{"id":"dfc8051e-8f98-41a7-aa57-3c7dc271a4d0","arxiv_id":"2505.22056","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A symmetry-enforced, quasi-automated pipeline generates second-principles electronic models that reproduce DFT band structures for SrTiO3 and LiF with two to three orders of magnitude fewer DFT calculations.","lead":"This paper presents a nearly automatic recipe for building fast, accurate electronic models of crystals from density functional theory calculations, and it tests the recipe on SrTiO3 and LiF. The recipe uses crystal symmetry to cut the number of costly calculations by a factor of hundreds to thousands, while keeping simulated band structures close to the reference DFT results.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Charge-state fidelity is asserted but never tested: fitted U/I electron-electron parameters are validated only on neutral random geometries, leaving the abstract's 'various charge states' claim unsupported.","rationale":"Reader's weakest_assumption is exactly U/I transferability; my check is designed to settle it. I agree with that identification. I examined possible internal inconsistencies in Sec. III B–C (ASR enforcement and quadratic couplings) and found the finite-difference construction self-consistent once the displacement parameter is tracked; the symmetrization relations Eqs. (23)–(27) are standard. The main weakness is empirical: the abstract advertises charge-state fidelity, but the validation protocol never changes charge in the test set. The random-geometry tests are held out only with respect to geometry; the same fitted U/I are never confronted with a held-out doping level or with doping combined with distortion. Because the polaron demonstration in Sec. VI A was inconclusive (no localization in either model or DFT), it cannot substitute as a charge-transfer test. The paper's own statements (Intro: 'does not include a comprehensive validation in more complex or application-driven scenarios'; Sec. VII: direct computation of U/I 'not yet available') acknowledge the gap. The secondary issue that test geometries are reused to select cutoffs (δrel, δf, δg) further lowers confidence in the reported error magnitudes, but the primary missing experiment is held-out charge states. Verdict stays CONDITIONAL: the methods and neutral-geometry evidence are solid enough for a conditional accept, but the central charge-state claim should be either demonstrated or softened.","tokens_in":28904,"tokens_out":11769,"duration_ms":139173,"concrete_test":"Perform a held-out charge-state validation for SrTiO3 using the existing 40-atom supercell and modelmaker pipeline. Compute DFT and second-principles Hamiltonians for q=+0.4, −0.4, +0.5, −0.5 e (spin-restricted and spin-polarized with |M|=|q|) at the RAG, and for ten random geometries with d=0.17 Å at q=+0.3 and q=−0.3 e (or ±0.4 e) not used in fitting the U/I parameters. Evaluate the goalfunction Θ of Eq. (31) for this set and compare with the neutral random-geometry results (Figs. 5–6). If per-configuration errors remain comparable and band deviations stay much smaller than typical band separations, the charge-state claim holds. If Θ or band errors increase materially (e.g., >2× the neutral-geometry value or >0.1 eV band shifts), U/I transferability fails and the abstract/claims should be restricted or the fitting expanded.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim—models reproduce DFT 'across various atomic configurations and charge states'—splits into two parts. The geometry part is genuinely tested: Sec. IV D defines held-out random-displacement test sets and Figs. 5–7 and 11–12 report small band errors up to d=0.17 Å. The charge-state part is not. The U/I electron-electron parameters are fitted to the Sec. IV C 2 training set (charges −0.3 to +0.3 e in 0.1 e steps, with spin-polarized variants; for LiF also neutral magnetizations 0.5–2.0 μB). The only reported validation uses random geometries and alternative phases, with no held-out doping level, no held-out magnetization, and no doped distorted geometry. Since U/I are the entire mechanism by which the model responds to added charge, spin, or electron-hole occupancy, polarons and excitons depend on their transferability. The one application that would exercise this response (electron doping in SrTiO3) did not localize charge, and the authors attribute the absence to LDA rather than comparing model vs DFT for that doped configuration. Sec. VII also states that direct computation of U/I is not yet available. Hence the abstract's charge-state fidelity claim rests on an untested transferability assumption. This is a validation gap, not an internal inconsistency: the ASR/finite-difference treatment and symmetrization appear internally consistent, and the neutral-geometry validation is a reasonable held-out test.