{"id":"6f8ae460-1e88-4702-853c-08055ed81300","arxiv_id":"2608.12142","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A refinement algorithm enforces monotonic mesh edges so that piecewise-linear approximations of implicit 2D scalar fields preserve the critical points of the underlying Morse function.","lead":"The paper presents a method to build a triangular mesh of an implicit 2D scalar field, like a neural network, that preserves the field's critical points by keeping every mesh edge monotonic. Generalists might read it to learn how to extract reliable topological features from neural implicit representations without dense sampling.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1's extremum-case proof rests on a false level-set containment claim; the central theorem needs an independent argument, and the algorithm still cannot certify the monotonicity precondition.","rationale":"The reader correctly identified the sharp-corner assertion as unsubstantiated in the rationale, but selected the sampling-detector gap as the weakest assumption. I agree the detector gap is a real practical limitation: the algorithm samples the projected gradient at spacing < w and cannot detect monotonicity violations narrower than w, so the final mesh is not certified to satisfy Theorem 1's precondition. However, the more load-bearing issue concerns Theorem 1 itself: if the extremum-case proof is invalid, the central theoretical contribution is not established, regardless of how well the algorithm samples. The theorem is likely true—a convex-hull argument over the link appears to repair the proof—but as written, the manuscript lacks that argument. Because the proof can plausibly be repaired and the theorem is not shown false, the verdict should remain CONDITIONAL rather than move to REJECT. The concrete test above would settle whether the theorem survives independently of the flawed level-set reasoning. If it succeeds, the paper still needs to address the detector guarantee (e.g., a feature-scale bound relating w to the field's Lipschitz/Hessian scale) before the method can claim topology preservation with confidence.","tokens_in":9405,"tokens_out":32163,"duration_ms":316319,"concrete_test":"Re-derive the extremum case of Theorem 1 without the sharp-corner claim. For an interior vertex v with link vertices p_i, edge monotonicity gives ∇f(v)·(p_i-v) ≥ 0 for all i. Since v lies in the convex hull of the p_i, there exist positive barycentric weights λ_i with Σ λ_i (p_i-v)=0. Dotting with ∇f(v) forces every term to vanish, so ∇f(v)·(p_i-v)=0 for all i; because the p_i-v span R^2, ∇f(v)=0. If this direct derivation succeeds, the theorem survives and only the exposition needs repair; if a legal star configuration with all link vertices on one side of a nonzero gradient exists, the theorem is false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim of the paper is Theorem 1, and its proof is not sound as written. In the extremum case of Part 1, the authors assert that the isocontour σ = f^{-1}(f(v)) cannot meet any edge of the star or link, and therefore 'σ must be contained entirely within a single triangle incident to x_m,' where it would form a sharp corner. The inference is invalid: a connected level-set component can pass through a vertex from one incident triangle to a non-adjacent triangle without crossing any edge, since the two triangles meet only at v. Thus σ need not lie in a single triangle. Moreover, if the stronger conclusion actually forced by the edge/link avoidance is that the component is just the point v, then there is no curve to which the differentiability argument applies. A correct proof of this case must use the fact that the link is a cycle surrounding v and sum the directional derivatives over the link to force ∇f(v)=0; the manuscript does not provide this step. Consequently, the paper's headline guarantee that monotone edges imply critical-point consistency is not established by the supplied argument. This is independent of, and more fundamental than, the reader's concern about the sampling detector not certifying monotonicity, which affects whether the algorithm ever meets the theorem's precondition.