{"id":"fade12c2-d7af-4ada-8d0e-ff9da44c26c9","arxiv_id":"2506.09435","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A Firedrake-based spectral element model for fully nonlinear potential flow water waves is verified through convergence and scaling tests and validated against experimental and analytical benchmarks.","lead":"This paper introduces FNPF-SEM, a parallel wave simulator for linear and fully nonlinear potential flow, built on the Firedrake framework. Tests show expected numerical convergence, good parallel scaling, and visual agreement with experiments and analytical solutions for wave-structure interaction.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Strong/weak scaling results may measure only the Laplace solve, not the full model time step, overstating parallel efficiency.","rationale":"The reader correctly identified the stagnation of EvaluateRHS and LaplaceUpdate at high core counts as a threat to the scalability claim. My reading sharpens this: the scaling sections appear to measure only the Laplace solve, per the explicit sentence in Section 4.1.2, so the reported efficiencies may not represent the full model at all. This is a load-bearing concern for the headline claim of 'good parallel efficiency.' It does not overturn the convergence verification or the qualitative validations, but it does mean the efficiency claim should be treated as conditional on showing full-step scaling. The reader's verdict of CONDITIONAL remains appropriate; my concern reinforces it rather than changing the verdict. I give partial agreement because the reader did not note the 'focus exclusively on this part of the code' statement, which is the strongest evidence that the scaling tests are Laplace-only. The proposed concrete test is straightforward and would settle the ambiguity without requiring new algorithm development.","tokens_in":16403,"tokens_out":4229,"duration_ms":46795,"concrete_test":"Re-run the strong-scaling cases in Table 3 (Mesh 1/2/3, p = 3/4/5) and the weak-scaling cases in Table 4 (p = 5) while timing the entire time-integration step (Steps 4.1–4.7), not just LaplaceSolve. Compute gamma_s and gamma_w from these full-step timings and compare with the reported values. If the full-step efficiency at Nc = 300 is substantially lower than the Laplace-only values, the scalability claim overstates practical parallel performance.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.1.2 profiles LaplaceSolve, EvaluateRHS, and LaplaceUpdate, showing that LaplaceSolve continues to scale while the other two stagnate above roughly 20 cores. The authors then state: 'Consequently, the following strong- and weak-scaling analyses focus exclusively on this part of the code.' Read literally, the strong-scaling efficiencies in Figures 6–7 and the weak-scaling efficiencies in Figures 8–9 are computed from LaplaceSolve timings only, not from the full time step (Steps 4.1–4.7 of the workflow). If so, the reported 'good parallel efficiency' characterizes only the Laplace component. The full model's efficiency is bounded by Amdahl's law: the constant-cost EvaluateRHS and LaplaceUpdate routines, whose relative share grows with core count (as the paper's own Figure 5 shows), will make the end-to-end speedup poorer than the Laplace-only curves suggest. The central claim that the solver scales efficiently to 300 cores for nonlinear wave simulations would therefore be overstated. Even if the authors intended the scaling timings to include the full step, the text is at minimum ambiguous on this point, and the omission of any full-step scaling data leaves the claim unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"FNPF-SEM is a new parallel spectral element solver implemented within the Firedrake framework for linear and fully nonlinear potential-flow water waves, including wave-structure interaction. The mathematical model is the classical Zakharov-form free-surface problem: a Laplace problem in the volume plus kinematic and dynamic free-surface evolution equations, discretized with continuous nodal Lagrange elements on vertically extruded, horizontally unstructured meshes, with explicit RK4 time integration, relaxation zones for generation and absorption, and a modal filter for stabilization. The paper verifies the implementation through h-convergence (O(h^p) for the recovered vertical velocity) and p-convergence for nonlinear periodic waves over a range of kh and steepnesses, profiles the computational routines, and presents strong and weak scaling tests up to 300 cores. Validation cases are the Beji-Battjes submerged bar, linear diffraction and nonlinear run-up on a vertical cylinder, and diffraction by a V-shaped breakwater, compared with analytical, experimental, and reference numerical data. The strong and weak scaling tests are stated to focus exclusively on the Laplace solve. The authors conclude that