{"id":"af640989-37cf-457e-b61f-0e0c9d70cd66","arxiv_id":"1908.03972","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A P-multigrid solver for high-order flux reconstruction accelerates locally preconditioned low-Mach Navier-Stokes simulations, with an empirical rule for choosing polynomial-level hierarchies.","lead":"A new numerical solver combines established high-order methods with a multigrid accelerator to simulate air flowing very slowly around wings. The paper shows how to choose the multigrid's internal polynomial levels for the fastest convergence and tests the solver on airfoil and wing flows.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The P-multigrid speedup is measured against single-level residuals, not against fair per-cycle cost of the smoother at each P-level; the hierarchy recommendation may conflate multigrid acceleration with the cost of doing many more smoothing sweeps on cheap coarse levels.","rationale":"The strongest_claim is the design recommendation for the polynomial hierarchy; the reader flags it as resting on hand-tuned parameters and absent sensitivity analysis. My concern is more specific: the convergence ranking across hierarchies is not work-normalized because the fixed smoothing schedule I{5-10-20} spends the most iterations on the cheapest, lowest-degree levels, so the reported speedup could partly be extra smoothing effort rather than multigrid coarse-grid correction. This is a genuine soft spot, but it is not sufficient to reject: the paper's central claim is an engineering heuristic, the same qualitative ranking is broadly consistent across two equations, 2D/3D cases, three polynomial degrees, and is also physically plausible (too-large degree gaps fail because coarse-grid corrections are ineffective). The accuracy validation (Table 1, Table 2) is imperfect but broadly consistent with expected orders, and the SD7003 application agrees reasonably with prior results. I also credit the explicit acknowledgment of the cost at Ma=0.01 as honest (\"computational cost of smoothing when Ma=0.01 is approximately as much as six times...\"), which supports the paper's good-faith engineering framing. The concern therefore argues for caution in how strongly the design rule is stated, but the current CONDITIONAL verdict already captures that; an independent check should be run before treating the hierarchy recommendation as robust, but no verdict change is required.","tokens_in":20198,"tokens_out":1715,"duration_ms":17173,"concrete_test":"Re-run the convergence comparisons of Section 4.2 with smoothing counts held constant in terms of work, not in terms of iterations: choose I at the coarse level so that the total V-cycle cost (measured in fine-level-equivalent element-Jacobi sweeps, e.g., cost ~ sum over levels of I_level * DOF_level) is equal across hierarchies. For example, compare P{4-2} with I{5-10} against P{4-1} with total work matched by reducing the coarse-level sweeps. If P{4-2} still wins by a comparable margin at equal work, the hierarchy recommendation stands as a multigrid effect; if the margin shrinks or reverses, the headline claim is an artifact of the smoothing schedule.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative claim is that hierarchies near {P0-P0/2-P0} or {P0-P0/2-P0/4-P0/2-P0} give the best convergence acceleration. The evidence is residual-versus-CPU-time and residual-versus-V-cycle plots (Figures 4-6, 8-10, 12-13). However, the comparison across hierarchies is not normalized for the cost or effect of the smoothing sweeps themselves. For two-level V-cycles, I{5-10} always spends more explicit element-Jacobi sweeps on the coarse low-degree level than on the fine level, and for three-level cycles I{5-10-20} spends 20 sweeps on the lowest level. A hierarchy with very different P-levels is therefore doing more smoothing operations on cheaper, lower-degree spaces; part of the apparent speedup of {P4-2}, {P5-3}, {P5-2} or {P4-2-1} may simply be the extra smoothing effort concentrated at low polynomial degree, rather than a multigrid coarse-grid correction effect. The paper does not report the cost of a smoothing sweep at each P-level, the work-per-V-cycle for each hierarchy, or any two-grid analysis (e.g., using the smoother alone on the coarse level as a reference). Because the recommended hierarchy is selected by ranking such plots, the design rule is not actually isolated from the smoothing schedule.