{"id":"c5c9ebc0-adab-490c-a5e8-5c8383bb28eb","arxiv_id":"2607.23133","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"On two plasma benchmarks, the unsplit Active Flux method is slightly more accurate and cheaper at coarse resolution in 2D, while the split-step method is slightly more accurate at fine resolution and more extensible.","lead":"Two recently proposed Active Flux numerical schemes for the Vlasov–Poisson plasma model are tested side-by-side on two standard plasma physics benchmarks. One scheme is modestly more accurate on coarse grids and cheaper in 2D; the other is slightly more accurate on fine grids and easier to extend to higher dimensions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Error-metric reference is a single unconverged PFC run; the comparative ranking may be an artifact of reference error.","rationale":"The reader's weakest assumption — that the PFC reference may not be accurate enough to support the ranking — is exactly the most load-bearing concern. The numerical comparison is the paper's main contribution, and it is computed entirely against one reference solution with no convergence verification. The cost claim in §2.4 is also based on term counts rather than runtime, but it is explicitly qualified ('according to this metric') and secondary to the numerical comparison. I therefore agree with the CONDITIONAL verdict: the qualitative comparison is plausible, but the lack of reference validation and reproducibility leaves the central ranking uncertain. No verdict change is needed beyond the reader's conditional recommendation.","tokens_in":8640,"tokens_out":13045,"duration_ms":116367,"concrete_test":"Recompute the two-stream ϵVP curves (Fig. 2) using a second, independently converged reference — e.g., a PFC run at N=1024 with CFL=1/(10π) or a high-order spectral/DG Vlasov–Poisson solver at equivalent resolution — and check whether the ordering of unsplit vs split-step BM at N=32,64,128 is preserved. If the ordering flips or the unsplit coarse-grid advantage disappears, the paper's comparative claim is reference-dependent.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quality ranking — unsplit slightly better on coarse grids, split-step/BM slightly better on fine grids — is measured by Eq. (13) against one PFC reference solution at N=512 with CFL=1/(5π). The manuscript provides no self-convergence study, Richardson estimate, or independent reference. In strong Landau damping, the Active Flux runs leave the reference curve after t≈50 even at 128×128 (Fig. 4b), and in the two-stream case the error growth is rapid after t≈40 (Fig. 2). If the PFC reference is not itself converged in that regime, the relative ordering of the two methods could reflect the reference's numerical dissipation or dispersion rather than the properties of the Active Flux schemes. Since the conclusion explicitly bases its numerical claim on 'our comparisons to the reference,' reference accuracy is load-bearing for the paper's main comparison.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper compares two recently proposed Active Flux methods for the 1D1V Vlasov–Poisson system: an unsplit, fully discrete method from [17] and a split-step method from [16] using Strang, Yoshida, or Blanes–Moan operator splitting. After reviewing the two discretizations, it develops a theoretical cost model based on term evaluations in §2.4 and presents numerical comparisons for the two-stream instability and strong Landau damping in §3, using a high-resolution split-step PFC solution (N=512, CFL=1/(5π)) as reference. The main conclusion is that the unsplit method is slightly more accurate on coarser grids and cheaper in two dimensions, while the split-step method is slightly more accurate on finer grids and more readily extensible to higher dimensions.","tokens_in":8887,"tokens_out":5288,"duration_ms":53142,"significance":"If the numerical ranking is robust, the paper provides useful practical guidance for choosing between two Active Flux formulations. It assembles two methods from the same family, evaluates them on standard benchmarks against an independent reference method (PFC, [5]), and makes the theoretical cost model explicit. The algorithmic descriptions are consistent with the cited sources and I found no internal inconsistency in the derivations as presented. However, the central quantitative claims currently rest on an unvalidated reference solution and on a term-count cost proxy rather than measured runtime; both are fixable and should be addressed before publication.","major_comments":[{"comment":"The error metric is computed against a single PFC reference run at N=512 with CFL=1/(5π). No self-convergence study, Richardson estimate, or independent second reference is provided. In strong Landau damping the Active Flux curves depart from the reference after t≈50 even at 128×128 (Fig. 4b), and in the two-stream case the error grows rapidly after t≈40 (Fig. 2). If the PFC reference is not converged in these regimes, the relative ordering 'unsplit slightly better on coarse grids' could be an artifact of reference dissipation/dispersion rather than a property of the Active Flux schemes. Because the conclusion's numerical claim is explicitly 'in our comparisons to the reference,' the reference must be validated (e.g., a finer PFC run with a Richardson estimate or a second independent method).","section":"Section 3, Eq. (13)"},{"comment":"The statement that the unsplit method 'is also cheaper for the two-dimensional case considered here' is based on the term-count proxy 9 vs 27 evaluations per point value, not on measured runtime, memory traffic, or total work including Poisson solves. The proxy omits the cost of the ODE solves for point values in the unsplit method and the multiple Poisson solves in the higher-order split-step variants. Please either report measured wall-clock times for the 2D experiments or qualify the conclusion as a theoretical term-count estimate rather than a