{"id":"5ccab56c-452f-415b-b42f-f7b03aa38faf","arxiv_id":"1908.01419","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A gradient sub-grid-scale model is derived for relativistic MHD and shows high correlation with unresolved residuals in Kelvin-Helmholtz simulations.","lead":"The authors derive the first complete set of small-scale correction terms for simulations of relativistic magnetized plasma turbulence. The terms match what high-resolution test simulations lose, suggesting cheaper simulations could capture physics that full-resolution runs currently miss.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The validation formula in Eq. (62) is not the box-filter SFS residual as stated; as written it gives a nonzero offset for constant fields, so Figs. 4-5 may not measure what the central claim asserts.","rationale":"The derivation of the SGS tensors is a substantial, self-contained contribution: it extends the gradient closure to general conservative systems with non-invertible conserved-to-primitive maps, and the non-relativistic limit checks against known expressions. I am not claiming the numerical results are fabricated; the mismatch in Eq. (62) may be a typo. But the paper's lack of released code and a-posteriori tests means the reader cannot adjudicate. The reader's CONDITIONAL verdict already captures the missing a-posteriori LES; my concern is more elementary: the printed formula for the target residual is not the SFS residual of the stated box filter, so the quantitative anchor of the central claim needs explicit verification. This does not force a rejection because the underlying derivation is still valuable, but it strengthens the conditions: the authors should correct or clarify Eq. (62) and provide the verification code or data. If the corrected computation reproduces the figures, the central numerical claim survives at the same conditional status; if it does not, the effective-resolution conclusion would need to be withdrawn or re-derived.","tokens_in":20348,"tokens_out":11820,"duration_ms":133111,"concrete_test":"Run the residual computation from Eq. (62) on a constant-field control case: it should return exactly zero for every filter factor S_f. Then recompute Figs. 4-5 using the correct box-filter residual, e.g. tau_fg(x_f) = S_f^{-3} sum_i f_i g_i - (S_f^{-3} sum_i f_i)(S_f^{-3} sum_i g_i) or its sign-per Eq. (20) counterpart, and check whether P >= 0.8 at S_f=2, P >= 0.5 at S_f=16, and C_best about 1 survive. If the stored snapshots reproduce the reported values under the corrected formula, the concern is resolved; if not, the central numerical claim is unsupported. Releasing the small SGS-verification script would settle the ambiguity permanently.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative claim rests entirely on the a-priori fits in Figs. 4-5. The target residual is evaluated by Eq. (62), which reads tau = S_f^{-3} [sum_i f_i sum_i g_i - sum_i f_i g_i]. With the box filter of Section VII.A, sum_i f_i = S_f^3 bar-f and sum_i f_i g_i = S_f^3 overline{fg}, so this expression equals S_f^3 bar-f bar-g - overline{fg}, not the residual bar-f bar-g - overline{fg} of the filter, nor its negative. For constant fields it returns S_f^3 - 1 instead of 0. Unless the code implements a different formula than the one printed, the Pearson coefficients and best-fit coefficients reported for tau_N, tau_T, and tau_M do not compare the SGS model with the SFS residuals defined in Eq. (20). Since no code or analysis scripts are released, a reader cannot determine whether this is a typographical error or an actual computational bug. This is the load-bearing weak point: the derived algebraic expressions may well be correct, but the numerical evidence for C_eff about 1 and the claimed factor-4-to-8 resolution gain depends on an unverifiable and, as printed, inconsistent residual evaluation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript extends the gradient sub-grid-scale (SGS) model to special-relativistic ideal MHD. Starting from a Gaussian-filter expansion of a generic conservative system, it derives closed-form SGS tensors for the continuity, momentum, energy, and induction equations [Eqs. (47)-(54)], checks the non-relativistic limit, and presents an a-priori validation using filtered snapshots of decaying Kelvin-Helmholtz turbulence at resolutions 128^3-1024^3. The validation reports Pearson correlations P≈0.8 or higher for filter factor S_f=2 and P≳0.5 for S_f=16, with best-fit coefficients Cbest≈1, and the authors infer an effective resolution