{"id":"f0e3d552-0017-4940-80dd-edb3b7b6e34e","arxiv_id":"2607.22916","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A developer-authored review of CP2K's methods for dynamics, transport, and spectroscopy, consolidating static-response, finite-temperature, real-time, and open-boundary capabilities with benchmark validations.","lead":"CP2K's developers review their own open-source simulation package, cataloguing the methods that turn electronic-structure calculations into molecular dynamics, transport, and spectra — from static response and perturbation theory to real-time, finite-temperature, and open-boundary simulations. A generalist might read it as a current map of what can be computed for disordered and condensed systems without leaving one software environment.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Gamma-point-only GW supercell reconstruction relies on unverified real-space decay for metals/narrow-gap systems; the 10k-atom claim is not supported.","rationale":"The reader's weakest_assumption identifies the same concern: the Gamma-point-only GW supercell reconstruction relies on real-space decay that is only validated for favorable gapped 2D systems. This is the most load-bearing issue because it directly affects an advertised headline capability ('up to 10,000 atoms') and the claim that the full k-dependence can be reconstructed at negligible cost. If the reconstruction fails for metals or narrow-gap systems, the GW large-supercell pillar of the review materially weakens. The concern is concrete and testable, and it does not collapse the central integration claim but does support a CONDITIONAL verdict. The alpha-COHSEX fitting issue (W12 inside the training set) is a secondary overfitting concern, but it is confined to one method, whereas the GW reconstruction limits an entire advertised scale. No other concern appears more central: the modular integration claim is well supported by examples, and the paper is transparent about several other limitations. Therefore I agree with the reader's weakest assumption and the CONDITIONAL verdict stands.","tokens_in":51017,"tokens_out":3506,"duration_ms":39568,"concrete_test":"Run the Gamma-point-only supercell GW with MIC on a small-gap 2D system (e.g., bilayer graphene or a TMD with band gap <0.5 eV) and on a 3D narrow-gap semiconductor (e.g., InAs) at supercells of 4×4, 8×8, and 12×12 primitive cells. Compare G0W0 quasiparticle gaps against full-k-point GW or plane-wave reference results. If the error does not systematically decrease toward zero with increasing supercell size, the MIC reconstruction is not valid beyond gapped 2D materials.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline GW capability for large supercells (Section VI E) uses the minimum-image convention (MIC) to reconstruct the full k-dependence of chi, W, and Sigma from Gamma-point data. This is exact only when chi and Sigma decay before periodic images overlap. The paper validates this solely on 8x8 TMD cells (~20 Å lateral, gapped 2D semiconductors) with errors below 20 meV. For metals, narrow-gap semiconductors, or 3D supercells, the real-space decay is algebraic or slow, so the MIC reconstruction becomes uncontrolled. The advertised 'possibly up to 10,000 atoms' is therefore not established for the general materials classes that the review claims to cover. This is a load-bearing correctness risk because it underpins a flagship capability and is presented with an 'exact in the large-cell limit' statement that is only literally true under exponential-decay assumptions. The paper does not flag this limitation or provide evidence for systems beyond gapped 2D semiconductors.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This is a comprehensive code review of CP2K covering the methods that connect electronic structure to dynamics, transport, and spectroscopic response. The paper is organized around the perturbation–response idea: static response calculations (DFPT, TDDFT, RT-TDDFT, GW, GW-BSE, X-ray, NMR/EPR, etc.) are embedded in finite-temperature sampling, AIMD/PIMD, time-correlation functions, or nonequilibrium propagation, and the same electronic-structure representation is used for Kubo transport, Hairy Probes, and DFT+NEGF. The manuscript provides detailed equations, input-keyword guidance, benchmark data against external codes (WIEN2k, BerkeleyGW, ORCA, GW5000, HAB databases), and a supplement with implementation-level working equations. The central claim is that CP2K offers a uniquely integrated, modular platform that carries a user from ground-state DFT through response and transport to spectra that include thermal disorder, anharmonicity, and nuclear quantum effects.","tokens_in":51111,"tokens_out":5245,"duration_ms":69860,"significance":"If the claims hold, this is a valuable reference for the community because it demonstrates—with concrete benchmarks—that one open-source package can unify quantum chemistry with statistical mechanics and transport. The paper is particularly strong where it anchors accuracy to third-party codes: spin–orbit splittings against WIEN2k (MAE below 20 meV), GW band structures against BerkeleyGW (within about 50 meV), and excitation-energy basis-set convergence against GW5000. The modular