{"id":"71980590-f7a2-438a-9baf-1936bb76b785","arxiv_id":"2506.15892","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Mesh morphing preserves mesh connectivity across geometric variations, enabling accurate ensembles and reduced-order models for dynamic rupture and subduction thermal sensitivity.","lead":"This paper shows that one computational mesh can be smoothly reshaped into many fault and slab geometries while keeping its internal structure intact, making it possible to build cheap surrogate models. The technique is demonstrated on earthquake rupture and subduction temperature models, where the surrogates evaluate new geometries orders of magnitude faster than full simulations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Slab2 uncertainty example lacks exact-mesh thermal verification: up to 5.5 km morphing error near the surface could contaminate the reported 40 K temperature variability.","rationale":"The core morphing and ROM demonstration is generally well supported: exact-versus-morphed comparisons exist for the dynamic rupture example and the curvature thermal example, and the held-out ROM tests show low errors. The reader's concern about the modified TPV13 parameters is legitimate but mostly affects the physical interpretation of the dip-sensitivity results, not the validity of the morphing method or the ROM construction itself. My concern is closer to the central claim: the Slab2 uncertainty example, one of the two headline applications, does not verify that thermal output on morphed meshes matches exactly generated meshes. Given the reported 5-5.5 km interface error in the upper 10-20 km and the large shallow temperature variability, an exact-mesh comparison is needed to separate genuine geometric sensitivity from morphing error. This is addressable with a small number of additional simulations, so the paper remains close to acceptance but should be revised to include that check or to qualify the 40 K conclusion.","tokens_in":34174,"tokens_out":7196,"duration_ms":79076,"concrete_test":"Run the thermal FOM on exactly generated meshes for at least β = -1, -0.5, 0, 0.5, 1 using the same model parameters as Section 4.4.4, and compare slab-interface temperature versus depth against the morphed-mesh results. Report the maximum and mean absolute temperature differences over 0-150 km depth. If these differences are small relative to the claimed 22-40 K variability (e.g., less than 5 K), the conclusion stands; if they are comparable, report the variability as a range that explicitly includes morphing uncertainty.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The abstract claims that in both applications morphed meshes 'lead to accurate simulation results that closely match those obtained using exactly generated meshes.' This is verified for the dynamic rupture example and for the curvature-based thermal example (Section 4.3.4), but not for the Slab2 uncertainty example in Section 4.4. There, the paper reports mesh quality and geometric accuracy, but no thermal simulation on exactly generated meshes is compared to the morphed-mesh results. The reported geometric error is up to 5-5.5 km in the upper 10-20 km of the mesh (Section 4.4.3), precisely where the temperature variability is largest: mean 22 K above 150 km depth and up to 40 K overall (Section 4.4.4). Because shallow slab depth strongly controls forearc temperatures, a 5 km interface offset could translate into temperature differences comparable to the claimed variability. The validation in the curvature example does not automatically transfer: the Slab2 example uses a younger plate age (8 Myr), a thinner slab, a different reference profile, and a different morphing parameterization (scaling of depth uncertainty), so the earlier agreement does not bound the error here. This is a specific, testable gap in the central claim for one of the two headline applications.