{"id":"64e8191e-7793-4fcd-99ef-e04dee2e4fa1","arxiv_id":"2608.05768","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"ROD boundary corrections are recast as per-quadrature-point polynomial corrections that avoid matrix inversion and recover high-order accuracy on linear meshes for curved domains.","lead":"The authors simplify a class of boundary-condition corrections for curved domains so that no small linear system needs to be inverted at each boundary point. They demonstrate that the cheaper corrections achieve high-order convergence in finite volume and discontinuous Galerkin schemes on fixed and moving meshes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The multi-dimensional pointwise consistency of the ROD-E/ROD-L2 corrections (Sec. 4.3) is asserted without proof, and the DG-P2 slip-wall results in Table 2 show orders below the design order, so the central claim is not fully secured.","rationale":"The reader correctly identifies the unproved multi-D consistency assertion as the weakest link. My analysis agrees: the algebraic derivations in Propositions 2 and 3 are correct for a single point, but the extension to a full boundary integral requires a consistency estimate that is not supplied. The Taylor argument sketched in the attack shows the claim is plausible under the conformal-chord hypothesis, but the paper never states or verifies the needed error estimates. The strongest direct evidence of a soft spot is the DG-P2 slip-wall convergence anomaly in Table 2, where the finest-grid orders are 2.46 and 2.76 rather than the design order 3. This is exactly the sort of signature one would expect if the pointwise correction produces a consistency defect that is not always O(h^{p+1}). I do not recommend changing the reader's conditional verdict: the numerical evidence is otherwise broad (fixed/moving, 2D/3D, Dirichlet and slip-wall, up to fifth order), and the concern is testable rather than demonstratively fatal. The condition on the verdict is precisely what the reader stated: either a consistency proof or a satisfactory explanation of the P2 slip-wall behavior. My concrete test would settle whether the anomaly is a genuine order reduction or a resolution artifact, and the flux-residual computation would provide a direct verification of the unproved consistency claim.","tokens_in":27033,"tokens_out":11854,"duration_ms":116265,"concrete_test":"Re-run the Table 2 DG-P2 slip-wall vortex on at least two finer meshes (extending the refinement sequence, e.g., h = 4.7e-3 and 2.35e-3), and separately compute the boundary flux consistency residual for a single curved face with a manufactured solution: on a sequence of meshes, evaluate max_face |∫_face ψ [F(v_h, n_comp) - F(u(x̃), n_comp)] dS|. If the P2 slip-wall order does not approach 3 and the residual decays slower than h^{p+1}, the multi-D pointwise-consistency claim in Sec. 4.3 fails; if both recover, the anomaly is a resolution or implementation artifact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that applying the 1D single-constraint correction independently at each boundary quadrature point preserves the design order in multi-D, 'while always preserving high-order consistency' (Sec. 4.3, also Sec. 1). This is not proven: Propositions 2 and 3 are algebraic identities for one point; the step from a pointwise identity to a global consistency estimate for the boundary flux is missing. A Taylor argument can be given: if the interior polynomial u_h is O(h^{p+1}) accurate at x̃ and the offset δ = ||x̃ - x̄|| = O(h²) for conformal chord faces, then v_h(x̃) = u_h(x̃) + α(u_D(x̄) - u_h(x̄)) is O(h^{p+1}) close to u(x̃), so the boundary flux residual is O(h^{p+1}) provided α - 1 = O(δ) and the ghost state is built from the corrected normal-velocity component for slip walls. But this argument requires several unstated hypotheses and does not appear in the paper. The numerical evidence is strong but not uniform: in Table 2 (DG-P2 slip wall), the finest-grid orders for ROD-E and ROD-L2 are 2.46 and 2.76, below the design order 3, and ROD-E and ROD-L2 differ on the intermediate mesh (1.43e-5 vs 1.76e-5) while they nearly coincide elsewhere. No explanation is given. If this anomaly persists under refinement, the assertion 'always preserving high-order consistency' is false for slip-wall P2, and the central claim narrows.