{"id":"5f863074-90d3-4d03-b98d-3564b4f52d3b","arxiv_id":"2412.02959","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A first-order Taylor expansion of nonlocal kinetic energy functionals about a sum of free-atom densities lets OF-DFT compute the costly nonlocal potential once, giving large speedups with minimal accuracy loss.","lead":"A tight-binding-inspired trick makes advanced nonlocal kinetic energy functionals in orbital-free DFT run 10 to 100 times faster by computing the expensive part only once. This could make these accurate functionals practical for very large material simulations.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"First-order truncation of the nonlocal KEDF is not error-controlled, and the paper only tests density closeness for Si; metals and III-V compounds remain unexamined.","rationale":"The reader's conditional verdict identifies the same load-bearing assumption: the optimized density must stay close to ρ0 for the first-order expansion to be valid. My stress-test sharpens this into a quantitative requirement. The paper provides only one illustrative density comparison (CD-Si) and does not quantify the truncation error of Eq. (4) for any system. For metallic and ionic systems, the free-atom superposition reference is expected to be substantially farther from the self-consistent density, so the central accuracy claim is not established for the full benchmark set. However, this is a missing validation, not a demonstrated internal inconsistency: the derivation is straightforward, the implementation is plausible, and the reported energy comparisons are consistent with the approximation holding where tested. The proposed concrete test—computing the linearization remainder and density deviation across all benchmark systems—would directly settle whether the concern lands. If the remainder is small and density deviations remain modest, the conditional acceptance can be upgraded; if not, the claim of general accuracy must be narrowed. The reader's CONDITIONAL verdict remains the appropriate assessment pending this check.","tokens_in":9950,"tokens_out":5931,"duration_ms":67127,"concrete_test":"For each benchmark system (especially Li, Al, Ga, and the III-V compounds), compute the converged OF-DFT density ρ* using the original KEDF, then evaluate the linearization remainder R = T_NL[ρ*] − (T_NL[ρ0] + ∫ V_T_NL[ρ0](r)(ρ*(r)−ρ0(r)) d³r). Also report the relative density deviation ‖ρ*−ρ0‖/‖ρ0‖. If |R| is comparable to or larger than the 2 meV/atom energy-convergence target for any tested system, the first-order truncation is not negligible, and the observed energy agreement between TB-KEDF and original KEDF would not be a reliable indicator of general accuracy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that TB-KEDFs 'nearly exactly reproduce' the accuracy of the original nonlocal KEDFs rests on Eq. (4), a first-order functional expansion around ρ0. No bound is given for the neglected second-order remainder, and the only density-deviation evidence in the paper is CD-Si (Fig. 1(a)), where the initial deviation is already 17–18% in the bonding region. For Li, Mg, Al, Ga, and the III-V semiconductors—systems with metallic delocalization or charge transfer—the superposition-of-free-atoms reference ρ0 is much less likely to stay close to the optimized density, yet no density-deviation or linearization-error data are reported. The paper's own conclusion states that effectiveness depends on maintaining a small difference between reference and optimized densities, which is exactly the unverified condition for most benchmark systems.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a tight-binding-inspired framework (TB-KEDF) for orbital-free DFT with nonlocal kinetic energy density functionals. The nonlocal part T_NL[ρ] is approximated by a first-order functional expansion around a reference density ρ0 defined as the superposition of free-atom densities (Eqs. (3)-(5)), so that the expensive nonlocal kinetic energy and potential are evaluated only once before the density optimization begins. The method is implemented in the ATLAS code and tested with revHC and LDAK-MGPA on bulk phases of Li, Mg, Al, Ga, and Si, nine III-V zincblende semiconductors, and 120 random Mg50 and Si50 clusters. The reported results show 10-100x speedups over direct energy minimization, 5-10x speedups relative to OE-SCF for the tested Si supercells, accuracy comparable to the original KEDFs, and improved numerical stability for the cluster calculations. The authors explicitly acknowledge that the framework's effectiveness depends on maintaining a small difference between the reference and optimized electron densities.","tokens_in":10047,"tokens_out":4970,"duration_ms":50464,"significance":"If substantiated, this is a practically valuable contribution because the nonlocal potential evaluation is the dominant cost in OF-DFT with state-of-the-art nonlocal KEDFs. The derivation of Eq. (4) is parameter-free, the reference density is computed independently from single-atom KS-DFT rather than from the functional being approximated, and the benchmarks cover a range of bulk and finite systems with accuracy assessed against external KS-DFT results. The speedups and stability improvements are concrete and potentially enabling for large-scale OF-DFT. However, the central first-order expansion is uncontrolled, and the evidence that the reference density