{"id":"bb84dc47-9911-401d-b745-39ef21352767","arxiv_id":"2607.25089","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Ab initio electron–phonon Hamiltonians plus nonperturbative Green–Kubo dynamics predict molecular-crystal transport and attribute DNTT localization to correlated acoustic on-site disorder, not independent hopping fluctuations.","lead":"A parameter-free computational framework predicts charge mobility in organic crystals from atomic structure by combining ab initio electron–phonon Hamiltonians with large-scale Green–Kubo dynamics. It matches experiments on five materials, revises the accepted bottleneck in DNTT, and flags phenacenes such as picene as a high-mobility design class.","discovery_kind":"new_method","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"The headline DNTT bottleneck reassignment rests on in-silico channel interventions inside a no-back-action mixed quantum–classical dynamics, and none of the four benchmark families covers the regime that claim lives in: strong, low-frequency (acoustic), correlated onsite-plus-hopping disorder in 2D.","rationale":"The reader identified the correct load-bearing family — the mixed quantum–classical partition with no back-action, the ω_c = max(2k_B T, |t|_max) cutoff, and the variational polaron trace-out — and priced it as medium correctness risk under an ACCEPT verdict. I agree with the identification and would sharpen it rather than replace it. The sharpening matters because the risk is not uniformly distributed across the paper's claims. The quantitative-capture claim (mobilities, exponents, optical fingerprints across five crystals) is triangulated by several independent supports: model-exact benchmarks across coupling regimes (g_P to 1.05, λ to 2t), ARPES checks of bandwidths and transfer integrals including the partially-dressed bandwidth matching experiment (Fig. S2), multi-material internal consistency (C8-DNTT vs C8-BTBT), and the explicit intrinsic-vs-extrinsic framing of the C8-DNTT tension. That claim would likely survive moderate distortion of C(t) by the MQC approximations. The DNTT mechanistic reassignment is different: it is the paper's flagship novelty, it is falsifiable in principle, but its dynamical evidence is generated by the very approximation under question, and the benchmark suite — while genuinely impressive and a real strength of the paper — does not include the regime the claim occupies (2D, strong low-frequency onsite coupling, correlated multi-channel modulation). That is a coverage gap, not a demonstrated error, and it is cheap to close because the authors already operate finite-temperature MPS and HMC-grade reference machinery. Hence CONDITIONAL rather than UNCHANGED: I would accept the framework and quantitative claims on current evidence, but attach one condition — a targeted exact benchmark in the DNTT-like parameter corner with the onsite-removal intervention reproduced in both solvers — before the bottleneck reassignment is treated as established rather than strongly suggested. This is consistent with the reader's medium correctness risk; it simply makes explicit which claim carries that risk and what retires it. Credit where due: the parameter-free workflow, shipped code (PyEPH) and data, grid-convergence checks (Fig. S5), and the multi-level validation strategy are real evidence and are why this is a condition, not a rejection.","tokens_in":31239,"tokens_out":2043,"duration_ms":1202893,"concrete_test":"Build a 1D (or small 2D cluster) multimode Holstein–Peierls model with DNTT-like parameters — t ≈ 60–90 meV; one dispersionless low-frequency mode ω ≈ 5–15 meV coupled both onsite (RMS onsite fluctuation ΔE_B ≈ 30–50 meV at 300 K) and to the hopping with the correlated sign structure of Fig. 2d — and solve it at T = 150–300 K with the authors' existing finite-temperature MPS machinery (Methods, Eqs. 20–21) and with the production MQC Green–Kubo solver. Compare (i) C(t) and μ(T) exponent, (ii) the memory kernel's decay within the explicit window, and (iii) the onsite-channel-removal intervention run in both solvers. If the exact reference shows the same intervention hierarchy (onsite removal → Drude-like σ(ω), exponent/mobility shifts within ~30%), the central concern is retired; if the exact dynamics reallocates weight to the hopping channel or changes the exponent, the DNTT reassignment","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's most novel claim is not the mobility numbers but the mechanistic reassignment for DNTT: transient localization driven by correlated onsite disorder from finite-q acoustic phonons, not independent hopping fluctuations. The static half of that claim is solid — the mode-resolved fluctuation decomposition (Fig. 