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes two boundary correction methods, ROD-E and ROD-L2, which recast the Reconstruction for Off-site Data (ROD) approach as pointwise polynomial corrections. In place of solving a K x K constraint system per boundary cell, the methods evaluate formulas (36) and (41) independently at each boundary quadrature point, using one scalar parameter alpha to enforce the boundary condition at the corresponding physical boundary point. The corrections are developed for a Runge-Kutta discontinuous Galerkin framework on fixed curved domains and an ADER arbitrary-Lagrangian-Eulerian finite volume framework on moving curved domains. Convergence studies are presented for manufactured solutions, a supersonic vortex with slip walls, the Kidder problem, and an oscillating cylinder, with claimed accuracy up to fifth order in two and three dimensions.","tokens_in":27337,"tokens_out":4019,"duration_ms":38228,"significance":"If the central claim is valid, the paper provides a practically useful simplification: it removes the need to assemble and invert the ROD constraint matrix in every boundary cell at every time step, and it makes the correction easy to port across DG, FV, and ADER codes. The algebraic derivations in Propositions 2 and 3 are clean and self-contained, and the numerical campaign is broad, covering two spatial dimensions, three spatial dimensions, fixed and moving meshes, Dirichlet and slip-wall conditions, and both DG and FV/ADER frameworks. The main weakness is that the extension from the single-point algebraic identities to a global multi-dimensional consistency statement is asserted rather than proved, and several convergence tables contain order reductions that are not addressed.","major_comments":[{"comment":"The paper's central claim, stated in the Introduction and again in Section 4.3, is that applying the one-dimensional correction independently at each boundary quadrature point preserves high-order consistency in multiple dimensions. Propositions 2 and 3, however, prove only pointwise algebraic identities for a single quadrature point: they show how to express v_h(xtilde) in terms of u_hat and u_D(bar x). The manuscript does not provide any estimate for the resulting boundary flux error, nor an analysis of how the correction interacts with the quadrature rule, the mapping M, and the DG/FV weak forms in equations (6) and (22). A Taylor-based consistency argument would require, among other things, a bound on alpha - 1, a comparison of u_h(xtilde) with u_h(bar x), and regularity assumptions on the boundary. As written, the statement 'while always preserving high-order consistency' is an assertion rather than a proved theorem, and this is load-bearing for the paper's main contribution.","section":"Section 4.3, Remark 4"},{"comment":"In the supersonic vortex slip-wall test, the finest-grid rho orders for DG-P2 are 2.46 for ROD-E and 2.76 for ROD-L2, both below the design order 3. On the intermediate mesh the two corrected methods also differ noticeably (1.43e-5 versus 1.76e-5 for rho), while they nearly coincide on the coarser meshes. No explanation is given. If this behavior persists under further refinement, it contradicts the claim that the corrections always preserve high-order consistency for slip-wall conditions. The paper should either explain the anomaly (e.g., as an asymptotic-range or time-integration effect) or restrict the consistency claim accordingly.","section":"Table 2, DG-P2 slip-wall row"},{"comment":"The moving-domain manufactured-solution tests show a similar pattern at the highest polynomial degree: in both 2D and 3D, the corrected FV-P4 results have final-grid rho orders of 3.17 and U1 orders of 3.87, well below the expected fifth order. These numbers appear in tables whose stated purpose is to demonstrate that the corrections recover the designed convergence rates. The manuscript gives no comment on these reductions. They should be discussed, and if they are due to the boundary correction rather than to solver tolerances or quadrature limits, the claim of arbitrary-order consistency needs to be qualified.","section":"Tables 3 and 4, FV-P4 rows"}],"minor_comments":[{"comment":"The displayed formula for alpha_ROD-E contains an extra transpose in the numerator, reading phi^T(xtilde) phi^T(bar x); it should be phi^T(xtilde) phi(bar x), as used in the proof.","section":"Equation (37)"},{"comment":"The notation ||phi(bar x)||_{M^{-1}} is used without