{"id":"c6415b75-2644-43b2-8168-b626f448f25e","arxiv_id":"2501.18065","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A corner-graded change of variables plus a change of unknown that absorbs the vanishing line element yields 8th-10th order convergence for 2D scattering integral equations with unbounded densities at corners.","lead":"A new numerical recipe restores high-order accuracy for electromagnetic scattering problems on objects with sharp corners, where the unknown solution becomes infinite and standard high-order methods stall. The authors smooth out the corner singularity by changing variables and unknowns, achieving errors near 10^-11 even a distance 10^-8 from the corner.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The regularity of the regularized density ψ is asserted rather than proved; because the required smoothness depends on the corner singularity exponent via β=p(1−ν)−1, a fixed p=4 or 6 is not guaranteed to sustain the claimed high order for arbitrary corners or for 3D edges.","rationale":"The reader's weakest assumption identifies the same soft spot: the corner-regularized unknown ψ is smooth only if the CoV order p is large enough relative to the corner singularity exponent, and this condition is asserted rather than demonstrated. My stress-test pass found no internal algebraic inconsistency in the derivation of Eq. (44), and the exact-solution test in Section 4.3 plus the corner-exponent recovery in Section 4.4 provide meaningful independent support for the 2D implementation on the tested geometries. The concern is therefore not that the reported numbers are wrong, but that the central claim generalizes beyond what is shown: the regularity exponent β=p(1−ν)−1 in Section 3.4 shows explicitly that the attainable convergence rate depends on the product of p and the geometry-dependent exponent ν, and no theorem, bound, or selection rule is supplied. Because the paper already receives a CONDITIONAL verdict, this concern reinforces the condition rather than changing it: the 2D results are credible, but the claim that the method works 'without requiring a priori analysis of the geometry' should be tempered until the p-regularity condition is analyzed, and the 3D generalization should be presented as an expectation rather than a direct consequence. A single near-crack exact-solution experiment with p=6 would test whether the empirical success extends into the regime where β is predicted to be small.","tokens_in":20811,"tokens_out":7415,"duration_ms":93415,"concrete_test":"Run the Fig. 9 manufactured exact-solution test (monopole point source inside a PEC scatterer) on a near-crack parallelogram with interior angle 0.001 rad using CoV order p=6, evaluating the field at d=10^-8 from the corner, and compare the observed convergence order with the predicted regularity bound β=p(1−ν)−1 computed from the known wedge exponent for that angle. If the observed order falls below the claimed 8th order, or if β is below the observed order, the missing p-selection rule is a genuine limitation; if the order holds and β is consistent, the empirical parameter choice is adequate for that regime.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The method's pivotal step is Eq. (42), ψ(θ)=φ(s(θ))L̃(θ), combined with the Section 3.2 assertion that 'the product of Jacobian and density has a number of bounded derivatives at the corner point.' This is the mechanism that makes the Chebyshev/RP discretization converge at high order, but its validity is angle-dependent and is never proved. Writing the wedge asymptotics as φ(s)∼s^{−ν}, with s(θ)∼θ^p and L̃(θ)∼θ^{p−1}, gives ψ(θ)∼θ^{p(1−ν)−1}; the paper itself uses this estimate in Section 3.4, where it states ψ_q(θ′)∼(θ′)^{(p−1)−νp}. Thus the achievable polynomial order is controlled by β=p(1−ν)−1, and reaching r-th order requires p>(r+1)/(1−ν). For a near-crack geometry (interior angle approaching zero), the standard wedge exponent ν approaches 1/2, so even p=6 yields only β≈2; the reported 8th–10th order at 0.01 rad is then not explained by the stated smoothness argument and may depend on features specific to the tested angles. The paper gives no automatic rule or a priori bound for selecting p for arbitrary geometry, and for 3D edges the analogous exponent is not known in closed form. Therefore the central claim that high-order accuracy is achieved 'without requiring a priori analysis of the geometry' rests on an unverified regularity assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops a high-order Nyström solver for 2D TM electromagnetic scattering by PEC scatterers with corners, targeting integral equations whose density blows up at corners. The method combines an operator-regularized combined-field integral equation (CFIE-R), a graded change of variables s(θ) of order p, a