{"id":"b83f9831-62e7-4785-a0ee-6f8c8a8d3910","arxiv_id":"2607.10592","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Transported empirical means of bundle-valued observations on manifolds obey dimension-free Hoeffding/Bernstein bounds plus an unavoidable holonomy bias floor controlled by curvature.","lead":"This paper derives finite-sample concentration bounds for averages of data that live in different vector spaces attached to points on a curved manifold. The bounds separate ordinary sampling noise from a curvature-driven bias that extra data cannot remove, with direct use in geometric machine learning.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The paper's strongest claim is a clean, modular extension of classical concentration: transport reduces the problem to Hilbert space (so Pinelis applies), while curvature supplies an additive deterministic floor that is both upper-bounded and minimax-lower-bounded inside the class of transport-based estimators. The mathematics is standard but carefully assembled; the matching lower bounds and the sphere numerical match (Appendix H) give independent support. The only genuine limitation is the scope of Assumption 1, which the reader already flags correctly: outside the Hadamard/normal-ball regime the transport rule becomes a modeling choice and Δ_hol is no longer a pure geometric constant. That limitation is acknowledged and quantified rather than hidden, so it does not undermine the claim inside the regime where the claim is made. No further load-bearing concern lands; the ACCEPT verdict stands.","tokens_in":46843,"tokens_out":518,"duration_ms":5594,"concrete_test":"Independently re-derive the holonomy operator-norm bound of Prop. 1 (or the general form Eq. 3) from the Ambrose-Singer theorem and the constant-curvature formula R(X,Y)Z = (1/r^{2})(⟨Y,Z⟩X - ⟨X,Z⟩Y) without using the paper's intermediate steps; confirm that the resulting Δ_hol ≲ (πρ^{2}/r^{2})B is recovered and that the Le Cam separation in Thm. 4 still produces a matching lower bound of the same order.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (bias-variance decomposition ||Ȳ_n - m⋆|| ≲ B/√n + Δ_hol with matching minimax lower bounds for transport-based estimators) is internally consistent under the stated assumptions. Once Assumption 1 reduces observations to i.i.d. Hilbert-space vectors, Theorems 1-2 follow from standard Pinelis inequalities; the holonomy term is cleanly isolated via curvature (Eq. 3, Prop. 1) and shown unavoidable by Le Cam constructions (Thm. 4). The reader's weakest assumption (geodesic uniqueness) is correctly identified as a modeling boundary rather than a flaw: outside normal balls the paper treats Δ_hol as an explicit additive modeling term, not a pure geometric constant. Sphere experiments match theory to ~3.7%. No hidden inconsistency or unsupported leap appears in the load-bearing chain.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper develops a non-asymptotic concentration theory for empirical means of bundle-valued observations on Riemannian manifolds. Observations live in fibers of a vector bundle and are reduced to a fixed reference fiber by parallel transport; the resulting estimator is analyzed via sharp Hilbert-space inequalities. The main results are dimension-free Hoeffding and Bernstein tail bounds (Theorems 1–2), an explicit bias–variance decomposition that isolates a curvature/holonomy-driven deterministic floor Δ_hol (Theorem 3, Eq. 3, Proposition 1), matching minimax lower bounds for the class of transport-based estimators (Theorem 4), a median-of-means estimator under second-moment assumptions (Corollary 2), and a CLT in the reference fiber (Theorem 5). Controlled experiments on the tangent bundle of the round sphere confirm both the n^{-1/2} stochastic decay and the holonomy floor, matching the sharp area formula to within a few percent.","tokens_in":47030,"tokens_out":1257,"duration_ms":38136,"significance":"If the claims hold—and the derivations appear sound—the work fills a genuine gap between classical Hilbert/Banach concentration and manifold statistics for data that live in varying fibers rather than a single vector space. Geometric ML pipelines (gauge-equivariant message passing, intrinsic regression residuals, diffusion-tensor averaging) routinely perform exactly this transport-and-average step; having finite-sample radii that separate sampling noise from an irreducible geometric floor is practically useful and theoretically clean. Strengths that should be credited include: complete appendix proofs of the main theorems via Pinelis inequalities and Ambrose–Singer holonomy; an exact closed-form holonomy formula on TS^2_r; minimax lower bounds that match the upper bounds up to universal constants within the natural estimator class; a robust MoM extension; and reproducible sphere experiments that validate the predicted floor quantitatively. The contribution is more geometric isolation and statistical packaging than new concentration technology, but that packaging is the right object for the applications.","major_comments":[{"comment":"Appendix H / Section 6.4: The holonomy-floor experiments implement Rule B as a synthetic post-composition of Rule A with a fixed rotation of angle θ=πρ², rather than as parallel transport along a genuinely distinct minimizing geodesic. Inside the normal-ball regime ρ<πr/2 used throughout the experiments, minimizing geodesics to x0 are unique, so