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents an improved, largely automated workflow for constructing the electronic part of second-principles DFT (SPDFT) models. The authors generalize the reference electron density to allow non-diagonal density matrices, reformulate linear and quadratic electron-lattice couplings using absolute atomic displacements with acoustic-sum-rule constraints, enforce space-group symmetry to reduce the number of required DFT calculations, and combine direct computation of one-electron and electron-lattice parameters with fitting of the electron-electron parameters U and I to a standardized training set of doped and spin-polarized DFT Hamiltonians. The workflow is applied to SrTiO3 and LiF; the authors report low goal-function values on random-displacement test sets, band-structure error bars at d = 0.17 Å, and a check against an alternative SrTiO3 phase. The central claim is that the resulting models reproduce DFT reference data with high fidelity across various atomic configurations and charge states.","tokens_in":29241,"tokens_out":11031,"duration_ms":111797,"significance":"If the geometry-transferability results hold, this is a substantial practical advance: the symmetry-based reduction in the number of first-principles calculations (e.g., from 57,840 to 15 for the SrTiO3 electron-lattice data at δrel = 2.0 Å) makes SPDFT model construction feasible for a wider class of materials, and the explicit goal-function metric provides a transparent quality assessment. The validation on held-out random geometries, the test on the antiferrodistortive phase of SrTiO3, and the careful documentation of cutoff choices are genuine strengths. The main weakness is that the charge-state half of the abstract's claim is not validated by any held-out test; because U and I are fitted parameters, their transferability to unseen doping levels, spin polarizations, and combined distortion-doping configurations is load-bearing for the promised polaron/exciton applications.","major_comments":[{"comment":"The abstract claims high fidelity 'across various atomic configurations and charge states,' but the charge-state part is not tested. The U and I parameters are fitted on the Sec. IV C 2 training set (charges from -0.3 to +0.3 e in 0.1 e steps, with fully spin-polarized variants, plus neutral magnetizations for LiF), while the Sec. IV D validation consists of random neutral geometries and alternative phases. No held-out doping level, held-out magnetization, or doped distorted geometry is reported. Since U and I carry the entire density/spin response of the model, the claim that the models reproduce DFT for arbitrary charge states is unsupported. Please add a cross-validation where at least one doping level or magnetization is excluded from the fit and the model error on that held-out configuration is reported, or compare model vs DFT for a doped distorted geometry (e.g., the electron-doped SrTiO3 configuration discussed in Sec. VI A).","section":"Sec. IV C 2, Sec. IV D, Abstract"},{"comment":"The same random-displacement test set appears to be used both for selecting the cutoffs δrel, δf, and δg and for reporting the final model errors. Fig. 5(a) is used to choose δrel, Fig. 5(b) to choose δf and δg, and Fig. 6 then reports the errors of the selected model on the same kind of test set; the analogous procedure is used for LiF in Figs. 11 and 12. This makes the reported errors an in-sample estimate of model performance after hyperparameter tuning. Please either reserve a separate final test set not used in any cutoff selection, or report the sensitivity of the final error to the cutoff choices so that the reader can assess the potential optimism bias.","section":"Sec. VI A, Figs. 5-6; Sec. VI B, Figs. 11-12"},{"comment":"The only application that would exercise the fitted U/I response outside the training set is the attempted electron doping of SrTiO3, but the outcome is reported only as a failure to localize charge, attributed to the LDA functional. No comparison of the second-principles model against DFT for that doped configuration is provided, so this episode neither validates nor invalidates the charge-state transferability of U/I. Please clarify whether such a comparison was performed, or state explicitly that it was not and that the charge-state transferability remains to be established.","section":"Sec. VI A, polaron paragraph"}],"minor_comments":[{"comment":"The captions call Θ the 'average error per calculation,' but the text (e.g., Sec. VI B) describes Θ as a sum over Hamiltonian matrix-element deviations (≈1 eV2 summed over 11,964 terms). Please clarify the normalization