{"id":"356cb430-7ab9-43b3-a98e-cc3c8608e4fd","arxiv_id":"2607.23208","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Stoichiometry-preserving ionic clusters plus multi-task fine-tuning of a pretrained atomistic potential predict detonation velocities of perovskite-type energetic materials from 25 labels, transferring to three newly synthesized ABX4 members at 92 m/s MAE.","lead":"Machine-learning screening of multi-ionic explosives is shown to work from small non-periodic clusters of their component ions, before a full crystal structure is known. A model trained on only 25 perovskite-type energetic materials guided the synthesis of three new materials and matched their Kamlet–Jacobs detonation velocities to within 92 m/s on average.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"OOD-new 'concordance' is measured against in-house K–J labels whose product-rule sensitivity is the same order as the claimed MAE; quantitative screening claim is not yet robust.","rationale":"The reader's weakest assumption identifies exactly the concern I find most load-bearing: the OOD-new reference velocities are computed in-house under a CO2-priority K–J rule, and the SI admits that a CO-priority variant shifts D by several hundred m·s−1. Since the claimed three-point MAE is 92 m·s−1 and the largest individual error is 219 m·s−1, an unquantified rule sensitivity of this magnitude undermines the quantitative 'concordance' claim. This is not a problem internal to the model's logic; it is an external-validation-label problem, but it is decisive for the strength of the headline claim because n=3 carries the whole pre-synthesis generalisation story.\n\nI still credit the paper's internal consistency and its several genuinely supportive diagnostics: the asymmetric cross-representation test (cluster-trained 405 m·s−1 on crystals vs crystal-trained 1314 m·s−1 on clusters), the DAP-core pretrained holdout, the perturbation ladder, and the transparent reporting of the DAI-10.5-40.5 failure all indicate the authors are not hiding the method's limitations. The concern I raise does not require rejecting the paper; it requires treating the 92 m·s−1 number as provisional until the reference-label product-rule sensitivity is quantified. That is precisely a CONDITIONAL posture, not an ACCEPT or REJECT. Since the reader already reached CONDITIONAL on essentially these grounds, my stress-test does not change the verdict.\n\nIf the concrete test shows that the CO-priority variant changes the MAE or ranking substantially, then the paper's strongest quantitative claim should be softened and the screening claim should be presented as rank-level rather than error-level. If the test shows only a small shift, the concern is resolved and the CONDITIONAL verdict could later be upgraded toward ACCEPT. Until then, UNCHANGED is the right recommendation.","tokens_in":37563,"tokens_out":4229,"duration_ms":44927,"concrete_test":"Recompute the K–J reference V_det for PEP, PEP-M, PEP-H, and DEP under the CO-priority product rule using the same refined densities and DFT heats of formation, and also under an independently implemented standard K–J product inventory (e.g., H2O/CO2 with no HCl priority, or a published K–J code). Then re-evaluate the five-fold MT-FT ensemble predictions from Table 1 against each alternative reference set, reporting per-material errors and the rank order. If the MAE remains ≤ ~100 m·s−1 and the PEP > PEP-H > PEP-M ranking is preserved, the 92 m·s−1 concordance is robust. If the MAE shifts to several hundred m·s−1 or the ranking changes, the central quantitative screening claim is not supported at the claimed precision and should be reframed as coarse ranking-level only.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative evidence for synthesis-facing screening is the three-point OOD-new MAE of 92 m·s−1 (Table 1) and the template-based ~100 m·s−1 deviation. But those reference velocities are not experimental detonation velocities: they are K–J calculations performed by the authors under a fixed CO2-priority reaction-balance rule (SI, 'Kamlet–Jacobs calculation protocol'). The SI explicitly states that a CO-priority variant shifts D by 'several hundred m·s−1 at the same ρ.' That shift is the same order of magnitude as the claimed MAE and comparable to the spread of the three predictions (8699–8871 m·s−1) around the references (8729–9090 m·s−1). PEP, the largest error (219 m·s−1), could easily move by several hundred m·s−1 under an alternative product rule, changing the reported 'concordance.' Because the training-set K–J labels come from heterogeneous literature sources with their own thermochemical conventions, the model may have