{"id":"ea7a4ba9-dfc9-45af-929c-347e3c3fb8af","arxiv_id":"2507.08456","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"A transformer that predicts vectors sampled along a polar spiral on a sphere reaches about 90 percent training accuracy but lower validation accuracy, so the proposed geometric ordering remains unvalidated.","lead":"This paper orders points on a sphere along a polar spiral so a transformer can treat spherical data as a sequence. The Spiroformer reaches about 90 percent training accuracy on Hamiltonian vector fields but overfits in validation, so the proposed geometric positional encoding is not yet demonstrated.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The evidence does not establish that spiral ordering conveys geometry: next-token accuracy on spiral samples is compatible with 1D sequence fitting, and no off-spiral or control-ordering evaluation is reported.","rationale":"The paper is a short proof-of-concept, and the authors are honest that Fig. 5 shows overfitting and that their implementations are preliminary. However, the paper's central claim is not supported by the experiment it reports. The training task reduces to 1D sequence prediction along a parameterized curve: input consecutive vector samples on a polar spiral, predict the next. Any reasonably smooth 1D target can be fit by a transformer without accessing the sphere's geometry. Since no held-out evaluation on off-spiral points is reported, and no comparison with random or alternate orderings is given, the observed about 90% training accuracy cannot be attributed to the spiral's geometric structure. Additionally, the polar spiral is not a space-filling curve; it has measure zero on S^2, contradicting the claim that it covers the sphere. These issues align with the reader's weakest_assumption, and the proposed concrete experiment would settle whether the ordering actually helps. Therefore no verdict change is needed.","tokens_in":5918,"tokens_out":6702,"duration_ms":81139,"concrete_test":"Train the identical architecture on the same 100-point sets under four orderings: the polar spiral, its reversal, a random permutation, and a space-filling spherical ordering (e.g., a Hilbert curve on an icosahedral grid). Keep architecture and training budget fixed. Then (a) compare next-sample validation accuracy on held-out Hamiltonian fields, and (b) evaluate each trained model on 1,000 uniformly sampled points off the spiral, using the model's position-dependent outputs, against the true Hamiltonian field. If spiral-order accuracy is not clearly above the random-order control and off-spiral RMSE is not substantially lower than a \"copy previous sample\" baseline, the geometric-reconstruction claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Sections 1.3 and 1.4) is that the Spiroformer \"learns to reconstruct Hamiltonian vector fields on the sphere.\" The only reported evidence is next-sample prediction accuracy along one polar spiral, about 90% in training with lower validation (Fig. 5). Because the training data are values of a smooth field sampled along a 1D curve, the target sequence v_{t+1} is essentially a function of the curve parameter t; a transformer with enough capacity can fit such a 1D sequence without learning anything about S^2. No evaluation is performed on points off the spiral, so the model is never asked to reconstruct the field on the sphere. There is also no comparison with a random ordering, a reversed spiral, or any other ordering of the same 100 points, which would be needed to attribute the accuracy to the spiral's geometry rather than to the generic sequential structure. Finally, the polar spiral is not a space-filling curve: its image has measure zero on S^2 and is not surjective, so Section 1.1's assertion that it \"cover[s] the sphere\" is false. The authors openly report overfitting and propose future fixes, but the current protocol cannot support the geometric-reconstruction claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes to extend transformers to geometric domains by ordering tokens along a spherical spiral, which it calls a 'space-filling curve.' It introduces the Spiroformer, a transformer trained to predict the next sample of a Hamiltonian vector field on the 2-sphere given a sequence of 100 vector samples along a polar spiral. The authors report roughly 90% training accuracy (Fig. 5), acknowledge lower validation performance and overfitting, and suggest future improvements including regularization and a SymFlux-like symbolic Hamiltonian recovery. The central claim, stated in Sections 1.3 and 1.4, is that the model 'effectively learns to reconstruct Hamiltonian vector fields on the sphere.'","tokens_in":6153,"tokens_out":2960,"duration_ms":35254,"significance":"The idea of using a curve that imposes a global order on a manifold to adapt sequence models to geometric data is potentially interesting, and the paper makes its data-generation pipeline concrete by using spherical harmonics and the Poisson bivector on S^2. However, the significance as presented is limited by the weakness of the experimental evidence and by a mathematically inaccurate core premise. The paper does not provide machine-checked proofs, reproducible code, or a parameter-free