{"id":"cc796cd2-1304-48a8-bc7e-088f969153ec","arxiv_id":"2512.07537","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"Transfer-trained MACE potentials reproduce XMCQDPT2-level excited-state surfaces for CH2NH2+, and a fitted geometric-survival model extracts state-specific lifetimes from the simulated dynamics.","lead":"The paper trains machine-learning potentials on costly quantum-chemistry data to simulate the ultrafast decay of a small charged molecule after UV excitation, mapping its photodissociation channels. It matters because it shows a practical recipe — transfer learning plus uncertainty corrections — for making high-accuracy excited-state dynamics affordable for model chromophores.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Kinetic support for the sigma-pi*/S0 pathway rests on an unquantified model-selection claim: no statistical comparison to a single-channel fit is reported, so 'requires branching' is not established.","rationale":"The paper's most valuable contribution is the transfer-learning MLIP and the ensemble-uncertainty correction, which are plausibly supported by test-set metrics and MECI validation. However, the headline claim goes further: it states that kinetic fits yield channel-specific lifetimes that support the sigma-pi*/S0 pathway. This support rests on a model-selection assertion that is never demonstrated with statistics. The two-channel model is fit to the population data, and its superior fit is then used as evidence that the second channel exists. This is circular in the absence of a formal comparison against a simpler model. The reader identified the LZBL hopping approximation as the weakest assumption, and that is a legitimate concern about the underlying dynamics. But the more load-bearing issue for the central claim is the missing alternative-model comparison, because even if LZBL were exact, the kinetic-fit argument would still be unsubstantiated. My proposed concrete test directly targets this gap. It would also address the related concern about the minor-channel lifetime being extrapolated beyond the 100 fs simulation window. If the two-channel model is statistically preferred and the lifetime is stable, the claim would be considerably stronger. Until such a test is reported, the verdict should remain conditional, with the condition being the provision of explicit model-comparison evidence. I therefore do not change the reader's verdict, but I sharpen the specific condition needed for acceptance.","tokens_in":15734,"tokens_out":4385,"duration_ms":39650,"concrete_test":"Refit the population curves in Fig. 4 with the model of Eq. 3 constrained to alpha=1 (no minor channel) and compare goodness-of-fit to the full model using an information criterion (AIC/BIC) or a likelihood-ratio test. Also, refit the full model using only the first 60 or 80 fs of data; if the fitted tau_b changes by more than its error, the minor-channel lifetime is an extrapolation artifact. If the single-channel model is statistically disfavored (e.g., delta-BIC > 10) and tau_b is stable under truncation, the pathway support would be substantially strengthened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that kinetic fits support the sigma-pi*/S0 pathway depends on the assertion in Sec. III.D that a high-quality fit is achievable \"only when the branching in S1 is explicitly included.\" This is a model-selection claim, but the paper provides no quantitative comparison: no chi-squared, AIC/BIC, residual analysis, or cross-validation for a one-channel (alpha=1) vs two-channel (Eq. 3) model. Since the two-channel model introduces additional parameters (P_LZ^b, T_b, alpha), it will generally fit better even if the second channel is absent, especially when the data extend only to 100 fs while the fitted minor-channel lifetime is ~222 fs (Table IV). The 16% branch fraction is not an observed product yield within the simulation window (the direct H2-loss yield is only 2% at 100 fs); it is a fitted extrapolation. Thus the claimed \"independent validation\" of the pathway is circular: the model assumes the pathway and then the fit \"confirms\" it. Additionally, the power-law description is contradicted by Eqs. 2-3: with N(t) linear in t, (1-P_LZ)^N is an exponential decay in t, not a power law.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops machine-learned interatomic potentials (MLIPs) targeting XMCQDPT2/SA(3)-CASSCF(12,12) accuracy for the lowest three singlet states of the methaniminium cation, using transfer learning, Δ-learning, and multi-state architectures. These potentials are used in Landau–Zener surface-hopping (LZBL) dynamics initiated from S2, and an ensemble uncertainty weighting is introduced to correct state populations. The authors report that transfer learning improves test-set errors, that MECI geometries are reproduced within ~0.1 Å, and that uncertainty corrections bring different MLIP models into closer agreement. They also introduce a 'wavepacket oscillation model' with parameters P_LZ and T, fit to the simulated population traces, from which channel-specific lifetimes are