{"id":"40a2b975-1d81-4a3f-8112-316d52db02d5","arxiv_id":"2508.04076","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A 3D element-splitting method predicts dynamic crack paths and branching by removing elements whose computed fracture energy release rate exceeds a critical value, with GPU acceleration.","lead":"This paper presents a computer method that simulates how cracks grow and branch in 3D materials under fast dynamic loading. It runs on GPUs, which may make large-scale fracture predictions faster and more practical for engineering designs.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The deactivation criterion in Eqs. (17)-(18) is a heuristic local product of projected edge stretch and maximum principal stress, not a derived or independently verified G; the central claim of accurate fracture simulation rests on this unverified equivalence.","rationale":"The reader's weakest_assumption already identifies the G-formula as underived. I agree and sharpen the point: the formula is not merely underived but structurally mismatched for mixed-mode states because it uses maximum principal stress and normal-projected edge stretch rather than the crack-plane traction and opening displacement. This is load-bearing because the entire element-deactivation rule, and hence every predicted crack path, depends on this local quantity. The benchmarks are qualitative and do not isolate the local G; the coarse-mesh failure to branch in §4.4 is consistent with a non-physical, mesh-sensitive criterion. Independent support is limited to a self-cited 2D study, which does not cover the 3D claim. A VCCT-based comparison on an existing benchmark would settle whether Eqs. (17)-(18) actually compute the energy release rate. Since this concern is exactly the basis of the reader's CONDITIONAL verdict, no verdict change is needed.","tokens_in":20586,"tokens_out":5262,"duration_ms":63556,"concrete_test":"On the Section 4.3 compact-compression model (three mesh sizes), compute the crack-front energy release rate at several time steps by two independent routes: (1) the proposed Eqs. (17)-(18), and (2) a virtual crack closure technique (VCCT) using the same nodal forces and displacements. If the ratio G_CEM/G_VCCT differs from 1 by more than ~20% on the finest mesh, or does not converge toward 1 with mesh refinement, the deactivation criterion is not the physical G and the central claim loses its fracture-mechanics basis. A simpler supplementary check is to load a small edge-cracked panel in pure mode I with known analytical G = K_I^2/E' and confirm Eqs. (17)-(18) reproduce it to within a few percent; this isolates the formula's dimensional and directional correctness from the complex benchmark dynamics.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3 introduces the fracture criterion via Eqs. (17)-(18): an element is deactivated when the computed quantity G exceeds the material G_c. This quantity is constructed from (a) the projection of edge stretch on the crack-surface normal and (b) the projection of the maximum principal stress on the same normal, summed for the two edges crossing the crack front. The paper calls this 'the fracture energy release rate' and claims it follows from the evolving topology, but it is not derived from the §2.1 variational principle nor from a J-integral or virtual crack closure argument. For a general 3D transient mixed-mode state, the product 1/2 sigma_n delta_n is not the energy release rate: the crack-plane traction should be the traction vector normal to the newly created surface, not the maximum principal stress, and delta_n should be the crack opening displacement, not an edge stretch whose direction is mesh-dependent. In shear-dominated regions (e.g., the Kalthoff-Winkler plate, §4.1), a large sliding displacement can have near-zero normal projection, causing G to underestimate the driving force and potentially suppressing valid crack growth; conversely, a compressive normal traction with large tensile principal stress elsewhere can trigger spurious deactivation. None of the benchmarks isolate these contributions: they report crack paths and dissipated energies, which are aggregate outputs that can coincidentally align even if the local criterion is wrong, and the paper itself notes that branching is mesh-sensitive (§4.4). Thus the central claim—that the method 'accurately simulates' fracture because it compares a physical G to G_c—is not supported unless Eqs. (17)-(18) are shown to be the actual energy release rate.