derivation; its only quantitative result is a single training/validation curve. If the evaluation were properly controlled and the space-filling claim corrected, the approach could be a modest contribution to geometric deep learning, but as it stands the supporting evidence does not establish the stated reconstruction claim.","major_comments":[{"comment":"The claim that the Spiroformer 'effectively learns to reconstruct Hamiltonian vector fields on the sphere' is not supported by the experimental protocol. The only reported measure is next-sample prediction accuracy along one polar spiral. Since the data are 100 samples of a smooth vector field along a 1D curve, the target v_{t+1} is essentially a function of the curve parameter t; a transformer with enough capacity can fit such a 1D sequence without learning the geometry of S^2. There is no evaluation on points off the spiral, no comparison with a random or reversed ordering of the same 100 points, and no quantitative error metric (e.g., mean squared error against true off-spiral vectors). The lower validation accuracy in Fig. 5 is consistent with interpolation along the curve rather than geometric reconstruction.","section":"Section 1.3 and 1.4, Fig. 5"},{"comment":"The polar spiral is not a space-filling curve. Its image is a 1D submanifold of S^2 with measure zero; it is not surjective, so it does not 'cover the sphere' or 'provide a continuous traversal of the sphere's surface' in the sense of a space-filling curve. This error is load-bearing because the title, abstract, and Section 1.1 motivate the entire method through the space-filling property. Either the terminology should be corrected to 'spiral ordering' or a genuinely space-filling construction (e.g., an approximation of a Hilbert curve on the sphere) should be used.","section":"Section 1.1, 'Spiral Over the Sphere'"},{"comment":"The sentence 'With larger sample sizes on machines with more local memory resources, validation could match training performance close to 90%' is speculation, not a result. Moreover, the reported 'high accuracy during training, ~90%' is based on a single curve with no error bars, no definition of accuracy (e.g., thresholded cosine similarity, classification, or regression), and no table of exact values. Without a precise definition and repeated runs, the number is not reproducible and does not support the reconstruction claim.","section":"Section 1.4 and Fig. 5"}],"minor_comments":[{"comment":"The term 'space-filling curve' is used inaccurately; a polar spiral on S^2 is not space-filling. Consider using 'spiral positionality' or 'spiral ordering' throughout.","section":"Abstract and Title"},{"comment":"The dataset consists of Hamiltonian vector fields generated from spherical harmonics through the authors' own poissongeometry modules and sampled on a geodesic sphere from geomstats. This is not inherently circular, but the absence of any independent data source or held-out field family makes the evaluation self-referential to the generation pipeline.","section":"Section 1.2, Dataset Generation"},{"comment":"The positional encoding is not specified; the reader does not know whether it is a learned embedding, a sinusoidal encoding, or a function of the spiral parameter t. The manuscript should state this explicitly.","section":"Section 1.3, Positional Encodings"},{"comment":"The axes and legend of Fig. 5 are not fully described; in particular, the y-axis is not labeled, and it is unclear whether the plotted quantity is accuracy, loss, or something else.","section":"Fig. 5"},{"comment":"Reference [7] contains a typo ('an others' should be 'and others').","section":"References"},{"comment":"The mention of SymFlux as a future direction is not elaborated; since it is used to motivate structure preservation, a brief explanation of how it would be adapted to spherical data would be helpful.","section":"Section 2, Conclusions"}],"recommendation":"reject","confidential_remarks":"This manuscript reads as an extended abstract rather than a complete research paper. The central experimental claim is not established by the reported single training/validation curve, and the mathematical use of 'space-filling curve' is incorrect. The paper's scope—a short position statement with preliminary results—might fit a workshop, but for a journal the evidence is insufficient. I recommend rejection; a resubmission would need a properly controlled evaluation with off-spiral tests, alternative orderings, error bars, and a corrected characterization of the spiral."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The Spiroformer idea is simple and not silly: give a spherical manifold a global order by following a polar spiral, and read vector-field samples into a transformer in that order. That is a legitimate and modestly novel extension of curve-ordering ideas to transformers on S^2. The paper also gives a coherent pipeline for generating Hamiltonian vector fields from spherical harmonics via sympy and poissongeometry, and it is honest about the overfitting visible in Fig. 5. Those are real merits.