{"id":"e3dc5339-06df-420f-b598-f2a8ca1f2d9c","arxiv_id":"2501.16529","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A parameter-free, locally computed artificial viscosity restores the global entropy inequality of flux-differencing DG while preserving high order accuracy and avoiding entropy conservative two-point fluxes.","lead":"This paper adds a smart automatic 'friction' to high-order discontinuous Galerkin simulations of fluids, choosing the amount locally from how much the entropy inequality is violated. This gives a simpler way to guarantee entropy stability without special two-point fluxes, with fewer oscillations than standard entropy-stable DG.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"BR-1 spurious gradient null modes can make the denominator in (24) vanish, so Lemma 2's sufficient condition (21) is unproven for such states; the entropy inequality (22) and the time-step claim rest on an empirical upwind-dissipation assumption.","rationale":"The reader's weakest assumption is exactly the behavior of BR-1 spurious gradient null modes near the denominator in (24), and Section 4.5 of the paper explicitly identifies this as an unproved point left to future work. This is the most load-bearing concern because the proof of Lemma 2 requires the sufficient condition (21) to hold with a finite viscosity coefficient; when the denominator vanishes while the numerator is negative, that condition cannot be met, so the central entropy inequality (22) is not established unconditionally. The regularization in Remark 3 mitigates division by exactly zero by returning zero viscosity, but that means the entropy inequality can fail on such an element, and for small nonzero denominators it can produce enormous coefficients that would violate the claimed maximum stable time-step property. The paper's numerical experiments are extensive and provide credible evidence that the upwind convective flux dissipates these modes in practice, but they do not constitute a proof, and the paper says so. Other concerns, such as entropy projection sensitivity near vacuum states and the empirical nature of the time-step claim, are real but secondary: they affect robustness and a headline assertion, whereas the spurious-mode issue touches the proof of the main theorem itself. Given the strong numerical support and the explicit limitation in the text, the appropriate verdict remains CONDITIONAL, identical to the reader's, so no verdict change is recommended.","tokens_in":25512,"tokens_out":4931,"duration_ms":54360,"concrete_test":"On a periodic 2D uniform triangular mesh, take degree N = 2 and initialize one cell (or a periodic pattern of cells) exactly with a spurious null-space mode of (16) from Figure 1 superimposed on a constant state, chosen so that δk(uh) from (20) is negative while the denominator in (24) is zero or near machine epsilon. Compute εk from (24) with the regularization of Remark 3 and take one explicit SSPRK step with the local Lax-Friedrichs flux. If εk ≈ 0 on the mode cell while δk < 0, then the sufficient condition (21) and the semi-discrete entropy inequality (22) fail at t = 0; if εk is O(10⁷), the hyperbolic CFL claim fails. A companion run with small random perturbations would show whether upwind dissipation removes the mode on the CFL-relevant timescale.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorem is Lemma 2 with Lemma 3, and its sufficient condition (21) requires the local viscous dissipation Dk = Σ_{i,j}(εk Kij Θj, Θi)_Dk to dominate −min(0,δk(uh)). Section 4.5 concedes that the BR-1 gradient discretization (16) admits non-constant spurious null-space modes for which Θ1 = ... = Θd = 0 (Figure 1, [48]); on such a state Dk = 0 regardless of εk. If δk(uh) < 0 there, no finite εk can satisfy (21), so the global entropy inequality (22) is not actually proved for general meshes and data. The regularized ratio in Remark 3 returns εk = 0 when the denominator is exactly zero (since ab/(δ+b²) with b = 0 is 0), and for small nonzero b it can amplify by O(1/√δ) ≈ 10⁷, which would destroy the claimed explicit hyperbolic time-step bound. The paper's response—that upwind convective fluxes rapidly dissipate these modes in practice—is numerical intuition, explicitly deferred to future work. Thus the headline claim that any standard weak-form DG code can be made entropy stable by adding this local parameter-free term is conditional on an unproven suppression mechanism.