pith. sign in

arxiv: 2606.31465 · v1 · pith:I5Q6CLRTnew · submitted 2026-06-30 · 📊 stat.ME · math.ST· stat.TH

Functional Principal Component Analysis for Manifold-Indexed Data

Chang Jun Im , Jeong Min Jeon This is my paper

Pith reviewed 2026-07-01 04:24 UTC · model grok-4.3

classification 📊 stat.ME math.STstat.TH
keywords functional principal component analysismanifold-indexed dataRiemannian manifoldintrinsic kernel estimationuniform convergence ratescovariance operatorspectral perturbationgeodesic distance
0
0 comments X

The pith

FPCA for scalar functions indexed by d-dimensional Riemannian manifolds achieves uniform convergence rates governed by the manifold's intrinsic dimension.

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

The paper develops a version of functional principal component analysis for data consisting of scalar-valued functions defined on a compact d-dimensional Riemannian manifold rather than on an interval. Estimators for the mean and covariance are constructed using kernels that depend on geodesic distance together with a correction for the manifold's volume measure. Uniform rates of convergence are proved for these estimators, for the covariance operator in Hilbert-Schmidt and operator norms, and for the eigenvalues and eigenfunctions obtained by spectral perturbation. The rates explicitly depend on d and recover the classical one-dimensional functional-data rates when the manifold is one-dimensional. The analysis accommodates arbitrary subject-specific sampling densities and both per-observation and per-subject weighting schemes.

Core claim

Intrinsic kernel estimators for the mean and covariance functions are constructed from geodesic distances on the manifold together with a Riemannian volume-density correction. These estimators are shown to satisfy uniform convergence rates whose dependence on sampling density is governed by the intrinsic dimension d of the indexing manifold. Hilbert-Schmidt and operator-norm bounds on the estimated covariance operator follow, and spectral perturbation then yields rates for the eigenvalues and eigenfunctions. The same rates hold under both equal-weight-per-observation and equal-weight-per-subject schemes and reduce exactly to the classical one-dimensional boundary case when d equals one.

What carries the argument

Intrinsic kernel estimators that replace Euclidean distance with geodesic distance on the manifold and incorporate a Riemannian volume-density correction.

If this is right

  • The sparse-to-dense transition threshold for consistent estimation scales with the intrinsic dimension d of the indexing manifold.
  • When the manifold is one-dimensional the rates coincide with those of classical functional data analysis on an interval.
  • The estimators remain consistent under subject-specific sampling frequencies that vary arbitrarily across observations.
  • Spectral perturbation applied to the covariance-operator bounds directly supplies rates for the eigenfunctions and eigenvalues used in dimension reduction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same geometric correction could be inserted into other functional-data procedures such as functional linear regression or clustering when the domain is a manifold.
  • Sampling designs for manifold-indexed experiments could be optimized by allocating more points where the volume density is low to maintain the predicted rates.
  • The framework suggests that ignoring the manifold geometry and working in ambient coordinates will incur a bias whose size grows with d.

Load-bearing premise

The intrinsic kernel classes satisfy VC-type empirical-process conditions that permit uniform stochastic bounds even when the kernels are not Lipschitz.

What would settle it

Empirical convergence rates on a two-dimensional manifold that fail to exhibit the predicted slower transition from sparse to dense regimes compared with the one-dimensional case.

Figures

Figures reproduced from arXiv: 2606.31465 by Chang Jun Im, Jeong Min Jeon.

Figure 1
Figure 1. Figure 1: Estimated mean HRTF surfaces for the SONICOM data, displayed on the unit [PITH_FULL_IMAGE:figures/full_fig_p043_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: First three eigenfunctions estimated by the proposed intrinsic sphere FPCA for [PITH_FULL_IMAGE:figures/full_fig_p044_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: First three eigenfunctions estimated by the naive azimuth–elevation FPCA [PITH_FULL_IMAGE:figures/full_fig_p098_3.png] view at source ↗
read the original abstract