implicitly learned a particular product-rule bias; comparing it to in-house CO2-priority references could make the agreement look better than it is. This is load-bearing because the three wet-lab-synthesized ABX4 materials are the only truly external test of the pre-synthesis screening claim, and their labels carry an unquantified, rule-dependent uncertainty of the same size as the error being claimed. The paper's own Limitations paragraph acknowledges the K–J anchor but does not quantify or bound the product-rule sensitivity. This concern does not by itself invalidate the approach—the model does place all three materials in a plausible high-performance regime—but it does mean the headline 92 m·s−1 MAE is not a stable quantitative statement until the reference-label sensitivity is resolved.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a two-stage workflow for few-shot prediction of detonation velocity in multi-ionic integrated energetic materials (MIXs). The first stage represents each candidate material as a non-periodic, stoichiometry-preserved formula-unit cluster built from ionic building blocks, thereby avoiding full crystal-structure prediction. The second stage adapts a pretrained MLIP backbone (DeepEMs-LAM) to sparse Kamlet–Jacobs (K–J) detonation-velocity labels via multi-task fine-tuning (MT-FT), retaining an energy–force head as physical regularization. The surrogate is evaluated on 25 structurally curated perovskite-type energetic materials (PEMs), giving a five-fold cross-validated MAE of 273 m/s with per-fold values and bootstrap confidence intervals. Additional analyses include literature out-of-distribution holdouts (four materials, MAE 35 m/s), representation probes, and wet-lab synthesis of three new ABX4 compounds (PEP, PEP-M, PEP-H) with an unseen ethylenediammonium B-site cation; against in-house K–J references the surrogate reportedly achieves a three-point MAE of 92 m/s, and a template-based pre-synthesis variant gives roughly 100 m/s deviation.","tokens_in":37864,"tokens_out":3041,"duration_ms":33277,"significance":"If the central result holds, the paper offers a practical screening strategy for data-scarce, structurally modular ionic energetic materials, with the notable advantage that candidates can be evaluated before a resolved periodic crystal structure is available. The work is commendably transparent: the cross-validation is reported with per-fold values, cluster-variant sensitivity is quantified, several adaptation ablations are compared, and the DAI-1_0.5_4_0.5 failure case is documented rather than hidden. The representation diagnostics (density probing, site-resolved embeddings, perturbation ladders) are informative and support the claim that the cluster representation retains physically meaningful local information. However, the central quantitative claim for the three new ABX4 materials is measured against in-house K–J labels whose product-rule sensitivity is the same order of magnitude as the claimed error; this is a load-bearing issue that must be addressed before the screening claim can be accepted.","major_comments":[{"comment":"The three-point OOD-new MAE of 92 m·s−1 is computed against K–J reference velocities generated under a fixed CO2-priority reaction-balance rule. The SI itself states that a CO-priority variant shifts D by 'several hundred m·s−1 at the same ρ.' This uncertainty is the same order of magnitude as the claimed MAE and is comparable to the largest per-material error (PEP, 219 m·s−1). Because the training labels are also drawn from heterogeneous literature sources with their own thermochemical conventions, the model may have implicitly learned a particular product-rule bias, making agreement with this specific in-house rule appear better than it is. Please quantify the sensitivity of all three references to alternative plausible product rules, propagate this uncertainty into the reported MAE, or explicitly reframe the three-point comparison as a ranking-level validation only.","section":"SI, 'Kamlet–Jacobs calculation protocol' (with Table 1)"},{"comment":"The pre-synthesis screening claim rests on the DAP-4 template scaffold, with a reported mean absolute deviation of approximately 100 m·s−1 for only three materials. The SI topology discussion acknowledges that the ABX4 packing differs substantially from the ABX3 perovskite cage, yet no per-material template errors or template-variant statistics are provided. Combined with the product-rule uncertainty above, this does not yet substantiate that the template-based clusters are a valid surrogate for unresolved ABX4 packing. Please report per-material template deviations, preferably against a range of K–J product