the construction correctly validates the operator-norm formula of Proposition 1 but does not exercise cut-locus multi-geodesic ambiguity. The manuscript should state this limitation explicitly and indicate whether the same quantitative agreement is expected (or how the theory changes) when support reaches the cut locus and the transport rule becomes a genuine modeling choice.","section":null},{"comment":"Theorem 4 and Remark 1: The minimax lower bounds are correctly scoped to transport-based estimators and require either pinched positive curvature or constant positive curvature for the geometric term. The upper bound (Eq. 3) is stated for general bundle curvature ζ. The paper should make more prominent, already in the main-body statement of Theorem 4, that the matching lower bound on the holonomy floor is a positive-curvature phenomenon and that under the Hadamard branch of Assumption 1 one has Δ_hol=0 by uniqueness, so the two-term rate collapses to the pure stochastic term. This is implicit but easy to miss.","section":null}],"minor_comments":[{"comment":"The three forms of the holonomy discrepancy (canonical section-dependent Δ(P,ẽP;s), section-uniform Δ^unif_hol, and the per-sample operator-norm form used in Theorem 11) are carefully distinguished in the appendix but appear somewhat abruptly in the main text. A short paragraph in Section 5 collecting the three definitions and their relative tightness would help readers.","section":null},{"comment":"Section 6.2 applications (gauge GNNs, DTI, Wasserstein tangent spaces) are useful recipes but contain no new numerical checks beyond the sphere validation. Even a small synthetic gauge-GNN aggregation example illustrating discrete cycle holonomy would strengthen the claim that the bias–variance decomposition is immediately actionable.","section":null},{"comment":"Corollary 1 notes that the Bernstein linear term 2B can be improved to B under exact recentering; this is correct but easy to overlook. Flagging the improved constant in the main display of the confidence radius would be helpful for practitioners.","section":null},{"comment":"Notation: main text uses Assumptions 1–3 while the appendix re-labels them A1–A3 for self-containment. A single sentence at the start of Appendix B noting the correspondence would reduce friction.","section":null},{"comment":"Table 1 is a clear comparison; the entry for this work correctly lists the holonomy floor as irreducible by data alone when Δ_hol>0. Consider adding a column or footnote indicating which frameworks handle heavy tails, since Corollary 2 is a genuine extension of Lugosi–Mendelson to the bundle setting.","section":null},{"comment":"Minor typographical/consistency items: the abstract and introduction both use “n^{-1/2}” and “n−1/2” interchangeably in places; standardize. In Appendix H the theoretical row of Table 4 is italicized as non-experimental—good—but the caption could state this more explicitly.","section":null}],"recommendation":"minor_revision","confidential_remarks":"Solid theory paper with complete proofs and matching numerics. The novelty is real but concentrated in the geometric isolation of holonomy rather than in new probabilistic technology; that is appropriate for the intended geometric-ML audience. Fit for a theory-oriented ML or geometric-statistics venue is good. No citation or novelty-disclosure concerns. I would not block on the experimental-design caveat above; it is a clarity fix, not a correctness issue."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing is that this paper gives the first clean finite-sample, dimension-free concentration for the estimators people actually use when they average tangent residuals, gauge features, or local frames: transport everything to a reference fiber and average. Once that is done you get the usual B/√n stochastic term plus an explicit additive holonomy floor Δ_hol controlled by bundle curvature (Eq. 3, sharp on the sphere via Prop. 1). Both pieces are shown minimax-unavoidable for any transport-based estimator (Thm. 4). That decomposition is the real contribution.\n\nWhat they do well is the reduction. Assumption 1 (Hadamard or normal ball) makes the transported samples i.i.d. Hilbert-space vectors, so Pinelis’s inequalities drop in directly for Hoeffding and Bernstein (Thms. 1–2, fully proved as Thms. 6–7). The holonomy term is isolated rather than swept under the rug, the sphere formula is exact, the Le Cam constructions for the lower bounds are explicit, and the S² experiments match theory to about 3.7 %. They also give MoM robustness and a CLT once the floor is negligible. The appendix is thorough and the citation pattern is honest about Pinelis, Ambrose–Singer, and the Fréchet literature.\n\nThe soft spot is exactly the one the reader flagged: outside normal balls the transport rule is a modeling choice, so Δ_hol is no longer a pure geometric constant. The paper is clear about this and treats it as an additive term rather than pretending it vanishes. That is a genuine scope limitation for global data, not a hole in the argument. Computational cost of transport and the usual practical issues of estimating B and σ² are secondary.