{"id":"335d4abe-93d3-492e-9180-a76bd4df7873","arxiv_id":"2502.02400","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"By matching edges of a point-cloud complex to minimising geodesics and minimizing lifted distances, one can label each cycle by its ambient homology class.","lead":"The paper presents a way to decide which loops detected in a point cloud on a manifold are truly loops of the manifold itself. It relies only on the manifold's known universal covering and on distances between points and their lifted copies, which makes it attractive for topological data analysis on tori, projective spaces, and other model manifolds.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The inference is constructive only if the deck-group minimization in Eq. (52) is actually solvable; the paper gives no algorithm, complexity bound, or decidability argument, and for a general compact Riemannian manifold Γ is an arbitrary finitely presented group.","rationale":"I read the paper in good faith and found the mathematical core—Corollary 3.16, Proposition 3.17, and Proposition 4.5—internally coherent: if one is given a realization functor satisfying (S1) and can solve the deck-group minimization in Eq. (52), then the transition homomorphism does recover the induced H1 map, and the groupoid machinery is sound. The reader's weakest assumption also pointed at the same two soft spots; I agree that the algorithmic premise of Eq. (52) is the more load-bearing one for the paper's central 'constructive' claim, while (S1) is less severe because the paper proves genericity for compact manifolds and for Cech complexes below the convexity radius. The concern is not that the mathematics is wrong but that the paper's stated pipeline cannot be executed from the inputs it claims suffice: no algorithm for the infimum over Γ is supplied, and for arbitrary compact Riemannian manifolds the deck group can be a finitely presented group for which decidability of such a minimization is not guaranteed. A concrete implementation on the paper's own hyperbolic example would settle whether the construction is genuinely effective. Since this matches the reader's conditional verdict, I recommend no change to the verdict; the paper remains acceptable only if this computational premise is either discharged or explicitly added as an assumption.","tokens_in":44647,"tokens_out":22587,"duration_ms":249934,"concrete_test":"For the genus-2 example of Example 3.2(iv), specify and implement a finite algorithm for Eq. (52): e.g., using a Dirichlet fundamental domain for the Fuchsian group, enumerate all deck transformations g with dtilde(ytilde_i, g·ytilde_j) ≤ d(x_i, x_j) (which is finite by properness), prove the enumeration terminates and the arg inf is found, and run it on one million random four-point samples to report uniqueness and any tie-breaking failures. If no such algorithm can be specified and verified, Eq. (52) is not a construction from the stated data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 4.5, and its applications in Propositions 4.6 and 4.8, rest entirely on solving t(ij) = arg inf_{g∈Γ} dtilde(ytilde_i, g·ytilde_j). The paper treats Γ as known and correctly notes (Prop. 4.3(ii)) that the infimum is attained by finitely many deck transformations, but it never explains how the arg inf is computed, certified, or bounded. For the flat torus and Klein bottle this is a finite lattice search, but for the genus-2 hyperbolic example Γ is a Fuchsian group; no fundamental-domain search, no finite enumeration radius, and no complexity estimate is supplied. More seriously, every finitely presented group arises as π1 of a compact smooth manifold, hence of a compact Riemannian manifold after choosing a metric. Thus, without an explicit restriction on Γ or an oracle for this minimization, the central claim that 'metric data on the point cloud and its fibre in the covering suffices' is not an effective construction. Eq. (52) is the key computational step: if it cannot be implemented, the transition homomorphism is not actually inferred, and the empirical labelling of principal persistence cycles in Section 4.4 is not reproducible from the stated inputs. This incompleteness is load-bearing for the paper's 'constructive method' claim, even though the surrounding groupoid and covering-space mathematics appears sound.