{"id":"976ad7ef-d15d-46b0-8408-fefe6a9c0511","arxiv_id":"2608.07401","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"DynaCrys couples space-group evolution with Wyckoff occupations and elements in a symbolic diffusion model and reports best-in-class stable-unique-novel rates on MP-20.","lead":"A new AI model called DynaCrys generates candidate crystal structures by letting the crystal's symmetry group change during generation instead of fixing it at the start. It reports the best balance of stable, unique, and novel crystals across two separate machine-learning evaluation engines.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Best-in-class claim is only as strong as the single-run, released-sample baselines; with WyckoffDiff absent and deltas near 0.4–0.9 pp on SUN@0, a fresh multi-seed baseline rerun is needed.","rationale":"The reader's weakest assumption is the fair representativeness of the baseline comparison, and that is exactly where the central claim is least secure. The comparison is asymmetric: the authors re-ran only DynaCrys and SGEquiDiff, while DiffCSP++, SymmCD, and WyFormer are represented by released sample sets that may come from different model versions, sampling schedules, or post-hoc filters. Appendix G.3 explicitly says the other baselines are single evaluated sets, so their run-to-run variance is unknown. The strict-stability differences are small (0.4–0.9 percentage points), making this a load-bearing condition rather than a cosmetic issue. WyckoffDiff's absence is material because the headline is about symmetry-aware crystal generation, WyckoffDiff's exact niche. I did not find a more serious internal flaw: Proposition 1's exactness argument is sound as written, the two-engine evaluation is a genuine strength, and the limitations section is candid about the MLIP stability proxies. The proposed test—multi-seed reruns of all baselines, including WyckoffDiff, under the same downstream protocol—would settle whether the ordering survives. Thus the reader's CONDITIONAL verdict stands, and no change is needed.","tokens_in":22017,"tokens_out":18986,"duration_ms":173395,"concrete_test":"Re-run each baseline from its official code and released checkpoint (DiffCSP++, SymmCD, WyFormer, SGEquiDiff, and WyckoffDiff) with five independent sampling seeds under the paper's exact downstream protocol (same relaxers, hull snapshot, matcher, and spglib settings), then recompute Table 1 as mean ± SD. If every DynaCrys configuration still exceeds every baseline on both engines at SUN@0 and SSUN@0, the concern is settled; if any baseline mean ties or surpasses a DynaCrys row on either engine, the strongest claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 5.1 states that 'publicly released samples are used when available; SGEquiDiff is sampled from its released checkpoint,' and Appendix G.3 says all baselines except SGEquiDiff are single evaluated sets. The strongest claim—all three DynaCrys configurations beat every baseline on SUN@0 and SSUN@0 under both engines—therefore depends on the released sample sets being representative of each baseline's true performance under the common protocol. The strict-stability deltas are small in absolute terms (e.g., CHGNet SUN@0: 9.39 vs 8.45 for SGEquiDiff; MACE-MP-0: 9.55 vs 9.11), and single-run baselines carry no error bars, so run-to-run variance or a suboptimal released checkpoint could change the ordering. The omission of WyckoffDiff, a directly competing symmetry-aware generator discussed in Section 2, further weakens the 'best-in-class' wording, since the claim is precisely about the symmetry-aware class. The paper's conditional wording is honest, but the headline comparison is not yet definitive.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"DynaCrys is a two-stage generative model for crystals in which the space group is part of the evolving diffusion state. A symbolic diffusion process jointly samples the space group, Wyckoff-row occupations, and elements, using a space-group transition kernel built from maximal translationengleiche and klassengleiche group–subgroup relations and a frozen, pretrained symmetry codebook. A legality-constrained stochastic decoder, whose exactness is stated in Proposition 1, produces a legal protostructure, and a symmetry-constrained geometry diffusion model generates the lattice and free Wyckoff coordinates. The method is evaluated on MP-20 with approximately 10,000 candidates per method, using two independent relaxation-and-evaluation engines (CHGNet and MACE-MP-0), and compared with seven baselines. The paper reports that all three DynaCrys configurations outperform every baseline on SUN@0 and SSUN@0, occupy the