the framework is accurate, efficient, and applicable to offshore engineering.","tokens_in":16646,"tokens_out":9149,"duration_ms":97062,"significance":"If the accuracy and scalability claims are fully supported, the Firedrake-based implementation is a useful engineering contribution: it brings high-order spectral element discretizations of FNPF equations to an open-source, MPI-capable framework with unstructured mesh support, and it provides a systematic benchmark suite covering convergence, scaling, and validation. The verification results are credible: the observed algebraic convergence order is consistent with the expected gradient-recovery loss, and the p-convergence in Appendix A shows exponential behavior. The validation benchmarks are well chosen and independent, with no fitted parameters. The paper also gives a clear breakdown of the computational routines, which is helpful for future optimization. If the requested evidence on end-to-end scaling and quantitative validation errors is supplied, the paper would support the advertised claims of accuracy, efficiency, and applicability.","major_comments":[{"comment":"The scaling analyses measure only the Laplace solve, not the full time step. The text states: \"Consequently, the following strong- and weak-scaling analyses focus exclusively on this part of the code.\" Figures 6-9 therefore report LaplaceSolve-only timings and efficiencies, while the abstract and conclusion claim that the model \"is scalable to full- and large-scale simulations\" and exhibits \"good parallel efficiency.\" This is load-bearing. Figure 4 shows that EvaluateRHS and LaplaceUpdate have constant runtimes beyond about 20 cores, and Figure 5 shows that the LaplaceSolve share of total time decreases as the core count grows. By Amdahl's law, the full-step speedups are strictly smaller than the Laplace-only curves, with an increasing gap at higher core counts. Please report strong and weak scaling for the complete time step (Steps 4.1-4.7), or alternatively provide Amdahl-corrected efficiency estimates using the measured cost shares.","section":"§4.1.2, Figs. 4-9, Eqs. (6)-(7)"},{"comment":"The validation is qualitative: the paper reports \"good visual agreement\" (bar test), \"good agreement\" (cylinder), and \"good qualitative agreement\" (breakwater), but no quantitative error metrics are given. The convergence tests in §4.1.1 establish discretization error only for periodic nonlinear waves, not for the inhomogeneous, structure-interaction cases in §4.2. To support the conclusion in §5 that the framework is accurate for wave-structure interaction, please add error measures such as RMS or L2 differences between numerical and experimental harmonic amplitudes in the bar test, relative errors in η_m/(2H) against MacCamy-Fuchs (TC1) and Kriebel (TC2), and profile errors against the Chang et al. reference for the breakwater. A mesh-resolution study for at least one of the structure cases is also needed to separate discretization error from physical modeling error; this is particularly important for the cylinder case, where the boundary is represented by affine geometry (§4.2.2).","section":"§4.2.1-4.2.3"},{"comment":"The modal filter is used for stabilization, but its effect on the solution is not quantified. The filter is taken from [23] and applied \"as needed\" to free-surface quantities, yet the paper does not report the filter order or strength used in the verification and validation runs, nor does it show that the filtered results are insensitive to those parameters. Since the filter modifies η, φη, or wη, the accuracy of the steep nonlinear results (e.g., ε/ε_max = 90% in Figure 3 and TC2 in Figure 12) may reflect the filter's influence as much as the spatial discretization. Please state the filter parameters and include a sensitivity study, e.g., varying the filter strength at fixed mesh resolution or comparing against an unfiltered run on a case that is stable without filtering.","section":"§3.2"}],"minor_comments":[{"comment":"The periodic boundary condition \"ϕ|Γp = ϕ|Γp\" is trivially true as written; please use notation distinguishing the two periodic faces, e.g., ϕ|Γp+ = ϕ|Γp−.","section":"§2.1, Eq. (1)"},{"comment":"The CFL time step uses u_max, \"the maximum velocity encountered in the simulation,\" but no a priori estimation procedure is given; for practical use, clarify how u_max is determined.","section":"§3.2"},{"comment":"The sentence \"A substantial reduction in time per stage is observed; the increase is from Nc = 1 to approximately Nc = 20\" is garbled; the second clause should refer to the reduction across this core range. Also, \"is has been\" later in the same section should read \"it has been.