{"id":"2e72ffe2-4674-456b-b8f0-9c1845f007d8","arxiv_id":"2607.04989","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.5,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"Local geodesic operations produce size- and quality-controlled intrinsic triangulations of closed discrete surfaces that support both refinement and coarsening and enable direct high-order meshing.","lead":"The paper gives a practical algorithm that rebuilds a triangle mesh on a closed 3D surface so every edge is a true shortest path along the surface and every face stays faithful to the original geometry. It works for both coarsening and refining, runs on thousands of real models, and directly yields high-order elements without first making a straight-edged mesh.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Heuristic opposite-angle swaps plus frequent circumcenter failures leave the claimed angle bounds unsatisfied on 72.5 % of successful models, so the quality-enforcement half of the strongest claim is only partially realized.","rationale":"The reader already isolated exactly this weakest assumption (heuristic Delaunay + circumcenter reliability) and correctly judged the overall contribution CONDITIONAL rather than ACCEPT. The algorithmic core—shortest-path edges, non-developable faces that permit coarsening, 99.6 % success on thousands of real models, and public Gmsh code—remains intact; the angle shortfall is a reported limitation, not a hidden contradiction. No stronger load-bearing flaw (e.g., topological invalidity, systematic geodesic failure, or non-reproducibility) appears in the text or experiments, so the reader’s verdict and confidence need no adjustment.","tokens_in":23136,"tokens_out":503,"duration_ms":19988,"concrete_test":"On the 4943 successful filtered Thingi10K models, compute the per-mesh minimum and maximum intrinsic angles and report their histograms/percentiles. If >20 % of models contain any angle <10° or >160°, the practical quality claim is weaker than stated and the CONDITIONAL verdict should carry a stronger caveat on angle enforcement.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim states that local geodesic operations produce an intrinsic mesh that “obeys a prescribed \tau… angle bounds.” Sec. 5.2 and Table 1 show that, under the very bounds used for the Thingi10K experiment (20°–140°), only 27.5 % of the 4943 successful runs actually meet them. The reason is structural: true geodesic circumcenters needed for triangle splits (Alg. 11, Sec. 4.4, Appendix A) are frequently non-existent (Fig. 6) or numerically unattainable, and the opposite-angle heuristic (Alg. 2) is only a planar-style proxy that can leave residual skinny triangles. Although the paper aborts operations that would worsen quality (Algs. 15–17) and reports the 27.5 % figure openly, the claim of systematic enforcement therefore rests on incomplete evidence; without residual-angle statistics it is unclear how severe the violations remain.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper presents a constructive algorithm for intrinsic triangulations of closed watertight discrete surfaces: edges are shortest geodesic paths and faces are unions of input primitives. Starting from an input triangulation, local operations (edge swaps, splits, collapses, and triangle splits) are performed intrinsically, driven by a characteristic-length field and angle-based quality criteria derived from exact geodesic distances. Distances use continuous Dijkstra (MMP/ICH) accelerated by an A* Euclidean lower-bound heuristic that reduces cost to roughly 3% of standard propagation. The framework supports both refinement and coarsening (overcoming a limitation of developable-triangle methods) and yields a foundation for direct high-order meshing. Validation on a filtered Thingi10K subset reports 99.6% success (4943/4963 models) under explicit time/memory limits, with open implementation in Gmsh.","tokens_in":23435,"tokens_out":1031,"duration_ms":16525,"significance":"If the claims hold, the work is a solid practical advance in discrete differential geometry and mesh generation: it produces isogeometric intrinsic meshes without parametrization or geometry alteration, enables coarsening (unlike prior developable intrinsic Delaunay schemes), and supplies a direct route to high-order elements that bypasses the classical linear-then-curve pipeline. The A* acceleration, fully specified pseudo-code, large-scale public-dataset validation, and open Gmsh code are concrete strengths that make the method usable rather than purely theoretical. The contribution is primarily algorithmic/engineering rather than a new existence theorem, but it fills a clear gap for closed organic surfaces.","major_comments":[{"comment":"Sec. 5.2 and Table 1: under the experimental bounds (20°–140°), only 27.5% of the 4943 successful runs strictly satisfy the prescribed angles. The abstract and introduction claim that “quality is enforced through angle-based criteria,” yet the text itself attributes the shortfall to non-existence or numerical failure of geodesic circumcenters (Sec. 4.4, Fig. 6) and the heuristic opposite-angle test (Alg. 2). Residual statistics—histograms of min/max angles, fraction of violating triangles, or maximum violation magnitude—for the remaining 72.5% are needed to quantify how well quality is actually controlled; without them the enforcement claim remains only partially evidenced.","section":"Sec. 5.2, Table 1"},{"comment":"Sec. 3.1.4 / Alg. 2 and Sec. 3.2.4 / Fig. 10: the opposite-angle-sum heuristic is adopted because true geodesic circumcenters need not exist or be unique. While it recovers the planar Delaunay criterion for developable triangles, no convergence guarantee is supplied for non-developable surfaces, and the paper itself exhibits unstable swap cycles that must be aborted by iteration limits. Empirical counts of residual non-locally-Delaunay edges after the main loop (or after each local operation) would clarify how often the heuristic leaves the mesh short of a usable Delaunay-like state.","section":"Sec. 3.1.4, Alg. 2"}],"minor_comments":[{"comment":"Fig. 20 color bars are labeled “0 1800 180”; the middle value appears to be a typographical artifact and should be clarified or removed.","section":"Fig. 20"},{"comment":"The A* speed-up is stated as “roughly 3%” both in the abstract and Sec. 4.3; a short table or plot of wall-clock ratios versus model size (or versus classical MMP) would make the claim more precise and reproducible.","section":"Sec. 4.3"},{"comment":"Notation for characteristic length occasionally switches between cl_min / cl_max and the integral form R dl/cl; a single consistent definition early in Sec. 3.2 would improve readability.","section":"Sec. 3.2"},{"comment":"Appendix A solutions for the circumcenter system are useful, but a brief numerical-stability note (conditioning when D_B or Y_C approach zero) would help implementers.","section":"Appendix A"}],"recommendation":"minor_revision","confidential_remarks":"Solid engineering contribution with unusually thorough public validation and open code; the quality-enforcement gap is real but openly reported and fixable with extra statistics. Suitable for a computational-geometry or meshing journal; novelty relative to Sharp et al. (intrinsic Delaunay) and Liu et al. (coarsening via flattening) is clear once the non-developable + coarsening combination is emphasized."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The real news is a complete local-optimization loop (swap/split/collapse/triangle-split) that works with non-developable intrinsic triangles whose edges are exact shortest geodesics. That finally lets you coarsen as well as refine while staying isogeometric to the input, and they ship it inside Gmsh. Prior developable-triangle methods could only refine; this removes that bottleneck and gives a direct path to high-order elements without the usual linear-then-curve pipeline.\n\nWhat they do well is engineering and honesty. Algorithms are fully specified with pseudo-code. Geodesics rest on standard continuous Dijkstra (MMP/ICH) plus a clean A* acceleration that drops cost to ~3 %. Validation on 4 943 filtered Thingi10K models yields 99.6 % success under explicit time/memory limits; residual failures are catalogued. Code is public. Self-citations stay within the expected geodesic and meshing literature. No circular fitting.\n\nThe soft spot the stress-test flags is real but already quantified by the authors: only 27.5 % of successful runs strictly meet the 20°–140° angle bounds they themselves used. True geodesic circumcenters are often missing or numerically unreachable, so triangle splits fall back on a planar-style opposite-angle heuristic that can leave residual skinny triangles. They abort operations that would worsen quality and report the 27.5 % figure, so the claim of systematic quality enforcement is overstated relative to the data, yet the size-control and coarsening claims still hold. Closed-surface restriction and lack of sharp-feature handling are acknowledged limitations, not hidden ones.\n\nThis is for people who need practical intrinsic remeshing or high-order surface meshes from discrete data. The math and citations look solid; the data are external and large. I would send it to referees—expect them to ask for residual-angle histograms and clearer language on what “quality enforcement” actually guarantees—but the core algorithmic advance is real and usable. Worth engaging.","headline":"Working open-source intrinsic mesher that finally coarsens non-developable discrete surfaces; angle-quality claim is only partially realized but openly reported.","tokens_in":23989,"tokens_out":489,"would_cite":true,"duration_ms":5073,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Local geodesic operations build intrinsic meshes that keep the original surface exact and support both coarsening and refinement.","keywords":["intrinsic mesh","discrete isogeometric triangulation","geodesic","shortest path","A* search","coarsening","direct high-order meshing"],"falsifier":"On a closed manifold from the filtered Thingi10K suite, run the algorithm with prescribed angle bounds and check whether the final intrinsic mesh still contains intersecting geodesics, violates the size