{"id":"7702438c-73c7-44b7-8012-7bbcd97d5ca9","arxiv_id":"2412.03477","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The semi-discrete Active Flux method is proven stationarity preserving for multi-dimensional linear acoustics on Cartesian grids, with explicit kernel bases and a 2D CFL limit near 0.28.","lead":"This paper proves that the semi-discrete Active Flux method, a high-order scheme for hyperbolic equations, preserves stationary vortex-like states of the linear acoustic equations in 2D and 3D on Cartesian grids. The result matters because stationarity preservation is what lets a scheme keep vortices and other balanced states without artificial diffusion.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorems 5.3 and 5.4 hinge on an unreproduced Mathematica rank computation to fix the exact kernel dimension; without an independent check, the upper bound in Definition 5.1 is unsecured in 2D.","rationale":"The reader's weakest assumption matches the central gap: the exact dimensionality of the kernel is asserted on the basis of a non-reproducible symbolic computation. This is genuinely load-bearing because the theorem statements claim a 1-dimensional (2D) and 5-dimensional (3D) kernel, and the upper-bound part of Definition 5.1 depends on there not being more than S·Ndof stationary modes. A modest excess of kernel elements in 2D (more than 4) would break the stationarity-preservation property as defined. The sufficiency direction is solid and independently supported by the well-prepared mode experiments in Sections 7.1.3 and 7.2.2, which gives partial confidence, but the necessity direction has no such support. I therefore agree with the reader's CONDITIONAL verdict: the paper should be accepted if the Mathematica verification is reproduced or replaced by a short proof; otherwise the exact kernel-dimension theorem remains unverified. No change to the reader's verdict is needed.","tokens_in":38484,"tokens_out":16738,"duration_ms":161417,"concrete_test":"Use an independent computer algebra system (e.g., SymPy or Sage) to compute the rank of the 12×12 matrix E from Appendix C over the field of rational functions in tx, ty; verify that the rank is 11, so ker E is exactly 1-dimensional for generic k. Repeat for the 32×32 matrix from Appendix D over Q(tx,ty,tz), checking that the rank is 27. Then numerically evaluate the rank on a fine grid of unit-circle samples (tx,ty,tz), including singular points such as tx=1, ty=1, tz=1, tx=−1, and confirm that dim ker E ≤ 4 in 2D and ≤ 16 in 3D at every sample, so the upper bound in Definition 5.1 holds at special wave vectors as well.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The stationarity-preservation claim in Theorem 5.3 (2D) and Theorem 5.4 (3D) requires the upper bound dim ker E(k) ≤ S·Ndof in Definition 5.1 (S=1, Ndof=4 in 2D, so ≤4; S=2, Ndof=8 in 3D, so ≤16). The lower bound (at least 1 in 2D, at least 5 in 3D) is proved constructively, but the assertion that there are no other kernel elements is delegated to Mathematica in both proofs ('verified using mathematica due to the excessive length of computations' in Theorem 5.3; 'we confirm using mathematica' in Theorem 5.4). This is the only evidence for the upper bound, and no notebook or reproduction script is shipped. The numerical tests in Section 7 verify only that the exhibited modes are stationary; they cannot rule out additional spurious kernel elements. If the true kernel dimension in 2D exceeded 4, the method would not be stationarity preserving under Definition 5.1. Additionally, the displayed basis vectors have denominators that vanish on subvarieties (e.g., tx=1, ty=1), and the paper does not analyse whether the exact dimension statement covers these cases; the lower bound likely persists there, but the upper bound is left unchecked on those subvarieties.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the semi-discrete (generalized) Active Flux method for the linear acoustic equations on Cartesian grids in two and three spatial dimensions. The method updates cell averages by flux quadrature and point values by finite-difference approximations obtained from a globally continuous biparabolic reconstruction. Using the discrete Fourier transform, the authors derive the evolution matrix E(k) and characterize its kernel. Their central theorems (Theorems 5.3 and 5.4) state that the upwind semi-discrete Active Flux method is stationarity preserving: the numerical stationary states consist exactly of divergence-free velocity reconstructions with constant