{"id":"bd2482b6-4d0f-4462-851c-2c6929ba4ecf","arxiv_id":"2607.24457","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Saturated MRI turbulence in differentially rotating neutron stars yields ℓ_mix ≈ (0.01–0.1) λ_MRI, largely independent of density, so standard GRLES mixing-length prescriptions overestimate transport by about an order of magnitude.","lead":"High-resolution 3D GRMHD runs of differentially rotating neutron stars measure MRI turbulence transport coefficients and find the mixing length is roughly 0.01–0.1 times the MRI wavelength and nearly density-independent. That undercuts common large-eddy closures used in merger and supernova modeling and argues that dimensional-analysis subgrid recipes are only qualitative.","discovery_kind":"extension","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"The headline ratio ℓ_mix/λ_MRI ~ 10^-2–10^-1 implies ℓ_mix is at or below one grid cell; with λ_MRI itself pinned near the resolution floor and dissipation purely numerical, the \"physically transferable\" ratio may instead be set by grid-scale dissipation.","rationale":"The reader's weakest_assumption correctly identifies transferability under marginal Q_MRI as the soft spot; this critique sharpens it into a quantitative observation (ℓ_mix lands at 0.1–1 cells) and adds the numerical-dissipation mechanism, but it is the same underlying concern, so agreement is \"agree.\" The paper is otherwise a competent, honest measurement study: the averaging framework (Favre vs volume weighting, Eqs. 12–13), the antisymmetric α_DYN diagnostic (Fig. 4), and the cross-model consistency between B1 and B2 lend real internal support, and the authors themselves flag every major idealization (j-constant rotation, Γ=2 EOS, poloidal seed, Q_MRI near threshold). The concern is not that the work is wrong but that its one transferable number is asserted rather than demonstrated to be resolution-independent — precisely what the reader's CONDITIONAL verdict already hinges on. Hence UNCHANGED: the existing verdict pending resolution studies is the right posture, and the proposed test is exactly the convergence check that would settle it.","tokens_in":29917,"tokens_out":1917,"duration_ms":68958,"concrete_test":"Re-run model B1 at 2× finest resolution (Δx ≈ 74 m, Q_MRI ~ 20 at saturation) with identical seed field and EOS, and compare three quantities against the fiducial run over the same physical saturated window: (i) ℓ_mix/λ_MRI per density bin — does it remain in (10^-2–10^-1)? (ii) ℓ_mix in physical units — does it stay fixed or scale with Δx (if ℓ_mix ∝ Δx, it is grid-set)? (iii) saturated λ_MRI — does it drop toward smaller physical scales or stay pinned near ~10 cells? The ratio is trustworthy only if (i) is stable while (ii) and (iii) do not track Δx. As a cheap auxiliary check on existing snapshots, evaluate the neglected Reynolds channel ⟨(ρh+b²)δ(Wv^ϖ)δ(Wv^φ)⟩ of Eq. (18) to confirm it is subdominant to the Maxwell term used for ν_T.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative claim is that ℓ_mix saturates at a fixed fraction (0.01–0.1) of λ_MRI, density-independent, and that this ratio — not the absolute λ_MRI — is the \"robust and physically transferable result\" for subgrid calibration (§IVF, Conclusion). The weakest point is internal and quantitative: by the paper's own numbers, saturated λ_MRI ≈ 1.5 km with Δx = 147.7 m (model B1, §IVB), i.e. Q_MRI ≈ 10, the bare minimum for the linear phase and far short of what the nonlinear cascade demands (the paper cites this itself, §IIB). The claimed ratio then places ℓ_mix at 0.1–1.0 finest-grid cells — i.e. the measured \"mixing length\" is a sub-grid quantity inferred from ν_T = ⟨−b^ϖb^φ⟩/σ^ϖ_φ (Eq. 25), where both the stress and the strain are evaluated at resolved scales but the turbulent cascade that should determine ν_T is truncated at the grid. Because the runs are ideal-MHD with no explicit viscosity or resistivity, dissipation is entirely numerical; the saturated state — including the level of Maxwell stress and hence ν_T — is generically controlled by the grid-scale dissipation rate. Two quantities being individually grid-influenced does not guarantee their ratio is converged: there is no demonstrated inertial range separating the driving scale from the dissipation scale (λ_MRI ~ 10 cells leaves essentially no room for one), so the cancellation the authors rely on when promoting