Pith. sign in

REVIEW 2 major objections 2 minor 1 cited by

Intrinsic Riemannian Cross-covariance for Manifold-valued Random Objects

T0 review · 2 major / 2 minor · reviewed 2026-06-27 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read An intrinsic cross-covariance for random objects on Riemannian manifolds is defined by parallel transport to a common tangent space.

desk verdict The paper defines manifold cross-covariance via parallel transport to a common tangent space, but path dependence on curved manifolds needs checking against the intrinsic claim. read the letter →

arxiv 2606.10212 v2 pith:US6MOS2Z submitted 2026-06-08 math.ST stat.MLstat.TH

Carlos Soto , Cheng Wang , Yujing Huang , Xiaoyu Chen This is my paper · ORCID
classification math.STstat.MLstat.TH
keywords Riemannianmanifoldcross-covarianceparalleltransportmanifold-valueddataintrinsicstatisticsSPDmatricesKendallshapespace
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper develops a definition of covariance and correlation for data points that live on curved Riemannian manifolds rather than flat Euclidean space. The key step is to use parallel transport to move the local variations around each point into one shared tangent space. A sympathetic reader would care because many modern data types, such as shapes or positive-definite matrices, naturally sit on such manifolds, yet standard covariance tools break down there. The construction is shown to be free of arbitrary coordinate choices and to recover the usual Euclidean properties when the manifold is flat.

What carries the argument

The intrinsic Riemannian cross-covariance, which maps variations at distinct base points to a single tangent space by parallel transport.

What would settle it

Demonstrating that the computed covariance matrix changes when a different base point or different transport path is chosen on the same dataset would falsify the claim of intrinsic, coordinate-independent covariance.

Watch

Extended reading notes

Core claim

The proposed intrinsic Riemannian cross-covariance is obtained by transporting local variations of manifold-valued random objects to a common tangent space via parallel transport. This yields a second-order descriptor that does not depend on arbitrary coordinate choices. The construction inherits the symmetry, positive-semidefiniteness, and other desirable properties of Euclidean covariance, and its asymptotic behavior is characterized.

Load-bearing premise

Parallel transport along the manifold can map local variations from distinct base points into one common tangent space without creating coordinate dependence or losing intrinsic information.

Editorial extensions

If this is right

  • It enables consistent second-order analysis for manifold-valued data such as shapes in Kendall space.
  • The asymptotic characterization supports large-sample inference on manifolds.
  • Estimators can be computed numerically on spheres and SPD manifolds.
  • Real-data experiments on heart valve shapes verify the coordinate-independence and other properties.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the construction works on general manifolds, it could be used to define manifold versions of canonical correlation analysis.
  • Extensions to time-series or spatial dependence on manifolds would follow naturally from the cross-covariance definition.
  • Comparison with extrinsic covariance approaches on the same data could quantify the gain from intrinsicness.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The paper introduces an intrinsic Riemannian cross-covariance for manifold-valued random objects. The approach defines covariance and correlation by transporting local variations to a common tangent space via parallel transport, yielding a second-order descriptor that is independent of arbitrary coordinate choices. It establishes that the proposed covariance inherits desirable properties of its Euclidean counterparts and characterizes its asymptotic behavior. Numerical studies on spheres and SPD manifolds, together with real-data experiments on heart valve shapes in Kendall's shape space, demonstrate the effectiveness of the estimators.

