{"id":"42ab747b-a5c5-4648-805d-b9519f571d45","arxiv_id":"1908.08158","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The global-best H(div) approximation error with divergence and boundary constraints is equivalent, up to a generic constant, to the sum of unconstrained local-best errors, yielding a local commuting projector and hp-optimal error estimates under minimal regularity.","lead":"For vector fields with finite divergence, the paper proves that the best global piecewise-polynomial approximation with constraints is equivalent to the sum of independent elementwise best approximations. It also builds a simple, local, commuting projector, delivering optimal hp error estimates and sharpened a priori bounds for mixed and least-squares finite element methods.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The p-robust local equivalence in Lemma A.1 rests on an unverified reading of [16, Cor. 3.4]; if that imported constant has any p-dependence, the hp-optimality in Theorem 3.6 collapses.","rationale":"The paper is well-written and the main construction is elegant. The projector is genuinely local and commuting, and I find no other internal gap in Sections 3-6: the patchwise construction, the use of equilibrated fluxes, and the application to mixed and least-squares methods are carefully argued. The reader's weakest assumption is indeed the one I would flag: Lemma A.1 makes a very strong claim on the basis of a single citation. The result of Costabel-McIntosh is deep and is not reproduced, and the authors' confidence that its constant is p-robust is plausible but unverified. Because Proposition 5.1 is built on Lemma A.1 and is the step that removes the p-logarithm, a p-dependent factor in Lemma A.1 would flow directly into Theorem 3.6. I therefore recommend CONDITIONAL rather than a blanket rejection: the proof is likely right, but the key external ingredient should be verified or made self-contained. The proposed numerical check on the minimal-norm lifting constant would be a quick falsification test; an analytical trace of [16, Cor. 3.4] would be the definitive verification.","tokens_in":24884,"tokens_out":9923,"duration_ms":102041,"concrete_test":"On a fixed shape-regular reference simplex K, compute for p=1,...,30 the quantity C_p = sup_{g in P_p(K), ||g||_{H^{-1}(K)}=1} inf_{v in RTN_p(K), div v = g} ||v||_{L^2(K)}, where ||g||_{H^{-1}} is the dual norm against H^1_0(K). This is the sharp stability constant underlying (A.2). If C_p is not uniformly bounded as p grows, Lemma A.1's p-independence is false, so Proposition 5.1 and Theorem 3.6 must be weakened. Alternatively, trace the proof of [16, Cor. 3.4] to extract its explicit dependence on p; either check settles whether the headline claim stands.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Appendix A's Lemma A.1 (Eq. A.1) is the hinge that converts divergence-constrained local minimizers into unconstrained ones with a constant independent of p. Its proof uses only the sentence 'It follows from [16, Corollary 3.4] that there exists v_K in RTN_p(K)...' to assert the existence of a polynomial lifting whose stability constant in (A.2) is C(d,kappa_K). The authors do not quote the corollary, and p-independence is not established in the text. This matters because Lemma A.1 is used in Lemma 4.6 (bound on sigma_a) and in Proposition 5.1 (via Lemma 5.3 and the localized estimate), and the latter is exactly what removes the p-logarithm in Theorem 3.6. If [16, Cor. 3.4] carries a hidden factor gamma(p) of any polynomial growth, then e_{K,p}(v) in Lemma A.1 and the right-hand side of (3.12) inherit gamma(p), and the bound (3.13) gains an extra p^alpha factor, destroying the claimed h^{min(s,p+1)}/(p+1)^s rate. The assumption is also not supported by the paper's own numerical or machine-checked evidence. Thus the central p-optimality claim is conditional on the correct p-robust reading of [16, Cor. 3.4].","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper establishes that, for arbitrary vector fields in H0,Gamma_N(div,Omega), the error of the global-best approximation by Raviart-Thomas-Nedelec finite element functions satisfying divergence and boundary constraints is equivalent, up to a generic constant, to the sum of elementwise local-best approximation errors without those constraints. The proof is constructive: the authors define a locally defined, commuting projector (Definition 3.1) built from elementwise local-best approximations and patchwise equilibrated flux reconstructions, prove its stability and approximation properties (Theorem 3.2), and derive the local-global equivalence (Theorem 3.3). They then prove an unbalanced, polynomial-degree-robust variant (Proposition 5.1) which, combined with the equivalence result, yields hp-optimal approximation rates with no logarithmic factors under only elementwise H^s regularity (Theorem 3.6). The last section applies these results to mixed and least-squares finite element methods for a model diffusion problem. The proofs are detailed and organized; the main external input is a locally constrained-unconstrained equivalence on a simplex (Lemma A.1) imported from Costabel-McIntosh [16, Corollary 