the ratio as transferable is an assumption, not a measurement. The authors are commendably explicit that λ_MRI is \"the characteristic scale the simulation resolves\" and that convergence tests are needed (Conclusion), but the paper still elevates the ratio to its headline subgrid-calibration result. A secondary, smaller concern: ℓ_mix is built from the Maxwell stress alone while the Reynolds channel of Eq. (18) is dropped \"as a modeling assumption, not a measured property of these runs\" — yet it is measurable from the same snapshots, and if it were non-negligible the effective ν_T","agreement_with_reader":"agree"},"referee_report":null,"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful content here is quantitative, not a new instability story. They run four 3D GR-Athena++ DR NS models (collapsing and long-lived), extract α_vis, ℓ_mix, and α_DYN from saturated Maxwell stress and fluctuation EMF, and report that ℓ_mix is largely density-independent and sits at ~0.01–0.1 λ_MRI—about an order of magnitude below the ℓ_mix ~ λ_MRI assumption used in calibrated GRLES. α_vis declines with density but with model-dependent slopes; α_DYN is antisymmetric as expected and only shallowly tracks v_A. Those numbers, and the density-independence claim, are what people building remnant subgrid models will actually use or argue with.\n\nWhat they do well: diagnostics are careful (Favre vs volume averages, Smagorinsky-type ν_T, least-squares α_DYN), saturation windows are stated, channel modes and dynamo antisymmetry are shown, and they compare collapsing vs non-collapsing cases. Circularity is low—they measure stresses and then compare to dimensional expectations rather than tuning a closure to reproduce itself. Idealizations (Γ=2, j-constant, pure poloidal seed, ideal MHD) are standard for this class and they own them in the conclusion.\n\nThe soft spot that matters is the one the stress note and the authors both hit. Saturated λ_MRI is only ~10 finest cells (B1: ~1.5 km vs Δx ~150 m), Q is near the linear-phase floor, dissipation is purely numerical, and the claimed ratio then puts ℓ_mix at roughly 0.1–1 cell. Promoting that ratio as the “physically transferable” calibration target is a stretch until higher-resolution runs show an inertial range. Dropping the Reynolds channel by assumption rather than measurement is a smaller, fixable gap—the same snapshots could have checked it. Neither sinks the paper; both belong in the referee report.\n\nThis is for people who write or use MRI subgrid models in BNS/CCSN remnants, not a general audience. Math and citation pattern look solid; self-cites to Radice GRLES are on-topic. I would send it to peer review and would cite the density-independence and ratio results with the resolution caveat attached. Worth a reading-group slot if anyone in the room is actively calibrating transport.","headline":"Useful coefficient measurements that challenge common GRLES closures, with the main caveat the authors already flag: the headline ratio may still be resolution-limited.","tokens_in":31110,"tokens_out":588,"would_cite":true,"duration_ms":20265,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"MRI turbulence in neutron stars mixes on a scale ten times smaller than subgrid models assume, so dimensional closures only work qualitatively.","keywords":["magnetorotational instability","neutron stars","GRMHD","turbulent transport","mixing length","mean-field dynamo","large-eddy simulation","differential rotation"],"falsifier":"A higher-resolution suite of the same models in which the quality factor is substantially increased and the saturated value of ℓ_mix/λ_MRI is shown either to stay near 0.01–0.1 or to rise toward unity.","tokens_in":30593,"feed_emoji":"🧲","tokens_out":880,"duration_ms":15886,"temperature":0.7,"pith_summary":"When a neutron star is spinning differentially, the magnetorotational instability drives turbulence that redistributes angular momentum and can decide whether the star collapses. Global simulations cannot resolve the tiny MRI wavelength, so large-eddy models insert effective transport coefficients chosen by dimensional analysis. This paper measures those coefficients directly in three-dimensional GRMHD runs of collapsing and non-collapsing stars. At saturation the mixing length tracks the MRI wavelength but sits an order of magnitude below the usual assumption that the two scales are