{"id":"64072d68-2769-4140-92af-c2380df29eb8","arxiv_id":"2607.23525","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.5,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Dual topological derivatives plus interface-enriched FEM let 3-D level-set topology optimization minimize aggregated boundary energy release rates from a single uncracked analysis while nucleating holes.","lead":"A 3-D topology optimization method designs brittle parts to resist cracking by estimating energy release rates from one uncracked stress analysis, using topological derivatives both to score the boundary and to open new holes. It matters for anyone who needs fracture-aware structural layouts without meshing every candidate crack.","discovery_kind":"extension","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"The fracture objective rests on a single weight-function matrix H, calibrated once on one cuboidal cracked specimen (App. B), applied pointwise at every enriched node on arbitrarily curved, evolving boundaries — and the paper never checks the surrogate ERR against an explicitly cracked analysis of a","rationale":"The reader identified the same soft spot: the asymptotic small-crack ERR formulas with fixed ε = 1% used as an optimizable surrogate for finite-crack fracture risk (Eqs. 30–34, §3.2). My stress-test sharpens where this assumption is least secure: not the asymptotic expansion itself (Alidoost et al. 2020 is a legitimate published basis), but (a) the universality of a single numerically calibrated H across all boundary geometries, and (b) the point-stress evaluation in precisely the high-gradient, high-curvature regions that dominate the p-mean objective. This is a correctness-risk concern, not circularity: the paper is internally consistent, the sensitivity pipeline is unusually complete and FD-verified, and the framework demonstrably produces sensible designs. The gap is the missing end-to-end validation — surrogate ERR vs. explicitly cracked analysis — which is entirely feasible with the authors' own published enriched-crack tooling and interaction-integral machinery. This reinforces rather than revises the reader's CONDITIONAL: the contribution is accept-shaped if the surrogate is validated against explicit cracked analyses at representative optimized designs (or its error bounds stated), and if the reported stress artifacts at slender cut elements are addressed. I see no load-bearing contradiction that would warrant REJECT, and the FD-verified adjoint plus consistent physical trends (spherical cavity under triaxial load, 45° rotation under shear) provide genuine independent support. Hence verdict UNCHANGED (CONDITIONAL stands), with the concrete cracked-design check as the decisive condition.","tokens_in":33839,"tokens_out":1856,"duration_ms":132400,"concrete_test":"Take the final optimized L-bracket design (§4.4) and the mode-I optimized cavity (§4.3). At the ~5 boundary nodes with the largest surrogate G, explicitly insert half penny-shaped cracks of radius ε = 0.01 with the paper's orientation rule, using the authors' own discontinuity-enriched 3-D formulation (Zhang et al. 2019) and compute K_I, K_II, K_III via the interaction integral (Eqs. 49–51, already implemented for App. B). Compare against the surrogate predictions from the uncracked analysis. If the relative error at the hottest nodes exceeds ~20–30%, or the ranking of the top nodes changes, the objective is mis-weighting fracture risk and the headline reductions need qualification; repeat at a mid-optimization sharp-ligament configuration to probe the asymptotics where curvature ~ ε.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that a single uncracked IGFEM analysis plus Eq. (31) — K = H σ′ with H a constant 3×3 weight matrix — gives ERRs faithful enough to steer optimization. Two conditions must hold: (i) H is universal, i.e., independent of local boundary curvature, feature size, and crack-to-surface geometry; (ii) the stress entering Eq. (31) can be taken as a point value (recovered nodal stress) rather than a weighted integral over the crack face. Neither is established. Appendix B derives H from one calibration: a cuboidal specimen with a penny-shaped crack under three load cases (Fig. 15), then reuses that same H at all boundary nodes in all examples, including the sharp-curved cavity morphologies of §4.3 and the L-bracket's evolving re-entrant corner. Yet the small-crack asymptotics (cited to Alidoost et al. 2020) assume the crack is small relative to both the stress-gradient length scale and the local surface radius of curvature. With ε fixed at 1% of characteristic size, this is exactly what fails at the sites that matter most: near stress concentrations the uncracked stress varies appreciably over length ε, and mid-optimization the boundary routinely passes through slender ligaments and