{"id":"3509b482-0419-49fa-87f3-11af467ffdf7","arxiv_id":"2607.23203","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"Machine-learned interatomic potentials and HSE06 DFT predict ultra-low lattice thermal conductivity and ZT above 1 over 600 K for half-Heusler LiCdSb.","lead":"Using DFT and a machine-learned interatomic potential, this paper computes a very low lattice thermal conductivity of 0.24 W/mK for the half-Heusler LiCdSb and predicts a thermoelectric figure of merit above 1 for temperatures over 600 K. If correct, LiCdSb would be a promising non-toxic material for high-temperature waste-heat recovery, but the supporting details and manuscript polish are currently insufficient.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The ZT>1 claim is unquantified: no ZT values above 500 K are given and no carrier concentration is specified, so the central result is not reproducible.","rationale":"I read the paper as a computational prediction that LiCdSb could achieve ZT>1 above 600 K, combining an HSE06 electronic band structure with an ML-based lattice thermal conductivity of 0.24 W/mK. For this to be true, the electronic transport model must be specified and realistic. The manuscript computes transport properties as functions of chemical potential (Section 3.3, Eqs. 6-8) but never states the carrier concentration or whether the reported ZT is the maximum over µ. Figure 11 shows curves, but numerical values above 500 K are not quoted. The 300 K comparison uses ZT_ML=0.17 vs experiment 0.10, which is not 'agreed well' but a 70% overestimate; this suggests the calculation is an upper bound at optimal doping, not a prediction for the measured sample. The deformation-potential relaxation time is also a coarse approximation. The reader's weakest assumption was MLIP transferability; I see that as secondary because the MLIP is used only to extract zero-temperature IFCs for the BTE, not to simulate high-T dynamics. The more decisive gap is the absence of a defined doping and the resulting unreproducible high-T ZT. A fixed-carrier-concentration re-analysis would settle whether the ZT>1 claim is physically meaningful. Given the incomplete specification and placeholder references, I concur with the reader's REJECT verdict.","tokens_in":28853,"tokens_out":8270,"duration_ms":73538,"concrete_test":"Using the same HSE06 band structure and ML-based κl, run BoltzTraP2 at fixed carrier concentrations n = 1e18, 5e18, 1e19, 5e19, 1e20 cm^-3 for both p- and n-type. Report the maximum ZT at 600, 700, 800, 900 K and the carrier concentration at which it occurs. If ZT never exceeds 1 at a fixed realistic concentration, the headline claim is an artifact of chemical-potential optimization. Also state explicitly whether the original ZT values in Fig. 11(i–l) are the maxima over µ.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract asserts ZT 'well above the benchmark value of ~1 beyond 600K', but the manuscript never reports the numerical ZT values above 500 K. Section 3.3 and Figure 11 show ZT curves, but no carrier concentration or chemical potential is given, and it is not stated whether the plotted ZT is the maximum over chemical potential (Eqs. 6-8 compute transport properties as functions of µ). The only quantitative comparison is ZT_ML = 0.17 at 300 K versus ZT_exp = 0.10, a 1.7× overestimate called 'agreed well'. This suggests the theoretical ZT may be an idealized limit at optimal doping, not a realistic prediction for a synthesizable sample. If the high-T ZT>1 is likewise the peak over µ, it is not a falsifiable claim about LiCdSb. The load-bearing condition — that the material can actually reach the high-T performance reported — is unsupported by any specification of the transport regime or dopability.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents a first-principles study of the half-Heusler compound LiCdSb. Structural, electronic, elastic, and thermodynamic properties are computed with DFT (GGA, GGA+SOC, HSE06), and thermoelectric transport coefficients are obtained with BoltzTraP2 under both CRTA and RTA. Lattice thermal conductivity is estimated with three approaches: the standard Slack model, a Slack model extended with temperature-dependent elastic constants (TDEC), and an on-the-fly machine-learned interatomic potential (MLIP). The authors report a room-temperature Kl of 0.24 W m^-1 K^-1, ZT values of 0.17-0.18 at 300 K, and claim that ZT exceeds ~1 above 600 K, making LiCdSb potentially promising for high-temperature thermoelectric applications.","tokens_in":29176,"tokens_out":5604,"duration_ms":56447,"significance":"If substantiated, the claim of ZT > 1 above 600 K would be a significant result for half-Heusler thermoelectrics, especially because the available experimental ZT is only ~0.10 at 300 K. The computational pipeline combining HSE06 electronic structure with MLIP-based thermal transport is modern and, in principle, well suited to this problem. The MLIP force/energy RMSEs are encouraging, and the comparison with