used in the plots.","section":"Figs. 5 and 11 captions"},{"comment":"The final sentence, 'More details will be given in a forthcoming simulation,' is an incomplete placeholder and should be replaced by a reference or removed.","section":"Sec. III D"},{"comment":"The statement that the LiF model has 'already proven capable' of describing optical properties such as excitons is not demonstrated in this manuscript; either show the data or refer explicitly to the forthcoming publication.","section":"Sec. I and Sec. VII"},{"comment":"Equation (29) appears to contain a typographical error: the first equality writes h^{DFT}_{a'b'}(u_{λ'}) on both sides, while the intended relation should connect the Hamiltonian at the symmetry-related geometry to the Hamiltonian at u_λ. Please correct the indices.","section":"Eq. (29)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript would be strengthened by a data/code availability statement, given the emphasis on a quasi-automated workflow and the existence of the modelmaker and scale-up codes. The charge-state validation gap is the main technical reason for major revision; the geometry validation is credible and the formalism appears internally consistent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The genuinely new thing is the pipeline: a quasi-automated construction of SPDFT electronic Hamiltonians with symmetry-constrained parameter extraction, independent wannierization of valence and conduction manifolds, a non-diagonal reference density matrix, and the ASR-preserving linear electron-lattice representation. The cost reduction is real - from 57,840 single-point DFT runs down to 15-209 for SrTiO3, and a similar factor for LiF. That matters. The band-structure comparisons for neutral random geometries up to 0.17 Angstrom displacements are convincing: error bars are shown, a goalfunction is used consistently, and the errors are genuinely small relative to band separations.\n\nWhere it goes soft is the abstract's 'various charge states' claim. The U/I electron-electron parameters are fitted to doped and spin-polarized training configurations, but the only held-out validation uses neutral random geometries and an alternative phase. There is no held-out doping level, no held-out magnetization, and no doped distorted geometry. Since the model's response to added charge, spin, or electron-hole occupancy is carried entirely by these fitted U/I integrals, their transferability to charge states outside the training set is asserted, not demonstrated. That is a validation gap - not an internal inconsistency. The symmetrization and finite-difference derivations are clear, and the neutral-geometry validation is a legitimate held-out test.\n\nOne milder issue: the test set used to tune cutoffs delta_rel, delta_f, delta_g is the same set used to report final errors, so the headline error bars are somewhat optimistic. The authors are also honest that their polaron attempt did not localize charge, that they attribute it to LDA rather than model error, and that direct computation of U/I is not yet available. That honesty is to their credit, but it means the advertised physics-level applications rest on unproven transferability. And the code isn't released, which limits reproducibility checks.\n\nFor a reader building SPDFT models, this is useful and worth engaging with. For a general condensed-matter audience, the value is narrower. I'd send it to peer review: the methodology is new, the neutral-geometry validation is solid, and the charge-state overclaim is fixable in revision rather than fatal. A serious referee should ask for either a held-out charge-state test or a toned-down abstract.","headline":"A genuinely useful automation of SPDFT electronic-model construction with solid neutral-geometry validation, but the abstract's charge-state fidelity claim is tested nowhere.","tokens_in":746,"tokens_out":822,"would_cite":true,"duration_ms":28188,"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":"A quasi-automated, symmetry-enforced pipeline builds SPDFT electron models that reproduce DFT band structures on geometries outside the training set, for SrTiO3 and LiF.","keywords":["second-principles DFT","Wannier functions","electron-lattice coupling","electron-electron interactions","space group symmetry","SrTiO3","LiF","polarons and excitons"],"falsifier":"Train a SrTiO3 model using only the paper's stated doping range, −0.3 to +0.3 e with spin polarizations up to 0.3 e, then compute the DFT Hamiltonian for a held-out +0.5 e doped supercell and compare the model's valence and t2g bands near the gap. If the error approaches the LDA gap of about 1.7 eV, the fitted $U$ and $I$ integrals have failed to transfer to an unseen charge state, and the central