{"id":"3b5e1a19-08d2-41e7-9a3b-5aee9bbd8022","arxiv_id":"2412.16741","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"Finite-temperature entropic effects, computed with machine-learned force fields and thermodynamic integration, increase the predicted concentration of the Te_i+1 defect in CdTe by two orders of magnitude, while harmonic approximations remain adequate.","lead":"The authors used machine-learned atomic force models to simulate point defects in cadmium telluride at operating temperatures. They found that thermal vibrations and structural flexibility can raise the predicted abundance of one defect by roughly 500 times, which changes how defects in solar-cell materials should be modelled.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The anharmonic TI free energy includes site-to-site migration entropy, which is then likely double-counted by the N_sites prefactor in Eq. (4); the claimed two-orders-of-magnitude concentration increase may be largely an artifact.","rationale":"I read the paper as a careful methodological study whose quantitative headline is the factor-of-500 concentration enhancement for Te_i+1. The paper has real strengths: the MLFF validation is thorough, the thermodynamic integration is converged to small error bars, cross-method agreement is shown, and the treatment of the Te reservoir across the melting transition is sophisticated. However, the most load-bearing assumption is the treatment of defect diffusion during TI. The paper itself flags this in the Methods note, but dismisses the contribution as small. The argument that the defect spends most time near local minima confuses the time fraction in wells with the entropic multiplicity of wells: even if the defect visits each minimum only briefly, the number of distinct minima it can access contributes kBT ln(number of minima) to the free energy. For a 64-atom supercell with 32 primitive cells and 12 equivalent interstitial sites per primitive cell, that is roughly 0.43 eV at 840 K, comparable to the entire claimed 0.5 eV thermal correction. If this entropy is already inside g_TI_f, then Eq. (4), which multiplies by N_sites, double-counts the site degeneracy and the concentration enhancement is largely an artifact. The reader's weakest-assumption statement identified the diffusion issue as one of two concerns, but framed it mainly as a question of MACE accuracy; my concern is more specific and quantitative. I therefore recommend the verdict remain CONDITIONAL, with the added condition that the authors either restrain the defect during the anharmonic TI or explicitly correct Eq. (4) for the migration entropy already sampled, and show that the factor-500 result survives that correction.","tokens_in":21283,"tokens_out":14595,"duration_ms":136465,"concrete_test":"Recompute the temperature-scaling TI for Te_i+1 at 840 K with the interstitial restrained to its initial lattice site (e.g., by a flat-bottom harmonic potential on the Te_i-Cd distance that prevents hopping), and compare the resulting g_TI^restrained with the published unrestrained value. If the unrestrained free energy is lower by approximately kBT ln(M*N_sites) ≈ 0.43 eV, or equivalently if using N_sites=1 in Eq. (4) reproduces the published concentration, then the site entropy is double-counted and the factor-500 claim must be revised. A complementary check is to count the distinct interstitial sites visited in the published 840 K TI trajectories; if the number approaches M*N_sites = 384, the migration-entropy contribution is not small.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim that thermal effects raise the Te_i+1 concentration by a factor of 500 rests on the assumption that the thermodynamic-integration free energy g_TI_f corresponds to formation at a fixed lattice site. The Methods explicitly note that defect diffusion occurs during the temperature-scaling runs and assume its contribution is small because the defect spends most time near local minima. That assumption is questionable for a multi-well system: at 840 K the interstitial hops between equivalent sites on a sub-nanosecond timescale, so the unconstrained TI partition function contains a factor M*N_sites, where M=32 primitive cells in the 64-atom supercell and N_sites=12 for the C2v ground state. Thus g_TI_f ≈ g_f^site - kBT ln(M*N_sites) ≈ g_f^site - 0.43 eV at 840 K. Eq. (4) then multiplies by N_sites/V, so the site degeneracy is counted twice, inflating the concentration by exp(0.43 eV/kBT) ≈ 400. This factor is the same order as the reported two-orders-of-magnitude enhancement. The good agreement between harmonic and anharmonic methods in Fig. 3b, cited as validation, is precisely what needs re-examination: if the anharmonic path carried an extra -0.43 eV of site entropy, it would not agree with a single-site harmonic calculation unless the harmonic result also implicitly includes migration entropy, which it should not.