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a groupoid-theoretic and covering-space framework for deciding, from a finite point cloud in a compact Riemannian manifold M with known universal covering p : M̃ → M, whether a one-cycle of a simplicial complex built on the point cloud is non-trivial in M. The main results relate the monodromy homomorphism of an induced covering to a transition homomorphism on the nerve of a good cover (Propositions 3.10 and 3.13), show that the first homology of the induced map is recovered as a map H₁(N(U)) → Γ_ab (Proposition 3.17), and—in the geometric part—construct the transition homomorphism from the assignment t(ij) = arg inf_{g∈Γ} d̃(ỹ_i, g·ỹ_j) under a uniqueness assumption on minimising geodesics (Proposition 4.5), with instantiations for Čech complexes of ε-thickenings for ε < conv(M) (Proposition 4.6) and for min-geodesic graphs (Proposition 4.8). The paper also proves that point clouds avoiding mutual cut loci are open and dense in configuration space on compact manifolds (Propositions 4.11–4.12, Corollary 4.13) and gives an empirical application to labelling principal persistence measures on four-point cycles for the torus, Klein bottle, RP², and the genus-2 surface (§4.4).","tokens_in":44951,"tokens_out":11026,"duration_ms":109865,"significance":"If the method can be made fully constructive, the paper provides a broad and conceptually clean answer to an important question in TDA: the metric data of the point cloud and of one lift per point do determine the induced homology homomorphism up to the inherent gauge freedom (conjugation). The groupoid formalism is appropriate and the chain of results from Section 2.3 to Proposition 4.5 is coherent; there are no fitted parameters, the main assumptions (S1) and (C1)–(C2) are stated explicitly, and the paper is honest about the limitations it does see (Remark 3.14 on pullback monodromy, Remark 4.14 on non-openness of unique-geodesic configurations). The generic-uniqueness results in Section 4.3 are a useful contribution in their own right. The central shortfall is that the passage from the mathematically well-defined Eq. (52) to an effective construction is not supplied, and the empirical section is not reproducible as described; these points are detailed in the major comments.","major_comments":[{"comment":"The abstract and Section 4 present the method as constructive, but the central computational step, t(ij) = arg inf_{g∈Γ} d̃(ỹ_i, g·ỹ_j) in Eq. (52), is not accompanied by an algorithm, a complexity bound, or a decidability statement for arbitrary deck groups Γ. Proposition 4.3(ii) proves only that the infimum is attained by finitely many elements; it does not explain how those elements are found. Since compact Riemannian manifolds in dimension at least 4 realize every finitely presented group as their fundamental group, and hence as deck group of the universal covering with an arbitrary metric, 'Γ is known' does not by itself make Eq. (52) computable. The constructive claim in the abstract and in Proposition 4.5 therefore needs either an explicit class of coverings for which the minimization is tractable (e.g., lattices and Fuchsian groups with fundamental-domain searches and explicit search radii), or a formal oracle model; as written, the 'metric data suffice' statement is a well-defined existence statement but not an effective construction.","section":"§4, Prop. 4.5, Eq. (52)"},{"comment":"The empirical section reports one million sampled four-point configurations on the torus, Klein bottle, RP², and the genus-2 surface and labels each persistent cycle by its homology class via Eq. (52). No numerical method is given for computing the arg inf over Γ, in particular for the Fuchsian group of the genus-2 surface, where neither a fundamental-domain search, nor a ball radius bound for the finite attainable set, nor a tolerance is specified. The figures (Figs. 2–5) are kernel density estimates without accompanying data or code, so the claimed decomposition of the principal persistence measure is not reproducible from the inputs described in the text.","section":"§4.4.2"},{"comment":"Assumption (S1) in Proposition 4.5 is the uniqueness hypothesis that makes the arg inf in Eq. (52) unambiguous. It is certified only in two regimes: ε < conv(M) for the Čech construction (Prop. 4.6) and generically in configuration space for compact manifolds (Cor. 4.13). The min-geodesic graph pipeline of Prop. 4.8, which is the one used in §4.4, has no verifiable per-edge certificate of uniqueness for a given point cloud; non-generic clouds (points on mutual cut loci) fall outside Cor. 4.13, and Remark 4.14 explicitly notes that unique-geodesic configurations need not form an open set. In the non-generic case where several g attain the infimum, the homology label depends on the choice of minimizer, so the paper should state what is checked or perturbed in practice to enforce (S1) in the experiments, or restrict the claims accordingly.","section":"§4.3, Cor. 4.13 and §4.2, Prop. 