3.4].","tokens_in":25172,"tokens_out":5151,"duration_ms":54451,"significance":"If the results hold, they settle a long-standing gap in the literature by providing a stable, local, commuting projector under minimal H(div) regularity, together with fully optimal hp-approximation estimates that require only elementwise minimal Sobolev regularity. The local-global equivalence in H(div) extends earlier scalar results of Veeser and others, and the applications to mixed and least-squares methods are immediate and useful. The construction based on equilibrated flux reconstructions is simple and elegant, and the proof structure makes the dependence of each constant transparent. The main achievement is the p-robust removal of the divergence constraint in Lemma A.1; this is also the only point where the manuscript relies on a deep external result without stating its precise content.","major_comments":[{"comment":"The proof of Lemma A.1, which is the load-bearing p-robust constrained-unconstrained equivalence on a simplex, rests entirely on the sentence 'It follows from [16, Corollary 3.4] that there exists v_K in RTN_p(K) ...' without quoting the corollary or explaining how the constant in (A.2) is independent of the polynomial degree. This lemma is used in the proof of Lemma 4.6 (final step of the proof of (4.6)) and in the proof of Proposition 5.1 (via Lemma 5.3), and it is exactly the mechanism that removes the p-logarithm in Theorem 3.6. If the constant in [16, Cor. 3.4] had any hidden dependence on p, then the bounds (A.1), (3.12), and (5.1) would inherit that dependence and the claimed rate (3.13) would lose its p-optimality. The authors should quote the precise statement of [16, Corollary 3.4], verify that it applies to the spaces RTN_p(K) rather than to the spaces of the de Rham complex treated there, and show in detail that the resulting constant depends only on d and kappa_K, or else provide a self-contained proof of Lemma A.1.","section":"Appendix A, Eq. (A.2); used in Lemma 4.6 and Proposition 5.1"},{"comment":"In the proof of Theorem 3.6 for the case p > s, the bound [e_{K,p-1}(v)]^2 <= C { [h_K^s/p^s ||v||_{H^s(K)}]^2 + delta_{s<1} [h_K/p ||div v||_K]^2 } is stated with a constant depending only on s, d, and kappa_T. Since this is the step that achieves the fully p-robust constant in (3.13), the authors should make explicit which approximation result supplies a constant independent of p for the (p-1)-degree local best approximation; the cited elementary hp-bounds usually contain constants that may depend on p, and this point is central to the advertised p-optimality.","section":"Section 5.2, Step 2"}],"minor_comments":[{"comment":"The notation L2(Omega) is introduced as L2(Omega; R^d), but scalar-valued L2 spaces also appear throughout; this overloading is clear from context but could be noted explicitly.","section":"Section 2.1"},{"comment":"The jump notation /llbracket v_T /rrbracket is introduced within the proof; defining it in Section 2 alongside the face notation would improve readability.","section":"Proof of Lemma 4.6"},{"comment":"The bound (5.8) cites [2, Lemma 4.1] for the hp-approximation estimate; since the constant is claimed to depend only on s, d, and kappa_T, a short justification of the elemental version of that lemma would be helpful.","section":"Section 5.2, Step 1"}],"recommendation":"major_revision","confidential_remarks":"The main revision point is the verification of Lemma A.1. I see no evidence of circularity or fabrication, and the overall structure of the proofs appears sound; however, the p-robustness claim of the central theorem is conditional on a correct reading of an imported external result, so I would not recommend acceptance until the authors supply the exact statement and a detailed derivation of (A.2)."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper is worth your time. It delivers the first H(div) projector I know of that is local, commuting, stable under the minimal H(div) regularity, and p-optimal, and it does so with a simple construction: elementwise local-best approximations followed by patchwise flux equilibration. The equivalence result between global-best and local-best approximations in H(div) (Theorem 3.3) is a substantive extension of Veeser's H^1 theorem, and the hp-approximation estimates in Theorem 3.6 are the natural payoff. The applications to mixed and least-squares methods are clean and correct.\n\nThe proofs are mostly detailed and honest. Using equilibrated flux reconstruction from a posteriori analysis in a pure approximation setting is a nice idea and it works. The dependence on earlier results by some of the same authors is legitimate, since those patchwise equilibration lemmas are exactly the right tools.\n\nThe one spot I cannot fully verify is Lemma A.1, the local p-robust equivalence between constrained and unconstrained best approximations. The proof rests on a single sentence that [16, Corollary 3.4] gives a polynomial lifting v_K with a constant independent of p. The corollary is not quoted, and p-independence is not argued in the text. This is the hinge for Proposition 5.1 and hence for the p-robust part of Theorem 3.6. I do not think this is a dealbreaker: the bound is in the H^{-1} norm, and for polynomial spaces the right inverse of divergence is p-robust in that norm, so the claim is plausible. But a referee should push for the exact statement or a self-contained proof. As written, it is a presentation gap, not a demonstrated error.