extracted. The paper claims that these kinetic fits require branching in S1 and thereby support the existence of a σπ*/S0 conical-intersection pathway for direct H2 loss.","tokens_in":16258,"tokens_out":3687,"duration_ms":35451,"significance":"If the MLIP methodology and the kinetic interpretation were both fully validated, the work would be a valuable contribution: it demonstrates a practical route to XMCQDPT2-fidelity excited-state dynamics with a large active space, and it proposes a transparent bridge between nonadiabatic surface hopping and classical kinetics. The transfer-learning gains (Table I) and MECI validation (Table II) are concrete and support the MLIP component. The uncertainty-weighting procedure (Eq. 1) is a useful diagnostic and the observation that it reconciles models is interesting. However, the kinetic model at the center of the paper's headline claims is internally inconsistent and its evidential basis for the σπ*/S0 pathway is not statistically established. These issues are load-bearing for the concluding claims about channel-specific lifetimes and pathway validation.","major_comments":[{"comment":"The lifetime extraction is circular. P_LZ and T are fitted to the simulated population curves in Fig. 4, and Eq. (9) then defines τ = T/(2P_LZ). Reporting these as 'extracted lifetimes' is a restatement of fit parameters, not an independent prediction. The only model-free observables are the average hopping times in Table III; the paper itself notes discrepancies between these and the fitted lifetimes. To support the claimed predictive character, the authors should either fit P_LZ and T from ab initio Landau–Zener parameters and oscillation frequencies and then compare with the simulated populations, or clearly label the results as a descriptive fit and provide goodness-of-fit statistics and parameter uncertainties.","section":"Sec. III.D, Eqs. (2)–(9)"},{"comment":"The 'power-law' label contradicts the model equations. With constant P_LZ and N(t) linear in t (Eqs. 6–7), (1−P_LZ)^{N(t)} = exp[N(t) ln(1−P_LZ)] is an exponential decay, not a power law. Furthermore, the geometric distribution with constant P_LZ is memoryless; the statement that 'each failed hopping attempt increases the probability of successful hopping' is inconsistent with the model's own assumption of independent passages. The terminology appears throughout the abstract and conclusions. The model should be renamed, or its equations modified (e.g., with a passage-dependent P_LZ) to genuinely produce nonexponential decay.","section":"Sec. III.D, Eqs. (2)–(7)"},{"comment":"The claim that a high-quality fit requires explicit S1 branching is not supported by any quantitative model comparison. No χ², AIC/BIC, residual analysis, or cross-validation is provided for a one-channel (α=1) versus two-channel (Eq. 3) model. Since the two-channel model adds three parameters (P_LZ^b, T_b, α), it will generally fit better even if the second channel is spurious. Moreover, the minor-channel lifetime of 222 fs greatly exceeds the 100 fs simulation window, and the 16% branch fraction is a fitted extrapolation, not an observed product yield (direct H2 loss is 2% at 100 fs). Please report a formal comparison against the single-channel model and, ideally, identify the σπ*/S0 pathway directly from trajectory geometries rather than from a fitted branch.","section":"Sec. III.D, Table IV"}],"minor_comments":[{"comment":"The text states 'Landau-Zener probability P_LZ^(b)=0.60' for the dominant channel; this should be P_LZ^(a)=0.60, consistent with Table IV.","section":"Sec. III.D, text near Table IV"},{"comment":"Typo: 'photodissotiation' should be 'photodissociation'; also group 7 contains 'disostiation'.","section":"Methods II.G heading"},{"comment":"Grammar: 'Figures S5 and S6 shows' should be 'show'.","section":"Fig. S5/S6 reference"},{"comment":"The statement 'available from the corresponding author upon reasonable request' is less reproducible than depositing the trained models and simulation data in a public repository; consider sharing at least the key training sets and fitted kinetic parameters.","section":"Data Availability"},{"comment":"Ref. 38 is a companion arXiv preprint; if it is under review, this should be stated explicitly. The claim that metrics are 'competitive with X-MACE' would benefit from a direct table or figure.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The MLIP and uncertainty-correction parts of the paper are solid and likely publishable after appropriate framing. The kinetic-model issues are substantive: the 'power-law' framework is actually exponential, and the pathway-support claim rests on an unquantified model-selection argument with a circular extraction of lifetimes. These need to be fixed or substantially softened before the paper can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear X,\n\nThe MLIP part of this paper is genuinely useful and I think the transfer-learning strategy is the real contribution. They train MACE models on CASSCF data and fine-tune on a compact set of ~1,450 XMCQDPT2 points, and they show consistent improvements in test-set errors over random initialization, multi-state, and Δ-learning. The MECI validation is also convincing: the best SS-TL model reproduces the three crossing geometries within 0.1 Å and 0.2 eV. The dynamics results, 600 trajectories with an ensemble-uncertainty correction that pulls different training protocols into closer agreement, are worth reporting. I'd cite that part.