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper trains MACE machine learning force fields on DFT data for CdTe and two charged defects, Te_i^+1 and V_Te^+2, and uses them to compute defect formation free energies at finite temperature. The authors compare harmonic, quasiharmonic, and fully anharmonic (thermodynamic integration) treatments and add electronic, spin, orientational, and structural entropy contributions. They report that Te_i^+1 is dynamically active at 300 K (configurational, orientational, and migrational motions) and that its formation free energy at 840 K differs by about 0.5 eV from the static 0 K internal-energy estimate, increasing the predicted concentration by a factor of about 500. For V_Te^+2, entropic effects are found to be negligible. The central claim is that finite-temperature entropy dominates the defect formation thermodynamics for defects that undergo structural reconstructions and have low-energy metastable configurations.","tokens_in":21565,"tokens_out":13521,"duration_ms":120303,"significance":"If the quantitative claim is correct, the paper provides a strong case against the standard static 0 K approximation for defect concentrations and demonstrates a practical MLFF-based route to anharmonic defect free energies. The study is carefully executed in many respects: the force fields are validated against independent DFT test sets, the potential-energy surface along the configurational path matches DFT, thermodynamic-integration convergence is checked with low standard errors, and the computed Te melting point (704 K) is close to experiment (722 K). The authors also make transparent use of open-source tools (doped, ShakeNBreak, calphy, phonopy). However, the central factor-of-500 concentration enhancement rests on the relation between the unconstrained thermodynamic-integration free energy and the fixed-site formation free energy used in Eq. (4). If the TI free energy includes the site-to-site migration entropy, the reported enhancement may be largely an artifact of double-counting the site multiplicity. This issue is load-bearing for the paper's main conclusion and must be resolved before the quantitative claims can be accepted.","major_comments":[{"comment":"The thermodynamic integration is performed on an unconstrained 65-atom supercell containing one Te_i^+1 interstitial, and the Methods explicitly note that defect diffusion occurs during the temperature-scaling runs. For a defect that can hop among the M=32 primitive cells of the 2x2x2 conventional supercell and N_sites=12 symmetry-equivalent interstitial sites per primitive cell, the unconstrained partition function contains a factor M*N_sites=384 relative to a defect pinned at a single site. This contributes a site entropy of k_B T ln(384) ≈ 0.43 eV at 840 K that is included in g_TI_f. Equation (4) then multiplies by N_sites/V, counting the site degeneracy a second time. The statement that the migration contribution is small because the defect spends most time near local minima is not a valid statistical-mechanical argument: the free-energy weight of additional basins is determined by their number and phase-space volume, not by the residence time in transition states. Since the double-counting factor is of the same order as the reported factor-of-500 concentration enhancement, the central claim of the paper is not established by the present calculations.","section":"Methods, Thermodynamic integration; Eq. (4)"},{"comment":"The agreement between the harmonic and anharmonic free energies in Fig. 3b is cited as validation of the harmonic approximation and of the decoupling assumption. However, if the anharmonic TI free energy includes the site-to-site migration entropy discussed above, it cannot agree with the harmonic free energy, which is computed for a single fixed interstitial site with phonopy. The authors need to reconcile this contradiction: either the TI path does not actually sample the full site multiplicity (in which case the 'fully anharmonic' label and the diffusion statement in Methods are misleading), or the harmonic calculation inadvertently contains the same multiplicity. A direct test would be to compute the TI free energy with the interstitial restrained to one lattice site (e.g., a harmonic umbrella restraint) and compare it with the unrestrained value; the difference quantifies the migration entropy and should be removed from g_TI_f before applying Eq. (4).","section":"Fig. 3b and Eq. (3)"}],"minor_comments":[{"comment":"The abstract states that thermal effects increase the predicted concentration by 'two orders of magnitude', while the main text reports a factor of 500; please use a single, consistent quantitative statement.","section":"Abstract and Section II.C"},{"comment":"The text defines N_sites as the number of symmetry-equivalent sites per primitive cell but calls V the 'crystallographic unit cell volume'. For CdTe the conventional cell is four times the primitive cell, so the concentration would be off by a factor of four if the conventional cell volume were used. Please specify explicitly which volume is used and keep the definitions consistent.","section":"Eq. (4)"},{"comment":"The