{"id":"0531c317-896a-47d6-95d5-445a8848518d","arxiv_id":"2504.14028","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":11,"one_line_summary":"A Bayesian-calibrated nonlinear Burgers model with evolving microstructure fits creep and forced-oscillation data for ice and olivine, and extrapolates to attenuation spectra at elevated stresses.","lead":"This paper fits a nonlinear viscoelastic model, complete with evolving grain size and crystal texture, to laboratory experiments on deformed ice and olivine using Bayesian statistics. It then uses the calibrated model to predict how ice dissipates energy under cyclic stress, a key input for tidal and glacier models.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Single-Kelvin-element transient model may mask underdetermination; out-of-sample attenuation prediction is not tested against a simpler baseline.","rationale":"The reader's weakest assumption is precisely the single Kelvin-Voigt element. The paper's own Discussion concedes that Andrade-type models with a spectrum of relaxation times fit lab data better, and that the single Kelvin element produces a Debye peak in attenuation. The nonlinearity and microstructure evolution are claimed to broaden the spectrum, but no demonstration is given. This is load-bearing because the out-of-sample attenuation prediction is the paper's headline extrapolative result, and the offset between the predicted and measured attenuation values (Fig. 6) is acknowledged by the authors. The concern is not that the model is wrong internally but that the comparison set is too narrow: with 11 parameters and no baseline comparison, the fits and predictions do not uniquely support the mechanistic interpretation. The concrete test (refit with linear or two-element Kelvin-Voigt and compare predictive scores) would settle whether the nonlinear microstructural machinery is needed. I agree with the reader's conditional verdict; if the baseline comparison confirms the more complex model's predictive advantage and code/data are released, ACCEPT would be warranted, but as written the evidence is insufficient for the strength of the claim.","tokens_in":13572,"tokens_out":1769,"duration_ms":14385,"concrete_test":"Refit the same ice (Qi-Goldsby) and olivine (forced oscillation) data with a linear Burgers model (same Maxwell spring and flow law, but a linear Maxwell dashpot) and with a two-element Kelvin-Voigt (or Andrade-type) anelastic spectrum, using the same NUTS sampler, priors, and observation error model. Compare (a) leave-one-out or cross-validation scores on the stress-strain and amplitude/phase data, and (b) the predicted attenuation spectrum at sigma_m = 1 MPa, sigma_0 = 0.17 MPa against McCarthy et al. (2016). If the linear or two-element model achieves comparable fits and predicts the attenuation spectrum at least as well, the central claim that the specific nonlinear microstructural formulation is needed for out-of-sample prediction is not supported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that a single nonlinear Burgers-type model, calibrated by MCMC, captures both steady and transient dynamics and predicts dynamics for data it was not trained on. The load-bearing assumption is in 'Transient Component' and 'Discussion & Conclusion': one linear Kelvin-Voigt element with a single relaxation time tau_K = eta_K/E_K is sufficient to represent the transient/anelastic response of both ice and olivine, with the nonlinear dashpot plus microstructure evolution invoked to broaden the relaxation spectrum without demonstration. The paper itself acknowledges (Discussion) that Andrade-type models with a spectrum of relaxation times fit laboratory data better, and that the single element yields a Debye peak in the attenuation spectrum (Fig. 6). The predicted attenuation spectrum is the Fourier image of this single relaxation time; the model is not benchmarked against a simpler linear Burgers model or an Andrade-type model with the same MCMC machinery. If a one-relaxation-time linear model (or a two-relaxation-time model) fits the same ice and olivine data equally well, then the specific mechanistic interpretation—that the nonlinear dashpot and microstructure evolution cause the broadening—is unsupported, and the out-of-sample prediction in Fig. 6 is not strong evidence for the model. This is a correctness risk rather than an internal inconsistency: the parameter count is 11, and the paper itself notes that a good fit has high probability if priors are too wide, but no posterior predictive checks against simpler alternatives are provided.