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript develops an implicit P-multigrid solver for high-order flux reconstruction/correction procedure via reconstruction (FR/CPR) discretizations of the locally preconditioned unsteady compressible Navier–Stokes equations at low Mach numbers. The solver combines a FAS-type V-cycle with an element Jacobi smoother and ESDIRK temporal integration, with local preconditioning applied only in the pseudo-time continuation to preserve time accuracy. Accuracy is verified on an isentropic vortex at Ma=0.005, the influence of polynomial-degree hierarchies is studied on 2D inviscid and viscous NACA0012 flows and a 3D sphere flow at Ma=0.001, and the solver is applied to an under-resolved transitional SD7003 wing at Re=60000 and Ma=0.1/0.01. The authors recommend hierarchies close to {P0-P0/2-P0} for two-level and {P0-P0/2-P0/4-P0/2-P0} for three-level V-cycles as giving the best convergence acceleration.","tokens_in":20563,"tokens_out":6101,"duration_ms":60737,"significance":"If the empirical hierarchy rule is robust, the paper offers a practical acceleration strategy for high-order low-Mach FR/CPR solvers, addressing a known stiffness problem in a way that is directly useful for coarse-resolution turbulence simulation. The manuscript is clearly organized, provides a complete algorithmic description, and includes a challenging 3D application with quantitative comparisons to previous experimental and numerical results. The central claims are conditional, however: the order-of-accuracy verification is incomplete for density and pressure at the finest resolutions, and the hierarchy recommendation is obtained with a fixed set of solver parameters and without work-per-cycle normalization, so its generality is not yet established.","major_comments":[{"comment":"The order-of-accuracy verification does not fully support the claim of verified high-order spatiotemporal accuracy. In Table 1, ESDIRK4 shows order drops to 0.79 for density and 1.39 for pressure at the finest time step, while velocity retains order 4.00. In Table 2, the P3 and P4 FR schemes similarly show density and pressure order collapse at the finest grid (e.g., P3 density order 0.49, P4 density order 0.02), while velocity continues to converge. The authors attribute these reductions to \"the accuracy limit of the solver,\" but do not identify this limit or explain why velocity is unaffected. Because the abstract and conclusions state that high-order spatiotemporal accuracy is verified, this inconsistency must be addressed, either by tightening the pseudo-time tolerance, analyzing the effect of the 1e-4 tolerance on each variable, or rephrasing the accuracy claim.","section":"§4.1, Tables 1 and 2"},{"comment":"The central hierarchy recommendation is based on a fixed set of solver parameters (I{5-10-20}, alpha_r=1, SER growth ratio 1.5, CFLinit=1e2, CFLmax=1e5, kappa=1) with no sensitivity study. It is therefore unclear whether the observed ranking of hierarchies, e.g., P{4-2} over P{4-1} or P{5-3-1} over P{5-4-1}, persists for other smoothing schedules, relaxation parameters, or CFL update strategies. A sensitivity study or a two-grid analysis is needed to show that the recommended rule \"polynomial degree difference close to half of the finer degree\" is a property of the multigrid correction rather than an artifact of the chosen parameters.","section":"§4.2, Figures 4–6, 8–10, 12–13"},{"comment":"The comparison across hierarchies is not normalized for the cost of smoothing sweeps at different polynomial degrees. All two-level cases use I{5-10} and all three-level cases use I{5-10-20}, so a V-cycle for P{5-4-1} performs the same number of sweeps as one for P{5-2-1}, but the sweeps at P4 are much more expensive than those at P2 or P1. The residual-versus-V-cycle plots partially mitigate this, but the CPU-time comparisons would be more interpretable if the paper reported the cost per smoothing sweep at each degree and the total work per V-cycle for each hierarchy. Without this information, the reader cannot separate the benefit of coarse-grid correction from the effect of concentrating smoothing effort on cheaper lower-degree spaces.","section":"§4.2, CPU-time and V-cycle comparisons"}],"minor_comments":[{"comment":"There is a typo: \"The ensure conservation\" should read \"To ensure conservation.\"","section":"§2.3, Eq. (15)"},{"comment":"The statement that P{3-2} and P{3-1} have \"almost the same computational cost\" is vague, since P{3-1} uses a cheaper coarse level; please clarify whether the similar cost arises because P{3-1} requires more V-cycles, which is not apparent from the figures.","section":"§4.2.1"},{"comment":"The choice of the global cutoff parameter kappa=2.5 for Ma=0.01, while kappa=1.0 for Ma=0.1, is stated without justification; a brief explanation of how this cutoff value was selected would improve reproducibility.","section":"§4.3"},{"comment":"The axis labels contain typos: \"Peudo iteration m\" and \"resisual\" should be \"Pseudo iteration m\" and \"residual.\"","section":"Figure 18"},{"comment":"The restriction and prolongation operators (PP0_P1 and IP1_P0) are not defined; specifying whether they are L2 projections or interpolation operators would make the algorithm fully reproducible.","section":"§3"},{"comment":"The convergence histories in Figures 4–6 are shown down to residual 1e-9, but the stopping criterion or target tolerance for the steady-state cases is not stated; please include it.","section":"§4.2.