falsifier":"Take a strictly mean convex thickening of a torus, choose boundary data with $[S]=\\partial\\alpha$ for an irrational class $\\alpha\\in H_2((M,\\partial M);\\mathbb{R})$, solve the equivariant $p$-Laplacian for $p\\to 1$ as in Lemma 3.5, and compare the total variation of the limiting least-gradient function $u$ with $\\sup\\int_M du\\wedge\\gamma$ over closed $L^\\infty$ forms $\\gamma$ with $\\|\\gamma\\|_{L^\\infty}\\le 1$; if these numbers differ, Theorem 1.2 is false.","tokens_in":14163,"feed_emoji":"🌊","tokens_out":15363,"duration_ms":129651,"temperature":0.7,"pith_summary":"The paper proves a continuous and topological version of the max flow/min cut theorem. On a compact Riemannian manifold whose boundary is strictly mean convex, it shows that for any prescribed relative homology class, the least possible mass of a $d-1$-current in that class equals the largest flux a divergence-free vector field of unit norm can carry across it. Earlier continuous analogues worked only on Euclidean domains and ignored topology; this one builds the topology into the statement through a homology class $\\alpha$ and a boundary cycle $S$. If the theorem is correct, it gives a variational duality relevant to holography and Teichmüller theory, and in dimensions up to seven it upgrades the optimal cut to a measured oriented minimal lamination.","feed_headline":"Flow and cut balance exactly on curved manifolds","feed_subtitle":"Least area in a homology class matches the largest flux of a divergence-free vector field.","key_machinery":"The load-bearing object is the least-gradient primitive $u$: an $\\hat\\alpha$-equivariant function of bounded variation on the universal cover $\\widetilde M$, whose total variation is minimal among functions with a given boundary trace. The minimizer $C^*$ is the current $\\psi\\mapsto\\int_M du\\wedge\\psi$, so its mass is $\\int_M |du|\\,dV$. Lemma 3.5 constructs a dual closed $(d-1)$-form $\\gamma$ with $\\|\\gamma\\|_{L^\\infty}\\le 1$, obtained as a limit of conjugate $p$-harmonic forms, satisfying $\\int_M |du|\\,dV=\\int_M du\\wedge\\gamma$; this $\\gamma$ is the flux form of the maximizing vector field. Existence of $u$ uses the mean curvature barrier condition, a weak form of strict mean convexity of $\\partial M$, which prevents the minimizing cut from collapsing to the boundary.","core_discovery":"The central discovery is Theorem 1.2: if $M$ is a compact oriented Riemannian manifold, $\\partial M$ is strictly mean convex, $\\alpha\\in H_{d-1}((M,\\partial M);\\mathbb{R})$, and $S$ is a closed $(d-2)$-current of finite mass in $\\partial M$ with $[S]=\\partial\\alpha$, then among all currents $C$ with $\\partial C=S$ and $[C]=\\alpha$ there is a mass-minimizing $C^*$, and $\\mathrm{mass}(C^*)=\\max_{F\\in\\mathcal{F}}\\int_{C^*}\\mathrm{flux}\\,F$, where $\\mathcal{F}$ is the set of measurable vector fields with $\\|F\\|_{L^\\infty}\\le 1$ and $\\mathrm{div}\\,F=0$. The same theorem implies that for $d\\le 7$, $C^*$ is a measured oriented minimal lamination.","pith_inferences":["The strict mean convexity hypothesis is likely essential: the paper's own example (Figure 1, right) shows duality fails without it, so one natural test is to perturb a non-mean-convex boundary by a small inward bend and watch whether the equality is restored as the perturbation vanishes.","The $p$-Laplacian proof of Lemma 3.5 suggests a numerical route to the graph discretization conjecture in Section 4.2: approximate the dual form $\\gamma$ by conjugate $p$-harmonic forms for $p$ close to $1$ on refined graphs, and compare discrete max flows to $\\mathrm{mass}(C^*)$.","In holography, the theorem makes the bit-thread picture exact in the mean-convex setting: entanglement entropy can be identified simultaneously with the minimal mass $\\mathrm{mass}(C^*)$ and with the maximum flux of a divergence-free unit vector field, the homology class supplying the topological selection rule."],"forward_implications":["If $d\\le 7$, the minimizer $C^*$ is a measured oriented minimal lamination: a closed