{"id":"94fff0d4-f0c2-47d7-ad81-297b327a5c6b","arxiv_id":"2607.13224","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Open generalized Gaussian quadrature points produce summation-by-parts operators with fewer points than closed rules for the same non-polynomial function space, at the cost of boundary extrapolation; the resulting schemes match polynomial convergence orders on Euler equations when the space is scale","lead":"This paper builds a type of numerical derivative operator, called FSBP, that no longer needs grid points at the boundary of each cell. By using open Gaussian quadrature rules, these operators need fewer points and still preserve conservation and stability; tests on advection, Burgers, and Euler equations show equal or better accuracy when the function space is scaled properly.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Bessel-space rank claim (dim G=27 vs [33]'s 48) is load-bearing: if wrong, the 14-node open GGQ is not exact for (F^2)' and the key efficiency comparison (14 vs 15 nodes) collapses.","rationale":"The reader's weakest assumption points to both the Chebyshev-set precondition and the SVD-based rank determination for the Bessel space. I single out the rank determination because it is concrete, decisive, and testable, and because the Bessel example is the only high-dimensional demonstration of the one-fewer-node advantage. If the true rank is 48 rather than 27, the constructed 14-node rule integrates only a subspace, the FSBP operator is not F-exact, and the claimed efficiency comparison with the 15-node closed rule collapses. The paper's own admission that Huybrechs' algorithm failed here (Sec. 4.3) and the absence of code/data heighten the risk. At the same time, the theoretical counting statement remains valid conditional on the existence of the underlying GGQs, and the exponential-space experiments (Sec. 5) appear internally consistent and reproduce the expected EOCs. Thus the appropriate disposition stays CONDITIONAL: the Bessel claims need independent verification, but the core framework and low-dimensional examples are not fundamentally undermined.","tokens_in":22262,"tokens_out":11152,"duration_ms":104422,"concrete_test":"Independently compute the Gram matrix of the 55 derivative products (J_mu J_nu)' on [0,25] using extended precision (e.g., mpmath with 100 digits), compute its singular values, and count those above thresholds 10^-8, 10^-12, 10^-15, 10^-20. Then evaluate the Table 2 quadrature on all 55 derivative products in this precision and record the maximum absolute error. If the rank at 10^-15 exceeds 27, or if the maximum quadrature error exceeds 1e-10, the Bessel claims in Sec. 4.3 and the counting comparison in Sec. 6 are unsupported; a rank of 27 with errors near machine precision would confirm the paper.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central efficiency claim for open GGQs—N=ceil(M/2) nodes for a 2m-dimensional system—rests on the constructed quadratures being genuinely exact for (F^2)'. In the Bessel example (Sec. 4.3), that exactness depends on the numerical rank determination dim G=27 via a 10^-14 SVD threshold (Remark 4.1), directly contradicting [33]'s value of 48. The authors attribute the discrepancy to ill-conditioning, but no independent computation or code is provided. If dim G is actually 48, a 14-node open rule (28 free parameters) cannot be exact for a 48-dimensional space; the reported 14-node quadrature (Table 2) would only be exact for a 27-dimensional subspace, so the FSBP operator would not be F-exact and the comparison with the 15-node closed rule (Table 3) would be invalid. This matters because Bessel is the only high-dimensional example supporting the efficiency claim; the exponential cases are low-dimensional and not contested. The paper itself flags that Huybrechs' algorithm failed here (Sec. 4.3), further underscoring the fragility.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops generalized function-space summation-by-parts (FSBP) operators on grids that need not include the interval endpoints, constructs F-exact extrapolation/boundary operators for such grids, and builds FSBP operators from open and closed generalized Gaussian quadratures (GGQs). The central claim is that an open GGQ exact for a 2m-dimensional system uses m interior nodes, while a closed GGQ needs m+1 nodes, so open Gaussian FSBP operators are one node more efficient than closed ones whenever an open formulation is admissible. The authors compare the resulting operators with least-squares FSBP operators and the closed GGQ operators of [33], and apply them to linear advection, Burgers, and compressible Euler problems, reporting improved accuracy at equal degrees of freedom and experimental orders of convergence equal to the node count for scaled exponential spaces.","tokens_in":22554,"tokens_out":13420,"duration_ms":132400,"significance":"If the construction and the reported quadratures are sound, the