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents an RBF-based mesh morphing workflow that deforms a reference mesh to represent different geometric configurations while preserving connectivity, and demonstrates it in two applications: 3D dynamic rupture (varying fault dip theta in a TPV13-style model) and 2D subduction thermal modeling (varying slab curvature alpha and Slab2-uncertainty scaling beta). For the dynamic rupture case, the authors compare accumulated slip, peak slip rate, 12 receiver velocity traces, and uplift profiles between morphed and exactly generated meshes, and build interpolated POD (iPOD) ROMs for receiver velocity and surface displacement, validating via leave-one-out cross-validation and held-out dips with a maximum reported error of 0.007 m. For the thermal case, they compare temperature-depth curves for the curvature parameterization and report mesh quality and geometric accuracy for the Slab2 uncertainty case. The central claim is that mesh morphing preserves connectivity and yields simulation outputs matching exactly generated meshes, thereby enabling ROM-based geometric sensitivity analysis.","tokens_in":34343,"tokens_out":8217,"duration_ms":73826,"significance":"If verified, this is a substantial methodological contribution: it removes the manual meshing bottleneck for geometric ensembles and, because connectivity is preserved, enables non-intrusive ROMs for geometric parameters. The verification is unusually thorough for a methods paper: exact-mesh comparisons cover multiple output quantities, ROM validation includes leave-one-out cross-validation and genuinely held-out dip values, and the code and data are openly archived. The Slab2 uncertainty example, however, currently lacks an exact-mesh thermal comparison, and the dynamic rupture sensitivity analysis relies on modified TPV13 parameters; both points need to be addressed before the claims in the abstract and conclusions are fully supported.","major_comments":[{"comment":"The Slab2 uncertainty example reports geometric accuracy (maximum 5-5.5 km offset in the upper 10-20 km of the mesh) and thermal variability (up to 40 K, mean 22 K above 150 km depth), but it never compares thermal simulation output on morphed meshes against simulations on exactly generated meshes. This is the only application in which the central claim of the abstract, that morphed meshes lead to accurate simulation results that closely match those obtained using exactly generated meshes, is not verified. Because the geometric error is concentrated exactly where the temperature variability is largest, the reported 40 K variability could be partly attributable to morphing error rather than to Slab2 depth uncertainty. Please run the thermal model on the exact meshes for the beta values and report the interface temperature and temperature-depth comparisons.","section":"Section 4.4 (Slab2 uncertainty example)"},{"comment":"The dip range theta in [50,70] degrees used for the ROMs and the sensitivity analysis is only accessible after modifying two TPV13 benchmark parameters: the nucleation-patch static friction coefficient is lowered from 0.54 to 0.48 and the bulk cohesion from 5.0e6 to 1.0e6 Pa. With the unmodified benchmark, rupture nucleates only for theta in [55,63] degrees. The paper should test whether the reported dip sensitivity and ROM accuracy are robust to the specific way the parameter space is widened (e.g., by instead changing the prestress or nucleation size), or should state explicitly that the dip-sensitivity conclusions are conditional on the modified setup. Without such a test or caveat, the quantitative statements about dip control of surface displacement are tied to a tuned model rather than to the published TPV13 benchmark.","section":"Section 3 and Table S3"}],"minor_comments":[{"comment":"The text reports a slight over-prediction of accumulated slip of 0.3 m/s at t = 1 s; accumulated slip has units of meters, so the stated units appear to be a typo and should be corrected.","section":"Section 3.3"},{"comment":"The text states that leave-one-out cross-validation errors are low for theta in [52,68] degrees, while Table 3 defines the interior as theta_int = [53,57] degrees; please reconcile these ranges.","section":"Section 5.1 and Table 3"},{"comment":"The number of retained POD modes is never reported; since this is a tunable parameter of the ROM construction, please specify it for each ROM and ideally show the sensitivity of the cross-validation error to the truncation rank.","section":"Sections 5.1 and 5.2"},{"comment":"The phrase 'Between 0 and 150 km depths' should read 'between 0 and 150 km depth', and the sentence reporting a mean variability of 3 K should explicitly state that this applies below 150 km depth.","section":"Section 4.4.4"},{"comment":"The caption does not identify the color or line mapping for the different alpha values; please add a legend or describe the mapping in the caption.","section":"Figure 12"},{"comment":"In the definition of discretization, 'dented by triangle' appears to be a typographical error for 'denoted by triangle'.","section":"Section 2.1"},{"comment":"The mesh quality