union of smooth zero-mean-curvature hypersurfaces with a transverse measure, so the optimal cut is a foliation by minimal leaves.","For rational homology classes $\\alpha\\in H_{d-1}((M,\\partial M);\\mathbb{Q})$ with rational $S$, $C^*$ can be chosen as a rational $d-1$-chain of disjoint area-minimizing hypersurfaces, recovering the classical integral Plateau solution.","Theorem 4.2 extends the same duality to $C^0$ elliptic integrands satisfying the mean curvature barrier condition, so the result holds for anisotropic area functionals, not only Riemannian metrics.","The $M'=S^1$ case of the least-Lipschitz lamination theorem from Teichmüller theory follows from Corollary 1.5, giving a max-flow/min-cut proof of that part of the theory.","The duality is sharp for every real homology class compatible with the boundary data, including irrational classes where the minimizer is genuinely a lamination rather than a chain."],"supporting_citations":[{"why":"Introduces the mean curvature barrier condition and solves the least-gradient Dirichlet problem that Theorem 3.4 adapts to the equivariant setting.","marker":"[JMN18]"},{"why":"Extends the least-gradient Dirichlet problem to discontinuous $L^1$ boundary data and provides the inverse trace theorem used to build admissible functions.","marker":"[Gó24]"},{"why":"Supplies the equivariant $p$-Laplacian and conjugate harmonic form construction that Lemma 3.5 generalizes.","marker":"[DU24b]"},{"why":"Gives the convex-duality theorem (Theorem 4.10(1)) that yields the maximizing vector field; the paper also rederives it via the $p$-Laplacian.","marker":"[Fed74]"},{"why":"Provides the extension of finite-mass currents to $L^\\infty$ forms that underpins the boundary operator for currents on manifolds-with-boundary.","marker":"[Anz83]"},{"why":"Supplies the BV trace theorem and measure-theoretic perimeter tools used throughout the proof.","marker":"[Giu84]"},{"why":"Supplies the classical Plateau solution for integral currents, used for the rational-chain version in Corollary 1.3.","marker":"[Sim83]"},{"why":"Shows a fat Cantor set trace on the disk admits no mass-minimizing primitive, motivating the finite-mass hypothesis on $S$.","marker":"[ST14]"},{"why":"Combines with Theorem 1.2 to conclude that $C^*$ is a measured oriented lamination, giving Corollary 1.5.","marker":"[Bac24a]"}],"fun_headline_variants":["Flow and cut balance via currents and laminations","Continuous duality of flow and cut on manifolds","Mass-minimizing currents achieve maximal flux","Topological max flow min cut theorem"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the boundary is strictly mean convex (equivalently, that positive mean curvature forms a dense subset of the boundary), because this barrier condition is what prevents the minimizing cut from collapsing onto the boundary.","fun_headline_variants_meta":{"raw":{"variants":["Flow and cut balance via currents and laminations","Continuous duality of flow and cut on manifolds","Mass-minimizing currents achieve maximal flux","Topological max flow min cut theorem"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00036,"raw_usage":{"total_tokens":1845,"prompt_tokens":742,"completion_tokens":1103,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":358,"completion_tokens_details":{"reasoning_tokens":1047}},"tokens_in":358,"tokens_out":1103,"duration_ms":8552,"temperature":1.0,"reasoning_tokens":1047,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:40:33.778799+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a strictly mean convex thickening of a torus, choose boundary data with $[S]=\\partial\\alpha$ for an irrational class $\\alpha\\in H_2((M,\\partial M);\\mathbb{R})$, solve the equivariant $p$-Laplacian for $p\\to 1$ as in Lemma 3.5, and compare the total variation of the limiting least-gradient function $u$ with $\\sup\\int_M du\\wedge\\gamma$ over closed $L^\\infty$ forms $\\gamma$ with $\\|\\gamma\\|_{L^\\infty}\\le 1$; if these numbers differ, Theorem 1.2 is false.","supporting_citations":[],"review_version":1}