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for physics.comp-ph and represents a useful incremental contribution. The main risk is that the hierarchy rule is presented as a general design guideline without sensitivity analysis; the authors should either add a parameter-sensitivity study or explicitly limit the claim to the tested configuration. The order-of-accuracy issue in Tables 1 and 2 should be resolved before the paper can claim verified high-order accuracy. I would not reject the manuscript on these grounds, as the issues appear addressable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague — this is a solid engineering paper, not a landmark. The new contribution is a working integration of local preconditioning, FR/CPR, ESDIRK, and P-multigrid into an unsteady low-Mach solver, plus a systematic empirical study of polynomial-degree hierarchies across 2D/3D, inviscid/viscous, and steady/unsteady cases. The consistent finding — use about P0-P0/2-P0 for two-level cycles and P0-P0/2-P0/4-P0/2-P0 for three-level cycles — is useful design guidance, even if it remains empirical. Those who build implicit high-order low-Mach solvers will get a practical recipe and a fair set of benchmarks.\n\nWhat it does well: accuracy is checked against an analytic isentropic vortex at Ma = 0.005, and the SD7003 transitional-flow results are compared with several published numerical datasets and one experiment. The paper is honest about order-of-accuracy reductions and about not knowing whether the Mach-number effect on the separation bubble is physical or an artifact of the preconditioning. Residual histories are shown against both V-cycles and CPU time, so the hierarchy comparisons are not just cycle counts.\n\nSoft spots. First, the order-of-accuracy tables are not clean: ESDIRK4 drops below design order for density and pressure on the finest time step, and the spatial refinement study shows order reductions for rho and p, attributed by the authors to the solver’s accuracy limit. Velocity errors are clean; rho and p are not. The limitation is stated, but it means the “high-order spatiotemporal accuracy” claim is only partially demonstrated.\n\nSecond, the hierarchy recommendation comes from a fixed smoothing schedule I{5-10-20}, an element Jacobi smoother with alpha_r = 1, a fixed SER growth ratio, and no sensitivity study. The stress-test concern about cost accounting mostly does not hold: the CPU-time plots include the cost of all smoothing sweeps, so a hierarchy that wins on CPU time is a fair engineering result. But the residual-versus-V-cycle plots are not cost-normalized, and there is no two-grid or smoother-only baseline, so we cannot separate coarse-grid correction from the effect of spending many cheap sweeps on low-degree levels. The recommendation is probably robust within this solver family — the pattern is consistent across several configurations — but it is not isolated from the smoother settings.\n\nThird, no code or data is shipped. That is normal for this subfield, but it makes the empirical rule harder to check.\n\nBottom line: send it to peer review. It deserves referee time. The main requests should be a sensitivity study of the hierarchy rule to smoother parameters and a cleaner accuracy study for rho and p.","headline":"Solid engineering paper worth a serious referee: packages known components into a working low-Mach P-multigrid solver and gives an empirically useful hierarchy rule, though the accuracy study has warts and the design rule lacks sensitivity analysis.","tokens_in":21040,"tokens_out":2539,"would_cite":true,"duration_ms":29298,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A P-multigrid solver for high-order flux reconstruction reaches its fastest convergence for low-Mach unsteady flows when coarse levels use polynomial degrees about half the fine degree, and it preserves high-order accuracy.","keywords":["P-multigrid","flux reconstruction","FR/CPR","low Mach number","local preconditioning","ESDIRK","unsteady Navier-Stokes","under-resolved simulation"],"falsifier":"Rerun the NACA0012 convergence comparisons with a different smoother configuration, for example $\\alpha_r=0.5$, smoothing counts $\\{10-20-40\\}$, or CFL growth ratio 2.0, and check whether the fastest hierarchy shifts away from $\\{P_0-P_0/2-P_0\\}$ and $\\{P_0-P_0/2-P_0/4-P_0/2-P_0\\}$. If the ranking changes, the claimed design rule is not robust.","tokens_in":19965,"feed_emoji":"🌀","tokens_out":7311,"duration_ms":67931,"temperature":0.7,"pith_summary":"The paper develops an implicit solver for low-Mach unsteady compressible flows using high-order flux reconstruction with ESDIRK time stepping and local preconditioning confined to pseudo-time. It claims