{"id":"2206fd06-f95d-411d-aa2b-2418ad5e9fcc","arxiv_id":"2412.06270","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A machine-learned potential predicts that beta Ti-Nb-Zr alloys near dynamical instability have bone-like Young's modulus, strong elastic anisotropy, and elinvar behavior.","lead":"A machine-learned atomic model for beta titanium alloys was trained on density functional theory data and used to simulate elastic properties from 300 to 1300 K. The simulations predict that Ti-Nb-Zr alloys with 12 to 17 percent niobium have a Young's modulus of 30 to 40 GPa, close to human bone, which could guide biomedical implant design.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed 30–40 GPa window at 12–17 at.% Nb is interpolated from single-SQS data with no direct DFT validation, and the nearby instability border is inferred from MTP MSD/stress rather than phonons, so the quantitative low-modulus prediction is not yet secured.","rationale":"The paper does many things right: the MLIP is not fitted to elastic constants, 0 K Cij are checked against PAW, and the 40-configuration test set covers the low-Nb unstable alloys. However, the headline application-relevant number, EH = 30–40 GPa for xNb = 12–17, appears in Fig. 7 as a smooth interpolation between the unstable x=6 and stable x=15 compositions. No simulation was run at x=12 or x=17, no second SQS per composition was used, and the test set does not include strained configurations from those compositions. The instability borders themselves are located by a minimum in MSD and by nonzero stresses in MTP-MD, not by phonon calculations; the paper says it cannot be sure the dynamical and mechanical instabilities coincide. Since the modulus minimum is an effect of proximity to instability, a small error in the border or in C′ near the border translates directly into a large error in EH. The 0 K validation already shows MTP/DFT differences of 3–6 GPa in C12 and C44, which is the same size as the claimed modulus window. None of this invalidates the qualitative conclusion that low-Nb Ti-Nb-Zr softens near instability, but it does mean the specific 30–40 GPa / 12–17 at.% claim is conditional on validation the paper does not provide.","tokens_in":20698,"tokens_out":8481,"duration_ms":93396,"concrete_test":"Run direct DFT (VASP, same PAW/GGA settings) stress-strain calculations on 2–3 independent 128-atom SQS cells for Ti82Nb12Zr6 and Ti77Nb17Zr6 at 300 K, using MTP-thermalized structures and the same ±2%/±4% strains as in §2.4. Compute C11, C12, C44 and EH with the same averaging. If DFT-based EH falls outside 30–40 GPa by more than the ~5 GPa experimental offset, or if C′ < 0 at either composition, the interpolated window is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim, the 12–17 at.% Nb / 30–40 GPa window in Fig. 7, is an interpolation between two compositions (6 and 15 at.% Nb) that lie on opposite sides of the mechanical-instability line, and neither the MTP nor DFT was directly used to simulate or validate the claimed window at 300 K. The only independent validation in the unstable/near-unstable regime (Supp. S2) is 40 MD snapshots from Ti93Nb1Zr6 and Ti88Nb6Zr6, testing energies and stresses, not elastic constants or phonons. The instability borders in Fig. 7 are inferred from MSD minima and stress fluctuations in MTP-MD (Figs. 3–4), with the paper explicitly acknowledging it cannot be certain that the dynamical and mechanical instabilities coincide and that pure β-Ti has non-Γ instabilities. If the true instability border for xNb = 12–17 is shifted by even 2–3 at.% (or if the MTP softens C′ and C44 there by the 3–6 GPa level seen in the 0 K unrelaxed SQS validation), the 30–40 GPa window moves or disappears. This is the load-bearing step because the biomedical low-modulus claim is the reason for the study.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper trains a moment tensor potential (MTP) for β-Ti94−xNbxZr6 alloys with x = 1, 6, 15, 22, and 30 at.% Nb using DFT data and an active-learning scheme, then uses MTP-MD (LAMMPS) to compute finite-temperature elastic constants and polycrystalline moduli over 300–1300 K. The central claims are: (i) an elinvar effect over a wide temperature range; (ii) a strongly nonlinear concentration dependence of elastic moduli near dynamical/mechanical instability that yields a predicted 30–40 GPa Young's modulus for compositions with 12–17 at.