change of unknown ψ(θ)=φ(s(θ))L̃(θ), precision-preserving treatment of near-corner source-target differences, and adaptive Gauss-Kronrod quadrature for precomputed weights. Numerical results for square, parallelogram, and teardrop cross-sections report relative errors near 10^-11 or below at distance 10^-8 from corners, observed convergence orders 8-10 for interior angles down to 0.01 radians, favorable conditioning and GMRES iteration counts versus MFIE, recovery of a theoretical corner exponent, and agreement with a commercial FDTD solver.","tokens_in":21124,"tokens_out":9857,"duration_ms":118804,"significance":"If the claims hold, the work is significant: it would remove a long-standing barrier for high-order integral-equation solvers on non-smooth geometries, avoiding singular basis functions or geometry-dependent exponents, and it offers a plausible route to 3D edge singularities. The paper has genuine strengths: an honest ablation study (Section 4.1), an exact-solution test for a monopole source (Fig. 9), recovery of the corner exponent as an independent post-hoc check (Fig. 16 right), and comparisons with a commercial solver (Fig. 14). No parameters are fit to the reported errors; the CoV order p is the only user parameter. The main caveat is that the central regularity assumption is asserted rather than proved, and the fixed-p strategy is not justified for arbitrary corner angles.","major_comments":[{"comment":"The claim that ψ(θ)=φ(s(θ))L̃(θ) has 'a number of bounded derivatives at the corner point' is not proved, and no rule is given for choosing p for a given geometry. The paper's own estimate ψ_q(θ′)∼(θ′)^{(p−1)−νp} implies that the number of bounded derivatives is controlled by β=p(1−ν)−1; for a near-crack geometry (ν→1/2) and p=6, β≈2, which is not by itself compatible with the 8th–10th order rates reported in Fig. 13. Because this regularity is the mechanism by which the Chebyshev/RP discretization is claimed to converge at high order, the central assertion 'without requiring a priori analysis of the geometry' is not yet established. Please provide a theorem or rigorous a priori bound connecting p, ν, and achievable order, or an automatic p-selection rule, or explicitly narrow the claim to the tested angles and p values.","section":"Section 3.2, Eq. (42); Section 3.4"},{"comment":"The main regularized equation contains index errors that make it ambiguous as printed: in the first source integral the unknown is written ψ_q(θ′) although the integration is over patch q′, so the displayed equation appears to couple every target point to the single unknown ψ_q(θ). The regularization terms contain similar notational inconsistencies between the target and source patch indices. Please correct the indices and verify that the discretized system uses ψ_{q′} in all source-patch integrals; as written, Eq. (44) cannot be implemented unambiguously.","section":"Eq. (44)"},{"comment":"The plane-wave convergence studies for the parallelogram, needle, and teardrop geometries use reference solutions computed by the same method on finer grids. The only fully independent accuracy tests are the monopole exact solution for the square cross-section (Fig. 9) and the FDTD field comparisons for the square at k=10 (Fig. 14). The reported 8th–10th order rates for the small-angle cases and the density-error results near corners therefore rest on self-convergence. Please provide an independent reference for at least the small-angle cases, or clearly state that the reported rates for those cases are self-convergence estimates.","section":"Section 4.3, Section 4.4"}],"minor_comments":[{"comment":"Equation (21) gives the same expression for the tangent vector τ and the normal vector n; the normal should be a 90-degree rotation of the tangent, e.g., n=(−dy/ds, dx/ds)/L.","section":"Eq. (21)"},{"comment":"The sign convention for the singularity exponent ν is inconsistent: Section 3.4 uses ν as a positive exponent in ψ_q(θ′)∼(θ′)^{(p−1)−νp}, while Section 4.4 states that φ(d)∼d^ν with ν=−1/3 for the same 90-degree corner. Please clarify the sign convention so the two statements are compatible.","section":"Section 3.4 versus Section 4.4"},{"comment":"References [24] and [25] are the same paper and should be merged.","section":"References"},{"comment":"The abstract's phrase 'near machine precision accuracy' is overstated given Remark 1, which states that the adaptive Gauss-Kronrod integration enforces a 10^-12 accuracy floor in the Matlab implementation; the reported errors near 10^-11 are not machine precision in double arithmetic.","section":"Abstract and Remark 1"},{"comment":"Section 4.1 describes the original formulation as achieving second-order convergence, while Section 4.3 (Fig. 10) describes the no-CoV method as exhibiting first-order convergence; these statements