{"id":"9707547f-7be2-431e-b0d1-024f27728f6a","arxiv_id":"2506.03025","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Introduces triangular histopolation and histopolation-regression schemes using Padua, approximate Fekete, and discrete Leja triangle selections, with theory and numerical comparisons.","lead":"This paper develops methods to reconstruct a function from its average values over triangles, using Padua, Fekete, and Leja triangle selection. The authors prove that the Padua-triangle approach is well-posed on fine regular meshes and show numerically that it converges on those meshes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proof of Props. 2 and 5 claims histopolation is equivalent to interpolation at mean-value points; this equivalence is false, so the logarithmic Lebesgue bound is not established as written.","rationale":"The reader's weakest assumption concerns the mesh regularity needed for the injective attribution map. That is a real limitation of the method's applicability, and the numerical experiments in Section 4.3 confirm it. However, the more load-bearing issue is in the proof of the central theoretical claim. Both Proposition 2 and Proposition 5 use the mean value theorem to reinterpret histopolation conditions as nodal interpolation conditions at points depending on f. This reinterpretation is not valid: the integral condition equates averages, not point values, and the MVT points for f and for the histopolant are generally different. The proof therefore does not establish unisolvence or the logarithmic Lebesgue bound. The claim may be salvageable by a correct perturbation argument for linear functionals, using a polynomial Markov inequality to show that each averaging functional is close to the corresponding Padua point evaluation in the dual norm. Such an argument is absent from the paper. Because the gap is in the proof rather than in a demonstrated counterexample to the theorem, the appropriate disposition remains a conditional acceptance with a required proof revision, matching the reader's overall verdict. The agreement is 'disagree' because the reader's identified weakest assumption is not the one that most threatens the central claim.","tokens_in":19420,"tokens_out":12609,"duration_ms":145398,"concrete_test":"Choose two disjoint subintervals of [-1,1], say t1=[-1,0], t2=[0,1], and m=2 (so M=3 with one extra interval, or use a bivariate example). Let f(x)=x^2. The averages are 1/3 on each interval, and MVT points are p_f,1=-1/sqrt(3), p_f,2=1/sqrt(3). The quadratic q(x)=a x^2 + b x + c satisfying both integral conditions is any polynomial with a+3c=1 and a/3+c=1/3? Solving gives a=1, c=0, b arbitrary, so q can be x^2 + b x. For b=1, q(p_f,1)=1/3 - 1/sqrt(3) != f(p_f,1). Thus q has the same averages as f but does not interpolate f at the MVT points. This directly disproves the equivalence step in Proposition 2. Then re-derive Proposition 2 using a dual-norm perturbation bound for the averaging functionals to see whether the theorem survives.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 2's proof asserts: by the mean value theorem, the histopolation condition on a triangle t is equivalent to an interpolation condition at a point p in t that depends on f. This is incorrect. For the histopolant q, the MVT gives mu_i(q)=q(p_q) with p_q in t, while mu_i(f)=f(p_f) with p_f in t; the equality mu_i(q)=mu_i(f) does not imply q(p_f)=f(p_f). The points p_q and p_f generally differ. Example: take t=[-1,1]^2, m=2, f(x,y)=x^2+y^2. The average of f over t is 2/3, so any mean-value point p_f has norm sqrt(2/3). The constant polynomial q=2/3 has the same average, so it satisfies the histopolation condition, but q(p_f)=2/3 is not equal to f(p_f)=2/3 for every such p_f? In fact it is equal for this particular q because q is constant. A better counterexample: q(x,y)=2/3 is indeed equal to the average everywhere, but the histopolation condition does not force q to agree with f at the MVT point of f; for q constant and f nonconstant, agreement at p_f would require f(p_f)=2/3, which holds for all p_f with norm sqrt(2/3). So this specific pair does not falsify the equivalence. Instead take m=1 on an interval to see the flaw: t=[-1,1], f(x)=x^2, average 1/3, MVT point p_f=1/sqrt(3). The linear histopolant q satisfying the average condition is q(x)=1/3 (if only this one condition is imposed), and q(p_f)=1/3 = f(p_f) by construction. To get a true counterexample, one needs m>=2 and at least two functionals so that the histopolant is not constant. For instance, on two disjoint triangles in 1D, m=2 with three functionals? More simply, the logical claim 'integral equality implies agreement at the MVT point of f' is false in general: for any nonconstant q with the same integral as f, the MVT point of q differs. Since the proof uses the MVT point of f as the perturbed interpolation node, the reduction to nodal interpolation fails.