Significance. If the construction is path-independent and the asymptotic results hold with proper error analysis, this would provide a fundamental tool for second-order statistics, dependence modeling, and dimension reduction on Riemannian manifolds common in machine learning (e.g., shapes, SPD matrices). The numerical and real-data experiments would strengthen its practical utility if they include appropriate baselines.

major comments (2)
  1. [Definition of the Riemannian cross-covariance (abstract and §3)] The central construction transports tangent vectors from distinct base points x, y to a common T_pM via parallel transport. On manifolds with nonzero sectional curvature, this transport is path-dependent (holonomy). The manuscript must specify the connecting curves (e.g., geodesics) and either prove invariance of the resulting bilinear form under path choice or discuss the resulting ambiguity. This directly affects the claim of independence from arbitrary choices and inheritance of Euclidean properties.
  2. [Asymptotic results and properties (abstract and §4)] The abstract asserts that the covariance 'inherits desirable properties of its Euclidean counterparts' and 'characterize[s] its asymptotic behavior,' yet supplies no derivations, proofs, or explicit error bounds. A concrete statement (e.g., the rate in the central limit theorem or consistency result) is required to verify these claims, especially given curvature effects on the transport.
minor comments (2)
  1. [Numerical experiments] Numerical studies are referenced without reported baselines, comparison methods, or quantitative metrics (e.g., MSE against Euclidean or other manifold covariance estimators).
  2. [Real-data experiments] The real-data section on Kendall shape space would benefit from explicit pseudocode or algorithmic description of how the cross-covariance estimator is implemented for the heart valve data.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive review. We address the two major comments point by point below, indicating the revisions we will make to the manuscript.

read point-by-point responses

  1. Referee: [Definition of the Riemannian cross-covariance (abstract and §3)] The central construction transports tangent vectors from distinct base points x, y to a common T_pM via parallel transport. On manifolds with nonzero sectional curvature, this transport is path-dependent (holonomy). The manuscript must specify the connecting curves (e.g., geodesics) and either prove invariance of the resulting bilinear form under path choice or discuss the resulting ambiguity. This directly affects the claim of independence from arbitrary choices and inheritance of Euclidean properties.

    Authors: We appreciate the referee for identifying this important technical point. Our construction uses parallel transport along the unique minimizing geodesic from each base point (x or y) to the common point p; this is the canonical choice on the manifolds studied (spheres, SPD matrices, Kendall shape space). While holonomy renders the transport path-dependent on manifolds with nonzero curvature, specifying the geodesic removes ambiguity for any fixed triple (x, y, p). In the revision we will (i) state this choice explicitly in §3, (ii) add a remark clarifying that the resulting bilinear form is independent of coordinate charts (as parallel transport is intrinsic) yet may depend on path when multiple geodesics exist, and (iii) note that the Euclidean-like properties hold under the chosen transport. We will also discuss the flat-manifold case where path independence is automatic. revision: yes

  2. Referee: [Asymptotic results and properties (abstract and §4)] The abstract asserts that the covariance 'inherits desirable properties of its Euclidean counterparts' and 'characterize[s] its asymptotic behavior,' yet supplies no derivations, proofs, or explicit error bounds. A concrete statement (e.g., the rate in the central limit theorem or consistency result) is required to verify these claims, especially given curvature effects on the transport.

    Authors: Section 4 and the supplementary material contain the consistency and central-limit results for the estimator, with rates derived via the exponential map and its differential to account for curvature and parallel transport. The estimator is shown to be sqrt(n)-consistent under standard moment and curvature bounds. To make these claims more transparent, we will insert a concise statement of the main asymptotic theorem (including the explicit rate) into the body of §4, together with a short proof sketch that highlights the role of the transport operator. Full derivations remain in the supplement. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; derivation relies on standard differential geometry

full rationale

The paper's central construction defines the Riemannian cross-covariance by transporting tangent vectors via parallel transport to a common tangent space, then claims this yields a coordinate-independent second-order statistic that inherits Euclidean properties. This definition is introduced directly from Riemannian geometry operations rather than being derived from or fitted to the target covariance itself. No equations, self-citations, or ansatzes are shown that reduce the claimed result to its own inputs by construction (e.g., no parameter fitted to data then renamed as prediction, no uniqueness theorem imported from the authors' prior work, and no renaming of known empirical patterns). The abstract and provided text position the result as self-contained against external benchmarks from differential geometry, with numerical studies serving as verification rather than circular justification. The path-dependence concern raised externally pertains to correctness of the intrinsic claim, not to any reduction of the derivation chain.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

