{"id":"7505972c-536e-46ad-bf54-9628bbd662ff","arxiv_id":"2607.26833","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Latent entropy of thermally fluctuating |mz| dynamics, inverted through fitted forward models, recovers Aex and Ku and detects grain boundaries in micromagnetic simulations more accurately than the temporal mean.","lead":"Simulations show that latent entropy from noisy magnetization time series can recover exchange stiffness and anisotropy, and flag grain boundaries, better than the plain time average. The method targets high-temperature magnets where domain textures wash out and usual imaging tricks fail.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"The one-known-parameter inversion is the load-bearing hinge; joint (Aex,Ku) is not identified from a single scalar descriptor.","rationale":"The reader already isolated the correct weakest assumption (one-of-{Aex,Ku} known; joint uniqueness not shown; ill-conditioning where sensitivity collapses). My stress test does not replace that diagnosis; it sharpens it as the single load-bearing hinge of the strongest claim. The simulation pipeline, latent-entropy superiority over μ, grain-boundary masking, and error maps are internally consistent under the stated protocol, so there is no grounds to move from CONDITIONAL to REJECT. Equally, the claim cannot be read as unconditional material-parameter recovery from magnetization-only data until level-set diameters (or an explicit two-descriptor/joint inversion) are shown. Keeping CONDITIONAL with the same scope the reader gave—accept-shaped as a computational methods result once joint non-identifiability and transfer are stress-tested and code released—is the honest adjustment. No stronger objection (e.g., internal inconsistency of the fits or failure of the heterogeneous demo under its own rules) lands on a careful read.","tokens_in":14679,"tokens_out":701,"duration_ms":13211,"concrete_test":"On the same uniform grid used for Fig. 3, treat both parameters as unknown: for each simulated Sobs (and separately μobs), compute the full level set {(A,K): Sfit(A,K)=Sobs} (or the μ analogue) and report its diameter in relative Aex and Ku. If typical level-set diameters exceed the claimed reconstruction errors (≳ few % in Aex or ≳10 % in Ku), joint non-identifiability is confirmed and the unconditional claim must be narrowed.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim that the pipeline “infers material parameters” rests on Eqs. (5a–5b) and §III: one of {Aex,Ku} is treated as known and the fitted scalar surface S(Ku,Aex) or μ(Ku,Aex) is inverted for the other. The forward maps (Eqs. 3–4, Fig. 2) are smooth and largely monotonic in each coordinate separately, but the paper never shows that a single observed Sobs (or μobs) selects a unique pair (Aex,Ku). Level sets of S are nontrivial curves in the (Aex,Ku) plane (visible in the surface tilt of Fig. 2 left and in the weaker ∂S/∂Ku* sensitivity of Fig. B2), so many distinct pairs produce the same descriptor. The reported sub-percent / few-percent errors (Fig. 3, App. C) are therefore conditional accuracies under an oracle that already supplies the second parameter; they do not establish unconditional recovery. The same conditioning appears in the Voronoi test (Fig. 4), where each grain is inverted with the true companion parameter known. §V acknowledges the one-parameter-known setup but still frames the result as parameter inference and experimental readiness. If neither parameter is independently known—or if unmodeled terms (dipolar, DMI, Ms(T), α) shift the descriptors—the inverted values are not identified. That is the single point on which the strongest claim stands or falls.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript proposes a magnetization-only pipeline for recovering local magnetic material parameters from thermally driven dynamics in regimes where texture-based methods fail. Using MuMax3 LLG simulations with a fluctuation–dissipation thermal field at T = 700 K, the authors extract pixel-wise latent entropy S and temporal mean μ of |mz(t)|, fit empirical forward surfaces S(Ku, Aex) and μ(Ku, Aex) (Eqs. 3–4), and invert them under the assumption that one of {Aex, Ku} is known (Eqs. 5a–5b). They report that S yields substantially lower relative errors than μ on a uniform parameter grid (Fig. 3; App. C) and on a Voronoi-heterogeneous film, where spatial gradients of the descriptors are also used to mask grain boundaries before interior-averaged inversion (Fig. 4; App. D). The work is carefully documented (fit metrics, sensitivity maps, error distributions, masking pipeline) and framed toward high-temperature experimental parameter mapping.","tokens_in":15045,"tokens_out":1687,"duration_ms":49588,"significance":"If the results hold