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a pipeline for meshing implicitly defined 2D scalar fields so that the resulting piecewise-linear (PL) approximation is topologically consistent with the underlying field. The central theoretical claim is Theorem 1: if every mesh edge is monotonic with respect to a Morse function f, then (1) every critical point of the PL interpolant coincides with a critical point of f, and (2) any critical point of f either coincides with a PL critical point or shares a triangle with another critical point of f. The method uses Delaunay refinement, samples projected gradients along edges to detect monotonicity violations, inserts vertices via 1D Newton solves, refines critical point locations, and then adds separatrix refinement. Experiments are reported on a synthetic Griewank-plus-Gaussian function and on an implicit terrain INR, with qualitative images and vertex counts.","tokens_in":9682,"tokens_out":8782,"duration_ms":99093,"significance":"If Theorem 1 were correct, it would provide a simple and attractive criterion for PL/continuous critical-point consistency, and the paper's pipeline would be a useful engineering contribution for implicit fields such as INRs. The paper has clear strengths: it is well written, the code is publicly released, the method requires only pointwise evaluations of the function and its derivatives, and the intended application to implicit neural representations is timely. However, the central theorem is false as stated. Section 3.1 contains an invalid level-set containment inference, and there is an explicit counterexample showing that monotonic edges do not prevent spurious PL critical points. Consequently, the paper's main theoretical claim cannot support the proposed method, and the experimental evidence is not quantitative enough to compensate.","major_comments":[{"comment":"The theorem is false as stated. Let f(x,y)=x, and let M be a triangulation of a small disk centered at the origin v=(0,0) whose link vertices all lie in the right half-plane x>0. On every edge of M, the restriction of f is affine, hence monotonic, so the mesh is monotonic with respect to f. The PL interpolant hat f has value 0 at v and positive values at every other vertex, so v is a strict local minimum of hat f. But ∇f(v)=(1,0)≠0, so v is not a critical point of f. This directly contradicts Part 1. The proof's assertion that the isocontour σ=f^{-1}(f(v)) 'must be contained entirely within a single triangle incident to x_m' is the exact false step: σ is the line x=0, which passes through v and continues through two (in general non-adjacent) incident triangles without crossing the interior of any edge. The differentiability argument then has nothing to apply to. If one insists on a Morse function with no critical points being vacuously Morse, this example suffices; alternatively, f(x,y)=x+y^2 with a sufficiently fine mesh gives the same local configuration with no critical point in the mesh neighborhood.","section":"Section 3.1, Theorem 1(1)"},{"comment":"The monotonicity detector does not certify the precondition of Theorem 1. The algorithm samples the projected gradient at spacing less than w and declares an edge monotonic when no sign change is observed. No bound is given relating w to the feature scale of f, such as a Lipschitz constant for the directional derivative along the edge. A non-monotonic feature narrower than w will therefore go undetected, and the final mesh may violate the monotonicity condition without the method knowing it. Since the algorithm can query f and its derivatives at arbitrary points, a rigorous certificate would require either a derivative-bounds argument or an adaptive subdivision rule with a proven stopping criterion; the paper provides neither.","section":"Section 4, Monotonic Edges"}],"minor_comments":[{"comment":"The statement that the method 'successfully matches all reference critical points' is not backed by a quantitative comparison; please report critical-point counts, false positives/negatives, and positional errors with respect to the reference mesh.","section":"Section 5.2"},{"comment":"The caption says '4567 fewer vertices (-42%)'; this should read '4,567 fewer vertices' and should specify the comparison baseline explicitly.","section":"Figure 4 caption"},{"comment":"The definition of a monotonic edge should state whether constant-valued edges are allowed; the proof of Theorem 1 uses 'strictly above' for extrema, suggesting strict monotonicity may be intended, and the distinction matters for the theorem's validity.","section":"Section 2"},{"comment":"The sentence 'As these contours expand, they must eventually intersect an edge' is terse; the proof should identify which edge is intersected and justify why continuing to increase the value forces a second intersection with the same edge.","section":"Section 3.1, Part 2 minimum case"},{"comment":"The claim that dense sampling ensures triangles 'do not contain pairs of critical points within a distance of 2R' is imprecise; it should be formulated as a geometric condition on the triangle circumradius and related explicitly to the scale of f.","section":"Section 4, Density Sampling"}],"recommendation":"reject","confidential_remarks":"The main theorem is false; a concrete counterexample is given in major comment 1. This is not merely a proof gap that can be filled by a local edit, since the stated claim fails in a simple construction. The sampling-based monotonicity certification is also heuristic, so even a corrected theorem would not fully cover the algorithm. The authors should be given the counterexample so they can reassess the contribution and either restate the theorem with additional hypotheses or reposition the paper as an empirical heuristic method."