should be reconciled.","section":"Section 4.1 versus Section 4.3"},{"comment":"The text attributes the superiority of CoV (32) over (34) to an 'evenly split distribution' in Section 3.2 but later attributes it to an 'uneven distribution' in Section 4.3; the wording should be harmonized.","section":"Section 4.3 discussion of CoV (32)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a good fit for the journal and the numerical evidence is substantial, but the central smoothness argument is asserted rather than proved; I would ask the authors to either prove the regularity claim under explicit hypotheses on p and ν, or provide an automatic p-selection rule, or substantially qualify the abstract's claim. The 3D generalization in the abstract is not demonstrated in this paper and should be phrased as an expectation, not a direct consequence. No concerns about citation practices."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is worth a serious read if you work on boundary integral equations for non-smooth geometries. The central trick is simple and effective: instead of approximating the unbounded density φ directly, the authors define ψ = φ L̃, absorbing the graded change-of-variables line element into the unknown. That turns a density that blows up like d^{-ν} into a function that vanishes at the corner, and the Chebyshev-based RP discretization converges at high order. The paper also deals carefully with the finite-precision problems that graded meshes create when source and target points both sit extremely close to the corner. These are genuine contributions, and the ablation study shows each piece matters.\n\nThe numerical evidence is convincing within 2D. There is a manufactured exact solution (monopole inside a PEC square) that gives 9th-order convergence to 1e-11 at distance 1e-8 from the corner. The recovered corner exponent ν = -1/3 matches theory to eight digits. Total fields agree well with a commercial FDTD solver. That is a solid validation package. The reference solutions for plane-wave cases are self-generated, which is common but less convincing; code release would settle it.\n\nThe main soft spot is the regularity argument for ψ. Smoothness of ψ is asserted, not proved. The number of bounded derivatives depends on p and the corner exponent ν. For very sharp corners ν approaches 1/2, so p=6 gives only about two derivatives. The paper's own Section 4.4 explains why observed 8th-10th order is still possible: the dominant error for moderate N comes from smooth patches away from corners, while near-corner error converges more slowly. So the concern is real but the paper already half-answers it. What is missing is a precise convergence-rate statement in terms of p, ν, and patch count, and an automatic rule for choosing p. The 3D generalization is an expectation, not a result, and should be labeled as such.\n\nThis deserves peer review. The method is novel, the 2D validation is strong, and the theoretical gap is fixable, not a dealbreaker. Send it out; ask for code and a sharper regularity statement, and push the p-selection question.","headline":"Strong 2D corner-regularized BIE method with credible high-order convergence; the 3D extrapolation is speculative and the regularity argument is asserted rather than proved, but the numerical evidence carries the paper.","tokens_in":21705,"tokens_out":2825,"would_cite":true,"duration_ms":27392,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65R20","65N38","78A45"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper shows that scattering integral equations whose density blows up at corners can be solved to high order by folding a corner-grading change of variables into a new unknown, without singular basis functions or a priori singularity…","keywords":["electromagnetic scattering","acoustic scattering","integral equations","domains with corners","unbounded solutions at corners","change of variables","Nyström method","high-order accuracy"],"falsifier":"Run the sharpest tested case (the $0.01$-radian parallelogram) with the change-of-variables order reduced to $p=2$ and measure the convergence slope of the field error at distance $d=10^{-8}$ from the corner: if the slope remains high, the smoothness of $\\psi$ is not the limiting premise, and if it drops to first or second order, the vanishing Jacobian is what carries the high-order convergence.","tokens_in":20568,"feed_emoji":"📐","tokens_out":9961,"duration_ms":102314,"temperature":0.7,"pith_summary":"This paper addresses a known failure of high-order boundary integral solvers: at a corner, the integral-equation density can blow up, and standard high-order discretizations drop to first-order accuracy. The authors claim that in the simplest setting where this happens (2D TM scattering, a Neumann-Helmholtz problem whose