architecture and the explicit tables of capabilities and finite-temperature strategies are genuinely useful. The main risk is that some headline scalability claims, especially the Γ-point-only supercell GW, rest on a narrow validation set and are stated more broadly than the evidence supports. This is a correctness concern for a flagship capability, not merely a presentation issue.","major_comments":[{"comment":"The Γ-point-only supercell GW implementation is advertised as being able to reach 'several thousand atoms, and possibly up to 10,000 atoms,' and the minimum-image reconstruction of χ and Σ is described as 'becoming exact in the limit of a large unit cell.' The only validation shown is for gapped 2D semiconductors: 8×8 TMD cells (about 20 Å lateral, convergence below 20 meV) and the 984-atom MoSe2/WS2 twisted bilayer. For metals, narrow-gap semiconductors, or 3D supercells, the real-space decay of G, χ, and Σ is algebraic or slow, so the microscopic reconstruction error is not controlled by the large-cell limit in any practical sense. This is load-bearing because the large-supercell GW capability and its accuracy at the advertised scale are part of the paper's central claim. Please either provide validation for a metallic or narrow-gap system and for a 3D supercell, or explicitly restrict","section":"Section VI E"},{"comment":"The Kubo conductivity implemented in CP2K is evaluated on a finite supercell with a phenomenological dissipation parameter η. The paper states that this yields 'equilibrium transport coefficients,' but it does not discuss how results depend on η or how a meaningful η→0 limit should be taken (e.g., via system-size extrapolation or a Drude–Lorentzian analysis of the finite-size spectrum). For metallic or low-carrier-density systems, the computed conductivity can vary strongly with η, and without a convergence protocol the label 'transport coefficient' is ambiguous. The implementation may be perfectly usable in practice, but the review should state the recommended procedure and the accuracy envelope, or explicitly flag this as a model-dependent estimate requiring extrapolation.","section":"Section IX D, Eq. (79)"},{"comment":"The α-COHSEX scheme uses an empirical scaling factor α that is fitted to the W12 water cluster and then transferred to other clusters. The paper does disclose this ('empirical scaling factor'), but the presentation may leave the impression that the resulting 21 meV MAE for electron affinities is a first-principles accuracy statement. Since α is a free parameter, the claim should be framed as a test of transferability rather than as an ab initio benchmark. Please add a sentence quantifying the sensitivity to α and clarifying that st-COHSEX is the parameter-free version, whereas α-COHSEX is a one-parameter model.","section":"Section VI F, Eq. (58)"}],"minor_comments":[{"comment":"The review relies heavily on the same author group's publications for method validation. While this is common in code reviews, the independence of the evidence would be strengthened by explicitly marking which benchmarks are third-party and which are author-generated. Some figures (e.g., Figs. 11, 12) are already labeled 'unpublished results by the authors'; please make the same distinction consistently for all benchmark claims.","section":"Throughout"},{"comment":"The phrase 'possibly up to 10,000 atoms' should be accompanied by a note that tractability does not imply accuracy. Even for tractable supercells, the minimum-image reconstruction error must be separately converged for each material class.","section":"Section VI E"},{"comment":"Table I lists BSE optical absorption among CP2K capabilities, while Section VII A states that linear-response GW-BSE is currently implemented only for finite systems. Please add a qualifier in the table row (e.g., 'molecular systems') so that the table does not imply periodic BSE is available.","section":"Table I / Section VII A"},{"comment":"The term 'Hairy Probes' is introduced as a CP2K-specific device. It may be helpful to note explicitly that this is a method name unique to the present implementation and to define the physical content at first use: a set of weakly coupled reservoir-like orbitals that impose local Fermi-level control. This would improve readability for readers outside the CP2K developer community.","section":"Section IX E"}],"recommendation":"major_revision","confidential_remarks":"The paper is a broad, useful code review with clear strengths: external benchmarks, detailed equation sets, and modular organization. The main technical concern is the overloading of the large-supercell GW claim: the minimum-image reconstruction is validated only on gapped 2D systems, yet the text suggests general applicability up to 10,000 atoms. This is fixable by adding explicit scope restrictions or additional benchmarks. There is also a moderate circularity issue common to developer-written code reviews: many capability claims cite the same author group's papers, and third-party validation is concentrated in a few benchmark sets. I would not reject on these grounds, but the revision should tighten the GW supercell claims and add appropriate caveats."