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an entropy correction artificial viscosity for standard weak-form discontinuous Galerkin (DG) discretizations of systems of conservation laws. The viscosity coefficient is chosen cell-locally to enforce a discrete cell entropy identity, which yields a global semi-discrete entropy inequality under a sufficient condition (21). The authors prove a superconvergence estimate for the volume entropy residual (Lemma 4), discuss the effect of inexact quadrature, compare the method with existing local entropy corrections and artificial viscosity approaches, and present extensive numerical experiments for the compressible Euler equations, including accuracy tests, shock tubes, the Shu-Osher problem, a 2D Riemann problem, and a long-time Kelvin-Helmholtz instability. The advertised advantages are a parameter-free, locally computable viscosity that recovers the same entropy inequality as flux differencing entropy stable DG methods, reduces oscillations, improves local linear stability, and preserves a hyperbolic explicit time-step restriction.","tokens_in":25787,"tokens_out":13556,"duration_ms":143326,"significance":"If the central claim held unconditionally, the paper would be a useful practical alternative to flux differencing entropy stable DG: it avoids the need for entropy conservative two-point fluxes and can be added to an existing weak-form DG code in a cell-local manner. Lemma 4's superconvergence estimate for the volume entropy residual is a genuine analytical contribution, and the numerical study is thorough, with reproducible code provided. However, the main advertised guarantees are conditional: Lemma 2 requires the sufficient condition (21), and the proposed coefficient (24) does not provably satisfy that condition for all admissible states because of the BR-1 spurious gradient null-space modes discussed in Section 4.5. The paper is transparent about this limitation, but the abstract's unconditional claims overstate what is proved.","major_comments":[{"comment":"The central result is conditional on the sufficient condition (21), but the proposed viscosity coefficient does not guarantee (21) on general meshes. Section 4.5 observes that the BR-1 gradient discretization (16) admits non-constant spurious null-space modes with Θ1=...=Θd=0, for which the denominator in (24) vanishes while the volume entropy residual δk(uh) need not be nonnegative. For such states no finite εk satisfies (21), so the global entropy inequality (22) is not proven. Remark 3's regularized ratio does not cure this: when the denominator is exactly zero it returns εk=0, and for small nonzero denominators it can amplify by O(1/√δ)≈10^7 with δ=10^{-14}, which would destroy the claimed explicit hyperbolic time-step bound. Because the analysis of these modes is explicitly deferred to future work, the abstract's statement that any standard weak-form DG code can be made entropy stable by adding this local parameter-free term is not supported. A concrete numerical test initializing a spurious null mode, or an analytical modification such as an LDG-type gradient discretization or a penalty that removes the null space, is needed before the unconditional claim can stand.","section":"§4.5, Eq. (24), Lemma 2, Remark 3"},{"comment":"The claim that the method preserves a hyperbolic maximum stable time-step size under explicit time-stepping is not established. No theorem or estimate in the paper bounds εk(uh); Section 6.6 reports only adaptive timestep counts and explicitly states that future work will analyze the maximum stable time-step restriction more rigorously. In view of the O(1/√δ) amplification in Remark 3, this advertised property should either be proved under the upwind-dissipation assumption or removed from the abstract and conclusion.","section":"§7, Abstract, §6.6"}],"minor_comments":[{"comment":"The statement that both factors in the Cauchy-Schwarz estimate are O(h^{N+1+√d}) is inconsistent with the final estimate (27), which is O(h^{2N+2+d}); the intermediate exponent should be O(h^{N+1+d/2}), since the scaling |D_k|=O(h^d) enters through the L2 norm on a d-dimensional simplex.","section":"§4.4, Eqs. (25)-(27)"},{"comment":"The notation uh ∈ [P^N(D_k)]^d is dimensionally wrong for systems: the number of solution components is n (for example, n=d+2 for the compressible Euler equations), not the spatial dimension d.","section":"Lemma 4 statement"},{"comment":"The denominator in the derivative-based local entropy correction has a duplicated subscript 'D_k D_k' in the inner product, which should be a single subscript.","section":"Eq. (30)"},{"comment":"The claim that the viscosity coefficients are parameter-free is slightly overstated, since the regularized ratio in Remark 3 uses a fixed tolerance δ=10^{-14}; although the authors report insensitivity to δ, the abstract should acknowledge this safeguard.","section":"Abstract and Remark 3"},{"comment":"Lemma 