top three positions in the explicit-symmetry block at the metastable threshold, achieve low median relaxed hull energies, low relaxation-induced RMSD, and competitive sampling times. Ablations include replacing the learned space-group proposal with a marginal distribution, varying the geometry reverse schedule, and checking symmetry-detection tolerance sensitivity.","tokens_in":22247,"tokens_out":6017,"duration_ms":54657,"significance":"If the empirical claims hold, DynaCrys is a meaningful advance in symmetry-aware crystal generation: it makes the space group a dynamically revised variable rather than a fixed conditioning input, and it provides a principled legality-constrained decoder with a stated exactness guarantee. The evaluation is unusually careful in several respects: two independent MLIP relaxation rulers, a common downstream protocol, approximately 10,000 candidates per method, five-run means with standard deviations for the proposed model and SGEquiDiff, and sensitivity analyses for symmetry detection, geometry steps, and the space-group coupling ablation. These strengths make the comparison substantially more credible than typical crystal-generation benchmarks. However, the headline 'best-in-class' claim is only as strong as the fairness of the baseline comparison, and the paper currently lacks statistical support for some of the small reported deltas and omits at least one direct symmetry-aware competitor. The contribution is therefore promising but needs strengthening before the central benchmark claim can be regarded as definitive.","major_comments":[{"comment":"The central claim that all three DynaCrys configurations outperform every baseline on SUN@0 and SSUN@0 is not yet supported by a significance analysis. All baselines except SGEquiDiff are single evaluated sets, and SGEquiDiff and DynaCrys are each summarized by only five runs. In the MACE-MP-0 panel, for example, DynaCrys-B has SUN@0 9.26±0.40 versus SGEquiDiff 9.11±0.23, so the mean gap is smaller than the reported dispersion; the same is true for DynaCrys-C under CHGNet (9.18±0.30 versus 8.45±0.34). The paper reports no paired or unpaired significance tests, and the deltas against the closest baseline are on the order of 0.1–0.9 percentage points. Because the 'best-in-class' wording depends on these orderings, the authors should either provide multi-seed baseline evaluations or formal significance tests, and they should describe how representative the released baseline sample sets are of each method's performance under the common protocol.","section":"§5.1, Table 1, Appendix G.3, Appendix H.1"},{"comment":"WyckoffDiff is discussed in Related Work as a symmetry-aware generator that represents Wyckoff occupations with space-group-specific discrete variables, yet it is absent from the evaluation. The headline claim is specifically about being best-in-class among symmetry-aware generative models, so omitting a directly competing method of this class weakens the claim. If WyckoffDiff cannot be evaluated under the common protocol, the authors should state this explicitly and qualify the 'best-in-class' wording to the set of methods actually compared; otherwise they should add it to Table 1.","section":"§2, Table 1"},{"comment":"The exactness statement in Proposition 1 is conditional on acceptance, but the sampling procedure does not specify what the reverse sampler does when all R proposal attempts are rejected. Appendix E.4 says only that the decoder 'reports a rejection to the reverse sampler.' If the reverse process then falls back to an unconstrained sample, stops, or re-noises, the final generated distribution is no longer the claimed constrained target, and the exactness guarantee in Eq. (9) does not cover the actual procedure. The authors should specify the rejection fallback precisely and analyze its effect on the generated distribution, or modify the sampler so that the claimed conditional distribution is realized.","section":"§4.3, Appendix E.4, Eq. (9)"}],"minor_comments":[{"comment":"The text says 'approximately 10,000 candidates per method,' but Appendix G.1 states that WyFormer's released set contains 9,999 structures; a more precise phrasing would avoid implying an exact common budget.","section":"§5.1, Appendix G.1"},{"comment":"The conclusion says DynaCrys models 'lead symmetry-aware generators,' while the abstract says 'best-in-class.' These formulations are not equivalent, and the stronger wording should be supported or softened consistently throughout the paper.","section":"§6"},{"comment":"The inline notation in Figure 1 appears corrupted, for example '𝑥!