\"","section":"§4.1.2"},{"comment":"The figure caption should explicitly identify which line corresponds to Mesh 1 and which to Mesh 2; the text does this, but the figure itself is ambiguous, especially in black-and-white printing.","section":"Figure 5"},{"comment":"The inequality \"−L2 − 2b ≤ (x, y) ≤ 2b + L2\" is not meaningful for a vector; please write it componentwise. Also, the definition of L2 as \"260 [m] ≈ 2L\" should be made clearer, since L2 denotes a length rather than a coordinate variable.","section":"§4.2.3"},{"comment":"The title page reads \"AN AR XIV PREPRINT\"; the space in \"AR XIV\" is erroneous and should read \"AN ARXIV PREPRINT.\"","section":"Title page"}],"recommendation":"major_revision","confidential_remarks":"The verification part of the manuscript is solid, and the benchmark choices are appropriate. The main reservations are the Laplace-only scaling analysis and the lack of quantitative validation metrics, both of which can be addressed within the manuscript's scope without changing the numerical method. The novelty is primarily the Firedrake implementation of an existing DTU SEM methodology, and the authors are transparent about that lineage. I do not see grounds for rejection, but the advertised scalability and accuracy claims need the additional evidence described in the major comments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this is a credible verification and validation paper for a known method, and the implementation work is real. But the headline parallel-efficiency numbers are computed from the Laplace solve only, not the full time step, so the claim that the solver scales to 300 cores is not supported by the data presented.\n\nWhat's genuinely new: the Firedrake implementation of the DTU spectral element FNPF method, along with h/p convergence studies and validation against the Beji-Battjes bar, MacCamy-Fuchs diffraction, Kriebel run-up, and Chang-breakwater data. The verification against stream-function waves is a solid check, and the free parameters are standard (CFL 0.95, relaxation zones, ramp-up). No fitted parameters. That is a useful public engineering tool—if the code were released.\n\nNow the soft spots. The scaling section is the biggest one. Section 4.1.2 profiles three routines and shows that EvaluateRHS and LaplaceUpdate flatten out above ~20 cores. The text then says the scaling analyses 'focus exclusively on this part of the code'—i.e., LaplaceSolve. So Figures 6-9 report speedup/efficiency for the Laplace solve, not the full time step. The paper's own Figure 5 shows that the non-scaling routines become a larger share of the total at higher core counts, so end-to-end speedup will be worse than those curves suggest. The conclusion says 'good parallel efficiency, particularly for large-scale simulations' without clarifying that this is a component measure. That's an overstated central claim, and it needs either full-step scaling data or a careful qualification.\n\nOther things: the validation cases are visual agreement, no error metrics; the cylinder boundary uses affine geometry, which limits the high-order claims at the structure; the modal filter is imported from [23] without sensitivity analysis. The code is not released, so the 'open-source framework' point only applies to Firedrake, not to this model.\n\nFor the right reader—someone wanting a verified FNPF-SEM baseline in Firedrake or a comparison point for their own wave tank—this is worth a serious referee. The numerical core is sound, and the flaws are in presentation and missing data, not in the math. I'd send it out, with a request for full-step scaling and code availability.","headline":"Credible verification of a Firedrake-based spectral element FNPF solver, but the scaling claims measure only the Laplace solve and overstate end-to-end parallel performance.","tokens_in":17112,"tokens_out":2568,"would_cite":false,"duration_ms":26643,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M60","65M70","76B15","76B07","65Y05"],"pacs":[],"model":"deepseek-v4-flash","headline":"A parallel spectral element solver for fully nonlinear potential-flow water waves is verified by convergence studies, scaling tests, and wave-structure benchmarks.","keywords":["spectral element method","fully nonlinear potential flow","water waves","wave-structure interaction","high-order finite elements","parallel computing","free surface","Firedrake"],"falsifier":"Run a full time-step strong-scaling experiment, including mesh update and free-surface evaluation, on a small free-surface-dominated mesh and compare total per-stage runtime with the Laplace-only scaling curve; if total efficiency drops far below the Laplace-only curve beyond 20 cores, the practical parallel performance of the whole solver is weaker than the Laplace-only numbers suggest.","tokens_in":16232,"feed_emoji":"🌊","tokens_out":9879,"duration_ms":99666,"temperature":0.7,"pith_summary":"This paper presents a parallel, high-order spectral element solver for linear and fully nonlinear potential-flow water waves, built on the