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful thing here is the map: Tables I and II connect observables (IR, Raman, XAS, EPR, transport) to the specific CP2K modules and to whether you need static response, MD snapshots, time correlations, or real-time propagation. For someone new to CP2K or choosing a method for finite-temperature spectra, that is genuinely valuable. The augmented MOLOPT basis set, benchmarked against GW5000, is a concrete new deliverable, not just a survey. The st-COHSEX coupling benchmark on HAB79, with no empirical scaling, is also a clean result. And the paper is unusually transparent about boundaries: NMR is not for conductors, RT-BSE is molecular only, st-COHSEX overestimates absolute IPs/EAs. That honesty counts.\n\nThe soft spots are real but localized. The alpha-COHSEX claim needs a correction: the single scaling parameter is fitted to W12, and the reported 21 meV MAE is over a cluster set that includes W12. That is an in-sample error. The paper should separate the fitting point from the test set and report the leave-one-out or out-of-sample number. This does not break the method's usefulness, but it overstates transferability.\n\nThe bigger concern is the Gamma-point-only GW supercell machinery. The paper says the minimum-image reconstruction becomes exact in the large-cell limit and validates it on 8x8 TMD cells (about 20 Å, gapped 2D semiconductors). For metals, narrow-gap systems, or 3D cells, the real-space decay of chi and Sigma is algebraic or slow, so the reconstruction is uncontrolled. The paper does not flag this, and the 'possibly up to 10 000 atoms' claim is not supported for those material classes. That is a load-bearing limitation for a flagship capability, but it is confined to that section; the rest of the review does not depend on it.\n\nThe paper is self-citation heavy, but that is normal for a developer code review, and there are external anchors (WIEN2k, BerkeleyGW, ORCA, NEVPT2). The 'invented Hairy Probes' flag from the reader is wrong: Hairy Probes is a published method (Refs. 59, 271, 272), not a product of the review.\n\nWho benefits: CP2K users planning finite-temperature spectroscopy or transport calculations, method developers looking for a gap map, and students wanting an entry point. The paper deserves a serious referee—it is a major public resource—but the referee should require the fit reporting to be fixed and the GW supercell claims to be scoped to systems where the decay assumption is checked. I would send it out.","headline":"A solid, useful developer-side review that maps CP2K's finite-temperature spectroscopy and transport methods; two real soft spots—the in-sample alpha-COHSEX fit and the unverified GW supercell decay assumption—keep it from being as definitive as it claims.","tokens_in":768,"tokens_out":1006,"would_cite":true,"duration_ms":36093,"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":"CP2K's modular perturbation–response machinery lets a single electronic-structure representation feed static response, thermal sampling, real-time propagation, and open-boundary transport alike.","keywords":["CP2K","Gaussian and plane-wave DFT","DFPT","TDDFT","GW-BSE","ab initio molecular dynamics","Kubo transport","finite-temperature spectroscopy"],"falsifier":"Compute GW quasiparticle band gaps of a metallic or narrow-gap three-dimensional supercell (for example, doped silicon or a small-gap semiconductor) at increasing supercell sizes, and compare the minimum-image reconstruction against a full k-point calculation; if the reconstructed gap does not converge to the full-k result before periodic images overlap, the central scalability claim for large-supercell GW fails.","tokens_in":50746,"feed_emoji":"⚛️","tokens_out":4882,"duration_ms":54880,"temperature":0.7,"pith_summary":"This review argues that CP2K has reached the point where one open-source electronic-structure platform can carry a calculation from a static, zero-temperature response property to an experimentally comparable observable that includes thermal motion, anharmonicity, and even explicit bias. The organizing idea is the perturbation–response relation: the same Gaussian and plane-wave electronic-structure representation can feed a density-functional perturbation theory, time-dependent density-functional theory, or GW-Bethe–Salpeter static response, an ab initio or path-integral molecular dynamics trajectory, a time-correlation spectrum, a Kubo conductivity, or an open-boundary nonequilibrium Green's function transport calculation. The paper's claim is that this unification is not just a matter of having many modules, but of a single architecture in which the same energy, force, state, and response machinery supports both static and dynamical observables. If true, a computational scientist can compute, in one package, spectra and transport coefficients that include thermal disorder, anharmonicity, and nuclear quantum effects, with benchmarked errors of tens of meV for band structures and couplings.","feed_headline":"One platform now computes spectra with thermal disorder and bias","feed_subtitle":"The same electronic-structure representation carries static response, MD sampling, time-correlation spectra, and open-boundary currents.","key_machinery":"The central object is the perturbation–response relation