pressure, with kernel dimensions 1 in 2D and 5 in 3D, matching the non-trivial stationary states of the PDE. The paper also provides an analysis of numerical diffusion and stability and presents numerical experiments for vortex, well-prepared mode, and traveling-wave tests.","tokens_in":38725,"tokens_out":3821,"duration_ms":39312,"significance":"If the results are correct, this is a valuable contribution: it shows that the semi-discrete Active Flux method preserves the same stationary states as the classical fully discrete Active Flux method, extending a desirable structure-preservation property to a method-of-lines formulation that is more easily applied to nonlinear problems. The Fourier-based framework is clean and provides explicit kernel vectors with no fitted parameters. The authors give constructive lower bounds for the kernel, show that well-prepared modes remain stationary to round-off, and demonstrate that the vortex test drives the discrete divergence relations to machine zero. These numerical experiments lend credible support to the main claim. The main weakness is that the sharpness of the kernel dimension (the upper bound in Definition 5.1) is established only by an unreproduced computer algebra computation, which is load-bearing for the stationarity-preservation claim.","major_comments":[{"comment":"The upper bound in Definition 5.1, namely dim ker E(k) ≤ S·Ndof, is the load-bearing part of the stationarity-preservation claim, but the only evidence offered is the statements 'verified using mathematica due to the excessive length of computations' (end of the proof of Theorem 5.3) and 'we confirm using mathematica' (proof of Theorem 5.4). No notebook, script, or algebraic certificate is provided, so the claim is not independently verifiable from the manuscript. Since the lower bound alone shows only that the exhibited modes are stationary and does not rule out additional spurious kernel elements, this gap directly affects the central theorem. Please provide a reproducible symbolic computation (e.g., a notebook or a rank computation over C(tx,ty,tz) with a certificate such as a maximal minor) or a hand proof that the homogeneous system has full rank on the relevant complement.","section":"Section 5.3.1, Theorem 5.3 and Theorem 5.4"},{"comment":"The displayed kernel vectors (93) and Q1,...,Q5 in Appendix D have entries with denominators that vanish on subvarieties of the torus, e.g., tx=1, ty=1, tz=1, 1+tx=0, and 1+ty=0. The paper does not analyze whether the exact equality dim ker E(k)=S holds on these subvarieties, either by continuity, by limits, or by separate computations. Since Definition 5.1 involves min_k over all k, including these special wave vectors, the 'for general k' statement is insufficient as written. Please specify the exceptional set of wave vectors (if any) and verify the dimension claim there or explain why the limit argument applies.","section":"Theorems 5.3 and 5.4 and Appendix D"},{"comment":"The paper defines stationarity preservation through a comparison of kernel dimensions, but the numerical tests in Section 7 (vortex decay to machine zero and well-prepared modes) can only confirm that the constructed vectors lie in the kernel; they cannot certify the absence of extra kernel elements. This does not invalidate the experiments, but it means the numerical evidence does not fill the gap left by the computer algebra step. A statement acknowledging this explicitly would help the reader calibrate the role of the Mathematica verification.","section":"Section 4.3 and Definition 5.1"}],"minor_comments":[{"comment":"There is an unbalanced parenthesis in the formula for det E(k): '... + 2c2(∆x2 − ∆x∆y + ∆y2)))' appears to have an extra closing parenthesis; please correct.","section":"Equation (96)"},{"comment":"The stability analysis reports that the instability appears 'between ∆t = 0.28 and ∆t = 0.3' for the 2D acoustics case. Since ∆x = ∆y = 1 in that figure, it may be helpful to state explicitly that ∆t equals the CFL number there, and to clarify whether this bound is for the given fixed wave-number parametrization.","section":"Section 6.2"},{"comment":"In the expression (2kz(ky − i), −2kxkz, 2ikx, 0), the symbol i is used for the imaginary unit, but this is not defined at that point; consider writing 'i' as 'i' with a comment that this is the imaginary unit, or use ı to avoid confusion with an index.","section":"Eq. (114) and Section 7.2.2"},{"comment":"The notation '1 + tx(4 + tx)' is unambiguous but could be typeset more clearly as '1 + 4tx + tx^2' for readability; similarly for other kernel vectors. This is only a presentation issue.","section":"General notation"}],"recommendation":"major_revision","confidential_remarks":"The paper is part of a well-established research program on Active Flux methods, and the citation practice appears appropriate. The main issue is the reliance on an unreproduced computer algebra computation for a load-bearing step; this is fixable and does not appear to indicate a fundamental flaw, but the manuscript should not be accepted as is. I would also encourage the editor to ask for the Mathematica notebook or a symbolic rank certificate as supplementary material."