REVIEW 2 major objections 4 minor 10 references
A note on the geodesic normal distribution on the sphere
T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read The paper derives a tangent-space-free formula for the anisotropic geodesic normal distribution on the sphere and claims its density contours are exactly spherical ellipses.
desk verdict The tangent-free rewrite of Hauberg's density is correct, but the paper's advertised spherical-ellipse characterization is false because the log map is not a global isometry. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The key object is the logarithm map $\operatorname{Log}_\mu(x)=\frac{x-(x^\top\mu)\mu}{\sin d_G(x,\mu)}\arccos(x^\top\mu)$, which sends the sphere minus the antipodal point into the tangent space at $\mu$. The paper uses the identity $\operatorname{Log}_\mu(x)^\top\Lambda\operatorname{Log}_\mu(x)=\arccos^2(x^\top\mu)\frac{\lambda_1(x^\top\eta)^2+\lambda_2(x^\top\xi)^2}{(x^\top\eta)^2+(x^\top\xi)^2}$ to eliminate the tangent space from the density. For the contours, the machinery is the equivalence between the level set of this quadratic form and the preimage under $\operatorname{Log}_\mu$ of an ellipse in the tangent space; the paper relies on $\operatorname{Log}_\mu$ being an isometry to transfer Euclidean focal-distance ellipses back to geodesic focal-distance ellipses on the sphere.
What would settle it
Pick $\alpha=\pi/6$, $\beta=\pi/12$, set $\gamma$ by $\cos\gamma=\cos\alpha/\cos\beta$, take $\mu=(1,0,0)$ and $\eta=(0,1,0)$, and generate points $x$ satisfying equation (7). Compare $d_G(x,f_1)+d_G(x,f_2)$ for $f_1=\exp_\mu(\gamma\eta)$ and $f_2=\exp_\mu(-\gamma\eta)$; if this sum is not constant on the curve, equation (7) does not define a spherical ellipse under definition (3).
Extended reading notes
Core claim
The central claim is that Hauberg's anisotropic geodesic normal density, originally written as $f_H(x;\mu,\Lambda)=C_H^{-1}\exp(-\tfrac12 \operatorname{Log}_\mu(x)^\top \Lambda \operatorname{Log}_\mu(x))$, can be rewritten exactly as equation (2), a function of the three coordinates $x^\top\mu$, $x^\top\eta$, $x^\top\xi$ with two concentration parameters $\lambda_1,\lambda_2$ and no tangent-space coordinates. This identity follows from expanding $\operatorname{Log}_\mu(x)$ in the orthonormal basis $\{\mu,\eta,\xi\}$. The paper then asserts that the level sets of this density, described by equation (7), are exactly ellipses on the sphere, where $\eta$ and $\xi$ give the axes and $\alpha,\beta$ are inversely related to $\lambda_1,\lambda_2$. It presents equation (7) as a new characterization of a spherical ellipse.
Load-bearing premise
The contour-ellipse proof assumes the logarithm map is an isometry of the whole sphere, so that a Euclidean ellipse in the tangent space maps to a curve whose points have constant sum of geodesic distances to the two foci; in fact the log map preserves distances only from the pole, not between arbitrary points.
Editorial extensions
If this is right
- The density in equation (2) allows practitioners to evaluate the geodesic normal distribution on $S^2$ without computing the log map or its inverse for each point.
- The parameters gain a geometric interpretation: $\eta$ and $\xi$ fix the axes of the contour ellipse, and $\lambda_1,\lambda_2$ are inversely related to the semi-axis lengths.
- Setting $\lambda_1=\lambda_2$ reduces the anisotropic density to the isotropic geodesic normal, giving a smooth family connecting isotropic and anisotropic models.
- The claimed equivalence in equation (7) would provide a closed-form algebraic equation for spherical ellipses that is an alternative to the focal-distance definition in equation (3).
Reading between the lines
- The same tangent-free substitution should extend the density formula to $S^p$ for $p>2$ by writing $\Lambda$ with $p$ positive eigenvalues and choosing an orthonormal basis of the tangent space.
- A direct numerical check of equation (7) against the focal-distance definition (3) would test whether the spherical-ellipse characterization holds for the full parameter range, especially near the antipode.
