{"id":"44e9046a-74b3-478a-91e6-955c465feba3","arxiv_id":"1909.01480","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Two approaches compute symmetric triangle quadrature rules that exactly integrate chosen sequences of singular functions; on a test electromagnetic integral, they reduce relative error by up to two orders of magnitude versus polynomial rules.","lead":"This paper presents two ways to build symmetric triangle integration rules that can handle singular functions, not just polynomials. The methods target electromagnetic simulation integrals with sharp boundary singularities, where polynomial rules converge slowly and non-monotonically.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reported gains rest on unverified completeness of the chosen singular sequence; Section 4 concedes cross terms are omitted, so the \"arbitrary functions\" claim is wider than demonstrated.","rationale":"The reader's conditional verdict is appropriate. The paper proposes a clear numerical construction, reports machine-precision residuals for explicitly listed function sequences, and shows an improvement on the test problem. The weakest point is not the nonlinear solver or the symmetry reduction but the representational assumption that a short finite sequence of one- or two-dimensional logarithmic singular functions captures the singular part of the true integrand. The paper itself supplies the missing-completeness admission in Section 4, where the one-dimensional characterization is acknowledged to omit cross terms, and in Section 7, where optimality and uniqueness are acknowledged as limited. A rule that exactly integrates a finite prescribed sequence cannot, without further argument, justify the title claim of integrating arbitrary functions; hence the two-order error gain should be read narrowly. This does not invalidate the demonstrated results, but it does mean the paper should be accepted with the condition that the generalizability claim be either narrowed or supported by additional tests. Since the reader already reached CONDITIONAL, no verdict change is needed.","tokens_in":11607,"tokens_out":17413,"duration_ms":187846,"concrete_test":"Construct the cross-term-enlarged one-dimensional sequence for Approach 1 by adding, for each polynomial group of degree k, the functions x^p y^q ln x and x^p y^q ln(1-x-y) for p+q=k, and re-run the Table 4 / Figure 7a computation for n = 12, 16, 25, 27. If the relative error for I_c changes by less than roughly half a digit, the omitted cross terms are not load-bearing for this integrand; if it improves materially, the reported gains are tied to the specific sequence and the general claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central effectiveness claim depends on the assumption that the finite singular sequences in Tables 2 and 3 faithfully represent the boundary singularity structure of I_c. Section 4 explicitly states that the one-dimensional characterization omits cross terms, offers three ad hoc remedies, and provides no completeness criterion; Section 6.1 selects the sequences by reference to [10] rather than by an error analysis. The resulting rules exactly integrate the finite lists such as x^{2j-1} ln x, but not unsampled cross terms of the form x^p y^q ln x that can appear in the boundary-layer expansion of the iterated integral. Consequently, the two-order-of-magnitude improvement in Figures 7a/8a is evidence for these particular sequences, not for \"arbitrary functions\"; a different singular integrand with the same edge and vertex singularities but different coupling could fall back to polynomial-rule performance. This is a generalization gap rather than an internal inconsistency, and the paper itself acknowledges related limitations in Section 7, but it is load-bearing because the abstract and title claim arbitrary functions.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes two numerical approaches for constructing symmetric quadrature rules on triangles that exactly integrate prescribed sets of functions, including functions with boundary singularities. Approach 1 solves a least-squares problem for orbit coordinates and weights, using polynomial rules as initial guesses and replacing some polynomial function groups with singular functions. Approach 2 builds tensor-product rules from one-dimensional generalized Gaussian rules and maps them to the triangle via a subdivision into three quadrilaterals. The methods are demonstrated on integrals with logarithmic singularities arising in electromagnetic simulations, where the new rules are reported to achieve relative errors up to two orders of magnitude smaller than standard polynomial rules for the singular test integral. The paper also discusses computational techniques, initial-guess strategies, and a uniqueness heuristic for Approach 1.","tokens_in":11776,"tokens_out":5373,"duration_ms":53912,"significance":"If the results are accepted, the paper provides a practical extension of symmetric triangle quadrature beyond polynomial integrands, addressing a real need in computational electromagnetics. Strengths include the clear description of both algorithms, the use of machine-precision residual checks for the constructed rules, and the direct comparison against established polynomial rules on a nontrivial singular test problem. The claim of 'arbitrary functions' is, however, broader than what is actually demonstrated: the construction depends on