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new thing here is the proof that the semi-discrete (generalized) Active Flux method preserves stationary states for linear acoustics in 2D and 3D, with explicit kernel bases and a diffusion/stability analysis. The sufficiency direction is done by hand and is clean: stationarity of point values plus continuity of normal derivatives forces the divergence of the reconstruction to vanish, and the average update becomes redundant. The numerical experiments back this up — the vortex test and the well-prepared Fourier modes decay to machine zero, and the 3D vortex ring behaves as advertised. There is no parameter fitting and no circularity; the reliance on earlier papers in the same program is normal and those results are properly cited.\n\nThe soft spot is real but narrow. The upper bound in Definition 5.1 — that the kernel has no extra elements beyond the explicitly constructed basis — is asserted on the basis of a Mathematica check that is not shipped. In Theorem 5.3 the paper says “verified using mathematica due to the excessive length of computations,” and Theorem 5.4 says the same. For a theorem whose force is the sharp equality of kernel dimensions, that is the load-bearing step, and an unreproduced computer algebra check is not the same as a proof. The stress-test worry about denominators vanishing on subvarieties (tx = 1, ty = 1) is also legitimate: the displayed basis vectors have poles there, and the paper does not explicitly analyze those wave vectors. I suspect the statements are still true by continuity or by a separate rank calculation, and the lower bound is constructive, so this is not a fatal flaw. But it is exactly the kind of gap a referee should ask to close: ship the notebook, add a short symbolic rank argument, or state the subvariety cases explicitly.\n\nThe stability analysis in Section 6 is more heuristic — eigenvalue plots and a CFL estimate — but that is secondary and clearly labeled. The 3D convergence study is honest about the coarse meshes and observed order around 2.5.\n\nWho gets value from this: anyone working on Active Flux, structure-preserving finite volume/finite difference methods, or low-Mach behavior. It deserves a serious referee. My recommendation is to send it to review, with the explicit request that the authors provide the missing computational verification in reproducible form before acceptance.","headline":"Solid extension of the Active Flux stationarity-preservation program to the semi-discrete variant; the central claim rests on an unshipped Mathematica rank computation, so the fix is to make that check reproducible.","tokens_in":39269,"tokens_out":1168,"would_cite":true,"duration_ms":15969,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M20","65M70","65M08","35E15"],"pacs":[],"model":"deepseek-v4-flash","headline":"For linear acoustics, semi-discrete Active Flux preserves all steady states in 2D and 3D.","keywords":["Active Flux","stationarity preserving","linear acoustics","Fourier transform","semi-discrete method","numerical diffusion","Cartesian grids","upwind scheme"],"falsifier":"Compute exactly the rank of the 12×12 evolution matrix $E(k)$ of Appendix C (and the 32×32 matrix for 3D) over the field of rational functions in $t_x,t_y,t_z$, for example by evaluating all minors symbolically at generic integer values of the translation factors; the theorem holds only if the generic rank is 11 in 2D and 27 in 3D, and any single wave vector with nullity greater than 1 (2D) or 5 (3D) would disprove stationarity preservation.","tokens_in":38268,"feed_emoji":"🌀","tokens_out":10635,"duration_ms":88646,"temperature":0.7,"pith_summary":"This paper proves that the semi-discrete (generalized) Active Flux method, which replaces the exact evolution operator of classical Active Flux with finite-difference approximations of spatial derivatives, retains