pith. sign in

arxiv: 2508.21656 · v2 · submitted 2025-08-29 · 🧮 math.ST · stat.TH

Asymptotic equivalence of non-parametric regression with spherical regressors and Gaussian white noise

Martin Kroll This is my paper

Pith reviewed 2026-05-18 20:26 UTC · model grok-4.3

classification 🧮 math.ST stat.TH
keywords asymptotic equivalenceLe Cam distancenonparametric regressionspherical t-designsSobolev ballsBesov ballsGaussian white noisespherical regressors
0
0 comments X

The pith

Regression on the sphere with t-designs or uniform points is asymptotically equivalent to Gaussian white noise experiments over Sobolev and Besov balls.

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

The paper shows that non-parametric regression experiments on the sphere, using either fixed spherical t-design sampling points or random uniform designs, become equivalent in Le Cam's sense to a sequence of Gaussian white noise experiments as the number of observations grows to infinity. This global equivalence holds over spherical Sobolev balls for both design types and over spherical Besov balls for the fixed design. A reader would care because the white noise model is simpler and better understood, so results about estimation rates and procedures can transfer directly to the spherical regression setting. The paper also gives matching non-equivalence results to show the smoothness requirements are essentially necessary.

Core claim

We show that the corresponding regression experiments are asymptotically equivalent, in the sense of Le Cam, to the same sequence of Gaussian white noise experiments as the sample size tends to infinity. More precisely, global asymptotic equivalence is established over spherical Sobolev balls (for both the fixed and the random uniform design case) and over spherical Besov balls (for the fixed design case). Matching non-equivalence results demonstrate that the imposed smoothness assumptions are essentially sharp.

What carries the argument

Le Cam asymptotic equivalence between the spherical regression experiment and the Gaussian white noise experiment, established uniformly over the function classes.

If this is right

  • Optimal estimation procedures and rates known for the white noise model apply directly to the spherical regression problem with the same asymptotic performance.
  • The equivalence allows uniform control of risks over the entire function class rather than pointwise.
  • For smoothness below the threshold the experiments remain distinguishable, so the white noise model cannot be used as a proxy.

Where Pith is reading between the lines

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

  • The same design conditions might permit equivalence results on other compact manifolds using comparable approximation properties.
  • Finite-sample simulations comparing risks in the regression and white noise settings could check how quickly the asymptotic equivalence appears.
  • Adaptive methods developed in white noise could be ported to spherical data without additional loss in rate.

Load-bearing premise

The sampling points must form spherical t-designs or random uniform designs on the sphere of arbitrary dimension, and the unknown functions must lie in spherical Sobolev or Besov balls of sufficient smoothness.

What would settle it

A calculation showing that the Le Cam distance between the regression experiment and the white noise experiment fails to tend to zero when the smoothness index drops below the threshold required for the given design and function class.

read the original abstract

We study the asymptotic behaviour of both spherical $t$-designs and random uniform designs as the set of sampling points in non-parametric regression with spherical regressors of arbitrary dimension. We show that the corresponding regression experiments are asymptotically equivalent, in the sense of Le Cam, to the same sequence of Gaussian white noise experiments as the sample size tends to infinity. More precisely, global asymptotic equivalence is established over spherical Sobolev balls (for both the fixed and the random uniform design case) and over spherical Besov balls (for the fixed design case). Matching non-equivalence results demonstrate that the imposed smoothness assumptions are essentially sharp.

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

2 major / 3 minor

Summary. The manuscript establishes that nonparametric regression experiments on the sphere, using either spherical t-designs or random uniform designs as sampling points, are asymptotically equivalent in the Le Cam sense to a sequence of Gaussian white noise experiments as the sample size n tends to infinity. Global equivalence is shown over spherical Sobolev balls for both design types and over spherical Besov balls for the fixed-design case only; matching non-equivalence results are provided to demonstrate that the required smoothness indices are essentially sharp.

