{"id":"2e2526e6-bfc4-4e4c-93ba-b3868d0764af","arxiv_id":"2607.25111","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Entropy stability of Active Flux-type schemes reduces to an entropy inequality on cell averages alone; a Tadmor-style blend yields a bound-preserving, non-oscillatory, entropy-diminishing scheme validated on KPP.","lead":"Active Flux schemes only need entropy control on cell averages, not point values. The authors build a bound-preserving, non-oscillatory, entropy-stable scheme and show it captures the correct KPP entropy solution.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"The per-edge entropy inequality (16) is guaranteed only by the exact cap ℓ=min(1,A/B); the limiter actually proposed, φ_α in (17), is provably below 1/B, not below A/B, so hypothesis (12) of Proposition 3.1 is not established for the scheme as written.","rationale":"The reader correctly identified the semi-discrete-only limitation and the unproved accuracy of the blend, and rated soundness 7 with the authors owning their gaps. My concern is adjacent but distinct and more specific: it is not that accuracy is unproven, but that the entropy inequality itself — the hypothesis of the paper's central proposition — is only established for the exact cap ℓ=min(1,A/B), while the advocated practical limiter (17) as written caps against the wrong hyperbola (1/B instead of A/B). The α≥1/4 condition in the text is exactly the discriminant condition for 1−αx ≤ 1/x, which strongly suggests the authors may have conflated the normalized (B/A) and unnormalized (B) arguments; if the code uses φ_α(B/A), the concern evaporates entirely and the construction is sound. This is why I do not move the verdict: the reader's CONDITIONAL already prices in unproven elements of the blend, and this gap has a trivially checkable, likely benign resolution. Credit where due: the structural insight (tested-identity rewriting in §2.2, so that point-value contributions carry an O(h) test-function factor and vanish under Lemma 2.6) is genuinely clean, the Tadmor per-edge reduction is standard and correct, and the KPP experiment with entropy-on-averages-only is real independent evidence. The uniform-L∞ hypothesis on point values vs. the observed ±25–39 overshoots is a lesser tension worth noting but not load-bearing, since per-run boundedness suffices formally. Net: keep CONDITIONAL; require the limiter normalization check and per-edge Θ audit before full acceptance.","tokens_in":16072,"tokens_out":6088,"duration_ms":55198,"concrete_test":"Two-part check: (i) inspect/replicate the implementation to determine whether (17) is evaluated as φ_α(B/A) (provably safe, by the α≥1/4 quadratic discriminant) or as φ_α(B) (not provably safe). (ii) In the KPP run with the Tadmor-only limiter, instrument the code to log, per edge per RK stage, Θ_e = A_e − ℓ_e B_e and the aggregate production Σ_e |e| min(Θ_e, 0). If min Θ_e is materially negative with φ_α(B) but the exact cap ℓ=min(1,A/B) yields the same KPP figures, the claim survives with a corrected formula; if the exact cap changes the results, the numerical evidence must be redone.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central structural claim (Prop. 3.1: entropy control of averages alone suffices) is conditional on the per-edge semi-discrete inequality (12), which §3.2 reduces to the scalar edge condition Θ(ℓ_e)=A−ℓ_e B ≥ 0 (16), with A = ⟨Δv, f̂LO⟩−Δψ·n ≥ 0 (Rusanov) and B = −⟨Δv, f̂HO−f̂LO⟩. The provably safe choice is ℓ=min(1, A/B) for B>0. But the paper then says \"in practice\" to use φ_α = 1 (B≤0), 1−αB (B>0), α≥1/4. The algebra behind α≥1/4 shows 1−αB ≤ 1/B for all B>0 (since αB²−B+1 ≥ 0 ⇔ α≥1/4) — i.e., φ_α stays below the hyperbola 1/B. The safe hyperbola, however, is A/B, and A/B < 1/B whenever A<1; A is an unnormalized, problem-dependent quantity (for Euler, A scales with states and mesh), and can be arbitrarily small in near-smooth regions. Example: A=0.1, B=0.9 gives Θ = A−(1−B/2)B = 0.1−0.495 < 0 — local entropy production on that edge, contradicting (12). Hence the scheme as written in (17) is not shown to satisfy the hypothesis that makes Prop. 3.1 applicable; the \"entropy satisfying\" property of the implemented method is empirical (KPP) rather than proved. This is repairable (use the exact cap, or φ_α(B/A), which the same discriminant argument shows is ≤ A/B), and the text is ambiguous enough that the code may do the safe thing — but as stated, the load-bearing hypothesis is unverified for the practical limiter. Secondary: the Fig. 3 experiment central to \"averages only\" reports point values u_σ∈[−25.6, 39.3]; Prop. 2.4/3.1 assume a uniform L∞ bound on u_σ, so that experiment sits at the edge of the theory's hypotheses.","agreement_with_reader":"partial"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The paper analyses entropy stability of Active Flux-type (PAMPA) schemes, which evolve both cell averages and point values on element boundaries. After a