\n\nThe soft spot is the wavepacket oscillation model and the way it is used to validate the sigma-pi*/S0 channel. The model itself is fine as a parametrization, but the paper calls it a 'power-law kinetics framework' and that is wrong: in Eqs. 2–3 the number of attempts N(t) is linear in t, so (1 - P_LZ)^N is an exponential decay in time, not a power law. The fitted P_LZ and T in Table IV are fit to the same population curves they're said to reproduce, so the claim of 'accurate reproduction' is a statement about the quality of the fit, not an independent validation. And the crucial claim that a high-quality fit requires explicitly including the S1 branching is supported only by assertion: no statistical comparison (chi-squared, AIC/BIC, residuals) is shown against a single-channel model. Since the two-channel model has more free parameters (P^b, T_b, alpha), it will fit better by construction. So the 16% branch fraction and ~222 fs lifetime are extrapolated parameters, not observed yields within the 100 fs window. The direct H2-loss yield at 2% is consistent with prior CASSCF, but that does not independently confirm the kinetic model's pathway assignment.\n\nI want to be fair: the LZBL protocol without nonadiabatic couplings is a legitimate choice and the paper transparently reports the limitations of the kinetic model as a fitting framework. The central mechanistic claim, however, is not established by the evidence presented. The paper deserves peer review because the MLIP methodology and dynamics results are solid and the kinetic modeling issue needs a rigorous statistical treatment. A serious referee should ask for the code and data, a corrected kinetic model framing, and an explicit model-selection analysis before the pathway-support claim can be accepted.\n\nMy advice: send it out, but expect the kinetic-model claims to be heavily revised.\n\nBest","headline":"Solid transfer-learned MLIP work for CH2NH2+ at XMCQDPT2 level, but the kinetic 'power-law' model is a fitted exponential in disguise and the claimed independent validation of the sigma-pi*/S0 pathway does not hold up.","tokens_in":16717,"tokens_out":2331,"would_cite":true,"duration_ms":20415,"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":"The paper claims that transfer-learned neural-network potentials fine-tuned on roughly 1,450 XMCQDPT2 points can drive nonadiabatic dynamics that expose a slow ~200 fs photodissociation channel in the methaniminium cation.","keywords":["machine learning interatomic potentials","nonadiabatic dynamics","conical intersections","transfer learning","XMCQDPT2","Landau-Zener surface hopping","methaniminium cation","power-law kinetics"],"falsifier":"Run the same S2-initiated dynamics with a surface-hopping method that uses explicit nonadiabatic coupling vectors at the XMCQDPT2 level and compare mean hopping times and S1 branching; if the 28.4 fs S1 time or the 84:16 split changes substantially, the Landau-Zener approximation is at fault. Alternatively, an ultrafast pump-probe experiment on CH2NH2+ that resolves a ~200 fs component in the S1 population would confirm the minor sigma-pi*/S0 channel.","tokens_in":15664,"feed_emoji":"🧪","tokens_out":7885,"duration_ms":63817,"temperature":0.7,"pith_summary":"The paper sets out to show that machine-learning potentials trained on a deliberately small set of high-level quantum-chemistry data can reproduce the full excited-state decay of the methaniminium cation, a molecule that models protonated Schiff bases and is relevant to Titan's atmosphere. The authors fine-tune an equivariant neural-network potential from a cheaper CASSCF description to XMCQDPT2 accuracy using about 1,450 reference points, then run 600-trajectory Landau-Zener surface-hopping dynamics from the S2 state. They also introduce a wavepacket oscillation model in which repeated attempts to cross a conical intersection produce power-law decay, letting them extract state-specific lifetimes rather than fit exponentials. Fits to the simulated populations require two distinct S1-to-S0 routes, with lifetimes of 25 fs and ~200 fs; the slow route supports a recently proposed photochemical pathway through a sigma-pi*/S0 conical intersection. If the approach holds, high-level nonadiabatic dynamics becomes affordable for other photochemical systems.","feed_headline":"High-level photodynamics from 1,450 quantum-chemistry points","feed_subtitle":"Transfer-learned potentials map CH2NH2+ decay and expose a rare 200-fs H2-loss route via a conical intersection.","key_machinery":"The load-bearing object is the multi-passage survival cascade model: a wavepacket oscillates near a conical intersection with period T, crossing the intersection twice per cycle, and