{"id":"f99699af-74d8-4283-83ee-d7458ea6070a","arxiv_id":"2607.07131","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":3,"one_line_summary":"A Hermitian linear vibronic coupling model, parameterized via first-principles and generative machine learning, successfully reproduces experimental absorption spectra and ultrafast population dynamics in silver nanoparticles up to 120 atoms.","lead":"This paper introduces a computational workflow (PyPC) that combines first-principles calculations with machine learning to simulate the quantum dynamics of plasmonic excitations in silver nanoparticles up to 120 atoms. It matters because it provides a scalable, atomistic method to predict ultrafast non-radiative decay in plasmonic nanocavities, a key step for designing quantum optical technologies.","discovery_kind":"new_method","skeptic_critique":{"model":"glm-5.2","headline":"For the largest clusters (Ag55, Ag120), the entire LVC Hamiltonian is GRCDE-generated with no parameter-level validation against FP data, despite TD-DFT+TB gradients being computable at these sizes.","rationale":"The reader correctly identified the state-independence assumption as a vulnerability, but the more load-bearing issue is the complete absence of parameter-level validation for the systems where the strongest claim is made. For Ag10, the paper validates GRCDE against FP data (Figure 6, pipeline II vs. pipeline I) and shows 'satisfactory agreement' — but this is a small cluster where the electronic structure is molecular, not plasmonic. For Ag55 and Ag120, where the electronic states are genuinely plasmonic and the GRCDE training data (from Ag10, Ag20) is most different in character, there is no analogous validation. The paper's own analysis in §3.1 shows that spectral densities J_n(ω) converge with cluster size, which supports the state-independence assumption, but this convergence is shown only for FP-computed data (Ag10, Ag20, Ag56) — not for GRCDE-generated data. The gap is specifically: does GRCDE reproduce the converged distributions, or does it introduce its own errors? The experimental spectral comparison is a real but weak constraint: a broad plasmonic peak with FWHM ~0.3–0.5 eV can be matched by many parameter combinations. The population dynamics (Figures 8c, 8d) have no experimental reference at all. The code availability (GitHub) is a genuine strength that enables reproducibility, but it does not substitute for the missing validation. The paper is a solid methodological contribution, but the claim of 'accurate prediction' for the largest systems is currently supported only by a necessary-but-not-sufficient spectral match. A single parameter-level cross-check at Ag55 would substantially strengthen or weaken the claim.","tokens_in":18740,"tokens_out":2640,"duration_ms":165182,"concrete_test":"Compute κ_i^(n) directly from TD-DFT+TB analytical gradients for 5–10 electronic states and all vibrational modes of Ag55 (Ih). Compare these FP-derived values against the GRCDE-generated values used in the Figure 8a spectrum. Report the mean absolute error and the distribution of errors. If the MAE exceeds ~0.02 eV (comparable to the typical κ scale shown in Figure 4f), or if systematic size-dependent bias is present, the GRCDE extrapolation is unreliable and the Ag55/Ag120 spectral agreement is not predictive.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim — that the workflow 'successfully reproduces the experimental absorption spectra' of Ag55 and Ag120 — rests on Hamiltonians built entirely from pipeline III (GRCDE-generated κ and λ), with validation limited to visual comparison of absorption spectra (Figures 8a, 8b). No quantitative error metric (RMSE, spectral overlap, linewidth error) is reported. More critically, the paper states that analytical excited-state gradients are available within TD-DFT+TB for these cluster sizes (§2.3, citing Havenridge et al. in AMS 2023), meaning κ parameters could in principle be computed directly for Ag55 or Ag120 — at least for a subset of states and modes — and compared against GRCDE predictions. This check is never performed. The GRCDE feature vector (electronic state energy, vibrational frequency, symmetry label) is low-dimensional and does not include descriptors that change with cluster size, such as surface-to-volume ratio, local coordination, or s–d transition character. If the coupling parameters depend on these omitted features, the GRCDE extrapolation from Ag10/Ag20 data to Ag55/Ag120 could be systematically