pith. sign in

arxiv: 2605.30300 · v1 · pith:MDXTFJ5Unew · submitted 2026-05-28 · 🧮 math.DG

Invariant statistical connections on the multivariate centered Gaussian model and their moduli spaces

Hideyuki Ishi , Hikozo Kobayashi , Takayuki Okuda This is my paper

Pith reviewed 2026-06-29 05:20 UTC · model grok-4.3

classification 🧮 math.DG
keywords statistical connectionsGaussian modelFisher metricmoduli spacesdually flatinvariant connectionshomogeneous manifoldsinformation geometry
0
0 comments X

The pith

The GL(n,ℝ)-invariant statistical connections on the multivariate centered Gaussian model with the Fisher metric are explicitly determined along with their moduli spaces.

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

This paper develops moduli spaces for invariant statistical connections on homogeneous Riemannian manifolds using two equivalence relations from a categorical viewpoint. It applies the framework to the space of zero-mean multivariate normal distributions with its Fisher metric. The authors then classify all connections invariant under GL(n, R) and under the isometry group of the metric. Special focus is given to the dually flat connections within this classification.

Core claim

We explicitly determine the GL(n,ℝ)-invariant and Isom(𝒩₀ⁿ, g^F)-invariant statistical connections on the multivariate centered Gaussian model, with particular emphasis on the dually flat case, and describe the corresponding moduli spaces defined via categorical equivalence relations.

What carries the argument

Moduli spaces of invariant statistical connections on homogeneous Riemannian manifolds, defined by two natural equivalence relations arising from a categorical viewpoint.

If this is right

  • The complete list of such invariant connections is now available for the Gaussian model.
  • The dually flat invariant connections can be used to define specific Bregman divergences on the model.
  • The moduli spaces classify these connections up to natural equivalences.
  • One can now compare the geometry induced by these connections across different dimensions n.

Where Pith is reading between the lines

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

  • Similar classifications could be carried out for other statistical models that are homogeneous spaces.
  • The categorical approach to moduli spaces might reveal connections between statistical geometry and other areas of differential geometry.
  • Explicit formulas for the connections could lead to new algorithms for statistical inference on Gaussian data.

Load-bearing premise

The two natural equivalence relations arising from a categorical viewpoint correctly capture the intended notion of equivalence for moduli spaces of invariant statistical connections on homogeneous Riemannian manifolds.

What would settle it

Constructing a statistical connection on the model that is invariant under GL(n, R) or the isometry group but lies outside the explicitly determined families would disprove the classification.

read the original abstract

We study invariant statistical connections on the space $\mathcal{N}_0^n$ of zero-mean multivariate normal distributions (the multivariate centered Gaussian model) equipped with the Fisher metric $g^F$. We introduce moduli spaces of invariant statistical connections on homogeneous Riemannian manifolds via two natural equivalence relations arising from a categorical viewpoint, and apply this framework to $(\mathcal{N}_0^n, g^F)$. We explicitly determine the $GL(n,\mathbb{R})$-invariant and $\mathrm{Isom}(\mathcal{N}_0^n, g^F)$-invariant statistical connections, with particular emphasis on the dually flat case, and describe the corresponding moduli spaces.

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 / 1 minor

Summary. The paper studies invariant statistical connections on the space 𝒩₀ⁿ of zero-mean multivariate normal distributions equipped with the Fisher metric g^F. It introduces moduli spaces of such connections on homogeneous Riemannian manifolds via two natural equivalence relations arising from a categorical viewpoint, explicitly determines the GL(n,ℝ)-invariant and Isom(𝒩₀ⁿ, g^F)-invariant statistical connections (with emphasis on the dually flat case), and describes the corresponding moduli spaces.

Significance. If the explicit determinations and parametrizations hold, the work supplies concrete invariant connections on a fundamental statistical model and a new categorical framework for moduli spaces that may extend to other homogeneous spaces in information geometry. The focus on dually flat structures is particularly relevant given their role in affine geometry and statistical inference.

major comments (1)
  1. [moduli spaces framework] The central construction of the moduli spaces rests on the assertion that the two natural equivalence relations from the categorical viewpoint correctly capture the intended notion of equivalence (§ on moduli spaces, as described in the abstract). This is presented without explicit comparison to standard equivalence notions for connections on homogeneous manifolds or verification that the relations are non-trivial and preserve the invariance and dually flat properties; a concrete check against the derived connections would strengthen the claim.
minor comments (1)
  1. [abstract] Notation for the space 𝒩₀ⁿ and the metric g^F should be defined at first use for readers outside information geometry.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and the recommendation for minor revision. We respond to the single major comment below.

read point-by-point responses

  1. Referee: [moduli spaces framework] The central construction of the moduli spaces rests on the assertion that the two natural equivalence relations from the categorical viewpoint correctly capture the intended notion of equivalence (§ on moduli spaces, as described in the abstract). This is presented without explicit comparison to standard equivalence notions for connections on homogeneous manifolds or verification that the relations are non-trivial and preserve the invariance and dually flat properties; a concrete check against the derived connections would strengthen the claim.