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper has a real new idea—use monotonic edges to guarantee critical-point consistency for PL approximations of implicit fields—and a refinement algorithm that clearly works in practice. But the proof of Theorem 1 has a genuine gap, and the experiments are too thin to verify the guarantee empirically.\n\nWhat is actually new here: the monotonicity characterization itself, and the pipeline that samples gradients along edges, inserts Newton-refined split points, then refines around separatrices. The theorem is a plausible statement and the algorithm is clearly described. The code is public, which helps.\n\nNow the soft spots, in order of importance. The proof of Theorem 1, Part 1, is not sound as written. The claim that an isocontour through an extremum vertex must be contained entirely within a single incident triangle is false: a smooth level curve can pass through the vertex from one triangle into a non-adjacent one without crossing any edge. The sharp-corner argument therefore doesn't go through. I agree with the stress-test that a correct proof needs a different argument—probably summing directional derivatives around the link to force the gradient to vanish. The theorem may still be true, but the manuscript doesn't establish it.\n\nSecond, the sampling-based monotonicity detector is heuristic. Sampling at spacing w cannot catch non-monotonic features narrower than w, so the algorithm never certifies the theorem's precondition. The paper repeatedly says \"approximately monotonic,\" but there is no bound tying w to the feature scale of f. That's a real gap between the theory and the method.\n\nThird, the experimental evaluation is qualitative: two datasets, no critical-point counts, no positional errors, no ablation of the monotonicity step. The INR comparison is muddied by persistence simplification on the reference grid. The code is available, but the paper itself doesn't give the reader enough to judge how often the method actually preserves topology.\n\nThese are specific issues, not a fundamental rejection. The direction is solid and the heuristic clearly helps in the shown examples. The paper deserves a serious referee, not a desk reject. The referee should ask for a corrected proof or a softened statement of Theorem 1, and a more quantitative evaluation. If the authors can close the proof gap, this becomes a solid contribution to topology-preserving meshing for implicit fields.","headline":"A genuinely useful meshing idea with a central theorem whose proof doesn't hold up as written—worth engaging, but the theoretical claim needs a corrected argument.","tokens_in":10167,"tokens_out":5133,"would_cite":false,"duration_ms":49124,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["68U05"],"pacs":[],"model":"deepseek-v4-flash","headline":"If every edge of a triangulation is monotone along an implicit scalar field, the piecewise-linear approximation cannot invent or silently lose critical points.","keywords":["Morse-Smale complex","implicit scalar fields","implicit neural representations","monotonic edges","Delaunay refinement","piecewise-linear interpolation","critical points","separatrix refinement"],"falsifier":"Take an implicit field that is monotone along an edge except for a tiny oscillation whose two sign changes of the directional derivative lie within a length shorter than $w$, run the pipeline with that $w$, and count critical points of the resulting PL mesh. If the mesh shows a spurious critical pair (or misses a true pair) with persistence above the numerical tolerance, the sampling-based monotonicity check has failed and the critical-point consistency claimed by the method does not hold for that input.","tokens_in":9221,"feed_emoji":"🗺️","tokens_out":7656,"duration_ms":72744,"temperature":0.7,"pith_summary":"This paper establishes a sufficient condition under which a piecewise-linear (PL) triangulation of an implicit 2D scalar field faithfully reproduces the field's critical points: every mesh edge must be monotone along the field, meaning the field value changes