analytical structural entropy gives s_struc(840 K) = 0.6 k_B (reported as 0.7 k_B in Fig. 3a), while the inherent-structures method gives 1.05 k_B. The text says these are in the same order, but the discrepancy is nearly a factor of two; a brief justification for using the analytical value and an estimate of its uncertainty would strengthen the analysis.","section":"Methods, Structural entropy"},{"comment":"The electronic entropy calculation relies on a mid-gap Fermi level for the defect supercell and an assumed excess electron concentration n = 10^15 cm^-3. Since the electronic entropy contribution is small (0.1 k_B), this does not affect the conclusions, but a sensitivity test or justification for these choices would be useful.","section":"Methods, Electronic entropy"}],"recommendation":"major_revision","confidential_remarks":"The site-degeneracy issue in the thermodynamic integration is the key point to resolve. If the TI free energy indeed contains the M*N_sites translational entropy, the main quantitative claim may be largely artifactual; if it does not, the Methods description needs to explain why the defect diffusion does not contribute, and the quoted error bars should reflect the resulting systematic uncertainty. This is a fixable issue in principle, but it is central enough that I cannot recommend acceptance without seeing the constrained-TI comparison or an equivalent quantitative treatment."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: this is a very solid computational paper. The headline result—thermal effects raise the predicted Te_i+1 concentration in CdTe by about two orders of magnitude—is not a fluke of the thermodynamic integration; the harmonic calculation independently gives the same order of magnitude. Second thing: there is a genuine, unanswered question about how the TI treats defect migration, and the authors' one-paragraph dismissal is the weakest part of the paper.\n\nWhat's new: a systematic comparison of harmonic, quasiharmonic, and fully anharmonic TI free energies for a very dynamic interstitial in CdTe, using one MLFF per system, with test-set errors, PES mapping against DFT, convergence checks, and a Te melting point reproduced at 704 K vs 722 K. They decompose the entropy into electronic, spin, orientational, structural, and vibrational pieces. That collection is a practical, reproducible workflow.\n\nThe soft spot is migration entropy. The Methods note that the interstitial diffuses during the temperature-scaling TI runs and assert the contribution is small because the defect spends most time near local minima. That argument isn't persuasive: if the defect can visit multiple sites, the partition function carries a factor proportional to M*N_sites. If g_TI_f includes that, then Eq. (4) double-counts N_sites. If it doesn't, the observed diffusion during the runs is inconsistent with the result. The harmonic-anharmonic agreement in Fig. 3b suggests the issue is not inflating the main claim—a 384-fold overcount would put a ~0.4 eV gap between the two curves—but the authors should prove it with a constrained-site TI or by subtracting kBT ln(M*N_sites) explicitly.\n\nMinor weaknesses: code and fitted models are promised on Zenodo but not yet there, so full reproducibility is not verifiable today. The electronic entropy uses mid-gap Fermi levels and a fixed DOS; reasonable, but a sensitivity scan would be welcome. The structural entropy from the inherent-structures method (1.05 k_B) sits above the analytical estimate (0.6 k_B) without a clean explanation.\n\nVerdict: the central message—thermal effects can change defect concentrations by orders of magnitude for reconstructing interstitials, and the harmonic approximation captures most of it—is well supported. The migration-entropy question and the data release should be handled in revision. This deserves a serious referee, not a desk reject.","headline":"Very solid benchmark of finite-temperature defect free energies; the main thermal-effect claim survives even if the TI migration-entropy handling needs a fix.","tokens_in":22072,"tokens_out":17446,"would_cite":true,"duration_ms":149862,"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":"Thermal entropy, not just the 0 K internal energy, sets the formation free energy of the tellurium interstitial in CdTe, raising its predicted equilibrium concentration by roughly two orders of magnitude.","keywords":["point defects","defect formation free energy","machine learning force fields","finite-temperature thermodynamics","thermodynamic integration","CdTe","tellurium interstitial","configurational entropy"],"falsifier":"Repeat the fully anharmonic thermodynamic-integration calculation for $\\mathrm{Te_i^{+1}}$ at 840 K using direct DFT energies on the same 65-atom supercell and compare the formation free energy with the machine-learned result; a disagreement larger than about $0.02$ eV per supercell (the reported TI error) would undermine the claimed 0.5 eV thermal shift and the roughly 500-fold concentration increase. An