{"id":"a6348742-457c-4ea9-8916-b3c05b033b03","arxiv_id":"2505.07752","paper_version":4,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Phase-field simulations with Bayesian active learning map the dendritic-to-planar transition in rapidly solidifying Fe-Cr and identify unstable intermediate microstructures in place of banding.","lead":"Simulations of fast-solidifying iron-chromium alloy map when the freezing front changes from dendrites to a smooth planar interface. The study combines phase-field simulations with machine learning to make the expensive mapping practical for additive manufacturing process design.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The precise PF/KGT match of G_max rests on an S=5 interface-width bracket (2e7–4e7 K/m) that is not shown to be converged: the paper's own S=8 test flips the G=1e7, V=0.3 m/s classification from planar to cellular, so the S-shift of the G_max bracket is unquantified.","rationale":"The reader's weakest assumption—quantitative fidelity of the S=5 upscaled PF model near the absolute stability threshold—is also the most load-bearing condition for the central claim. The KGT agreement is only as strong as the PF bracket, and the single interface-width sensitivity test (S=8, §4.1.5) shows the planar/non-planar classification is S-dependent: at G=10^7 K/m, V=0.3 m/s, S=5 labels planar and S=8 labels cellular. The S=8 series covers only G≤10^7 K/m, so the bracket-defining values (2×10^7, 4×10^7 K/m) were never tested at another S; the direction of the observed shift (wider interface → cells persist to higher V) means the converged G_max could move above the S=5 bracket, which would place KGT's 3.16×10^7 K/m outside the PF brackets. The paper deserves credit for reporting the S=8 comparison, the single S=3 check (which sided with S=5 for the wavy-channel artifact), and the explicit frozen-temperature caveat (§4.1.4, Conclusions); those honest disclosures are what make the concern precisely testable rather than speculative. Note also that the PF/KGT match is on an emergent boundary, not on the calibrated k_V/m_V inputs, so it is not trivially circular. Two secondary reservations: 'precisely matched' exceeds what a factor-of-two bracket can establish, and the frozen-temperature approximation (Eq. 14) is acknowledged to need quantitative study at high V; neither changes the verdict alone. The concern is addressable by targeted S=3 runs at the bracket points, so CONDITIONAL (not REJECT) is the right verdict, and my read does not alter it.","tokens_in":39299,"tokens_out":18885,"duration_ms":155898,"concrete_test":"Run c∞=17 wt% PF simulations at S=3 (closer to the sharp-interface S=1 limit) at the bracket-defining points: (G,V)=(10^7, 0.3) where S=5 and S=8 disagree; G=2×10^7 at the velocities that are non-planar at S=5; and G=4×10^7 across all six velocities (planar at S=5). Compare planar/non-planar labels with S=5 and S=8. If S=3 reproduces the S=5 labels at every point, the bracket is converged and the KGT agreement stands; if S=3 matches S=8 at any point, or shifts the 2×10^7/4×10^7 classifications, the PF boundary is S-dependent and the claimed 'precise match' requires revision. A reduced discriminating subset is (1e7, 0.3), (2e7, 0.3), (2e7, 0.6), (4e7, 0.3), (4e7, 0.6). Complement: recompute the KGT G_max with the PF-consistent k_V/m_V curves (already computed for the pink dotted line in Fig. 7) at S=5 and S=8 to bound calibration sensitivity.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—KGT accurately predicts the maximum gradient G_max above which the interface is planar at any velocity (Abstract; §4.1.1; Conclusions)—is supported only by a factor-of-two bracket: cells/dendrites at G=2×10^7 K/m, all-planar at G=4×10^7 K/m, enclosing the KGT values 3.16×10^7 (2D) and 3.00×10^7 K/m (3D). Every simulation defining this bracket used a single upscaled interface width S=5, A=11 (Table 1). The paper's only interface-width test, the S=8, A=18 series (§4.1.5, Fig. 6 vs Fig. 4), shows the classification at G=10^7 K/m, V=0.3 m/s changes from planar (S=5, Fig. 4c5) to cellular (S=8, Fig. 6c5): a wider interface 'promotes cellular growth at the expense of planar growth in the vicinity of the transition.' Because the S=8 series covers only G≤10^7 K/m, neither bracket point (2×10^7, 4×10^7) nor any intermediate G has been tested at S=8, or at S<5, except one S=3 run for a different artifact (wavy channels at G=10^6, V=0.06, which sided with S=5). The direction of the observed shift—expanding the non-planar region as S grows—implies the converged G_max could lie above the S=5 bracket; if it exceeded 4×10^7 K/m, KGT's 3.16×10^7 would fall outside the PF brackets and the claimed 'precise match' (Conclusions) would fail. The frozen-temperature approximation (Eq. 14) is a second, explicitly acknowledged risk (§4.1.4), but it is independent of the interface-width question and secondary to it for the G_max claim. Separately, 'precisely matched' overstates what a factor-of-two bracket can establish.