biased even if the state-independence assumption (the reader's concern) holds. The spectral agreement with experiment is necessary but not sufficient: multiple parameter sets can reproduce a single broad absorption feature, especially when damping and broadening are applied. Without at least one parameter-level cross-check at intermediate size, the 'accurate prediction' framing for the largest systems is not independently supported.","agreement_with_reader":"partial"},"referee_report":{"model":"glm-5.2","summary":"This manuscript introduces the Python Plasmonic Cavity (PyPC) platform for parameterizing linear vibronic coupling (LVC) Hamiltonians to describe plasmonic excitations in silver nanoclusters. The approach combines first-principles (FP) calculations for small clusters with a generative machine-learning method (GRCDE) to extrapolate LVC parameters for larger systems. The authors demonstrate their workflow on Ag clusters ranging from 10 to 147 atoms, reporting reproduction of experimental absorption spectra and ultrafast population dynamics for bright and dark states. The work addresses a genuine methodological gap — the atomistic quantum dynamics of plasmonic nanoparticles in the 20–120 atom regime — and the combination of LVC Hamiltonians with ML-based parameter generation is a reasonable and potentially impactful strategy. However, the validation of the GRCDE-generated parameters for the largest clusters (Ag55, Ag120) is limited to qualitative spectral comparison, and a parameter-level cross-check against FP data at intermediate sizes, which appears feasible given the stated availability of TD-DFT+TB gradients, is not performed.","tokens_in":19803,"tokens_out":1691,"duration_ms":257912,"significance":"The manuscript tackles a real and difficult problem: extending fully quantum dynamical treatment of plasmonic excitations beyond the few-atom regime while retaining an atomistic, non-phenomenological description of non-radiative decay. The release of the PyPC code as an open-source tool (GitHub link provided) is a positive step for reproducibility. The identification of statistical regularities in LVC parameter distributions (e.g., the state-independence of κ/ω for larger clusters, the linear scaling of mode counts by irreducible representation) provides useful empirical foundations for the extrapolation strategy. The application to clusters with over 100 atoms and thousands of electronic states, propagated via ML-MCTDH, represents a meaningful scale-up compared to prior LVC studies.","major_comments":[{"comment":"§3.2, Figures 8a–b: The claim that the workflow 'successfully reproduces the experimental absorption spectra' of Ag55 and Ag120 rests entirely on GRCDE-generated LVC parameters (pipeline III) with validation limited to visual comparison of broad spectral features. No quantitative error metric (e.g., RMSE, spectral overlap, peak position error, linewidth error) is reported. Given that broad plasmonic absorption bands can be reproduced by multiple parameter sets — especially when damping is applied — the spectral agreement alone is necessary but not sufficient to validate the GRCDE-extrapolated Hamiltonian. The authors should provide at least one quantitative comparison metric between computed and experimental spectra for all clusters shown in Figs. 7–8.","section":null},{"comment":"§3.2 and §2.3: For Ag55 and Ag120, the entire LVC Hamiltonian (both κ and λ) is GRCDE-generated with no parameter-level validation against FP data. The manuscript states (§2.3, citing Havenridge et al., Ref. 41) that analytical excited-state gradients are available within TD-DFT+TB, meaning κ parameters could in principle be computed directly for at least a subset of states and modes at these sizes. A direct comparison of GRCDE-predicted κ values against FP-computed κ values for, e.g., 5–10 electronic states and a subset of vibrational modes of Ag55 would substantially strengthen the claim that the GRCDE extrapolation is reliable. This check is absent and is the single most important missing validation.","section":null},{"comment":"§2.2, Eq. (5): The GRCDE feature vector for predicting κ and λ consists of (electronic state energy, vibrational frequency, symmetry label). This is a low-dimensional descriptor set that does not include cluster-size-dependent features such as surface-to-volume ratio, local coordination number, or s–d transition character. If the coupling parameters depend on these omitted features, the extrapolation from Ag10/Ag20 