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The authors combine a nonlinear Burgers-type viscoelastic model with Hamiltonian Monte Carlo (No-U-Turn Sampler) inference, incorporating evolving grain size and crystallographic preferred orientation (CPO) in a Maxwell dashpot with a power-law flow law and a single linear Kelvin-Voigt element for transient/anelastic response. They fit three constant-strain-rate stress-strain experiments on polycrystalline ice and a forced-oscillation experiment on olivine, reporting posterior distributions and retrodiction bands. They then use the ice-calibrated model to compute an attenuation spectrum for comparison to McCarthy et al. (2016), finding a qualitative Debye-peak shape and nonlinear amplitude dependence but quantitative deviations from the measured attenuation. The paper claims that the method constrains a nonlinear viscoelastic model that captures both steady and transient dynamics and can predict dynamics for untrained data.","tokens_in":13905,"tokens_out":6596,"duration_ms":62699,"significance":"The paper is valuable in demonstrating the feasibility of Bayesian calibration of a microstructurally informed Burgers model against two distinct laboratory datasets, and the ice retrodiction bands and olivine amplitude/phase fits are encouraging. The model also makes falsifiable predictions for attenuation at elevated stress levels. If the predictive claim were quantitatively established, the approach could improve geodynamic modeling by providing parameter uncertainties and a systematic route from laboratory creep and oscillation data to constitutive laws. However, the central predictive claim is not yet established: the out-of-sample attenuation comparison is qualitative, and the use of a single Kelvin-Voigt relaxation time means that the predicted spectral shape is largely fixed by fitted parameters rather than independently tested. The paper's own caveats and the absence of comparison to simpler baselines leave the mechanistic interpretation underdetermined. With additional model-comparison and posterior-predictive diagnostics, the framework would be a solid contribution.","major_comments":[{"comment":"The central claim that the model \"can also predict dynamics for data it was not trained on\" is not quantitatively supported by the out-of-sample attenuation test. The model curves in Fig. 6 visibly deviate from the McCarthy et al. (2016) data, and the manuscript's explanations (GBS regime, ambient pressure, cracking, low-frequency microstructure evolution) are plausible but unverified. With only a handful of data points and no uncertainty bands on the predicted attenuation, the comparison establishes at most a qualitative resemblance in the form of a Debye peak and high-temperature background. The authors should either quantify the prediction error and propagate posterior uncertainty into the attenuation spectra, or soften the abstract's predictive claim to reflect that the extrapolation is qualitative.","section":"Predictions, Fig. 6"},{"comment":"The transient/anelastic element is a single linear Kelvin-Voigt circuit, so the model has exactly one relaxation time tau_K = eta_K/E_K. The position and shape of the Debye peak in Fig. 6 are therefore largely determined by the fitted values of eta_K and E_K; the assertion that the nonlinear dashpot and microstructure evolution broaden the relaxation spectrum is not demonstrated. A simpler linear Burgers model (or a two-Kelvin-element version) fitted with the same MCMC approach might reproduce the ice and olivine data equally well while yielding a different attenuation spectrum. Reporting posterior-predictive checks and comparing information criteria against such baselines is necessary before the mechanism can be attributed to the nonlinear/microstructural ingredients.","section":"Transient Component and Discussion"},{"comment":"The model has 11 free parameters, but the ice calibration uses only three stress-strain curves, and the paper itself notes (Discussion) that a good fit has high probability if the prior space is too large. The posterior convergence of E_K and eta_K is reported, but no posterior correlations or identifiability analysis is shown. The statement that the model captures the experiment \"uniquely with this mechanistic formulation\" (Results, ice) is stronger than the evidence. The authors should report posterior correlation matrices and, at minimum, a complexity measure such as the effective number of parameters to show that the data can constrain the 11-dimensional parameter vector.","section":"MCMC with Ice Data and Materials and Methods"}],"minor_comments":[{"comment":"The text says the attenuation data are \"sometimes reffers to as high-temperature background\"; \"reffers\" should be \"referred\" and the phrasing should be corrected.","section":"Predictions"},{"comment":"\"Despite