gain of 4 to 8. The derivation is a mathematical closure and is not a fit to the simulation data, but the a-priori test is the only numerical evidence for the model's performance.","tokens_in":20578,"tokens_out":8634,"duration_ms":87378,"significance":"If the quantitative claims hold, this would be the first gradient SGS formulation for relativistic MHD with a general equation of state, with a plausible path to GRMHD binary-merger simulations and a possible factor-of-four-to-eight effective resolution gain. The derivation is a genuine extension: it confronts the nontrivial conserved-to-primitive inversion and reproduces known non-relativistic results as a consistency check. The model contains no fitted free parameters in its functional form; only the empirical amplitude Cbest is adjusted in the validation, which is an appropriate a-priori procedure. However, the numerical evidence is not yet reliable and no code or analysis scripts are provided, so the central validation must be regarded as unverified.","major_comments":[{"comment":"The formula used to evaluate the a-priori SFS residual is inconsistent with the box-filter definition stated in the same section. With bar-f = (1/S_f^3) sum_i f_i and overline{fg} = (1/S_f^3) sum_i f_i g_i, the printed right-hand side equals S_f^3 bar-f bar-g - overline{fg}, not the SFS residual bar-f bar-g - overline{fg} of Eq. (20) and the preceding illustrative definition. For constant fields it returns S_f^3 - 1 instead of 0. Since Figs. 4 and 5 and the reported values P≥0.8 and Cbest≈1 are computed from this expression, the central quantitative claim is not supported as printed. The authors should correct Eq. (62) (or clarify the normalization of the sums), rerun the fits, and, preferably, release the analysis code so that the implemented expression can be checked.","section":"§VII.A, Eq. (62)"},{"comment":"The validation is purely a-priori, and the conclusion overreaches when it states that the model yields a gain in effective resolution of at least a factor 4 and possibly up to a factor 8. The high-resolution reference runs are not a clean representation of the unresolved scales: §VI.C reports a change of slope in all energy spectra at high k due to numerical dissipation and describes the discretization as an implicit LES. The filtered residuals from such runs therefore mix numerical dissipation with true sub-filter physics. An a-posteriori LES with the SGS terms included, or at least a convergence study showing that the fitted residuals are insensitive to the reference resolution, is needed before the resolution-gain estimate can be taken as established.","section":"§VIII vs. §VI.C"},{"comment":"The first-order gradient expansion is applied for filter factors up to S_f=16, for which the expansion parameter xi k^2 is not small relative to unity at the resolved grid scale. The paper does not test whether the second-order terms in Eqs. (15)-(16) change the predicted tensors, so the regime of validity of the truncation is unclear. The reported P≥0.5 at S_f=16 is encouraging but should be accompanied by a filter-size convergence check before the model is claimed to represent scales up to 16 grid cells.","section":"§III and §VII.B"}],"minor_comments":[{"comment":"The auxiliary expressions (51)-(54) are presented without derivation or verification; a short appendix showing how they follow from the inverse-function-theorem Jacobians would help readers trust and reimplement them.","section":"§V.B, after Eq. (50)"},{"comment":"The initial-condition line 'Bx = Bx0, Bx = By0, Bz = Bz0' contains a typo: the second equality should presumably be By = By0.","section":"§VI.A, Eq. (59)"},{"comment":"The statement that 'we have explored different initial conditions, finding mainly the same results' is not supported by any figure or table; please show at least one representative additional case or remove the claim.","section":"§VII.B"},{"comment":"Minor typos include 'Large-Eddie-Simulations' and 'Minkovski metric' in the abstract/introduction; these should be corrected.","section":"§II"}],"recommendation":"major_revision","confidential_remarks":"The paper's novelty is real and the derivation appears sound, but the main numerical validation hinges on a single printed formula whose normalization is wrong. If the authors can correct it and confirm the reported numbers, or release the scripts, the paper could become acceptable. I would also ask the editor to require that the analysis code or data products be made available for verification."