\n\nWhat is new here is really the specific spiral-ordered positionality application; the components are established. The references to spherical transformers and equivariant spherical networks are adequate, though thin. The authors' reliance on their own Poisson-geometry packages for data generation is not a problem in itself, but without shipped code or an independent check it makes the single reported curve harder to trust.\n\nThe soft spots are large and load-bearing. Section 1.3 claims the model 'effectively learns to reconstruct Hamiltonian vector fields on the sphere.' The only evidence is next-sample accuracy along the spiral, ~90% training with lower validation in Fig. 5. The model never sees a point off the spiral, so it is never tested on reconstruction over S^2. A smooth vector field sampled on a 1D curve yields a sequence that a transformer can fit as a function of the curve parameter, without learning any geometry. No random ordering, reversed spiral, or alternate sampling of the same points is reported, so the contribution of the spiral order itself is not isolated. There are no error bars, baselines, held-out field families, or quantitative error metric, and the Section 1.4 guess that validation could reach 90% with more resources is exactly that. The term 'space-filling curve' is also wrong for this spiral: its image has measure zero on S^2. That is minor terminology, but it points at the missing off-spiral evaluation.\n\nThis paper is for people working on positional encodings on manifolds, and it is a useful cautionary example of confusing next-token fit on a 1D sample with geometric reconstruction. It is not ready for archival peer review as is; I would not send it out. Desk rejection with an invitation to resubmit with controls—off-spiral test points, ordering baselines, error bars, and code—is the fair outcome. I would read the revision.","headline":"A modestly novel spiral-ordering positionality idea whose only experimental support is a single training curve, so the geometric-reconstruction claim is untested as written.","tokens_in":6670,"tokens_out":3592,"would_cite":false,"duration_ms":41124,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["68T07","53D05","53D17"],"pacs":[],"model":"deepseek-v4-flash","headline":"A spiral space-filling curve supplies the global order transformers lack on the sphere, and a small transformer trained along it reaches roughly 90% training accuracy at predicting spherical Hamiltonian vector fields.","keywords":["transformers","space-filling curve","spherical spiral","positional encoding","Hamiltonian vector fields","spherical harmonics","geometric deep learning","manifolds"],"falsifier":"Evaluate the trained model on vectors at points randomly sampled across the whole sphere rather than on the spiral; if prediction error at those points is no better than guessing the mean field, the model has learned curve-following, not spherical geometry. A complementary test is to retrain on the same vector samples in a randomly permuted order—if training accuracy stays near 90%, the spiral's geometric ordering is not the cause of the result.","tokens_in":5707,"feed_emoji":"🌀","tokens_out":7187,"duration_ms":75230,"temperature":0.7,"pith_summary":"Transformers rely on a fixed global order—text, time, or grid position—but manifolds like the sphere have no canonical ordering. This paper proposes that a space-filling curve, specifically a polar spherical spiral, can supply that order, converting geometric data into a sequence a transformer can process. The authors build the Spiroformer, which reads samples of Hamiltonian vector fields along the spiral and predicts the next sample, and report about 90% training accuracy, with validation still lagging due to overfitting. The point they are trying to establish is that this spiral-based positionality is a viable way to import the full transformer toolkit into geometric domains.","feed_headline":"Spiral curve gives transformers a global order on the sphere","feed_subtitle":"Spiroformer turns vector-field samples on the sphere into a sequence, reaching ~90% training accuracy.","key_machinery":"The central object is the polar spherical spiral, the curve $(x,y,z)=(\\sin t \\cos ct, \\sin t \\sin ct, \\cos t)$ for $t \\in [0,\\pi]$, used as a space-filling curve that assigns a total order to points of $S^2$. The transformer's positional encoding is tied to position along this curve, so attention operates along the spiral rather than on the sphere directly. The training signal comes from Hamiltonian vector fields $X_H$ built with the Poisson bivector $\\pi = \\sin\\theta\\,\\partial_\\theta \\wedge \\partial_\\phi$ applied to spherical harmonics, and the task is next-sample prediction with causal masking. Together these convert the geometric problem into a sequence-to-sequence one.","core_discovery":"The paper's central claim is that attention heads placed along a space-filling curve give a transformer the global ordering it needs to handle manifold-valued data, and that this works concretely for reconstructing Hamiltonian vector fields on the 2-sphere. The Spiroformer takes sequences of vector samples generated from spherical-harmonic Hamiltonians, uses the spiral's parameter to define positional encodings, and learns to predict the next vector in the sequence, achieving roughly 90% training accuracy (Fig. 5). On the authors' framing, the model 'effectively learns