these incogruencies\" should be \"incongruencies\" or \"inconsistencies\".","section":"Discussion & Conclusion"},{"comment":"The caption states that the temperature in the numerical solutions was 260 K, while the main text reports experiments at 263 K; please clarify whether this difference is intentional.","section":"Figure 3 caption"},{"comment":"The olivine strain data were detrended by fitting and subtracting a polynomial before computing the oscillatory strain rate; the uncertainty of this correction is not propagated into the MCMC, and its influence on the inferred Kelvin-Voigt parameters should be discussed.","section":"MCMC with Olivine Data"},{"comment":"The text says grain-size evolution was incorporated and later states that microstructural analysis showed negligible grain-size evolution; the non-convergence of the grain-size parameter in Fig. S5B is consistent with that, but the main text should state this more directly to avoid apparent contradiction.","section":"Olivine Data and Priors"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is likely to be of interest to the journal's readership, and the ice and olivine fits are encouraging. My main concern is that the paper's headline predictive claim is not yet a clean test: the attenuation comparison is qualitative, and no simpler baseline is considered. I recommend major revision rather than rejection, with emphasis on adding posterior-predictive checks and a linear-Burgers or multi-relaxation-time comparison."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this is a real attempt at unifying transient and steady-state viscoelasticity in ice and olivine with a microstructure-aware Burgers model, and the fits look good. But the paper oversells the out-of-sample attenuation prediction: because the attenuation spectrum is essentially the Fourier transform of a single Kelvin-Voigt relaxation time, the Debye peak is built into the model rather than learned. A simpler linear Burgers fit might do as well. That doesn't sink the paper, but it means the mechanistic story about nonlinear dashpot and microstructure broadening the spectrum is not yet supported.\n\nWhat's new: they combine nonlinear flow law, grain-size and CPO evolution, and a Kelvin-Voigt transient element in one Bayesian framework, and apply the same machinery to constant-strain-rate ice creep and forced olivine oscillation. As far as I can tell, nobody has done that combination. The ice stress-strain fits, including the early peak and subsequent softening, are genuinely good; the olivine amplitude-phase match is also convincing. The posterior inferences on n and epsilon_c1 converging to sensible values is a nice sanity check. Credit also for transparency: they explicitly list the discrepancies with McCarthy et al. and possible causes (GBS regime, ambient pressure, cracking).\n\nSoft spots: the single KV element is the load-bearing assumption. The paper itself admits Andrade-type models fit lab data better, and the claim that the nonlinear dashpot plus microstructure evolution broadens the spectrum is asserted, not demonstrated. There is no benchmark against a linear Burgers model, a two-element KV, or an Andrade model with the same MCMC machinery. If a simpler model fits the same data equally well, the \"unified\" interpretation weakens. Also, no code or data are shipped, so the MCMC results aren't independently checkable. Minor: Figure 3 caption says 260 K while the text says 263 K. That's a flag for sloppiness, not a fatal issue. The priors on QM and beta are narrow, which is justified but should be discussed more.\n\nWho this is for: geodynamicists who use Burgers-type rheology and experimentalists designing attenuation experiments. The paper deserves a serious referee, but the referee should push for baseline comparisons and, ideally, a release of code and data. If the authors can show that their nonlinear model outperforms a simpler linear one on the same data, the mechanistic claim would stand much stronger.