{"id":"be5ae748-af34-43ba-96e8-76a1f4e8c1ac","arxiv_id":"2502.03114","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Using molecular dynamics and Wigner transport theory, the authors show that thermal conductivity of amorphous HfO2 increases continuously from 50K to 2000K, driven by low-frequency vibrations and convection near the glass transition.","lead":"This paper uses machine-learning molecular dynamics and two independent thermal transport theories to compute how well amorphous hafnia conducts heat from 50K to 2000K. It finds that thermal conductivity rises continuously with temperature, contradicting a recent prediction of an unusual drop.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Wigner-based high-T mechanism rests on extrapolating temperature-independent D(ω,T) from 900 K to 2000 K, exactly where quasiparticles break down; if this fails, the dual-GK/WTE claim loses one leg.","rationale":"Read in good faith: this is a competent computational study with useful validations—structure factors against experiment/GAP, rWTE versus q=0 WTE, and a physically motivated decomposition of GK conductivity. The core MD-GK trend (κ increasing to 2000 K) is a direct simulation result and should be taken seriously. The most fragile part is not the low-T rWTE but the extension of WTE into the supercooled liquid. The temperature-independence of D(ω,T) is asserted rather than derived, tested only below 900 K, and used precisely in the range where quasiparticle concepts fail. This is exactly the assumption the reader flagged, and it is load-bearing for the abstract's 'both Green-Kubo and Wigner transport theories' formulation. I therefore keep the CONDITIONAL verdict: the manuscript should either demonstrate D(ω,T) stability up to 2000 K or limit the claim to the MD-GK trend and describe the WTE extension as a hypothesis.","tokens_in":10348,"tokens_out":11090,"duration_ms":116002,"concrete_test":"Compute D(ω,T) at 1100 K and 1400 K from the same NEP-MD trajectories using a frequency-resolved heat-current autocorrelation (or rWTE at 1100 K if modes remain defined), and re-evaluate Eq. 3 using D(ω,T) together with the MD VDOS at each temperature. If the resulting κLD+MD at 1200–2000 K departs from the MD-GK values by more than the GK statistical uncertainty, the temperature-independence assumption is falsified and the Wigner-based convection mechanism loses quantitative support.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim is that κ of a-HfO2 increases continuously to 2000 K based on both Green-Kubo and Wigner transport theories. The high-T Wigner leg depends on Eq. 3 together with the stated observation that D(ω,T) remains nearly temperature-independent in a-HfO2. The support for this is Fig. 3c, which only goes to 900 K, and the extrapolation reaches 2000 K across the regime where the paper itself says quasiparticle frequencies and linewidths become ill-defined (power-spectrum disruption at 1200 K). Thus D(ω,900 K) is combined with MD VDOS at 1500–2000 K (Fig. 2c) to produce κLD+MD. If D(ω,T) changes above 900 K, the agreement with MD-GK in Fig. 3b is coincidental rather than evidence for the low-frequency-VDOS convection mechanism. The direct MD-GK trend would still stand, but the claim that Wigner transport theory also yields the increase would not be established. The manuscript also provides no error bars or validation of the high-T heat-current convention, but the D extrapolation is the sharpest internal-consistency risk.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports molecular dynamics simulations with a machine-learned neuroevolution potential to study thermal transport of amorphous HfO2 from 50 K to 2000 K. At low temperatures the authors use the regularized Wigner transport equation and at high temperatures the Green-Kubo method, and they claim that the thermal conductivity increases continuously with temperature up to 2000 K, in contrast to a previous QHGK prediction of decreasing conductivity above 900 K. To connect the two regimes, they introduce a phenomenological extension of the Wigner framework in which the frequency-dependent diffusivity D(ω,T) is assumed nearly temperature-independent above 600 K, so that D(ω,900K) can be combined with MD-computed VDOS at higher temperatures through Eq. (3) to reproduce the MD-GK conductivity. The paper attributes the increase to enhanced low-frequency vibrational density of states and associated heat convection in the supercooled