to reconstruct Hamiltonian vector fields on the sphere,' so the discovery is a new positional-ordering mechanism, with the overfitting gap acknowledged as an open technical issue.","pith_inferences":["Editorial: the ~90% training accuracy admits a mundane alternative explanation—along a smooth spiral, consecutive points are heavily correlated, so a model that learns local interpolation could succeed without any global geometric representation; a shuffled-order control would distinguish these.","Editorial: the mechanism should be tested for curve-specificity: swapping the polar spiral for another space-filling curve (e.g., a Hilbert-like curve on a cubed sphere or a Peano curve) on the same task would show whether the benefit comes from the particular spiral or from the general property of having a locality-preserving total order.","Editorial: the Hamiltonian structure is not essential to the ordering idea; rerunning the same setup with generic smooth functions would clarify whether the spiral positionality, rather than the symplectic structure, is what the transformer exploits."],"forward_implications":["If the spiral ordering works, any manifold admitting a space-filling curve can be fed into a standard transformer by using the curve as the positional encoding, bypassing custom spherical attention layers.","The ~90% training accuracy shows that an autoregressive transformer can track the local evolution of spherical Hamiltonian vector fields along the curve, which is the first concrete evidence for the approach.","The validation gap implies a concrete next step: enlarge the dataset (e.g., more than 100 samples per field) and regularize, which the authors expect could bring validation close to training accuracy.","A structure-preserving variant that recovers the symbolic Hamiltonian (via symbolic regression) would guarantee the model outputs are genuinely Hamiltonian, a direction the paper identifies as future work."],"supporting_citations":[{"why":"Supplies the transformer architecture and positional encoding machinery that the Spiroformer is built on.","marker":"[18]"},{"why":"Symbolic Poisson geometry computation used to generate the Hamiltonian vector fields from spherical harmonics.","marker":"[8]"},{"why":"Numerical Poisson geometry module used to evaluate the symbolic vector fields on the discrete sphere.","marker":"[9]"},{"why":"Provides the discrete spherical-coordinate sampling of the sphere on which the vector fields are evaluated.","marker":"[16]"},{"why":"Standard reference for Hamiltonian mechanics and symplectic formalism that defines the target vector fields.","marker":"[1]"}],"fun_headline_variants":["Spiral attention orders transformers on the sphere","Spiroformer: space-filling curve unlocks sphere data","Space-filling spiral gives transformers global order","Transformer on a sphere? Follow a polar spiral"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim rests on the assumption that training the transformer to predict the next vector sample along the spiral measures understanding of the sphere's geometry, even though the model never sees off-spiral data and consecutive samples are strongly correlated by the smoothness of the field.","fun_headline_variants_meta":{"raw":{"variants":["Spiral attention orders transformers on the sphere","Spiroformer: space-filling curve unlocks sphere data","Space-filling spiral gives transformers global order","Transformer on a sphere? Follow a polar spiral"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":9.3e-05,"raw_usage":{"total_tokens":883,"prompt_tokens":728,"completion_tokens":155,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":344,"completion_tokens_details":{"reasoning_tokens":95}},"tokens_in":344,"tokens_out":155,"duration_ms":2462,"temperature":1.0,"reasoning_tokens":95,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:18:41.067036+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the trained model on vectors at points randomly sampled across the whole sphere rather than on the spiral; if prediction error at those points is no better than guessing the mean field, the model has learned curve-following, not spherical geometry. A complementary test is to retrain on the same vector samples in a randomly permuted order—if training accuracy stays near 90%, the spiral's geometric ordering is not the cause of the result.","supporting_citations":[{"cited_title":"Vaswani, N","cited_arxiv_id":null,"evidence_quote":"Supplies the transformer architecture and positional encoding machinery that the Spiroformer is built on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Symbolic Poisson geometry computation used to generate the Hamiltonian vector fields from spherical harmonics."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Numerical Poisson geometry module used to evaluate the symbolic vector fields on the discrete sphere."},{"cited_title":"Miolane, N","cited_arxiv_id":null,"evidence_quote":"Provides the discrete spherical-coordinate sampling of the sphere on which the vector fields are evaluated."},{"cited_title":"Arnold, A","cited_arxiv_id":null,"evidence_quote":"Standard reference for Hamiltonian mechanics and symplectic formalism that defines the target vector fields."}],"review_version":1}