sharp transient features (the paper itself reports such transients, §4.3/§4.4). In those regimes a point-stress, constant-H evaluation can mis-rank candidate crack sites, and since the objective is a p-mean (p=8) that heavily weights the largest G_i, an error concentrated at the hottest node directly steers the design. The paper's own validation only covers the adjoint sensitivities (App. H, FD checks) — i.e., the surrogate is differentiated consistently — not whether the surrogate approximates true fracture driving force. No example inserts an actual crack into an optimized design to confirm the claimed ERR reduction (e.g., the 44%/63%/27% peak-ERR reductions of §4.3) survives a real cracked analysis.","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript presents a 3-D topology optimization framework for fracture mitigation in brittle solids. The design is described by an RBF-parameterized level set function (with a new quadratic interpolation of the zero contour and a PDE-free re-initialization), analyzed with the interface-enriched generalized finite element method (IGFEM) on fixed structured meshes. Topological derivatives are used twice: (i) a Garreau-type asymptotic expression drives hole nucleation, and (ii) the small-crack asymptotic formulas of Alidoost et al. (2020) are used to evaluate energy release rates (ERRs) for virtual half penny-shaped cracks at every enriched boundary node, from a single analysis of the uncracked geometry. Crack orientation follows a maximum-hoop-stress criterion; stresses entering the ERR evaluation are obtained via a 3-D SIP stress-recovery procedure; boundary ERRs are aggregated with a p-mean (p=8) objective; and full adjoint sensitivities are derived and verified against finite differences (App. H). Three examples are shown: a triaxially loaded cube (shape, then topology optimization), mode I/II/III cavity tailoring, and an immersed 3-D L-bracket.","tokens_in":34266,"tokens_out":3659,"duration_ms":230020,"significance":"If the surrogate ERR evaluation is trustworthy, this is a significant contribution: it is, to my knowledge, the first 3-D fracture-aware topology optimization that evaluates ERRs at all boundary sites from a single uncracked analysis, avoiding per-crack-configuration solves, and the first to combine this with topological-derivative hole nucleation (removing the pre-seeded-hole limitation of Zhang et al. 2022). The manuscript ships several concrete strengths: a complete, finite-difference-verified adjoint sensitivity derivation for a nontrivial chain (quadratic LSF intersection, surface normals, principal-direction rotation, SIP-recovered stresses), a demonstrated ablation showing that both quadratic LSF interpolation and re-initialization are needed for the sphere benchmark (Fig. 9), and examples whose qualitative behavior matches fracture-mechanics intuition (near-sphere under triaxial tension, 45° alignment under shear, re-entrant corner rounding). The examples are reproducible in principle from the stated parameters. The load-bearing question is whether the constant weight-function ERR surrogate remains accurate on the evolving, curved, slender geometries the optimizer produces","major_comments":[{"comment":"The central claim — that ERRs faithful enough to steer optimization are obtained from one uncracked analysis via K = H σ′ — rests on a weight-function matrix H calibrated once, on a single cuboidal specimen with a penny-shaped crack under three load cases (App. B, Fig. 15), and then applied pointwise at every enriched node in all examples. The small-crack asymptotics (Alidoost et al. 2020) require the crack to be small relative to both the local stress-gradient length scale and the local surface radius of curvature. Neither condition is checked anywhere in the paper, and both are most likely violated exactly where the objective is most sensitive: with p=8 the p-mean is dominated by the hottest sites, which are stress concentrations (steep stress gradients over the crack radius ε = 1% of characteristic size) and, mid-optimization, sharp transient features — the authors themselves report s","section":"§3.2, Eqs. (30)–(34); App. B"},{"comment":"Only configurations with K_I > 0 are retained in the ERR evaluation and aggregation (stated after Eq. (33) and again at the end of App. C). This truncation makes each G_i a nonsmooth function of the design variables at the surface K_I = 0: the objective in Eq. (29) is then only piecewise differentiable, and the adjoint derivation in App. F (Eqs. (79)–(80)) treats G_i as smooth everywhere. Since enriched nodes continually cross the K_I = 0 threshold as the boundary evolves, the authors should (a) state how the K_I > 0 filter is implemented in the sensitivity chain (subgradient, smooth clamp, or exclusion with a vanishing-contribution argument), and (b) confirm that the