density diverges at corners), a regularized combined-field integral equation can be made high-order again by a change of variables whose Jacobian vanishes at the corner, with that vanishing line element folded into the unknown. The new unknown is smooth where the old density is singular, so the Chebyshev-based Nyström discretization converges at observed 8th to 10th order, giving field errors below $10^{-11}$ at distance $10^{-8}$ from the corner for interior angles as small as $0.01$ radians. This matters because the method needs neither the singular exponent nor singular basis functions, and the authors expect the same strategy to extend to 3D Maxwell problems, where unbounded edge currents cannot be avoided by formulation choice.","feed_headline":"Solver keeps 9th-order accuracy at 1e-8 from a corner","feed_subtitle":"Folding the change of variables into the unknown absorbs the corner blow-up and restores 8th–10th order convergence.","key_machinery":"Two mechanisms carry the argument. First, the corner-regularized change of unknown $\\psi_q(\\theta)=\\phi(s_q(\\theta))\\widetilde{L}_q(\\theta)$ in equation (44): the change of variables $s(\\theta)$ (based on the function $w(\\theta)$ of equation (32), with $s(\\theta)\\sim\\theta^p$ as $\\theta\\to 0$) grades the mesh toward the corner, and multiplying the unbounded density $\\phi$ by the vanishing line element $\\widetilde{L}_q$ transfers the singularity into the known geometry factor, leaving a smooth unknown for Chebyshev approximation. Second, the precision-preserving quadrature layer: near-corner source-target differences are evaluated through Taylor expansions of the parametrization around the corner to avoid catastrophic cancellation, and the $1/|r-r'|$ peak of the normal-derivative kernel is integrated with adaptive Gauss-Kronrod quadrature rather than fixed Fejér rules. Together these make the regularized equation (44) numerically tractable at the extremely fine mesh spacings the grading produces.","core_discovery":"The paper's central claim is that the infinite-density corner difficulty can be removed by a change of unknown rather than by adding singular basis functions. Starting from the operator-regularized combined field integral equation (CFIE-R), the authors apply a change of variables $s(\\theta)$ whose Jacobian and its first $p-1$ derivatives vanish at the corner, and they define a new unknown $\\psi_q(\\theta)=\\phi(s_q(\\theta))\\widetilde{L}_q(\\theta)$, where $\\widetilde{L}_q$ is the line element in the new variable. Because $\\widetilde{L}_q$ vanishes at the corner, $\\psi$ is smooth where the original density $\\phi$ blows up, so the Chebyshev-based rectangular-polar Nyström discretization can approximate it accurately. With CoV order $p=4$ and $p=6$, plus precision-preserving treatments for near-corner and self interactions, the computed field converges at observed rates between 8th and 10th order and reaches relative errors below $10^{-11}$ at distance $10^{-8}$ from the corner, for square, parallelogram, and teardrop cross-sections including interior angles down to $0.01$ radians. The method also reproduces the theoretical corner exponent $\\nu=-1/3$ for a $90^\\circ$ corner to about eight digits, and it stays resonance-free and well-conditioned where the underlying MFIE becomes singular.","pith_inferences":["If the same smoothness premise holds for edge singularities, the change-of-unknown mechanism should transfer to 3D Maxwell problems, where the tangential current component diverges at edges in every formulation; the paper presents this as expected future work, and the mechanism does not require knowing the singular exponent.","The reported 12-digit error floor is attributed to the Gauss-Kronrod quadrature's accuracy limit in the implementation; replacing it with higher-precision adaptive integration is a direct, testable route to machine-precision corner fields.","The text notes that smooth patches dominate the error at coarse resolutions; an error-balancing strategy that concentrates unknowns near corners only as the mesh refines could reduce cost for a given tolerance.","If the central claim is right, the practical consequence is that singular-basis expansions—whose exponents for 3D corners are not known in closed form and would require their own numerical solution—can be bypassed entirely."],"forward_implications":["Field evaluations at distance $d=10^{-8}$ from a corner reach relative errors below $10^{-11}$, with observed convergence between 8th and 10th order, for interior angles as small as $0.01$ radians.","The corner-regularized CFIE-R system remains resonance-free and well-conditioned at wavenumbers where the MFIE matrix is singular, and GMRES reaches a $10^{-5}$ residual in fewer than about 15 iterations across the tested wavenumber range.","Because