metrics for theta = 70 and theta = 80 are identical in the table; if this reflects symmetry or rounding, please state this explicitly.","section":"Table 2"},{"comment":"The text refers to 'vertical ground velocity u_z' while the ROM is for vertical displacement at the final time step; please use consistent terminology throughout the section.","section":"Section 5.2"}],"recommendation":"major_revision","confidential_remarks":"The missing exact-mesh thermal verification for the Slab2 example is the main substantive gap and is straightforward to address with the machinery already used in Section 4.3.4. The modified TPV13 parameters are disclosed, but their implications for the sensitivity analysis should be more prominently framed. The method itself and the dynamic rupture verification are strong, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Gabrielle and coauthors have written a careful methods paper. The genuinely new thing is coupling RBF mesh morphing with POD-based reduced-order models for dynamic rupture; to my knowledge that combination hasn't been done before. The verification effort is real: they compare morphed-mesh accumulated slip, peak slip rate, receiver velocities, and uplift against exactly generated meshes, and report RMS velocity errors around 1e-3 m/s. The ROM validation with leave-one-out plus held-out dips (50.5 to 69.5) is credible, with a maximum surface displacement error of 0.007 m. The code and data are on Zenodo, so the reproducibility claim is checkable.\n\nThe soft spots are specific and fixable. Most important: the Slab2 uncertainty example never runs the thermal model on the exact meshes. The abstract and conclusions claim that in both applications morphed meshes 'closely match' exactly generated meshes, but for Slab2 only the geometry is compared. The morphing error is up to 5–5.5 km in the upper 10–20 km of the mesh, exactly where the reported temperature variability is largest (mean 22 K above 150 km, up to 40 K). A 5 km offset in slab depth can plausibly shift forearc temperatures by tens of K, so part of that 40 K 'uncertainty' could be morphing error. This is directly testable: they already have the exact meshes; run the thermal model on them and compare. The earlier curvature validation does not transfer cleanly because the Slab2 case uses a younger plate age (8 Myr), a thinner slab, and a different morphing parameterization.\n\nSecond, the speedup claims are overstated by about two orders of magnitude if read as per-simulation: 10^9x and 10^8x, versus roughly 10^7 and 10^6 for a single full-order run. The 10^9x number only holds if you divide by the total cost of all 21 simulations, which is not how the abstract phrases it. Third, the modified TPV13 parameters (static friction 0.54 to 0.48, bulk cohesion 5e6 to 1e6 Pa) are disclosed, but the paper does not test whether the dip sensitivity or the ROM conclusions hold under the original benchmark values. That is a minor but relevant caveat. Finally, the POD truncation rank is never stated, which is an easy reproducibility fix.\n\nNone of these are fatal. The central method is sound, the rupture verification is thorough, and the limitations section is unusually candid. This paper deserves a serious referee and, after a revision that closes the Slab2 validation gap and corrects the speedup arithmetic, should be accepted. I would send it to G3 or a comparable journal.","headline":"A careful, honest methods paper that makes geometric UQ practical for rupture and subduction thermal models; needs a revision to fix the Slab2 validation gap and the speedup arithmetic.","tokens_in":34956,"tokens_out":4045,"would_cite":true,"duration_ms":37131,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Deforming one mesh into many fault geometries produces simulation outputs accurate enough to train fast surrogate models for geometric uncertainty.","keywords":["mesh morphing","reduced-order models","dynamic rupture","subduction thermal structure","sensitivity analysis","uncertainty quantification","radial basis functions","fault dip"],"falsifier":"Re-run the dip study on the unmodified TPV13 benchmark: if rupture fails to nucleate for dips outside 55–63 degrees, or if a ROM trained on modified-benchmark runs fails to match unmodified-benchmark surface displacements within the reported 0.007 m error, the dip-sensitivity and ROM conclusions hold only for the tuned setup.","tokens_in":2070,"feed_emoji":"🌋","tokens_out":2092,"duration_ms":77967,"temperature":0.7,"pith_summary":"This paper seeks to