\n\nThis is for people who build or analyze geometric pipelines (gauge GNNs, manifold regression, DTI averaging). It is not a new concentration inequality in the abstract; it is the right packaging of classical tools for a setting that previously lacked non-asymptotic guarantees. The math is solid, the data are pure validation, and nothing is circular. I would send it to referees without hesitation and would cite the bias–variance split and the sphere formula myself.","headline":"Clean, usable concentration theory for transported bundle means that correctly isolates a curvature-driven holonomy floor with matching lower bounds and sphere validation.","tokens_in":47604,"tokens_out":532,"would_cite":true,"duration_ms":9569,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60E15","62H11","53C29"],"pacs":[],"model":"grok-4.5","headline":"Averaging vectors living on different points of a curved manifold leaves a curvature-driven error floor that more data cannot remove.","keywords":["concentration inequalities","Riemannian statistics","vector bundles","parallel transport","holonomy","geometric machine learning","bias-variance decomposition"],"falsifier":"On the unit sphere, draw samples from a geodesic ball of fixed radius ρ, form the transported mean under two different transport rules, and check whether the gap between the two means remains constant (and matches 2 sin(π\rho^{2}/2)) as sample size grows from hundreds to tens of thousands; if the gap shrinks or systematically deviates from the formula, the claimed error floor is false.","tokens_in":47761,"feed_emoji":"🌐","tokens_out":724,"duration_ms":8864,"temperature":0.7,"pith_summary":"Many geometric machine-learning pipelines produce observations that live in different local vector spaces attached to different points of a manifold. To average them one must first parallel-transport them into a single reference space; that transport is path-dependent once geodesics are no longer unique. The paper proves that the resulting empirical mean obeys a clean bias-variance split: ordinary sampling noise that decays like 1 over square-root of sample size, plus a deterministic holonomy bias controlled by the curvature of the bundle and the diameter of the data support. Both pieces are shown to be unavoidable for any transport-based estimator, and the theory supplies explicit finite-sample concentration radii, a robust median-of-means variant, and a central-limit theorem once the geometric floor becomes negligible. Controlled experiments on the sphere confirm that the predicted error floor appears exactly where the formulas say it should.","feed_headline":"Curvature sets an error floor no amount of data can erase","feed_subtitle":"Transporting vectors across a manifold leaves a holonomy bias that decays only when the data region shrinks","key_machinery":"The bias-variance decomposition ||Ȳ_n - m⋆|| ≲ B/√n + Δ_hol, where the stochastic piece is controlled by dimension-free Hilbert-space Hoeffding/Bernstein inequalities and the geometric floor Δ_hol is bounded by the operator norm of the bundle curvature times the squared diameter of the support (with a sharp closed-form expression on the round sphere).","core_discovery":"Once bundle-valued observations are parallel-transported to a fixed reference fiber, the transported empirical mean concentrates about its expectation at the classical Euclidean rate, but any residual path-dependence of the transport injects an irreducible, curvature-controlled bias Δ_hol that is independent of sample size; the two terms together form a minimax-optimal bias-variance decomposition for every transport-based estimator.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Holonomy bias sets a curvature floor data cannot erase","Transported means concentrate, but holonomy leaves fixed bias","Bundle averages hit n^{-1/2} rate plus irreducible holonomy error","Parallel transport injects sample-size-independent curvature bias","Curvature and loops fix a bias floor no more samples remove"],"cache_read_input_tokens":32896,"weakest_assumption_plain":"The argument needs a measurable, essentially unique way to choose the transport paths from every data point to the reference point; outside a normal ball or a non-positively curved cover that uniqueness fails and the holonomy term becomes a modeling choice rather than a pure geometric constant.","fun_headline_variants_meta":{"raw":{"variants":["Holonomy bias sets a curvature floor data cannot erase","Transported means concentrate, but holonomy leaves fixed bias","Bundle averages hit n^{-1/2} rate plus irreducible holonomy error","Parallel transport injects sample-size-independent curvature bias","Curvature and loops fix a bias floor no more samples remove"]},"model":"grok-4.5","effort":"low","cost_usd":0.003428,"raw_usage":{"total_tokens":1128,"prompt_tokens":787,"num_sources_used":0,"completion_tokens":68,"cost_in_usd_ticks":34280000,"prompt_tokens_details":{"text_tokens":787,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":273,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":787,"tokens_out":68,"duration_ms":7240,"temperature":1.0,"reasoning_tokens":273,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-14T10:35:52.095487+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"On the unit sphere, draw samples from a geodesic ball of fixed radius ρ, form the transported mean under two different transport rules, and check whether the gap between the two means remains constant (and matches 2 sin(π\rho^{2}/2)) as sample size grows from hundreds to tens of thousands; if the gap shrinks or systematically deviates from the formula, the claimed error floor is false.","supporting_citations":[],"review_version":1}