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper combines high-fidelity phase-field (PF) simulations with a Gaussian-process Bayesian active-learning framework to map solidification microstructure selection in the (c∞, G, V) space for a binary Fe-Cr alloy as a surrogate for 316L stainless steel, focusing on the dendritic/cellular-to-planar transition near absolute stability. The central physical claim is that the classical KGT model accurately predicts the maximum temperature gradient Gmax above which planar growth is stable across the entire velocity range, with KGT giving Gmax ≈ 3.16×10^7 K/m in 2D and the PF simulations bracketing it between 2×10^7 and 4×10^7 K/m. The paper also reports an unstable 'intermediate' microstructure regime at low G and proposes a simple velocity-shifted KGT limit as a computationally cheap approximation of the PF stability boundary. The Bayesian active-learning methodology is demonstrated to refine the decision boundary efficiently, and the paper includes a retrospective comparison of acquisition strategies.","tokens_in":39782,"tokens_out":3104,"duration_ms":33501,"significance":"If the central claim is correct, the paper offers a practical and computationally cheap route to identify safe process windows for rapid solidification in additive manufacturing, and it advances the use of Bayesian-guided phase-field exploration in higher-dimensional parameter spaces. The paper provides a large, carefully classified dataset of PF simulations (54 full-factorial cases plus 50 Bayesian-selected cases, with an additional S=8 series), and the KGT comparison is an independent analytical benchmark rather than a fit to the PF output. The identification of an unstable intermediate regime in a non-banding alloy is a novel and interesting physical observation. However, the strength of the central quantitative claim (exact KGT match for Gmax) is currently not commensurate with the supporting evidence, which is only a factor-of-two bracket obtained with a single upscaled interface width.","major_comments":[{"comment":"The conclusion that 'the calculated value of G above which planar growth is stable across the entire V range is precisely matched between PF and KGT models' is an overstatement of the evidence. The PF determination of Gmax rests entirely on the S=5, A=11 series: cells/dendrites persist at G=2×10^7 K/m, while all interfaces are planar at G=4×10^7 K/m (Fig. 4; Section 4.1.1). This is a factor-of-two bracket enclosing the KGT value of 3.16×10^7 K/m, not a 'precise match.' More importantly, the only interface-width test, the S=8 series in §4.1.5 and Fig. 6, shows that increasing S from 5 to 8 at G=10^7 K/m and V=0.3 m/s changes the classification from planar to cellular, shifting the planar boundary in the direction of suppressing planar growth. Because the S=8 series only covers G≤10^7 K/m, neither bracket point (2×10^7 or 4×10^7) nor any intermediate G has been tested at S=8 or at S<5 (except one S=3 run for a different artifact). The observed S-dependence implies that the converged Gmax could lie above the S=5 bracket; if it exceeded 4×10^7 K/m, the KGT value would fall outside the PF bounds and the central claim would fail. I request either an explicit S-convergence study at the bracket points (and ideally at intermediate G values), or a substantial softening of the claim to state that the PF results bracket the KGT prediction within a factor of two for a single, unverified interface width.","section":"§4.1.1, Conclusions, and Figs. 4, 6, 8"},{"comment":"The frozen temperature approximation, Eq. (14), is a second load-bearing assumption for the Gmax claim. The authors acknowledge in §4.1.4 that latent heat release at high V can strongly affect banding instability and microstructural length scales, and they argue that the Fe-Cr alloy's absence of banding mitigates the effect, with G interpretable as an effective gradient. However, the quantitative comparison with KGT is a sharp numerical statement: a shift in the effective gradient due to latent heat could change Gmax enough to alter the agreement. The manuscript itself notes that 'a deeper quantitative study of the effect of latent heat release on the present results remain needed in the future,' which is appropriate for a limitation, but the Conclusions then state the match