comparable, and it shows almost no density dependence. The dynamo coefficient correlates only shallowly with Alfvén speed and both quantities fluctuate strongly in space and time. The practical claim is that dimensional closures can capture the qualitative effect of MRI turbulence, but quantitative predictions of collapse, jets, or ejecta require fully resolved GRMHD.","feed_headline":"MRI mixes neutron stars ten times weaker than models assume","feed_subtitle":"Direct GRMHD measurements show dimensional transport closures work only qualitatively","key_machinery":"Direct extraction of residual Maxwell stress and turbulent electromotive force from azimuthally averaged, fully resolved GRMHD snapshots, closed as a Smagorinsky-type eddy viscosity ν_T = ℓ_mix c_s and a mean-field α-dynamo coefficient; the measured ratio ℓ_mix/λ_MRI is the central diagnostic.","core_discovery":"At saturation, the effective mixing length of MRI-driven turbulence in differentially rotating neutron stars is largely independent of density and satisfies ℓ_mix ≈ (10^{-2}–10^{-1}) λ_MRI—roughly an order of magnitude below the common subgrid assumption ℓ_mix ∼ λ_MRI—while α_DYN correlates only shallowly with v_A; both coefficients are highly intermittent, so dimensional-analysis transport coefficients reproduce MRI effects only qualitatively and fully resolved GRMHD is required for quantitative predictions.","pith_inferences":["If the ratio remains small at higher resolution, calibrated LES for BNS remnants will need a reduced transport length rather than a simple rescaling of existing closures.","The absence of a clean density trend for ℓ_mix suggests that merger-remnant LES should treat the mixing length as a global fraction of the local driving scale, not a local function of ρ.","Intermittency of α_DYN implies that mean-field dynamo terms in long-term remnant evolutions may need stochastic or spatially fluctuating amplitudes."],"forward_implications":["Subgrid models that set ℓ_mix ∼ λ_MRI systematically overestimate turbulent angular-momentum transport in neutron-star interiors.","Density-dependent mixing-length prescriptions used in many GR large-eddy simulations are not supported by the measured turbulence.","Collapse time, large-scale field growth, and outflow launching in merger remnants cannot be predicted quantitatively from dimensional closures alone.","The same order-of-magnitude offset and intermittency appear in both collapsing and long-lived models once MRI turbulence is established."],"fun_headline_variants":["MRI mixing length in neutron stars falls 10x below model assumptions","Neutron-star MRI turbulence mixes far weaker than λ_MRI scaling predicts","Resolved GRMHD finds MRI mixing largely density-independent and intermittent","Dimensional MRI closures only qualitative; mixing length ~0.01–0.1 λ_MRI","MRI transport coefficients in NS deviate an order of magnitude from expectations"],"cache_read_input_tokens":128,"weakest_assumption_plain":"That the measured saturated ratio of mixing length to MRI wavelength is a physical property of the turbulence rather than an artifact of the grid, even though the saturated MRI wavelength itself is only a few cells across.","fun_headline_variants_meta":{"raw":{"variants":["MRI mixing length in neutron stars falls 10x below model assumptions","Neutron-star MRI turbulence mixes far weaker than λ_MRI scaling predicts","Resolved GRMHD finds MRI mixing largely density-independent and intermittent","Dimensional MRI closures only qualitative; mixing length ~0.01–0.1 λ_MRI","MRI transport coefficients in NS deviate an order of magnitude from expectations"]},"model":"grok-4.5","effort":"low","cost_usd":0.004478,"raw_usage":{"total_tokens":1360,"prompt_tokens":866,"num_sources_used":0,"completion_tokens":85,"cost_in_usd_ticks":44784000,"prompt_tokens_details":{"text_tokens":866,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":409,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":866,"tokens_out":85,"duration_ms":6825,"temperature":1.0,"reasoning_tokens":409,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T14:24:48.161740+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"A higher-resolution suite of the same models in which the quality factor is substantially increased and the saturated value of ℓ_mix/λ_MRI is shown either to stay near 0.01–0.1 or to rise toward unity.","supporting_citations":[],"review_version":1}