stays close to the optimized density is limited to a single system (CD-Si), so the generality of the 'nearly exactly reproduce' accuracy claim is not yet fully established.","major_comments":[{"comment":"The first-order functional expansion in Eq. (4) neglects the second-order remainder without any estimate or bound. The only density-deviation data presented are for CD-Si (initial reference-density deviations of 17.1% and 18.0% in the bonding region), yet the method is applied to Li, Mg, Al, Ga, and III-V semiconductors, where a superposition of free-atom densities is expected to deviate more from the self-consistent density. To support the claim that TB-KEDFs 'nearly exactly reproduce' the accuracy of the original KEDFs, please report for representative bulk systems (at least one metal and one III-V compound) the maximum |δρ/ρ0| at the converged density and the linearization error, for example |T_NL[ρ_OF] - (T_NL[ρ0] + ∫ V_T_NL[ρ0] δρ)| / T_NL[ρ_OF]. Without such diagnostics, the central approximation remains unverified outside a single semiconducting case.","section":"Section II, Eq. (4) and Fig. 1(a)"},{"comment":"The bulk-property comparison reports MAEs and MAPEs with respect to KS-DFT, showing that TB-KEDFs and original KEDFs have similar deviations from KS-DFT. However, because the original KEDFs themselves have substantial errors for some systems (e.g., bulk-modulus MAPEs exceeding 60% in Fig. 2(c)), similarity to KS-DFT does not directly quantify how closely the TB-KEDFs reproduce the functionals they are meant to reconstruct. Please add a direct comparison between TB-KEDF and original-KEDF predictions for E_R, V0, and B0 across all tested phases, for instance mean absolute differences or a parity plot. This would substantiate the wording 'nearly exactly reproduce the accuracy of the original KEDFs'.","section":"Section IV, Fig. 2"},{"comment":"The efficiency comparison reports wall times for revHC and LDAK-MGPA in Fig. 1(b), but Table I lists nonlocal and semilocal call counts only for revHC. Since LDAK-MGPA has a different and generally more expensive kernel, and the abstract and conclusions claim 'orders-of-magnitude' efficiency improvements for both functionals, the LDAK-MGPA call-count data should be presented in the main text rather than only in the Supplemental Material, or the efficiency claim should be restricted to the systems and functionals for which complete data are shown.","section":"Section IV, Table I and Fig. 1(b)"}],"minor_comments":[{"comment":"The horizontal-axis label appears garbled as '/glyph1197umber of atoms' in the manuscript text; it should read 'Number of atoms'.","section":"Fig. 1(b)"},{"comment":"The bulk-derived local pseudopotentials are cited only for Mg, Al, and Si (Ref. [45]); please specify which pseudopotentials were used for Li, Ga, and the III-V compounds, and whether the same pseudopotentials were used for the single-atom reference densities.","section":"Section III"},{"comment":"The claim of improved numerical stability is supported with convergence statistics for clusters (Fig. S1 and the average 12.6 steps for TB-revHC), but no convergence statistics are given for bulk systems; please either add such data or restrict the stability claim to the finite systems tested.","section":"Section IV"},{"comment":"There are several minor typographical issues, including 'Chac´ on' in Ref. [23], 'Fig.4' in the text before Fig. 4, and inconsistent spacing in '10 −5 eV/atom'; these should be corrected during production.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern is legitimate and lands on a load-bearing point: the first-order expansion is not error-controlled, and the only density-closeness evidence is CD-Si. That said, the paper's benchmarks are broad and the limitation is explicitly acknowledged, so the issue is fixable within the manuscript's scope by adding density-deviation and linearization-error diagnostics for metals and III-V compounds, plus a direct TB-versus-original accuracy comparison. I see no evidence of circularity or invented free parameters. The manuscript is a reasonable fit for the journal, but the 'nearly exactly reproduce' and 'orders-of-magnitude' claims should be matched with the requested quantitative support before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does something genuinely useful: it linearizes nonlocal KEDFs around a superposition of free-atom densities, turning the expensive nonlocal potential into a one-shot calculation. That is a real step beyond OE-SCF, which still re-evaluates the nonlocal potential at every SCF cycle. The derivation is clean, there are no fitted parameters, and the benchmarks cover a reasonable set of bulk phases, III-V semiconductors, and random clusters. The speedup numbers (10-100x versus direct minimization, 5-10x versus OE-SCF) are plausible given that the nonlocal call count drops to exactly one. I also like that TB-revHC converges where revHC itself fails; that is a practical win, not just a cosmetic one.