experimental ZT at 300-500 K is a useful benchmark. However, the paper currently does not provide the quantitative inputs needed to assess the central claim: no numerical ZT values above 500 K, no carrier concentration or chemical potential, and no demonstration that the MLIP trained at 300 K is transferable to 900 K. The significance is therefore potential rather than established.","major_comments":[{"comment":"The central claim 'ZT well above the benchmark value of ~1 beyond 600 K' is never quantified. The text and Figure 11 give no numerical ZT values above 500 K, no chemical potential or carrier concentration at which the curves are evaluated, and no statement whether the plotted ZT is the maximum over chemical potential. Since Eq. (8) defines all transport coefficients as functions of μ and the p-/n-type labels are not tied to a specific μ or carrier density, the high-temperature claim is not reproducible. Please report the peak ZT, the corresponding μ or carrier concentration, and the temperature for each functional and κl model.","section":"Abstract and §3.3, Fig. 11"},{"comment":"The MLIP is trained on 10,000 AIMD steps (10 ps) at 300 K only, but is used in Phono3py to compute second- and third-order IFCs and Kl over 300-900 K. The force/energy RMSEs of 0.010 eV/Å and 0.032 eV are parity metrics on training configurations; they do not establish transferability of the anharmonic potential surface to high temperature or to displaced configurations far from the training set. Please validate the MLIP phonon dispersions and Kl against direct DFT/DFPT or Phono3py results at least at 300 K with independent supercells, and ideally at elevated temperatures. Also report convergence of the 4×4×4 supercell and the 30×30×30 q-point mesh.","section":"§2.2 and Fig. 2"},{"comment":"The RTA transport results and all ZT values obtained with finite relaxation time depend on the deformation potential constants Ed, effective masses, and elastic constants through Eq. (10). The numerical values of Ed for the CBM and VBM are not reported, and the main text does not give the actual τ(T) values or the carrier concentration used. Without these inputs, the RTA panels in Figure 11 and the comparison with experiment cannot be reproduced or independently checked. Provide a table of Ed, τ0/τ(T), and the μ or carrier concentration used for each panel.","section":"§3.3, Eq. (10), Fig. S5"},{"comment":"The rejection of the QHA results because C44 becomes negative near ~600 K is not sufficiently justified. The paper argues that no phase transition is known, but a negative C44 in QHA can also indicate a numerical/methodological artifact or a genuine tendency toward mechanical instability that a static QSA would miss. Since the Slack+TDEC Kl curve and the corresponding ZT panels in Fig. 11 rely on the QSA elastic constants, this choice should be supported by convergence tests (elastic constants vs q-mesh, smearing, volume sampling) or by comparison with any available experimental elastic data.","section":"§3.2, Fig. 5"}],"minor_comments":[{"comment":"Placeholder references [126], [128], and [129] (e.g., 'A. B. Surname', 'F. N. Hyphenated-Lastname') must be replaced. The manuscript appears to contain template entries.","section":"References"},{"comment":"Figure 11 has panels (a)-(m), but the caption does not describe individual panels, line styles, or the exact definition of p-/n-type doping (fixed μ, fixed carrier concentration, or maximized over μ). Please expand the caption.","section":"Fig. 11 caption"},{"comment":"The phrase 'agreed well' is used for ZT_ML = 0.17 versus ZT_exp = 0.10 at 300 K; this is a 1.7× overestimate. The manuscript should describe this as qualitative agreement and note that the 500-K comparison (0.37 vs 0.32) is closer.","section":"Abstract and §3.3"},{"comment":"Several typos and notation inconsistencies should be corrected: 'anab-initiomolecular dynamics', 'Thermo_PW' vs thermo_pw, 'TEDC' vs TDEC, 'Boltztrap2' vs BoltzTraP2, and inconsistent κ_l/Kl notation.","section":"Throughout"},{"comment":"The text says the FMLP model was trained using 'the machine learning module' but does not state which code/package (e.g., VASP ML module, NEP, GAP) was used, nor the number of training/validation structures. Please specify.","section":"§2.2"}],"recommendation":"major_revision","confidential_remarks":"I agree with the reader's overall assessment that the paper is not acceptable in its present form, but I do not see a fundamental error that would force rejection. The main gaps are missing quantitative transport inputs (μ/n, Ed, τ, numerical ZT above 500 K) and the lack of MLIP transferability validation. These can be addressed in revision. The placeholder references and incomplete figure captions also need attention before resubmission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this one is not ready for circulation. The headline claim (ZT>1 above 600 K) is asserted without numbers or doping levels, and the reference list contains four placeholder entries (refs 126-129). That combination makes the paper unpublishable as-is, and the central result is not reproducible.