Significance. If the central claims hold, the work extends classical Le Cam equivalence results from Euclidean domains to spheres of arbitrary dimension, supplying a rigorous basis for transferring minimax rates, adaptive estimation procedures, and risk bounds from the white-noise model to spherical regression. The treatment of both deterministic t-designs and random uniform designs, together with the inclusion of Besov spaces, strengthens the applicability to modern spherical data settings.

major comments (2)
  1. [§4] §4 (Le Cam distance for random designs): the uniform control of the deficiency distance over the Sobolev ball appears to rely on concentration of the empirical measure around the uniform measure; the argument should explicitly bound the remainder term when the dimension d is allowed to grow (even slowly) with n, as the abstract claims arbitrary dimension.
  2. [Theorem 5.2] Theorem 5.2 (non-equivalence for Besov balls): the construction used to exhibit positive Le Cam distance when the smoothness index falls below the threshold must be checked for compatibility with the spherical t-design quadrature error; if the counter-example functions are not well approximated by the design, the sharpness claim may not be fully load-bearing.
minor comments (3)
  1. [Definition 2.3] The notation for the spherical Sobolev and Besov norms (Definition 2.3) should include an explicit reference to the underlying spherical harmonic expansion to avoid ambiguity when d varies.
  2. [Figure 1] Figure 1 (illustration of t-designs) would benefit from a caption that states the exact degree and number of points used, facilitating reproducibility.
  3. [Introduction] A short remark comparing the obtained equivalence rates with the corresponding Euclidean white-noise results (e.g., from Brown & Low or Nussbaum) would help readers gauge the dimensional dependence.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading of the manuscript, the positive overall assessment, and the recommendation for minor revision. We address the two major comments point by point below, indicating the revisions made.

read point-by-point responses

  1. Referee: [§4] §4 (Le Cam distance for random designs): the uniform control of the deficiency distance over the Sobolev ball appears to rely on concentration of the empirical measure around the uniform measure; the argument should explicitly bound the remainder term when the dimension d is allowed to grow (even slowly) with n, as the abstract claims arbitrary dimension.

    Authors: We thank the referee for this observation. The results are stated for spheres of arbitrary but fixed dimension d, with n tending to infinity; this is the regime in which the abstract and all theorems are formulated. For any fixed d the empirical measure concentrates around the uniform measure at the required rate, and the resulting bound on the deficiency distance is uniform over the Sobolev ball. To make the dependence on d fully transparent we have added a short remark after the proof of the random-design equivalence (new Remark 4.3) that records the explicit remainder term and notes that the same argument continues to hold provided d grows slower than any positive power of log n. No alteration of the main statements is required. revision: yes

  2. Referee: [Theorem 5.2] Theorem 5.2 (non-equivalence for Besov balls): the construction used to exhibit positive Le Cam distance when the smoothness index falls below the threshold must be checked for compatibility with the spherical t-design quadrature error; if the counter-example functions are not well approximated by the design, the sharpness claim may not be fully load-bearing.

    Authors: We appreciate the referee drawing attention to this point. The counter-example functions in the proof of Theorem 5.2 are linear combinations of spherical harmonics of degree roughly equal to the critical smoothness index. Because a t-design integrates exactly all harmonics up to degree t and t is chosen larger than this degree, the quadrature error vanishes identically on the leading terms. The tail of the harmonic expansion is controlled in the Besov norm and contributes an o(1) term to the Le Cam distance. We have inserted a brief paragraph immediately after the construction (new display (5.8) and the following two sentences) that makes this verification explicit. The sharpness statement therefore remains valid for both the t-design and the uniform-design settings. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained via standard Le Cam theory and external results

full rationale

The paper establishes global asymptotic equivalence (in Le Cam distance) between nonparametric regression on the sphere (with t-design or random uniform sampling) and a Gaussian white noise experiment, uniformly over spherical Sobolev balls (both designs) and Besov balls (fixed design). This is achieved by verifying the usual conditions for Le Cam equivalence—such as contiguity, local asymptotic normality, and approximation of the regression operator by the white noise model—using known properties of spherical designs and Sobolev/Besov spaces on the sphere. These supporting facts are drawn from prior independent literature on harmonic analysis and design theory, not from any self-referential definition, fitted parameter renamed as prediction, or load-bearing self-citation chain within the paper. The matching non-equivalence results for insufficient smoothness are likewise external sharpness statements. No equation reduces to its own input by construction, and the central claim retains independent mathematical content.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper builds on standard frameworks from asymptotic statistics without introducing new free parameters, ad-hoc axioms, or invented entities; it relies on established properties of spherical designs and function spaces.

axioms (2)
  • standard math Le Cam's theory of asymptotic equivalence of statistical experiments applies to the regression and white noise models under the given conditions.
    Invoked to establish the equivalence of the experiments as sample size tends to infinity.
  • domain assumption Spherical Sobolev and Besov balls are well-defined function classes on the sphere with the necessary embedding and approximation properties.
    Used as the parameter spaces over which global equivalence holds.

pith-pipeline@v0.9.0 · 5619 in / 1433 out tokens · 38976 ms · 2026-05-18T20:26:36.224701+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.