a definition; since M is a matrix, the reader must infer the M^{-1}-weighted norm. Defining this norm explicitly would improve clarity.","section":"Equation (42)"},{"comment":"The header row repeats 'rho rho U1' three times without labeling the three blocks as 'w/o correction', 'ROD-E', and 'ROD-L2'. This makes the table harder to read than the corresponding tables in the rest of the paper.","section":"Table 1"},{"comment":"The text contains a typo: 'discontinous Galerkin' should be 'discontinuous Galerkin'. Also, the reliance on the unpublished preprint [30] for the stability properties of ROD-L2 should be indicated more clearly in the main text, since those properties are used to motivate the method.","section":"Introduction, Section 4.3"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. The paper turns ROD boundary corrections into a per-quadrature-point polynomial correction, eliminating the linear-system inversion in each boundary cell. That is a real practical win, especially for moving meshes. The second thing: the multi-D consistency claim is asserted rather than proved, and one convergence table has an order drop the authors don't explain.\n\nWhat is new and good: Propositions 2 and 3 are correct derivations for ROD-E and ROD-L2, with a clear complexity reduction from O(K^3+K^2D+KD) to O(KD) (or O(KD^2) for L2). The extension to ADER-ALE FV on moving 2D/3D meshes is a genuine addition over the 1D analysis in [30]. The numerical assessment is thorough: fixed and moving domains, Dirichlet and slip-wall, up to fifth order in 2D and 3D, with big error reductions over uncorrected boundaries. The Kidder problem and oscillating cylinder give credible qualitative evidence.\n\nSoft spots. First, the step from a per-point algebraic identity to global high-order consistency for a multi-D boundary integral is missing. The paper says 'while always preserving high-order consistency' in the intro and Sec. 4.3, but no error estimate or Taylor argument appears. A referee should ask for that. Second, Table 2's DG-P2 slip-wall rows are a real worry: ROD-E rho order drops to 2.46 on the finest mesh, below the design order 3, and ROD-E/ROD-L2 diverge on the intermediate mesh (1.43e-5 vs 1.76e-5) without comment. It could be mesh-related or a bug; it needs explanation. Third, no code or data is released, which makes the anomaly harder to chase.\n\nWho should read it: anyone implementing high-order boundary conditions on linear meshes in FV or DG, particularly in ALE moving-mesh codes. It gives a simple recipe and strong benchmarks. The paper deserves peer review; conditional acceptance on a consistency proof (or a careful statement of hypotheses) and an explanation of the Table 2 slip-wall behavior. If that anomaly persists, the claim 'always preserving high-order consistency' is too broad.","headline":"Useful matrix-inversion-free boundary correction with broad numerical support, but multi-D pointwise consistency is asserted rather than proved and one slip-wall table shows an unexplained order drop.","tokens_in":27894,"tokens_out":3523,"would_cite":true,"duration_ms":30508,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M60","65M08","65N50","76N15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that ROD-type high-order boundary conditions on curved domains can be imposed pointwise as simple polynomial corrections, eliminating the per-cell constraint-matrix inversion and preserving high-order accuracy on fixed and…","keywords":["curved domains","high-order boundary conditions","Reconstruction for Off-site Data","polynomial correction","discontinuous Galerkin","ADER-ALE finite volume","compressible Euler equations","moving meshes"],"falsifier":"Run a DG-P3 manufactured-solution test on a domain with a single circular-arc boundary face, so any local consistency defect cannot cancel through symmetry, and measure the $L^2$ error convergence in all conserved variables. If the pointwise correction introduces a defect of order below $p+1$ that happens to cancel only on the concentric-geometry tests in the paper, the observed order will drop below the designed value on this single-arc geometry.","tokens_in":26774,"feed_emoji":"📐","tokens_out":7754,"duration_ms":66177,"temperature":0.7,"pith_summary":"The paper claims that the Reconstruction for Off-site Data (ROD) boundary treatment, normally posed as a constrained minimization that needs inversion of