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops polynomial histopolation and combined histopolation-regression on triangulated planar domains. It selects M triangles from a triangulation by association with Padua points, approximate Fekete selection, or discrete Leja sequences, and solves for a bivariate polynomial of total degree m whose averages over the selected triangles match given cell averages, optionally with a least-squares regression on the remaining triangles. The main theoretical results are Proposition 1 (injectivity of the Padua attribution map under a mesh-size condition), Proposition 2 (unisolvence of Padua triangles for histopolation under a stronger mesh-size condition), and Proposition 5 (a logarithmic-squared Lebesgue constant bound). Numerical experiments compare the three triangle-selection strategies on Friedrichs-Keller and random meshes.","tokens_in":19820,"tokens_out":14605,"duration_ms":165883,"significance":"If the theoretical results were established, Proposition 5 would show that histopolation from cell averages on Padua triangles is as well conditioned as Padua interpolation, and the combination with regression would give a practical scheme for average data. The paper contributes useful algorithms (Algorithms 1-4) and a careful numerical comparison, including the observation that Fekete and Leja triangles remain viable on irregular meshes where the Padua attribution map frequently fails. The numerical evidence in Figures 8-13 is informative. However, the central perturbation argument is not currently valid, and the sufficient condition in Proposition 2 is far too restrictive to cover the experiments; the theoretical claims therefore need substantial revision before the paper can be accepted.","major_comments":[{"comment":"The proofs of Propositions 2 and 5 rest on the claim that a histopolation condition on a triangle t is equivalent to an interpolation condition at a point p in t that depends only on f. This is not correct: the mean value theorem gives mu_t(f)=f(p_f) and mu_t(q)=q(p_q) with, in general, p_f different from p_q; the equality mu_t(q)=mu_t(f) does not imply q(p_f)=f(p_f). A one-dimensional instance suffices to see the flaw: on t=[0,1], take f(x)=x and q(x)=6(x-1/2)^2. Both have average 1/2; the unique mean-value point for f is p_f=1/2, and q(p_f)=0, while f(p_f)=1/2. Thus the histopolation operator is not a nodal interpolation operator at f-dependent points, and the cited perturbation result [35, Proposition 1] is not applicable as written. The same gap invalidates the proof of the bound L_m(T_Pad_M) <= K/(1-alpha) (ln m)^2 in Proposition 5.","section":"§2.3 and §2.6, Propositions 2 and 5"},{"comment":"The sufficient condition (9) is not satisfied by the Friedrichs-Keller experiments reported in Figures 4, 6, and 8-11. For T_reg_N one has h_max=4/sqrt(N)=2*sqrt(2)/n, while the largest admissible degree in (7) grows like m ~ (pi/2) sqrt(n); hence h_max is of order n^{-1}, whereas the right-hand side alpha/(2K(m ln m)^2) is of order n^{-1}(ln n)^{-2}. The actual h_max therefore exceeds the Proposition 2 threshold by a factor of order (ln n)^2, so Propositions 2 and 5 do not explain the observed convergence and Lebesgue constants. The authors should either prove a less restrictive sufficient condition or explicitly state that the numerical results lie outside the scope of the stated sufficient condition.","section":"§2.3, Eq. (9), and §4.1"},{"comment":"In the proof of Proposition 3, the vector b has N entries, not D entries, so the inequality ||b(f)||_1 <= D ||f||_infinity is false when N > D. It should read ||b(f)||_1 <= N ||f||_infinity, and the first term in the definition of eta_d should be N rather than D. Since the numerical bound in Figure 3 is based on these quantities, the displayed formulas need correction and the figure should be recomputed if the code used D. As stated, Eq. (19) is not a valid upper bound for the operator norm of the histopolation-regression operator.","section":"§2.5, Proposition 3 and Eq. (19)"}],"minor_comments":[{"comment":"The sentence 'we must guarantee that the distance between two Padua points is less than h_max' should read 'greater than h_max'; otherwise it contradicts the displayed inequality that follows.","section":"§2.2, proof of Proposition 1"},{"comment":"The legend 'Padova triangles' should be 'Padua triangles' to match the terminology