finite-difference verification in App. H exercises at least one design variable whose perturbation moves a node across the K_I = 0 boundary; as presented, Fig. 19 does not indicate whether this regime is covered.","section":"§3.2 and App. C, Eq. (34)/(59); App. F"},{"comment":"The triaxial-cube example is described as verifying the formulation, but the check is qualitative: convergence to a near-sphere (r_x=0.443, r_y=0.435, r_z=0.429) and a uniform ERR distribution are compared against physical expectation, not against any independent fracture calculation. Since this is the paper's only 'verification' example for the ERR surrogate, and the exact solution for the optimal cavity under hydrostatic tension is known to be a sphere, the example verifies the optimization machinery but not the fracture quantities. A quantitative benchmark — e.g., comparing the surrogate G field on the optimized sphere against explicitly cracked analyses at several boundary sites — would substantially strengthen the paper and is the natural companion to the validation requested in the first comment.","section":"§4.1, Fig. 8"},{"comment":"The L-bracket is the paper's flagship demonstration on a 'complex engineering structure', yet the fracture interpretation is problematic as presented. The objective is lower in early iterations than at convergence, which the text attributes to the volume constraint only being satisfied after iteration 80 — i.e., the low early values occur at infeasible, bulkier designs. The headline comparison (initial vs. final ERR fields, bottom of Fig. 14) therefore compares designs at different volumes, and the claim of improved fracture resistance conflates material removal with corner rounding. The authors should report the peak/p-mean ERR trajectory restricted to feasible iterates, and ideally compare the final design against a volume-matched un-optimized L-bracket (sharp corner, same V_s) so that the fracture benefit of the optimization itself is isolated. In addition, the discretization has only","section":"§4.4, Fig. 14"}],"minor_comments":[{"comment":"The section heading '3 Examples of citations, figures, tables, references' is a leftover template title and does not describe the content (topology description, problem formulation, stress recovery, sensitivities). Please rename.","section":"§3 (heading)"},{"comment":"Typo: 'Furthre details' → 'Further details'. Also 'a multi-point constraint system is setup' → 'is set up'.","section":"§2.2, penultimate paragraph"},{"comment":"Typo: 'Wee see that the hole nucleation procedure...' → 'We see...'. Also 'This rounding continuous progressively' → 'continues'.","section":"§4.4"},{"comment":"The citations 'osh [2004]' and 'Andrew [2000]' appear garbled: the former is presumably Osher & Fedkiw (the 2004 Appl. Mech. Rev. entry in the bibliography is a review of their book), and Andrew [2000] is a book review of Sethian's monograph. Please cite the primary sources.","section":"§3.5 and references"},{"comment":"The parenthesization in the first line of Eq. (38) is garbled ('∂(KU)/∂s_j − ) ) ∂F/∂s_j )'), apparently from a line-break artifact. Please reformat.","section":"Eq. (38)"},{"comment":"The number of enriched nodes |ι_w| changes from iteration to iteration as the interface sweeps the mesh, so the p-mean normalization changes with the design. A sentence clarifying that this poses no issue for the per-iteration MMA treatment (and whether it contributes to the observed objective oscillations) would help the reader.","section":"§3.2, Eq. (29)"},{"comment":"The crack radius ε = 1% of characteristic size, the nucleation threshold ϕ < −4h, the 5-iteration nucleation cadence, and the per-example move limits and regularization schedules (§4.3) constitute a sizable heuristic set. The authors acknowledge this in §5, but a short sensitivity study on at least ε (does the ranking of candidate sites, not just the scale of G, depend on it?) would be valuable.","section":"§4, parameters"},{"comment":"The caption reports 'close agreement' but gives no quantitative statement (e.g., minimum relative error achieved and the step size at which it occurs). Please state the best relative error and discuss the upturn/plateau at small Δs_j.","section":"App. H, Fig. 19c"},{"comment":"The reported improvement from shape to topology optimization (0.0358 → 0.0286) is at fixed volume constraint; please state explicitly whether the three-cavity final design and the single-cavity design are compared at the same V_s, since cavity nucleation transiently changes the solid volume.","section":"§4.2, Fig. 10"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a competent and well-engineered extension of the group's prior 2-D work (Zhang et al. 2022), and the self-citation pattern is appropriate for a continuation. My concern is purely technical: the paper's novelty and practical value hinge on the fidelity of the constant-weight-function ERR surrogate away from its calibration geometry, and the manuscript contains no direct evidence on this point — no comparison against an explicitly cracked analysis anywhere. The requested validation is feasible with tools the authors clearly have (they compute interaction integrals in App. B), so I regard this as addressable within one revision cycle. The leftover template heading in §3 and citation artifacts suggest the submission was prepared hastily; the editor may wish to note this."