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a three-dimensional Crack Element Method (CEM) for transient dynamic fracture in quasi-brittle materials. The method combines an element-splitting algorithm with the Edge-based Smoothed Finite Element Method (ES-FEM) on tetrahedral and hexahedral meshes. Crack growth is modeled by deactivating an element when a locally computed quantity G, defined in Eqs. (17)-(18), exceeds the material critical energy release rate G_c. The local G is constructed from edge stretches and maximum principal stresses projected onto the crack-surface normal. The paper reports five benchmark simulations: the Kalthoff-Winkler plate, an anchorage pull-out test, a compact compression specimen, and two crack-branching problems under Neumann and Dirichlet boundary conditions. All 3D simulations are run with GPU acceleration, and the paper claims accurate prediction of single-crack propagation and complex crack branching, with branching emerging without an explicit branching criterion.","tokens_in":20977,"tokens_out":4851,"duration_ms":62089,"significance":"If the proposed fracture criterion is valid, the method could be practically useful: it is GPU-accelerated, operates on unstructured tetrahedral meshes, and appears to capture curved and branching crack paths in several classical benchmarks with only qualitative comparison to experiment and prior simulations. The paper's strengths include mesh-sensitivity studies in all five examples, comparison of dissipated energy and load-displacement curves to published methods, and a clear presentation of the element-splitting topology. The central weakness is that Eq. (17)-(18) are presented as a definition rather than derived from the variational principle in Eq. (2), and they are not checked against a known analytical solution or an independent energy-release-rate calculation. The stress-test concern about the physical equivalence of the local G is therefore substantive and is not resolved by the aggregate benchmark comparisons.","major_comments":[{"comment":"The central fracture criterion is a definition, not a derived or independently validated energy release rate. Eqs. (17)-(18) define G as half the product of the projection of an edge stretch and of the maximum principal stress onto the crack-surface normal. For a general three-dimensional mixed-mode state this is not the fracture energy release rate: G should involve the traction vector on the newly created crack surface and the crack opening displacement, not the maximum principal stress and a mesh-dependent edge stretch. In shear-dominated regions such as the Kalthoff-Winkler plate (§4.1), a large sliding displacement can have a small normal projection, potentially suppressing valid crack growth; conversely, a compressive normal traction combined with a large tensile principal stress elsewhere could trigger spurious deactivation. No benchmark isolates these contributions, and crack-pat","section":"§3, Eqs. (16)-(18)"},{"comment":"The variational principle in Eq. (2) contains the fracture-energy term ∫ G_c dΓ, but the implemented method does not minimize this functional; it uses the separate element-wise deactivation rule of Section 3. The paper states that the fracture energy release rate is 'derived' from the evolving topology, but no derivation connects Eq. (2) to Eqs. (17)-(18). Clarify whether Eqs. (17)-(18) are an approximation to the true energy release rate or a heuristic criterion. This distinction is load-bearing because the paper's central claim of accuracy rests on the physical status of this quantity.","section":"§2.1 and §3"},{"comment":"The accuracy claims need quantitative mesh-convergence evidence. In the anchorage pull-out example (§4.2), the coarse mesh is reported to delay crack initiation and generate a crack surface that penetrates the whole body, with a final dissipated energy substantially different from the finer meshes (Figure 18). In §4.4, the coarse mesh fails to branch, and in §4.5 the two highest-velocity cases produce 'more scattered final dissipated energy' across meshes. These results show mesh sensitivity in exactly the cases that support the branching claims. Please provide quantitative metrics, such as crack-surface angle, branching-point location, or energy error as a function of mesh size, or explicitly state the resolution requirement for the method to be predictive.","section":"§4.2, §4.4, §4.5"},{"comment":"The method section states that the remaining portions of a split parallelepiped element can be replaced by triangular prisms 'without significant loss of accuracy', but no analysis or numerical test is provided. Since all benchmarks in Section 4 use tetrahedral elements, the ES-H-FEM/hexahedral version is not validated. Either add a hexahedral-mesh benchmark or explicitly scope the claims of the paper to the tetrahedral CEM formulation.","section":"§3, hexahedral elements"}],"minor_comments":[{"comment":"There are many typographical errors that should be corrected: 'Griffth' for Griffith, 'banching' for branching, 'Poisson ration', 'facture', 'Beisdes', and 'the we consider' are examples. The quality of the manuscript would be improved by a careful proofreading pass.","section":"Throughout"},{"comment":"Several matrix equations are garbled in the submitted text, making the ES-FEM discretization difficult to follow. Please ensure the mathematical typesetting is complete and unambiguous.","section":"Eqs. (6), (7), (9), (10)"},{"comment":"The reference to Xie et al. (2025) is incomplete: it contains only 'URL: ...  ...' and no title, venue, or full citation. This is the predecessor work on which the 3D G-formulation is based, so it must be cited completely.","section":"References"},{"comment":"The claim that branching is 'completely spontaneous without any locally/globally defined criteria' is overstated. Branching is emergent, but it is influenced by the mesh topology and by the element-splitting algorithm; the criterion is local but it is still a criterion. Please rephrase to reflect this.","section":"§4"},{"comment":"The text says the dissipated energy results 'align well with the theoretical value', but the theoretical value is not shown or defined in the figure or text. Please include the reference value or remove this claim.","section":"Figure 12 and §4.1"}],"recommendation":"major_revision","confidential_remarks":"The paper's central novelty is the 3D element-splitting topology and the associated local G. The authors' own 2D predecessor (Xie et al. 2025) is cited as the source of the criterion, but the paper does not clearly delineate what is new beyond the geometric extension. This is relevant for evaluating the novelty claim and should be clarified during revision. I do not see evidence of fabrication or improper data handling; the comparisons with published results appear honest, including cases where the method fails on coarse meshes."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Alright, quick take. This paper is a real extension of the 2D crack element method to 3D. The element-splitting patterns for tets and hexes, the local G formula, and the GPU implementation are new relative to prior work, including their own 2D paper. And the headline result holds up in the benchmarks: branching appears without an explicit branching criterion, and the crack paths broadly match Kalthoff-Winkler, the concrete pull-out, the compact compression, and the Ozbolt compact-tension experiments. That's worth something; most 3D methods need ad hoc rules to branch.\n\nThe soft spot is the load-bearing one, and the stress-test note lands. Equations (17)–(18) define G as a product of the normal projection of edge stretch and the normal projection of maximum principal stress. That is not the fracture energy release rate in any standard sense. The crack-plane traction is not the maximum principal stress, and the edge stretch is not the crack opening displacement, especially in shear-dominated regions like the Kalthoff-Winkler plate. The paper says the formula is \"derived\" from the evolving topology, but no derivation is given, and it is not checked against an analytical solution or a J-integral/VCCM computation. The benchmarks are mostly visual crack paths plus dissipated energy curves; those are aggregate outputs and can coincide even if the local criterion is wrong. The paper itself reports that branching is mesh-sensitive (§4.4), and the coarse mesh misses it, which is consistent with a criterion that is not robustly physical.\n\nOther gaps are smaller but real: GPU speedup is never quantified—no wall-clock times or speedup factors—and no code or data are provided. The dissipated-energy comparison in the compact compression case is 2D-normalized, which is fine but should be stated more carefully. The co-planarity proof in the appendix is a nice formal touch.\n\nSo: the paper is not ready to be relied on as a validated fracture law, but it is a serious engineering method paper with a credible 3D extension and honest benchmark selection. I'd send it to peer review, with the request that the authors either derive or numerically verify the G formula, add a convergence study, quantify the GPU speedup, and release code/data. If the G question is resolved, this could be a useful tool for industrial-scale dynamic fracture.","headline":"A genuine 3D extension of the 2D crack element method that does produce spontaneous branching, but the central G-criterion is asserted rather than derived, so treat it as a promising numerical demo, not a validated fracture law.","tokens_in":21457,"tokens_out":2643,"would_cite":false,"duration_ms":32917,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["74R10","74S05","65M60"],"pacs":[],"model":"deepseek-v4-flash","headline":"A three-dimensional crack element method that splits elements when a local fracture energy release rate exceeds the critical value reproduces both single dynamic cracks and spontaneous branching, with GPU acceleration, across