used in the text.","section":"Figures 8-11"},{"comment":"The text reports that about 30% of attempts failed to produce a Padua histopolator, but Figure 13 plots errors for the Padua method; the authors should state how unsuccessful runs are treated, for instance whether the plotted errors are conditional means over successful realizations.","section":"§4.3.1"},{"comment":"The numerical evaluation relies on a Gaussian quadrature rule for the triangle averages, but no order or implementation detail is given; specifying it would improve reproducibility.","section":"Algorithm 4 and Section 4"},{"comment":"The relation zeta_d + eta_d approximately n^{3.64} is an empirical fit to computed values, not a proved estimate; the text should label it as such.","section":"Figure 3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of the journal and the numerical study is useful. The main obstacle is the invalid mean-value perturbation argument in the proofs of Propositions 2 and 5; if the authors replace it with a correct direct estimate comparing the averaging functional with the Padua point evaluation, the paper could become acceptable. I do not see evidence of duplicate publication or inappropriate citation practice."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a useful paper for anyone working on reconstruction from cell-averaged data. The Padua-triangle construction is new, the combined histopolation-regression framework is a natural extension of the authors' segmental work, and the comparison with Fekete/Leja triangles is informative. The experiments are honest: they show convergence, but also report that Padua triangles often fail on random meshes.\n\nThe weak point is the theory. Proposition 2's proof says the histopolation condition is equivalent to interpolation at an unknown mean-value point. That is not right. The MVT gives separate points for the function and the polynomial, and equality of averages does not imply equality at a common point. The 'perturbed node' depends on f, so the stability result from [35] does not apply. The same gap propagates to Proposition 5, so the logarithmic Lebesgue bound is not established as written. I suspect the result is repairable with a direct estimate of the averaging functional as a perturbation of point evaluation, but the current proof is not that.\n\nThere is also a mismatch between the sufficient condition h_max = O((m ln m)^-2) and the mesh sizes in the experiments, so the numerical convergence is not covered by the theorems. Proposition 4 is referenced to an unpublished preprint, and the numerics would benefit from code and error bars. These are secondary.\n\nWho should read it: numerical analysts working on histopolation, cell-average data, or mesh-based approximation. It deserves a serious referee, but revision should be major: fix the proof of Proposition 2 and restate what is actually proven, or restrict the claims. I would not cite the main theorem in its current form.","headline":"A practical histopolation toolbox, but the proof of the central theorem rests on a false equivalence between histopolation and mean-value interpolation.","tokens_in":20438,"tokens_out":6512,"would_cite":false,"duration_ms":72487,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["33F05","41A05","41A10"],"pacs":[],"model":"deepseek-v4-flash","headline":"On sufficiently fine regular meshes, reconstructing a bivariate function from triangle averages is as stable as interpolating at Padua points, with a Lebesgue constant bounded by a constant times $(\\ln m)^2$.","keywords":["polynomial histopolation","histopolation-regression","Padua points","Padua triangles","Lebesgue constant","discrete Leja sequences","approximate Fekete points","triangulation"],"falsifier":"Compute $L_m(T_M^{\\mathrm{Pad}})$ numerically for regular Delaunay meshes with $n$ satisfying (7) and growing $m$: if $L_m(T_M^{\\mathrm{Pad}})/(\\ln m)^2$ is unbounded, the claimed $K(1-\\alpha)^{-1}(\\ln m)^2$ bound is false as stated. A direct check of Proposition 1 is to run Algorithm 1 on a random mesh of about 2000 triangles with $m=7$; the paper itself reports that only 25 of the required 36 triangles are found, so the histopolator is undefined.","tokens_in":19212,"feed_emoji":"📐","tokens_out":11508,"duration_ms":105044,"temperature":0.7,"pith_summary":"The paper asks whether a polynomial can be recovered from cell averages over a triangulation, when only the average of the function on each triangle is known. It proposes selecting a subset of triangles so that the equal-average conditions are both solvable and well-conditioned, and shows that on the so-called Padua triangles this holds under a