liquid.","tokens_in":10604,"tokens_out":6029,"duration_ms":58353,"significance":"If the central claim holds, the paper resolves a notable discrepancy in a-HfO2 and extends the study of glass thermal transport across the glass transition, with practical implications for gate dielectrics and thermal barrier coatings. The study has clear strengths: the NEP potential is validated against experimental structure factors and earlier GAP simulations, the low-temperature rWTE results are benchmarked internally against bare WTE and QHGK, and the MD-GK calculations are a standard approach with a useful decomposition into potential, kinetic, and cross terms. However, the high-temperature Wigner-based leg of the argument rests on an unvalidated assumption about the temperature independence of D(ω,T), and the mechanistic conclusion about low-frequency modes is partly built into that assumption. The direct MD-GK trend is likely robust, but the claim that Wigner transport theory independently supports the increase is not fully established.","major_comments":[{"comment":"The reconstruction κLD+MD above 900 K relies on the statement that D(ω,T) remains nearly temperature-independent in a-HfO2, but the supporting calculation in Fig. 3c is shown only for 100-900 K. The reconstruction uses D(ω,900K) together with MD VDOS at 1500-2000 K, i.e., across the temperature range where the authors themselves find quasiparticle frequencies and linewidths ill-defined at 1200 K and where atomic diffusion is significant. If D(ω,T) changes above 900 K, the agreement between κLD+MD and κMD-GK in Fig. 3b is coincidental rather than evidence for the low-frequency-VDOS mechanism. Please provide direct evidence that D(ω,T) is temperature-independent in the supercooled liquid, for example by extracting frequency-resolved diffusivities from MD at 1200-2000 K or by a decomposition of the MD heat flux that isolates the low-frequency contribution, or restrict the Wigner-based claim to temperatures where the assumption has actually been checked.","section":"Thermal transport crossover; Eq. (3), Fig. 3b-c"},{"comment":"Eq. (3) is introduced as a mathematical rewriting of the rWTE, but for the supercooled liquid it is applied under an explicitly phenomenological hypothesis. The mechanistic conclusion that the enhanced low-frequency VDOS plays a key role in increasing κ is largely a consequence of the model: with C(ω,T)=kB and D(ω,T) fixed, Eq. (3) forces κ to follow the temperature dependence of the VDOS-weighted diffusivity. Reproducing MD-GK above 900 K therefore validates the assumed form but does not independently establish the causal role of low-frequency modes. The authors should state this limitation explicitly and, ideally, test the hypothesis against a direct MD-based modal decomposition of the heat current.","section":"Thermal transport crossover; Eq. (3)"},{"comment":"The central high-temperature trend rests on the κMD-GK values, but Fig. 3a-b report no statistical uncertainties, convergence checks with respect to simulation time, system size, or heat-current convention. Given that the increase above 900 K is the main claim and that the paper proposes to overturn a previous QHGK result, the authors should provide error bars or at least report the standard deviation across independent runs and the characteristic correlation time of the heat-current autocorrelation function.","section":"Thermal transport crossover; Fig. 3a-b"}],"minor_comments":[{"comment":"In the introduction, 'an decrease' should be 'a decrease'.","section":"Introduction"},{"comment":"The label 'corrected κ~T trend' is unclear; please specify how the low-temperature rWTE and high-temperature MD-GK values are combined into a single trend.","section":"Fig. 3b"},{"comment":"The symbols V, Nc, and η are used before being defined; please define them in the main text or point to the SI at first use.","section":"Eq. (1)"},{"comment":"The sentence 'This may be attributed to the interplay between the low-frequency VDOS and the velocity operator' is qualitative; please quantify the compensation between VDOS growth and velocity-operator decrease, or remove the speculative explanation.","section":"Thermal transport crossover"},{"comment":"Fig. 3c would benefit from a legend or explicit labels for each temperature, since the reported 'minimal differences' are difficult to assess from the figure as printed.","section":"Fig. 3c"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a serious computational study with a clear central claim. The stress-test concern about