Functional principal component analysis (FPCA) is a central tool for dimension reduction and covariance analysis in functional data analysis. We study FPCA for discretely observed scalar-valued functional data indexed by a compact d-dimensional Riemannian manifold M; that is, each subject is modeled as a random function from M to R. This setting is distinct from manifold-valued functional data, where the function values themselves lie on a manifold. We develop intrinsic kernel estimators for the mean and covariance functions using geodesic distances and a Riemannian volume-density correction. The proposed framework accommodates general subject-specific sampling frequencies and includes both equal-weight-per-observation and equal-weight-per-subject schemes. The uniform stochastic analysis uses VC-type empirical-process conditions for intrinsic kernel classes, together with clustered empirical-process compatibility conditions, allowing non-Lipschitz kernels under the stated assumptions. We establish uniform convergence rates for the mean and covariance estimators, Hilbert-Schmidt and operator-norm error bounds for the estimated covariance operator, and convergence rates for eigenvalues and eigenfunctions via spectral perturbation. The rates show that the sparse-to-dense transition is governed by the intrinsic dimension of the indexing manifold, reducing to the classical one-dimensional boundary when d=1. Simulations on S^1 and S^2 and a SONICOM head-related transfer function analysis illustrate the method and show modest but consistent improvements over a coordinate-based baseline when intrinsic geometry is ignored.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper develops intrinsic kernel estimators for the mean and covariance of scalar-valued functions indexed by a compact d-dimensional Riemannian manifold, using geodesic distances and Riemannian volume-density corrections. It accommodates subject-specific sampling designs with both per-observation and per-subject weighting. Under VC-type empirical-process conditions on intrinsic kernel classes and clustered empirical-process compatibility conditions (allowing non-Lipschitz kernels), the authors derive uniform convergence rates for the mean and covariance estimators, Hilbert-Schmidt and operator-norm bounds on the estimated covariance operator, and eigenvalue/eigenfunction convergence rates via spectral perturbation. These rates depend on the intrinsic dimension d and reduce to the classical one-dimensional case when d=1. The claims are illustrated by simulations on S^1 and S^2 and an application to SONICOM head-related transfer function data, showing modest gains over a coordinate-based baseline.

Significance. If the stochastic analysis holds, the work provides a geometrically intrinsic extension of FPCA to manifold-indexed functional data, with rates that correctly reflect the role of intrinsic dimension in the sparse-to-dense transition. This is a substantive advance for applications where the indexing domain has non-Euclidean geometry. The explicit treatment of both weighting schemes and the reduction to the d=1 boundary case are clear strengths.

major comments (1)
  1. [Assumptions and uniform stochastic analysis] Assumptions (abstract and the uniform stochastic analysis section): The clustered empirical-process compatibility conditions are asserted to enable the uniform rates and d-dependent sparse-to-dense transition for the specific geodesic-distance kernels with Riemannian correction under subject-specific sampling. However, no explicit verification or argument is supplied that these kernels satisfy the compatibility conditions when d>1. This is load-bearing for the central claims on uniform convergence, operator-norm bounds, and the dimension-dependent rates.
minor comments (2)
  1. [Abstract] Abstract: the phrasing 'allowing non-Lipschitz kernels under the stated assumptions' could be expanded to indicate whether the kernels are Lipschitz or not in the manifold setting.
  2. [Simulations] Simulations section: the comparison to the coordinate-based baseline would benefit from reporting the precise manifold dimension and sampling design used in the S^2 case to make the improvement attributable to intrinsic geometry fully transparent.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thorough review and for identifying this important point about the assumptions. We address the major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: Assumptions (abstract and the uniform stochastic analysis section): The clustered empirical-process compatibility conditions are asserted to enable the uniform rates and d-dependent sparse-to-dense transition for the specific geodesic-distance kernels with Riemannian correction under subject-specific sampling. However, no explicit verification or argument is supplied that these kernels satisfy the compatibility conditions when d>1. This is load-bearing for the central claims on uniform convergence, operator-norm bounds, and the dimension-dependent rates.

    Authors: We agree that an explicit verification for the geodesic-distance kernels with Riemannian volume-density correction is not supplied in the current manuscript and would strengthen the presentation. In the revision we will add a short dedicated paragraph (or appendix subsection) showing that these kernels satisfy the VC-type conditions and the clustered empirical-process compatibility conditions for d>1. The argument relies on the fact that, locally, the manifold is Euclidean, the geodesic distance is bi-Lipschitz in normal coordinates, and the volume-density factor is C^infty; these properties ensure that the kernel classes have polynomial covering numbers controlled by the intrinsic dimension d, exactly as required by the general theorems. The same local-chart argument also verifies the compatibility conditions under subject-specific sampling designs. We will make this verification self-contained and independent of the main convergence theorems. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper derives uniform convergence rates, HS/operator-norm bounds, and spectral perturbation results for eigenvalues/eigenfunctions from standard empirical-process theory applied to intrinsic kernel estimators under explicitly stated VC-type conditions and clustered compatibility assumptions. These assumptions are external inputs to the argument rather than outputs, and no equations or claims reduce by construction to fitted parameters, self-definitions, or self-citation chains. The sparse-to-dense transition governed by manifold dimension d follows directly from the perturbation analysis without circular reduction. The derivation is self-contained against external benchmarks in empirical process theory.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review yields no concrete free parameters or invented entities; the central claims rest on standard empirical-process assumptions whose details are not supplied.