truncation-error analysis and a Lax–Wendroff-type theorem with consistency assumptions refined relative to earlier work (the point-value residual condition (11) is deliberately weak), the authors prove their central structural result (Proposition 3.1): if the semi-discrete scheme for the cell averages satisfies a per-edge entropy inequality (12), then any L²-limit satisfies the entropy inequality, with no condition on the point values beyond boundedness. They then construct a blended low/high-order flux for the average update in Tadmor's framework, reducing the per-edge condition to a scalar inequality A−ℓB≥0 (16), and combine this with previous bound-preserving and oscillation-elimination limiters. The method is demonstrated on the Kurganov–Petrova–Popov test case, where the entropy correction is shown numerically to select the correct entropy solution even when the point values are left unlimited. The authors state openly that the entropy property is semi-discrete only and that compatibility with formal high-order accuracy is not proved.","tokens_in":16540,"tokens_out":9427,"duration_ms":84077,"significance":"If the limiter issue in §3.2 is repaired, this is a useful and cleanly argued contribution: it reduces entropy stability of Active Flux-type schemes to a property of the cell-average update alone (Prop. 3.1), refines the consistency assumptions relative to [1], and delivers a monolithic BP + oscillation-elimination + entropy construction demonstrated on the notoriously entropy-sensitive KPP problem, where the corrected scheme visibly selects the right solution while remaining much sharper than Rusanov. The entropy construction is derived from standard convex-entropy calculus rather than fitted, and the falsifiable per-edge condition (16) is a concrete, checkable design criterion. The honest statement of open issues (semi-discrete only; accuracy not yet proved) is a strength, though the absence of any convergence study weakens the high-order claim.","major_comments":[{"comment":"§3.2, Eqs. (16)–(17): Prop. 3.1 applies only if the per-edge condition (12) holds, which §3.2 reduces to Θ(ℓ)=A−ℓB≥0 with A=⟨Δv,f̂LO⟩−Δψ·n≥0 (Rusanov) and B=−⟨Δv,f̂HO−f̂LO⟩. The provably safe cap is ℓ≤A/B. The limiter actually proposed, φ_α=1−αB for B>0 with α≥1/4, is only shown to lie below 1/B (discriminant αB²−B+1≥0). But A is unnormalized and can be arbitrarily small (A≈(α_e/2)|Δv|² in weakly varying regions), so φ_α can exceed A/B: e.g. A=0.1, B=0.9, α=1/2 gives Θ=0.1−0.495<0. Hence hypothesis (12) is not established for the scheme as written in (17); the entropy property of the implemented method is empirical. The fix looks easy (exact cap, or ℓ=φ_α(B/A), for which the same discriminant argument gives φ_α(B/A)≤A/B). Please state precisely what is proved for the limiter actually used, and re-run the tests with a provably safe variant.","section":"§3.2, Eqs. (16)-(17)"},{"comment":"The paper claims a high-order method, but the entropy blend's effect on formal accuracy is supported neither analytically (acknowledged in §5) nor numerically: the only test is KPP, with no convergence or EOC study on a smooth problem. In smooth regions B=−⟨Δv,f̂HO−f̂LO⟩ should be O(h³) (Δv=O(h), flux discrepancy O(h²)), so φ_α=1−αB deactivates at a rate plausibly compatible with the O(h³) truncation target of §2.2. This should at least be checked numerically (smooth Euler or advection test with EOC table); without it, the 'high order' character of the entropy-corrected scheme is unverified.","section":"§4-5, accuracy of the blended scheme"},{"comment":"The entropy inequality is proved only for the semi-discrete scheme (Tadmor's framework), while the computations use SSP Runge–Kutta in time. Lemmas 3.2–3.3 show the first-order building blocks do admit fully discrete entropy inequalities under CFL, so the gap is not intrinsic to the fluxes but to the blending construction. Given that the title and abstract advertise 'entropy satisfying' methods, the paper should either provide a fully discrete statement (e.g. via a convex-limiting interpretation of the blend under a CFL condition) or explicitly discuss the status and practical relevance of the semi-discrete-only guarantee.","section":"§3.2 vs. time discretisation (§2.1, §5)"},{"comment":"Prop. 3.1 inherits hypothesis 4 of Prop. 2.4: a uniform L∞ bound on the point values u_σ. Yet the experiment central to the headline claim that 'only the averages matter' (Fig. 3, entropy limiter on averages, unlimited high-order point update) reports u_σ∈[−25.56,39.29]. On a fixed mesh this is harmless, but the uncontrolled point values are exactly what the theorem needs bounded as h→0; moreover wild point values feed f̂HO and can inflate B, aggravating the issue in comment 1. Please discuss the status of this hypothesis; a mesh-refinement study showing that the averages converge to the KPP entropy solution would substantiate the claim.","section":"Prop. 3.1, hypothesis 4 of Prop. 2.4; Fig. 