2d, Figs. S4–S6) is a property of the ab initio Hamiltonian and does not depend on the dynamics solver. The dynamical half, however, is established entirely by interventions run inside the approximate solver: removing the onsite EPC channel restores a Drude-like σ(ω) (Fig. 2f), and removing onsite disorder extends the delocalization length and suppresses the localized tail (Fig. 3d). Those interventions inherit the solver's central approximation (Eqs. 22–27): slow phonons are sampled from the bare Boltzmann/Wigner distribution and propagated as classical fields with no electronic back-action, so transient localization is produced by construction from quasistatic bare-disorder sampling, and lattice relaxation/self-trapping response of the slow modes — the physics most capable of reshuffling the onsite-vs-hopping attribution for a strongly coupled slow mode — is absent. The authors acknowledge the detailed-balance violation (SI §1) and lean on the benchmark suite (Fig. 2a) for validation. But the suite's coverage is: 1D Peierls (hopping modulation only), 1D Holstein (single-mode onsite), 2D Peierls (hopping only, and only against HMC with analytic continuation), and a multimode Holstein–Peierls whose nine Holstein modes are high-frequency intramolecular modes of the kind the polaron transform traces out. The DNTT-relevant corner — a strongly coupled, low-frequency (ℏω ≪ k_B T) acoustic-like mode modulating onsite energy and several hoppings with correlated signs, in a 2D lattice — appears in no benchmark. A secondary, related exposure: for DNTT the explicit C(t) window is ~80 fs (Fig. S9) while the relevant acoustic periods are hundreds of fs, so the Mo","agreement_with_reader":"agree"},"referee_report":{"model":"moonshotai/kimi-k3","summary":"The manuscript presents an end-to-end first-principles framework for charge transport in molecular crystals. Starting from DFT/DFPT electron–phonon Hamiltonians, the authors develop a gauge-invariant maximal-localization scheme (Eq. 13) to obtain compact real-space electron–phonon couplings, then propagate carriers on domains of up to 648 molecules using a mixed quantum–classical Green–Kubo solver: high-frequency modes are traced out via a variational polaron transform, low-frequency modes are sampled classically, and a Mori memory kernel extends the current correlation to long times. The solver is benchmarked against TD-DMRG, HEOM, and HMC references on 1D/2D Peierls, Holstein, and multimode Holstein–Peierls models (Fig. 2a). Applied to DNTT, pentacene, picene, C8-BTBT, and C8-DNTT, the framework reproduces measured mobilities, temperature exponents, and optical-conductivity lineshapes, and is cross-checked against ARPES bandwidths and transfer integrals (Fig. S2). The headline mechanistic claim is that DNTT's transient localization is driven by correlated onsite disorder from finite-q acoustic phonons rather than independent hopping fluctuations, supported by mode-resolved fluctuation decomposition (Fig. 2d) and channel-removal interventions (Figs. 2f, 3d). A two-axis transport map (connectivity × low-frequency disorder) and a phenacene design suggestion follow.","tokens_in":25653,"tokens_out":2996,"duration_ms":19173,"significance":"If the results hold, this is a substantial advance: quantitative, unfitted mobility and temperature-exponent predictions across five molecular crystals from ab initio Hamiltonians, at a domain scale (160,000+ phonon modes) not previously accessible to nonperturbative dynamics. Particular strengths worth naming: (i) the Hamiltonian is fixed by DFT/DFPT and dynamics parameters by internal criteria, so the agreement with FET/THz mobilities and exponents (Fig. 2b) is not a fit; (ii) independent ARPES validation of bandwidths and transfer integrals, including a nontrivial check that partial polaron dressing matches experiment (Fig. S2); (iii) direct head-to-head benchmarks of the solver against numerically exact MPS on a 9+1-mode Holstein–Peierls model; (iv) public data and the PyEPH code, supporting reproducibility. The DNTT bottleneck reassignment, if robust, corrects a widely cited mechanistic picture and the two-axis map offers genuinely falsifiable design guidance for the phenacene family.","major_comments":[{"comment":"The paper's most novel claim — that DNTT's transient localization is driven by correlated onsite disorder rather than independent hopping fluctuations — rests dynamically on channel interventions (Fig. 2f: onsite removal restores Drude-like σ(ω); Fig. 3d: onsite removal suppresses the localized delocalization tail) performed inside the no-back-action solver of Eqs. 22–27. In that scheme slow phonons are sampled from the bare Boltzmann/Wigner distribution and evolve as classical