training data to Ag55/Ag120 targets could be systematically biased. The authors should discuss whether such size-dependent descriptors were considered and why they were excluded, or provide evidence that the chosen features are sufficient.","section":null},{"comment":"§3.1, Fig. 4: The load-bearing assumption that the distribution of normalized on-diagonal parameters (κ/ω) is independent of the specific plasmonic electronic state for sufficiently large clusters is supported by visual inspection of spectral densities J_n(ω) for Ag10, Ag20, and Ag56 (Figs. 4a–d). However, no quantitative test of this independence (e.g., a Kolmogorov–Smirnov test or variance ratio across states) is provided. Since this assumption underpins the entire GRCDE sampling strategy for κ, a quantitative justification — even a simple statistical test on the Ag20 or Ag56 data — would strengthen the extrapolation strategy.","section":null}],"minor_comments":[{"comment":"§2.1, Eq. (1): The notation mixes hats and plain symbols for operators (e.g., Q̂_i in the first term but Q_i elsewhere). Consistent operator notation would help reproducibility.","section":null},{"comment":"§2.2: The kernel bandwidth used in the univariate KDE for GRCDE is not specified. Since this is a free parameter that affects the predicted coupling constants, its value (or selection procedure) should be reported.","section":null},{"comment":"§2.5: The damping time τ = 150 fs is used for Ag10 spectra (Fig. 6) but the caption of Fig. 7 states that no damping function is applied. The rationale for applying or not applying damping should be clarified, and the τ value should be justified.","section":null},{"comment":"Fig. 5: The panel labels (a), (b), (c) are referenced in the text but the distinction between 'full set' and 'reduced set' of parameters is not immediately clear from the figure legend. A more descriptive legend within the figure would help.","section":null},{"comment":"§3.1: The statement 'the distribution of first-order LVC parameters turns out to be independent to the electronic state' is a key claim. The phrase 'turns out to be' should be replaced with a more precise statement of the evidence supporting this claim.","section":null},{"comment":"Table S1 and Figures S2–S3 (referenced in §2.5) are mentioned as detailing efficiency and scalability but are in the Supporting Information. A brief summary of the scaling in the main text would be helpful.","section":null},{"comment":"§3.2: The population dynamics for Ag55 and Ag120 (Figs. 8c–d) show summed bright and dark state populations but do not identify individual dark states. Given that the paper claims to 'effectively capture the dynamics of dark states,' showing at least one individual dark-state trajectory would strengthen this claim.","section":null},{"comment":"The reference list contains an arXiv preprint stamp date of 8 Jul 2026, which appears to be a typographical error in the preprint stamp.","section":null}],"recommendation":"major_revision","confidential_remarks":"The reader's concern about circularity is not strictly correct — the training data (Ag10, Ag20, Ag56) and validation targets (Ag55, Ag120) are different systems, so the spectral agreement is not circular in the formal sense. However, the skeptic's concern about the absence of parameter-level validation at intermediate sizes is well-placed and is the primary reason for the major_revision recommendation. The GRCDE method is essentially a conditional density estimator on a low-dimensional feature space; without at least one direct FP cross-check at the target scale, the claim of 'accurate' parameter generation is not adequately supported. The authors appear to have the computational tools (TD-DFT+TB gradients) to perform this check, so it should be feasible within a revision cycle."},"author_rebuttal":null,"desk_editor":{"model":"glm-5.2","letter":"Short version: this paper combines the LVC Hamiltonian with ML-MCTDH quantum dynamics and a machine-learning extrapolation scheme (GRCDE) to compute absorption spectra and population dynamics for silver nanoclusters up to 120 atoms. That size regime has not been accessible to fully quantum wavepacket dynamics before, and the code (PyPC) is public. The core idea — that first-order vibronic coupling parameters, normalized by frequency, become approximately state-independent for sufficiently large clusters — is plausible and supported by the spectral density analysis in §3.1. The Ag10 benchmark (Fig. 6) showing that