claim of high-fidelity reproduction across charge states is falsified.","tokens_in":28682,"feed_emoji":"⚛️","tokens_out":8653,"duration_ms":80657,"temperature":0.7,"pith_summary":"This paper attempts to turn second-principles DFT (SPDFT) model construction into a quasi-automated, reproducible procedure: a user supplies a reference crystal structure and a handful of physical thresholds, and the pipeline returns an electronic Hamiltonian that tracks DFT band structures at a far lower computational cost. The key division of labor is that one-electron hoppings and electron-lattice couplings are computed directly from first-principles data, while the effective electron-electron response parameters are fitted to a small, standardized training set of doped and spin-polarized supercells. Space-group symmetry is enforced on every tensor, which removes spurious parameter differences and reduces the number of required DFT calculations by orders of magnitude. If the method performs as claimed, large-scale simulations of polarons and excitons in materials such as SrTiO3 and LiF become practical.","feed_headline":"Symmetry cuts the cost of building DFT-accurate electron models","feed_subtitle":"For SrTiO3 and LiF, the pipeline reproduces band structures on distorted geometries, enabling polaron and exciton studies.","key_machinery":"The central object is the SPDFT energy functional of Eq. (9), written in a basis of localized Wannier orbitals (orbitals extracted from the DFT wavefunctions): the one-electron hopping matrix $\\gamma_{ab}$, the linear and quadratic electron-lattice couplings $\\mathbf{f}_{ab,\\lambda}$ and $\\overleftrightarrow{g}_{ab,\\lambda\\upsilon}$, and the effective electron-electron integrals $U_{ab,a'b'}$ and $I_{ab,a'b'}$. Space-group operations of the reference geometry are applied to every tensor, grouping symmetry-equivalent parameters and letting one DFT calculation stand for all symmetry-equivalent distortions; the resulting symmetry constraints are enforced with Lagrange multipliers during the electron-electron fit. The electron-electron integrals are obtained by solving the linear fitting equations (32)-(33) against Hamiltonians from the training set, and their influence is screened by a spatial cutoff $\\delta_{ree}$ and a goal-function threshold $\\delta_\\Theta$ that retain only the integrals that matter. This machinery converts the old fitting problem into a mostly computed, symmetry-consistent, and systematically refinable model.","core_discovery":"The paper claims that a largely automated protocol can construct SPDFT electronic models that reproduce DFT reference data across atomic configurations and charge states. The protocol works in a basis of localized Wannier orbitals, separates valence and conduction manifolds during wannierization so the reference density stays well defined, and enforces the space group of the reference structure on all Hamiltonian, electron-lattice, and electron-electron parameters. One-electron and electron-lattice terms are computed directly (hoppings from wannierized Hamiltonians; couplings by finite differences), while the electron-electron integrals $U_{ab,a'b'}$ and $I_{ab,a'b'}$ — effective responses of the Hamiltonian to changes in density and spin — are fitted to a standardized training set that includes hole/electron doping and spin-polarized configurations. For SrTiO3 and LiF, the resulting models keep deviations in the valence and lower conduction bands small even for random atomic displacements up to 0.17 Å, with errors much smaller than typical band separations, and the training set shrinks to a few hundred single-point calculations.","pith_inferences":["The paper tests random geometries within the same charge-state range as the training set, so a natural extension is to hold out entire charge states or spin configurations; the fitted $U$ and $I$ parameters either transfer, which would confirm the claim, or they do not, which would set the method's actual domain of validity.","Because the fitted electron-electron integrals are effective density responses rather than bare Coulomb integrals, their values could be read across materials as a screening descriptor; the paper does not discuss this, but the model files it produces would permit such a comparison.","The LDA-based SrTiO3 model did not localize electron polarons, and the paper attributes this to self-interaction error; a direct test of the pipeline is to rebuild it from hybrid-functional data, where localization should emerge if the fitted parameters carry the needed physics.","For metals, where the reference density matrix must be non-diagonal and geometry-dependent, the subtraction procedure that separates electron-electron from electron-lattice effects is more