4.8"}],"minor_comments":[{"comment":"The notation in the proof is inconsistent: 'C = C(EX)' should read 'C̄ = C₁(EK)' and 'B = B(EK)' should be 'B̄ = B₁(EK)'; the text also concludes 'H₀(X) ∼= H₀(X)' where the second factor should be H₀(ΠX); and the phrase 'the , induces' contains a stray comma.","section":"Lemma 2.20 proof"},{"comment":"There are repeated typos: 'manfiolds' (§4.1 header and Definition 4.10), 'meausure', 'empiricla', 'folowing' (§4.4.2), 'unviersal' (§4.4.2), 'cna' (Introduction), and 'the such data' (§4.4.2); these should be corrected in a revision.","section":"Throughout"},{"comment":"The citation 'Vidit Nanda's lecture notes (?)' contains a dangling '(?)' and should be completed with a full reference or removed.","section":"Appendix A, Lemma A.3"},{"comment":"The map π₀(ι) appears without a definition of ι; the maps in the pointed-set exact sequence should be labelled consistently.","section":"§3.1, Eq. (28)"},{"comment":"The phrase 'When we lift any such geodesic' should specify that the lift is based at the chosen x̃ ∈ p⁻¹(x), and the proof carries some claims forward (e.g., d(x,y) = L) before they are established, which makes the ordering hard to follow.","section":"Prop. 4.3 proof"},{"comment":"The caption states homology classes are (n,m) ∈ Z ⊕ Z₂; the sign convention used to identify the classes should be specified, since the abelianization of the Klein bottle group has a canonical Z₂ summand only up to choice of generator.","section":"Figure 3 caption"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of math.AT and would be a solid contribution once the algorithmic gap in Eq. (52) is addressed, either by restricting to a tractable class of coverings or by making the computational model explicit. The empirical Section 4.4 would profit considerably from data and code sharing; as it stands it is illustrative rather than decisive. I would not reject on the basis of the empirical weakness alone, but the central 'constructive method' claim should be aligned with what is actually established. The relationship to the prior work [21] and [22] is stated clearly enough."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a legitimately useful paper for TDA people who work with point clouds on manifolds with known universal covers, and the main mathematical engine is sound. The generalization from the Z^n periodic case in [21,22] to arbitrary deck groups is real, and the geodesic-lift labelling in Prop 4.5 is a clean idea. I buy Propositions 4.5, 4.6, 4.8; the genericity result (open dense NC point sets) is nice, and the principal-persistence decomposition is a genuine application.\n\nThe soft spot the reader flagged is real and load-bearing for the 'constructive' claim. Eq (52) asks for the arg inf over all deck transformations for every edge. The paper proves the minimizer set is finite (Prop 4.3(ii)) but gives no algorithm, fundamental-domain search, or complexity bound. Since every finitely presented group is pi_1 of a compact manifold, you cannot assume Gamma is tractable in general. For the flat torus and Klein bottle you have a lattice search; for the genus-2 hyperbolic example you need actual hyperbolic distance computations with a known fundamental domain. The paper should either restrict the claim to such tractable classes or explicitly state that 'constructive' means existence of a finite description, not an effective computation. As written, the phrase 'constructive method' overpromises.\n\nThe empirical section is also thin: no code, no data, no error bars, and the text admits it is an illustration. That is fine for a math paper, but it should be framed as a demonstration, not as part of the main claim. Minor presentation issues: the proof of Lemma 2.20 has notation slips, and Prop 4.18 is hard to follow. These are minor and fixable.\n\nBottom line: I agree with the conditional verdict. The covering-space/groupoid mathematics is coherent and the core inference mechanism is valuable. I would send this to a serious referee, but with the expectation of major revision: add the algorithmic discussion (or an explicit restriction), make the experiments reproducible, and clean up the presentation. It is not desk-reject material, and it is not yet ready as-is.","headline":"Solid math and a real generalization of ambient-cycle labeling to arbitrary universal coverings, but the 'constructive' claim needs an algorithmic caveat and the experiments need reproducibility.","tokens_in":45450,"tokens_out":2614,"would_cite":false,"duration_ms":29280,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["55N31","57M10","53C22","18B40"],"pacs":[],"model":"deepseek-v4-flash","headline":"If the universal