five benchmark","keywords":["3D crack branching","crack element method","dynamic crack propagation","ES-T/H-FEM","fracture energy release rate","GPU acceleration","quasi-brittle materials","element splitting"],"falsifier":"Compute $G$ from Eqs. (17)–(18) element by element in a 3D simulation and compare it with a direct energy-balance estimate (change in stored strain energy divided by the newly created crack area from the same mesh). A systematic mismatch, or a mesh-converged crack path that misses the measured ~65° Kalthoff angle or the observed concrete branching pattern, would falsify the claim.","tokens_in":20507,"feed_emoji":"💥","tokens_out":11562,"duration_ms":118580,"temperature":0.7,"pith_summary":"The paper aims to establish that transient dynamic fracture in three dimensions—curved single cracks and crack branching—can be simulated with an element-level rule: split a tetrahedral (or hexahedral) element along a plane set by its edge quadrature points, compute a local fracture energy release rate $G$ from edge stretch and maximum principal stress projected on the crack-front normal, and deactivate the element when $G$ exceeds the critical value $G_c$. The authors derive an original three-dimensional formula for $G$ from the evolving topology of split elements and discretize the formulation with edge-based smoothed finite elements (ES-T-FEM/ES-H-FEM). Five benchmarks—Kalthoff–Winkler shear impact, concrete anchorage pull-out, PMMA compact compression, and two crack-branching plates under Neumann and Dirichlet boundary conditions—yield crack paths and dissipated energies consistent with experiments and reference simulations. This matters because three-dimensional crack branching has no universally accepted criterion; here branching emerges without any explicit branching rule, and GPU acceleration makes the 3D computation practical.","feed_headline":"Element splitting reproduces 3D crack branching on GPUs","feed_subtitle":"A local energy-release-rate rule drives element deactivation, matching impact, pull-out, and branching experiments.","key_machinery":"The load-bearing object is the three-dimensional crack element: a tetrahedron or parallelepiped hexahedron whose edge-quadrature-point set defines candidate crack planes. The fracture energy release rate $G$ (Eqs. (17)–(18)) is constructed from the projected edge stretch and projected maximum principal stress relative to the candidate plane's unit normal, and the threshold $G > G_c$ triggers element splitting/deactivation. Edge-based smoothed finite element discretization (ES-T-FEM for tetrahedra, ES-H-FEM for hexahedra) provides smoothed strains over edge-sharing element patches and a lumped mass matrix for explicit central-difference time stepping; GPU parallelization carries the 3D benchm","core_discovery":"The central discovery is that a crack front in 3D can be advanced piecewise by splitting elements, and that the energy-release-rate needed to decide the split can be read off the split element's own topology. For a constant-strain tetrahedron, the candidate crack surface is either a quadrilateral or a triangle formed by edge quadrature points; the fracture energy release rate $G$ is built from the stretch of the two cracked edges and the maximum principal stress on the opposite edge, both projected on the unit normal to the candidate plane (Eqs. (17)–(18)). When $G$ exceeds $G_c$, the new crack surface follows that pattern and the element is deactivated, while partially cracked hexahedra are","pith_inferences":["Editorial inference: Because the criterion is purely local and removes elements rather than inserting new surfaces, the same splitting procedure could serve as a low-cost screening tool for industrial fatigue or impact assessments, where full crack-front tracking is currently too expensive.","Editorial inference: The paper validates $G$ indirectly through crack paths and energy curves; an element-by-element energy-balance check that compares Eqs. (17)–(18) with the actual strain-energy drop per unit new crack area would sharpen the criterion's physical status.","Editorial inference: The coplanarity proof in Appendix A.1 guarantees the quadrilateral crack face lies in a plane for the described quadrature-point configuration; whether a similar construction works for general, non-parallelepiped hexahedral meshes remains an open extension."],"forward_implications":["3D dynamic crack branching can be captured without any explicit branching criterion; branching appears spontaneously when enough elements satisfy $G > G_c$.","Crack path direction in the Kalthoff–Winkler benchmark stays within the experimental range even on a coarse mesh with roughly 20,000 tetrahedra, far fewer than reference simulations.","Fine and medium meshes give dissipated-energy and load-displacement curves close to