rules, and discuss under which packing conditions the DAP-4 scaffold is expected to fail.","section":"Results, 'Synthesis validation in the H2en2+-based ABX4 branch' (Extended Data Fig. 4)"},{"comment":"The in-domain cross-validation result (MAE 273 m/s, CI [201, 350]) is honestly reported, but the comparison against the global-mean predictor (690 m/s) is a weak baseline. The more relevant baselines—composition-only inputs and periodic-crystal inputs—are only partially reported (e.g., the periodic-crystal control is in Extended Data Fig. 2 and the composition-only baseline appears in probes). For the central representation claim, please report the matched-input CV MAE for a composition-only stoichiometric model and for the periodic-crystal model on the same fold assignments, so that the added value of the cluster representation over composition-level inputs is quantified directly on the same splits.","section":"Methods, 'Data curation' and Extended Data Table 2"}],"minor_comments":[{"comment":"The Kamlet–Jacobs equation is garbled in the main text (Greek symbols and subscripts are corrupted). Please typeset the equation with standard notation and define all variables explicitly.","section":"Eq. (1)"},{"comment":"The phrase 'experimentally derived Kamlet–Jacobs detonation velocities' is ambiguous: K–J values are empirical estimates, not direct experimental measurements. Please say 'K–J velocities computed from experimentally refined crystal structures and thermochemical inputs' or similar.","section":"Abstract and Table S2"},{"comment":"The notation DAI-1_0.5_4_0.5 is typeset inconsistently and may be confused with a subscripted formula. Please define this material name in a glossary and use a consistent format.","section":"Throughout"},{"comment":"The reference list contains self-citations that appear in both main-text and supplementary lists (e.g., refs 26, 36, 41). Please unify the numbering or clearly separate primary and supplementary citations to avoid confusion.","section":"References"},{"comment":"The panel compares the new materials with RDX, HMX, and CL-20, but the K–J references for the new materials and the literature values for benchmarks may use different conventions. Please state the source of the benchmark velocities in the caption, or move this comparison to the SI.","section":"Figure 5d"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is methodologically interesting and the experimental synthesis of three new ABX4 compounds is a genuine contribution. The main concern is not the use of K–J endpoints per se, but the fact that the three-point validation is measured against in-house labels with a product-rule sensitivity of the same magnitude as the reported error. A revision that quantifies this uncertainty, or appropriately softens the claim, would make the paper publishable. I do not see grounds for rejection: the central idea is defensible and the authors have been unusually transparent about limitations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should read this one for the representation idea, not for the quantitative screening claim. The stoichiometric vacancy cluster — non-periodic, charge-balanced, preserving A/B/X site identity — is genuinely new and well motivated for multi-ionic materials where crystal-structure prediction is the bottleneck. The wet-lab synthesis of three new ABX4 materials with an unseen B-site cation is real external validation, and the paper does several things right: per-fold CV numbers with bootstrap CIs, split-seed robustness, a documented failure case (DAI-1_0.5_4_0.5) that is honestly excluded, and an asymmetric cross-representation test showing cluster-trained models degrade gracefully on crystals while crystal-trained models fail badly on clusters. That asymmetry is a strong argument that the cluster representation captures useful local signal rather than just memorizing geometry.\n\nThe soft spot is where the stress-test note lands. The three-point OOD-new MAE of 92 m/s is measured against Kamlet–Jacobs reference velocities computed in-house under a fixed CO2-priority reaction-balance rule. The SI says a CO-priority variant shifts D by several hundred m/s at the same density. That is the same order of magnitude as the claimed agreement, and PEP's 219 m/s error could easily be absorbed or inflated by that choice. So the headline number is not a stable quantitative statement until the reference-label sensitivity is bounded. This does not sink the paper — the model places all three materials in the right high-performance regime — but it should push the authors to either compute K–J under multiple product rules and report the range, or reframe the OOD-new result as qualitative concordance.