    Authors: We agree that an explicit comparison to standard equivalence notions would strengthen the exposition. In the revised version we will add a short subsection in the moduli-spaces section that (i) recalls the usual notions of equivalence for affine connections on homogeneous Riemannian manifolds (automorphism-induced and gauge equivalence) and (ii) verifies that the two categorical relations are distinct from these while preserving GL(n,ℝ)-invariance and dual flatness. We will then apply both relations to the explicit one-parameter family of dually flat GL(n,ℝ)-invariant connections already derived in the paper, confirming that the resulting moduli spaces are non-trivial. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces a categorical framework for moduli spaces of invariant statistical connections on homogeneous Riemannian manifolds and applies it to explicitly construct the GL(n,ℝ)- and Isom(𝒩₀ⁿ, g^F)-invariant connections on the centered Gaussian model, with emphasis on the dually flat subfamily. No derivation step reduces by the paper's own equations to a fitted input, self-definition, or self-citation chain; the invariance conditions and moduli parametrization are presented as independent constructions from the given metric and group actions. The central claims rest on direct determination rather than renaming or smuggling prior ansatzes.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard background from information geometry and differential geometry together with the new categorical equivalence relations; no free parameters or invented entities are visible in the abstract.

axioms (2)
  • domain assumption The Fisher metric g^F equips the space of centered Gaussians with a homogeneous Riemannian structure.
    Invoked when the paper states the manifold (𝒩₀ⁿ, g^F) and studies its invariant connections.
  • ad hoc to paper Two natural equivalence relations arising from a categorical viewpoint define the moduli spaces of invariant statistical connections.
    Introduced in the abstract as the framework applied to the Gaussian model.

pith-pipeline@v0.9.1-grok · 5638 in / 1368 out tokens · 26370 ms · 2026-06-29T05:20:15.646012+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

54 extracted references · 38 canonical work pages

  1. [1]

    Amari.Differential-Geometrical Methods in Statistics, volume 28 ofLecture Notes in Statistics

    S.-i. Amari.Differential-Geometrical Methods in Statistics, volume 28 ofLecture Notes in Statistics. Springer- Verlag, New York, 1985.https://doi.org/10.1007/978-1-4612-5056-2

    work page doi:10.1007/978-1-4612-5056-2 1985

  2. [2]

    S.-i. Amari. Information Geometry of Positive Measures and Positive-Definite Matrices: Decomposable Dually Flat Structure.Entropy, 16(4):2131–2145, 2014.https://doi.org/10.3390/e16042131

    work page doi:10.3390/e16042131 2014

  3. [3]

    Amari and H

    S.-i. Amari and H. Nagaoka.Methods of Information Geometry, volume 191 ofTranslations of Mathematical Monographs. American Mathematical Society, Providence, RI; Oxford University Press, Oxford, 2000.https: //doi.org/10.1090/mmono/191

    work page doi:10.1090/mmono/191 2000

  4. [4]

    N. Ay, J. Jost, H. V. Lˆ e, and L. Schwachh¨ ofer. Information geometry and sufficient statistics.Probab. Theory Related Fields, 162(1-2):327–364, 2015.https://doi.org/10.1007/s00440-014-0574-8

    work page doi:10.1007/s00440-014-0574-8 2015

  5. [5]

    N. Ay, J. Jost, H. V. Lˆ e, and L. Schwachh¨ ofer.Information Geometry, volume 64 ofErgebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics. Springer, Cham, 2017.https: //doi.org/10.1007/978-3-319-56478-4

    work page doi:10.1007/978-3-319-56478-4 2017

  6. [6]

    A learnable spatial mapping for decoding the directional focus of auditory attention using EEG,

    I. Ayadi, F. Bouchard, and F. Pascal. Elliptical Wishart Distribution: Maximum Likelihood Estimator from Information Geometry. InIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5, 2023.https://doi.org/10.1109/ICASSP49357.2023.10096222

    work page doi:10.1109/icassp49357.2023.10096222 2023

  7. [7]

    J. Burbea. Informative Geometry of Probability Spaces.Exposition. Math., 4(4):347–378, 1986

    1986

  8. [8]