stationarity preservation for linear acoustics on Cartesian grids. In two dimensions the evolution matrix has a one-dimensional kernel, and in three dimensions a five-dimensional kernel, whose elements correspond exactly to divergence-free velocity reconstructions with constant pressure. These discrete stationary states coincide with those of the classical Active Flux method in 2D, so the easier-to-implement semi-discrete version loses none of the structure preservation. The analysis also shows that the upwind Jacobian splitting is essential: a Rusanov-type splitting destroys stationarity preservation entirely. Numerical experiments confirm third-order convergence for smooth waves and the long-time preservation of vortices and vortex rings.","feed_headline":"Semi-discrete Active Flux keeps acoustic steady states in 2D and 3D","feed_subtitle":"The method's discrete steady states are exactly the divergence-free velocity fields, matching classical Active Flux.","key_machinery":"The central object is the evolution matrix $E(k)$ obtained by applying the discrete Fourier transform to the semi-discrete update equations; its kernel is the set of numerical stationary states. The argument that kernel vectors correspond exactly to divergence-free reconstructions rests on the $P^{2,2}$-projected divergence of Definition 5.2 and the unisolvence of the broken space $D^{2,2}_{\\mathrm{br}}$ with respect to the eight pointwise degrees of freedom (Lemma 5.1), which lets the paper pass from eight (or 26) pointwise divergence-vanishing conditions to the vanishing of the divergence polynomial itself. The upwind Jacobian splitting matters because it adds the requirement that normal derivatives be continuous at stationary states, and the resulting count of constraints leaves exactly one free parameter in 2D (and five in 3D).","core_discovery":"The central discovery is that the semi-discrete Active Flux method, defined by updating cell averages through exact flux quadrature and point values through one-sided finite differences obtained from a globally continuous biparabolic reconstruction, is stationarity preserving for the linear acoustic equations in two and three spatial dimensions. Concretely, the discrete Fourier transform reduces the method to an evolution matrix $E(k)$, and the paper proves (Theorems 5.3 and 5.4) that for general wave vectors the kernel of $E(k)$ is one-dimensional in 2D and five-dimensional in 3D, consisting precisely of the Fourier modes whose reconstruction has vanishing divergence and constant pressure. The unique kernel vector in 2D is given explicitly, and the 3D basis is provided in an appendix; in both cases the kernels match the analytic stationary states of the acoustic system (the divergence-free velocity fields with zero pressure) plus, in 3D, higher-order terms. The proof uses a unisolvence property of the broken divergence space $D^{2,2}_{\\mathrm{br}}$ to show that vanishing of the divergence at all point values forces the polynomial divergence to vanish identically, together with continuity of normal derivatives that the upwind splitting enforces at stationarity.","pith_inferences":["The same Fourier-kernel analysis could be applied to other linear symmetrizable hyperbolic systems (e.g., linearized Euler or Maxwell equations) to characterize which splittings and reconstruction spaces are stationarity preserving; the general criterion of Definition 5.1 is system-agnostic.","Because the kernel dimension is determined solely by the reconstruction space and the upwind constraints, replacing the biparabolic reconstruction with higher-degree polynomials would likely enlarge the set of preserved stationary states; testing $P^{3,3}$ reconstructions would show whether the matching of classical and semi-discrete stationary states persists.","A practical implication for code designers: the non-physical eigenmode that limits the CFL number is a separate degree-of-freedom mode (the point values vs. averages); damping or filtering that mode could push the stability limit closer to the classical method's CFL.","The Mathematica rank check could be replaced by a constructive symbolic proof (e.g., computing the Smith normal form of $E(k)$ over the polynomial ring), which would remove the only computational step in the proof and make the theorem fully analytic."],"forward_implications":["Semi-discrete Active Flux can be applied to nonlinear systems of conservation laws without sacrificing correct representation of steady states of the linearized equations, since the structure