\n\nNow the soft spots. The load-bearing assumption is that the optimized density stays close to the reference superposition. The paper only demonstrates that for CD-Si, and the deviations there are already 17-18% in the bonding region before optimization. For the metals and III-V compounds, there is no density-deviation or linearization-error data at all. The authors themselves admit in the conclusion that the framework's effectiveness depends on maintaining a small difference between reference and optimized densities, but they never verify that condition for most benchmarks. The fact that the final bulk properties are close to the original KEDFs is encouraging, but it is indirect evidence; the error in the functional expansion is not formally controlled, and the claim that the reconstructed functionals 'nearly exactly reproduce' the originals is too strong as stated.\n\nReproducibility is a secondary concern. The implementation is in ATLAS, but no code or data are provided, and the supplemental material is not available on arXiv. That matters more than usual because the speedup is an implementation-dependent claim and the accuracy comparison depends on the exact reference densities and convergence criteria.\n\nThis paper is for the OF-DFT community, specifically anyone using nonlocal KEDFs with density-dependent kernels who wants to push to larger systems. It deserves a serious referee, not because it is revolutionary, but because it is a solid, simple idea with clear practical value. The referee should ask for density-closeness checks on at least one metallic and one III-V system, and for code or detailed input data. Those are addressable issues. Worth engaging with.","headline":"A clean linearization of nonlocal KEDFs that delivers real speedups and stable convergence, but the load-bearing assumption of density closeness is verified for only one of the benchmark systems.","tokens_in":10607,"tokens_out":2245,"would_cite":true,"duration_ms":23587,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["71.15.Mb"],"model":"deepseek-v4-flash","headline":"By expanding nonlocal kinetic energy functionals around a superposition of atomic densities, the paper makes accurate orbital-free DFT calculations orders of magnitude faster at nearly unchanged accuracy.","keywords":["orbital-free density functional theory","kinetic energy density functional","tight-binding approximation","nonlocal KEDF","first-order functional expansion","revHC","LDAK-MGPA","numerical stability"],"falsifier":"Run the reconstructed and original KEDFs on a system where the optimized density is far from the superposition of free-atom densities, such as a strongly ionic interface, a charged defect, or a hydrogen-bonded solid, and compare total energies and forces. If the TB-KEDF deviates from the original KEDF by more than the claimed accuracy threshold, the first-order truncation is the cause; a quantitative version is to measure the density difference in the bonding region and correlate it with the energy error.","tokens_in":9737,"feed_emoji":"⚡","tokens_out":8376,"duration_ms":67807,"temperature":0.7,"pith_summary":"This paper claims that the most accurate family of orbital-free density functional theory (OF-DFT) kinetic energy functionals—the nonlocal ones with density-dependent kernels—can be rebuilt in a tight-binding style so that their expensive part is evaluated only once per calculation. The reconstruction expands the nonlocal kinetic energy around a reference density made by superimposing free-atom densities, keeping only the first-order term. If the claim holds, the dominant computational bottleneck of nonlocal KEDFs largely disappears, making these accurate functionals practical for large-scale simulations. The paper demonstrates the strategy on the revHC and LDAK-MGPA functionals across metals, semiconductors, and clusters, reporting comparable accuracy to the original functionals with up to two orders of magnitude less wall time.","feed_headline":"Computing nonlocal kinetic energy once speeds orbital-free DFT by 100x","feed_subtitle":"Accurate nonlocal functionals become practical for large-scale simulations with accuracy nearly unchanged.","key_machinery":"The load-bearing object is the tight-binding density expansion $\\rho(r) = \\rho_0(r) + \\delta\\rho(r)$, with $\\rho_0$ built as the sum of free-atom densities, combined with a first-order functional Taylor expansion of the nonlocal KEDF. This turns the double-integral nonlocal term, which normally must be recomputed at every self-consistent step, into a constant plus a single linear integral whose kernel is the nonlocal kinetic potential evaluated at $\\rho_0$. That potential is computed once during initialization, so the number of nonlocal potential calls during density optimization drops from hundreds to one.","core_discovery":"The central claim is that a first-order functional expansion of a nonlocal KEDF about the superposition of free-atom densities, $\\rho_0$, reconstructs the functional accurately enough for self-consistent OF-DFT, while reducing the nonlocal kinetic energy and potential to a one-shot initialization. Writing $\\rho = \\rho_0 + \\delta\\rho$, the nonlocal term $T_{NL}[\\rho]$ is replaced by $T_{NL}[\\rho_0] + \\int V^T_{NL}[\\rho_0](r)\\delta\\rho(r)d^3r$. Because both $T_{NL}[\\rho_0]$ and its functional derivative depend only on $\\rho_0$, they are computed once, and subsequent density optimization only evaluates the cheap Thomas-Fermi, von Weizsäcker, and linear-response terms. Benchmarking against the original revHC and LDAK-MGPA functionals, the reconstructed TB-KEDFs reproduce equilibrium energies, volumes, and bulk moduli to comparable accuracy and reproduce KS-DFT energy orderings, while requiring roughly 10 to 100 times less time than direct energy minimization and roughly 5 to 10 