{"id":"6cd8e564-72fe-4400-ae6f-4d6ceb1447c1","arxiv_id":"2506.02776","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A chromium-diffusion phase-field model reproduces weight loss and corrosion depth of 9Cr steel in static liquid lithium, showing near-surface grain density governs intergranular corrosion.","lead":"This paper presents a phase-field model that simulates how liquid lithium corrodes ferritic/martensitic steels along grain boundaries, using chromium concentration to track the corrosion front. The model reproduces measured weight loss and corrosion depth for a 9Cr steel in static lithium and identifies near-surface grain density as the main driver of corrosion.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central validation is undermined by circular calibration: the chemical free-energy curvature A (Table 1) is explicitly fitted to the same Xu et al. data used to claim agreement, so Fig. 19 does not independently test the model.","rationale":"I selected the A-calibration issue as the most load-bearing because the central claim is an empirical validation statement. The paper's own limitation passage states that A is selected based on agreement with the same experimental dataset. A model with a parameter tuned to the target cannot, by itself, establish that it captures the physics; it only shows the fitting procedure can reproduce the data. The reader's chosen weakest assumption (lp) is also serious: the paper admits corrosion depth varies with lp in a way that is 'purely artificial', and the constant-product rescaling changes the effective GB diffusivity. However, the constant-product ansatz is a standard way to preserve integrated GB transport, and the paper quantifies its effect. The A calibration is more direct: there is no independent measurement of A, no sensitivity study, and no external prediction. This warrants keeping the verdict CONDITIONAL: the model may still be useful for ranking microstructures (the 2D sensitivity trends and the saturation/dynamic distinction are valuable), and the concern could be retired by the proposed sweep. I do not move to REJECT because the paper is transparent about the limitation and because a single fitted parameter, if shown to be non-influential, would not invalidate the framework. The verdict should remain CONDITIONAL pending the A-sensitivity check.","tokens_in":20557,"tokens_out":8722,"duration_ms":102047,"concrete_test":"Rerun the reference validation case of §4.2 (20 µm average grain size, 6 surface GBs, ten microstructures, concentration-sink boundary) with A = 5×10^8 and 5×10^10 N/m^2 while keeping every other Table 1 entry fixed. Report the mean weight loss at 100, 250 and 500 h and the mean corrosion depth at 250 and 500 h against the experimental data/error bars in Fig. 19. If either A value changes these outputs by more than the experimental relative error, the reported match is attributable to the fitted A and the central validation claim is not independent; if the outputs are essentially unchanged, the circular-calibration objection does not land.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing issue is not the smeared-GB representation itself but the calibration of the chemical free-energy curvature A. In Methods (§4.2), A = 5×10^9 N/m^2 is introduced without an independent source, and the Discussion admits: \"The chemical free energy density curvature parameter A is selected based on the accuracy of the phase-field predictions to the experimental data.\" The abstract's claim that the framework \"reproduces experimental measurements\" is therefore at least partly a fitting outcome: one parameter has been aimed at the same Xu et al. [44] data used for validation. The constant-product lp issue is real and is acknowledged by the authors as producing a purely artificial corrosion-depth dependence (Sec. 2.3), but it is a numerical-representation concern; A is a direct tuning knob on the validation target. If changing A within a plausible range moves the predicted weight loss/corrosion depth outside the experimental uncertainty in Fig. 19, the \"reproduces measurements\" claim is not a genuine confirmation of the physics. If it does not move them, the concern is retired. The paper's sensitivity analyses and saturation/dynamic distinction are useful independent contributions, but they do not supply a parameter-free validation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a phase-field model of intergranular liquid lithium corrosion in ferritic/martensitic steels, using the chromium concentration as the transported species and a stationary grain-boundary field to enhance diffusion along grain boundaries. The model is calibrated and validated against experimental weight loss and corrosion depth data for a 9 wt% Cr steel exposed to static lithium at 600 °C, with