(#)∈𝒳%&’(%,' and '𝑍!,#𝑍$,#'; these fragments should be replaced with the intended mathematical expressions.","section":"Figure 1"},{"comment":"The model vocabulary contains only 168 of the 230 space-group types. The paper should state prominently that generation is restricted to training-supported space groups and discuss the implications for coverage and for the Eff. SGs comparison in Table 2.","section":"Appendix A.1"},{"comment":"The quality–efficiency comparison includes only five methods, and the timing rows for the other baselines are absent; the sentence 'DynaCrys-small exceeds every baseline in this comparison' should make clear that the 'baselines' here are only those with same-machine timing.","section":"Table 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is well executed and the method is interesting, but the 'best-in-class' benchmark claim needs additional baseline rigor before publication. In particular, the absence of WyckoffDiff and the lack of significance tests or multi-seed baseline runs are likely to be raised by other readers; these should be addressed head-on. The rejection-fallback ambiguity in Appendix E.4 is also a correctness gap that should be fixed rather than left implicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The actual novelty is the dynamic space-group diffusion: G co-evolves as a symbolic variable through a group-subgroup graph instead of being fixed before generation. That is a real departure from SymmCD, WyckoffDiff, SGEquiDiff, and the other static-G generators, and the legality-constrained decoder has a neat exactness argument (Proposition 1, Appendix E) that checks out. I also think the evaluation is well above the usual bar: two independent MLIP rulers, ~10k candidates per method, across-run standard deviations for the multi-run methods, a symmetry-tolerance sensitivity table, a marginal-G ablation that cleanly isolates the mechanism, and a same-machine timing comparison. The authors also state the obvious limitations (MLIP stability proxies, finite hull snapshot) without burying them.\n\nThe soft spot is the one the stress-test note flags. The 'best-in-class' claim rests on comparing five-run means for DynaCrys and SGEquiDiff against single evaluated sets for the other baselines, using released samples. The strict-stability deltas are small—about 0.9 pp over SGEquiDiff on CHGNet SUN@0 and 0.4 pp on MACE—and without baseline error bars or fresh multi-seed reruns, those gaps could shift. WyckoffDiff is discussed in Section 2 and then never evaluated, which matters because the stated claim is specifically about the symmetry-aware class. That said, the authors disclose the single-run baseline limitation in Appendix G.3, and the advantage is consistent across both independent engines, so this is a robustness concern rather than a sign of a broken comparison. The marginal-G ablation is the most convincing piece: replacing the learned space-group proposal with the training marginal drops SUN@0 from 9.28 to 3.01, which directly supports the dynamic mechanism.\n\nOverall, the central mechanism is credible and well-ablated; the headline superiority is plausible but not yet definitive. I would send this to peer review and, as a referee, ask for a fresh multi-seed rerun of at least SGEquiDiff and WyckoffDiff plus code and data release. The paper deserves referee time, and if I were working on symmetry-aware generation I would cite it for the dynamic-G idea and the decoder construction.","headline":"Genuinely new mechanism in symmetry-aware crystal generation; strong benchmark evidence, but the best-in-class headline leans on single-run released-sample baselines and omits WyckoffDiff.","tokens_in":22810,"tokens_out":2041,"would_cite":true,"duration_ms":18593,"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":"DynaCrys treats the crystallographic space group as part of the evolving diffusion state, revising it jointly with Wyckoff occupations and elements, and reports best-in-class stable, unique, and novel crystal discovery rates on MP-20…","keywords":["crystal generation","space-group diffusion","Wyckoff positions","group-subgroup relations","symmetry codebook","legality-constrained decoding","stable-unique-novel rate","discrete diffusion"],"falsifier":"Re-generate every baseline from its current official checkpoint on the same machine with several independent seeds, run all samples through the identical relaxation, hull, matching, and symmetry pipeline, and compare mean and variance of SUN@0/SSUN@0 on both engines; the claim would fail if any baseline's mean at the strict threshold matches or exceeds the corresponding DynaCrys value on either engine.","tokens_in":21791,"feed_emoji":"💎","tokens_out":11897,"duration_ms":90010,"temperature":0.7,"pith_summary":"The