fields without electronic back-action, so quasistatic-disorder-induced localization is partially built in, and the lattice relaxation/self-trapping response of the strongly coupled slow modes — the physics most likely to reshuffle the onsite-vs-hopping attribution — is absent by construction. The benchmark suite (Fig. 2a) does not cover this regime: 1D Peierls is hopping-only, 1D Holstein is single-mode onsite, th","section":"Fig. 2f, Fig. 3d; Methods Eqs. (22)–(27)"},{"comment":"The cutoff ω_c = max(2k_BT, |t|_max) is the single most consequential heuristic in the method: it decides which modes are dressed analytically and which are sampled classically, and it is temperature-dependent, so it directly shapes the predicted temperature exponents that serve as a key experimental validation (Fig. 2b). Yet no sensitivity analysis is shown. Please report how μ magnitudes, exponents a in μ ∝ T^(−a), and the DNTT σ(ω) lineshape change under a reasonable variation of ω_c (e.g., ±50% or 2k_BT vs |t|_max separately), and under Wigner vs Boltzmann sampling for the slow modes. If the exponents are robust, this substantially strengthens the paper; if not, the uncertainty should be stated.","section":"Methods, after Eq. (22); Fig. 2b"},{"comment":"The Green–Kubo mobility integral relies on extending the explicitly propagated C(t) via the Mori memory kernel, valid only when M(t) decays within the explicit window. Fig. S9b shows DNTT has a notably long-lived memory kernel — the material for which the quantitative mobility agreement is most heavily advertised. Please quantify: the propagation window, the residual magnitude of M(t) at the window edge for DNTT, and the variation of the extracted μ as the reconstruction window is shortened/lengthened. Without this, the DNTT number in Fig. 2b–c carries an unquantified extrapolation error.","section":"Methods, memory-kernel reconstruction (Eq. 19); Fig. S9b"}],"minor_comments":[{"comment":"Fig. 2 caption/panel text contains typos: 'collevctive motions', and in Fig. 4 'photon absorbtion'; also 'fromab ini-tioelectron–phonon' in the abstract/introduction has a spacing error.","section":"Figs. 2 and 4; abstract"},{"comment":"Fig. 3a: the axis positions are described as 'relative rankings inferred from' spectra and transfer-integral networks, and the mobility background is a visual interpolation. A quantitative definition of the two coordinates (even a heuristic one, e.g., integrated low-frequency fluctuation variance vs. signed plaquette product) would make the map more useful and let future materials be placed reproducibly.","section":"Fig. 3a"},{"comment":"The 2D Peierls benchmark against HMC (Ref. 37) relies on analytic continuation of imaginary-time data, which is known to be ill-conditioned for real-time dynamics. This should be flagged explicitly in the text so readers can weigh that benchmark accordingly.","section":"Fig. 2a; SI benchmarking discussion"},{"comment":"C8-DNTT: the discrepancy with THz-probe mobility is handled reasonably, but the comparison would be cleaner if the text stated the probe dependence (THz vs FET vs cAFM) quantitatively rather than qualitatively 'narrowing this gap'.","section":"Fig. 2b discussion"},{"comment":"Eq. (13): the localization functional and the role of the δ offsets should be given one more sentence of motivation; as written the weighting choice appears ad hoc, and the physical interpretation of the branch-resolved R_p index is only clarified in the Methods paragraph after Eq. (15).","section":"Methods, Eq. (13)"},{"comment":"Data/code: the repository links are given, but versioned DOIs are promised only 'upon publication'. Depositing an archival release at revision would make the claims checkable now.","section":"Data and code availability"}],"recommendation":"major_revision","confidential_remarks":"A strong and potentially field-shaping manuscript with unusually good validation discipline. My one substantive reservation is that the headline DNTT mechanistic reassignment — prominently featured in the abstract — is established by interventions inside an approximate solver in a parameter regime the benchmark suite does not cover. This is fixable within the manuscript's scope (the authors have tensor-network tooling in-house and could run the targeted 2D control), so I recommend major rather than minor revision; if the authors supply the cutoff-sensitivity and kernel-convergence analyses plus one targeted benchmark, I would expect to recommend acceptance. No concerns about citation balance or scope fit."