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The real news is the 3-D leap. They take the 2-D fracture-aware IGFEM/level-set idea from Zhang et al. 2022 and make it work in three dimensions: half-penny cracks at enriched boundary nodes, ERRs from 3-D topological derivatives after one uncracked analysis, SIP stress recovery, quadratic edge intersection for the LSF, signed-distance re-init without a Hamilton–Jacobi solve, and—new relative to that paper—topological derivatives also used to nucleate holes. Dual TD plus fixed-mesh enriched analysis is the package they brand ∂^{2}(TO).\n\nWhat they did well is the engineering completeness. Formulation, adjoint, and FD checks (App. H) are unusually thorough for a TO paper. Examples behave as physics would suggest: triaxial cube goes spherical under shape-only and multi-cavity when nucleation is allowed; mode I/II/III cavities reorient or modulate sensibly; immersed L-bracket rounds the re-entrant corner and redistributes material. Self-citations are prior tools, not circular claims. Circularity burden is low.\n\nThe soft spot that matters is the one the stress-test flags, and it is real but not fatal. Weight matrix H is calibrated once on a cuboidal penny-crack specimen (App. B) and reused everywhere. Small-crack asymptotics need ε small relative to stress gradients and local curvature; they fix ε at 1% of domain size and still optimize through ligaments and sharp transients where that assumption is weakest. They never close the loop by inserting an actual crack into an optimized geometry and checking that the reported peak-G drops survive a cracked analysis. Sensitivity consistency is verified; surrogate fidelity to true fracture driving force is not. Hyperparameters (nucleation threshold/frequency, move limits, regularization schedule, volume continuation) are also hand-tuned per example. Residual cut-element stress artifacts remain after SIP; they say so themselves.\n\nNone of that invents a contradiction in the stated numerical claims. It is a methods paper that delivers a working 3-D pipeline on synthetic benchmarks. For people already doing enriched TO or fracture-aware design it is worth reading and citing; for a general solid-mechanics audience it is incremental but clean. I would send it to referees. Ask them to demand either a cracked-geometry validation on one optimized design or a clear limitation statement on the constant-H/point-stress surrogate, plus tighter discussion of the free parameters. Engage.","headline":"Solid 3-D methods extension of Zhang et al. 2022: dual topological derivatives + IGFEM give a usable fracture-mitigation TO pipeline, with one real open question on whether the constant-H ERR surrogate stays faithful on curved evolving boundaries.","tokens_in":33753,"tokens_out":616,"would_cite":true,"duration_ms":14566,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A single uncracked 3-D analysis plus dual topological derivatives can drive topology optimization that lowers energy release rates along the entire boundary of brittle solids.","keywords":["topology optimization","interface-enriched GFEM","linear elastic fracture mechanics","topological derivatives","brittle fracture","energy release rate","level-set method","stress recovery"],"falsifier":"Re-analyze any optimized geometry with explicit finite cracks of realistic size placed at the high-ERR sites predicted by the topological-derivative formula; if the true energy-release rates or crack paths differ sharply from the surrogate ranking, or if a stress-only redesign of equal volume outperforms it under the same fracture test, the central claim fails.","tokens_in":33382,"feed_emoji":"⚒️","tokens_out":967,"duration_ms":20611,"temperature":0.7,"pith_summary":"This paper claims that fracture risk in three-dimensional brittle solids can be reduced by topology optimization without ever meshing or analyzing cracked geometries. Geometry is represented by a level-set field built from radial basis functions; stresses come from one interface-enriched finite-element solve of the intact body. Topological derivatives then do two jobs at once: they nucleate new holes inside the solid, and they convert the recovered boundary stress into energy-release-rate estimates for tiny half-penny cracks assumed