\n\nMinor concerns: the OOD aggregates are n=4 and n=3; the central dataset table is gated until publication; no commit hash anchors the code; and the production pipeline leans heavily on self-cited prior work. None of those is disqualifying, and the Limitations paragraph is unusually candid — it says the ABX4 validation is anchored on K–J rather than measured detonation velocities, which is exactly the right caveat.\n\nI would send this to peer review rather than desk-reject it. The representation is novel, the experiments are real, and the failure analysis is a model of transparency. But the referees should push for a quantified treatment of the product-rule ambiguity before the 92 m/s claim goes public. The paper is worth engaging with, and I'd cite it for the cluster representation even if I treat the quantitative transfer claims with caution.","headline":"Genuinely new cluster representation and honest reporting, but the 92 m/s 'concordance' rests on in-house K–J labels whose product-rule sensitivity is the same size as the claimed error.","tokens_in":38576,"tokens_out":1183,"would_cite":true,"duration_ms":14545,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Detonation velocity of multi-ionic explosives can be predicted from a stoichiometric ionic cluster alone, before a crystal structure is known.","keywords":["stoichiometric cluster representation","multi-ionic integrated explosives","Kamlet–Jacobs detonation velocity","multi-task fine-tuning","machine-learned interatomic potentials","perovskite-type energetic materials","few-shot property prediction","pre-synthesis screening"],"falsifier":"Recompute the K–J reference velocities for PEP, PEP-M, and PEP-H using the CO-priority product-balance rule; if the surrogate's errors against those labels are several hundred m/s larger than 92 m/s, the claimed concordance depends on a product-rule choice rather than on the cluster representation.","tokens_in":37276,"feed_emoji":"💥","tokens_out":5937,"duration_ms":52464,"temperature":0.7,"pith_summary":"The paper sets out to show that the detonation velocity of a multi-ionic energetic material—a property usually thought to require the fully resolved crystal—can be predicted from a non-periodic, charge-balanced formula-unit cluster that preserves the identities and mutual orientation of the A, B, and X ionic building blocks. The proposed mechanism is to fine-tune a pretrained machine-learned interatomic potential with a multi-task objective: one head reads the sparse Kamlet–Jacobs detonation-velocity labels, while the inherited energy–force head keeps acting as physical regularization. On a curated set of 25 perovskite-type energetic materials, cross-validated predictions reach a mean absolute error of 273 m/s, about a 60% reduction over a composition-blind baseline. The same surrogate, without retraining, predicts three newly synthesized H2en2+-based ABX4 materials—an unseen stoichiometry and an unseen B-site cation—with a mean absolute error of 92 m/s against Kamlet–Jacobs references, and a template-based variant built on a DAP-4 scaffold stays near 100 m/s deviation. The authors conclude that the stoichiometric-cluster level is a usable pre-synthesis screening representation for data-scarce multi-ionic materials, provided the candidates preserve recognizable building blocks and local packing compatibility.","feed_headline":"Forecast detonation velocity from ionic clusters, no crystal needed","feed_subtitle":"Fine-tuning a pretrained atomistic model on 25 perovskite explosives transfers to new ABX4 salts at 92 m/s error.","key_machinery":"The load-bearing mechanism is the stoichiometric ionic-cluster representation: each candidate material is reduced to a non-periodic, charge-neutral formula-unit cluster that preserves the chemical identity of A, B, and X building blocks, their mutual orientation, and a coarse packing motif, by removing one formula unit from a supercell and wrapping it in a 100 Å box without periodic boundary conditions. This cluster is fed to a pretrained DPA3-based MLIP backbone (the authors' DeepEMs-LAM) that is fine-tuned with multi-task fine-tuning (MT-FT): one head predicts Kamlet–Jacobs detonation velocity, while the inherited energy–force head is kept as physical regularization for the sparse labels.","core_discovery":"The central discovery is that the stoichiometric-cluster level of description—a charge-balanced, non-periodic formula-unit cluster built from a known ionic packing template—retains enough relational information about how A-, B-, and X-site ions co-assemble that detonation velocity can be predicted before a full periodic crystal structure is resolved. The