additional sensitivity studies varying near-surface grain density, grain size, smeared grain-boundary thickness, and the effect of saturation versus a concentration sink. The central claims are that the framework reproduces the experimental measurements and that near-surface grain density dominates corrosion severity while grain size controls susceptibility to intergranular corrosion.","tokens_in":20800,"tokens_out":2916,"duration_ms":30328,"significance":"If the central validation claim is sound, the model would be a practically useful mesoscale tool for ranking microstructural features for liquid-lithium compatibility, and the saturation-versus-sink comparison is a valuable conceptual contribution. The paper has notable strengths: a thermodynamically consistent derivation from Eq. (2) onward, explicit treatment of grain-boundary diffusion through Eq. (11), a documented implementation in COMSOL with promised code availability, and a systematic set of sensitivity analyses over statistically sampled microstructures. However, the validation is weakened by the calibration of the chemical free energy curvature parameter A against the same experimental data used for Fig. 19, and by the acknowledged artificial dependence of corrosion depth on the numerical grain-boundary thickness l_p. These issues must be addressed before the predictive claims can be taken as established.","major_comments":[{"comment":"The validation in Fig. 19 is partly circular. Section 4.2 says the chemical free energy curvature parameter A is chosen from similar studies, but the Discussion explicitly states that 'The chemical free energy density curvature parameter A is selected based on the accuracy of the phase-field predictions to the experimental data.' Since A appears in Eq. (5) and controls the chemical driving force, the abstract's claim that the framework 'reproduces experimental measurements' is at least in part a fitting outcome rather than an independent test. Please quantify the sensitivity of the predicted weight loss and corrosion depth in Fig. 19 to A (for example ±50% variation) and, if possible, provide an independent determination of A from CALPHAD or first-principles data, or identify an out-of-sample prediction that does not rely on the same Xu et al. [44] dataset.","section":"§4.2, Table 1, and §3"},{"comment":"The corrosion-depth validation is not robust to the numerical grain-boundary thickness l_p. The paper shows in Fig. 11(b) that corrosion depth varies strongly with l_p (50, 100, and 200 nm), and the Discussion states that the observed correlation is 'purely artificial and attributed to the constant product approach.' Because l_p is a computational smearing thickness rather than a physical quantity, the agreement between the predicted and experimental corrosion depth at 250 h in Fig. 19(b) may depend on the particular choice l_p = 100 nm. The authors should either demonstrate that the experimental depth match is insensitive to l_p within a physically reasonable range, or provide an independent justification for the chosen l_p that does not derive from the target experimental data.","section":"§2.3, Eq. (11), and Fig. 11"},{"comment":"The 3D demonstration reports a weight loss of 80.93 g/m² after 500 h, which is two orders of magnitude larger than the 2D result of 1.16 g/m². The paper attributes this to the interface kinetics coefficient L being tailored to 2D microstructures. Since the abstract claims the formulation applies to arbitrary 2D and 3D polycrystalline geometries, the 3D result should be clearly framed as a proof-of-concept only, and the abstract or conclusions should not imply quantitative 3D predictive capability without a 3D-calibrated L.","section":"§3, Fig. 16"}],"minor_comments":[{"comment":"The stationary grain-boundary field η(x) is introduced in Eq. (12) without a physical interpretation beyond interpolating diffusivity. It would improve clarity to state explicitly that η represents a smeared Cr-depletion zone and to discuss how this choice relates to the physical thickness δ_gb.","section":"§4.1, Eq. (12)"},{"comment":"The saturation analysis uses a 1 µm liquid layer chosen specifically to reach saturation within 6000 h, and the paper notes this conflicts with experimental behavior. This is an acknowledged limitation, but the wording in Section 2.4 could more clearly distinguish between a numerical convenience and a physically representative liquid volume.","section":"§2.4 and Fig. 14"},{"comment":"The value of A = 5 × 10^9 N/m² has no source reference in Table 1, unlike the other parameters; the table should indicate whether this value is fitted, taken from literature, or otherwise justified.","section":"Table 1"},{"comment":"The data availability statement says data are available upon reasonable request, while code availability is promised