monotonically along each edge. The paper proves that a PL mesh satisfying this condition cannot invent critical points, and any critical point it misses must be paired with another critical point inside the same triangle. It then presents a refinement algorithm that detects monotonicity violations by sampling the projected gradient along edges, splits offending edges with one-dimensional Newton steps, and further refines near separatrices, using only pointwise evaluations of the field and its derivatives. This matters because implicit neural representations and other implicit fields are increasingly used in visualization, yet standard Morse–Smale complex extraction assumes an explicit mesh; the paper gives a route to a topologically consistent explicit mesh from such fields.","feed_headline":"Monotone edges guarantee meshes keep a field's critical points","feed_subtitle":"A proof plus edge-refinement lets implicit 2D fields yield Morse–Smale complexes without dense uniform sampling.","key_machinery":"The load-bearing object is the monotone edge: an edge $v_i v_j$ is monotone with respect to $f$ when the restriction of $f$ to the segment between its endpoints is a monotone function of position along the segment, equivalently when the directional derivative $g'(t)$ along the edge does not change sign, so any isocontour crosses the edge at most once. Theorem 1 converts this one-dimensional property into a local counting argument inside the star of each vertex and inside single triangles: contour branches forced by a PL critical point, or by a critical point of $f$ inside a triangle, would have to cross some edge twice unless the critical points are genuine and appropriately paired. The algorithm's practical machinery is a Delaunay-refinement loop that samples the projected gradient on each edge at spacing less than $w$, uses one-dimensional Newton updates to locate sign changes of $g'(t)$, splits edges there, then refines near separatrices so PL steepest paths align with the true integral lines.","core_discovery":"The paper's central claim is Theorem 1: if $\\hat f$ is the piecewise-linear interpolation of a Morse function $f$ on a triangulation whose every edge is monotone with respect to $f$, then every critical point of $\\hat f$ is a critical point of $f$, and every critical point of $f$ either appears as a critical point of $\\hat f$ or shares a triangle with at least one other critical point of $f$. The proof is a case analysis on isocontours: a monotone edge can be crossed by a given isocontour at most once, so the contour geometry forced by a PL extremum or saddle cannot be realized by a smooth Morse function unless the critical point is genuine, and an isolated critical point trapped inside a triangle would force an isocontour to cross some edge twice. The paper wraps this theoretical condition in a four-stage pipeline—density sampling, monotonicity-driven edge splitting, Newton-based critical-point repositioning, and separatrix refinement—that realizes the condition approximately with pointwise queries only, and demonstrates in experiments that the resulting meshes recover reference critical points with far fewer vertices than uniform sampling.","pith_inferences":["The theorem supplies a potential certificate: if a mesher could verify edge monotonicity with interval arithmetic or Lipschitz bounds on $g'$ instead of sampling, the critical-point guarantee would become rigorous; the paper itself relies on sampling at scale $w$.","The 'shares a triangle' bound suggests an adaptive multi-scale strategy: choose the sampling radius $R$ from a persistence or feature-scale estimate so that unresolved critical pairs are pushed below the desired scale, turning the refinement loop into a scale-space Morse–Smale complex extractor.","A natural 3D analogue would require a monotonicity condition on tetrahedral faces as well as edges, because isosurfaces can re-enter a face even when edges are monotone; the paper leaves 3D as future work."],"forward_implications":["On a monotone-edge mesh, every PL critical point is a genuine critical point of $f$, so spurious extrema and saddles do not need to be removed by persistence-based simplification.","If a true critical point is missing from the PL mesh, at least one other true critical point lies in the same triangle; topology can only be lost as a close pair, never as an isolated feature.","Topology-preserving meshing of an implicit field reduces to enforcing a local, checkable condition on edges, rather than solving a global optimization over vertex placement.","For implicit neural representations, the pipeline makes Morse–Smale complex extraction practical at modest mesh sizes: in