authors show that a pretrained energetic-material MLIP backbone, adapted by multi-task fine-tuning with an energy–force head as physical regularization, learns descriptors that linearly encode crystal density and oxygen balance, organize atom embeddings by ionic site, and respond strongly only when fragment identity or local organization is","pith_inferences":["If the cluster-level transfer holds beyond the three ABX4 examples, the same workflow could be applied to energetic cocrystals and coformer-ratio variants, where the representation bottleneck is identical (which molecular ions pair and how they pack) — a testable extension the authors mention as an outlook.","The dependence of the validation labels on the CO2-priority reaction-balance rule suggests that the claimed 92 m/s concordance should be re-read as a concordance with one specific thermochemical convention; an independent detonation-velocity measurement or a protocol-averaged label would be a stronger anchor.","The finding that the surrogate's descriptors linearly encode crystal density without density supervision hints that the energy–force head is effectively teaching the model an equation-of-state prior; one could test this by ablating the force head and checking whether density readability and OOD transfer collapse together.","A minimal falsification test: synthesize or simulate an ABX4 material with a genuinely new A-site family (not one of the six in training) and check whether template-based prediction degrades beyond the ~100 m/s template deviation seen here."],"forward_implications":["For multi-ionic materials whose properties emerge from local co-assembly, screening can proceed from stoichiometric clusters built on a related template, before a single-crystal structure is resolved.","The same MT-FT recipe—pretrained atomistic backbone plus energy–force regularization—can be reused for other sparse property labels in the MIX family (e.g., detonation pressure, oxygen balance) without additional crystal-structure prediction.","The failure on the ordered double-perovskite DAI-1_0.5_4_0.5 (overestimate ~1300 m/s with small ensemble spread) shows that cluster-based screening is reliable only within the family of recognizable A/B/X building blocks; materials combining site patterns absent from training need domain checks.","Representation probes imply that the surrogate's usefulness degrades gracefully under shell-preserving perturbations (template substitution, rigid motions, distance stretches), so template-built clusters can stand in for resolved clusters with roughly 100 m/s added deviation.","Because the crystal-trained model fails badly on cluster inputs while the cluster-trained model degrades only mildly on crystal inputs, cluster training is the deployable direction for pre-synthesis screening."],"fun_headline_variants":["Ionic clusters forecast detonation velocity without full crystal","Pretrained model fine-tuned on 25 explosives forecasts new salts","Cluster learning skips crystal structure for explosive screening","Stoichiometric clusters predict detonation speed from ions alone","No crystal? No problem: ionic clusters predict detonation velocity"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the Kamlet–Jacobs reference velocities—computed in-house under a fixed CO2-priority product-balance rule—are correct enough to judge the surrogate, because switching to a CO-priority rule shifts those references by several hundred m/s, the same order as the claimed 92 m/s error.","fun_headline_variants_meta":{"raw":{"variants":["Ionic clusters forecast detonation velocity without full crystal","Pretrained model fine-tuned on 25 explosives forecasts new salts","Cluster learning skips crystal structure for explosive screening","Stoichiometric clusters predict detonation speed from ions alone","No crystal? No problem: ionic clusters predict detonation velocity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000587,"raw_usage":{"total_tokens":2646,"prompt_tokens":847,"completion_tokens":1799,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":591,"completion_tokens_details":{"reasoning_tokens":1731}},"tokens_in":591,"tokens_out":1799,"duration_ms":13856,"temperature":1.0,"reasoning_tokens":1731,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:17:13.385154+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the K–J reference velocities for PEP, PEP-M, and PEP-H using the CO-priority product-balance rule; if the surrogate's errors against those labels are several hundred m/s larger than 92 m/s, the claimed concordance depends on a product-rule choice rather than on the cluster representation.","supporting_citations":[],"review_version":1}