measured cost comparison.","section":"Section 2.4 and Conclusion"},{"comment":"The central claims that the unsplit method is 'slightly better' on coarse grids and the split-step/BM method 'slightly better' on finer grids are supported only by curves; no numerical error values, error tables, or observed convergence rates are given. This makes it hard to verify the relative ordering or judge whether the differences are meaningful. Please include a table of ϵVP at representative times/resolutions and, if possible, estimated convergence rates.","section":"Section 3, Figs. 2 and 4"}],"minor_comments":[{"comment":"'and for H_{i+1/2,j} respectively' should be reworded to indicate that the H flux is approximated analogously; as written it is incomplete.","section":"Section 2.2, after Eq. (4)"},{"comment":"The sentence 'Applying second-order Strang splitting reduces the number of Poisson solves required per time step to one' should explicitly state that this applies to the split-step method and only for Strang splitting, not for Yoshida or Blanes–Moan.","section":"Section 2.4"},{"comment":"Please define the discrete electric field energy in the text; the expression 1/2 Δx ∑ E_i^2 appears only in the narrative and is not labeled as an equation.","section":"Section 3.2"},{"comment":"References [3] and [21] are the same paper (Rossmanith and Seal, J. Comput. Phys. 230, 2011). Duplicate references should be merged.","section":"References"},{"comment":"The d≥4 break-even statement is explicitly qualified as 'according to this metric,' but the conclusion drops this qualifier when saying the unsplit method is 'cheaper' in 2D. Align the wording.","section":"Section 2.4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a side-by-side comparison by the two groups that developed the methods; self-citations are expected and appropriate. The main risk is not circularity but reference validity. If the authors add a convergence study of the PFC reference and a measured cost comparison, the paper would likely be acceptable. The numerical comparison is otherwise useful and within the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis is a useful, no-frills benchmark of two existing Active Flux schemes for 1D1V Vlasov-Poisson. The direct head-to-head comparison is new, and the paper is honest about what it cannot claim. If you work with Active Flux or semi-Lagrangian methods, it is worth your time; if not, it is not going to change your view of the field.\n\nThe new content: a pointed comparison of the split-step (Hensel et al.) and unsplit (Kiechle et al.) methods on the two-stream instability and strong Landau damping, at several grid resolutions; a simple term-count cost model; and the observation that the electric-field time-derivative formulas (7)–(8) are genuinely one-dimensional, which blocks a straightforward unsplit extension to 2D2V or 6D. The numerical experiments are cleanly set up, and the writing is clear. I appreciate that the authors note in Section 2.4 that the unsplit method need not use a full tensor-product grid, so the cost comparison is explicitly a naive upper bound.\n\nThe soft spots are mostly at the evidence level. The relative ranking is measured against a single high-resolution PFC reference run (N=512, CFL=1/(5π)), and there is no self-convergence study or Richardson estimate to show that this reference is itself converged in the long-time regimes where the Active Flux solutions depart from it (e.g., after t≈50 in strong Landau damping at 64×64). The error metric (13) is an L1 relative difference; with no error bars or runtime measurements, \"slightly better\" is a thin basis for the conclusion. The cost model is a term-count proxy, not measured runtime, and the serendipity-element remark suggests the unsplit method could be cheaper than the naive count. None of these problems is fatal to the qualitative conclusion that both methods work and are comparable, but they do make the quantitative ranking provisional.\n\nThe paper also correctly flags its own limitation: the unsplit method's extension to higher dimensions is not immediate because Eqs. (7)–(8) rely on 1D structure. That is a real structural insight.\n\nWho this is for: researchers choosing between Active Flux variants for Vlasov-Poisson, and developers of kinetic solvers who want a compact comparison. It is not a breakthrough, but it is a solid, honest piece of comparative work. It deserves a serious referee — someone who can push for reproducibility details (code/data release, reference convergence) and a runtime break-even measurement.\n\nMy recommendation: yes, send it to peer review, but make sure the revision addresses the reference validation and the cost-measurement caveats.","headline":"A clean, honest benchmark of two Active Flux variants for Vlasov-Poisson; the quantitative ranking is provisional until the reference solution is shown converged.","tokens_in":9307,"tokens_out":2474,"would_cite":true,"duration_ms":24080,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M08","65M25","35Q83"],"pacs":["52.65.Ff"],"model":"deepseek-v4-flash","headline":"For the Vlasov–Poisson system, the unsplit Active Flux method is slightly more accurate on coarse grids and cheaper in two dimensions, while the split-step version is more accurate on fine grids and easier to extend to higher dimensions.","keywords":["Active Flux","Vlasov–Poisson system","finite volume methods","semi-Lagrangian methods","operator splitting","third-order accuracy","plasma kinetics","dimensional scaling"],"falsifier":"Run the two-stream instability with both methods against an independent, converged reference—for example, two references at different resolutions with Richardson extrapolation, or a manufactured solution—and compare the relative error at t>40 on 64x64 and 128x128 grids. If the unsplit method no longer shows lower error than the split-step method on coarse grids, the paper's central ranking would be