% Nb at room temperature; and (iii) strong anisotropy of directional Young's modulus in the vicinity of instability. The MTP is validated against PAW-DFT at 0 K on relaxed and unrelaxed SQS supercells, and on an independent 40-configuration test set drawn from the unstable low-Nb alloys.","tokens_in":20968,"tokens_out":5926,"duration_ms":53837,"significance":"If the central prediction is reliable, the paper would extend machine-learned interatomic potentials to a regime—dynamically unstable near-instability compositions—where classical potentials fail, and would identify a concrete composition window for low-modulus biomedical Ti alloys. The workflow is a strength: the MTP is fitted to DFT energies/forces/stresses, not to experimental elastic constants; the elastic properties emerge from MD stress–strain simulations; the training dataset and SQS structures are deposited; and an active-learning protocol is used to include low-temperature unstable configurations. The qualitative trends (decreasing modulus with decreasing Nb, increased anisotropy near instability, elinvar-like weak temperature dependence of the stable alloys) match available experiments and prior DFT studies. However, the central quantitative claim—the 30–40 GPa window at 12–17 at.% Nb—is sensitive to interpolation between sparse compositions and to the indirect inference of the instability border; this sensitivity is not yet quantified.","major_comments":[{"comment":"The statement that β-Ti94−xNbxZr6 with 12–17 at.% Nb exhibits EH between 30 and 40 GPa is an interpolation across only five simulated compositions (x = 1, 6, 15, 22, 30), with only Ti79Nb15Zr6 lying inside the claimed window; the boundaries at x = 12 and x = 17 are not directly simulated. Because the paper itself emphasizes a strongly nonlinear concentration dependence of the moduli near instability (Section 3.3, Fig. 5f and Fig. 6), the color-map interpolation in Fig. 7 cannot be assumed accurate across the entire window. Please perform MTP-MD for at least one or two additional compositions inside the window (e.g., x = 12 and x = 14) and, ideally, targeted DFT validation at those compositions at 0 K or at 300 K, to support the specific 30–40 GPa claim.","section":"3.4, Fig. 7"},{"comment":"The instability border shown as dashed/solid lines in Fig. 7 is inferred from the minimum of the MSD versus temperature (Fig. 3) and from the appearance of nonzero stress components (Fig. 4). These are indirect criteria, and the paper explicitly acknowledges in Section 3.3 that 'we cannot be certain that dynamical and mechanical instabilities coincide' and that pure β-Ti has non-Γ instabilities. Because the claimed 30–40 GPa modulus window is located immediately adjacent to this border, a shift of the border by only 2–3 at.% Nb or by 50–100 K would move or eliminate the window. The authors should substantiate the stability boundary with a more direct method, such as phonon dispersions for representative SQSs at 0 K, or at least a finite-temperature spectral-function analysis using the MTP, cross-checked against DFT for a small number of configurations.","section":"3.2–3.3, Figs. 3–5"},{"comment":"The validation of the MTP in the near-instability regime is limited: the independent test set (Supp. S2) comprises only 40 MD snapshots from the two low-Nb alloys and checks energies and stresses, not elastic constants; the 0 K unrelaxed-SQS comparison (Supp. S5) shows systematic errors ΔC12 ≈ 5–6 GPa and ΔC44 ≈ 3–4 GPa. Since the predicted softening is driven by C′ = (C11 − C12)/2 approaching zero, a 5–6 GPa error in C12 translates to a ~2.5–3 GPa error in C′, which could shift the composition at which C′ crosses zero by several at.% Nb and materially change the predicted EH. This error budget is not reflected in the uncertainty of the central 30–40 GPa window; please quantify the propagation of the MTP error (e.g., by re-fitting or bootstrap) and, if feasible, recompute the finite-temperature EH for at least one composition in the claimed window directly from DFT MD.","section":"Supplement S2, S5 (Fig. S3, Fig. S6, Table S1)"}],"minor_comments":[{"comment":"The critical temperatures are stated inconsistently: the text says 'Ti93Nb1Zr6 is dynamically unstable below 450–500 K and Ti88Nb6Zr6 is unstable bellow 350–400 K,' and then gives critical temperatures of '~400 and ~5 00 K, respectively.' The numbers appear to be swapped; please correct and ensure consistency with the instability borders in Fig. 7.","section":"3.2, last paragraph"},{"comment":"There are several typos: 'aroud' should be 'around', 'expalined' should be 'explained', 'disscus' should be 'discuss', and 'emphisize' should be 'emphasize'.","section":"2.4"},{"comment":"The phrase 'underesteemates C44' should read 'underestimates C44'.","section":"3.1"},{"comment":"The word 'anomality' should be 'anomaly' in 'This anomality can be attributed to the dynamical instability.'","section":"3.2"},{"comment":"The central claim of a 30–40 GPa EH window is stated without an explicit uncertainty estimate; please report the numerical uncertainty of the predicted EH values (e.g., from the averaging over orientations and from the MTP energy/stress errors) and add error bars or a confidence interval to Fig. 7.","section":"3.4"}],"recommendation":"major_revision","confidential_remarks":"The paper presents a carefully executed active-learning MTP workflow and a relevant biomedical materials question, and the qualitative predictions align with prior experiments and DFT. However, the main quantitative prediction—the 12–17 at.