the user-supplied function sequence, and the numerical gains are shown only for a particular singular integrand with sequences borrowed from a companion reference. The paper would be more compelling with a reproducibility package containing the actual rules and with a more cautious statement of the scope.","major_comments":[{"comment":"The actual quadrature rules are not reported anywhere in the manuscript: Tables 4 and 5 give only orbit counts and maximum function-group indices, while Figures 4-6 show point distributions but no coordinates or weights. Since the paper's contribution is a set of symmetric triangle rules, this omission prevents readers from using or independently verifying the rules. Provide the computed points and weights as supplementary material or in an appendix.","section":"Section 6 and Appendix A"},{"comment":"The phrase 'arbitrary functions' overstates the demonstrated scope. Approach 1 requires the user to prescribe a finite sequence of functions, and the reported accuracy for I_c relies on singular sequences taken from Reference [10] without a completeness or convergence analysis. Section 4 explicitly concedes that one-dimensional characterization omits cross terms, and Section 6.1 gives no argument that the chosen sequences capture all relevant singular behavior. The authors should either soften the title/abstract claims to 'prescribed function sequences' or provide a theoretical or systematic numerical justification for the generalization.","section":"Abstract and Section 1"},{"comment":"Approach 2 inherits the same completeness limitation in a more direct form: the tensor product of one-dimensional rules does not integrate cross terms such as x^p y^q ln x that can appear in edge and vertex expansions of the test integrand. The paper notes this issue for Approach 1 in Section 4 but does not discuss its effect on the monotonic convergence claimed for Approach 2 in Section 6.3. A comparison of Approach 2's error behavior with and without such cross terms, or at least a discussion of why they are negligible for the test cases, would strengthen the empirical claims.","section":"Section 5 (Approach 2)"}],"minor_comments":[{"comment":"The displayed objective function appears to have a typesetting error: the exponent on the relative error is missing or misplaced, and there is an unbalanced parenthesis. Please correct the formula.","section":"Equation (3)"},{"comment":"For n = 1, the table lists ng = d = 1; while the centroid rule indeed integrates linear functions on the reference triangle, the entry would benefit from a short explanation because standard polynomial-degree conventions sometimes count only constants for a one-point rule.","section":"Table 1"},{"comment":"The legends and curves for 'Approach 1, 1D Singularities' and 'Approach 1, 2D Singularities' may be difficult to distinguish in grayscale; consider using distinct markers or line styles.","section":"Figures 7 and 8"},{"comment":"The sentence 'the ability to integrate f(α,β) indicates the ability to integrate f(β,α)' is correct for symmetric rules but would be clearer if the symmetry group of the equilateral triangle is explicitly invoked, since the reduction in the number of functions relies on it.","section":"Section 3.1"}],"recommendation":"major_revision","confidential_remarks":"The overlap with Reference [10] (by four of the same authors) is substantial: the singular function sequences, the reference solutions, and the motivating integrals all come from that companion paper. The editor may wish to verify that the present manuscript offers sufficient independent contribution beyond [10] and that the reference solutions are not circular, even though the quadrature construction itself is self-contained. The missing data for the actual rules is a reproducibility concern that should be addressed before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a genuinely useful methods paper. The two approaches are clearly described and the numerical results show real gains on a practical EM integral. The title and abstract promise \"arbitrary functions,\" but the evidence only covers specific singular sequences, and there is no error analysis or completeness argument. Still, I think it deserves peer review.\n\nWhat's new: Approach 1 generalizes the standard symmetric polynomial-rule optimization to arbitrary function sequences by grouping functions and replacing high-degree polynomial groups with singular functions. Approach 2 takes one-dimensional generalized Gaussian rules, takes an outer product, and maps to three quadrilaterals around the triangle; that gives symmetric rules that are monotone in n. Both are natural extensions of Ma/Rokhlin/Wandzura and Vipiana et al., but the symmetric triangle construction with function groups is new as far as I know.\n\nWhat's done well: the paper is honest about the difficulties. It openly says the solver depends on initial guesses, uses random perturbations with 40 Latin hypercube samples, and that rules may not be unique. It explains the cross-term limitation of one-dimensional characterizations and offers three remedies. It computes to 64-digit precision with a residual tolerance of 1e-150, which is strong evidence the listed rules actually integrate the stated sequences. The test problem is relevant and the two-order-of-magnitude improvements in Figures 7a/8a are plausible.