3"}],"minor_comments":[{"comment":"The role of ℓ_e is inconsistent: in (3c) ℓ_e multiplies the high-order flux, while in (14) (1−ℓ_e) multiplies ∫f(u)·n and ℓ_e multiplies f̂LO; (16) then treats ℓ as the high-order weight. Please harmonise.","section":"§2.1 (3c) vs. §3.2 (14)"},{"comment":"Eq. (1): the entropy inequality should read ∂η/∂t + div g(u) ≤ 0 (div is missing).","section":"§1, Eq. (1)-(2)"},{"comment":"Lemma 3.3: the update is written 'u^{n+1}_K = u^{n+1}_K − ...'; the right-hand side should be u^n_K.","section":"§3.1, Lemma 3.3"},{"comment":"Lemma 3.2, item 2: the CFL is stated as Δt/|K| ≤ 1/6, but the proof uses 6αΔt/|K| ≤ 1; the statement should include the wave-speed factor.","section":"§3.1, Lemma 3.2"},{"comment":"Eq. (13): index mismatch n^{T^K_j} vs. T^K_i; also η(u)^{T^K_i} is defined with a sum over σ_j but written σ_l.","section":"§3.1, Eq. (13)"},{"comment":"Figure 2 is load-bearing for the choice (17) but its dotted curves are never identified: is the hyperbola 1/B or A/B (and for which A)? Labelling it would make the gap in comment 1 explicit to readers.","section":"§3.2, Figure 2"},{"comment":"Table 1: several entries read '11.' with a trailing period; please clean the formatting.","section":"§4, Table 1"},{"comment":"Footnote 2 (wrong global extrema in the published [3]) is important for interpreting the BP-only KPP results; it deserves to be in the main text with a precise statement of what was mis-set.","section":"§4, footnote 2"},{"comment":"Typos: 'consistant' (Prop. 2.4), 'arewriting' (§2.2), 'sens of finite element' (§2.1), 'from bellow' (Prop. 2.4), 'thermo-dynamically consistant' (§1), 'Rusanov soluton' (Fig. 6 caption).","section":"throughout"},{"comment":"Since the scheme evolves η(u_σ) through v(u_σ)^TωΦ in Prop. 3.1's proof, a remark clarifying that boundedness of v(u_σ) follows from hypothesis 4 and strict convexity of η would help the reader.","section":"§3, proof of Prop. 3.1"}],"recommendation":"major_revision","confidential_remarks":"The reference list is heavily self-referential: items [1]–[7] all involve the authors, and the manuscript reads as an incremental note on the authors' PAMPA/Active Flux program (they themselves call it \"this note\"). The entropy component is a genuine addition over [3,5], but the editor may wish to consider whether the paper meets the journal's bar for standalone novelty, and whether the reliance on companion papers [3,5] for key limiter definitions makes the manuscript sufficiently self-contained. I also note the v2 date and several 2026-dated references; nothing improper, but the results depend on material that has not yet appeared in final form."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful core here is structural, not the limiter stack. For this DoF split, their testing identity shows that a semi-discrete entropy inequality on cell averages alone is enough for the entropy Lax–Wendroff argument; point values drop out. That is a real, citable observation for Active Flux / PAMPA-type schemes, and Prop. 3.1 is cleanly set up from the same rewriting they use for consistency.\n\nWhat they do well: the truncation and LW sections refine the point-residual consistency assumptions relative to their earlier work, the first-order BP/entropy lemmas are standard and correctly cited, and the KPP experiment where entropy-on-averages alone already selects the right solution (with wild point values) is the right numerical check for the claim. Stacking with their existing BP and OE limiters is engineering, not novelty, but it is done monolithically and the figures are readable.\n\nSoft spots, in proportion. Entropy is only semi-discrete and they own that they have no proof the blend keeps formal order—fine, common. The real write-up issue is the practical ℓ. They correctly reduce to Θ = A − ℓB ≥ 0 and mention the safe cap min(1, A/B), then switch to φ_α = 1−αB (α=1/2), which only stays under 1/B, not A/B. When A is small, that can produce local entropy production on an edge, so the implemented scheme is not shown to satisfy the hypothesis of Prop. 3.1; KPP success is empirical. Easy fix (use A/B, or scale), but a referee should force the algebra to match the code. Secondary: Fig. 3’s point range sits outside the L∞ bound used in the LW hypotheses. Only one scalar test, no systems, no code.