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The thing worth knowing is that they actually built the missing pipeline: full solid-state DFPT electron–phonon Hamiltonians, gauge-invariant real-space localization, then nonperturbative Green–Kubo on domains of hundreds of molecules with the full phonon spectrum. No prescribed mechanism, no few-mode reduction, no empirical force field. That alone is new and useful.\n\nWhat they do well is multi-observable, not just mobility chasing. Model dynamics match TD-DMRG/HEOM/HMC in the Holstein–Peierls regimes they tested. Material runs hit FET/THz μ(T) and exponents for DNTT and C8-BTBT, ARPES bandwidths and transfer integrals line up, and the optical-conductivity interventions (onsite-only vs remove-onsite) are clean. Circularity is low: Hamiltonian from DFT/DFPT, cutoffs and polaron parameters fixed internally, not fitted to μ. Code and data are shipped. The two-axis map and the phenacene/picene pointer follow naturally from the calculations rather than from slogans.\n\nThe soft spot is real but bounded. The static half of the DNTT claim—mode- and channel-resolved disorder dominated by correlated acoustic onsite fluctuations—is a property of the ab initio Hamiltonian and does not need the dynamics solver. The dynamical half (remove onsite → Drude-like σ(ω), longer delocalization) lives inside the mixed quantum–classical partition: slow modes sampled without electronic back-action, fast modes traced out by variational polaron. Their benchmark suite does not fully cover the DNTT corner (strong, low-frequency, correlated onsite-plus-hopping in 2D). They know the detailed-balance issue and lean on short-time Green–Kubo plus memory kernels, which is the right mitigation, but it is still an exposure. C8-DNTT magnitude vs one THz probe is a secondary tension they already flag. DFT/PBE+D3 EPC error is the usual residual.\n\nThis is for people who care about intrinsic crystal-domain transport and design rules, not device scatter. Math and citation pattern look solid; no load-bearing fitting dressed up as prediction. I would bring it to reading group, cite the workflow and the DNTT static diagnosis, and send it to peer review. Engage.","headline":"Real end-to-end ab initio transport workflow with a credible DNTT reassignment; the static EPC story is solid, the dynamical half sits on a standard but under-benchmarked mixed QC partition.","tokens_in":26316,"tokens_out":586,"would_cite":true,"duration_ms":18749,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A parameter-free first-principles method predicts molecular-crystal charge transport and shows DNTT’s localization comes from correlated acoustic on-site disorder, not independent hopping fluctuations.","keywords":["charge transport","molecular semiconductors","electron-phonon coupling","Green-Kubo dynamics","transient localization","DNTT","picene","phenacenes"],"falsifier":"Recompute DNTT mobility, temperature exponent, and optical conductivity with a dynamics method that restores electronic back-action on the acoustic modes (or with a different fast/slow cutoff) on the same ab initio Hamiltonian; if the onsite-disorder bottleneck and displaced-Drude peak disappear or the mobility temperature power changes materially, the central mechanistic claim fails. Independently, single-crystal measurements that cleanly separate intrinsic mobility from contacts and grain boundaries should converge to the predicted intrinsic values and temperature exponents.","tokens_in":25896,"feed_emoji":"⚡","tokens_out":1047,"duration_ms":29906,"temperature":0.7,"pith_summary":"Predicting how charge moves through organic molecular crystals has stayed hard because electrons couple to many phonons of different frequencies and ranges, putting carriers outside simple band or hopping pictures. This paper builds a fully first-principles route: it takes solid-state electron–phonon Hamiltonians, localizes them without gauge artifacts, and propagates carriers across hundreds of molecules with the full phonon spectrum using nonperturbative Green–Kubo dynamics. Across five crystals the calculations match measured mobilities, temperature trends, and optical-conductivity shapes. For the low-mobility workhorse DNTT they overturn the usual story, tracing transient localization to collective acoustic phonons that create correlated on-site energy disorder rather than independent intermolecular hopping noise. The same results organize the materials on a two-axis map of electronic connectivity versus low-frequency disorder and flag the phenacene family, with picene as the high-mobility example, as a promising underexplored class.","feed_headline":"Acoustic on-site disorder, not hop noise, bottlenecks DNTT","feed_subtitle":"Parameter-free first-principles dynamics match mobilities and flag phenacenes as a high-mobility class","key_machinery":"Gauge-invariant real-space localization of the full ab