field, or leaves angles outside the requested range on more than a few percent of models after the iteration limit.","tokens_in":24065,"feed_emoji":"△","tokens_out":813,"duration_ms":6698,"temperature":0.7,"pith_summary":"This paper shows how to remesh a closed discrete surface so that every edge is a shortest geodesic path and every face is a union of original input triangles, without ever changing the geometry. Starting from a watertight triangulation, the algorithm applies ordinary local operations—edge swaps, splits, collapses and triangle splits—but evaluates lengths, angles and intersections intrinsically on the surface. Size is controlled by a characteristic-length field and quality by intrinsic-angle bounds. Exact geodesic distances come from a continuous Dijkstra method sped up by an A* heuristic that cuts cost to roughly three percent of ordinary propagation. Because the triangles need not be developable, the same framework can coarsen as well as refine, removing a central restriction of earlier intrinsic methods. The resulting mesh is therefore an exact, isogeometric representation that can be handed directly to high-order polynomial fitting, skipping the usual “linear mesh then curve” pipeline. Nearly five thousand filtered models from Thingi10K succeed under these rules.","feed_headline":"Geodesic local ops remesh surfaces while keeping geometry exact","feed_subtitle":"Coarsening and refinement both work; high-order meshes follow without intermediate curving","key_machinery":"Intrinsic local operators (edge swap/split/collapse and triangle split) whose decisions rest on exact shortest-path geodesics computed by continuous Dijkstra with A* guidance, plus a simple opposite-angle swap heuristic that replaces the classical empty-circumcircle test.","core_discovery":"A complete intrinsic meshing pipeline exists: repeated local geodesic operations driven by a characteristic-length field and an opposite-angle heuristic produce triangulations whose edges are shortest geodesics, whose faces inherit the input geometry, that obey size and angle constraints wherever feasible, and that support both refinement and coarsening on closed discrete surfaces.","pith_inferences":["The same local geodesic operators could be adapted to open surfaces once boundary and feature edges are treated as constrained non-geodesic paths.","Because each intrinsic triangle is already a piecewise-linear patch of the original mesh, it supplies an exact domain for integration or for fitting non-polynomial bases without geometric error.","Parallel execution of independent cavities would be the natural next performance step once sequential geodesic queries dominate runtime."],"forward_implications":["Coarsening and refinement of discrete surfaces can be performed while preserving the exact input geometry.","High-order finite-element meshes can be generated directly from geodesic edges without first building and then curving a linear mesh.","Size- and angle-controlled remeshing becomes available for any watertight closed triangulation, independent of parametrization.","Exact geodesic distances need only be computed locally and can be accelerated by simple Euclidean lower bounds."],"fun_headline_variants":["Local geodesic ops build intrinsic meshes keeping geometry exact","Intrinsic remeshing supports both coarsening and refinement on surfaces","Geodesic edges and angle criteria produce size-controlled intrinsic meshes","A* geodesics drive local ops for full intrinsic triangulation pipeline","Closed surfaces get intrinsic triangulations via geodesic local ops"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"The opposite-angle swap heuristic plus occasional circumcenter insertion is enough to reach a usable Delaunay-like mesh even when true geodesic circumcenters are missing or non-unique.","fun_headline_variants_meta":{"raw":{"variants":["Local geodesic ops build intrinsic meshes keeping geometry exact","Intrinsic remeshing supports both coarsening and refinement on surfaces","Geodesic edges and angle criteria produce size-controlled intrinsic meshes","A* geodesics drive local ops for full intrinsic triangulation pipeline","Closed surfaces get intrinsic triangulations via geodesic local ops"]},"model":"grok-4.5","effort":"low","cost_usd":0.00452,"raw_usage":{"total_tokens":1270,"prompt_tokens":727,"num_sources_used":0,"completion_tokens":84,"cost_in_usd_ticks":45200000,"prompt_tokens_details":{"text_tokens":727,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":459,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":727,"tokens_out":84,"duration_ms":4272,"temperature":1.0,"reasoning_tokens":459,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-11T10:27:20.717814+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On a closed manifold from the filtered Thingi10K suite, run the algorithm with prescribed angle bounds and check whether the final intrinsic mesh still contains intersecting geodesics, violates the size field, or leaves angles outside the requested range on more than a few percent of models after the iteration limit.","supporting_citations":[],"review_version":1}