delicate and remains largely untested; applying the pipeline to a simple metal would reveal whether quasi-automation survives metallic occupations."],"forward_implications":["Materials with strong electron-lattice coupling, such as transition-metal perovskites, can be simulated in supercells large enough for polaron formation and hopping without a self-consistent DFT loop at every step.","The real-time time-dependent SPDFT extension can compute optical spectra with explicit electron-hole interactions, making exciton binding and dynamics in wide-gap insulators like LiF directly accessible.","Because symmetry turns tens of thousands of candidate DFT calculations into a few hundred, systematic model generation for new materials becomes affordable, and the models become comparable across different systems and users.","The explicit hierarchy of cutoffs ($\\delta_{rh}$, $\\delta_{rel}$, $\\delta_f$, $\\delta_g$, $\\delta_{ree}$, $\\delta_\\Theta$) gives a principled route to refinement: tighten a cutoff and the model improves in a controlled way."],"supporting_citations":[{"why":"Defines the SPDFT energy expansion and the reference-electron-density concept that this work systematizes.","marker":"[23]"},{"why":"Supplies the localized Wannier-function basis in which one-electron and interaction parameters are expressed.","marker":"[30]"},{"why":"Provides the second-principles lattice-model Taylor expansion around the reference geometry that the electronic model is designed to complement.","marker":"[31]"},{"why":"Introduces the real-time time-dependent SPDFT formalism that the generated electronic models would feed for optical spectra.","marker":"[36]"},{"why":"Provides the DFT engine in which the training calculations and the multi-manifold wannierization are performed.","marker":"[37]"},{"why":"Provides the Wannier-interpolation code from which the one-electron Hamiltonian and position matrix elements are obtained.","marker":"[38, 39]"},{"why":"Supplies the systematic stepwise-regression fitting procedure adapted here for the electron-electron parameters.","marker":"[45]"}],"fun_headline_variants":["Automated symmetry-aware path to DFT-accurate electron models","Space-group symmetry cuts cost, keeps DFT accuracy","DFT-accurate electron models for oxides and insulators","Systematic SPDFT models: symmetry-reduced, DFT-faithful"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the electron-electron interaction parameters fitted to a small set of doped and spin-polarized supercell calculations transfer unchanged to other geometries, larger supercells, and charge states outside that training set.","fun_headline_variants_meta":{"raw":{"variants":["Automated symmetry-aware path to DFT-accurate electron models","Space-group symmetry cuts cost, keeps DFT accuracy","DFT-accurate electron models for oxides and insulators","Systematic SPDFT models: symmetry-reduced, DFT-faithful"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000254,"raw_usage":{"total_tokens":1578,"prompt_tokens":963,"completion_tokens":615,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":579,"completion_tokens_details":{"reasoning_tokens":544}},"tokens_in":579,"tokens_out":615,"duration_ms":5953,"temperature":1.0,"reasoning_tokens":544,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:16:44.393530+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Train a SrTiO3 model using only the paper's stated doping range, −0.3 to +0.3 e with spin polarizations up to 0.3 e, then compute the DFT Hamiltonian for a held-out +0.5 e doped supercell and compare the model's valence and t2g bands near the gap. If the error approaches the LDA gap of about 1.7 eV, the fitted $U$ and $I$ integrals have failed to transfer to an unseen charge state, and the central claim of high-fidelity reproduction across charge states is falsified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the SPDFT energy expansion and the reference-electron-density concept that this work systematizes."},{"cited_title":"com/mzjb/DeepH-pack, accessed: 1 April 2025","cited_arxiv_id":null,"evidence_quote":"Supplies the localized Wannier-function basis in which one-electron and interaction parameters are expressed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the second-principles lattice-model Taylor expansion around the reference geometry that the electronic model is designed to complement."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the real-time time-dependent SPDFT formalism that the generated electronic models would feed for optical spectra."},{"cited_title":"Garc ´ ıa, N","cited_arxiv_id":null,"evidence_quote":"Supplies the systematic stepwise-regression fitting procedure adapted here for the electron-electron parameters."}],"review_version":1}