mesh-fineness condition: the histopolation problem is unisolvent and its Lebesgue constant grows at most like the square of a logarithm. The same construction is combined with a least-squares regression on the remaining triangles, and compared with two greedy alternatives, approximate Fekete and discrete Leja triangles. The significance is that average-based reconstruction, which is the natural data format in many applications, can be made nearly as stable as point interpolation at the well-known Padua nodes.","feed_headline":"Histopolation on Padua triangles is as stable as Padua interpolation","feed_subtitle":"On fine regular meshes, fitting a polynomial to triangle averages is nearly as stable as sampling at Padua points.","key_machinery":"The load-bearing object is the Padua triangle set $T_M^{\\mathrm{Pad}}$, obtained by assigning each of the $M$ Padua points to the first triangle of the triangulation that contains it. Its injectivity is guaranteed by Proposition 1 via an explicit comparison of $h_{\\max}$ with the minimal Padua-point distance. The argument then runs through the mean value theorem, which rewrites every histopolation condition $\\frac{1}{|t_i|}\\int_{t_i} p\\,dxdy = \\mu_i(f)$ as $p(\\xi_i)=f(\\xi_i)$ for an unknown $\\xi_i\\in t_i$; with $\\|x^{\\mathrm{Pad}}_i-\\xi_i\\|_2 \\le h_{\\max}$, norming-set stability for perturbed nodes converts the Padua points' near-optimal conditioning into unisolvence and the logarithmic-squared Lebesgue bound. For the regression variant, the additional machinery is the constrained least-squares system (13) with Lagrange multipliers, whose direct elimination gives the operator-norm bound in Proposition 3. The greedy alternatives replace the geometric attribution by pivoted QR (approximate Fekete triangles) or pivoted LU (discrete Leja triangles) on the collocation matrix $W_{i,j}=\\mu_i(p_j)$.","core_discovery":"The central claim is that bivariate polynomial histopolation can be built on triangles that contain the Padua points, provided the mesh is fine enough for the attribution map from Padua points to triangles to be injective. Under the explicit condition on the maximal edge length $h_{\\max}$ of Proposition 1, the set of Padua triangles has exactly $M=(m+1)(m+2)/2$ elements and is unisolvent for histopolation in $\\mathbb{P}_m(\\mathbb{R}^2)$ (Proposition 2). The key mechanism is the mean value theorem: each average condition over a triangle is equivalent to a point evaluation at some unknown node inside that triangle, so histopolation becomes a perturbation of interpolation at the Padua points. Because the perturbation is controlled by $h_{\\max}$, stability results for perturbed norming sets give a Lebesgue constant bound $L_m(T_M^{\\mathrm{Pad}}) \\le K(1-\\alpha)^{-1}(\\ln m)^2$ (Proposition 5), matching the Padua points up to the factor $(1-\\alpha)^{-1}$. The paper also shows numerically that the Padua-triangle histopolator converges for all three test functions, while the Fekete and Leja greedy selections suffer conditioning peaks on the same meshes.","pith_inferences":["A practical cost not emphasized in the paper: on the regular Delaunay mesh used in the experiments, the injectivity condition (7) forces the triangle count to grow roughly like $m^4$ for degree $m$, so the near-optimal conditioning is purchased with a fine mesh; applications needing high degree on coarse meshes would have to rely on the Fekete and Leja fallback.","The same mean-value perturbation argument should work for any point set whose Lebesgue constant grows slowly, not only Padua points; testing other low-Lebesgue node sets on the same triangulations would isolate whether the logarithmic-squared bound is a property of the node set or of the histopolation mechanism.","The 30% failure rate on randomized structured meshes suggests an adaptive strategy: try Padua attribution first, and when it fails, switch to a greedy triangle extraction; the paper's experiments indicate such a hybrid would keep stability on regular regions and robustness on irregular ones."],"forward_implications":["If the bound in Proposition 5 holds, triangle-average data on fine regular meshes can be used with the same confidence as point values at Padua nodes, including the logarithmic-squared stability that prevents Runge-type blow-up.","The histopolation-regression operator is a linear projector, so the Lebesgue inequality applies and the extra non-selected triangle averages can only improve the approximation constant; the numerics show faster convergence for smooth functions and removal of instability peaks for rough ones.","For irregular meshes the Fekete and Leja greedy extractions