cover of the ambient manifold is known, distances between lifted sample points suffice to decide which cycles of a point cloud are genuine loops of the ambient space, and the paper gives the explicit homology-labelling map.","keywords":["topological data analysis","universal covering","monodromy","fundamental groupoid","edge groupoid","minimising geodesics","ambient cycles","principal persistence measures"],"falsifier":"On the flat unit torus, represent lifts in $\\mathbb{R}^2$ and compute $t(ij) = \\arg\\min_{g\\in\\mathbb{Z}^2} \\|\\tilde x_i - g - \\tilde x_j\\|$; for any four-point configuration that Lemma 4.15 says yields a persistent cycle, compare $\\tau = t_{\\mathrm{ab}}(01)+t_{\\mathrm{ab}}(12)+t_{\\mathrm{ab}}(23)+t_{\\mathrm{ab}}(30)$ in $\\mathbb{Z}^2$ with the actual winding of the interpolating geodesic loop around the torus — a single mismatch would falsify the reconstruction. A complementary test on the real projective plane, with lifts in $S^2$ and a pair of antipodal points where minimising geodesics are not unique, would show whether the uniqueness premise (S1) is genuinely needed for the homology recovery.","tokens_in":44442,"feed_emoji":"🕳️","tokens_out":17684,"duration_ms":142708,"temperature":0.7,"pith_summary":"Topological data analysis usually reports cycles in a point cloud's complex without saying whether those cycles are genuinely non-trivial in the space the points were sampled from. This paper shows that if the ambient manifold's universal cover and its deck group $\\Gamma$ are known, the question becomes computable from metric data alone: for each edge of the complex, take the deck transformation that brings one chosen lift closest to the other, and record it as $t(ij)$. These edge labels assemble into the transition homomorphism of the covering induced on the complex, and its abelianisation is exactly the inclusion-induced map on first homology, $H_1(K) \\to H_1(M) \\cong \\Gamma_{\\mathrm{ab}}$. So a cycle is an ambient cycle precisely when its labelled edges sum to a non-zero element of $\\Gamma_{\\mathrm{ab}}$. The method is constructive for Čech complexes below the convexity radius and for graphs whose edges are minimising geodesics, which exist for generic point clouds on compact manifolds, and the paper demonstrates it by labelling the four-point persistence cycles of uniform samples on the torus, Klein bottle, projective plane, and a genus-two surface by their ambient homology classes.","feed_headline":"Nearest-lift distances reveal which point-cloud loops are real holes","feed_subtitle":"Distances between lifts in the universal cover assign each data cycle its ambient homology class, telling real loops from contractible ones.","key_machinery":"The load-bearing object is the transition homomorphism $t: EK \\to \\Gamma$ of the induced covering, computed edge-by-edge as $t(ij) = \\arg\\inf_{g\\in\\Gamma} \\tilde d(\\tilde x_i, g\\cdot \\tilde x_j)$. Two bridges make this metric computation carry topological content. First, the equivalence between the edge groupoid $EK$ of the complex and the fundamental groupoid $\\Pi|K|$, realised through the realisation and snapping homomorphisms $R$ and $S$: a groupoid homomorphism defined on edges, satisfying the cocycle condition of Lemma 2.8, stands in for homotopy classes of paths. Second, the monodromy correspondence: a transition homomorphism of a $\\Gamma$-covering encodes exactly the same data as the monodromy homomorphism $\\mu: \\Pi X \\to \\Gamma$, the deck transformation picked out by lifting paths. The geometric fact tying these together is that a minimising geodesic in $M$ lifts to a minimising geodesic in $\\tilde M$ ending at the nearest point of the relevant fibre, so the nearest-lift arg inf reproduces the monodromy on the unique geodesic class. Groupoid homology then converts $t$ into the explicit homology map $\\tau: H_1(N(U)) \\to \\Gamma_{\\mathrm{ab}}$ that recovers $H_1(f)$.","core_discovery":"The central claim is that the induced map on first homology, $H_1(f): H_1(K) \\to H_1(M)$, for a simplicial complex $K$ built on a finite point sample in a compact Riemannian manifold $M$ can be recovered from metric data on $M$ and on the universal covering $p: \\tilde M \\to M$. Concretely, choose arbitrary lifts $\\tilde x_i \\in p^{-1}(x_i)$ for the vertices, and for each edge $ij$ define $t(ij) = \\arg\\inf_{g\\in\\Gamma} \\tilde d(\\tilde x_i, g\\cdot \\tilde x_j)$, where $\\Gamma$ is the deck group of the covering. Under the assumption that the edge's homotopy class in $M$ contains a unique minimising geodesic, the values $t(ij)$ form a groupoid homomorphism $t: EK \\to \\Gamma$, namely the transition homomorphism of