paper claims that a crystal generator performs better when the crystallographic space group is not fixed in advance but is revised during generation together with the symbolic description of the crystal. DynaCrys encodes a crystal first as a protostructure—the space group, a set of Wyckoff orbit occupations, and element choices—and applies a coupled discrete diffusion process over these variables, followed by a symmetry-constrained diffusion over lattice and free coordinates. Evaluating about 10,000 candidates per method under two independent relaxation-and-evaluation pipelines, the paper reports that all three DynaCrys configurations outrank every baseline on the strict stable–unique–novel rate and its symmetry-gated variant, and lead the explicit-symmetry block on the metastable thresholds. The same configurations also produce structures that relax with consistently low displacement and sample quickly. If the comparison is fair, dynamic space-group diffusion is an effective foundation for symmetry-aware crystal generation.","feed_headline":"Dynamic space-group diffusion leads crystal discovery benchmarks","feed_subtitle":"It reports top stable-unique-novel rates on MP-20 under two independent evaluation engines.","key_machinery":"Two mechanisms carry the argument. The first is the coupled symbolic diffusion over protostructures: the space-group channel diffuses over the undirected graph of maximal translationengleiche and klassengleiche group–subgroup relations, while a shared pretrained symmetry codebook supplies embeddings c_G and c_{G,r} for space groups and their Wyckoff rows. The codebook's row vectors act as the clean centers of a continuous row channel, as the tied weights of the space-group-conditional row classifier, and as conditioning for the geometry model, so both stages speak the same Wyckoff vocabulary. The second mechanism is the legality-constrained stochastic decoder: after drawing G, K, and the active-slot set with probabilities proportional to products of active logits—computed by elementary symmetric polynomials e_K(ω)—the decoder rejects skeletons that violate row capacity or atom-count constraints, and Proposition 1 shows the accepted output is an exact sample from the constrained distribution without evaluating its global normalizer. The geometry stage then projects every lattice and coordinate update onto the crystal-family and free-anchor subspaces of G, keeping symmetry exact at intermediate steps.","core_discovery":"On its own terms, the paper's central claim is that a space group should be treated as a dynamic diffusion variable rather than a conditioning context fixed before sampling. The model state is a protostructure P=(G,K,{(Z_i,r_i)}), where G is the space group, K the number of orbit occupations, r_i a Wyckoff row in that space group, and Z_i the element. The symbolic forward process mixes a structured group–subgroup transition over maximal translationengleiche and klassengleiche relations with a reset to the training marginal, giving the space group a graph-based corruption channel; survivors, births, and padding handle a changing number of occupations. During reverse sampling, a legality-constrained decoder draws from the denoiser's factorized distribution restricted to legal states—groups with valid rows, capacity and atom-count caps—without computing the global normalizer, using an elementary-symmetric-polynomial proposal. The decoded protostructure then conditions a geometry diffusion whose lattice and fractional-coordinate updates are projected onto the subspace allowed by G, so every intermediate geometry obeys the decoded symmetry. Under the common protocol with two independent relaxation-and-evaluation engines, the paper reports that all three configurations of DynaCrys exceed every baseline on SUN@0 and SSUN@0, occupy the top three explicit-symmetry positions on SUN@0.1 and SSUN@0.1, and retain competitive sample-quality, displacement, and speed metrics.","pith_inferences":["If the coupling is the mechanism rather than the architecture, the same symbolic diffusion could be attached to a different geometry model or to property-conditioned objectives (band gap, formation energy, synthesizability) with minimal changes.","The group–subgroup kernel uses only maximal one-hop relations; extending it to longer paths or to experimentally observed transformation mechanisms might alter discovery rates and would test whether maximal relations are the right transition geometry.","The exact constrained-decoder idea is not crystal-specific: any discrete diffusion whose target is a subset of finite states can use an elementary-symmetric rejection proposal whenever the unconstrained