the terrain INR experiment it matched all reference critical points with 6,359 vertices, far fewer than a 500×500 grid."],"supporting_citations":[{"why":"supplies the PL Morse-Smale extraction convention used to trace separatrices through mesh edges and triangle interiors","marker":"[1]"},{"why":"stability of persistence diagrams; motivates why dense sampling alone cannot guarantee topological consistency","marker":"[6]"},{"why":"Poisson disk sampling that provides the initial point distribution for the refinement loop","marker":"[2]"},{"why":"background theory of Delaunay mesh generation on which the refinement strategy builds","marker":"[4]"},{"why":"provides the Delaunay triangulation implementation used for mesh updates","marker":"[25]"},{"why":"supplies the trained terrain implicit neural representation and the reference critical points for the INR experiment","marker":"[11]"},{"why":"introduces the periodic-activation INR formulation that defines the target class of implicit fields","marker":"[28]"}],"fun_headline_variants":["Monotone edges ensure PL meshes preserve critical points","Implicit 2D fields: monotone-edge meshing yields Morse-Smale complexes","Topology-preserving meshing: enforce monotonicity on each edge","Want correct critical points in meshes? Make every edge monotone","Monotone edges: a simple condition for topology-preserving meshing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that sampling the projected gradient at spacing $w$ along each edge detects every change in the sign of the directional derivative; if a non-monotonic wiggle is narrower than $w$, the edge is declared monotone and the final mesh can violate the precondition of Theorem 1 without the algorithm knowing.","fun_headline_variants_meta":{"raw":{"variants":["Monotone edges ensure PL meshes preserve critical points","Implicit 2D fields: monotone-edge meshing yields Morse-Smale complexes","Topology-preserving meshing: enforce monotonicity on each edge","Want correct critical points in meshes? Make every edge monotone","Monotone edges: a simple condition for topology-preserving meshing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000876,"raw_usage":{"total_tokens":3805,"prompt_tokens":979,"completion_tokens":2826,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":2741}},"tokens_in":595,"tokens_out":2826,"duration_ms":20301,"temperature":1.0,"reasoning_tokens":2741,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:14:10.515674+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take an implicit field that is monotone along an edge except for a tiny oscillation whose two sign changes of the directional derivative lie within a length shorter than $w$, run the pipeline with that $w$, and count critical points of the resulting PL mesh. If the mesh shows a spurious critical pair (or misses a true pair) with persistence above the numerical tolerance, the sampling-based monotonicity check has failed and the critical-point consistency claimed by the method does not hold for that input.","supporting_citations":[{"cited_title":"A topological hierarchy for functions on triangulated sur- faces.IEEE Transactions on Visualization and Computer Graphics, 10(4):385–396, 2004","cited_arxiv_id":null,"evidence_quote":"supplies the PL Morse-Smale extraction convention used to trace separatrices through mesh edges and triangle interiors"},{"cited_title":"Stability of persistence diagrams","cited_arxiv_id":null,"evidence_quote":"stability of persistence diagrams; motivates why dense sampling alone cannot guarantee topological consistency"},{"cited_title":"Fast Poisson disk sampling in arbitrary dimensions","cited_arxiv_id":null,"evidence_quote":"Poisson disk sampling that provides the initial point distribution for the refinement loop"},{"cited_title":"CRC Press Boca Raton, 2013","cited_arxiv_id":null,"evidence_quote":"background theory of Delaunay mesh generation on which the refinement strategy builds"},{"cited_title":"2D conforming triangulations and meshes","cited_arxiv_id":null,"evidence_quote":"provides the Delaunay triangulation implementation used for mesh updates"},{"cited_title":"ImplicitTerrain: a con- tinuous surface model for terrain data analysis","cited_arxiv_id":null,"evidence_quote":"supplies the trained terrain implicit neural representation and the reference critical points for the INR experiment"},{"cited_title":"Implicit neural representations with periodic activation functions.Advances in neural information processing sys- tems, 33:7462–7473, 2020","cited_arxiv_id":null,"evidence_quote":"introduces the periodic-activation INR formulation that defines the target class of implicit fields"}],"review_version":1}