refuted.","tokens_in":8596,"feed_emoji":"⚡","tokens_out":4538,"duration_ms":40204,"temperature":0.7,"pith_summary":"This paper establishes a practical trade-off between the two Active Flux approaches currently proposed for the Vlasov–Poisson system, the standard kinetic model of a collisionless electrostatic plasma. The unsplit method, which evolves all interface point values and the electric field in one fully coupled time step, is roughly as accurate as the split-step method while being cheaper in the two-dimensional setting studied here. The split-step method, which decomposes the dynamics into alternating one-dimensional advection stages, is somewhat more accurate on fine grids and generalises more directly to higher dimensions and electromagnetic problems. On two benchmark problems, both methods match a high-resolution reference until the solution develops unresolved small-scale structure, after which the coarse-grid comparison becomes less reliable. The practical conclusion is that the better method depends on resolution, dimension, and available computing budget.","feed_headline":"Unsplit Active Flux edges coarse grids; split scales to 4D+","feed_subtitle":"Cost and accuracy trade-off: unsplit is cheaper in 2D, split-step becomes cheaper from four dimensions up.","key_machinery":"The Active Flux reconstruction: a compact, globally continuous finite-volume representation carrying cell averages plus shared point values on cell interfaces. Instead of Riemann solvers, the point values are advanced by characteristic tracing—backward ODEs for the unsplit method, successive one-dimensional advection updates for the split-step method. This reconstruction lets numerical fluxes be evaluated directly from the continuous representation, giving third-order accuracy with low dissipation on coarse grids.","core_discovery":"The paper's central finding is that the two Active Flux formulations for the one-dimensional 1D1V Vlasov–Poisson system have opposite strengths. The unsplit method is fully discrete: it solves the characteristic ODEs backwards in time, approximates the electric field within the step via a Taylor expansion using moment equations, and performs a single Poisson solve per step. The split-step method instead alternates x- and v-advection sub-steps, repeating Poisson solves and reconstruction evaluations. On coarse grids the unsplit method has slightly lower error against a high-resolution reference and lower cost; on fine grids the split-step method with an optimised fourth-order splitting is sli","pith_inferences":["The coarse-grid ranking rests on a single high-resolution reference; an independent convergence study could plausibly reverse the 'unsplit slightly better on coarse grids' conclusion.","The cost metric counts reconstruction term evaluations only; a full hardware comparison including Poisson solves and memory traffic might move the crossover dimension.","If the unsplit method is to reach Vlasov–Maxwell, it will need a different way to represent the field in time than the one-dimensional moment identities used here; the split method already has a demonstrated path.","For practitioners, a practical default could be: use the unsplit method for two-dimensional electrostatic runs on coarse grids, and the split-step method for high-resolution or four-dimensional-and-above simulations."],"forward_implications":["Both Active Flux variants provide third-order accurate, low-dissipation Vlasov–Poisson solutions that track a high-resolution reference through the linear and early nonlinear phase.","The unsplit method is cheaper in the two-dimensional case: one Poisson solve per step and fewer polynomial term evaluations than a nine-stage fourth-order splitting.","The split-step method extends naturally to higher dimensions and to Vlasov–Maxwell, while the unsplit method lacks an obvious generalisation of its electric-field time expansion beyond one-dimensional problems.","By the paper's term-evaluation count, the split method becomes the more cost-effective choice in four or more dimensions.","Using a splitting scheme with more stages improves split-step accuracy but also raises its cost, so the crossover dimension depends on the splitting chosen."],"fun_headline_variants":["Unsplit Active Flux better on coarse grids, split on fine","Grid size decides Active Flux choice for Vlasov-Poisson","Vlasov-Poisson: unsplit wins coarse, split wins fine","Active Flux trade-off: unsplit coarse, split fine","Coarse unsplit, fine split: Active Flux comparison"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The comparison treats one finely resolved run as the exact answer; if that reference is not converged, the reported error ranking between the two methods could be an artifact of comparing to a slightly wrong solution.","fun_headline_variants_meta":{"raw":{"variants":["Unsplit Active Flux better on coarse grids, split on fine","Grid size decides Active Flux choice for Vlasov-Poisson","Vlasov-Poisson: unsplit wins coarse, split wins fine","Active Flux trade-off: unsplit coarse, split fine","Coarse unsplit, fine split: Active Flux comparison"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000306,"raw_usage":{"total_tokens":1537,"prompt_tokens":640,"completion_tokens":897,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":384,"completion_tokens_details":{"reasoning_tokens":823}},"tokens_in":384,"tokens_out":897,"duration_ms":9245,"temperature":1.0,"reasoning_tokens":823,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:28:30.085077+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the two-stream instability with both methods against an independent, converged reference—for example, two references at different resolutions with Richardson extrapolation, or a manufactured solution—and compare the relative error at t>40 on 64x64 and 128x128 grids. If the unsplit method no longer shows lower error than the split-step method on coarse grids, the paper's central ranking would be refuted.","supporting_citations":[],"review_version":1}