the method needs no a priori singularity analysis or singular bases, the same solver handles corners of different angles and curved boundaries (square, parallelogram, teardrop) with essentially the same accuracy.","Each ingredient is necessary: omitting the change of unknown makes the scheme worse than the unregularized formulation, and omitting the change of variables reduces convergence to first order with errors above $10^{-2}$ even past 1000 unknowns."],"supporting_citations":[{"why":"supplies the operator-regularized CFIE-R equation that the corner-regularized formulation (44) extends","marker":"[11]"},{"why":"provides the Chebyshev rectangular-polar Nyström discretization that the paper adapts to graded meshes","marker":"[12]"},{"why":"the related Maxwell RP solver whose patch parametrization and differentiation routines are reused","marker":"[28]"},{"why":"source of the change-of-variables function w(θ) used to grade the mesh toward corners","marker":"[16]"},{"why":"the graded-mesh Nyström method for corner domains whose regularization idea is generalized to unbounded densities","marker":"[31]"},{"why":"the adaptive Gauss-Kronrod quadrature used to precompute the singular near-corner interaction weights","marker":"[18]"},{"why":"supplies the theoretical corner singularity exponent ν = −1/3 used to validate the recovered density asymptotics","marker":"[14]"}],"fun_headline_variants":["Change of unknown absorbs corner blow-up, restores 10th order","Smooth new variable tames infinite corner density","No singular basis: change of unknown fixes corner blow-up","Variable switch gives 8th-10th order accuracy at sharp corners","Corner blow-up absorbed, accuracy 10^-11 near edge"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method's accuracy rests on the assumption that the chosen change-of-variables order $p$ makes the product of the density and the Jacobian, $\\psi(\\theta)=\\phi(s(\\theta))\\widetilde{L}(\\theta)$, smooth at the corner; the paper verifies this empirically for the tested angles rather than proving it or giving an automatic rule for $p$.","fun_headline_variants_meta":{"raw":{"variants":["Change of unknown absorbs corner blow-up, restores 10th order","Smooth new variable tames infinite corner density","No singular basis: change of unknown fixes corner blow-up","Variable switch gives 8th-10th order accuracy at sharp corners","Corner blow-up absorbed, accuracy 10^-11 near edge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000663,"raw_usage":{"total_tokens":3085,"prompt_tokens":1057,"completion_tokens":2028,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":673,"completion_tokens_details":{"reasoning_tokens":1941}},"tokens_in":673,"tokens_out":2028,"duration_ms":15823,"temperature":1.0,"reasoning_tokens":1941,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T00:49:04.955114+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the sharpest tested case (the $0.01$-radian parallelogram) with the change-of-variables order reduced to $p=2$ and measure the convergence slope of the field error at distance $d=10^{-8}$ from the corner: if the slope remains high, the smoothness of $\\psi$ is not the limiting premise, and if it drops to first or second order, the vanishing Jacobian is what carries the high-order convergence.","supporting_citations":[{"cited_title":"P., Elling, T., and Turc, C","cited_arxiv_id":null,"evidence_quote":"supplies the operator-regularized CFIE-R equation that the corner-regularized formulation (44) extends"},{"cited_title":"P., and Garza, E","cited_arxiv_id":null,"evidence_quote":"provides the Chebyshev rectangular-polar Nyström discretization that the paper adapts to graded meshes"},{"cited_title":"A Chebyshev-based high-order-accurate integral equation solver for Maxwell’s equations","cited_arxiv_id":null,"evidence_quote":"the related Maxwell RP solver whose patch parametrization and differentiation routines are reused"},{"cited_title":"Inverse acoustic and electromagnetic scattering theory, 1998","cited_arxiv_id":null,"evidence_quote":"source of the change-of-variables function w(θ) used to grade the mesh toward corners"},{"cited_title":"A Nystr¨ om method for boundary integral equations in domains with corners.Numerische Mathematik 58 , 1 (1990), 145–161","cited_arxiv_id":null,"evidence_quote":"the graded-mesh Nyström method for corner domains whose regularization idea is generalized to unbounded densities"},{"cited_title":"J., and Rabinowitz, P","cited_arxiv_id":null,"evidence_quote":"the adaptive Gauss-Kronrod quadrature used to precompute the singular near-corner interaction weights"},{"cited_title":"P., Ovall, J","cited_arxiv_id":null,"evidence_quote":"supplies the theoretical corner singularity exponent ν = −1/3 used to validate the recovered density asymptotics"}],"review_version":1}