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take. The genuinely new thing is real: this is the first gradient sub-grid-scale closure for special-relativistic ideal MHD with arbitrary equation of state, and the general balance-law formulation with the conserved-to-primitive inversion is a proper extension, not a cosmetic tweak. The non-relativistic limit checks out, and the authors are upfront about gauge dependence, shocks, and numerical dissipation in their own high-resolution runs. That part deserves credit.\n\nThe soft spot is the validation. Eq. (62) as printed is not the SFS residual of Eq. (20). With a box filter, sum_i f_i = S_f^3 * bar_f and sum_i (f_i g_i) = S_f^3 * overline{fg}, so the printed expression equals S_f^3 * bar_f * bar_g - overline{fg}, not bar_f * bar_g - overline{fg}. For constant fields it gives (S_f^3 - 1) f g rather than zero. If the code implements what is printed, Figs. 4-5 are not measuring the advertised residuals. If it is a typo and the code is correct, the Pearson values may be fine, but no code or analysis scripts are released, so a reader cannot tell. This matters because the central quantitative claim - Cbest ~ 1 and a resolution gain of 4 to 8 - rests entirely on those fits.\n\nThe validation is also a-priori only: Cbest is fitted to the same residuals it is then judged against, there are no error bars, and no actual LES with the SGS term is run. The factor-4-to-8 gain is an extrapolation, not a demonstrated result. Equations (51)-(54) are dense but appear internally consistent; they are the sort of expressions that need a companion derivation or a reproducibility artifact.\n\nThe citation pattern is fine. The non-relativistic gradient model and the earlier Smagorinsky GRMHD work are properly acknowledged, and the self-citation to Ref. [28] is for the model being extended, which is legitimate.\n\nWho is this for: numerical relativists and code developers who want to try explicit LES in GRMHD mergers. The derivation alone is a contribution. But the paper currently overclaims on the strength of an a-priori test with a possibly misprinted residual formula. I would send it to peer review with the expectation of major revision: fix Eq. (62), show the constant-field sanity check, release the filtering scripts, and either add a minimal a-posteriori test or soften the resolution-gain claim. I would not desk-reject it.","headline":"Novel and useful derivation of a relativistic MHD gradient SGS closure, but the a-priori validation is compromised by an apparent error in Eq. (62) and by missing code or a-posteriori tests, so the resolution-gain claim is overstated.","tokens_in":21141,"tokens_out":3475,"would_cite":true,"duration_ms":34646,"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":"Gradient model supplies the missing small-scale terms in relativistic MHD turbulence.","keywords":["sub-grid-scale model","gradient model","relativistic magnetohydrodynamics","large eddy simulation","Kelvin-Helmholtz instability","MHD turbulence","binary neutron star mergers","a-priori filtering test"],"falsifier":"Run an actual coarse-grid LES with the SGS tensors (47)--(54) for the same Kelvin-Helmholtz setup and compare its kinetic and magnetic energy spectra with the $1024^3$ DNS; if adding the model fails to reproduce the magnetic-energy growth and spectra of the fine run, or if it destabilizes the scheme, the central claim is refuted. A cheaper check is to compute the Pearson coefficient for a filter factor above 16 or across a strong shock, where the gradient expansion and the shock-capturing dissipation are expected to break down.","tokens_in":20120,"feed_emoji":"🌪️","tokens_out":6758,"duration_ms":60888,"temperature":0.7,"pith_summary":"This paper extends the gradient sub-grid-scale (SGS) model, a closure based on the Taylor expansion of the spatial filter, to the full special-relativistic ideal magnetohydrodynamics (MHD) system with a general equation of state. The authors derive explicit closed-form expressions for the sub-filter-scale residuals that appear in the filtered continuity, momentum, energy, and induction equations when the conserved-to-primitive inversion is not analytic. They then test these expressions a priori on box simulations of the relativistic Kelvin-Helmholtz instability at resolutions from $128^3$ to $1024^3$, comparing the modeled tensors with the residuals obtained by filtering the high-resolution data. The central quantitative result is that the gradient tensors match the true residuals with Pearson coefficients above about $0.8$ for a