each crossing succeeds with Landau-Zener probability P_LZ; the chance of surviving N attempts is (1-P_LZ)^N, which produces power-law kinetics and a mean lifetime T/(2P_LZ). This converts a single quantum transition probability into a classical rate constant without assuming exponential decay. On the machine-learning side, the enabling mechanism is transfer learning: a cheaper pretrained model supplies the representation, and a small fine-tuning set transfers it to XMCQDPT2 fidelity; an ensemble's energy-standar","core_discovery":"On the paper's own terms, the central discovery is that a single-state transfer-learning protocol—pretrain an equivariant neural network on roughly 50,000 CASSCF geometries, then fine-tune on ~1,450 XMCQDPT2 points—yields potential energy surfaces that reproduce XMCQDPT2 reference conical intersections to within ~0.2 eV in energy and ~0.1 Å in geometry. These potentials drive Landau-Zener surface-hopping dynamics that map all decay channels after S2 excitation, giving uncertainty-corrected mean hopping times of 5.0 fs for S2→S1 and 28.4 fs for S1→S0, and a product distribution dominated by CN cleavage with ~2% direct H2 loss. Fitting the populations with the new wavepacket oscillation model","pith_inferences":["If the transfer-learning recipe generalizes, the same ~1,500-point fine-tuning could bring XMCQDPT2-fidelity dynamics to larger protonated Schiff bases and other chromophores where cheaper CASSCF datasets already exist.","A concrete testable prediction follows from the 16% branching but only 2% yield at 100 fs: trajectories longer than 100 fs should show delayed H2-loss products appearing on the ~200 fs timescale.","The model's lifetime formula T/(2P_LZ) suggests that experimental pump-probe transients could be fit directly with power laws to estimate single-passage Landau-Zener probabilities and oscillation periods, rather than phenomenological exponentials.","The mismatch between the 13.6 fs fitted S2 lifetime and the 5.0 fs trajectory-averaged hopping time hints that the constant-period, constant-probability assumption breaks down near the Franck-Condon region; a time-dependent version of the cascade model would reconcile the two."],"forward_implications":["High-level multi-reference nonadiabatic dynamics can be run on the fly with roughly 1,450 XMCQDPT2 reference points, as long as a cheaper pretrained potential is available for transfer learning.","Ensemble uncertainty weighting changes real conclusions: it shortens the S2→S1 mean hopping time from 14.6 fs to 5.0 fs and brings models trained differently into agreement, so it should be a standard part of ML-driven dynamics.","Exponential decay is not the right default for repeated conical-intersection passages; lifetimes extracted from single-exponential fits to ultrafast populations need to be revisited under the multi-passage picture.","The S1 population decay cannot be fit without splitting it into two routes; the minor sigma-pi*/S0 route has a ~200 fs lifetime and 16% branching, supporting the recently discovered photochemical pathway.","Branching ratios among photodissociation products are robust to the choice of MLIP, with CN cleavage dominant and direct H2 loss at ~2% within 100 fs."],"fun_headline_variants":["Transfer learning unlocks full photodynamics from 1,450 points","Rare H2-loss route from transfer-learned potentials","Transfer-learned potentials map all decay channels after S2","Uncertainty corrections align MLIP dynamics with quantum rates"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire dynamical picture rests on Landau-Zener single-passage probabilities—computed without nonadiabatic coupling vectors—fully determining whether a trajectory hops, and the kinetic model then assumes those probabilities, the oscillation periods, and the independence of repeated passages all stay constant.","fun_headline_variants_meta":{"raw":{"variants":["Transfer learning unlocks full photodynamics from 1,450 points","Rare H2-loss route from transfer-learned potentials","Transfer-learned potentials map all decay channels after S2","Uncertainty corrections align MLIP dynamics with quantum rates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000763,"raw_usage":{"total_tokens":3298,"prompt_tokens":897,"completion_tokens":2401,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":641,"completion_tokens_details":{"reasoning_tokens":2333}},"tokens_in":641,"tokens_out":2401,"duration_ms":16789,"temperature":1.0,"reasoning_tokens":2333,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T17:54:00.976294+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same S2-initiated dynamics with a surface-hopping method that uses explicit nonadiabatic coupling vectors at the XMCQDPT2 level and compare mean hopping times and S1 branching; if the 28.4 fs S1 time or the 84:16 split changes substantially, the Landau-Zener approximation is at fault. Alternatively, an ultrafast pump-probe experiment on CH2NH2+ that resolves a ~200 fs component in the S1 population would confirm the minor sigma-pi*/S0 channel.","supporting_citations":[],"review_version":1}