establish that RBF-based mesh morphing, which deforms a single reference mesh into many geometric configurations while preserving mesh connectivity, is a practical and general tool for quantifying geometric uncertainty in computational geophysics. The authors show that morphed meshes for 3D dynamic earthquake rupture (fault dip from 40 to 80 degrees) and 2D subduction thermal models (slab curvature and Slab2 depth uncertainty) retain acceptable quality, and that simulations on them closely match results from exactly generated meshes. Because connectivity is preserved, the outputs can train data-driven reduced-order models, which reproduce full-model behavior with errors of roughly 1e-3 m/s in receiver velocities and at most 0.007 m in held-out surface displacement, at speedups up to 1e9 times. If true, this makes geometric sensitivity analysis and uncertainty quantification feasible for problems that were previously limited by manual remeshing and computational cost.","feed_headline":"Deformed meshes reproduce earthquake and subduction simulations","feed_subtitle":"Surrogate models trained on the morphed meshes answer in 1e-4 seconds, up to 1e9 times faster than full runs.","key_machinery":"The central object is the RBF mesh-morphing interpolant: displacements prescribed along boundary and interface curves or surfaces are interpolated by radial basis functions (linear kernel augmented with a polynomial basis), then evaluated at every mesh vertex to displace the whole mesh without changing connectivity. Successive morphing steps enforce geometric constraints such as fault planarity, uniform slab width, and straight subdomain boundaries. The preserved connectivity is what makes the output suitable for data-driven ROMs built from proper orthogonal decomposition, allowing snapshots from different geometries to share a common representation.","core_discovery":"The central claim is that mesh morphing preserving connectivity yields simulation outputs suitable both as accurate replacements for exact-mesh runs and as training data for non-intrusive reduced-order models. For the TPV13-3D dynamic rupture benchmark, morphed faults with dip between 40 and 80 degrees lie within 140–150 m of exactly meshed faults, which is comparable to the local mesh resolution, and simulated receiver velocities agree to roughly 1e-3 m/s RMS. A correction step keeps the morphed fault planar to within 1 m. Using interpolated proper orthogonal decomposition with a quintic RBF interpolant on POD coefficients, a surface-displacement ROM evaluated at held-out dips between 50.5 and 69.5 degrees reaches a maximum L-infinity error of 0.007 m. For subduction thermal models, morphing slab curvature over the global range changes slab-interface temperature by up to 85 K, while morphing within Slab2 depth uncertainty changes it by up to 40 K.","pith_inferences":["A natural extension not demonstrated in the paper is simultaneous morphing of multiple geometric parameters (e.g., dip and strike, or curvature and depth), which would allow higher-dimensional geometric uncertainty quantification with the same connectivity-preserving framework.","Because the reordering permutation fixes the local-time-stepping output ordering, the methodology likely carries over to other solvers that permute output, enabling ROM construction without modifying the forward code.","The reported temperature sensitivities could be propagated into derived quantities such as dehydration depths or the thermally controlled seismogenic zone limits, which the paper leaves implicit.","A useful cross-check would be to compare morphed-mesh dip sensitivities against an independently generated set of meshes without connectivity constraints at matched resolution, to isolate geometric accuracy from discretization error common to both morphed and exactly generated meshes."],"forward_implications":["Geometric uncertainty quantification becomes practical for dynamic rupture and subduction thermal models, as ensembles of morphed meshes can be generated in tens of seconds on a laptop without manual remeshing.","Reduced-order models trained on morphed-mesh outputs can replace full simulations for rapid sensitivity analysis, with evaluations in about 1e-4 seconds and speedups of 1e8 to 1e9 relative to forward models.","Fault dip is shown to be a first-order control on surface ground motion, with vertical displacement varying by up to 2.9 m across the 50–70 degree dip range at the final simulation time.","Slab interface curvature across the global range changes