under the stated one-parameter-known protocol, the paper offers a practical route to local Aex or Ku inference and grain-boundary localization precisely where conventional domain-wall-width methods break down—strongly fluctuating, polycrystalline, or multiphase magnets at elevated temperature. The systematic comparison of latent entropy against the temporal mean, with transparent sensitivity and residual analyses (App. B–C), is a genuine methodological contribution: S remains informative where μ saturates at high Aex. The Voronoi demonstration that descriptor gradients can both segment grains and support interior parameter recovery is relevant to real microstructures. Strengths include reproducible simulation choices, closed-form/numerical inverses, and full error bookkeeping. The main scientific value is conditional on clearer validation hygiene and on not overstating unconditional joint recovery of (Aex, Ku).","major_comments":[{"comment":"§III, Eqs. (5a–5b) and the abstract/§V framing: all reported reconstructions treat one of {Aex, Ku} as known and invert a single scalar descriptor for the other. Level sets of S(Ku, Aex) (Fig. 2 left; weaker ∂S/∂K*u in Fig. B2) are nontrivial curves, so a lone Sobs (or μobs) does not identify a unique pair. The sub-percent/few-percent errors in Fig. 3 and the Voronoi results in Fig. 4 are therefore conditional accuracies under an oracle companion parameter. This scope is stated in §III but is easy to miss in the abstract and outlook, which speak of “material-parameter inference” and experimental readiness more broadly. Please (i) state the one-known-parameter protocol prominently in the abstract and conclusions, and (ii) either demonstrate a joint (Aex, Ku) inversion using both S and μ together (as §V itself suggests) or quantify non-uniqueness (e.g., level-set widths / condition numbers","section":"§III, Eqs. (5a–5b); abstract; §V"},{"comment":"§IV.A and Fig. 3: the uniform-sample relative-error maps appear to invert the fitted surfaces on the same (Ku, Aex) campaign used to build Eqs. (3)–(4). In that case the quoted average |rel. err.| values (0.66% vs 4.79% for Aex; 4.32% vs 20.48% for Ku) largely restate forward-fit fidelity and local conditioning rather than out-of-sample predictive accuracy. Please reserve a held-out grid (or an independent dense sampling) never used in the fit, report reconstruction errors on that set, and keep training-set residuals in App. B. The Voronoi test is closer to a transfer check but still uses the same forward models and known companion parameters; it does not replace a held-out uniform benchmark.","section":"§IV.A, Fig. 3; App. B–C"},{"comment":"§II.A / Hamiltonian (1) and §V: the forward models are trained in a micromagnetic world containing only exchange, uniaxial anisotropy, and the thermal field—no dipolar fields, DMI, Ms(T), or α variation. Descriptor shifts from any of these will bias the inverted Aex or Ku under the current maps. For the central claim of experimental applicability at high T, at least a limited robustness check is needed (e.g., freeze the fitted S/μ models and resimulate a few grid points with demagnetization on, or with modest Ms or α offsets) and the failure modes should be stated quantitatively next to the error maps. Without that, the path from simulation inversion to “experimental parameter extraction” remains an untested extrapolation.","section":"§II.A, Eq. (1); §V"}],"minor_comments":[{"comment":"Fig. 1 workflow caption and main text: specify explicitly that the observable is |mz| (magnitude), and why the sign is discarded—readers may wonder about up/down domain information.","section":"Fig. 1; §II.A"},{"comment":"Eqs. (3)–(4): the empirical forms are flexible but unmotivated. A short remark on why this exponential/saturation structure was chosen (vs. a simpler interpolant or a scaling-inspired ansatz) would help, even if full physics-informed models are left to future work.","section":"§III, Eqs. (3)–(4)"},{"comment":"App. D masking: the hysteresis thresholds (0.28, 0.85) for |∇S| and (0.04, 0.18) for |∇μ| and the 19 nm exclusion width are central free parameters. Please note how they were chosen and whether results are stable under modest threshold changes.","section":"Appendix D"},{"comment":"Table D1 and Fig. 4: merged regions (footnotes *, †) make some “region-averaged” errors hard to interpret. Mark merged regions graphically in Fig. 4 or in the pixel-wise panels of Fig. D2.","section":"Fig. 4; Table D1; Fig. D2"},{"comment":"Nbins = 9 and fixed global bin edges are stated but not motivated. A one-sentence sensitivity note (or pointer that empty-bin avoidance drove the choice at 700 K) would suffice.","section":"§II.B"},{"comment":"Minor typography: “INFERENCE STRA TEGY” and “INFERENCE OF MA TERIAL