paper makes a useful contribution: it provides a clean extension of the FSBP framework to open grids, introduces F-exact extrapolation operators in this setting, and demonstrates that open GGQ-based operators can match or beat closed ones on non-polynomial spaces in genuinely time-dependent nonlinear problems. The scaled-versus-fixed exponential study in Sec. 5.3 is a particularly honest and informative comparison. However, the theoretical counting claim is stated more generally than the construction supports, and the high-dimensional Bessel example, which is load-bearing for the efficiency claim, is not independently verifiable from the manuscript because no code or exactness residuals are supplied.","major_comments":[{"comment":"The summary statement that 'an open GGQ thus requires N=ceil(M/2) points and a closed GGQ exactly one point more' is only valid when the resulting node count satisfies N >= dim(F), as required for the Vandermonde matrix in Lemma 2.4 to have linearly independent columns (and for the orthonormalization step in Algorithm 2.1 to be possible). For polynomial spaces, F=P_d, one has M=dim((F^2)')=2d, so the minimal open GGQ has d nodes, strictly fewer than dim(P_d)=d+1; no F-exact FSBP operator exists at that node count (e.g., the 2-point Gauss rule on P_2 cannot satisfy the x^2 derivative condition). All examples in Secs. 4-5 satisfy M>=2K (after enrichment in the E_2 case), so the issue does not invalidate those operators, but the general claim in Sec. 6 is false as written. Please restrict the counting statement to M>=2K (or N>=dim(F)) and discuss the polynomial case explicitly.","section":"Sec. 6 and Lemma 2.4"},{"comment":"The Bessel example is load-bearing for the claimed one-node saving in a high-dimensional setting, but the reported dimension dim G=27 versus [33]'s 48 rests on a numerical SVD rank decision with a 10^-14 sigma_max threshold and on an approximate Gauss-Legendre Gram matrix. No code or scripts are provided, so the assertion that [33]'s value is 'incorrect' cannot be checked. If dim G were 48, a 14-node open rule with 28 parameters could not be exact for a 48-dimensional space, and Table 2 would not define an F-exact FSBP operator; the comparison in Table 3 would then be invalid. Please provide a reproducible script and report exactness residuals, e.g., max_{mu,nu} |∫ (J_mu J_nu)' dx - sum_i w_i (J_mu J_nu)'(x_i)| for all 55 pairs, computed in high precision and with a well-conditioned rank determination. It would also help to state whether the measure-continuation method [67] is known to g","section":"Sec. 4.3, Remark 4.1, Tables 2-3"}],"minor_comments":[{"comment":"In Definition 3.1, the closed GGQ formula sums from i=1 to m, but the rule has m+1 nodes and weights; the upper limit should be m+1. As written, the formula is internally inconsistent.","section":"Eq. (3.2)"},{"comment":"There are several cross-reference errors: the text refers to 'Theorem 4.1 below' where the actual item is Remark 4.1; Sec. 6 cites 'Theorem 3.3' for the surplus-degrees-of-freedom discussion, but only Remark 3.3 exists; and Sec. 5.3 refers to 'Theorem 5.1' when it means Remark 5.1.","section":"Sec. 4.3, Sec. 5.3, Sec. 6"},{"comment":"The caption 'not exact for (F^2)' is a conclusion drawn from the authors' own rank computation. Since the table reports a rule from [33], it would be clearer to label it as 'claimed by [33] to be exact; the present authors find it not exact,' so that the reader can distinguish the data from the interpretation.","section":"Sec. 4.3, Table 1 caption"}],"recommendation":"major_revision","confidential_remarks":"The main uncertainty is the Bessel rank claim. Because the paper's central efficiency comparison depends on a 14-node rule being exact for a space whose dimension the authors themselves determine only numerically, I would request source code or detailed reproducible scripts, plus exactness residuals, before final acceptance. The theoretical overstatement about the general counting statement should also be corrected, although it does not affect the specific examples presented."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper is a real step forward for FSBP operators. The new piece is the systematic use of open generalized Gaussian quadratures: they extend the FSBP framework to grids that do not contain the boundary points, construct F-exact extrapolation operators, and show that an open GGQ needs ceil(M/2) points for a 2m-dimensional system while a closed GGQ needs one more. That counting argument is clean, and the numerical experiments on Burgers and Euler are more demanding than what earlier FSBP papers typically show. At equal node counts the open operators are exact for larger spaces, and in the Euler runs they reach the same EOC as polynomial operators with the same number of nodes. Credit is due for the construction details and for honest disclosure of the algorithm failure on the Bessel space.