\n\nWho should read this: anyone working on hp approximation, mixed FEM, or polynomial-degree-robust estimates. I would bring it to a reading group, I would cite it, and I would send it to a serious journal. The main revision request is to expand the proof of Lemma A.1.","headline":"A strong paper that constructs the first local, commuting, minimal-regularity H(div) projector with p-optimal approximation; the only soft spot is the too-terse import of a Costabel–McIntosh bound in Lemma A.1.","tokens_in":25737,"tokens_out":11479,"would_cite":true,"duration_ms":108880,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N30","65N15","41A10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Global and local best-approximation errors in H(div) are equivalent up to a constant, yielding hp-optimal convergence rates under minimal regularity.","keywords":["H(div) approximation","Raviart-Thomas-Nédélec elements","local-best approximation","global-best approximation","commuting projector","hp approximation estimates","flux equilibration","minimal regularity"],"falsifier":"On a fixed shape-regular simplex, compute the ratio between the divergence-constrained local-best error and the unconstrained one from Lemma A.1 for p=1 through 20 and for a collection of smooth fields; if the ratios grow without bound in p, the p-robustness the final theorem needs is false.","tokens_in":24686,"feed_emoji":"📐","tokens_out":8812,"duration_ms":77797,"temperature":0.7,"pith_summary":"The paper shows that, for an arbitrary vector field in H(div), the error of the best global approximation by H(div)-conforming piecewise polynomial Raviart-Thomas-Nédélec elements of degree p, with the divergence and boundary fluxes constrained, is comparable to the sum of independent elementwise best approximation errors with no such constraints. The comparison constant depends only on the space dimension, the shape regularity of the mesh, and the polynomial degree. The proof constructs a simple, local projector that commutes with the divergence and is stable in L2 up to an hp data-oscillation term, built by gluing elementwise local-best approximations through equilibrated flux reconstructions on vertex patches. From this equivalence, the paper derives hp-optimal approximation rates $h^{{min(s,p+1)}}$/(p+1)^s requiring only elementwise H^s regularity, with no logarithmic factors in p. It then applies these bounds to mixed and least-squares mixed finite element methods to obtain optimal a priori error estimates.","feed_headline":"Global and local H(div) approximations are equivalent up to a constant","feed_subtitle":"A local commuting projector yields hp-optimal rates with no log factors.","key_machinery":"The load-bearing object is the projector P_T^p defined in Definition 3.1: on each element, τ_T is the divergence-constrained Raviart-Thomas-Nédélec best approximation of v; over each vertex patch ω_a, an equilibrated flux σ_a is chosen by a constrained minimization (3.2); summing σ_a over all vertices yields an H(div)-conforming piecewise polynomial that commutes with the divergence. The argument that converts the elementwise minimizers into a global H(div) function is the patchwise flux equilibration bound of Lemma 4.4, and the p-robust passage from constrained to unconstrained local errors is Lemma A.1, which imports the result [16, Corollary 3.4].","core_discovery":"The central claim is Theorem 3.3: for any v in H0,Γ_N(div,Ω), the global-best approximation error E_{T,p}(v) defined in (3.10), with divergence fixed to Π_T^p(∇·v) and normal flux constrained on Γ_N, satisfies [E_{T,p}(v)]^2 ≤ C Σ_K [e_{K,p}(v)]^2 ≤ C [E_{T,p}(v)]^2, where e_{K,p}(v) is the unconstrained elementwise best error (3.11). The upper bound is obtained by taking the global competitor to be a new projector P_T^p(v) whose per-element error is controlled by local-best errors over the element and its neighbours; the lower bound is immediate from the definitions. A second, one-sided bound with p replaced by p−1 and a constant independent of p is proved using an unbalanced patchwise construction; combining the two bounds gives the hp-optimal estimate (3.13) under only elementwise H^s regularity.","pith_inferences":["The same equilibrated-flux construction should carry over to H(curl) and to tensor-valued H(div) settings, yielding analogous local commuting projectors; the paper does not treat these cases.","The one-sided p-robust bound suggests a concrete algorithmic prescription for hp-adaptive codes: compute local minimizers one degree lower than the target space to keep all constants independent of p.","A direct numerical test of Lemma A.1 on anisotropic elements would show whether the shape-regularity parameter alone controls the constant, or whether a more detailed geometric quantity enters; the paper assumes the former."],"forward_implications":["The equivalence