slab-interface temperatures by up to 85 K, and Slab2 depth uncertainty translates to at most 40 K of temperature uncertainty, with a mean of 22 K above 150 km depth.","The general, solver-agnostic nature of the approach suggests it can extend to other mesh-based geophysical simulations and, with careful treatment of intersections, to multi-fault systems."],"supporting_citations":[{"why":"Supplies the RBF mesh morphing technique and the polynomial-augmented interpolation formalism on which Algorithms 1–4 are based.","marker":"Sieger et al., 2014"},{"why":"Established RBF-based mesh deformation for engineering meshes, providing the methodological foundation adapted here.","marker":"de Boer et al., 2007"},{"why":"Defines the TPV13-3D dynamic rupture benchmark whose geometry, stress, and friction parameters the rupture example morphs and modifies.","marker":"Harris et al., 2018"},{"why":"Provides the Slab2 subduction zone geometry model and its depth uncertainty fields, which define the beta morphing example.","marker":"Hayes et al., 2018"},{"why":"Shows that parabolic profiles y=-alpha x^2 fit Slab2 geometries within depth uncertainty, supplying the alpha parameterization used in the curvature example.","marker":"England and May, 2021"},{"why":"The kinematic-dynamic thermal model, including equations, mesh refinement, and parameter values, is the model run on morphed subduction meshes.","marker":"Hobson and May, 2025"},{"why":"Introduces the interpolated proper orthogonal decomposition (iPOD) approach used to build the receiver-velocity and surface-displacement ROMs.","marker":"Ly & Tran, 2001"},{"why":"GMSH generates the reference meshes and the exactly generated comparison meshes that establish accuracy baselines for morphed meshes.","marker":"Geuzaine and Remacle, 2009"}],"fun_headline_variants":["Morphing meshes powers 1e9x faster quake and subduction surrogates","Mesh morphing yields surrogates with 1e9x speedup for geophysics","Deformed meshes train ROMs: 1e9x faster, 1e-3 m accurate","Morphing fault geometry enables 1e9x speedup in rupture simulations","Mesh morphing: accurate surrogates for fault dip and slab shape"],"cache_read_input_tokens":36992,"weakest_assumption_plain":"The dip-sensitivity and ROM results depend on modifying the TPV13 benchmark by lowering the nucleation static friction from 0.54 to 0.48 and the bulk cohesion from 5e6 to 1e6 Pa, because under the unmodified benchmark rupture nucleates only for dips in 55–63 degrees.","fun_headline_variants_meta":{"raw":{"variants":["Morphing meshes powers 1e9x faster quake and subduction surrogates","Mesh morphing yields surrogates with 1e9x speedup for geophysics","Deformed meshes train ROMs: 1e9x faster, 1e-3 m accurate","Morphing fault geometry enables 1e9x speedup in rupture simulations","Mesh morphing: accurate surrogates for fault dip and slab shape"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000234,"raw_usage":{"total_tokens":1552,"prompt_tokens":1058,"completion_tokens":494,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":674,"completion_tokens_details":{"reasoning_tokens":378}},"tokens_in":674,"tokens_out":494,"duration_ms":4613,"temperature":1.0,"reasoning_tokens":378,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:30:29.072329+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the dip study on the unmodified TPV13 benchmark: if rupture fails to nucleate for dips outside 55–63 degrees, or if a ROM trained on modified-benchmark runs fails to match unmodified-benchmark surface displacements within the reported 0.007 m error, the dip-sensitivity and ROM conclusions hold only for the tuned setup.","supporting_citations":[{"cited_title":", Menzel, S","cited_arxiv_id":null,"evidence_quote":"Supplies the RBF mesh morphing technique and the polynomial-augmented interpolation formalism on which Algorithms 1–4 are based."},{"cited_title":", Barall, M","cited_arxiv_id":null,"evidence_quote":"Defines the TPV13-3D dynamic rupture benchmark whose geometry, stress, and friction parameters the rupture example morphs and modifies."},{"cited_title":"\\ May, D","cited_arxiv_id":null,"evidence_quote":"Shows that parabolic profiles y=-alpha x^2 fit Slab2 geometries within depth uncertainty, supplying the alpha parameterization used in the curvature example."},{"cited_title":"\\ Tran, H T","cited_arxiv_id":null,"evidence_quote":"Introduces the interpolated proper orthogonal decomposition (iPOD) approach used to build the receiver-velocity and surface-displacement ROMs."}],"review_version":1}