\n\nSoft spots: no point/weight tables or code shipped, so the rules are not directly usable without re-running the optimization, which is a nontrivial step. The \"arbitrary functions\" claim is broader than demonstrated; the examples are limited to logarithmic singularities from EM, and there's no argument that the chosen sequences capture the full singularity structure. Section 4 explicitly acknowledges cross terms are omitted; the stress-test's concern about unsampled cross terms is legitimate, though it's a generalization gap, not an internal inconsistency. The reference solutions come from [10], which shares authors, so the benchmark isn't fully independent. No error analysis.\n\nFor whom: computational scientists working on singular integrals in EM or potential theory. I'd cite it if I were in that area. It deserves a serious referee; the referee should ask for reproducibility (tables or code) and a more modest claim about scope.","headline":"Solid methods paper on symmetric triangle rules for singular integrands; the 'arbitrary functions' title oversells slightly, but the construction is sound and the gains on the EM test problem are real.","tokens_in":12314,"tokens_out":2058,"would_cite":true,"duration_ms":19543,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65D32","65D30"],"pacs":[],"model":"deepseek-v4-flash","headline":"Symmetric triangle quadrature rules can be built to exactly integrate logarithmic singular functions, and on a representative electromagnetic integral they reduce relative error by two orders of magnitude compared with polynomial rules.","keywords":["symmetric quadrature rules","triangle quadrature","singular integrands","generalized Gaussian quadrature","logarithmic singularities","electromagnetic integrals"],"falsifier":"Take the same construction and apply it to an integrable boundary singularity outside the logarithmic family, such as $r^{-1/2}$ where $r$ is distance to a vertex, and check whether the relative error still stays two orders of magnitude below polynomial rules; if it does not, the method's demonstrated value is specific to the logarithmic singularity sequence rather than to arbitrary functions.","tokens_in":11394,"feed_emoji":"📐","tokens_out":8002,"duration_ms":72646,"temperature":0.7,"pith_summary":"This paper develops two ways to compute symmetric quadrature rules on triangles that exactly integrate user-selected sequences of functions, not just polynomials. The motivation is that polynomial rules converge slowly and non-monotonically for integrands with boundary singularities, such as the logarithmic singularities that arise in electromagnetic reaction integrals. Approach 1 solves a scaled nonlinear least-squares problem for the orbit points and weights, starting from known polynomial rules and replacing the highest-degree polynomial groups with singular functions; it suits moderate point counts. Approach 2 tensor-products one-dimensional generalized Gaussian rules and maps the unit square to three quadrilaterals inside the triangle, suiting larger point counts and giving monotonic error decrease. On the singular test integral $I_c$, both approaches generally beat polynomial rules, with Approach 1 often giving relative errors about two orders of magnitude smaller at the same number of points.","feed_headline":"New triangle rules cut singular-integral error by 100x","feed_subtitle":"For logarithmic edge and vertex singularities, the new symmetric rules beat polynomial rules by two orders of magnitude.","key_machinery":"The machinery is the orbit decomposition of symmetric triangle rules combined with function groups. A rule's points live in three orbit types: one centroid point (type 0), three points on medians (type 1), and six points off the medians (type 2); points in an orbit share one weight, so the unknowns are per orbit. The paper groups the target sequence into function groups $f_k$, each either a singularity or the monomials of a fixed degree, and requires the rule to integrate all preceding groups. Approach 1 solves for the orbit unknowns with a nonlinear least-squares objective in which each function's quadrature sum is compared with its exact integral; Approach 2 bypasses the multivariate optimization by building a one-dimensional generalized Gaussian rule for the singularity sequence and tensor-producting it, then mapping it to the triangle via a bilinear transformation of the three quadrilateral subdomains.","core_discovery":"The central claim is that a symmetric triangle rule can be designed to integrate exactly a prescribed function sequence that mixes monomials with singular terms, e.g. $x^{2j-1}\\ln x$ for $j\\in\\mathbb{N}^+$, while retaining the standard orbit structure of symmetric triangle rules (centroid, three median points, six off-median points). Approach 1 obtains such a rule by minimizing a scaled least-squares objective over orbit coordinates and weights, seeded with polynomial-rule points and weights, and accepting only interior points; the cost is a reduction in the maximum polynomial degree integrated exactly. Approach 2 instead forms the product of one-dimensional rules that integrate the same one-dimensional singularity sequence and maps the product rule to the triangle through a bilinear map of three quadrilateral subdomains, producing $n=3n'^2$ points. For the singular integral $I_c$, the resulting rules yield relative errors that are often two orders of magnitude below those of polynomial quadrature rules at comparable point counts, while for