independent check would be a temperature-dependent experimental measurement of interstitial concentration in CdTe, if the predicted slope of $\\ln[c]$ versus $1/T$ can be resolved.","tokens_in":21080,"feed_emoji":"⚛️","tokens_out":10914,"duration_ms":87569,"temperature":0.7,"pith_summary":"The paper argues that the usual shortcut for predicting point-defect concentrations—using the 0 K internal energy of the defect's ground-state structure as a stand-in for the formation free energy—can fail badly for defects that are dynamic at operating temperatures. Using CdTe as a test case, it trains a machine-learned force field to follow the tellurium interstitial $\\mathrm{Te_i^{+1}}$, which flips between two near-degenerate configurations, migrates between sites, and rotates its Te–Te bond on the nanosecond timescale at 300 K. Combining harmonic, quasiharmonic, and fully anharmonic thermodynamic-integration free energies with analytical estimates of spin, electronic, orientational, and configurational entropy, the paper finds that the formation free energy $g_f$ at the 840 K annealing temperature is about 0.5 eV below the 0 K value $u_f(0\\,\\mathrm{K})$, raising the predicted interstitial concentration by roughly two orders of magnitude. By contrast, the tellurium vacancy $V_{\\mathrm{Te}}^{+2}$, which has no accessible metastable states and no symmetry-breaking, shows a negligible thermal correction. The message is that finite-temperature entropy dominates defect thermodynamics precisely when a defect reconstructs its local structure and has low-energy competing configurations.","feed_headline":"Thermal motion raises predicted CdTe defect count 500-fold","feed_subtitle":"Machine-learned force fields show entropy, not static energy, controls the tellurium interstitial at 840 K.","key_machinery":"The central machinery is a machine-learned force field trained on DFT data for each species in the defect reaction, used as a surrogate to run nanosecond NPT molecular dynamics and thermodynamic integration. The free-energy calculation proceeds by non-equilibrium thermodynamic integration—a reversible-switching method that starts from an Einstein crystal of independent harmonic oscillators, integrates to the anharmonic crystal at 100 K, and then sweeps temperature up to 840 K—with the electronic and spin entropies added analytically; separately, the harmonic and quasiharmonic vibrational free energies are obtained from phonon calculations. The defect's active degrees of freedom (configurational, orientational, and migrational) are the mechanism that generates the entropy, and the paper's decomposition shows the vibrational contribution dominates.","core_discovery":"For a defect whose local geometry is bistable—$\\mathrm{Te_i^{+1}}$ in CdTe, with a $\\mathrm{C}_{2v}$ split-interstitial ground state and a $C_s$ metastable configuration only 18 meV higher—the thermal formation free energy is not the 0 K internal energy. Molecular dynamics with a machine-learned force field shows the interstitial changes configuration, hops between lattice sites, and reorients within nanoseconds at 300 K, with barriers of 28–100 meV. Summing the entropic contributions, the vibrational term dominates ($4.2\\,k_{\\mathrm{B}}$ at 840 K), followed by structural and spin terms; the net effect lowers $g_f$ by 0.5 eV relative to $u_f(0\\,\\mathrm{K})$ and multiplies the predicted equilibrium concentration by a factor ~500. The anharmonic thermodynamic-integration result agrees with the harmonic approximation to within the computed error, because anharmonic contributions largely cancel between the defective and pristine supercells. $V_{\\mathrm{Te}}^{+2}$ behaves classically: it stays in its $T_d$ ground state and its formation free energy shifts by only 0.08 eV, showing the contrast is defect-specific, not a general failure of the 0 K picture.","pith_inferences":["This result suggests that defect-population models which solve for the Fermi level self-consistently will see temperature-dependent shifts in charge-state transitions for any defect with an accessible metastable manifold, not just in CdTe.","A concrete extension would be to run the same protocol on a reconstructive defect in a soft lattice such as a halide perovskite, where anharmonic cancellation between bulk and defect is less likely and the harmonic approximation may break down.","Because the electronic entropy term was evaluated with a functional that underestimates the band gap, the small electronic contribution is the least certain part of the entropy budget; training the force field on hybrid-functional data would test whether that term changes the total formation free energy by more than the reported thermodynamic-integration error.","The paper's recommendation to train on both pristine and defective supercells has a practical consequence: defect-focused machine-learned force fields trained only on defective cells may be unsuitable for computing absolute formation energies in larger supercells."],"forward_implications":["The predicted equilibrium