Firedrake finite element framework with native MPI parallelism. The authors aim to establish that the solver, FNPF-SEM, can serve as a general-purpose wave model for offshore engineering: unstructured meshes represent complex geometry, high-order polynomial bases keep numerical diffusion and dispersion low, and the parallel implementation targets large-scale, long-time simulations. Verification shows algebraic h-convergence at the expected rate and spectral p-convergence, while strong and weak scaling tests show good parallel efficiency, especially on large problems where the Laplace solve dominates. Validation against analytical solutions and experimental data for harmonic generation over a submerged bar, wave run-up on a vertical cylinder, and V-shaped breakwater diffraction supports the paper's conclusion that the framework is accurate and applicable.","feed_headline":"Nonlinear wave solver scales to 300 cores, passes benchmarks","feed_subtitle":"High-order spectral elements reproduce experimental wave run-up and harmonic generation at scale.","key_machinery":"The load-bearing object is the Zakharov-form fully nonlinear potential flow model: the Laplace equation $\\nabla^2\\phi = 0$ in the fluid with Dirichlet data $\\phi_\\eta$ on the free surface and no-flow conditions on bottom and walls, coupled to evolution equations for the surface elevation $\\eta$ and surface potential $\\phi_\\eta$. A mapping that extracts free-surface values from the volume mesh and a vertical mesh update tie the surface and volume discretizations together, and the Laplace solve is the dominant cost that carries the scalability analysis. Stabilization comes from vertically aligned structured elements, over-integration of the nonlinear free-surface terms, and a modal filter applied to the highest modes of free-surface quantities.","core_discovery":"The central claim is that FNPF-SEM reproduces the expected high-order accuracy of spectral element discretizations of the Zakharov free-surface potential-flow equations while remaining parallelizable. The Laplace problem for the velocity potential is solved on an extruded, vertically moving prismatic mesh with nodal Lagrange basis functions of order p, and the free-surface evolution equations for $\\eta$ and $\\phi_\\eta$ are integrated with explicit fourth-order Runge-Kutta, with over-integration and a mild modal filter to control aliasing instabilities. The paper reports $O(h^p)$ convergence under h-refinement, exponential convergence under p-refinement, near-ideal strong scaling for larger meshes and higher polynomial order, near-constant per-core runtime in weak scaling up to 300 cores, and agreement with the chosen analytical and experimental benchmarks. The authors conclude that these results confirm the accuracy and applicability of the proposed framework.","pith_inferences":["The profiling suggests a practical ceiling for small or free-surface-dominated problems: the full time step's parallel efficiency may saturate near 20 cores because the free-surface evaluation and mesh update routines stop scaling, even though the Laplace-only scaling curves look better.","Since the cylinder boundary is currently approximated with affine geometry and submerged or floating bodies are not yet supported, a body-fitted high-order curvilinear boundary treatment is the natural next step for accurate near-structure predictions.","The solver's reliance on a code-generation finite element environment means performance portability, for example to many-core or GPU architectures, could come from swapping solver backends rather than rewriting the model equations.","All validated cases are non-breaking waves of moderate steepness; testing near the breaking limit would stress the stabilization and filtering machinery more severely than the presented benchmarks do."],"forward_implications":["Users can choose either mesh refinement (h) or polynomial order increase (p), with predictable algebraic or spectral error reduction.","On large problems the model maintains near-ideal parallel efficiency up to the tested 300 cores, making long-time simulations of engineering-scale domains practical.","Because the Laplace solve is the dominant cost, further improvements to its solver or preconditioner translate almost directly into whole-model speedups.","The validated benchmarks support the use of the solver for predicting wave run-up, harmonic generation, and diffraction around offshore structures.","The unstructured horizontal meshing opens the path to regional-scale wave propagation and irregular coastlines, which the paper names as ongoing work."],"supporting_citations":[{"why":"Supplies the stabilized nodal spectral element method, including vertical element alignment, over-integration, and the modal filter used to control aliasing instabilities, and the expected O(h^p) convergence rate.","marker":"[23]"},{"why":"Documents