- If the contour-ellipse claim holds, the geodesic normal distribution could serve as a statistical model attached to spherical ellipses, with potential uses in shape analysis and directional clustering.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper presents a tangent-space-free expression for Hauberg's anisotropic spherical normal distribution, obtained by rewriting the Mahalanobis exponent of the log-map coordinates directly in terms of dot products on the sphere (Section 3, Eq. (2)). It then claims that the level sets of this density are exactly spherical ellipses, and proposes Eq. (7) as a new characterization of spherical ellipses (Section 4). The algebraic reformulation leading to Eq. (2) is correct, but the geometric conclusion is not. The argument that the exponential image of a Euclidean tangent-space ellipse is a spherical ellipse relies on treating the log map as a global isometry, which it is not; the paper's own focal-distance relation in Eq. (5) contradicts the Euclidean focal distance used in the derivation of Eq. (7).
Significance. If correct, the ellipse-contour claim would give the anisotropy parameters of the geodesic normal distribution a clear geometric interpretation and would provide a new equation for spherical ellipses. The tangent-free density formula (2) is indeed a valid and potentially convenient algebraic reformulation of Hauberg's density. However, the central geometric claim fails: Eq. (7) defines the exponential image of a Euclidean ellipse, which is not the constant-sum-of-geodesic-distances locus defined in Eq. (3). The paper's main advertised contribution, that the density contours are true spherical ellipses, is therefore unsupported and false.
major comments (2)
- [Section 4, after Eq. (7)] The sentence 'Since Log_mu is an isometry' uses the log map in a way that is false. Log_mu preserves radial distances from mu, in the sense that dG(x,mu)=||Log_mu(x)||, but it does not preserve distances between two arbitrary points. The argument requires exactly such pairwise distance preservation in order to transfer the focal property of the Euclidean ellipse E* to a sum of geodesic distances on the sphere: one would need dG(x,Exp_mu(gamma eta))=||Log_mu(x)-gamma eta|| for points off the radial geodesic, which is not true on a curved sphere. Consequently, the set E defined by Eq. (7) is the exponential image of a Euclidean ellipse, but it is not the locus defined by Eq. (3). This invalidates the conclusion that the density contours of fGN in Eq. (2) are spherical ellipses.
- [Section 4, Eqs. (5) and (7)] The two characterizations are mutually inconsistent regarding the focal distance. In Eq. (5), a spherical ellipse with semiaxes alpha and beta has focus angle gamma satisfying cos(gamma)=cos(alpha)/cos(beta). In the tangent-space derivation leading to Eq. (7), the Euclidean ellipse E* has semifocal distance gamma_E=sqrt(alpha^2-beta^2). For alpha=pi/4 and beta=pi/6 these are approximately 0.6155 and 0.5854, respectively; they agree only in the circular limit beta=alpha. Since Log_mu is not an isometry, there is no mechanism by which the Euclidean focal distance becomes the correct geodesic focal distance. Thus Eq. (7) and Eq. (3) define different curves in general, so the claim that Eq. (7) represents a spherical ellipse with the stated axes is not correct.
minor comments (4)
- [Section 2] The word 'fuction' should be 'function'.
- [Section 4, Eq. (6)] Equation (6) is printed without visible fraction bars in the text; the intended equation appears to be x_1^2 cos^2(gamma)/cos^2(alpha) + x_2^2 sin^2(gamma)/sin^2(alpha) = 1. Please fix the typesetting.
- [Section 4, after Eq. (7)] The symbol gamma is used both for the Euclidean semifocal distance in the tangent-space ellipse E* and for the geodesic focus angle in Eqs. (3)-(5). This notation collision obscures the inconsistency between the two uses; distinct symbols should be introduced.
- [Section 2] The statement 'This map is an isometry' is imprecise. Log_mu is a radial isometry with respect to mu (it preserves distances from mu), but it is not a global isometry between the tangent space and the sphere. Clarifying this would help prevent the misuse in Section 4.