% Nb / 30–40 GPa window—is an interpolation that is not yet supported by direct simulation at the claimed boundary compositions, and the instability border is inferred indirectly rather than from phonon calculations. These issues are addressable within the scope of the manuscript by adding targeted compositions and a phonon-based stability check. I therefore recommend major revision rather than rejection. The data-deposition statement and the fact that elastic properties are not fitted targets are positive aspects that should be retained."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid MLIP study that extends a known workflow to a new system and produces a plausible, practically interesting prediction of low Young's modulus in Ti-Nb-Zr near dynamical instability. The main quantitative claim, the 12–17 at.% Nb window at 30–40 GPa, is less secure than the rest of the paper because it rests on interpolation between an unstable and a stable composition, with no direct DFT or phonon validation in that window. I would send it to review, with the referee instructed to push for extra configurational sampling and a direct check of the predicted window.\n\nWhat is actually new: the same group has already used MTPs for elinvar in pure beta-Ti and for Ti0.5Al0.5N. Here they apply the method to a ternary Ti-Nb-Zr system and map how elastic constants, polycrystalline moduli, and directional Young's modulus evolve with Nb content and temperature, including the region where the bcc phase is dynamically unstable. The paper ships a working MTP, a clear active-learning protocol, and public data in an Open Materials Database, which is genuinely useful. The 0 K comparison against DFT on SQS cells is honest about the discrepancies: C12 off by 5–6 GPa, C44 by 3–4 GPa. The experimental comparison for Young's modulus follows the right trend, even if absolute values are a few GPa off.\n\nThe soft spots are where the main claim lives. The 30–40 GPa window at 12–17 at.% Nb is an interpolation between Ti88Nb6Zr6, dynamically unstable at 300 K, and Ti79Nb15Zr6, stable. The MTP was actively trained mostly in the stable, higher-temperature regime; only a 40-configuration test set from the two low-Nb alloys checks the unstable region, and it tests energies and stresses, not elastic constants or phonon instabilities. The instability borders are inferred from MSD minima and stress fluctuations, not from direct phonon calculations, and the authors themselves say they cannot be certain that the mechanical and dynamical instabilities coincide. Given the 3–6 GPa errors in C12 and C44, a small shift in the instability border could move the predicted window or eliminate it. Single SQS per composition is also a thin basis for a concentration map.\n\nNone of this sinks the paper. The qualitative story—instability-induced softening and elinvar behavior near the border—is well supported, and the workflow can be reproduced and checked. The reader for this paper is someone working on computational design of biomedical titanium alloys or on MLIPs for unstable systems. They will get a clear method and a concrete target composition to test. Whether the exact window survives a more thorough check is an open empirical question, not a reason to reject.\n\nMy recommendation: accept for peer review, and in the report ask for at least a second SQS per composition and direct DFT (or at minimum phonon) validation at two points inside the claimed 12–17 at.% window. With that, the quantitative claim could become solid.","headline":"Solid MLIP study with an honest validation effort, but the headline low-modulus window is interpolated across an instability border and needs direct checking before it should be quoted.","tokens_in":21541,"tokens_out":4132,"would_cite":false,"duration_ms":36983,"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":"Simulation finds bone-like stiffness in Ti-Nb-Zr alloys.","keywords":["machine learning interatomic potential","moment tensor potential","titanium alloys","elastic moduli","dynamical instability","Young's modulus","biomedical implants","elinvar effect"],"falsifier":"Measure the room-temperature Young's modulus of single-phase β-Ti-Nb-Zr samples with 12–17 at.