1 is stated with a reference to [13] and a 'straightforward modification' but no proof; a short sketch of the localization argument would make the paper more self-contained.","section":"Lemma 1"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the manuscript is well written and the numerical study is substantial. The main gap, namely the unproven control of BR-1 spurious gradient null-space modes in Eq. (24), is openly acknowledged by the author but is load-bearing for the advertised unconditional entropy stability and time-step claims. I believe the issue is fixable within the manuscript's scope, for example by changing the gradient discretization or by stating the theorem with the extra hypothesis that the denominator in (24) is controlled along the discrete trajectory, so I recommend major revision rather than rejection. The paper fits the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is one of the better papers in the entropy-correction/artificial-viscosity line. The construction is materially different from Abgrall-Offner-Ranocha: a consistent BR-1 viscous discretization, a volume entropy residual based on the entropy projection, and a sharp O(h^{2N+2+d}) estimate for that residual. Lemma 2 is correct as a conditional: any epsilon satisfying (21) yields the same global entropy inequality as flux differencing DG (Lemma 3 handles the interface fluxes). The experiments are extensive and supportive—optimal convergence, reduced oscillations, better local linear stability than flux-differencing schemes—and the Julia code is public. That is real work, and the residual estimate is a genuine sharpening of the earlier O(h^{N+1+d}) bounds.\n\nThe soft spots are both about what is promised versus what is proved. The abstract says the method preserves a hyperbolic maximum stable time-step. What the paper actually shows is adaptive timestep counts for a few test problems, and the text itself defers rigorous analysis to future work. The regularized ratio in Remark 3 can amplify a near-zero denominator by O(1/sqrt(delta)) ~ 1e7; that is not obviously consistent with the time-step claim.\n\nThe more substantive issue is the BR-1 spurious gradient null-space modes. Section 4.5 concedes that for some non-constant states the gradient discretization returns Theta_i = 0 on an element. On such a state the denominator in (24) is zero regardless of epsilon. If the volume entropy residual is negative there, no finite epsilon satisfies (21), and the regularized ratio returns epsilon = 0 exactly when the denominator is zero—so the entropy inequality (22) is not actually guaranteed on such states. The paper's answer, that upwind convective fluxes dissipate these modes quickly, is numerical intuition and the paper says so. That does not invalidate the conditional theorem, but it does mean the headline 'any standard weak-form DG code can be made entropy stable by adding this local parameter-free term' is not fully established. It is the one point I would want the author to address head-on, either by proof or by a stated assumption.\n\nMinor: Section 4.4 has an exponent typo—sqrt(d) should be d/2—but the final estimate is consistent.\n\nRecommendation: send to review. The core construction and the residual estimate deserve referee time. The referee should ask for the time-step claim to be softened and for the null-mode concern to be addressed, but the paper is honest about both limitations and the numerics are strong.","headline":"Entropy-correction viscosity with a genuinely sharper residual estimate and strong numerics, but the time-step and BR-1 null-mode guarantees are softer than the abstract implies.","tokens_in":26297,"tokens_out":4156,"would_cite":true,"duration_ms":42691,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M60","65M12","35L65"],"pacs":[],"model":"deepseek-v4-flash","headline":"A local, parameter-free artificial viscosity recovers the entropy-stability guarantee of flux-differencing DG without entropy-conservative two-point fluxes.","keywords":["entropy stability","discontinuous Galerkin","artificial viscosity","entropy correction","conservation laws","high order methods","entropy projection","compressible Euler equations"],"falsifier":"Initialize a degree N=2 periodic DG simulation with one of the spurious BR-1 null-space modes shown in Figure 1, such as a checkerboard gradient mode, and monitor the viscosity coefficient (24). If the coefficient grows without bound as the denominator approaches zero, or if the explicit step size collapses to an O($h^{2}$) parabolic restriction on a genuinely hyperbolic problem, the load-bearing assumption fails.","tokens_in":25306,"feed_emoji":"🧮","tokens_out":7477,"duration_ms":68877,"temperature":0.7,"pith_summary":"This