only after article acceptance. For reproducibility, the authors should consider making the code available at the time of submission or at least provide a permanent repository DOI.","section":"§5, Data and Code Availability"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of npj Materials Degradation and has a solid model-derivation core. The main concern is the calibration-vs-validation circularity around parameter A and the artificial l_p dependence of corrosion depth; these need to be resolved before the quantitative claims are accepted. The authors' own Discussion acknowledges both issues, which is encouraging, but the abstract and conclusions currently state the validation more strongly than the evidence supports."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things before reading this paper. First, the model is a sensible, computationally cheap extension of the authors' earlier phase-field corrosion framework to a fusion-relevant problem: liquid lithium intergranular corrosion of F/M steels. The stationary eta field that assigns enhanced GB diffusivity without a full multiphase-field is a neat trick. Second, the abstract's claim that the framework \"reproduces experimental measurements\" is only partly supported. The chemical free energy curvature A is, by the paper's own admission in Section 3, selected to match the Xu et al. data used for validation. So Figure 19 is not an independent test of the model; it shows a fit. That doesn't make the paper worthless, but it should make you cautious about citing it as a validated predictive tool.\n\nWhat is genuinely new and good: the application to liquid Li IGC is new, and the sensitivity analysis separating near-surface grain density from grain size is well done. The R^2 = 0.9973 for inverse projected grain size against weight loss, aligning with Bhave et al., is a strong qualitative finding. The saturation-versus-sink comparison is thoughtful: it clarifies that in static conditions corrosion halts, while in dynamic loops it continues, and the microstructural factors switch roles. That is a real contribution.\n\nThe soft spots are mostly acknowledged in the paper. The l_p dependence of corrosion depth is called \"purely artificial,\" which undercuts the quantitative depth match. The saturation time of 6000 h disagrees with the experimental 250 h equilibration, so that section is more illustrative than predictive. The 3D result is unvalidated and the interface kinetics L is tailored to 2D. None of these are fatal to the qualitative conclusions, but they mean the quantitative validation is thinner than it first appears.\n\nWho is this for? People working on phase-field corrosion modeling, or fusion blanket materials who want a cheap mesoscale tool for ranking microstructures. The derivation is standard and clearly presented, and the code will be available. It deserves a serious referee; I would send it out, but with a clear request to address the circular calibration — for example, a sensitivity study over A or a cross-validation against another experiment would turn the fitting exercise into something closer to a test.\n\nReading group: maybe; it is worth an hour of discussion on calibration honesty in phase-field models.","headline":"A sensible phase-field model for liquid-lithium intergranular corrosion, but the validation claim is softer than the abstract suggests because the free-energy curvature A is fitted to the same data used for comparison.","tokens_in":21359,"tokens_out":2436,"would_cite":false,"duration_ms":24603,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["81.65.Kn","81.05.Bx"],"model":"deepseek-v4-flash","headline":"This paper establishes that a chromium-diffusion phase-field model reproduces measured liquid-lithium corrosion of a 9 wt% Cr steel and separates the roles of surface grain-boundary density and grain size in intergranular attack.","keywords":["liquid lithium corrosion","intergranular corrosion","phase-field model","ferritic/martensitic steel","grain boundary diffusion","chromium depletion","microstructural sensitivity","saturation"],"falsifier":"Expose 9 wt% Cr ferritic/martensitic specimens with controlled near-surface grain densities and grain sizes to static lithium at 600 °C and compare the model's predicted scalings: weight loss rising about 15% per added surface grain, corrosion depth nearly independent of surface grain density, and deeper, more variable penetration for 40 µm grains; any clear violation of these scalings, or a diffusion measurement that contradicts the re-scaled grain-boundary diffusivity, would settle the claim.","tokens_in":20345,"feed_emoji":"🧪","tokens_out":11212,"duration_ms":98463,"temperature":0.7,"pith_summary":"The paper develops a mesoscale phase-field model in which the normalized chromium concentration is the transported