P ARAMETERS” in the section headings appear to contain stray spaces; “T emporal” in Table B1 likewise.","section":"§III–IV headings; Table B1"}],"recommendation":"major_revision","confidential_remarks":"The latent-entropy idea and the high-T motivation are a good fit for a condensed-matter / magnetism journal. The main risk is over-selling: the strongest numerical claims are conditional oracle inversions on in-sample surfaces. If the authors add held-out errors, tighten the one-known-parameter framing, and run a small robustness check (or a joint S+μ inversion), this should become a solid methods paper. I would not reject on the one-parameter protocol alone—it is a legitimate materials-science use case when bulk Ku or Aex is known—but the abstract currently reads stronger than the theorems support."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The useful takeaway is simple. In MuMax3 at 700 K they show that pixel-wise latent entropy of |mz(t)|, inverted through an empirical surface, recovers Aex (and less well Ku) more accurately than the temporal mean, and that spatial gradients of either descriptor can mask Voronoi grain boundaries before region-wise inversion. That end-to-end pipeline—descriptor maps, fitted S(Ku,Aex)/μ(Ku,Aex), error grids, heterogeneous test—is the new piece; the latent-entropy definition itself is their prior work.\n\nWhat they do carefully is worth credit. Fit quality is high (R² 0.9997 / 0.9962), residuals and sensitivity maps are shown, and they document the failure modes instead of hiding them: low-Aex/low-Ku loss of contrast for S, high-Aex saturation for μ, systematically worse Ku recovery. The 19 nm exclusion tied to exchange length and the hysteresis-threshold masking are practical and reproducible on paper. Citations to the materials range and to imaging modalities look appropriate.\n\nThe soft spot that actually matters is the one the stress-test flags. Every inversion (Eqs. 5a–b, Fig. 3, Fig. 4) treats one of {Aex, Ku} as known. Level sets of S are curves, not points; a single Sobs does not pick a unique pair. The sub-percent / few-percent errors are therefore conditional accuracies under an oracle companion parameter, not unconditional recovery. Uniform-grid numbers also partly restate fit fidelity on the same campaign. Unmodeled dipolar/DMI, Ms(T), or α shifts are left for later. They acknowledge the one-known setup in the discussion, but the abstract and framing still say “infer material parameters.” That is a scoping issue, not a hidden contradiction—the math is honest once you read §III.\n\nWho it is for: people doing high-T micromagnetic imaging, magnetocalorics, or permanent-magnet microstructure who already have one parameter from bulk data and want local maps where domain walls are gone. Not a general inverse-magnetism result yet.\n\nI would send it to referees as a computational methods paper. Ask for code/data, an explicit joint-identifiability / level-set figure, and clearer abstract language. No desk reject.","headline":"Solid simulation methods paper: latent entropy beats the mean for recovering Aex/Ku from hot |mz| dynamics and for grain-edge contrast, but every reported accuracy assumes the other parameter is already known.","tokens_in":15746,"tokens_out":585,"would_cite":true,"duration_ms":11410,"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":"Latent entropy from hot, fluctuating magnetization dynamics recovers exchange stiffness and anisotropy better than the temporal mean, and can map grain boundaries.","keywords":["latent entropy","micromagnetics","exchange stiffness","magnetic anisotropy","thermal fluctuations","parameter inference","grain boundaries","magnetization dynamics"],"falsifier":"On a uniform film with independently known exchange and anisotropy, measure time-resolved |mz| at high temperature, compute latent entropy with the paper’s fixed binning, invert the published S-model, and check whether the recovered parameter falls outside the reported few-percent error bands; systematic failure would refute the claim.","tokens_in":15480,"feed_emoji":"🧲","tokens_out":926,"duration_ms":19936,"temperature":0.7,"pith_summary":"Magnetic materials are controlled by local parameters such as exchange stiffness and anisotropy, but those parameters are hard to read out when thermal noise washes out domain walls and other textures. This paper shows that you can still recover them from magnetization time series alone by computing simple statistical descriptors—especially latent entropy, which measures how unpredictably the magnetization jumps between coarse-grained states—and inverting fitted models that link those descriptors to the material constants. In micromagnetic simulations at 700 K, latent entropy yields substantially smaller reconstruction