paper proposes making a standard weak-form discontinuous Galerkin (DG) discretization of a nonlinear conservation law entropy stable by adding an artificial viscosity term, with a coefficient chosen cell-by-cell from the local violation of a cell entropy inequality. The coefficient is parameter-free and locally computable, and it requires none of the entropy-conservative two-point flux formulas that flux-differencing entropy stable DG methods depend on. The main theorem states that with this coefficient the semi-discrete scheme satisfies the same global entropy inequality satisfied by entropy-stable flux-differencing DG methods, provided interface fluxes are entropy stable and evaluated using the entropy projection. If correct, any existing weak-form DG code can gain an entropy stability guarantee with a small local modification, while preserving high order accuracy and a hyperbolic explicit time-step bound.","feed_headline":"Local viscosity restores entropy stability without flux formulas","feed_subtitle":"A cell-local, parameter-free coefficient gives the same entropy guarantee as flux differencing, without two-point fluxes.","key_machinery":"The load-bearing object is the cell-local ratio defining the viscosity coefficient: the numerator is the volume entropy residual $\\delta_k(u_h)$ from (20), a local measure of the failure of the discrete chain-rule identity, and the denominator is the viscous entropy dissipation $\\sum_{i,j}(\\epsilon_k K_{ij}\\Theta_j,\\Theta_i)_{D_k}$, where $\\Theta_i$ is a BR-1 DG approximation of $\\partial \\Pi^N v(u_h)/\\partial x_i$. The entropy projection $\\tilde u = u(\\Pi^N v(u_h))$ is the other essential ingredient: evaluating interface fluxes at $\\tilde u$ makes inter-element contributions dissipative, and under nodal collocation this projection collapses to evaluation of $u_h$ at quadrature points. The proofs use a Roe-type linearization matrix to show that $\\delta_k$ is a commutator of projection with the flux Jacobian, giving the superconvergence estimate that keeps the viscosity small for smooth data.","core_discovery":"The central claim is that entropy stability can be enforced by a local viscous term of BR-1 type. Concretely, define the volume entropy residual $\\delta_k(u_h)$ as the cell-local failure of the discrete entropy identity, and choose the viscosity coefficient as $\\epsilon_k(u_h) = -\\min(0,\\delta_k(u_h)) / \\sum_{i,j}(K_{ij}\\Theta_j,\\Theta_i)_{D_k}$, using a regularized ratio when the denominator is small. Lemma 2 proves that if the viscous entropy dissipation is at least $-\\min(0,\\delta_k(u_h))$, the DG scheme satisfies the global semi-discrete entropy inequality; Lemma 3 upgrades this to a global inequality when the interface flux is entropy stable and evaluated at the entropy projection $\\tilde u = u(\\Pi^N v(u_h))$. The paper also proves that the volume entropy residual is superconvergent, of order $O(h^{2N+2+d})$ for smooth solutions, which explains why the viscosity vanishes for smooth flows and for linear advection. The resulting inequality is the same one satisfied by flux-differencing entropy stable DG methods, achieved without entropy-conservative two-point fluxes.","pith_inferences":["Because the viscosity coefficient is essentially an entropy-residual sensor, it could be combined with a shock-capturing viscosity by taking the maximum of the two coefficients, a route the paper mentions in an appendix.","The superconvergence estimate suggests the coefficient itself could serve as a mesh-adaptive or p-adaptive indicator for smooth versus under-resolved regions, a use the paper does not pursue.","If the spurious gradient mode issue is resolved by switching to non-central gradient discretizations such as local DG, the coefficient would be even smaller; the paper notes this as future work.","The construction should extend to any system with a strictly convex entropy, including shallow water or magnetohydrodynamics, since only the entropy variables and the viscous Jacobian enter the coefficient."],"forward_implications":["Any standard weak-form DG code can be made entropy stable by inserting the BR-1 viscous term and the cell-local coefficient, without changing the convective discretization.","The approach removes the need for explicit entropy-conservative two-point flux formulas, which may be unavailable or expensive for some systems.","High order accuracy is preserved, and numerical experiments show the viscosity coefficient converges to zero for smooth solutions and is sharply localized near shocks.","The scheme is less dissipative than flux-differencing entropy stable DG in the accuracy tests reported, and