that P-multigrid acceleration works best when the polynomial degree roughly halves between adjacent levels, and that this rule holds for two- and three-level V-cycles across P3, P4, and P5 discretizations and across 2D and 3D test cases. The payoff if true is a simple, practical recipe for reducing the cost of high-order low-Mach simulations, with demonstrated application to under-resolved transitional flow over a wing.","feed_headline":"Halved-degree multigrid speeds low-Mach flow solves","feed_subtitle":"Coarse levels at half the fine polynomial degree converge fastest in high-order low-Mach simulations.","key_machinery":"The central object is the P-multigrid V-cycle on a hierarchy of polynomial spaces, with an element Jacobi smoother performing a few relaxed updates per level. The smoother update is $q^{m+1}=q^m+\\alpha_r D^{-1}F$, where $D$ is the block diagonal of the linearized pseudo-time operator and $\\alpha_r=1$ in this study. The load-bearing idea is the degree-halving hierarchy: adjacent levels should differ by roughly half the finer polynomial degree, so coarse-grid corrections stay effective instead of introducing new errors. This rule is what carries the convergence acceleration.","core_discovery":"The paper establishes that an implicit P-multigrid V-cycle can be combined with locally preconditioned flux reconstruction and ESDIRK time integration without destroying high-order accuracy. Its central finding is empirical: the fastest convergence for both two-level and three-level cycles occurs when coarse polynomial degrees are near half the fine degree, specifically hierarchies close to $\\{P_0-P_0/2-P_0\\}$ and $\\{P_0-P_0/2-P_0/4-P_0/2-P_0\\}$. In the tested cases, three-level cycles beat two-level cycles, and excessive degree gaps such as $\\{P_5-P_1\\}$ either stall or contaminate convergence. Local preconditioning appears only in the pseudo-transient-continuation loop, so the unsteady physical time integration retains the design order of the ESDIRK method.","pith_inferences":["The degree-halving rule likely transfers to other compact high-order discretizations such as discontinuous Galerkin or spectral difference, because the mechanism is coarse-level correction effectiveness rather than the flux-reconstruction formulation; this is an inference, not tested in the paper.","A two-grid local-mode analysis of the element Jacobi smoother with preconditioned eigenvalues could turn the empirical rule into a predictive criterion and show how the optimal ratio shifts with cell aspect ratio or Reynolds number.","The paper's comparison of Ma = 0.1 and Ma = 0.01 wing flows leaves the cause of the difference in lift and separation bubble ambiguous; a controlled study with matched cutoff parameter and matched numerical dissipation could separate weak-compressibility effects from solver-settings effects.","Extending the solver to Newton-Krylov with P-multigrid as a preconditioner, which the paper names as future work, is the natural test of whether the same hierarchy rule accelerates very-low-Mach nonlinear solves."],"forward_implications":["For a chosen fine polynomial degree, setting the coarse level near $P_0/2$ and the next near $P_0/4$ for three levels should cut both CPU time and V-cycle count relative to single-level iteration.","Three-level cycles are preferable to two-level cycles in the tested low-Mach regime.","Hierarchies with large adjacent degree gaps should be avoided, since they can stall convergence or cause residual oscillation.","The combination of pseudo-time local preconditioning and P-multigrid preserves nominal high-order spatiotemporal accuracy at Ma = 0.005.","The solver supports under-resolved transitional low-Mach wing simulations, with lift, drag, and separation/reattachment predictions consistent with prior studies."],"supporting_citations":[{"why":"Supplies the primitive-variable preconditioning matrix and reference-velocity cutoff that remove low-Mach stiffness.","marker":"[7]"},{"why":"Demonstrates the baseline implicit preconditioned flux reconstruction method with dynamic meshes that this paper extends with P-multigrid.","marker":"[8]"},{"why":"Introduces the flux reconstruction spatial discretization used throughout.","marker":"[9]"},{"why":"Provides the correction-procedure formulation that defines the discrete residual and correction field.","marker":"[11]"},{"why":"Contributes the element Jacobi and line-Jacobi smoother and the V-cycle p-multigrid machinery for high-order discretizations.","marker":"[18]"},{"why":"Supplies the unstructured-grid V-cycle forcing-term and prolongation steps adopted in the algorithm.","marker":"[20]"},{"why":"Supports the choice of P-multigrid with explicit and implicit smoothers and hierarchy testing.","marker":"[21]"},{"why":"Provides the ESDIRK3 and ESDIRK4 high-order implicit time integrators used for unsteady accuracy.","marker":"[13]"},{"why":"Preliminary