symmetry-reduced and GRCDE-generated λ parameters reproduce the full-FP spectrum is a useful sanity check. Reproducing experimental absorption lineshapes for Ag55 and Ag120 (Fig. 8) is a real result, even if qualitative. The population dynamics — bright-state lifetimes of ~10–20 fs — are consistent with prior work and physically reasonable. Code availability is a genuine plus. The stress-test concern lands. The paper states that analytical TD-DFT+TB gradients are available for Ag55 and Ag120 sizes (§2.3, citing Havenridge et al.). That means κ parameters could be computed directly for at least a subset of states and modes at those sizes and compared against GRCDE predictions. This check is never done. For the largest clusters, the entire Hamiltonian is GRCDE-generated, and validation is limited to visual spectral comparison. No RMSE, no spectral overlap metric, no linewidth error. The GRCDE feature vector (state energy, vibrational frequency, symmetry label) is low-dimensional and omits size-dependent descriptors like surface-to-volume ratio or local coordination. If coupling parameters depend on those, the extrapolation could be systematically biased. The reader's concern about state-independence is related but slightly off-target — the bigger issue is that the ML extrapolation is never validated at the parameter level for the systems where it matters most. The circularity burden is moderate, not severe: training data (Ag10, Ag20, Ag56) and validation targets (Ag55, Ag120) are genuinely different systems, so the spectral agreement is not trivially circular. But without one parameter-level cross-check at intermediate size, the 'accurate prediction' framing for the largest clusters is not fully earned. This is a solid methods paper with a real new capability. It deserves a serious referee who should ask for at least one direct κ comparison at Ag55 or Ag120 size, and ideally a quantitative error metric on the spectra.","headline":"LVC + ML-MCTDH workflow reaches 120-atom plasmonic clusters, but the largest systems lack parameter-level validation against the FP methods that are computationally available at those sizes.","tokens_in":19753,"tokens_out":600,"would_cite":true,"duration_ms":145991,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["31.15.xv","78.67.Bf"],"model":"glm-5.2","headline":"Statistical LVC parameters predict plasmonic decay in 120-atom silver clusters","keywords":[],"falsifier":"If the absorption spectrum or population dynamics of a 55- or 120-atom silver cluster computed with the GRCDE-sampled Hamiltonian disagreed significantly with experiment, or if the state-independence of normalized kappa parameters were shown to fail for a subset of modes or states that dominate the plasmonic response, the extrapolation strategy would be invalid.","tokens_in":19025,"feed_emoji":" nanop","tokens_out":922,"duration_ms":189384,"temperature":0.7,"pith_summary":"Plasmonic excitations in metal nanoparticles are collective oscillations of free electrons driven by light, but modeling their quantum dynamics is hard because non-radiative decay channels and dense manifolds of electronic states make explicit simulations intractable beyond about 20 atoms. This paper introduces a Hermitian linear vibronic coupling (LVC) framework, automated by a platform called PyPC, that parameterizes the Hamiltonian using first-principles data for small clusters and a generative machine-learning method (GRCDE) to extrapolate coupling constants for larger ones. The key enabler is the empirical observation that first-order vibronic coupling constants, when normalized by vibrational frequency, become nearly independent of the specific electronic state as clusters grow past about 20 atoms. This statistical regularity lets the authors sample coupling parameters from small-cluster datasets and construct full Hamiltonians for silver nanoparticles with 55 and 120 atoms, each involving thousands of parameters and hundreds of vibrational modes. The resulting quantum dynamics, propagated with the ML-MCTDH method, reproduce experimental absorption spectra and predict ultrafast bright-state lifetimes of roughly 10 to 20 femtoseconds.","feed_headline":"Statistical LVC parameters predict plasmonic decay in 120-atom silver clusters","feed_subtitle":"A machine-learned vibronic coupling model reproduces absorption spectra and ultrafast lifetimes in nanoparticles too large for direct first-","key_machinery":"The LVC Hamiltonian is a Taylor expansion of the diabatic potential energy