is 'precise' without flagging this condition. I request that the conclusions explicitly condition the KGT agreement on the frozen-temperature approximation, or provide an order-of-magnitude estimate of the latent-heat correction to Gmax.","section":"§4.1.4 and Eq. (14)"}],"minor_comments":[{"comment":"The notation 'V→V*[max{TS(V)}/V a]' in Fig. 8 is difficult to parse; please define the shifted velocity variable explicitly in the caption or text.","section":"Fig. 8 caption and Section 4.1.1"},{"comment":"The manual fixing of kernel length scales in the final iteration is mentioned in the caption of Fig. 12 and in §2.3.2, but the consequence—a slight increase in entropy between iterations 4 and 5—is only discussed in the figure caption. Please move this into the main text for clarity.","section":"§2.3.2 and Fig. 12"},{"comment":"Equations (A.1) and (A.2) have slightly inconsistent notation (e.g., 'c' and 'ϕ' are used without explicit definitions in this appendix); please make the notation self-consistent with the main text.","section":"Appendix A"},{"comment":"The paper repeats the phrase 'precisely matched' in the Abstract, §4.1.1, and Conclusions; given Major Comment 1, this wording should be revised to reflect the actual bracket-based evidence.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for Acta Materialia and the Bayesian-guided PF integration is a valuable contribution. The main issue is that the headline quantitative claim (exact KGT match for Gmax) is not supported by the convergence evidence the paper itself provides; the S=8 test points toward a systematic S-dependence that could invalidate the bracket. I believe this is fixable either by a targeted convergence study at the two bracket gradients or by explicitly reframing the claim as a factor-of-two bracket under the chosen numerical parameters. The frozen-temperature caveat should also be connected to the conclusions. I would not reject the manuscript, as the physical findings about the intermediate regime and the method validation are solid."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper is worth reading if you care about microstructure selection in rapid solidification or about using Bayesian active learning to make expensive phase-field campaigns tractable. The genuinely new piece is the mapping itself: for a binary Fe-Cr surrogate of 316L, the authors combine a quantitative phase-field model with a GP classifier to trace the dendritic-to-planar boundary across composition, velocity, and gradient, and they identify an unstable intermediate 'wavy' microstructure at low G and high V, rather than a stable cellular regime or banding. That result is plausible and physically interesting, and the Bayesian workflow is presented with a retrospective comparison against uninformed and grid-based sampling, which is more than most such papers do.\n\nThe KGT comparison is also a real benchmark: the analytical model is not fitted to the phase-field output, and the phase-field model itself has prior validation against Al and Mg banding experiments. That said, the central quantitative claim is oversold. The PF evidence for the maximum gradient above which growth is planar everywhere is a bracket: cells at 2e7 K/m, planar at 4e7 K/m, with KGT giving about 3.16e7 K/m. Calling that 'precisely matched' in the conclusions is not supported. The bracket is consistent with the KGT value, but it does not pin it down, and the paper's own S=8 test shows that at one condition the planar classification flips to cellular, which signals sensitivity to the upscaled interface width. That S=8 test does not cover the bracket points, so the magnitude of the shift in G_max is unknown. This is the softest spot in the paper, and it is addressable: run the 2e7 and 4e7 conditions at S=8, or at least soften the language to 'consistent within a factor of two.'\n\nThe frozen temperature approximation is a second acknowledged limitation, but it is secondary for the G_max claim and the authors flag it honestly. No code or data are released, which limits reproducibility, but the supplementary material does include full parameter lists and classifications.