filter factor of $2$ and still above about $0.5$ for a filter factor of $16$, with best-fit pre-coefficients close to $1$. If this carries over to an actual simulation, the model would let a coarse grid reproduce dynamics normally requiring four to eight times finer resolution, which is directly relevant to magnetic-field amplification in binary neutron star mergers.","feed_headline":"Gradient model closes small-scale gap in relativistic MHD simulations","feed_subtitle":"Matches true sub-filter residuals with correlation above 0.8; promises a 4-8x effective resolution gain.","key_machinery":"The load-bearing object is the double-gradient operator $H(X)=\\nabla(dX/dC^b)\\cdot\\nabla C^b$, obtained from the gradient expansion of the inverse Gaussian filter $G^{-1}=1+\\xi\\nabla^2+O(\\xi^2)$. It has the property $H(C^a)=0$ for conserved fields and obeys a generalized Leibniz rule $H(XY)=XH(Y)+YH(X)+2\\nabla X\\cdot\\nabla Y$, which is what lets products like $hW^2 v^i v^k$ be expanded without explicitly inverting $C^a(P)$. The whole SGS model is the statement $\\tau_X \\simeq -\\xi H(X)$ for any field $X$, with equations (47)--(54) giving the explicit $H$ terms for the relativistic MHD fluxes.","core_discovery":"The claim is that for a generic system of conservation laws $\\partial_t C^a + \\partial_k F^a_k(P)=0$, including the relativistic ideal MHD case where the map $C^a=f^a(P)$ is not analytically invertible, every sub-filter-scale residual can be approximated to first order in the filter width by an expression of the form $\\tau^a_k=-\\xi\\,\\nabla(dF^a_k/dC^b)\\cdot\\nabla C^b$, with $\\xi$ determined by the filter width. Applied to special-relativistic ideal MHD, this yields the SGS tensors in Eqs. (47)--(50) together with the auxiliary double-gradient terms (51)--(54), which involve the primitive fields, their gradients, and the Jacobian of the inverse transformation. The a-priori comparison against filtered high-resolution Kelvin-Helmholtz turbulence shows high correlation and order-one coefficients, which the authors take as evidence that the functional form captures the unresolved dynamics for filter factors up to 16.","pith_inferences":["The a-priori test establishes correlation, not causation: the decisive evidence would be an a-posteriori LES in which the SGS tensors are actually evolved, and that test is the natural next step implied by this paper.","The formulation assigns a SGS term to the continuity equation while the energy equation gets none; if that asymmetry is physical, it predicts that unresolved dynamics preferentially alter rest-mass transport, which could be checked by comparing coarse LES and fine DNS mass-flux statistics.","Because the filtering operation is tied to the 3+1 foliation, a gauge-independent SGS model for GRMHD may be impossible by construction; implementations will likely need to recompute the tensors for each gauge choice.","The model's backscatter capability is its main advantage over purely dissipative closures, but it also means stability is not guaranteed; a testable extension is to quantify how much numerical dissipation must be added to offset the model's anti-diffusive contributions."],"forward_implications":["For any equation of state, the filtered relativistic ideal MHD equations can be closed with explicit SGS terms that require only gradients of the primitive fields and the Jacobian of the conserved-to-primitive map, so implementation is a matter of adding terms to an existing code.","Because best-fit pre-coefficients stay close to 1 and vary little in time, a fixed pre-factor, without a dynamic Germano-type procedure, is sufficient in this Kelvin-Helmholtz regime.","The retained correlation at $S_f=16$ indicates the model can substitute for a resolution increase of a factor 4 to 8, which translates directly into a computational-cost saving in production simulations.","Extending the derivation from Minkowski to a curved background is expected to be straightforward because the metric varies more smoothly than the turbulent fields, so the same SGS tensors can be used in GRMHD binary-neutron-star simulations.","The model is not a substitute for resolution near strong shocks, where the numerical dissipation of high-resolution shock-capturing schemes overwhelms the SGS terms."],"supporting_citations":[{"why":"Defines the standard purely dissipative closure that the gradient model aims to replace and that is the baseline for comparison.","marker":"[25]"},{"why":"Introduces the Taylor or gradient expansion of sub-grid-scale terms