errors than the ordinary time average, both on uniform films and on a polycrystalline Voronoi sample where the same maps also locate grain boundaries. A sympathetic reader cares because the method needs only dynamical magnetization data, works where texture-based recipes fail, and points toward local parameter mapping in real high-temperature experiments.","feed_headline":"Hot magnetization noise reveals exchange and anisotropy","feed_subtitle":"Latent entropy beats the time average and maps grain boundaries when textures vanish","key_machinery":"Latent entropy: a scalar that quantifies the stochasticity of transitions among a fixed set of discretized |mz| bins along each pixel’s time series. It carries dynamical transition structure that remains informative even when the mean magnetization saturates, and is the descriptor whose forward model is inverted for material parameters.","core_discovery":"From thermally driven out-of-plane magnetization trajectories the authors extract pixel-wise latent entropy and temporal mean, fit empirical forward models of each descriptor versus exchange stiffness and uniaxial anisotropy, and invert those models to recover one parameter when the other is known. Latent entropy systematically outperforms the temporal mean (average absolute relative errors of roughly 0.66% versus 4.79% for exchange and 4.32% versus 20.48% for anisotropy on the uniform grid) and, on a heterogeneous Voronoi film, also supplies spatial gradients that detect grain boundaries so that interior parameters can be predicted grain by grain.","pith_inferences":["If experimental noise floors still preserve transition statistics among a few |mz| bins, laboratory TR-MOKE or STXM movies could feed the same inversion without new theory.","Treating temperature or damping as additional unknown parameters would require multi-descriptor or multi-temperature measurements to restore uniqueness.","The 19 nm exclusion zone around boundaries sets a practical lower grain-size limit; smaller grains would need thinner masks or joint multi-pixel models.","Dipolar or Dzyaloshinskii–Moriya terms, once included in the training simulations, could turn the same pipeline into a local probe of interfacial chirality."],"forward_implications":["Local exchange and anisotropy can be mapped from time-resolved magnetization imaging without relying on domain-wall width or other textures.","Grain boundaries in polycrystalline magnets become visible as spatial gradients of latent entropy or temporal mean.","The same descriptors remain usable at elevated temperature where conventional texture methods break down.","Joint use of latent entropy and temporal mean, or physics-informed forward models, is expected to tighten predictions further.","Extension to three-dimensional and multiphase microstructures is natural because the method never needs a projected wall profile."],"fun_headline_variants":["Latent entropy maps exchange and anisotropy from hot mag noise","Magnetization fluctuations alone recover stiffness and anisotropy","Latent entropy beats time mean for parameter inference","Entropy from mag trajectories detects grain boundaries","Thermal mag dynamics yield local exchange and anisotropy"],"cache_read_input_tokens":128,"weakest_assumption_plain":"One of the two material parameters must already be known so the other can be read off by inverting a single scalar descriptor; if both are unknown, or if unmodeled interactions shift the descriptors, the inversion is not uniquely identified.","fun_headline_variants_meta":{"raw":{"variants":["Latent entropy maps exchange and anisotropy from hot mag noise","Magnetization fluctuations alone recover stiffness and anisotropy","Latent entropy beats time mean for parameter inference","Entropy from mag trajectories detects grain boundaries","Thermal mag dynamics yield local exchange and anisotropy"]},"model":"grok-4.5","effort":"low","cost_usd":0.001742,"raw_usage":{"total_tokens":828,"prompt_tokens":752,"num_sources_used":0,"completion_tokens":56,"cost_in_usd_ticks":17424000,"prompt_tokens_details":{"text_tokens":752,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":20,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":752,"tokens_out":56,"duration_ms":2240,"temperature":1.0,"reasoning_tokens":20,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-30T19:43:33.154520+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On a uniform film with independently known exchange and anisotropy, measure time-resolved |mz| at high temperature, compute latent entropy with the paper’s fixed binning, invert the published S-model, and check whether the recovered parameter falls outside the reported few-percent error bands; systematic failure would refute the claim.","supporting_citations":[],"review_version":1}