a $K\\times K$ linear system in each boundary cell, can be rewritten as a pointwise polynomial correction that needs no matrix inversion. In the corrected form, a single boundary constraint is enforced at the physical-boundary point corresponding to each quadrature point on the computational boundary, and the modified polynomial value is obtained directly from evaluations of the basis functions. The claim is developed and tested in a Runge-Kutta discontinuous Galerkin framework on fixed curved domains and in an ADER arbitrary-Lagrangian-Eulerian finite volume framework on moving curved domains. If correct, high-order accuracy on simple simplicial linear meshes becomes substantially cheaper and easier to implement, especially for moving boundaries.","feed_headline":"Curved-boundary fix restores high order, drops matrix solve","feed_subtitle":"Applying one boundary condition per quadrature point recovers up to fifth-order accuracy on straight-edged meshes.","key_machinery":"The key object is the signed-distance mapping between computational and physical boundaries, $\\bar{x}=\\tilde{x}+d(\\tilde{x})n$, together with the single-constraint ROD polynomial generated along that mapping. The correction formula $v_h(\\tilde x)=[\\varphi(\\tilde x)-\\alpha\\,\\varphi(\\bar x)]^T\\hat u+\\alpha\\,u_D(\\bar x)$ is obtained by solving the one-constraint optimality system analytically; $\\alpha$ is the ratio of two inner products of basis evaluations at $\\tilde{x}$ and $\\bar{x}$, with the mass matrix inserted for ROD-L2. This machinery converts a boundary-condition imposition that required assembling and inverting $\\Phi^T(\\bar{x})\\Phi(\\bar{x})$ into a scalar evaluation per quadrature point, so the per-boundary-cell cost drops from $O(K^3+K^2D+KD)$ to $O(KD)$ for ROD-E and $O(KD^2+KD)$ for ROD-L2.","core_discovery":"The central discovery is that the ROD constrained-minimization problem reduces, at each boundary point, to a closed-form correction of the internal polynomial. For a point $\\tilde{x}$ on the computational boundary mapped to $\\bar{x}$ on the physical boundary, the corrected value is $v_h(\\tilde x)=[\\varphi(\\tilde x)-\\alpha\\,\\varphi(\\bar x)]^T\\hat u+\\alpha\\,u_D(\\bar x)$, where $\\varphi$ is the polynomial basis, $\\hat u$ the internal degrees of freedom, and $\\alpha$ is chosen so that $v_h(\\bar x)=u_D(\\bar x)$ exactly. Proposition 2 gives $\\alpha_{\\mathrm{ROD-E}}=\\varphi^T(\\tilde x)\\varphi(\\bar x)/\\|\\varphi(\\bar x)\\|_2^2$ for the Euclidean distance, and Proposition 3 gives the mass-matrix-weighted analogue for the $L^2$ distance. Applied independently at every boundary quadrature point, these corrections replace the cell-wise ROD polynomial and its $K\\times K$ constraint system. The paper presents this as a new scheme in multiple dimensions, distinct from the original ROD, and supports it with convergence tests up to fifth order in 2D and fourth order in 3D for Euler flows.","pith_inferences":["Because the correction depends only on basis evaluations at the mapped point, the same formula could be inserted into any polynomial-based discretization with a boundary cell, including nodal spectral element methods or embedded ghost-cell solvers, without per-cell matrix assembly.","For high polynomial degrees, ROD-L2's mass-matrix-weighted projection should control the boundary error better than ROD-E's Euclidean projection; a systematic stability comparison across $p$ and mesh families would test whether the extra $O(KD^2)$ cost is worthwhile.","The real computational win is on moving meshes: fixed meshes can precompute the constraint-matrix inverse, but moving meshes cannot, so eliminating that inversion removes a bottleneck that grows with polynomial degree and refinement.","The pointwise decoupling would allow mixed boundary-condition types on different parts of the same boundary face, such as Dirichlet on one segment and slip-wall on another, without extending the constraint system, which the original cell-wise ROD cannot do without additional constraints."],"forward_implications":["No constraint matrix is assembled or inverted per boundary cell: ROD-E costs $O(KD)$ and ROD-L2 costs $O(KD^2+KD)$ instead of the $O(K^3+K^2D+KD)$ of the original ROD method.","On fixed simplicial meshes, uncorrected Dirichlet and slip-wall boundary conditions limit convergence to second order; the ROD-E and ROD-L2 corrections restore the