\n\nRecommendation: send to peer review, with requests for a baseline comparison and code/data release.","headline":"Good idea, plausible fits, but the attenuation prediction rests on a single fitted relaxation time and needs a baseline comparison before the mechanistic story holds.","tokens_in":14456,"tokens_out":2083,"would_cite":true,"duration_ms":19196,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single nonlinear Burgers-type viscoelastic model, calibrated by Bayesian inference from laboratory experiments, reproduces both steady and transient deformation of ice and olivine and predicts attenuation data it was not trained on.","keywords":["viscoelasticity","Burgers model","Markov chain Monte Carlo","microstructure evolution","ice rheology","olivine rheology","seismic attenuation","Bayesian inference"],"falsifier":"Run forced-oscillation experiments on ice at elevated median stresses around 1.5, 2, and 4 MPa, where the paper predicts increasing attenuation and a flattening of the attenuation peak; if the peak persists instead of being suppressed at high stress, or if energy loss does not depend on oscillation amplitude, the single-element transient branch is wrong. A sharper test is the width of the attenuation peak at one stress: a single relaxation time gives a peak of one fixed width, while a distribution of relaxation times gives a broader peak.","tokens_in":13359,"feed_emoji":"🧊","tokens_out":12507,"duration_ms":103212,"temperature":0.7,"pith_summary":"The paper attempts to show that one mechanistic model can span the full viscoelastic behavior of geologic materials: a nonlinear Burgers-type circuit whose parameters are learned from laboratory data by Markov chain Monte Carlo. The authors fit this model to constant-strain-rate compression tests on polycrystalline ice and to a forced-oscillation experiment on olivine, then use the ice calibration to predict the attenuation spectrum of a different ice experiment that was not part of the fitting data. If the claim holds, geodynamic and glaciological modelers would gain a single constitutive description that ties evolving grain size and crystal orientation to both steady flow and transient creep, and they could use lab-tuned models to forecast behavior in conditions experiments have not yet reached.","feed_headline":"One nonlinear Burgers model fits ice and olivine experiments","feed_subtitle":"Bayesian inference ties grain size and orientation to steady and transient creep, then predicts unseen attenuation data.","key_machinery":"The load-bearing object is the nonlinear Burgers model, a mechanical circuit in which a Maxwell element (an elastic spring in series with a nonlinear viscous dashpot) is connected in series with a Kelvin-Voigt element (a spring parallel to a dashpot). The Maxwell dashpot follows a dislocation-creep flow law with a stress exponent, an activation enthalpy, and explicit evolution of average grain size $d$ and crystal-orientation factor $F$; the Kelvin-Voigt element supplies the transient and anelastic response, linearized from a hyperbolic-sine backstress law on the assumption that the effective stress on the transient dashpot is small. This one circuit carries the entire argument because every deformation mode — elastic, transient, and steady-state flow — is represented in the same equations, so a single posterior over the parameters must simultaneously satisfy loading, oscillation, and attenuation observations.","core_discovery":"The central claim is that a nonlinear Burgers model with a power-law Maxwell dashpot and a single linear Kelvin-Voigt element, augmented with phenomenological relaxation equations for grain size and crystallographic preferred orientation, is sufficiently identifiable from laboratory data to represent ice and olivine in the dislocation-creep regime. After Bayesian inference, the posterior distributions for the stress exponent and the critical strain for grain-size evolution converge to physically plausible values ($n \\approx 4$ and $\\varepsilon_{c1} \\approx 0.03$ for ice; $n \\approx 3$ for olivine), the model captures the peak stress and steady-state flow of ice stress-strain curves and the amplitude and phase of olivine forced oscillations, and the calibrated ice model reproduces the shape of an attenuation spectrum measured in a separate experiment, including a Debye peak and a power-law high-temperature background. The paper further claims predictive power: extrapolating the ice model to higher median stresses yields amplitude-dependent attenuation with an increasing power-law background and a suppressed Debye peak, which future experiments could verify.","pith_inferences":["The claim that nonlinearity and microstructure evolution broaden the relaxation spectrum beyond the single spring-and-dashpot element is asserted rather than demonstrated; computing the effective relaxation-time distribution from the model's small-oscillation response at different mean stresses would test it directly.","Because the predicted attenuation peak is the frequency-domain image of a single relaxation time, a high-resolution attenuation measurement that resolves the peak width would discriminate this model from models with a continuous relaxation spectrum.","The same inference pipeline could be pointed at other strain-rate histories, such as stress-relaxation or stress-drop experiments, to check whether the posterior parameters