the smooth integral $I_s$ the polynomial rules remain best and Approach 1 remains comparable.","pith_inferences":["If the $x^{2j-1}\\ln x$ sequence is assumed complete, the same construction should transfer to other integrable boundary singularities (e.g., $x^{-1/2}$ or $x^p\\ln x$) by swapping the singularity sequence, but the paper does not test this.","Because Approach 2 builds triangle rules from one-dimensional rules and a bilinear quadrilateral map, its accuracy is largely set by the quality of the one-dimensional generalized Gaussian rule; tuning that rule for a different singularity family may be cheaper than rerunning Approach 1's optimization.","The reported gains for $I_c$ suggest that replacing polynomial triangle rules in boundary-element or method-of-moments codes with these rules could reduce the number of integration points needed for comparable accuracy, but that impact is an extrapolation beyond the paper's test integrals."],"forward_implications":["For the singular integral $I_c$, the new rules reach relative errors roughly two orders of magnitude smaller than polynomial rules at several point counts (e.g., $n=27$ for Approach 1).","Approach 2's relative error decreases monotonically with the number of points, guaranteeing that adding points improves accuracy for singular integrands.","Because both approaches preserve the symmetric orbit structure, the computed rules map to arbitrary triangles by an affine transformation with weights scaled by twice the area.","Because the function sequence is user-selected, a user can trade polynomial degree for the ability to integrate a chosen set of singular functions; the maximum polynomial degree drops as singular functions are added.","For a smooth test integrand ($I_s$), polynomial rules remain the most accurate, while Approach 1 retains comparable efficiency."],"supporting_citations":[{"why":"Establishes the orbit decomposition of symmetric triangle quadrature rules that the paper's point layouts inherit.","marker":"[1]"},{"why":"Supplies the polynomial symmetric rules used as initial guesses and the orbit counts $(n_0,n_1,n_2)$ listed in Table 1.","marker":"[2]"},{"why":"Provides the one-dimensional generalized Gaussian quadrature and continuation method that Approach 2 builds upon via outer products.","marker":"[7]"},{"why":"Extends one-dimensional rules to triangles asymmetrically and motivates the symmetric construction; also supplies the electromagnetic singular integral context.","marker":"[8]"},{"why":"Provides the singular function sequences $x^{2j-1}\\ln x$ and the radial-angular reference solutions used to evaluate the test integrals.","marker":"[10]"},{"why":"Supplies the improved transformation used to compute the reference solutions for $I_c$ and $I_s$.","marker":"[15]"},{"why":"Provides alternative symmetric polynomial rule orbit counts and algebraic solving used to obtain additional solutions when the primary choices fail.","marker":"[4]"}],"fun_headline_variants":["Triangle rules slash singular-integral error 100x","100x less error for singular triangle integrals","Symmetric rules beat polynomial by 2 orders","New triangle quadrature for singular integrands","Two orders better: triangle rules for singular functions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The accuracy gains hinge on the assumption that the chosen logarithmic sequence $x^{2j-1}\\ln x$ (or its two-dimensional counterpart) completely captures the singular behavior of the integrand on the triangle edges and vertices, since the paper offers no completeness proof for this representation.","fun_headline_variants_meta":{"raw":{"variants":["Triangle rules slash singular-integral error 100x","100x less error for singular triangle integrals","Symmetric rules beat polynomial by 2 orders","New triangle quadrature for singular integrands","Two orders better: triangle rules for singular functions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000412,"raw_usage":{"total_tokens":2099,"prompt_tokens":877,"completion_tokens":1222,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":1152}},"tokens_in":493,"tokens_out":1222,"duration_ms":9189,"temperature":1.0,"reasoning_tokens":1152,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:16:17.064360+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same construction and apply it to an integrable boundary singularity outside the logarithmic family, such as $r^{-1/2}$ where $r$ is distance to a vertex, and check whether the relative error still stays two orders of magnitude below polynomial rules; if it does not, the method's demonstrated value is specific to the logarithmic singularity sequence rather than to arbitrary functions.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the orbit decomposition of symmetric triangle quadrature rules that the paper's point layouts inherit."},{"cited_title":"Vipiana, D","cited_arxiv_id":null,"evidence_quote":"Extends one-dimensional rules to triangles asymmetrically and motivates the symmetric construction; also supplies the electromagnetic singular integral context."},{"cited_title":"Characterization and Integration of the Singular Test Integrals in the Method-of-Moments Implementation of the Electric-Field Integral Equation","cited_arxiv_id":"1911.02107","evidence_quote":"Provides the singular function sequences $x^{2j-1}\\ln x$ and the radial-angular reference solutions used to evaluate the test integrals."}],"review_version":1}