times less than an orbital-ensemble solver. The paper also finds the reconstructed functionals converge reliably for disordered clusters where the original revHC functional fails to converge.","pith_inferences":["Editorial inference: the same one-shot expansion could be combined with higher-order corrections in $\\delta\\rho$, analogous to many-body tight-binding, to extend accuracy to systems with stronger charge redistribution at modest extra cost.","Editorial inference: the choice of $\\rho_0$ is the transferability knob; using nonlocal-pseudopotential atomic densities, which the paper flags as future work, should widen the range of elements and bond types for which the first-order truncation is valid.","Editorial inference: the method's promise suggests a natural test on heterogeneous interfaces or charged defects, where density deviations from superposition are large and the claimed accuracy should degrade measurably."],"forward_implications":["OF-DFT with accurate nonlocal KEDFs becomes feasible for systems of tens of thousands of atoms, where the original functionals are too costly.","The reconstruction inherits the accuracy of the parent KEDF when the optimized density stays near $\\rho_0$, so future improvements to nonlocal kernels can be adopted without multiplying the cost.","The reported stability gain means functionals that are accurate but hard to converge, such as revHC on disordered clusters, become usable in practical structure searches.","Because only one nonlocal potential evaluation is needed, the bottleneck shifts from the KEDF evaluation to density optimization and electrostatics."],"supporting_citations":[{"why":"supplies the tight-binding ansatz of expanding about a reference density that the reconstruction adapts.","marker":"[37]"},{"why":"defines the Huang-Carter nonlocal KEDF whose revised version is one of the two benchmarked functionals.","marker":"[28]"},{"why":"provides the revised HC functional (revHC) whose accuracy and convergence are tested in the TB scheme.","marker":"[32]"},{"why":"gives the functional-integration construction underlying the LDAK family of nonlocal KEDFs.","marker":"[29]"},{"why":"defines the LDAK-MGPA nonlocal KEDF and its local-density-approximation kernels used in the benchmarks.","marker":"[34]"},{"why":"is the orbital-ensemble solver baseline that the TB-KEDF speedups are compared against.","marker":"[36]"},{"why":"is the real-space orbital-free DFT code where the framework is implemented for all benchmark calculations.","marker":"[38]"}],"fun_headline_variants":["One-time nonlocal kinetic energy gives 100x faster OF-DFT","Nonlocal KEDF computed once: 100x speedup for orbital-free DFT","Tight-binding trick cuts nonlocal KEDF cost by 100x","First-order expansion makes nonlocal KEDF 100x cheaper","Reusing free-atom density reconstructs KEDF with 100x speedup"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole speedup rests on the assumption that the true electron density stays close to the superposition of free-atom densities, so that keeping only the first term in the expansion is accurate; the paper states this dependence explicitly.","fun_headline_variants_meta":{"raw":{"variants":["One-time nonlocal kinetic energy gives 100x faster OF-DFT","Nonlocal KEDF computed once: 100x speedup for orbital-free DFT","Tight-binding trick cuts nonlocal KEDF cost by 100x","First-order expansion makes nonlocal KEDF 100x cheaper","Reusing free-atom density reconstructs KEDF with 100x speedup"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00023,"raw_usage":{"total_tokens":1549,"prompt_tokens":1079,"completion_tokens":470,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":695,"completion_tokens_details":{"reasoning_tokens":367}},"tokens_in":695,"tokens_out":470,"duration_ms":5264,"temperature":1.0,"reasoning_tokens":367,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T22:54:18.343784+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the reconstructed and original KEDFs on a system where the optimized density is far from the superposition of free-atom densities, such as a strongly ionic interface, a charged defect, or a hydrogen-bonded solid, and compare total energies and forces. If the TB-KEDF deviates from the original KEDF by more than the claimed accuracy threshold, the first-order truncation is the cause; a quantitative version is to measure the density difference in the bonding region and correlate it with the energy error.","supporting_citations":[{"cited_title":"Huang and E","cited_arxiv_id":null,"evidence_quote":"defines the Huang-Carter nonlocal KEDF whose revised version is one of the two benchmarked functionals."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the revised HC functional (revHC) whose accuracy and convergence are tested in the TB scheme."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the functional-integration construction underlying the LDAK family of nonlocal KEDFs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"defines the LDAK-MGPA nonlocal KEDF and its local-density-approximation kernels used in the benchmarks."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"is the orbital-ensemble solver baseline that the TB-KEDF speedups are compared against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"is the real-space orbital-free DFT code where the framework is implemented for all benchmark calculations."}],"review_version":1}