more locally linearly stable, with smaller spurious oscillations on shock problems.","Explicit time-stepping does not collapse to a parabolic step restriction in the tested cases, even when large viscosity is generated near shocks."],"supporting_citations":[{"why":"Supplies the local entropy-correction idea that this paper modifies and sharpens.","marker":"[2]"},{"why":"Provides the BR-1 viscous discretization and its localizable stability estimate used in Lemma 1.","marker":"[13]"},{"why":"Defines the flux-differencing entropy stable DG framework whose global entropy inequality Lemma 3 recovers.","marker":"[16]"},{"why":"Introduces the entropy projection and the nodal/vmodal entropy stable DG formulations used throughout.","marker":"[11]"},{"why":"Gives the Roe-type linearization theorem used to prove the superconvergence estimate for the volume entropy residual.","marker":"[36]"},{"why":"Establishes the entropy-conservative two-point flux machinery that the paper seeks to avoid.","marker":"[78]"},{"why":"Provides a derivative-based entropy correction that the paper compares against and sharpens.","marker":"[28]"},{"why":"Motivates enforcing a cell entropy inequality on nodal DG through subcell limiting.","marker":"[55]"}],"fun_headline_variants":["Parameter-free local viscosity enforces entropy stability","No two-point fluxes: local viscosity gives entropy stable DG","Cell-local viscosity matches flux-differencing entropy guarantee","Viscosity replaces flux differencing for entropy stable DG","Local viscosity recovers same entropy inequality without flux formulas"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The scheme assumes that the spurious gradient modes of the BR-1 viscosity operator are damped by the upwind convective flux in practice, so that the denominator in the viscosity coefficient does not become pathologically small; if it did, the viscosity would blow up and the time-step claim would fail.","fun_headline_variants_meta":{"raw":{"variants":["Parameter-free local viscosity enforces entropy stability","No two-point fluxes: local viscosity gives entropy stable DG","Cell-local viscosity matches flux-differencing entropy guarantee","Viscosity replaces flux differencing for entropy stable DG","Local viscosity recovers same entropy inequality without flux formulas"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000294,"raw_usage":{"total_tokens":1746,"prompt_tokens":1016,"completion_tokens":730,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":632,"completion_tokens_details":{"reasoning_tokens":654}},"tokens_in":632,"tokens_out":730,"duration_ms":7786,"temperature":1.0,"reasoning_tokens":654,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T12:36:01.652210+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Initialize a degree N=2 periodic DG simulation with one of the spurious BR-1 null-space modes shown in Figure 1, such as a checkerboard gradient mode, and monitor the viscosity coefficient (24). If the coefficient grows without bound as the denominator approaches zero, or if the explicit step size collapses to an O($h^{2}$) parabolic restriction on a genuinely hyperbolic problem, the load-bearing assumption fails.","supporting_citations":[{"cited_title":"Entropy stable modal discontinuous Galerkin schemes and wall boundary conditions for the compressible Navier-Stokes equations","cited_arxiv_id":null,"evidence_quote":"Provides the BR-1 viscous discretization and its localizable stability estimate used in Lemma 1."},{"cited_title":"On discretely entropy conservative and entropy stable discontinuous Galerkin methods","cited_arxiv_id":null,"evidence_quote":"Introduces the entropy projection and the nodal/vmodal entropy stable DG formulations used throughout."},{"cited_title":"Numerical approximation of hyperbolic systems of conservation laws , volume 118","cited_arxiv_id":null,"evidence_quote":"Gives the Roe-type linearization theorem used to prove the superconvergence estimate for the volume entropy residual."},{"cited_title":"The numerical viscosity of entropy stable schemes for systems of conservation laws","cited_arxiv_id":null,"evidence_quote":"Establishes the entropy-conservative two-point flux machinery that the paper seeks to avoid."},{"cited_title":"High order entropy pre- serving ADER-DG schemes","cited_arxiv_id":null,"evidence_quote":"Provides a derivative-based entropy correction that the paper compares against and sharpens."},{"cited_title":"High order entropy stable discontinuous Galerkin spectral element methods through subcell limiting","cited_arxiv_id":null,"evidence_quote":"Motivates enforcing a cell entropy inequality on nodal DG through subcell limiting."}],"review_version":1}