the Firedrake framework that provides finite element assembly, MPI parallelism, and extruded mesh support.","marker":"[31]"},{"why":"Describes the automated finite element composition and solver infrastructure used for the Laplace solve.","marker":"[51]"},{"why":"Provides the experimental bar data used to validate high-order harmonic generation over a submerged bar.","marker":"[9]"},{"why":"Provides the analytical linear diffraction solution used to validate the linear vertical cylinder run-up case.","marker":"[45]"},{"why":"Provides the experimental run-up data used to validate the nonlinear vertical cylinder case.","marker":"[39]"},{"why":"Provides the reference solution for wave diffraction around a V-shaped breakwater.","marker":"[16]"},{"why":"Gives the Zakharov free-surface formulation on which the evolution equations are based.","marker":"[63]"},{"why":"Supplies the relaxation zone method used for wave generation and absorption.","marker":"[41]"},{"why":"Provides the stream function theory used to initialize nonlinear wave conditions.","marker":"[28]"}],"fun_headline_variants":["FNPF-SEM: spectral element waves scale to 300 cores","Nonlinear wave solver: high-order accuracy, parallel scaling","Firedrake-based wave model passes benchmarks at scale","Spectral element wave simulations: p-convergence, 300 cores","High-order wave-structure solver validated, parallel"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The scalability claims rest on the Laplace solve remaining the dominant routine, while the paper's own profiling shows the two other main routines stagnate above about 20 cores.","fun_headline_variants_meta":{"raw":{"variants":["FNPF-SEM: spectral element waves scale to 300 cores","Nonlinear wave solver: high-order accuracy, parallel scaling","Firedrake-based wave model passes benchmarks at scale","Spectral element wave simulations: p-convergence, 300 cores","High-order wave-structure solver validated, parallel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001026,"raw_usage":{"total_tokens":4336,"prompt_tokens":964,"completion_tokens":3372,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":580,"completion_tokens_details":{"reasoning_tokens":3288}},"tokens_in":580,"tokens_out":3372,"duration_ms":28638,"temperature":1.0,"reasoning_tokens":3288,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:47:38.041618+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a full time-step strong-scaling experiment, including mesh update and free-surface evaluation, on a small free-surface-dominated mesh and compare total per-stage runtime with the Laplace-only scaling curve; if total efficiency drops far below the Laplace-only curve beyond 20 cores, the practical parallel performance of the whole solver is weaker than the Laplace-only numbers suggest.","supporting_citations":[{"cited_title":"Journal of Computational Physics 318: 1–21","cited_arxiv_id":null,"evidence_quote":"Supplies the stabilized nodal spectral element method, including vertical element alignment, over-integration, and the modal filter used to control aliasing instabilities, and the expected O(h^p) convergence rate."},{"cited_title":"Imperial College London and University of Oxford and Baylor University and University of Washington, first edition edition","cited_arxiv_id":null,"evidence_quote":"Documents the Firedrake framework that provides finite element assembly, MPI parallelism, and extruded mesh support."},{"cited_title":"ACM Transactions on Mathematical Software 43(3)","cited_arxiv_id":null,"evidence_quote":"Describes the automated finite element composition and solver infrastructure used for the Laplace solve."},{"cited_title":"Coastal Engineering 23(1): 1–16","cited_arxiv_id":null,"evidence_quote":"Provides the experimental bar data used to validate high-order harmonic generation over a submerged bar."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the analytical linear diffraction solution used to validate the linear vertical cylinder run-up case."},{"cited_title":"part ii: Wave run-up","cited_arxiv_id":null,"evidence_quote":"Provides the experimental run-up data used to validate the nonlinear vertical cylinder case."},{"cited_title":"Coastal Engineering 68: 56–66","cited_arxiv_id":null,"evidence_quote":"Provides the reference solution for wave diffraction around a V-shaped breakwater."},{"cited_title":"Journal of Applied Mechanics and Technical Physics 9: 190–194","cited_arxiv_id":null,"evidence_quote":"Gives the Zakharov free-surface formulation on which the evolution equations are based."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the relaxation zone method used for wave generation and absorption."},{"cited_title":"Computers & Geosciences 14(3): 357–368","cited_arxiv_id":null,"evidence_quote":"Provides the stream function theory used to initialize nonlinear wave conditions."}],"review_version":1}