\n\nWho is this for? Researchers in solidification and additive manufacturing who want an efficient way to explore process-parameter space with a quantitative PF model. It is a solid engineering-scale contribution, not a breakthrough, and it deserves a serious referee. I would send it to review with a request to address the convergence question and to calibrate the strength of the claims to the evidence.","headline":"A useful demonstration of Bayesian-guided phase-field mapping for rapid solidification in Fe-Cr, but the paper's claim of a precise PF/KGT match on G_max overstates what a factor-of-two bracket can establish.","tokens_in":40336,"tokens_out":1254,"would_cite":true,"duration_ms":15480,"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":"The classical KGT analytical model accurately predicts the temperature gradient above which a planar solid-liquid interface is stable at any growth velocity during rapid solidification, matching expensive phase-field simulations in a…","keywords":["rapid solidification","microstructure selection","phase-field modeling","KGT model","absolute stability","Bayesian active learning","Gaussian process classification","Fe-Cr alloy"],"falsifier":"Rerun the same Fe-Cr simulations at $S=3$ or $S=1$ for the points near $G=2\\times 10^7$ K/m and $V=0.3$ m/s; if cellular patterns persist where $S=5$ produced planar growth, then the KGT $G_{\\max}$ agreement is a numerical artifact rather than a physical prediction.","tokens_in":39112,"feed_emoji":"🔬","tokens_out":10092,"duration_ms":88971,"temperature":0.7,"pith_summary":"The paper sets out to establish that microstructure selection in rapid solidification can be predicted without running expensive simulations: the classical KGT analytical model gives the value of the thermal gradient above which planar growth is stable for every growth velocity. Comparing with Bayesian-guided phase-field simulations in a Fe-Cr surrogate of 316L stainless steel, the paper finds KGT's $G_{\\max}\\approx 3.16\\times 10^7$ K/m lies between the simulation bracket of $2\\times 10^7$ and $4\\times 10^7$ K/m. It also claims that the absolute stability velocity depends on the gradient, contrary to the usual $G$-independent formula, and that a shifted KGT limit using the maximum of the solidus temperature $T_S(V)$ is a simple and accurate estimate. If true, process windows for fully planar solidification in additive manufacturing could be set by a cheap analytical calculation, with the phase-field and Bayesian machinery used to refine details such as cells versus dendrites and unstable intermediate patterns.","feed_headline":"KGT formula nails the planar-solidification limit in rapid growth","feed_subtitle":"Phase-field simulations confirm the analytical prediction, so process windows can be set without costly mapping.","key_machinery":"The argument is carried by four objects: the KGT dendrite-tip model, an analytical marginal-stability calculation of tip radius and temperature as functions of $V$ and $G$; an upscaled quantitative phase-field model of rapid solidification with an enhanced interface diffusivity, run in 2D with interface-width scaling $S=5$ and diffusion coefficient $A=11$; the frozen-temperature approximation, which imposes a moving linear thermal profile and neglects latent-heat release; and a Gaussian-process classifier with an informative prior based on the maximum of the solidus temperature $T_S(V)$, which actively selects new simulation points. The factor $V_{\\max\\{T_S\\}}/V_a$ is the device that converts the $G$-independent classical absolute-stability velocity into a $G$-dependent boundary that matches the simulations.","core_discovery":"The paper's central claim is that the classical KGT analytical model, used with a 2D tip solution, predicts the value of the thermal gradient $G$ above which a planar interface is stable across the entire growth-velocity range, and that an upscaled phase-field model confirms it: KGT gives $G_{\\max}\\approx 3.16\\times 10^7$ K/m while the simulations put cells at $2\\times 10^7$ K/m and full planarity at $4\\times 10^7$ K/m. A second claim is that the absolute-stability velocity is not independent of $G$: the simulations and KGT both show the boundary moving to lower $V$ as $G$ rises, and a simple shift of the KGT limit by the ratio $V_{\\max\\{T_S\\}}/V_a$ captures the boundary better than the classical $V_a$ formula. Third, the transition from dendritic to cellular morphology with increasing $G$ near this limit is a genuine feature, and in the low-$G$ dendritic regime there is no steady cellular-dendritic intermediate: the interface goes through an unstable 'wavy' structure that never stabilizes, which the authors interpret as the Fe-Cr counterpart of the banding instability seen in other alloys.","pith_inferences":["This is an inference, not a paper claim: if the $S$-dependence reported in Section 4.1.5 is generic, convergence studies at $S=3$ or $S=1$ should become standard before any upscaled phase-field map near absolute stability is used quantitatively.","Inference: the wavy 'intermediate' structures may be the Fe-based analogue of banding, the same instability expressed without periodic