concentration of the tellurium interstitial in CdTe at annealing temperatures is roughly two orders of magnitude higher than the static 0 K estimate, so relative defect populations and derived carrier concentrations shift accordingly.","The 0 K description remains adequate for defects like the tellurium vacancy that keep a single high-symmetry ground state and lack low-energy metastable configurations.","For dynamic defects, the harmonic approximation to the vibrational formation entropy matches the fully anharmonic thermodynamic-integration result at 840 K, validating cheaper phonon-based studies in similar cases.","High-temperature applications—thermoelectrics, catalysts, fuel cells—are where thermal corrections to defect formation are large enough to matter, and charge-transition levels may also move with temperature.","Machine-learned force fields need training sets that include both pristine and defective supercells if they are to be used for absolute defect formation energies in larger cells."],"supporting_citations":[{"why":"Defines the two low-energy Te_i^+1 configurations (C2v ground state and Cs metastable state) whose 18 meV separation is the basis of the bistability argument.","marker":"[12]"},{"why":"Provides the machine-learned interatomic potential architecture used to run the molecular dynamics and thermodynamic-integration simulations.","marker":"[83]"},{"why":"Supplies the non-equilibrium thermodynamic-integration and reversible-scaling implementation used for the fully anharmonic free energies.","marker":"[81]"},{"why":"Identifies ground-state and metastable defect structures, feeding the structural entropy contributions.","marker":"[85]"},{"why":"Automates defect-supercell setup, charge corrections, and the spin and orientational degeneracy factors entering the formation free energy.","marker":"[92]"},{"why":"Provides the harmonic and quasiharmonic phonon framework used to compute the vibrational free-energy contributions.","marker":"[86]"},{"why":"Establishes the defect-formation free-energy formalism and the decomposition into vibrational, spin, electronic, orientational, and structural entropy.","marker":"[2]"},{"why":"Supplies the inherent-structures method used as an independent cross-check of the configurational entropy.","marker":"[25]"}],"fun_headline_variants":["Entropy boosts CdTe interstitial concentration 500-fold","Finite-temperature effects multiply CdTe defect count by 500","Machine learning force fields reveal 500x defect concentration jump","Vibrational entropy drives 500-fold rise in CdTe defect density","Thermal fluctuations raise Te interstitial count 500-fold in CdTe"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The calculation rests on the machine-learned force field reproducing the interstitial's energy landscape accurately enough to resolve free-energy differences of a few hundredths of an electron-volt, and on the assumption that defect migration during thermodynamic integration contributes little to the free energy.","fun_headline_variants_meta":{"raw":{"variants":["Entropy boosts CdTe interstitial concentration 500-fold","Finite-temperature effects multiply CdTe defect count by 500","Machine learning force fields reveal 500x defect concentration jump","Vibrational entropy drives 500-fold rise in CdTe defect density","Thermal fluctuations raise Te interstitial count 500-fold in CdTe"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000194,"raw_usage":{"total_tokens":1384,"prompt_tokens":1008,"completion_tokens":376,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":624,"completion_tokens_details":{"reasoning_tokens":290}},"tokens_in":624,"tokens_out":376,"duration_ms":3705,"temperature":1.0,"reasoning_tokens":290,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:15:19.295161+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the fully anharmonic thermodynamic-integration calculation for $\\mathrm{Te_i^{+1}}$ at 840 K using direct DFT energies on the same 65-atom supercell and compare the formation free energy with the machine-learned result; a disagreement larger than about $0.02$ eV per supercell (the reported TI error) would undermine the claimed 0.5 eV thermal shift and the roughly 500-fold concentration increase. An independent check would be a temperature-dependent experimental measurement of interstitial concentration in CdTe, if the predicted slope of $\\ln[c]$ versus $1/T$ can be resolved.","supporting_citations":[{"cited_title":"Batatia, D","cited_arxiv_id":null,"evidence_quote":"Provides the machine-learned interatomic potential architecture used to run the molecular dynamics and thermodynamic-integration simulations."},{"cited_title":"Menon, Y","cited_arxiv_id":null,"evidence_quote":"Supplies the non-equilibrium thermodynamic-integration and reversible-scaling implementation used for the fully anharmonic free energies."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the harmonic and quasiharmonic phonon framework used to compute the vibrational free-energy contributions."}],"review_version":1}