the temperature independence of D(ω,T) is legitimate and is the main reason I cannot recommend acceptance in present form. The MD-GK result is standard and appears internally consistent, and the low-temperature rWTE part is well benchmarked, so I do not see grounds for rejection. With direct evidence for the D(ω,T) assumption or a suitably reduced Wigner-based claim, this paper could become publishable in a strong journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the headline: this is a credible computational challenge to a recent QHGK prediction of decreasing κ in a-HfO2. The authors combine NEP MD with Green-Kubo and rWTE, and they report a continuous increase in κ up to 2000 K, with a mechanistic explanation involving low-frequency modes and heat convection. The paper does a lot right: the glass transition is characterized carefully, the NEP potential is validated against experiment and GAP, and the low-temperature rWTE is benchmarked against bare WTE and QHGK. The observation that α_l/α_g approaches the universal ratio at slower cooling rates is a nice bonus.\n\nThe soft spots are real but not fatal. The high-temperature Wigner leg depends on the claim that D(ω,T) is nearly temperature-independent above 600 K, as computed up to 900 K, and then combined with MD VDOS at 1500–2000 K to reconstruct κ. That is an extrapolation across exactly the regime where quasiparticles break down, so the agreement with MD-GK above 900 K is suggestive rather than proof of the mechanism. The paper is honest about the phenomenological nature of this extension, but it still underpins the abstract's claim that 'both Green-Kubo and Wigner transport theories' show the increase. If D changes above 900 K, the Wigner leg weakens. The direct MD-GK trend would still stand, but the dual-theory claim would be overstated.\n\nOther concerns are minor: no error bars are given in the main text, no experimental κ comparison is attempted, and no code or potential is shipped. Given the controversial prediction, those would be nice to have, but they are addressable.\n\nWho this is for: anyone working on thermal transport in glasses or disordered oxides. It deserves a serious referee; the question of whether κ rises across the glass transition is important, and the paper gives a testable, well-defined computational scenario. My recommendation: send it to review, but flag the D(ω,T) extrapolation as the key issue, and ask for error bars and a stronger discussion of the classical-heat-capacity limitation at high T.","headline":"This paper credibly challenges a recent QHGK prediction by finding increasing κ up to 2000 K in a-HfO2, but its high-temperature Wigner extension rests on an extrapolated temperature-independent diffusivity that deserves scrutiny.","tokens_in":11131,"tokens_out":1855,"would_cite":true,"duration_ms":17166,"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":"Amorphous hafnia's thermal conductivity rises continuously from 50 K to 2000 K, overturning a predicted high-temperature drop and tracing the increase to low-frequency vibrations that carry heat by convection across the glass transition.","keywords":["amorphous hafnia","thermal conductivity","glass transition","Wigner transport equation","Green-Kubo method","machine-learned potential","low-frequency vibrations","heat convection"],"falsifier":"Compute $D(\\omega,T)$ directly from molecular dynamics at 1200 K, 1500 K, and 1800 K by decomposing the heat flux into mode contributions; if the extracted low-frequency diffusivity deviates from the lattice-dynamics $D(\\omega,900\\,\\mathrm{K})$ by more than the simulation uncertainty, the paper's reconstruction of $\\kappa$ and its conclusion that enhanced VDOS drives convection would fail.","tokens_in":10141,"feed_emoji":"🔥","tokens_out":12173,"duration_ms":102061,"temperature":0.7,"pith_summary":"Amorphous hafnia (a-HfO2), a material used in memory devices and thermal barrier coatings, has been predicted to conduct heat worse as it gets hotter, the opposite of most glasses. This paper argues that prediction is wrong. Using molecular dynamics driven by a machine-learned interatomic potential, together with two transport theories, the authors find that the thermal conductivity of a-HfO2 rises continuously from 50 K to 2000 K, even across the glass transition near 1500 K. The rise comes from heat convection carried by diffusing atoms in the supercooled liquid, and the authors trace it to an enhanced density of low-frequency vibrational modes dominated by