surface to first order in dimensionless normal coordinates, with on-diagonal terms (kappa) describing how each excited state's potential shifts along each vibrational mode and off-diagonal terms (lambda) describing interstate coupling. The GRCDE method treats each coupling constant as a scalar target conditioned on features (electronic state energy, vibrational frequency, mode symmetry), uses kernel density estimation on a reference dataset from small clusters to estimate the conditional distribution, and takes its mean as the predicted parameter. The Hamiltonian is then propagated with the multilayer multiconfigurat ","core_discovery":"The central claim is that the first-order LVC coupling constants for plasmonic silver clusters converge to a state-independent statistical distribution as cluster size increases, and that this distribution, sampled via kernel density estimation from smaller-cluster first-principles data, is sufficient to construct accurate vibronic Hamiltonians for clusters of 100+ atoms. The resulting Hermitian Hamiltonians reproduce experimental absorption spectra and yield physically meaningful population dynamics for both bright and dark plasmonic states.","pith_inferences":[],"forward_implications":["If the state-independence of normalized LVC parameters holds for other noble metals and alloy compositions, the same pipeline could model gold and mixed metal plasmonic nanoparticles without new first-principles gradient data for each target size.","The Hermitian LVC framework could be extended to include coupling between the plasmonic cavity and nearby molecular emitters, enabling quantum dynamics simulations of polaritonic systems at the single-emitter level without phenomenological dissipation parameters.","The statistical sampling approach could be combined with higher-order vibronic coupling terms (quadratic or cubic) to systematically improve accuracy for systems where the first-order approximation breaks down.","The predicted ultrafast bright-state lifetimes of 10 to 20 fs for large clusters provide quantitative targets for ultrafast pump-probe experiments on size-selected silver nanoparticles."],"fun_headline_variants":["LVC coupling constants converge to size-independent distribution in silver clusters","Machine-learned vibronic model reproduces spectra in 100+ atom silver nanoparticles","First-principles LVC captures bright and dark plasmonic state dynamics","Hermitian vibronic Hamiltonian extends plasmonic modeling to 100-atom clusters","Statistical sampling of coupling constants predicts plasmonic lifetimes"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The load-bearing premise is that the distribution of first-order vibronic coupling constants, normalized by vibrational frequency, is effectively independent of which specific plasmonic electronic state is considered once the cluster is large enough (roughly 20+ atoms). If this statistical independence breaks down for certain geometries, compositions, or size regimes, the machine-learned Hamiltonians for 55- and 120-atom clusters would not faithfully represent the true vibron","fun_headline_variants_meta":{"raw":{"variants":["LVC coupling constants converge to size-independent distribution in silver clusters","Machine-learned vibronic model reproduces spectra in 100+ atom silver nanoparticles","First-principles LVC captures bright and dark plasmonic state dynamics","Hermitian vibronic Hamiltonian extends plasmonic modeling to 100-atom clusters","Statistical sampling of coupling constants predicts plasmonic lifetimes","PyPC workflow parameterizes LVC from DFT for large silver plasmonic systems","Vibronic coupling model recovers experimental absorption in silver nanoparticles"]},"model":"glm-5.2","effort":"low","cost_usd":0.0,"raw_usage":{"total_tokens":1393,"prompt_tokens":491,"completion_tokens":902,"prompt_tokens_details":null},"tokens_in":491,"tokens_out":902,"duration_ms":66913,"temperature":1.0,"reasoning_tokens":837,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-09T19:21:05.852207+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"If the absorption spectrum or population dynamics of a 55- or 120-atom silver cluster computed with the GRCDE-sampled Hamiltonian disagreed significantly with experiment, or if the state-independence of normalized kappa parameters were shown to fail for a subset of modes or states that dominate the plasmonic response, the extrapolation strategy would be invalid.","supporting_citations":[],"review_version":1}