to start at every enriched boundary node, oriented by the maximum-hoop-stress rule. Those rates are folded into a single p-mean objective that the optimizer minimizes under a volume limit. On a triaxially loaded cube, mode-I/II/III cavity problems, and a fully immersed L-bracket, the method produces rounded, multi-cavity layouts whose peak boundary energy-release rates drop substantially relative to the starting designs. A sympathetic reader cares because the usual cost of fracture-aware design—one expensive cracked analysis per candidate crack site—is replaced by a single uncracked solve plus cheap asymptotic formulas, making three-dimensional fracture-driven topology optimization practical on fixed background meshes.","feed_headline":"One uncracked 3-D solve drives fracture-aware topology design","feed_subtitle":"Dual topological derivatives nucleate holes and score boundary energy-release rates without meshing cracks","key_machinery":"∂²(TO): the dual use of three-dimensional topological derivatives—once to nucleate holes, once to convert recovered uncracked stresses into energy-release rates of virtual half-penny cracks at enriched boundary nodes—feeding a p-mean objective on a level-set geometry analyzed by IGFEM.","core_discovery":"Dual topological derivatives, paired with interface-enriched finite-element analysis and non-local stress recovery, let a designer evaluate and minimize energy release rates along the entire free boundary of a three-dimensional brittle solid from one analysis of the uncracked geometry, while the same derivatives nucleate holes so that topology and shape evolve together.","pith_inferences":["If the asymptotic surrogate remains accurate under moderate mesh coarsening, the same pipeline could be embedded in real-time design loops for additively manufactured brittle parts.","The dual-derivative idea may transfer to other local failure indicators (fatigue, corrosion) that admit topological expansions, not only brittle fracture.","Persistent stress over-estimation in slender cut elements, noted by the authors, suggests that further enrichment or cut-element stabilization could tighten the gap between surrogate and true ERR."],"forward_implications":["Fracture-aware three-dimensional topology optimization can run on fixed structured meshes without remeshing cracked configurations.","Hole nucleation and boundary motion can be driven by the same asymptotic quantity, reducing dependence on a pre-seeded initial design.","Mode-I, II and III cavity problems yield distinct optimal void shapes (elongated, rotated/separated, wavy), showing the objective distinguishes fracture mechanisms.","On the classical L-bracket the method automatically rounds the re-entrant corner and inserts interior voids while lowering peak boundary energy-release rate."],"fun_headline_variants":["Dual topological derivatives score boundary ERRs from one uncracked 3-D solve","Fracture-aware 3-D topology from a single uncracked enriched FE analysis","Hole nucleation and ERR minimization via dual topological derivatives","Minimize boundary energy release rates without meshing cracks in 3-D","Dual derivatives nucleate holes and rank ERRs in one uncracked 3-D solve"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"Tiny half-penny cracks whose energy-release rates are read from the stress of the uncracked body at boundary nodes remain a faithful, optimizable stand-in for real fracture risk as the three-dimensional design changes.","fun_headline_variants_meta":{"raw":{"variants":["Dual topological derivatives score boundary ERRs from one uncracked 3-D solve","Fracture-aware 3-D topology from a single uncracked enriched FE analysis","Hole nucleation and ERR minimization via dual topological derivatives","Minimize boundary energy release rates without meshing cracks in 3-D","Dual derivatives nucleate holes and rank ERRs in one uncracked 3-D solve"]},"model":"grok-4.5","effort":"low","cost_usd":0.00497,"raw_usage":{"total_tokens":1364,"prompt_tokens":749,"num_sources_used":0,"completion_tokens":86,"cost_in_usd_ticks":49704000,"prompt_tokens_details":{"text_tokens":749,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":529,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":749,"tokens_out":86,"duration_ms":8882,"temperature":1.0,"reasoning_tokens":529,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-30T20:03:05.193809+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Re-analyze any optimized geometry with explicit finite cracks of realistic size placed at the high-ERR sites predicted by the topological-derivative formula; if the true energy-release rates or crack paths differ sharply from the surrogate ranking, or if a stress-only redesign of equal volume outperforms it under the same fracture test, the central claim fails.","supporting_citations":[],"review_version":1}