\n\nWhat is actually new: an on-the-fly MLIP calculation of lattice thermal conductivity for LiCdSb, compared with Slack and Slack+TDEC estimates, and combined with HSE06 transport. The MLIP phonon dispersion and parity RMSEs (~0.01 eV/Å forces, 0.032 eV energies) look reasonable. The paper also does the usual groundwork well: structure, Bader charges, elastic constants, and a comparison of GGA/SOC/HSE06 band structures.\n\nThe soft spots, in order of severity. First, the ZT>1 statement: the abstract gives ZT at 300 and 500 K (0.17-0.18 and 0.37, vs experimental 0.10 and 0.32), but never reports the ZT above 500 K. No carrier concentration or chemical potential is specified for the ZT curves. So the paper's most exciting sentence is not backed by any number a reader could check. Second, the MLIP is trained on 10 ps of AIMD at 300 K and then used to compute Kl up to 900 K. There is no transferability test—no comparison of MLIP forces against DFT at higher temperatures. The RMSEs are on training data. Third, the placeholder references (e.g. 'A. B. Surname' etc.) are a clear sign the draft is unfinished. Fourth, prior theory for this compound (Pallavi et al., ref [88]) is cited for the lattice parameter but not compared for transport, which undercuts the 'new ZT>1' claim. Fifth, the QSA-over-QHA choice is post-hoc; the authors justify it, but it still biases the Kl result.\n\nNone of this is fatal in the sense that the methods are wrong. The MLIP workflow is sensible and the experimental comparison at 300/500 K is decent. But the load-bearing claim needs quantitative backing, and the manuscript needs a hygiene pass. Who reads this: computational thermoelectrics people, especially those using MLIPs for Kl. If the authors fix the references, report ZT values and doping, and add MLIP transferability checks, it could be a useful case study. In current form I would not put a serious referee on it; I'd desk-reject and invite a resubmission.","headline":"Unquantified ZT>1 claim and placeholder references make this an unfinished draft, despite a sensible MLIP-based Kl workflow.","tokens_in":29620,"tokens_out":4171,"would_cite":false,"duration_ms":38123,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"LiCdSb, a half-Heusler semiconductor, is predicted to have ultralow lattice thermal conductivity (0.24 W/mK at 300 K) and a thermoelectric figure of merit above 1 beyond 600 K, via a machine-learned interatomic potential.","keywords":["half-Heusler","lattice thermal conductivity","machine-learned interatomic potential","thermoelectric figure of merit ZT","LiCdSb","density functional theory","Boltzmann transport","hybrid functional"],"falsifier":"Measure the lattice thermal conductivity of LiCdSb at 600-900 K experimentally (e.g., on a dense polycrystalline pellet via laser flash), or compute κl from DFT-quality third-order interatomic force constants at high temperature; if the measured/computed κl comes out above about 0.5 W/mK in that range, the ZT>1 prediction collapses. A cheaper check: run the same MLIP workflow with AIMD data points at 700 K and 900 K and compare force predictions to fresh DFT forces.","tokens_in":28752,"feed_emoji":"⚡","tokens_out":5708,"duration_ms":47844,"temperature":0.7,"pith_summary":"This paper is trying to establish that LiCdSb, a lithium-cadmium-antimony half-Heusler compound, is a promising high-temperature thermoelectric material because its lattice thermal conductivity is very low. Using density functional theory with a hybrid functional for an accurate band gap, and a machine-learned interatomic potential to compute lattice thermal conductivity cheaply, the authors obtain a room-temperature value of 0.24 W/mK and a thermoelectric figure of merit ZT that rises above the benchmark of 1 beyond 600 K. The paper matters because high-temperature waste-heat recovery needs materials with high ZT and low thermal conductivity, and a non-toxic, earth-abundant candidate would be valuable. It also demonstrates a cost-saving pipeline: replace expensive anharmonic phonon calculations with a machine-learned potential, yielding results in qualitative agreement with experiment at 300 K.","feed_headline":"Machine-learned potential finds ZT>1 in half-Heusler LiCdSb","feed_subtitle":"LiCdSb's predicted heat conductivity of 0.24 W/mK at 300 K pushes its thermoelectric efficiency past the benchmark at high temperature.","key_machinery":"The load-bearing machinery is the on-the-fly machine-learned interatomic potential (MLIP): a potential trained on 10,000 ab initio molecular dynamics steps at 300 K (10 ps), from which second- and third-order interatomic force constants are extracted and fed into the phonon Boltzmann transport equation to obtain κl. This replaces the expensive DFT-based anharmonic force-constant calculation. Alongside it, the paper couples a hybrid HSE06 band structure with Boltzmann transport for electronic coefficients, and