% Nb and well-characterized composition; if the modulus is clearly outside 30–40 GPa, the central prediction is wrong. A complementary check is to compute phonon dispersions for Ti$_{93}$Nb$_1$Zr$_6$ and Ti$_{88}$Nb$_6$Zr$_6$ near the predicted critical temperatures: the absence of imaginary modes below roughly 500 K and 390 K would contradict the claimed dynamical instability.","tokens_in":20460,"feed_emoji":"🦴","tokens_out":9995,"duration_ms":89322,"temperature":0.7,"pith_summary":"This paper seeks to establish that a machine-learned interatomic potential can reliably simulate the finite-temperature elastic properties of multicomponent $\\beta$-Ti$_{94-x}$Nb$_x$Zr$_6$ alloys even in the vicinity of dynamical instability. The authors train a moment tensor potential with an active-learning loop and show that it reproduces density-functional-theory elastic constants at 0 K across five Nb concentrations. Their central prediction is that near the boundary of dynamical and mechanical instability, the elastic moduli depend strongly nonlinearly on Nb content, producing a room-temperature Young's modulus of 30–40 GPa for alloys with 12–17 at.% Nb, matching human bone. They also predict elinvar behavior, meaning the moduli stay nearly constant over a wide temperature range, and strong anisotropy of directional Young's modulus that could enable texture-based design of implants. If correct, the work offers a practical computational route to designing low-stiffness biomedical titanium alloys without expensive ab initio molecular dynamics.","feed_headline":"Simulation finds bone-like stiffness in Ti-Nb-Zr alloys","feed_subtitle":"A machine-learned potential predicts 30-40 GPa stiffness at 12-17% Nb, close to human bone.","key_machinery":"The central machinery is a moment tensor potential, a machine-learned interatomic potential that expresses total energy, forces, and stresses as linear combinations of tensor-valued basis functions of local atomic environments. Training proceeds by active learning: configurations with high extrapolation grade are collected from molecular dynamics and re-evaluated with density functional theory, iteratively extending the dataset to include the distorted, low-Nb, low-temperature configurations that appear near instability. Elastic constants are then extracted from stress–strain relations in NVT molecular dynamics on supercells of 128,000 atoms, and polycrystalline moduli are obtained by Voigt–Reuss–Hill averaging. The key physical quantity is the tetragonal shear modulus $C' = (C_{11}-C_{12})/2$: as composition and temperature approach the boundary where $C'\\to 0$, the material softens, the moduli become strongly nonlinear in composition, and directional anisotropy grows, producing the predicted low-modulus window.","core_discovery":"The paper claims that a suitably trained moment tensor potential captures the temperature- and composition-dependent elastic response of $\\beta$-Ti$_{94-x}$Nb$_x$Zr$_6$ solid solutions, including the weakly unstable low-Nb, low-temperature regime. On this basis the authors predict that Ti$_{93}$Nb$_1$Zr$_6$ and Ti$_{88}$Nb$_6$Zr$_6$ become dynamically and mechanically unstable below roughly 500 K and 390 K respectively, identified through anomalies in lattice parameters, mean-square displacements, and stress components. In the stable region, the computed Hill-averaged Young's modulus for 12–17 at.% Nb falls between 30 and 40 GPa at room temperature, comparable to human bone, and all mechanically stable compositions show elinvar-like weak temperature dependence of the elastic moduli. Near instability the alloys also exhibit strong elastic anisotropy, with the largest directional Young's modulus along [111] and the smallest along [100], which the authors connect to the same $C'$ softening that drives the low-modulus window.","pith_inferences":["This suggests the active-learning approach could be transferred to other metastable bcc alloys such as Ti-Mo or Ti-V, where the same $C'$-driven softening mechanism should produce analogous low-modulus windows.","If the 30–40 GPa window is confirmed experimentally, it would indicate that composition alone, without cold work or aging, can bring a $\\beta$-titanium alloy to bone-like stiffness, which would simplify implant manufacturing.","The instability boundaries inferred from mean-square displacement and stress anomalies could be sharpened by direct anharmonic phonon calculations; a mismatch there would not overturn the modulus prediction but would refine the stability map.","The predicted anisotropy suggests that polycrystalline samples with random texture should exhibit a wider modulus spread than strongly textured ones; comparing textured and untextured samples of the same composition would test the texturing design idea."],"forward_implications":["If the prediction is correct, $\\beta$-Ti$_{94-x}$Nb$_x$Zr$_6$ alloys with 12–17 at.