learned from constant strain rate and forced oscillation agree out of sample.","The same inference approach could be used to test uniqueness: fitting experiments with different transient loading histories and checking whether the inferred spring-and-dashpot values coincide would show how much of the model's structure is actually demanded by the data."],"forward_implications":["Geodynamic and ice-sheet models could replace separate steady-state and transient parametrizations with a single nonlinear Burgers law whose transient branch is tied to measurable microstructure.","The calibrated ice model can generate attenuation spectra for stress amplitudes and frequencies not yet explored in the laboratory, directly informing tidal-dissipation models for icy moons and ice-shelf flexure.","Posterior convergence for individual parameters, such as the Kelvin-Voigt spring and dashpot, reveals which deformation mechanisms a given experiment actually constrains, guiding the design of future transient-creep and attenuation experiments.","The same framework can be applied to other deformation mechanisms and materials by changing the stress and grain-size exponents, allowing composite flow laws to be tested against several experimental geometries at once."],"supporting_citations":[{"why":"Supplies the constant-strain-rate ice compression data and observed dynamic recrystallization used for the ice MCMC fits.","marker":"(58)"},{"why":"Provides the stress-reduction backstress experiments that motivate the effective-stress formulation of transient creep.","marker":"(41)"},{"why":"Supplies the dislocation-theory flow law with hyperbolic-sine stress dependence that is linearized for the Kelvin-Voigt transient branch.","marker":"(56)"},{"why":"Provides the phenomenological equations used to evolve grain size and crystallographic preferred orientation with strain.","marker":"(19)"},{"why":"Establishes the previous nonlinear Burgers-type MCMC analysis of dunite that this study extends with explicit microstructure evolution and forced-oscillation data.","marker":"(36)"},{"why":"Provides the ice attenuation spectrum used as the out-of-sample prediction target for the calibrated model.","marker":"(61)"},{"why":"Introduces the No-U-Turn Sampler, the adaptive Hamiltonian Monte Carlo variant used for posterior inference.","marker":"(68)"},{"why":"Supplies olivine flow-law parameters and grain-boundary-sliding constraints used in the olivine priors.","marker":"(54)"},{"why":"Supplies ice flow-law stress-exponent constraints used in the ice priors.","marker":"(33)"}],"fun_headline_variants":["Bayesian model ties ice and olivine creep to microstructure","Unified viscoelastic model learned from ice and olivine data","Bayesian inference predicts unseen attenuation in ice","Microstructure-flow link revealed by MCMC for ice and olivine","A single Burgers model unifies ice and olivine rheology"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that one spring-and-dashpot element, with a single stiffness and a single viscosity, is enough to represent the short-term and recoverable deformation of both ice and olivine in the dislocation-creep regime, so that any spread of relaxation times comes from the nonlinear flow law and evolving grain structure rather than from a distribution built into the model.","fun_headline_variants_meta":{"raw":{"variants":["Bayesian model ties ice and olivine creep to microstructure","Unified viscoelastic model learned from ice and olivine data","Bayesian inference predicts unseen attenuation in ice","Microstructure-flow link revealed by MCMC for ice and olivine","A single Burgers model unifies ice and olivine rheology"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000365,"raw_usage":{"total_tokens":1971,"prompt_tokens":960,"completion_tokens":1011,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":576,"completion_tokens_details":{"reasoning_tokens":923}},"tokens_in":576,"tokens_out":1011,"duration_ms":7522,"temperature":1.0,"reasoning_tokens":923,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:58:45.421464+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run forced-oscillation experiments on ice at elevated median stresses around 1.5, 2, and 4 MPa, where the paper predicts increasing attenuation and a flattening of the attenuation peak; if the peak persists instead of being suppressed at high stress, or if energy loss does not depend on oscillation amplitude, the single-element transient branch is wrong. A sharper test is the width of the attenuation peak at one stress: a single relaxation time gives a peak of one fixed width, while a distribution of relaxation times gives a broader peak.","supporting_citations":[],"review_version":1}