\n\nThis is for people already in Active Flux / residual / entropy-stable FV. Worth a serious referee; not a desk reject. I would engage after they tighten the limiter statement.","headline":"Clean structural result for Active Flux entropy (averages suffice), but the practical Tadmor limiter as written does not actually guarantee the edge inequality the proof needs.","tokens_in":17542,"tokens_out":513,"would_cite":false,"duration_ms":30683,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M08","65M12","35L65"],"pacs":[],"model":"grok-4.5","headline":"For Active Flux schemes, entropy stability only needs to be enforced on cell averages; point values can be left alone.","keywords":["Active Flux","entropy stability","Tadmor framework","cell averages","bound preserving","hyperbolic conservation laws","KPP problem","blended fluxes"],"falsifier":"Run the KPP problem with the entropy blend turned off on the averages while keeping high-order point updates: if the rotational wave structure collapses to a wrong entropy solution, the claim that averages alone control entropy fails.","tokens_in":17074,"feed_emoji":"📐","tokens_out":833,"duration_ms":14138,"temperature":0.7,"pith_summary":"Active Flux-type methods track two kinds of unknowns: cell averages and point values on element boundaries. This paper shows that, for a hyperbolic system with one entropy, a semi-discrete entropy inequality on the averages alone is enough for the limit solution to satisfy the entropy inequality; the point-value degrees of freedom drop out of the argument. The authors then build a single blended scheme that is bound-preserving, non-oscillatory, and entropy-diminishing by blending a high-order flux with a Rusanov flux on each edge, choosing the blend parameter so that Tadmor’s edge-wise entropy production stays non-negative. On the KPP problem, known to be extremely sensitive to entropy, the entropy blend alone already produces the correct rotational wave structure (with the expected wiggles), confirming that controlling averages is sufficient. The same construction extends unchanged to polygonal meshes.","feed_headline":"Entropy for Active Flux only needs cell averages","feed_subtitle":"Point values drop out of the entropy argument; a simple edge blend recovers the correct KPP solution","key_machinery":"Tadmor-style edge entropy production Θ(ℓ) for the blended numerical flux: the largest admissible blend parameter ℓ★ is taken so that Θ(ℓ) ≥ 0 on every edge, guaranteeing a semi-discrete entropy inequality for the cell averages alone.","core_discovery":"In Active Flux-type schemes the only quantity that must satisfy a semi-discrete entropy inequality is the cell average; the point-value updates play no role in the entropy Lax–Wendroff argument. Consequently an edge-wise blend between a high-order flux and a Rusanov flux, chosen so that Tadmor’s production term remains non-negative, yields a limit solution that obeys the entropy inequality.","pith_inferences":["Because point values are free of the entropy constraint, one can design them purely for accuracy or positivity while the averages carry the thermodynamic consistency.","The same average-only argument should apply to other hybrid DOF schemes (residual-distribution, certain DG–FV hybrids) that already possess a conservative cell-average update.","A fully discrete entropy proof would still be needed before the method can be certified for long-time or steady-state entropy-critical flows."],"forward_implications":["Entropy-stable Active Flux schemes can be obtained by acting only on the finite-volume update of the averages.","The same edge-blend construction carries over unchanged to polygonal and virtual-element meshes.","Bound-preserving, oscillation-control and entropy conditions can be combined into a single monolithic limiter without separate stages.","Higher-order Active Flux extensions need only an entropy-stable average update; point residuals may keep their existing high-order form."],"fun_headline_variants":["Active Flux entropy hinges only on cell averages","Point values irrelevant to Active Flux entropy proof","Edge blend fixes entropy for Active Flux on KPP","Tadmor entropy via averages alone in Active Flux","Cell-average entropy inequality suffices for Active Flux"],"cache_read_input_tokens":512,"weakest_assumption_plain":"Entropy is enforced only at the semi-discrete level; the paper does not prove that the blend preserves the formal high-order accuracy of the unblended scheme.","fun_headline_variants_meta":{"raw":{"variants":["Active Flux entropy hinges only on cell averages","Point values irrelevant to Active Flux entropy proof","Edge blend fixes entropy for Active Flux on KPP","Tadmor entropy via averages alone in Active Flux","Cell-average entropy inequality suffices for Active Flux"]},"model":"grok-4.5","effort":"low","cost_usd":0.003391,"raw_usage":{"total_tokens":1177,"prompt_tokens":809,"num_sources_used":0,"completion_tokens":58,"cost_in_usd_ticks":33908000,"prompt_tokens_details":{"text_tokens":809,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":310,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":809,"tokens_out":58,"duration_ms":6634,"temperature":1.0,"reasoning_tokens":310,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T01:35:53.216421+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Run the KPP problem with the entropy blend turned off on the averages while keeping high-order point updates: if the rotational wave structure collapses to a wrong entropy solution, the claim that averages alone control entropy fails.","supporting_citations":[],"review_version":2}