axioms (1)
  • domain assumption VC-type empirical-process conditions for intrinsic kernel classes together with clustered empirical-process compatibility conditions
    Invoked to obtain uniform stochastic analysis and allow non-Lipschitz kernels.

pith-pipeline@v0.9.1-grok · 5772 in / 1175 out tokens · 44535 ms · 2026-07-01T04:24:18.545178+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [1]

    and Taachouche, N

    Bouzebda, S. and Taachouche, N. (2023). Rates of the strong uniform consistency for the kernel-type regression function estimators with general kernels on manifolds. Mathematical Methods of Statistics , 32, 27-80

    work page 2023

  2. [2]

    Cai, T. T. and Yuan, M. (2011). Optimal estimation of the mean function based on discretely sampled functional data: Phase transition. The Annals of Statistics , 39, 2330-2355

    work page 2011

  3. [3]

    Chavel, I. (2006). Riemannian Geometry: A Modern Introduction (2nd ed.). Cambridge University Press

    work page 2006

  4. [4]

    and Kahan, W

    Davis, C. and Kahan, W. M. (1970). The rotation of eigenvectors by a perturbation. III. SIAM Journal on Numerical Analysis , 7, 1-46

    work page 1970

  5. [5]

    and M\" u ller, H.-G

    Dai, X. and M\" u ller, H.-G. (2018). Principal component analysis for functional data on Riemannian manifolds. Biometrika , 105(1), 177-190

    work page 2018

  6. [6]

    Do Carmo, M. P. (1992). Riemannian Geometry . Birkh\"auser

    work page 1992

  7. [7]

    and Mason, D

    Einmahl, U. and Mason, D. M. (2005). Uniform in bandwidth consistency of kernel-type function estimators. The Annals of Statistics , 33, 1380-1403

    work page 2005

  8. [8]

    Engel, I., Daugintis, R., Vicente, T., Hogg, A. O. T., Pauwels, J., Tournier, A. J. R. and Picinali, L. (2023). The SONICOM HRTF dataset. Journal of the Audio Engineering Society , 71, 241--253

    work page 2023

  9. [9]

    and Nickl, R

    Gin \'e , E. and Nickl, R. (2016). Mathematical Foundations of Infinite-Dimensional Statistical Models. Cambridge University Press

    work page 2016

  10. [10]

    and Wang, J.-L

    Hall, P., M\" u ller, H.-G. and Wang, J.-L. (2006). Properties of principal component methods for functional and longitudinal data analysis. The Annals of Statistics , 34, 1493-1537

    work page 2006

  11. [11]

    M., Park, B

    Jeon, J. M., Park, B. U. and Van Keilegom, I. (2021). Additive regression for non-Euclidean responses and predictors. Annals of Statistics , 49, 2611-2641

    work page 2021

  12. [12]

    and Hsing, T

    Li, Y. and Hsing, T. (2010). Uniform convergence rates for nonparametric regression and principal component analysis in functional data. The Annals of Statistics , 38, 3321-3351

    work page 2010

  13. [13]

    and Yao, F

    Lin, Z. and Yao, F. (2019). Intrinsic Riemannian functional data analysis. Annals of Statistics , 47, 3533-3577

    work page 2019

  14. [14]

    and Pollard, D

    Nolan, D. and Pollard, D. (1987). U-processes: Rates of convergence. The Annals of Statistics , 15, 780-799

    work page 1987

  15. [15]

    Pelletier, B. (2005). Kernel density estimation on Riemannian manifolds. Statistics and Probability Letters , 73, 297-304

    work page 2005

  16. [16]

    Pelletier, B. (2006). Non-parametric regression estimation on closed Riemannian manifolds. Journal of Nonparametric Statistics , 18, 57-67

    work page 2006

  17. [17]

    Ramsay, J. O. and Silverman, B. W. (2005). Functional Data Analysis . Springer

    work page 2005

  18. [18]

    and Yao, F

    Shao, L., Lin, Z. and Yao, F. (2022). Intrinsic Riemannian functional data analysis for sparse longitudinal observations. Annals of Statistics , 50, 1696-1721

    work page 2022

  19. [19]

    and Wellner, J

    van der Vaart, A. and Wellner, J. A. (1996). Weak Convergence and Empirical Processes: With Applications to Statistics . Springer

    work page 1996

  20. [20]

    and Wang, J.-L

    Yao, F., M\" u ller, H.-G. and Wang, J.-L. (2005). Functional data analysis for sparse longitudinal data. Journal of the American Statistical Association , 100, 577-590

    work page 2005

  21. [21]

    and Wang, J.-L

    Zhang, X. and Wang, J.-L. (2016). From sparse to dense functional data and beyond. The Annals of Statistics , 44, 2281-2321

    work page 2016