remain well defined where Padua attribution fails, but their Lebesgue constants show isolated peaks and their histopolators do not converge for all test functions, so they are fallback strategies rather than equivalent replacements.","The construction is not tied to squares: the same framework extends to general polygonal domains for Fekete and Leja triangles, and the Padua argument carries over to other low-Lebesgue point sets and to higher-dimensional simplices, as the paper's Remark 5 states."],"supporting_citations":[{"why":"Establishes unisolvence of the Padua points and the Lebesgue bound $L_m\\le K(\\ln m)^2$ that Propositions 2 and 5 inherit.","marker":"[5]"},{"why":"Supplies the norming-set stability threshold for perturbed nodes that turns the mean-value perturbation into unisolvence and the $(1-\\alpha)^{-1}$ factor.","marker":"[35]"},{"why":"Provides the abstract Lebesgue-constant characterization for averaging functionals used to identify the operator norm with $L_m(T_M)$.","marker":"[1]"},{"why":"Source from which the identity $\\|\\Pi_m\\|_{\\mathrm{op}}=L_m(T_M)$ for disjoint triangles is derived.","marker":"[12]"},{"why":"Defines the Fekete-segment selection problem in one dimension that the paper extends to Fekete triangles.","marker":"[14]"},{"why":"Supplies the column-pivoted QR greedy algorithm used to extract approximate Fekete triangles.","marker":"[6]"},{"why":"Supplies the greedy Leja-sequence construction adapted to discrete Leja triangles.","marker":"[18]"}],"fun_headline_variants":["Histopolation on Padua triangles rivals interpolation stability","Fine meshes make Padua-triangle histopolation as stable as interpolation","Padua-triangle histopolation matches interpolation stability on fine meshes","Padua-triangle histopolation: stability like interpolation on fine meshes","Padua-triangle histopolation rivals interpolation stability"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method needs a mesh so fine that every Padua point lies in its own private triangle, and that fails on many irregular meshes: about 30% of randomized structured attempts and almost all completely random Delaunay attempts in the paper's tests.","fun_headline_variants_meta":{"raw":{"variants":["Histopolation on Padua triangles rivals interpolation stability","Fine meshes make Padua-triangle histopolation as stable as interpolation","Padua-triangle histopolation matches interpolation stability on fine meshes","Padua-triangle histopolation: stability like interpolation on fine meshes","Padua-triangle histopolation rivals interpolation stability"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000912,"raw_usage":{"total_tokens":3945,"prompt_tokens":997,"completion_tokens":2948,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":2858}},"tokens_in":613,"tokens_out":2948,"duration_ms":20297,"temperature":1.0,"reasoning_tokens":2858,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:11:10.372205+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $L_m(T_M^{\\mathrm{Pad}})$ numerically for regular Delaunay meshes with $n$ satisfying (7) and growing $m$: if $L_m(T_M^{\\mathrm{Pad}})/(\\ln m)^2$ is unbounded, the claimed $K(1-\\alpha)^{-1}(\\ln m)^2$ bound is false as stated. A direct check of Proposition 1 is to run Algorithm 1 on a random mesh of about 2000 triangles with $m=7$; the paper itself reports that only 25 of the required 36 triangles are found, so the histopolator is undefined.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes unisolvence of the Padua points and the Lebesgue bound $L_m\\le K(\\ln m)^2$ that Propositions 2 and 5 inherit."},{"cited_title":"Dolomite s Res","cited_arxiv_id":null,"evidence_quote":"Supplies the norming-set stability threshold for perturbed nodes that turns the mean-value perturbation into unisolvence and the $(1-\\alpha)^{-1}$ factor."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the abstract Lebesgue-constant characterization for averaging functionals used to identify the operator norm with $L_m(T_M)$."},{"cited_title":"preprint (2024)","cited_arxiv_id":null,"evidence_quote":"Source from which the identity $\\|\\Pi_m\\|_{\\mathrm{op}}=L_m(T_M)$ for disjoint triangles is derived."},{"cited_title":"BIT Numer","cited_arxiv_id":null,"evidence_quote":"Defines the Fekete-segment selection problem in one dimension that the paper extends to Fekete triangles."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the column-pivoted QR greedy algorithm used to extract approximate Fekete triangles."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the greedy Leja-sequence construction adapted to discrete Leja triangles."}],"review_version":1}