hafnium atoms. If correct, the work settles a disputed trend and offers a way to compute heat transport in glasses up to and beyond the glass transition.","feed_headline":"Amorphous hafnia's heat conduction rises continuously to 2000 K","feed_subtitle":"Machine-learned simulations overturn the predicted drop and trace the rise to atomic vibrations.","key_machinery":"The central object is the frequency-dependent thermal diffusivity $D(\\omega,T)$, defined from the regularized Wigner transport equation (rWTE), a lattice-dynamics theory for glasses in which vibrational modes are broadened by disorder and anharmonicity and quantum statistics are retained. The load-bearing identity is $\\kappa(T) = \\int_0^\\infty g(\\omega)\\, C(\\omega,T)\\, D(\\omega,T)\\, d\\omega$, which splits the conductivity into the vibrational density of states $g(\\omega)$, the specific heat $C(\\omega,T)$, and the diffusivity. The paper's mechanistic claim is that $D(\\omega,T)$ is nearly constant above 600 K, so the temperature dependence of $\\kappa$ across the glass transition is set almost entirely by $g(\\omega)$, whose low-frequency part grows and crosses over from $\\omega^2$ (Debye) to $\\omega$ (liquid-like) scaling. This lets the authors reconstruct $\\kappa$ in the supercooled liquid from a 900 K lattice-dynamics diffusivity combined with molecular-dynamics VDOS at higher temperature, and it connects heat convection to low-frequency propagating modes dominated by heavy Hf atoms.","core_discovery":"The authors establish, through molecular dynamics simulations with a machine-learned interatomic potential, that the thermal conductivity $\\kappa$ of amorphous HfO2 increases continuously with temperature up to 2000 K. At low temperatures (50–900 K), $\\kappa$ is computed with the regularized Wigner transport equation (rWTE), which accounts for anharmonicity and Bose–Einstein statistics, and it rises with temperature. Above about 1200 K, atomic diffusion broadens vibrational modes beyond the Lorentzian quasiparticle picture, so lattice-dynamics methods no longer apply; here Green–Kubo molecular dynamics captures an additional convective (kinetic) contribution that grows sharply near the glass transition at about 1500 K and keeps $\\kappa$ increasing. By rewriting the rWTE in frequency space, the authors define a frequency-dependent thermal diffusivity $D(\\omega,T)$ and show it is nearly temperature-independent above 600 K. Combining this $D(\\omega,900\\,\\mathrm{K})$ with the molecular-dynamics vibrational density of states at higher temperatures reproduces the Green–Kubo conductivity, demonstrating that the supercooled-liquid rise is driven by the growing low-frequency VDOS, which switches from Debye $\\omega^2$ scaling to liquid-like linear scaling, rather than by changes in how individual modes diffuse heat.","pith_inferences":["If this mechanism generalizes, other amorphous oxides whose heavy cations diffuse before crystallization should also show a convective upturn in $\\kappa$ near $T_g$, whereas rigid network glasses such as a-SiO2 should not; this is a testable distinguishing prediction not made in the paper.","The near-constancy of $D(\\omega,T)$ suggests a practical shortcut: predict high-temperature glass conductivity by reweighting an equilibrium VDOS with one a priori diffusivity, cutting the cost of simulations in supercooled regimes.","The result implies that the common notion of a 'minimum thermal conductivity' bounding glass transport from above does not apply once diffusion contributes; near the glass transition the convective channel can push $\\kappa$ above the plateau."],"forward_implications":["Amorphous HfO2 does not show the previously predicted drop in $\\kappa$ at high temperature; its conductivity rises steadily up to 2000 K.","Lattice-dynamics-only methods (rWTE and QHGK) underestimate $\\kappa$ once atomic diffusion sets in near 1200 K, so the paper's two-method protocol (rWTE below, Green–Kubo above) is needed to cover the full range up to and beyond the glass transition.","Because $D(\\omega,T)$ is nearly temperature-independent above 600 K, high-temperature $\\kappa$ can be estimated by combining a single low-temperature lattice-dynamics diffusivity with the molecular-dynamics VDOS, avoiding expensive direct simulation at every temperature.","The evolution of the low-frequency VDOS from $\\omega^2$ to $\\omega$ scaling across $T_g$ gives a microscopic signature that the phonon quasiparticle picture is breaking down and convective