\n\nThe soft spot is exactly where the stress-test note points. The Bessel example is the only high-dimensional case, and its validity depends on dim G = 27, obtained from an SVD threshold of 1e-14 and contradicting the 48 reported in [33]. If the true dimension is closer to 48, the 14-node rule cannot be exact for (F^2)' and the claimed efficiency gain collapses for that example. The paper offers a plausible explanation — ill-conditioning — but no code, no data, and no independent computation. That is a verifiability gap, not a proven flaw. The authors did not hide the difficulty; they report that Huybrechs' algorithm failed and that they switched to measure-continuation. A referee should be able to rerun this with supplied scripts.\n\nThe overlap with the concurrent work [4] is disclosed but not addressed in depth. That is fine for a preprint, but the comparison should be sharpened in revision. The extensive self-citation is mostly legitimate: these are the people who built the FSBP framework, and they cite their own foundational papers where relevant.\n\nOverall, the paper deserves a serious referee. The central idea is sound, the low-dimensional examples are convincing, and the Bessel issue is specific and addressable. I would send it to review, with a request that the authors provide reproducible code and independently verify the rank claim before acceptance.","headline":"Open GGQs give a clean one-node saving over closed GGQs and hold up in nontrivial hyperbolic tests; the Bessel rank discrepancy is a verifiability problem, not a demonstrated error.","tokens_in":23055,"tokens_out":1210,"would_cite":true,"duration_ms":15685,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N12","65D25"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that open generalized Gaussian quadrature constructs summation-by-parts operators for non-polynomial function spaces with exactly one fewer grid point than closed rules, and that with function-space-exact extrapolation they","keywords":["FSBP operators","generalized Gaussian quadrature","open quadrature rules","summation-by-parts","non-polynomial approximation spaces","hyperbolic conservation laws","extrapolation operators","Chebyshev sets"],"falsifier":"On the Bessel space spanned by J_0,…,J_9 on [0,25], independently recompute dim((F^2)′) and test the published 14-node rule for exactness on an L2-orthogonal basis of that space; if the true dimension is 48 (as a related study reported) rather than 27, or if any of the 27 exactness conditions fail by more than quadrature roundoff, the claimed 14-node exact open rule—and with it the one-fewer-node advantage in that example—collapses.","tokens_in":22132,"feed_emoji":"📐","tokens_out":5875,"duration_ms":44182,"temperature":0.7,"pith_summary":"This paper extends function-space summation-by-parts (FSBP) operators—discrete derivative operators that mimic integration by parts for general, non-polynomial approximation spaces—to grids that omit the interval endpoints, and shows that open generalized Gaussian quadratures (GGQs) build such operators with exactly one fewer grid point than closed GGQs while preserving exactness for the same space (F^2)′. The counting is simple: an open GGQ uses m interior nodes to integrate a 2m-dimensional system exactly, whereas a closed GGQ needs m+1 nodes with two pinned to the boundary, so N = ⌈dim((F^2)′)/2⌉ for open and one more for closed. To make open nodes usable, the paper constructs F-exact extrapolation operators that recover boundary and interface values from interior nodes, and assembles the boundary matrix B = t_R t_R^T − t_L t_L^T. On the 1-D compressible Euler equations, scaled exponential open operators reach experimental convergence equal to their number of nodes, matching polynomial operators with the same node count, while fixed-scale exponentials degrade to second or third order. If the construction is right, open Gaussian FSBP operators are the most node-efficient structure-preserving discretizations of the tested non-polynomial spaces, and the same counting will apply to any admissible space.","feed_headline":"Open Gaussian nodes cut one point from closed counterparts","feed_subtitle":"Structure-preserving solvers on non-polynomial spaces gain the same accuracy with one less degree of freedom per element.","key_machinery":"The central object is the (F^2)′-exact generalized Gaussian quadrature in its open form: m interior nodes and positive weights that integrate the 2m-dimensional space (F^2)′ exactly, hence with dim((F^2)′)=M requiring N=⌈M/2⌉ nodes, versus m+1 for the closed form. The companion mechanism is the F-exact extrapolation operator E, whose rows are the least-squares evaluation vectors t_L, t_R mapping interior function values to the endpoints; the boundary matrix is B = t_R t_R^T − t_L t_L^T, so that f^T B g = fg| boundaries, preserving the discrete integration-by-parts property on point sets that exclude the endpoints.","core_discovery":"The central claim is a counting