Approach relies on standard differential geometry without new free parameters or invented entities visible in the abstract.

assumptions (1)
  • standard math Riemannian manifolds admit a well-defined parallel transport operation
    Invoked to move local variations to a common tangent space; standard background result in differential geometry.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Intrinsic Riemannian Cross-covariance for Manifold-valued Random Objects." pith.science (2026). https://pith.science/paper/US6MOS2Z

@misc{pith2026260610212,
  author       = {Pith},
  title        = {Pith review of: Intrinsic Riemannian Cross-covariance for Manifold-valued Random Objects},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/US6MOS2Z}},
  note         = {Machine review of arXiv:2606.10212}
}
read the original abstract

Covariance estimation yields a fundamental second-order statistic underlying representation learning, dimension reduction, and dependence modeling. While covariance has been well understood in Euclidean spaces, it is ill-defined for random objects residing on nonlinear Riemannian manifolds, which increasingly arise in modern machine learning applications involving shapes, symmetric positive definite (SPD) matrices, etc. This paper introduces an intrinsic Riemannian cross-covariance for manifold-valued random objects. Our approach defines covariance and correlation by transporting local variations to a common tangent space via parallel transport, yielding a second-order descriptor that is independent of arbitrary coordinate choices. We establish that the proposed covariance inherits desirable properties of its Euclidean counterparts and characterize its asymptotic behavior. Numerical studies on spheres and SPD manifolds, together with real-data experiments on heart valve shapes in Kendall's shape space, demonstrate the effectiveness of our estimators and verify the stated properties. Our results position the Riemannian covariance as a fundamental tool for second-order learning and analysis in non-Euclidean representation spaces.

Figures

Figures reproduced from arXiv: 2606.10212 by the authors.

Figure 1
Figure 1. Geodesic Evaluation on Sphere. The shaded area indicates 95% confidence intervals [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Comprehensive Comparison on Sphere with parallel-transported dataset. Correlation, [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Comprehensive Comparison on the SPD Manifold with parallel-transported dataset. Correla [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: An illustration of a dataset (n = 100) on S 2 and its dependent dataset, together with the respective Fréchet mean and the joint Fréchet mean (midpoint of two Fréchet mean). area represents the confidence interval of 2 standard errors. It can be seen that our estimates…
Figure 5
Figure 5. Figure 5: Comparison of Riemannian correlation (Ours), [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]
Figure 6
Figure 6. Figure 6: Comparison of Riemannian correlation, A&P correlation and CCA estimator. Solid line correspond to samples sharing the same Fréchet mean while dotted line represents results from data {Yi} transported to (0, −1, 0). Parameters: τ = π/6, η = π/4, with noise level σϵ = 0.…
Figure 7
Figure 7. Figure 7: Comparison of Riemannian correlation (Ours), [PITH_FULL_IMAGE:figures/full_fig_p026_7.png]
Figure 8
Figure 8. Figure 8: Convergence of Correlation Estimates on SPD Manifold. [PITH_FULL_IMAGE:figures/full_fig_p027_8.png]
Figure 9
Figure 9. Figure 9: Geodesic Evaluation on SPD Manifold. Left: Original dataset (non-transport). Right: Parallel-transported dataset (transport). Our method produces nearly constant estimates (SE ≈ 0.008) across all evaluation points, consistent with footpoint invariance (Theorem 3.6). Th…
Figure 10
Figure 10. Figure 10: Comprehensive Comparison on SPD Manifold. [PITH_FULL_IMAGE:figures/full_fig_p028_10.png]
Figure 11
Figure 11. Figure 11: Geodesic Simulation and Transport Effects on SPD Manifold. [PITH_FULL_IMAGE:figures/full_fig_p029_11.png]
Figure 12
Figure 12. Figure 12: Angle Effect and Independence on SPD Manifold. [PITH_FULL_IMAGE:figures/full_fig_p029_12.png]
Figure 13
Figure 13. Figure 13: Effect of Mean Separation on Estimator Performance. [PITH_FULL_IMAGE:figures/full_fig_p030_13.png]
Figure 16
Figure 16. Figure 16: 30 [PITH_FULL_IMAGE:figures/full_fig_p030_16.png]
Figure 14
Figure 14. Figure 14: Illustrative examples of inner and outer ventricular walls from the four groups. [PITH_FULL_IMAGE:figures/full_fig_p031_14.png]
Figure 15
Figure 15. Figure 15: Footpoint invariance on connectome data. Our method (orange) is constant; A&P (blue) [PITH_FULL_IMAGE:figures/full_fig_p031_15.png]
Figure 16
Figure 16. Figure 16: Bootstrap 95% CI for Riemannian correlation. Sign reversal between groups is significant [PITH_FULL_IMAGE:figures/full_fig_p031_16.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Riemannian View on Active Subspaces