expressed through a Gaussian and plane-wave (GPW) dual representation of the electronic structure, augmented by the all-electron GAPW extension. The GPW representation expands Kohn–Sham orbitals in atom-centered Gaussians while representing the electron density on a plane-wave grid; GAPW decomposes the density into a smooth part and atom-centered hard components, removing the core density from the grid and suppressing force oscillations. This dual representation provides smooth, repeatable energies and forces, makes response tensors and trajectories share the same Hamiltonian and overlap matrices, and supplies the locality that low-sca","core_discovery":"The central claim is that CP2K's response formalism unifies static and dynamical spectroscopy and transport. On the static side, the paper describes variational DFPT for electric, magnetic, and nuclear perturbations, linear-response TDDFT with Tamm–Dancoff and spin–orbit corrections, transition-potential and linear-response X-ray methods, GW and GW-BSE, and active-space embedding. On the dynamical side, the same electronic-structure layer feeds AIMD, path-integral MD, Ehrenfest and surface-hopping dynamics, real-time TDDFT, time-correlation spectra, Kubo transport, Hairy Probes open-boundary DFT, and DFT+NEGF. The paper's distinctive assertion is that a single electronic-structure representa","pith_inferences":["Editorial inference: if the real-space decay of the polarizability and self-energy is the load-bearing premise for large-supercell GW, then the advertised scalability to several thousand atoms is conditional on that decay, and should be tested on metals, narrow-gap systems, and three-dimensional supercells where the decay is slower.","Editorial extension: the perturbation–response viewpoint suggests a design principle for future modules: every new static response capability should have a defined trajectory-based or ensemble-based counterpart, or the unification claim weakens.","Editorial extrapolation: a natural stress test of the unified claim is to compare, for the same system, a time-correlation infrared spectrum against a snapshot-averaged DFPT spectrum; agreement would validate that the static and dynamical routes are truly consistent within one representation."],"forward_implications":["If the unified architecture holds, vibrational, electronic, X-ray, and magnetic spectra can be computed with thermal disorder and nuclear quantum effects in one package, rather than by stitching together separate codes.","Equilibrium conductivities from Kubo transport and biased currents from DFT+NEGF can be obtained from the same Hamiltonian, enabling operando comparison with experiment for disordered and driven systems.","GW quasiparticle band structures of large two-dimensional moiré systems become tractable: 984-atom cells are demonstrated, and the paper indicates that thousands of atoms are within reach.","Static approximations such as COHSEX yield smooth quasiparticle energy surfaces suitable for geometry optimization and molecular dynamics, avoiding the discontinuities that can plague perturbative GW.","Benchmarks in the paper put typical errors at roughly 20–50 meV for spin–orbit splittings and band gaps, and about 10 meV for charge-transfer couplings, making quantitative predictions plausible in those regimes."],"fun_headline_variants":["CP2K: one framework for spectroscopy, dynamics, and transport","Thermal sampling and open boundaries in CP2K's response engine","From static response to real-time dynamics in CP2K","CP2K unifies Kubo transport and time-correlation spectra"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the irreducible polarizability and self-energy of a large supercell decay in real space fast enough that the Γ-point-only minimum-image reconstruction of chi, W, and Sigma is accurate before periodic images overlap; the paper validates this only for gapped two-dimensional semiconductors in roughly 20-Angstrom cells, leaving metals, narrow-gap systems, and three-dimensional supercells without controlled error bounds.","fun_headline_variants_meta":{"raw":{"variants":["CP2K: one framework for spectroscopy, dynamics, and transport","Thermal sampling and open boundaries in CP2K's response engine","From static response to real-time dynamics in CP2K","CP2K unifies Kubo transport and time-correlation spectra"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000197,"raw_usage":{"total_tokens":1239,"prompt_tokens":817,"completion_tokens":422,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":561,"completion_tokens_details":{"reasoning_tokens":349}},"tokens_in":561,"tokens_out":422,"duration_ms":5458,"temperature":1.0,"reasoning_tokens":349,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T01:48:38.455586+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute GW quasiparticle band gaps of a metallic or narrow-gap three-dimensional supercell (for example, doped silicon or a small-gap semiconductor) at increasing supercell sizes, and compare the minimum-image reconstruction against a full k-point calculation; if the reconstructed gap does not converge to the full-k result before periodic images overlap, the central scalability claim for large-supercell GW fails.","supporting_citations":[],"review_version":2}