Circularity Check
No circularity: the paper's alternative density is an algebraic reformulation of Hauberg's model, and the spherical-ellipse claim is an attempted theorem with a false isometry premise, not an input recycled as a conclusion.
full rationale
The derivation is self-contained and does not reduce to its own inputs. Equation (1) follows by direct algebra from the orthonormal expansion of x and the definition of the logarithm map; equation (2) is an algebraic rewrite of Hauberg's anisotropic geodesic normal density, not a new fitted object. Section 4 defines spherical ellipses independently via the geodesic-distance sum in equation (3), derives equivalent trigonometric forms, and then introduces equation (7) as a candidate locus. The load-bearing step after equation (7) is the sentence 'Since Log_mu is an isometry, this implies that E is the locus of points x in S2 such that dG(x,f1)+dG(y,f2)=2 alpha.' This step is mathematically false, because Log_mu preserves only distances from mu, not arbitrary pairwise geodesic distances, so the tangent-space ellipse does not pull back to a geodesic spherical ellipse. However, a false mathematical inference is not circular reasoning: the conclusion is not assumed as a premise, no parameter is fitted and renamed as a prediction, and the paper does not rely on any self-citation as evidence. The claimed spherical-ellipse characterization is therefore unsupported as written, but that is a correctness risk rather than a circularity. Since there are no fitted inputs, no self-citation chains, and no premise equivalent to the conclusion, the circularity score is 0.
Assumptions & free parameters
assumptions (3)
- ad hoc to paper The logarithm map Log_mu is an isometry between the Euclidean tangent space T_mu and the sphere S^2 with geodesic distance, preserving all pairwise distances.
- standard math Symmetric positive semi-definite Lambda can be diagonalized with orthonormal eigenvectors mu, eta, xi.
- standard math A spherical ellipse is the locus of points with constant sum of geodesic distances to two non-antipodal foci.
Cite this review
Pith. "Pith review of A note on the geodesic normal distribution on the sphere." pith.science (2026). https://pith.science/paper/NEAC6VUV
@misc{pith2026241114899,
author = {Pith},
title = {Pith review of: A note on the geodesic normal distribution on the sphere},
year = {2026},
howpublished = {\url{https://pith.science/paper/NEAC6VUV}},
note = {Machine review of arXiv:2411.14899}
}
read the original abstract
This paper presents an alternative formulation of the geodesic normal distribution on the sphere, building on the work of Hauberg (2018). While the isotropic version of this distribution is naturally defined on the sphere, the anisotropic version requires projecting points from the hypersphere onto the tangent space. In contrast, our approach removes the dependence on the tangent space and defines the geodesic normal distribution directly on the sphere. Moreover, we demonstrate that the density contours of this distribution are exactly ellipses on the sphere, providing intriguing alternative characterizations for describing this locus of points.
Figures
Reference graph
Works this paper leans on
-
[1]
Bhattacharya, A. and Bhattacharya, R. (2012). Nonparametric inference on manifolds: with applications to shape spaces , volume 2. Cambridge University Press
work page 2012
-
[2]
Coeurjolly, J.-F. and Bihan, N. L. (2012). Geodesic normal distribution on the circle. Metrika , 75(7):977--995
work page 2012
-
[3]
Glaeser, G., Stachel, H., and Odehnal, B. (2016). The Universe of Conics: From the ancient Greeks to 21st century developments . Springer
work page 2016
-
[4]
Hauberg, S. (2018). Directional statistics with the spherical normal distribution. In 2018 21st International Conference on Information Fusion ( FUSION ) , pages 704--711, New York. IEEE
work page 2018
-
[5]
Kent, J. T. (1982). The F isher- B ingham distribution on the sphere. Journal of the Royal Statistical Society, Series B (Methodological) , 44(1):71--80
work page 1982
-
[6]
Mardia, K. V. and Jupp, P. E. (2000). Directional Statistics . Wiley Series in Probability and Statistics. Wiley, Chichester
work page 2000
-
[7]
Paine, P., Preston, S. P., Tsagris, M., and Wood, A. T. (2018). An elliptically symmetric angular gaussian distribution. Statistics and Computing , 28:689--697
work page 2018
-
[8]
Pennec, X. (2006). Intrinsic statistics on riemannian manifolds: Basic tools for geometric measurements. Journal of Mathematical Imaging and Vision , 25:127--154
work page 2006
Show all 10 references
-
[9]
and Wood, A
Scealy, J. and Wood, A. T. (2019). Scaled von mises--fisher distributions and regression models for paleomagnetic directional data. Journal of the American Statistical Association , 114(528):1547--1560
2019
-
[10]
and Suh, C
You, K. and Suh, C. (2022). Parameter estimation and model-based clustering with spherical normal distribution on the unit hypersphere. Computational Statistics & Data Analysis , 171:107457
2022
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.