the covering of $|K|$ induced by $f$. By the monodromy correspondence, $t$ carries the same information as the monodromy homomorphism $\\mu: \\Pi|K| \\to \\Gamma$, which is the pullback of the monodromy of the universal covering, so $t$ embeds $\\pi_1(|K|)$ into $\\Gamma$ up to conjugation. Applying groupoid homology, the map $\\tau: H_1(K) \\to \\Gamma_{\\mathrm{ab}}$ given by $\\tau([\\sum_i a_i \\sigma_i]) = \\sum_i a_i\\, t_{\\mathrm{ab}}([\\sigma_i])$ recovers $H_1(f)$ up to the natural isomorphisms $H_1(K) \\cong H_1(N(U))$ and $H_1(M) \\cong \\Gamma_{\\mathrm{ab}}$. In this way, the homology class of every cycle of the complex, and in particular whether it is trivial in the ambient manifold, is read off a finite collection of nearest-lift computations.","pith_inferences":["The nearest-lift step is the computational heart of the method, and for the model spaces the paper lists it is directly solvable: on the flat torus it is a nearest-lattice-point search in $\\mathbb{Z}^2$, and on the hyperbolic disk a bounded search over generator combinations; an algorithmic treatment with complexity bounds for general $\\Gamma$ is the natural next step but is not given in the paper","The criterion that a cycle is ambient exactly when $\\tau \\neq 0$ in $\\Gamma_{\\mathrm{ab}}$ suggests a practical noise-versus-signal pipeline: colour each point of a persistence diagram by its ambient homology class, treating the zero class as sampling noise; this is testable on the torus, where the correct class of a geodesic loop is directly visible.","The restriction to first homology is forced by the fundamental groupoid, which only sees one-dimensional topology; whether the same nearest-lift idea has a higher-degree analogue through the universal cover's cohomology or higher homotopy is a question the paper leaves open."],"forward_implications":["For Čech complexes with radius below the convexity radius of the manifold, the good-cover property means the induced map $H_1(\\check{\\mathrm{C}}ech_\\epsilon(X)) \\to H_1(M)$ is computable from pairwise distances on the point cloud and pairwise distances between chosen lifts, with no further information about $M$.","For min-geodesic graphs, whose edges are mapped to minimising geodesics, the same nearest-lift formula yields the transition homomorphism, and conversely any edge labelling into $\\Gamma$ whose values attain the infimum is the transition homomorphism of some min-geodesic embedding.","Generic point clouds on compact manifolds satisfy the uniqueness premise: point sets with no point in the cut locus of another form an open dense subset of the configuration space, so the construction is well-defined for almost every configuration.","Because the homology map $\\tau$ lands in $\\Gamma_{\\mathrm{ab}}$, it is independent of the arbitrary choices of atlases and lifts, allowing cycles from different samples to be compared by ambient homology class; this abelianisation sidesteps the conjugation ambiguity that affects the statement at the level of $\\pi_1$.","Applied to four-point cycles, the method decomposes the first principal persistence measure by ambient homology class, and the empirical experiments on the torus, Klein bottle, real projective plane, and genus-two surface show that non-trivial ambient cycles can exceed the persistence bounds proven for simply-connected constant-curvature spaces and cannot approach the origin of the persistence dia"],"supporting_citations":[{"why":"Supplies the covering-space theory — path lifting, monodromy, deck transformations, and classification of coverings — on which the monodromy and induced-covering constructions rest.","marker":"[13]"},{"why":"Provides the geometric premises: convexity radius, cut locus, and the fact that pairs outside mutual cut loci have unique minimising geodesics, which underpin assumption (S1).","marker":"[5]"},{"why":"Cited for the fact that geodesic balls of radius below the convexity radius form a good cover, enabling the Čech construction of Proposition 4.6.","marker":"[8]"},{"why":"Provides the Riemannian-covering facts — geodesics lift to geodesics of the same length and deck actions are isometries — used in Proposition 4.3 and Corollary 4.4.","marker":"[9]"},{"why":"Supplies the stability of the cut locus under perturbation of the base point, used to show that point clouds with unique minimising geodesics form an open (and dense) set.","marker":"[7]"},{"why":"Defines the principal persistence measures and the four-point condition for a non-trivial first persistence generator, which the application in Section 4.4 decomposes by ambient homology class.","marker":"[11]"},{"why":"Underpins the equivalence between transition