density factorizes over slots, provided acceptance rates stay practical.","Because novelty is defined only against the MP-20 training partition, a natural external validation would be to search generated 'novel' candidates against broader experimental and computed structure databases and to test the most stable ones with higher-fidelity relaxation."],"forward_implications":["Across both engines, all three DynaCrys configurations are reported to exceed every baseline on SUN@0 and SSUN@0; the canonical configuration reaches 9.39% and 9.55% on the two engines.","At the metastable threshold, the three configurations take the top three explicit-symmetry spots on SUN@0.1 and SSUN@0.1.","Replacing the state-dependent space-group proposal with the training-set marginal drops the canonical configuration's SUN@0 from 9.28% to 3.01% at a fixed seed, identifying the coupling as the active ingredient.","The reduced-size DynaCrys-small is 1.59× faster than the canonical configuration while staying ahead of every timed baseline on all four SUN/SSUN columns.","Generated structures relax with lower mean RMSD than every baseline except one, with matched-pair rates around 96%, so the symmetry-enforced geometries are close to their relaxed endpoints."],"supporting_citations":[{"why":"Supplies the Wyckoff-position tables, space-group vocabulary, and t/k group–subgroup definitions that define the protostructure state.","marker":"(Aroyo, 2016)"},{"why":"Supplies the group–subgroup relations used to build the space-group corruption graph.","marker":"(Bärnighausen, 1980)"},{"why":"Supplies the discrete diffusion marginal-kernel formalism underlying the symbolic forward process.","marker":"(Austin et al., 2021)"},{"why":"Supplies the MP-20 dataset, training/test split, and evaluation conventions, plus a baseline.","marker":"(Xie et al., 2022)"},{"why":"Supplies the DiffCSP++ geometry-diffusion framework that the symmetry-constrained stage builds on, and a baseline.","marker":"(Jiao et al., 2024)"},{"why":"Defines the stable–unique–novel criterion used for the main discovery comparison and provides a baseline.","marker":"(Zeni et al., 2025)"},{"why":"Provides the SGEquiDiff symmetry-aware baseline, sampled from its released checkpoint under the common protocol.","marker":"(Chang et al., 2025)"},{"why":"Provides the CHGNet relaxation engine used as the primary stability ruler.","marker":"(Deng et al., 2023)"},{"why":"Provides the MACE-MP-0 relaxation engine used as the independent second ruler.","marker":"(Batatia et al., 2025)"},{"why":"Supplies the DDPM reverse-sampling equations used by the geometry stage.","marker":"(Ho et al., 2020)"}],"fun_headline_variants":["Space group itself becomes a diffusion variable in DynaCrys","Crystal diffusion that evolves symmetry on the fly","DynaCrys: symmetry co-evolves with atoms in diffusion","Dynamic space-group diffusion beats crystal benchmarks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison's load-bearing premise is that the baseline numbers are honest representatives of each baseline's typical performance—one released sample or single run, evaluated under the same protocol—so the reported margins mean DynaCrys is genuinely ahead rather than ahead of unusually weak baseline sets.","fun_headline_variants_meta":{"raw":{"variants":["Space group itself becomes a diffusion variable in DynaCrys","Crystal diffusion that evolves symmetry on the fly","DynaCrys: symmetry co-evolves with atoms in diffusion","Dynamic space-group diffusion beats crystal benchmarks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000184,"raw_usage":{"total_tokens":1342,"prompt_tokens":996,"completion_tokens":346,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":612,"completion_tokens_details":{"reasoning_tokens":281}},"tokens_in":612,"tokens_out":346,"duration_ms":4065,"temperature":1.0,"reasoning_tokens":281,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:27:13.334947+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-generate every baseline from its current official checkpoint on the same machine with several independent seeds, run all samples through the identical relaxation, hull, matching, and symmetry pipeline, and compare mean and variance of SUN@0/SSUN@0 on both engines; the claim would fail if any baseline's mean at the strict threshold matches or exceeds the corresponding DynaCrys value on either engine.","supporting_citations":[{"cited_title":"They define two separate evaluation rulers: values are compared across methods within an engine, not across engines","cited_arxiv_id":null,"evidence_quote":"Provides the MACE-MP-0 relaxation engine used as the independent second ruler."}],"review_version":1}