uses a temperature-dependent elastic-constant (TDEC, quasi-static) extension of the Slack model as a cheaper cross-check. The MLIP is the piece doing the central work: it produces the u","core_discovery":"On the paper's own terms, the central discovery is that LiCdSb combines a HSE06-corrected electronic structure—a direct gap of 0.92 eV with light electrons and coexisting light and heavy holes—with a machine-learning-derived lattice thermal conductivity of 0.24 W/mK at room temperature, low enough that the thermoelectric figure of merit ZT = S²σT/(κe+κl) reaches values well above 1 for temperatures beyond 600 K. The authors argue that the ML-based κl reproduces the experimental ZT at 300 K (computed 0.17 vs experimental 0.10) more closely than the standard Slack model, with the agreement ranking ML + HSE06 > Slack+TDEC + HSE06 > Slack + HSE06. If correct, LiCdSb is an ultralow-κl half-Heusle","pith_inferences":["Inference: The ZT>1 claim relies on the MLIP extrapolating to 900 K from a 300 K training set; a transferability test (e.g., a handful of AIMD forces at 600-900 K) would settle whether the anharmonicity is genuinely captured or under-captured.","Inference: The paper's qualitative 300 K agreement (0.17 vs 0.10 experimental ZT) leaves room for the true high-temperature ZT to be lower; the plotted ZT>1 region may be an upper bound unless the carrier concentration is optimized—the paper does not report the chemical potential or doping level used in Figure 11.","Inference: Because LiCdSb is a Zintl-type half-Heusler with underbonded Cd, the same MLIP workflow could be applied to isoelectronic siblings (LiZnSb, LiMgSb, etc.) to see whether ultralow κl is a family trait rather than a single-compound accident.","Inference: A direct experimental measurement of κl on a single crystal at 300-900 K would be the cleanest test; if κl rises above about 0.5 W/mK at temperature, the ZT>1 claim would not survive."],"forward_implications":["If κl = 0.24 W/mK holds, LiCdSb's lattice thermal conductivity is among the lowest computed for half-Heuslers, making the material a candidate for thermoelectric generators operating on waste heat above 600 K.","The ML-assisted pipeline shows that machine-learned interatomic potentials trained on short AIMD trajectories can substitute for direct DFT anharmonic phonon calculations, making high-throughput screening of thermoelectric materials more feasible.","The computed ZT at 300 K (0.17) sits close to the experimental value (0.10) when HSE06 electronic structure is combined with the ML κl, indicating that both electronic-structure accuracy and lattice-transport accuracy are needed for quantitative predictions.","Below about 600 K the material's ZT is modest, so the payoff is specifically in high-temperature applications; doping or alloying that preserves the low κl could push ZT further.","The temperature-dependent elastic analysis (TDEC) supports the thermal stability picture, although the paper flags that the quasi-harmonic elastic constants behave anomalously; this tempers but does not remove the conclusion."],"fun_headline_variants":["ML-assisted κl gives LiCdSb ZT>1 at high temperature","LiCdSb thermoelectric: machine-learned κl yields ZT>1","Half-Heusler LiCdSb: low thermal conductivity via ML, ZT>1","LiCdSb's 0.24 W/mK from ML points to ZT>1 above 600K","Machine learning predicts LiCdSb ZT>1 for high-temp energy"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The paper's prediction of ultralow lattice thermal conductivity, and hence ZT>1 above 600 K, rests on the assumption that the machine-learned interatomic potential trained on only 10 ps of 300 K ab initio molecular dynamics remains accurate for anharmonic phonon-phonon interactions across the full 300-900 K range.","fun_headline_variants_meta":{"raw":{"variants":["ML-assisted κl gives LiCdSb ZT>1 at high temperature","LiCdSb thermoelectric: machine-learned κl yields ZT>1","Half-Heusler LiCdSb: low thermal conductivity via ML, ZT>1","LiCdSb's 0.24 W/mK from ML points to ZT>1 above 600K","Machine learning predicts LiCdSb ZT>1 for high-temp energy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000246,"raw_usage":{"total_tokens":1437,"prompt_tokens":869,"completion_tokens":568,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":465}},"tokens_in":613,"tokens_out":568,"duration_ms":5038,"temperature":1.0,"reasoning_tokens":465,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:18:21.870319+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the lattice thermal conductivity of LiCdSb at 600-900 K experimentally (e.g., on a dense polycrystalline pellet via laser flash), or compute κl from DFT-quality third-order interatomic force constants at high temperature; if the measured/computed κl comes out above about 0.5 W/mK in that range, the ZT>1 prediction collapses. A cheaper check: run the same MLIP workflow with AIMD data points at 700 K and 900 K and compare force predictions to fresh DFT forces.","supporting_citations":[],"review_version":1}