that the paper generalizes.","marker":"[26]"},{"why":"Supplies the gradient-diffusion SGS model for incompressible MHD whose relativistic extension is derived here.","marker":"[27]"},{"why":"Previous non-relativistic compressible MHD a-priori fitting; supplies the methodology and the comparison baseline.","marker":"[28]"},{"why":"Previous explicit LES for GRMHD with a dissipative closure; the relativistic comparison target.","marker":"[24]"},{"why":"Provides the inverse-filter gradient-expansion series used to derive the first-order closure.","marker":"[37]"},{"why":"Local simulations of the magnetized Kelvin-Helmholtz instability in neutron-star mergers; defines the test scenario.","marker":"[40]"},{"why":"Second-order Godunov scheme for relativistic MHD; supplies the numerical setup and initial perturbation style for the KHI tests.","marker":"[41]"},{"why":"High-resolution binary-neutron-star merger simulations showing Kelvin-Helmholtz amplification of magnetic fields; the motivation for needing a sub-grid model.","marker":"[14]"}],"fun_headline_variants":["Gradient SGS model matches MHD turbulence residuals","Relativistic MHD LES gets gradient sub-grid model","First gradient SGS model for relativistic MHD","Gradient SGS model unlocks small-scale MHD dynamics","Relativistic MHD turbulence: gradient model resolves"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The validation assumes that the residuals computed by filtering a high-resolution simulation over scale factors up to 16 are a faithful stand-in for the unresolved terms a coarser LES would actually need, while the true sub-grid scales below the grid spacing and the extra numerical dissipation of the high-resolution run are not part of the comparison; no a-posteriori LES is performed.","fun_headline_variants_meta":{"raw":{"variants":["Gradient SGS model matches MHD turbulence residuals","Relativistic MHD LES gets gradient sub-grid model","First gradient SGS model for relativistic MHD","Gradient SGS model unlocks small-scale MHD dynamics","Relativistic MHD turbulence: gradient model resolves"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00043,"raw_usage":{"total_tokens":2262,"prompt_tokens":1077,"completion_tokens":1185,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":693,"completion_tokens_details":{"reasoning_tokens":1108}},"tokens_in":693,"tokens_out":1185,"duration_ms":9581,"temperature":1.0,"reasoning_tokens":1108,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:13:40.874427+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an actual coarse-grid LES with the SGS tensors (47)--(54) for the same Kelvin-Helmholtz setup and compare its kinetic and magnetic energy spectra with the $1024^3$ DNS; if adding the model fails to reproduce the magnetic-energy growth and spectra of the fine run, or if it destabilizes the scheme, the central claim is refuted. A cheaper check is to compute the Pearson coefficient for a filter factor above 16 or across a strong shock, where the gradient expansion and the shock-capturing dissipation are expected to break down.","supporting_citations":[{"cited_title":"Smagorinsky","cited_arxiv_id":null,"evidence_quote":"Defines the standard purely dissipative closure that the gradient model aims to replace and that is the baseline for comparison."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the Taylor or gradient expansion of sub-grid-scale terms that the paper generalizes."},{"cited_title":"M¨ uller and D","cited_arxiv_id":null,"evidence_quote":"Supplies the gradient-diffusion SGS model for incompressible MHD whose relativistic extension is derived here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Previous explicit LES for GRMHD with a dissipative closure; the relativistic comparison target."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the inverse-filter gradient-expansion series used to derive the first-order closure."},{"cited_title":"Obergaulinger, M","cited_arxiv_id":null,"evidence_quote":"Local simulations of the magnetized Kelvin-Helmholtz instability in neutron-star mergers; defines the test scenario."},{"cited_title":"Beckwith and J","cited_arxiv_id":null,"evidence_quote":"Second-order Godunov scheme for relativistic MHD; supplies the numerical setup and initial perturbation style for the KHI tests."},{"cited_title":"Kiuchi, P","cited_arxiv_id":null,"evidence_quote":"High-resolution binary-neutron-star merger simulations showing Kelvin-Helmholtz amplification of magnetic fields; the motivation for needing a sub-grid model."}],"review_version":1}