transport is taking over."],"supporting_citations":[{"why":"Supplies the regularized Wigner transport equation used to compute the low-temperature (50–900 K) conductivity.","marker":"[24]"},{"why":"The prior quasi-harmonic Green-Kubo calculation that predicted a decrease in kappa above 900 K; this paper's central claim directly contradicts and explains that result.","marker":"[30]"},{"why":"Describes the neuroevolution machine-learning potential used for all molecular dynamics simulations, including heat-flux and VDOS calculations.","marker":"[31]"},{"why":"Provides the machine-learning training set (Gaussian approximation potential data) on which the neuroevolution potential is trained, plus structural references for the amorphous phase.","marker":"[15]"},{"why":"Establishes the linear-in-frequency vibrational density of states in liquids, the signature used to interpret the low-frequency VDOS enhancement across the glass transition.","marker":"[41]"}],"fun_headline_variants":["Amorphous hafnia heat conduction rises up to 2000 K","Machine-learning simulations reverse hafnia heat-drop prediction","Hafnia glass thermal conductivity climbs with temperature","Overturning predictions: hafnia glass heat conduction rises","Simulations reveal hafnia heat flow rises with temperature"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the frequency-dependent thermal diffusivity $D(\\omega,T)$ remains nearly unchanged above about 600 K, so that the 900 K diffusivity can stand in for higher temperatures when reconstructing $\\kappa$; if $D$ varies substantially between 900 K and 2000 K, the mechanistic explanation loses quantitative support even though the directly simulated Green–Kubo trend would still stand.","fun_headline_variants_meta":{"raw":{"variants":["Amorphous hafnia heat conduction rises up to 2000 K","Machine-learning simulations reverse hafnia heat-drop prediction","Hafnia glass thermal conductivity climbs with temperature","Overturning predictions: hafnia glass heat conduction rises","Simulations reveal hafnia heat flow rises with temperature"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000686,"raw_usage":{"total_tokens":3181,"prompt_tokens":1085,"completion_tokens":2096,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":701,"completion_tokens_details":{"reasoning_tokens":2014}},"tokens_in":701,"tokens_out":2096,"duration_ms":15852,"temperature":1.0,"reasoning_tokens":2014,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T05:51:19.366297+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $D(\\omega,T)$ directly from molecular dynamics at 1200 K, 1500 K, and 1800 K by decomposing the heat flux into mode contributions; if the extracted low-frequency diffusivity deviates from the lattice-dynamics $D(\\omega,900\\,\\mathrm{K})$ by more than the simulation uncertainty, the paper's reconstruction of $\\kappa$ and its conclusion that enhanced VDOS drives convection would fail.","supporting_citations":[{"cited_title":"Thermal conduc- tivity of glasses: First-principles theory and applications,","cited_arxiv_id":null,"evidence_quote":"Supplies the regularized Wigner transport equation used to compute the low-temperature (50–900 K) conductivity."},{"cited_title":"Vibrational and thermal properties of amor- phousaluminafromfirstprinciples,","cited_arxiv_id":null,"evidence_quote":"The prior quasi-harmonic Green-Kubo calculation that predicted a decrease in kappa above 900 K; this paper's central claim directly contradicts and explains that result."},{"cited_title":"Neuroevolution machine learning poten- tials: Combining high accuracy and low cost in atomistic sim- ulations and application to heat transport,","cited_arxiv_id":null,"evidence_quote":"Describes the neuroevolution machine-learning potential used for all molecular dynamics simulations, including heat-flux and VDOS calculations."},{"cited_title":"Machine-learned interatomic potentials by active learning: amorphousandliquidhafniumdioxide,","cited_arxiv_id":null,"evidence_quote":"Provides the machine-learning training set (Gaussian approximation potential data) on which the neuroevolution potential is trained, plus structural references for the amorphous phase."},{"cited_title":"Universal law for the vibra- tional density of states of liquids,","cited_arxiv_id":null,"evidence_quote":"Establishes the linear-in-frequency vibrational density of states in liquids, the signature used to interpret the low-frequency VDOS enhancement across the glass transition."}],"review_version":1}