field and grain boundaries are marked by an extra stationary field with enhanced diffusivity. Its central claim is that this model, without any front-tracking device, reproduces the experimentally measured weight loss and corrosion depth of a 9 wt% Cr ferritic/martensitic steel in static liquid lithium at 600 °C. Using the validated model, the paper argues that corrosion severity is set by the number of grain-boundary entry points at the exposed surface, while susceptibility to intergranular penetration is set by the bulk grain-boundary density, i.e., by grain size. If correct, the model provides a microstructure-ranking tool for liquid-lithium compatibility and a way to compare static saturation-limited corrosion with the indefinite dissolution expected in dynamic breeder loops.","feed_headline":"Surface grain density sets how fast lithium eats steel","feed_subtitle":"A phase-field model reproduces 600°C lithium weight loss and shows which microstructures resist intergranular attack.","key_machinery":"The central object is a thermodynamically consistent phase-field model of a binary Fe–Cr alloy in contact with liquid lithium, with the phase field $\\phi$ and normalized chromium concentration $c$ as the two variables; the corrosion front is not tracked but emerges from Allen–Cahn relaxation of a free-energy functional whose chemical driving force is the difference between $c$ and the equilibrium solid and liquid concentrations. Grain boundaries are encoded by an independent stationary field $\\eta$, and the load-bearing identity is $D'_{\\mathrm{gb}} = (\\delta_{\\mathrm{gb}}/l_p)D_{\\mathrm{gb}}$, which rescales the physical grain-boundary diffusivity by the ratio of the physical Cr-depletion width to the numerical smearing thickness, so fast intergranular transport is captured without resolving nanometre-sized features. This machinery is what allows intergranular corrosion to appear naturally, with no special treatment of the moving interface.","core_discovery":"The paper's central claim is that intergranular corrosion by liquid lithium can be captured from chromium diffusion alone: with a stationary field $\\eta$ marking grain boundaries and the constant-product identity $D'_{\\mathrm{gb}} = (\\delta_{\\mathrm{gb}}/l_p)D_{\\mathrm{gb}}$ relating the physical Cr-depletion width $\\delta_{\\mathrm{gb}}$ to the numerical smearing width $l_p$, the model reproduces the experimental weight-loss and corrosion-depth curves of a 9 wt% Cr ferritic/martensitic steel at 600 °C. The same framework then shows that weight loss scales with near-surface grain density—about 15% per additional surface grain—whereas corrosion depth responds mainly to grain size through the length and branching of the grain-boundary network. With static saturation, microstructures that corrode fastest also saturate the lithium soonest, so their penetration depth plateaus shallower; with a concentration sink, corrosion continues indefinitely and depth becomes nearly insensitive to microstructure.","pith_inferences":["A testable extension would be to prepare specimens with controlled surface grain-boundary densities but identical bulk grain sizes; the model predicts the 500-hour weight loss should rise about 15% per additional surface grain, which is a sharper fingerprint than a bulk grain-size effect.","Because the paper itself reports that corrosion depth depends on the numerical smearing width $l_p$ in a way it calls purely artificial, the quantitative depth match may be partly tied to a non-physical parameter; measuring effective grain-boundary transport in the same steel would show whether the match reflects physical fidelity.","The saturation calculations imply a scaling rule for static systems: time to saturation should increase with the lithium volume per exposed grain-boundary area, so varying the liquid-to-specimen volume ratio in experiments would test the mechanism directly.","The 2D-to-3D weight-loss gap suggests the interface kinetics coefficient should be tied to material parameters rather than fitted to 2D data before the model is used to rank real 3D components."],"forward_implications":["Holding grain size at 20 µm, going from 5 to 6 to 7 exposed surface grain boundaries raises the 500-hour weight loss by roughly 15% per added grain, while average corrosion depth stays near 12 µm.","Reducing average grain size from 40 µm to 10 µm dramatically increases intergranular attack, so that after 30,000 hours of sink-driven corrosion the fine-grained microstructure is largely engulfed by lithium-filled boundaries.","Under static saturation, the 7-GB and 10 µm microstructures reach saturation first; the 5-GB and 40 µm microstructures corrode longer, with the 40 µm case still unsaturated at 6000 hours.","The concentration-sink model, which mimics dynamic breeder-loop