initio electron–phonon Hamiltonian, followed by mixed quantum–classical Green–Kubo dynamics (classical sampling of slow phonons plus variational polaron dressing of fast modes, with memory-kernel reconstruction of the current correlation). This object lets mobility, optical conductivity, and real-space coherence emerge without a prescribed transport regime and supplies the two coordinates of the transport map: electronic connectivity and low-frequency electron–phonon disorder.","core_discovery":"Transport regimes and bottlenecks in molecular semiconductors can be computed directly from ab initio electron–phonon Hamiltonians at domain scale, without assuming a mechanism or reducing the coupling to a few modes. The resulting dynamics reproduce experimental mobilities, temperature exponents, and optical fingerprints, and show that DNTT’s transient localization is driven by correlated on-site disorder from thermally populated acoustic phonons rather than independent hopping fluctuations.","pith_inferences":["If acoustic on-site disorder is the generic bottleneck in frustrated herringbone packings, pressure, side-chain, or dielectric engineering that hardens those acoustic branches should raise mobility more than further π-extension alone.","The framework’s domain-scale coherence maps suggest analogous two-axis organization may apply to soft hybrid semiconductors and 2D molecular frameworks where low-frequency lattice modes likewise dominate.","Alkali-intercalated picene superconductivity is consistent with the same weak-disorder, high-coherence corner of the map; charge-doped phenacenes may therefore be natural candidates for other collective electronic phases."],"forward_implications":["Design rules shift from maximizing transfer integrals alone toward simultaneously securing constructive connectivity and suppressing low-frequency acoustic electron–phonon disorder.","Phenacenes, sharing picene’s armchair frontier-orbital topology, are predicted as a coherent high-mobility class rather than isolated lucky compounds.","The same localized Hamiltonians become a common input for other nonperturbative solvers, turning mechanism diagnosis into a shared, material-specific starting point.","Machine-learning or semiempirical surrogates for the phonon and coupling step can turn the two-axis map into a high-throughput screen for new molecular semiconductors.","Channel-intervention diagnostics (onsite-only versus hopping-only) become a practical way to assign optical and mobility fingerprints to specific phonon channels."],"fun_headline_variants":["DNTT bottleneck is acoustic on-site disorder, not hop noise","Ab initio dynamics pin DNTT localization to acoustic disorder","Full-phonon Green-Kubo map overturns DNTT hop-fluctuation view","Parameter-free e-ph dynamics flag phenacenes as high-mobility class","Transport regimes emerge from domain-scale ab initio e-ph Hamiltonians"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"Low-frequency phonons are treated as classical fields sampled without forces back from the electrons, with a fixed frequency cutoff separating them from high-frequency modes that are removed analytically by a polaron transform.","fun_headline_variants_meta":{"raw":{"variants":["DNTT bottleneck is acoustic on-site disorder, not hop noise","Ab initio dynamics pin DNTT localization to acoustic disorder","Full-phonon Green-Kubo map overturns DNTT hop-fluctuation view","Parameter-free e-ph dynamics flag phenacenes as high-mobility class","Transport regimes emerge from domain-scale ab initio e-ph Hamiltonians"]},"model":"grok-4.5","effort":"low","cost_usd":0.003423,"raw_usage":{"total_tokens":1133,"prompt_tokens":735,"num_sources_used":0,"completion_tokens":81,"cost_in_usd_ticks":34228000,"prompt_tokens_details":{"text_tokens":735,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":317,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":735,"tokens_out":81,"duration_ms":6354,"temperature":1.0,"reasoning_tokens":317,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T01:37:38.071475+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Recompute DNTT mobility, temperature exponent, and optical conductivity with a dynamics method that restores electronic back-action on the acoustic modes (or with a different fast/slow cutoff) on the same ab initio Hamiltonian; if the onsite-disorder bottleneck and displaced-Drude peak disappear or the mobility temperature power changes materially, the central mechanistic claim fails. Independently, single-crystal measurements that cleanly separate intrinsic mobility from contacts and grain boundaries should converge to the predicted intrinsic values and temperature exponents.","supporting_citations":[],"review_version":1}