designed order, observed up to fifth order in 2D and fourth order in 3D.","The same pointwise correction works in RK-DG, FV, and ADER frameworks whenever the internal solution has a polynomial representation, by replacing the degree-of-freedom vector $\\hat u$.","On moving curved domains, the correction needs only evaluations of basis functions at the mapped boundary points, so it avoids the per-time-step, per-cell matrix inversion that the original ROD formulation requires.","The moving-domain tests, including a horizontally oscillating cylinder, show that the correction suppresses spurious boundary layers and entropy production near curved walls."],"supporting_citations":[{"why":"Introduces the original ROD least-squares formulation for finite volumes, the constrained minimization the paper recasts.","marker":"[24]"},{"why":"Presents ROD for discontinuous Galerkin with multiple boundary constraints; system (31) is the target the pointwise correction avoids.","marker":"[26]"},{"why":"One-dimensional analysis showing ROD as a polynomial correction, which the paper extends to multiple dimensions and several frameworks.","marker":"[30]"},{"why":"Defines the shifted boundary polynomial correction whose pointwise form the new corrections resemble and extend.","marker":"[27]"},{"why":"Extends shifted boundary polynomial corrections to moving curved boundaries in ALE methods, the moving-domain context assessed here.","marker":"[29]"}],"fun_headline_variants":["Closed-form ROD correction drops matrix solve, hits 5th order","Straight mesh edges hit 5th order with matrix-free ROD","No linear system: ROD becomes closed-form boundary correction","Matrix-free ROD: high-order curved boundaries from straight meshes","ROD without the solve: polynomial corrections for curved boundaries"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that imposing one boundary condition per boundary quadrature point, independently, preserves the designed high order in multiple dimensions; Section 4.3 asserts this while always preserving high-order consistency but supplies no proof, so the numerical convergence tables are the only support.","fun_headline_variants_meta":{"raw":{"variants":["Closed-form ROD correction drops matrix solve, hits 5th order","Straight mesh edges hit 5th order with matrix-free ROD","No linear system: ROD becomes closed-form boundary correction","Matrix-free ROD: high-order curved boundaries from straight meshes","ROD without the solve: polynomial corrections for curved boundaries"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000899,"raw_usage":{"total_tokens":3960,"prompt_tokens":1121,"completion_tokens":2839,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":737,"completion_tokens_details":{"reasoning_tokens":2750}},"tokens_in":737,"tokens_out":2839,"duration_ms":21931,"temperature":1.0,"reasoning_tokens":2750,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T23:49:44.225497+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a DG-P3 manufactured-solution test on a domain with a single circular-arc boundary face, so any local consistency defect cannot cancel through symmetry, and measure the $L^2$ error convergence in all conserved variables. If the pointwise correction introduces a defect of order below $p+1$ that happens to cancel only on the concentric-geometry tests in the paper, the observed order will drop below the designed value on this single-arc geometry.","supporting_citations":[{"cited_title":"Costa, S","cited_arxiv_id":null,"evidence_quote":"Introduces the original ROD least-squares formulation for finite volumes, the constrained minimization the paper recasts."},{"cited_title":"Santos, A","cited_arxiv_id":null,"evidence_quote":"Presents ROD for discontinuous Galerkin with multiple boundary constraints; system (31) is the target the pointwise correction avoids."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"One-dimensional analysis showing ROD as a polynomial correction, which the paper extends to multiple dimensions and several frameworks."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the shifted boundary polynomial correction whose pointwise form the new corrections resemble and extend."},{"cited_title":"Boscheri, M","cited_arxiv_id":null,"evidence_quote":"Extends shifted boundary polynomial corrections to moving curved boundaries in ALE methods, the moving-domain context assessed here."}],"review_version":1}