statement plus an operationalization. For a finite-dimensional function space F ⊂ C^1, an FSBP operator exists exactly when a positive quadrature exact for (F^2)′ = {(fg)′ | f,g∈F} exists. Generalized Gaussian quadrature theory provides such quadratures with minimal node counts: an open GGQ has N = ⌈dim((F^2)′)/2⌉ nodes, and a closed GGQ has exactly one more. The paper constructs these open FSBP operators by first computing the open GGQ on (F^2)′, setting the norm matrix P to the quadrature weights, and recovering the antisymmetric part of Q from Q_A V = P V′ – BV/2; the boundary matrix B is built from least-squares F-exact extrapolation of the solution to the","pith_inferences":["Editorial inference: the counting identity N = ⌈dim((F^2)′)/2⌉ suggests an information-theoretic floor for any F-exact FSBP operator: no discretization of a space whose derivative products span M dimensions can use fewer than ⌈M/2⌉ degrees of freedom and remain exact, so open GGQ constructions are optimal whenever they exist.","Editorial inference: the same open-GGQ construction can be transplanted to other nodal methods—discontinuous Galerkin and flux-reconstruction schemes built on non-polynomial bases would inherit the node savings, though they would also inherit the need for extrapolated interface values.","Editorial inference: the Bessel-space episode indicates that the practical bottleneck is not the counting but the computation: when the standard GGQ algorithm fails to converge, dimension determination by numerical rank becomes the deciding step; an independent check of exactness of the published 14-node rule on all 27 basis functions would settle whether the reported savings are realized.","Editorial inference: a testable extension is to run adaptive local scaling for exponential spaces so that the basis is matched to element width (as the 'scaled' operators do) and check whether the node-order convergence persists for systems with shocks and non-smooth data."],"forward_implications":["For any Chebyshev-admissible function space, open GGQs cut the node count for structure-preserving differentiation from N to N−1 relative to closed GGQs, narrowing the gap to the interpolatory ideal N = dim(F).","Open Gaussian FSBP operators make non-polynomial spaces practical in time-dependent conservation-law solvers: in the Euler test, scaled exponential operators reached the same experimental order of convergence as polynomial operators with the same node count.","The exactness gain is real: a four-node open operator can be exact on a larger space than a four-node closed operator, which translates into smaller errors at equal degrees of freedom (observed in the exponential boundary-layer advection test).","Closed GGQs remain optimal among operators that include both boundary nodes, so the choice between open and closed is a trade-off between node efficiency and having boundary data directly available.","The extrapolation step is the price of openness; without a stable F-exact extrapolation operator, the open construction does not yield a usable boundary or inter-element coupling."],"fun_headline_variants":["Open Gaussian nodes: one less point than closed","One node less: open Gaussian FSBP operators","Save a node: open beats closed in Gaussian FSBP","Non-polynomial FSBP gains efficiency with open Gaussian","Open nodes cut one point per element in FSBP"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that for each function space F, the derivative-product space (F^2)′ is (or can be enriched to) a Chebyshev set, so that a positive open generalized Gaussian quadrature with exactly ⌈dim((F^2)′)/2⌉ nodes exists and can actually be computed; the paper itself observes this can fail, as the standard algorithm did not converge for the Bessel space.","fun_headline_variants_meta":{"raw":{"variants":["Open Gaussian nodes: one less point than closed","One node less: open Gaussian FSBP operators","Save a node: open beats closed in Gaussian FSBP","Non-polynomial FSBP gains efficiency with open Gaussian","Open nodes cut one point per element in FSBP"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000217,"raw_usage":{"total_tokens":1265,"prompt_tokens":728,"completion_tokens":537,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":462}},"tokens_in":472,"tokens_out":537,"duration_ms":29577,"temperature":1.0,"reasoning_tokens":462,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T05:49:16.131917+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On the Bessel space spanned by J_0,…,J_9 on [0,25], independently recompute dim((F^2)′) and test the published 14-node rule for exactness on an L2-orthogonal basis of that space; if the true dimension is 48 (as a related study reported) rather than 27, or if any of the 27 exactness conditions fail by more than quadrature roundoff, the claimed 14-node exact open rule—and with it the one-fewer-node advantage in that example—collapses.","supporting_citations":[],"review_version":1}