conditions, gives an average corrosion depth around 66 µm after 30,000 hours and implies that structural components would need routine replacement within about three years.","The 3D simulation produces a much larger weight loss (80.93 g/m² vs 1.16 g/m² after 500 hours), indicating that the 2D-calibrated interface kinetics coefficient needs re-derivation before quantitative 3D use."],"supporting_citations":[{"why":"Supplies the experimental weight-loss and corrosion-depth data for a 9 wt% Cr F/M steel in static lithium at 600 °C used to calibrate and validate the model.","marker":"[44]"},{"why":"Provides the saturated chromium concentration after equilibrium and the prior-austenite grain size used to set the reference microstructure.","marker":"[20]"},{"why":"Supplies the chromium diffusion coefficients in the grain bulk and along grain boundaries.","marker":"[76]"},{"why":"Provides the measured 10–15 nm Cr-depleted zone thickness used as the physical width $\\delta_{\\mathrm{gb}}$.","marker":"[79]"},{"why":"Supplies the constant-product approximation that defines the effective grain-boundary diffusivity $D'_{\\mathrm{gb}}$ in terms of $\\delta_{\\mathrm{gb}}$ and $l_p$.","marker":"[70]"},{"why":"Provide grain-boundary interfacial energy values for BCC metals used to set the phase-field parameters.","marker":"[77, 78]"},{"why":"Gives the comparable inverse-projected-grain-size versus weight-loss correlation (R² ≈ 0.9966) that supports the central role of grain-boundary entry points.","marker":"[43]"}],"fun_headline_variants":["Near-surface grain density drives lithium corrosion weight loss","Lithium eats steel faster when surface grain density is high","Surface grain count predicts lithium corrosion rate in steel","Grain size, not density, governs lithium penetration depth in steel"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quantitative validation rests on the constant-product treatment $D'_{\\mathrm{gb}} = (\\delta_{\\mathrm{gb}}/l_p)D_{\\mathrm{gb}}$, which rescales the grain-boundary chromium diffusivity by the ratio of a physical depletion width to a numerical smearing width chosen by the modeler; if that rescaling does not faithfully represent real grain-boundary transport, the claimed reproduction of the measured corrosion depth is not established.","fun_headline_variants_meta":{"raw":{"variants":["Near-surface grain density drives lithium corrosion weight loss","Lithium eats steel faster when surface grain density is high","Surface grain count predicts lithium corrosion rate in steel","Grain size, not density, governs lithium penetration depth in steel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000361,"raw_usage":{"total_tokens":1941,"prompt_tokens":924,"completion_tokens":1017,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":540,"completion_tokens_details":{"reasoning_tokens":952}},"tokens_in":540,"tokens_out":1017,"duration_ms":10810,"temperature":1.0,"reasoning_tokens":952,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:16:44.763636+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Expose 9 wt% Cr ferritic/martensitic specimens with controlled near-surface grain densities and grain sizes to static lithium at 600 °C and compare the model's predicted scalings: weight loss rising about 15% per added surface grain, corrosion depth nearly independent of surface grain density, and deeper, more variable penetration for 40 µm grains; any clear violation of these scalings, or a diffusion measurement that contradicts the re-scaled grain-boundary diffusivity, would settle the claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the experimental weight-loss and corrosion-depth data for a 9 wt% Cr F/M steel in static lithium at 600 °C used to calibrate and validate the model."},{"cited_title":"& Yeliseyeva, O","cited_arxiv_id":null,"evidence_quote":"Provides the saturated chromium concentration after equilibrium and the prior-austenite grain size used to set the reference microstructure."},{"cited_title":"& Pokorn´ a, A","cited_arxiv_id":null,"evidence_quote":"Supplies the chromium diffusion coefficients in the grain bulk and along grain boundaries."},{"cited_title":"& Masamura, K","cited_arxiv_id":null,"evidence_quote":"Provides the measured 10–15 nm Cr-depleted zone thickness used as the physical width $\\delta_{\\mathrm{gb}}$."},{"cited_title":"K., Jiang, C., Jiang, W","cited_arxiv_id":null,"evidence_quote":"Supplies the constant-product approximation that defines the effective grain-boundary diffusivity $D'_{\\mathrm{gb}}$ in terms of $\\delta_{\\mathrm{gb}}$ and $l_p$."},{"cited_title":"& Tonks, M","cited_arxiv_id":null,"evidence_quote":"Gives the comparable inverse-projected-grain-size versus weight-loss correlation (R² ≈ 0.9966) that supports the central role of grain-boundary entry points."}],"review_version":1}