conference study of polynomial-degree hierarchies for steady flows whose conclusions are here extended to unsteady low-Mach cases.","marker":"[52]"},{"why":"Defines the SER CFL evolution algorithm used in the pseudo-transient continuation.","marker":"[56]"}],"fun_headline_variants":["Half-degree P-multigrid levels fastest for low-Mach flow","Three-level P-multigrid beats two-level at low Mach","Low-Mach solver speeds up with halved polynomial degree","Implicit P-multigrid accelerates low-Mach unsteady flows","Half-degree coarse levels speed up high-order low-Mach solver"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the hierarchy ranking observed with the paper's fixed solver settings, including Jacobi relaxation $\\alpha_r=1$, smoothing counts $\\{5-10-20\\}$, SER CFL growth ratio 1.5, smoother update intervals, and cutoff parameter $\\kappa$, remains valid when those settings change; if the ranking is an artifact of the tuning, the recommended degree-halving rule collapses.","fun_headline_variants_meta":{"raw":{"variants":["Half-degree P-multigrid levels fastest for low-Mach flow","Three-level P-multigrid beats two-level at low Mach","Low-Mach solver speeds up with halved polynomial degree","Implicit P-multigrid accelerates low-Mach unsteady flows","Half-degree coarse levels speed up high-order low-Mach solver"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000202,"raw_usage":{"total_tokens":1435,"prompt_tokens":1053,"completion_tokens":382,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":669,"completion_tokens_details":{"reasoning_tokens":292}},"tokens_in":669,"tokens_out":382,"duration_ms":4270,"temperature":1.0,"reasoning_tokens":292,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:56:10.343371+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Rerun the NACA0012 convergence comparisons with a different smoother configuration, for example $\\alpha_r=0.5$, smoothing counts $\\{10-20-40\\}$, or CFL growth ratio 2.0, and check whether the fastest hierarchy shifts away from $\\{P_0-P_0/2-P_0\\}$ and $\\{P_0-P_0/2-P_0/4-P_0/2-P_0\\}$. If the ranking changes, the claimed design rule is not robust.","supporting_citations":[{"cited_title":"Preconditioning applied to variable and constant density Flows,","cited_arxiv_id":null,"evidence_quote":"Supplies the primitive-variable preconditioning matrix and reference-velocity cutoff that remove low-Mach stiffness."},{"cited_title":"An Implicit High-Order Preconditioned Flux Re- construction Method for Low-Mach-Number Flow Simulation with Dy- namic Meshes,","cited_arxiv_id":null,"evidence_quote":"Demonstrates the baseline implicit preconditioned flux reconstruction method with dynamic meshes that this paper extends with P-multigrid."},{"cited_title":"A flux reconstruction approach to high-order schemes including discontinuous Galerkin methods,","cited_arxiv_id":null,"evidence_quote":"Introduces the flux reconstruction spatial discretization used throughout."},{"cited_title":"A unifying lifting collocation penalty formulation including the discontinuous Galerkin, spectral vol- ume/difference methods for conservation laws on mixed grids,","cited_arxiv_id":null,"evidence_quote":"Provides the correction-procedure formulation that defines the discrete residual and correction field."},{"cited_title":"p-Multigrid solution of high-order discontinuous Galerkin discretizations of the com- pressible Navier–Stokes equations,","cited_arxiv_id":null,"evidence_quote":"Contributes the element Jacobi and line-Jacobi smoother and the V-cycle p-multigrid machinery for high-order discretizations."},{"cited_title":"A p-multigrid discontinuous Galerkin method for the Euler equations on unstructured grids,","cited_arxiv_id":null,"evidence_quote":"Supplies the unstructured-grid V-cycle forcing-term and prolongation steps adopted in the algorithm."},{"cited_title":"A p-multigrid spectral difference method with explicit and implicit smoothers on unstructured triangular grids,","cited_arxiv_id":null,"evidence_quote":"Supports the choice of P-multigrid with explicit and implicit smoothers and hierarchy testing."},{"cited_title":"Implicit time integration schemes for the unsteady compressible Navier–Stokes equations: laminar flow,","cited_arxiv_id":null,"evidence_quote":"Provides the ESDIRK3 and ESDIRK4 high-order implicit time integrators used for unsteady accuracy."},{"cited_title":"A p-multigrid flux reconstruction method for the steady Navier-Stokes equations,","cited_arxiv_id":null,"evidence_quote":"Preliminary conference study of polynomial-degree hierarchies for steady flows whose conclusions are here extended to unsteady low-Mach cases."},{"cited_title":"Experiments with implicit upwind methods for the Euler equations,","cited_arxiv_id":null,"evidence_quote":"Defines the SER CFL evolution algorithm used in the pseudo-transient continuation."}],"review_version":1}