makes global H(div) approximation error bounds reduce to elementwise computations, so the global error can be bounded by summing independent local problems on each element.","The projector P_T^p is a commuting local projection valid for all of H0,Γ_N(div,Ω), so it supplies a Fortin operator for mixed methods without any extra smoothness assumption.","For vector fields with only elementwise H^s regularity, the approximation rate is h^{min(s,p+1)}/(p+1)^s in the weighted H(div) norm, with no logarithmic factors in p and no global regularity requirements.","In mixed finite element methods, the flux error is exactly a constrained global-best error, so the new bounds give fully optimal hp a priori error estimates; least-squares mixed methods inherit the same rate for the flux plus a best-approximation term for the gradient."],"supporting_citations":[{"why":"Supplies the p-robust equivalence between divergence-constrained and unconstrained best approximation on a single simplex used in Lemma A.1, the hinge of the whole argument.","marker":"[16]"},{"why":"Gives the two-dimensional patchwise flux equilibration stability used in Lemma 4.4.","marker":"[8]"},{"why":"Gives the three-dimensional patchwise flux equilibration stability used in Lemma 4.4.","marker":"[30]"},{"why":"Provides the prior local bounded cochain projectors whose stability properties are sharpened here.","marker":"[31]"},{"why":"Establishes the H1 local-global best approximation equivalence that this paper extends to H(div).","marker":"[45]"},{"why":"Supplies the standard hp approximation bounds used to turn the local errors into rates.","marker":"[2]"},{"why":"Gives a commuting p-version projector with optimal p-approximation but requiring higher regularity, the gap this paper removes.","marker":"[36]"}],"fun_headline_variants":["Local-global H(div) equivalence up to constant, via stable projector","hp-optimal H(div) estimates without log factors from local projector","Local and global best H(div) errors equivalent up to constant","Stable commuting projector gives optimal hp rates in H(div)"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof rests on the imported fact that, on one triangle or tetrahedron, forcing the divergence to a polynomial projection costs no more than a factor depending only on the simplex shape, not on the polynomial degree; if that factor secretly grows with degree, the p-optimal rates of Theorem 3.6 collapse.","fun_headline_variants_meta":{"raw":{"variants":["Local-global H(div) equivalence up to constant, via stable projector","hp-optimal H(div) estimates without log factors from local projector","Local and global best H(div) errors equivalent up to constant","Stable commuting projector gives optimal hp rates in H(div)"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000274,"raw_usage":{"total_tokens":1666,"prompt_tokens":1001,"completion_tokens":665,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":617,"completion_tokens_details":{"reasoning_tokens":588}},"tokens_in":617,"tokens_out":665,"duration_ms":6223,"temperature":1.0,"reasoning_tokens":588,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:47:30.253669+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On a fixed shape-regular simplex, compute the ratio between the divergence-constrained local-best error and the unconstrained one from Lemma A.1 for p=1 through 20 and for a collection of smooth fields; if the ratios grow without bound in p, the p-robustness the final theorem needs is false.","supporting_citations":[{"cited_title":"On Bogovski ˘ ı and regularized Poincar´ e integral operators for de Rham complexes on Lipschitz domains","cited_arxiv_id":null,"evidence_quote":"Supplies the p-robust equivalence between divergence-constrained and unconstrained best approximation on a single simplex used in Lemma A.1, the hinge of the whole argument."},{"cited_title":"Equilibrated residual error estimates are p-robust","cited_arxiv_id":null,"evidence_quote":"Gives the two-dimensional patchwise flux equilibration stability used in Lemma 4.4."},{"cited_title":"Stable broken H 1 and H(div) polynomial extensions for polynomial- degree-robust potential and flux reconstruction in three space dimensions","cited_arxiv_id":null,"evidence_quote":"Gives the three-dimensional patchwise flux equilibration stability used in Lemma 4.4."},{"cited_title":"S., and Winther, R","cited_arxiv_id":null,"evidence_quote":"Provides the prior local bounded cochain projectors whose stability properties are sharpened here."},{"cited_title":"Approximating gradients with continuous piecewise polynomial funct ions","cited_arxiv_id":null,"evidence_quote":"Establishes the H1 local-global best approximation equivalence that this paper extends to H(div)."},{"cited_title":"The h-p version of the finite element method with quasi-uniform meshes","cited_arxiv_id":null,"evidence_quote":"Supplies the standard hp approximation bounds used to turn the local errors into rates."},{"cited_title":"M., and Rojik, C","cited_arxiv_id":null,"evidence_quote":"Gives a commuting p-version projector with optimal p-approximation but requiring higher regularity, the gap this paper removes."}],"review_version":1}