% Nb give room-temperature Young's modulus of 30–40 GPa, matching the range needed for bone-implant compatibility.","The predicted elinvar behavior means that in mechanically stable compositions, elastic moduli remain nearly constant from 300 to 1300 K, so implants made from these alloys would resist stiffness changes with body temperature and processing heat.","The strong directional anisotropy, with maximum Young's modulus along [111] and minimum along [100], implies that controlling crystallographic texture during thermomechanical processing can tune the macroscopic stiffness without changing composition.","The identified instability boundaries, Ti$_{93}$Nb$_1$Zr$_6$ below about 500 K and Ti$_{88}$Nb$_6$Zr$_6$ below about 390 K, define a composition-temperature map that experimental synthesis could target to obtain metastable bcc alloys."],"supporting_citations":[{"why":"Establishes the moment tensor potential functional form used for the interatomic potential.","marker":"[28]"},{"why":"Provides the MLIP package and its active-learning machinery for iterative training.","marker":"[29]"},{"why":"Shows that the same potential class reproduces AIMD elastic constants and elinvar behavior in β-Ti, the methodological template.","marker":"[30]"},{"why":"Supplies the active-learning workflow for computing elastic properties of alloys that is extended to the unstable regime.","marker":"[31]"},{"why":"Prior demonstration of low elastic moduli and strong anisotropy near mechanical instability in bcc Ti-V alloys, motivating the target.","marker":"[54]"},{"why":"Introduces special quasirandom structures used to model chemical disorder in the alloys.","marker":"[61]"},{"why":"Experimental report of elinvar/invar gum metal used as the benchmark for the predicted elinvar behavior.","marker":"[73]"},{"why":"Experimental evidence of elinvar effect in a Ti-Nb-Zr alloy, supporting that weak temperature dependence is native to the β phase.","marker":"[77]"},{"why":"Experimental Young's modulus data for Ti-Nb-Zr alloys used to benchmark the room-temperature predictions.","marker":"[78]"}],"fun_headline_variants":["Machine learning finds bone-soft Ti-Nb-Zr alloys near instability","Simulation: titanium alloys soften to bone level under instability","Elinvar effect and bone-like stiffness predicted in Ti-Nb-Zr alloys","Machine-learned potential reveals soft titanium alloys mimic bone"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The predictions rest on the assumption that the machine-learned potential is accurate for the strongly distorted, low-symmetry atomic environments that appear in the unstable low-Nb alloys at 300–500 K, even though it was mostly trained on stable compositions and is validated by only 40 independent low-Nb configurations.","fun_headline_variants_meta":{"raw":{"variants":["Machine learning finds bone-soft Ti-Nb-Zr alloys near instability","Simulation: titanium alloys soften to bone level under instability","Elinvar effect and bone-like stiffness predicted in Ti-Nb-Zr alloys","Machine-learned potential reveals soft titanium alloys mimic bone"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000601,"raw_usage":{"total_tokens":2807,"prompt_tokens":942,"completion_tokens":1865,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":558,"completion_tokens_details":{"reasoning_tokens":1791}},"tokens_in":558,"tokens_out":1865,"duration_ms":14444,"temperature":1.0,"reasoning_tokens":1791,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T19:51:15.388550+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the room-temperature Young's modulus of single-phase β-Ti-Nb-Zr samples with 12–17 at.% Nb and well-characterized composition; if the modulus is clearly outside 30–40 GPa, the central prediction is wrong. A complementary check is to compute phonon dispersions for Ti$_{93}$Nb$_1$Zr$_6$ and Ti$_{88}$Nb$_6$Zr$_6$ near the predicted critical temperatures: the absence of imaginary modes below roughly 500 K and 390 K would contradict the claimed dynamical instability.","supporting_citations":[{"cited_title":"Shapeev, E.V","cited_arxiv_id":null,"evidence_quote":"Shows that the same potential class reproduces AIMD elastic constants and elinvar behavior in β-Ti, the methodological template."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the active-learning workflow for computing elastic properties of alloys that is extended to the unstable regime."},{"cited_title":"Skripnyak, F","cited_arxiv_id":null,"evidence_quote":"Prior demonstration of low elastic moduli and strong anisotropy near mechanical instability in bcc Ti-V alloys, motivating the target."},{"cited_title":"Saito, T","cited_arxiv_id":null,"evidence_quote":"Experimental report of elinvar/invar gum metal used as the benchmark for the predicted elinvar behavior."},{"cited_title":"Dubinskiy, G","cited_arxiv_id":null,"evidence_quote":"Experimental evidence of elinvar effect in a Ti-Nb-Zr alloy, supporting that weak temperature dependence is native to the β phase."},{"cited_title":"Kim, H.Y","cited_arxiv_id":null,"evidence_quote":"Experimental Young's modulus data for Ti-Nb-Zr alloys used to benchmark the room-temperature predictions."}],"review_version":1}