preservation is inherited from the reconstruction space, not from the exact evolution operator.","The discrete stationary states in 2D are identical to those of classical Active Flux, so vortices and other divergence-free flows are preserved by both schemes; the semi-discrete method can serve as a drop-in replacement that is easier to extend to nonlinear problems.","The upwind splitting is the structure-preserving choice; switching to a Rusanov-type Jacobian splitting (39) produces a method with no non-trivial stationary states, as $\\det E(k) \\neq 0$ for generic $k$, so the choice of numerical diffusion terms controls stationarity preservation.","The maximum stable time step for the semi-discrete method on linear acoustics is roughly $\\Delta t \\approx 0.28$ (with $\\Delta x = \\Delta y = 1$), about half that of the classical method, so third-order accuracy is retained at the cost of more steps per unit time.","In 3D the 5-dimensional kernel includes the two analytic non-trivial stationary modes plus three higher-order modes, meaning the method captures all stationary states of the PDE at the discrete level."],"supporting_citations":[{"why":"supplies the definition of stationarity preservation (criterion (68)) and the analytic stationary states of linear systems that the paper compares against.","marker":"[Bar19]"},{"why":"established the stationarity preservation of classical Active Flux and gave the divergence-free reconstruction conditions (their eqs. (6.25)–(6.26)) that Corollary 5.2 matches.","marker":"[BHKR19]"},{"why":"provides the exact evolution operator for linear acoustics used by the classical Active Flux method, which the semi-discrete method replaces with finite differences.","marker":"[BK22]"},{"why":"introduced the multi-dimensional Active Flux method and its application to linear acoustics, defining the degrees of freedom used throughout.","marker":"[ER13]"},{"why":"introduced the semi-discrete Active Flux formulation for nonlinear problems, the method whose structure preservation this paper analyzes.","marker":"[Abg22]"},{"why":"extended Active Flux to arbitrary order and (together with [Abg22]) motivated the semi-discrete version analyzed here.","marker":"[AB23a]"},{"why":"provided the well-prepared stationary-mode construction used in the 2D numerical test of Section 7.1.3.","marker":"[Bar20]"},{"why":"explained that stability of semi-discrete methods is governed by the non-physical eigenvalue, which the diffusion analysis in Section 6 relies on.","marker":"[Roe21]"}],"fun_headline_variants":["Active Flux variant keeps acoustic steady states exact in 2D and 3D","Semi-discrete Active Flux: exact steady states for linear acoustics","Fourier analysis shows Active Flux preserves acoustic steady states","New proof: semi-discrete Active Flux preserves acoustic steady states","Multi-D semi-discrete Active Flux matches classical steady states"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The assertion that the evolution matrix has no stationary modes beyond the ones listed is verified by a computer algebra system rather than by a hand-written proof, so the kernel-dimension equality stands or falls with that computation.","fun_headline_variants_meta":{"raw":{"variants":["Active Flux variant keeps acoustic steady states exact in 2D and 3D","Semi-discrete Active Flux: exact steady states for linear acoustics","Fourier analysis shows Active Flux preserves acoustic steady states","New proof: semi-discrete Active Flux preserves acoustic steady states","Multi-D semi-discrete Active Flux matches classical steady states"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000731,"raw_usage":{"total_tokens":3258,"prompt_tokens":915,"completion_tokens":2343,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":531,"completion_tokens_details":{"reasoning_tokens":2253}},"tokens_in":531,"tokens_out":2343,"duration_ms":15154,"temperature":1.0,"reasoning_tokens":2253,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T22:20:28.602800+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute exactly the rank of the 12×12 evolution matrix $E(k)$ of Appendix C (and the 32×32 matrix for 3D) over the field of rational functions in $t_x,t_y,t_z$, for example by evaluating all minors symbolically at generic integer values of the translation factors; the theorem holds only if the generic rank is 11 in 2D and 27 in 3D, and any single wave vector with nullity greater than 1 (2D) or 5 (3D) would disprove stationarity preservation.","supporting_citations":[],"review_version":1}