bands, so time-resolved experiments on rapidly solidified steels could look for lateral interface motion instead of banded solute layers.","Inference: the same Gaussian-process classifier with a $T_S(V)$-based prior could be run in reverse to locate banding onset in alloys that do band, searching the region where $dT_S/dV>0$ as an inexpensive pre-screening step before phase-field runs."],"forward_implications":["Process engineers can use KGT alone to set a safe planar-growth window by choosing $G$ above $G_{\\max}$, without running phase-field simulations for every velocity.","The transition from dendrites to cells with increasing $G$ near the planar limit is a real feature, so rapid-solidification maps should show a cellular regime only there, not as a stable band between dendritic and planar at low $G$.","The absolute stability velocity depends on $G$, so the common practice of treating $V_a$ as a constant will misplace the boundary at high thermal gradients.","A Bayesian-guided phase-field campaign of roughly one hundred simulations can map a three-dimensional selection diagram in composition, velocity, and gradient, a budget too small for a comparable full factorial study."],"supporting_citations":[{"why":"Supplies the linear-stability definition of the absolute stability threshold that the paper compares against and reinterprets.","marker":"[8]"},{"why":"Provides the experimental and theoretical basis for using the maximum of the solidus temperature as the better planar-limit velocity.","marker":"[17]"},{"why":"Supplies the upscaled quantitative phase-field model of rapid solidification used for all microstructure simulations.","marker":"[38, 39]"},{"why":"Supplies the KGT dendrite-tip model whose planar stability limit is compared with the phase-field results.","marker":"[59, 60]"},{"why":"Supplies the Fe-Cr pseudo-binary alloy parameters and the 316L surrogate baseline on which the simulations are built.","marker":"[61]"},{"why":"Supplies the Gaussian-process classification with informative priors that the Bayesian active-learning loop uses.","marker":"[64]"}],"fun_headline_variants":["KGT nails planar solidification limit in rapid growth","Phase-field simulations confirm KGT's planar threshold","Rapid solidification: KGT predicts planar boundary across speeds","Unstable wavy microstructures emerge at low thermal gradient","Dendritic to planar: KGT and simulations converge"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the upscaled phase-field model with $S=5$ is converged enough near absolute stability; the paper itself reports that at $S=8$ the predicted planar transition at $V=0.3$ m/s and $G=10^7$ K/m disappears, so if $S=5$ is not the limiting behavior, the KGT agreement could be numerical.","fun_headline_variants_meta":{"raw":{"variants":["KGT nails planar solidification limit in rapid growth","Phase-field simulations confirm KGT's planar threshold","Rapid solidification: KGT predicts planar boundary across speeds","Unstable wavy microstructures emerge at low thermal gradient","Dendritic to planar: KGT and simulations converge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000317,"raw_usage":{"total_tokens":1847,"prompt_tokens":1055,"completion_tokens":792,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":671,"completion_tokens_details":{"reasoning_tokens":713}},"tokens_in":671,"tokens_out":792,"duration_ms":7864,"temperature":1.0,"reasoning_tokens":713,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:08:35.941398+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Rerun the same Fe-Cr simulations at $S=3$ or $S=1$ for the points near $G=2\\times 10^7$ K/m and $V=0.3$ m/s; if cellular patterns persist where $S=5$ produced planar growth, then the KGT $G_{\\max}$ agreement is a numerical artifact rather than a physical prediction.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the linear-stability definition of the absolute stability threshold that the paper compares against and reinterprets."},{"cited_title":"Tourret, J","cited_arxiv_id":null,"evidence_quote":"Provides the experimental and theoretical basis for using the maximum of the solidus temperature as the better planar-limit velocity."},{"cited_title":"Pinomaa, M","cited_arxiv_id":null,"evidence_quote":"Supplies the Fe-Cr pseudo-binary alloy parameters and the 316L surrogate baseline on which the simulations are built."},{"cited_title":"Physics-Informed Gaussian Process Classification for Constraint-Aware Alloy Design","cited_arxiv_id":"2502.11369","evidence_quote":"Supplies the Gaussian-process classification with informative priors that the Bayesian active-learning loop uses."}],"review_version":1}