homomorphisms on the nerve and Γ-coverings up to isomorphism, used in Proposition 3.9 and in reconstructing the covering from t.","marker":"[15]"},{"why":"Supplies the edge groupoid and combinatorial edge homotopy on which the paper's discrete groupoid formalism is built.","marker":"[6]"}],"fun_headline_variants":["Lift distances expose which data loops are true holes","Universal cover metrics label real cycles in point clouds","Nearest lifts decode ambient homology of point samples","Monodromy from lifts tells real holes from loops","Point-cloud cycles classified by lift distances in cover"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction assumes the universal cover and its deck group are explicitly known and the per-edge nearest-lift minimisation can actually be solved, and it needs each edge to sit in a unique minimising-geodesic homotopy class — a uniqueness the paper proves for generic point clouds and for Čech complexes below the convexity radius.","fun_headline_variants_meta":{"raw":{"variants":["Lift distances expose which data loops are true holes","Universal cover metrics label real cycles in point clouds","Nearest lifts decode ambient homology of point samples","Monodromy from lifts tells real holes from loops","Point-cloud cycles classified by lift distances in cover"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000179,"raw_usage":{"total_tokens":1392,"prompt_tokens":1127,"completion_tokens":265,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":743,"completion_tokens_details":{"reasoning_tokens":205}},"tokens_in":743,"tokens_out":265,"duration_ms":3511,"temperature":1.0,"reasoning_tokens":205,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T12:16:22.779125+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On the flat unit torus, represent lifts in $\\mathbb{R}^2$ and compute $t(ij) = \\arg\\min_{g\\in\\mathbb{Z}^2} \\|\\tilde x_i - g - \\tilde x_j\\|$; for any four-point configuration that Lemma 4.15 says yields a persistent cycle, compare $\\tau = t_{\\mathrm{ab}}(01)+t_{\\mathrm{ab}}(12)+t_{\\mathrm{ab}}(23)+t_{\\mathrm{ab}}(30)$ in $\\mathbb{Z}^2$ with the actual winding of the interpolating geodesic loop around the torus — a single mismatch would falsify the reconstruction. A complementary test on the real projective plane, with lifts in $S^2$ and a pair of antipodal points where minimising geodesics are not unique, would show whether the uniqueness premise (S1) is genuinely needed for the homology recovery.","supporting_citations":[{"cited_title":"Algebraic Topology","cited_arxiv_id":null,"evidence_quote":"Supplies the covering-space theory — path lifting, monodromy, deck transformations, and classification of coverings — on which the monodromy and induced-covering constructions rest."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the geometric premises: convexity radius, cut locus, and the fact that pairs outside mutual cut loci have unique minimising geodesics, which underpin assumption (S1)."},{"cited_title":"ˇCech cocycles for differential character- istic classes: an infinity-Lie theoretic construction","cited_arxiv_id":null,"evidence_quote":"Cited for the fact that geodesic balls of radius below the convexity radius form a good cover, enabling the Čech construction of Proposition 4.6."},{"cited_title":"Differential Geometry and Lie Groups , volume 12","cited_arxiv_id":null,"evidence_quote":"Provides the Riemannian-covering facts — geodesics lift to geodesics of the same length and deck actions are isometries — used in Proposition 4.3 and Corollary 4.4."},{"cited_title":"Stability of the cut locus and a central limit theorem for Fr´ echet means of Riemannian manifolds","cited_arxiv_id":null,"evidence_quote":"Supplies the stability of the cut locus under perturbation of the base point, used to show that point clouds with unique minimising geodesics form an open (and dense) set."},{"cited_title":"Curvature Sets Over Persistence Diagrams","cited_arxiv_id":null,"evidence_quote":"Defines the principal persistence measures and the four-point condition for a non-trivial first persistence generator, which the application in Section 4.4 decomposes by ambient homology class."},{"cited_title":"Springer, 1966","cited_arxiv_id":null,"evidence_quote":"Underpins the equivalence between transition homomorphisms on the nerve and Γ-coverings up to isomorphism, used in Proposition 3.9 and in reconstructing the covering from t."},{"cited_title":"Combinatorial group theory: a topological approach","cited_arxiv_id":null,"evidence_quote":"Supplies the edge groupoid and combinatorial edge homotopy on which the paper's discrete groupoid formalism is built."}],"review_version":1}