published transient-dynamic results for the anchorage pull-out and compact compression tests.","Under Dirichlet boundary conditions the method reproduces experimentally observed single-crack and branching patterns in concrete compact tension, including the rate-dependent transition; under Neumann traction loading, a sufficiently fine mesh (typically over 80,000 elements) is needed for branching."],"supporting_citations":[{"why":"Prior 2D CEM whose element-failure criterion and fracture-energy-release-rate computation are extended to 3D.","marker":"Xie, Wu, Xu, Perez and Li (2025)"},{"why":"Supplies the ES-T-FEM discretization used to approximate the 3D displacement field.","marker":"He, Li, Zhong, Cheng, Zhang, Liu, Li and Zhou (2013)"},{"why":"Provides the edge-notched plate shear-impact experiment used as the first 3D benchmark.","marker":"Kalthoff (1988)"},{"why":"Supplies the PMMA compact compression experiment with a curved 3D crack path.","marker":"Rittel and Maigre (1996)"},{"why":"Supplies concrete compact tension experiments with rate-dependent single-crack and branching patterns.","marker":"Ožbolt et al. (2013)"},{"why":"Three-dimensional dynamic crack-growth reference for comparing pull-out dissipated energy and load-displacement curves.","marker":"Duan et al. (2009)"},{"why":"Phase-field dynamic brittle fracture model used as a reference for dissipated energy in the branching benchmark.","marker":"Borden et al. (2012)"},{"why":"Three-dimensional crack propagation modeling in concrete whose pull-out results are compared with the dynamic CEM solution.","marker":"Gasser and Holzapfel (2005)"},{"why":"XFEM-based 3D crack initiation/propagation analysis; its quasi-static pull-out results serve as comparison.","marker":"Areias and Belytschko (2005)"},{"why":"Nonlocal gradient damage model used to compare dissipated energy and mesh efficiency in the Kalthoff and compact-compression benchmarks.","marker":"Tran and Bui (2024)"}],"fun_headline_variants":["3D crack branching via element splitting, GPU-accelerated","Element-splitting rule drives 3D crack growth on GPUs","GPU-accelerated 3D crack method captures branching","Crack growth: read energy release from split tetrahedra","GPU-accelerated element-splitting simulates 3D cracking"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The load-bearing premise is that the local quantity in Eqs. (17)–(18)—edge stretch times maximum principal stress projected on the crack-front normal—equals the true fracture energy release rate per unit crack area; if that equivalence fails under a general stress state, the $G > G_c$ deactivation rule loses its physical grounding.","fun_headline_variants_meta":{"raw":{"variants":["3D crack branching via element splitting, GPU-accelerated","Element-splitting rule drives 3D crack growth on GPUs","GPU-accelerated 3D crack method captures branching","Crack growth: read energy release from split tetrahedra","GPU-accelerated element-splitting simulates 3D cracking"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000827,"raw_usage":{"total_tokens":3398,"prompt_tokens":640,"completion_tokens":2758,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":384,"completion_tokens_details":{"reasoning_tokens":2669}},"tokens_in":384,"tokens_out":2758,"duration_ms":23198,"temperature":1.0,"reasoning_tokens":2669,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T00:55:25.704909+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $G$ from Eqs. (17)–(18) element by element in a 3D simulation and compare it with a direct energy-balance estimate (change in stored strain energy divided by the newly created crack area from the same mesh). A systematic mismatch, or a mesh-converged crack path that misses the measured ~65° Kalthoff angle or the observed concrete branching pattern, would falsify the claim.","supporting_citations":[{"cited_title":", author Maigre, H","cited_arxiv_id":null,"evidence_quote":"Supplies the PMMA compact compression experiment with a curved 3D crack path."},{"cited_title":", author Song, J.H","cited_arxiv_id":null,"evidence_quote":"Three-dimensional dynamic crack-growth reference for comparing pull-out dissipated energy and load-displacement curves."},{"cited_title":", author Holzapfel, G.A","cited_arxiv_id":null,"evidence_quote":"Three-dimensional crack propagation modeling in concrete whose pull-out results are compared with the dynamic CEM solution."},{"cited_title":", author Belytschko, T","cited_arxiv_id":null,"evidence_quote":"XFEM-based 3D crack initiation/propagation analysis; its quasi-static pull-out results serve as comparison."},{"cited_title":", author Bui, T.Q","cited_arxiv_id":null,"evidence_quote":"Nonlocal gradient damage model used to compare dissipated energy and mesh efficiency in the Kalthoff and compact-compression benchmarks."}],"review_version":1}