    math.NA 2026-07 conditional novelty 7.0 of 10

    A parallel-transport-based intrinsic generalization of active subspaces to Riemannian manifolds, with second-order intrinsic/extrinsic equivalence and 2-sphere demonstrations.

Reference graph

Works this paper leans on

12 extracted references · cited by 1 Pith paper

  1. [1]

    Variance:Cov(X, X) =Var(X)

  2. [2]

    Symmetry:Cov(X, Y) = Cov(Y, X)

  3. [3]

    Scaling:Cov(aX, Y) =aCov(X, Y)

  4. [4]

    Uncorrelatedness: IfXandYare independent, thenCov(X, Y) = 0

  5. [5]

    Invariance to constant:Cov(X+c, Y) = Cov(X, Y)

  6. [6]

    Variance of a constant:Cov(X, c) = 0

  7. [7]

    More generally, Cov  X i aiXi, X j bjYj   = X i X j aibj Cov(Xi, Yj)

    Additivity:Cov(X 1 +X 2, Y) = Cov(X 1, Y) + Cov(X2, Y). More generally, Cov  X i aiXi, X j bjYj   = X i X j aibj Cov(Xi, Yj). With the variance given by Var(X) = Cov(X, X), the Pearson correlation is defined as ρ(X, Y) = Cov(X, Y)p Var(X) p Var(Y) Thus we have the following properties for correlation:

  8. [8]

    Bounded:ρ(X, Y)∈[−1,1]

Show all 12 references
  1. [9]

    Symmetry:ρ(X, Y) =ρ(Y, X)

  2. [10]

    Scale invariance:ρ(aX+b, cY+d) =sign(ac)ρ(X, Y)

  3. [11]

    16 D Covariance Under Basis Rotation The parallel transport can be interpreted as a rotation of basis when the manifold is viewed as embedded in some ambient space, i.e

    Self-correlation: if X is nondegenerate,ρ(X, X) = 1. 16 D Covariance Under Basis Rotation The parallel transport can be interpreted as a rotation of basis when the manifold is viewed as embedded in some ambient space, i.e. Rn. Suppose that we want to move a vector w from the t...

  4. [12]

    Proof of Theorem 3.6

    Consequently, tr ΣA&P X,Y (q) ̸= tr ΣA&P X,Y (p) , which proves that the Abuqrais-Pigoli covariance depends on the choice of footpoint. Proof of Theorem 3.6. tr [ΣX,Y (p)] = tr E (Γp µ logµ X)(Γ p ν logν Y) T =E tr (Γp µ logµ X)(Γ p ν logν Y) T =E tr (Γq pΓp µ logµ X)(Γ q pΓp ...

    2016

Pith tools

Reviewed June 27, 2026 · model on record in the stance chip above.