    Cardoso, A

    I. Cardoso, A. Cosgaya, and S. Reggiani. The moduli space of left-invariant metrics on six-dimensional characteristically solvable nilmanifolds.Math. Nachr., 298(5):1496–1520, 2025.https://doi.org/10.1002/ mana.202400213 INVARIANT STATISTICAL CONNECTIONS ON THE MULTIVARIATE CENTERED GAUSSIAN MODEL AND THEIR MODULI SPACES 39

    2025

  9. [9]

    Cosgaya and S

    A. Cosgaya and S. Reggiani. Isometry groups of three-dimensional Lie groups.Ann. Global Anal. Geom., 61(4):831–845, 2022.https://doi.org/10.1007/s10455-022-09835-3

    work page doi:10.1007/s10455-022-09835-3 2022

  10. [10]

    Deki´ c, M

    A. Deki´ c, M. Babi´ c, and S. Vukmirovi´ c. Classification of Left Invariant Riemannian Metrics on Com- plex Hyperbolic Space.Mediterr. J. Math., 19(5):Paper No. 232, 18, 2022.https://doi.org/10.1007/ s00009-022-02152-w

    2022

  11. [11]

    Dolcetti and D

    A. Dolcetti and D. Pertici. Differential properties of spaces of symmetric real matrices.Rend. Semin. Mat. Univ. Politec. Torino, 77(1):25–43, 2019

    2019

  12. [12]

    Furuhata

    H. Furuhata. Hypersurfaces in statistical manifolds.Differential Geom. Appl., 27(3):420–429, 2009.https: //doi.org/10.1016/j.difgeo.2008.10.019

    work page doi:10.1016/j.difgeo.2008.10.019 2009

  13. [13]

    Furuhata and I

    H. Furuhata and I. Hasegawa. Submanifold Theory in Holomorphic Statistical Manifolds. InGeometry of Cauchy-Riemann submanifolds, pages 179–215. Springer, Singapore, 2016.https://doi.org/10.1007/ 978-981-10-0916-7_7

    2016

  14. [14]

    Furuhata, J.-i

    H. Furuhata, J.-i. Inoguchi, and S.-P. Kobayashi. A characterization of the alpha-connections on the statistical manifold of normal distributions.Inf. Geom., 4(1):177–188, 2021.https://doi.org/10.1007/ s41884-020-00037-z

    2021

  15. [15]

    Hashinaga

    T. Hashinaga. On the minimality of the corresponding submanifolds to four-dimensional solvsolitons.Hi- roshima Math. J., 44(2):173–191, 2014.https://doi.org/10.32917/hmj/1408972906

    work page doi:10.32917/hmj/1408972906 2014

  16. [16]

    Hashinaga and H

    T. Hashinaga and H. Tamaru. Three-dimensional solvsolitons and the minimality of the corresponding sub- manifolds.Internat. J. Math., 28(6):1750048, 31, 2017.https://doi.org/10.1142/S0129167X17500483

    work page doi:10.1142/s0129167x17500483 2017

  17. [17]

    Hashinaga, H

    T. Hashinaga, H. Tamaru, and K. Terada. Milnor-type theorems for left-invariant Riemannian metrics on Lie groups.J. Math. Soc. Japan, 68(2):669–684, 2016.https://doi.org/10.2969/jmsj/06820669

    work page doi:10.2969/jmsj/06820669 2016

  18. [18]

    Inoguchi

    J.-i. Inoguchi. Holomorphic statistical 2-manifolds and almost contact statistical 3-manifolds.Differential Geom. Appl., 103:Paper No. 102366, 2026.https://doi.org/10.1016/j.difgeo.2026.102366

    work page doi:10.1016/j.difgeo.2026.102366 2026

  19. [19]

    Inoguchi and Y

    J.-i. Inoguchi and Y. Ohno. Homogeneous statistical manifolds.Inf. Geom., 8(2):285–341, 2025.https: //doi.org/10.1007/s41884-025-00172-5

    work page doi:10.1007/s41884-025-00172-5 2025

  20. [20]

    S. Ivanov. On dual-projectively flat affine connections.J. Geom., 53(1-2):89–99, 1995.https://doi.org/10. 1007/BF01224043

    1995

  21. [21]

    Kobayashi, Y

    H. Kobayashi, Y. Ohno, T. Okuda, and H. Tamaru. The moduli spaces of left-invariant statistical structures on Lie groups.Inf. Geom., 9(1):235–278, 2026.https://doi.org/10.1007/s41884-026-00200-y

    work page doi:10.1007/s41884-026-00200-y 2026

  22. [22]

    Kobayashi and T

    H. Kobayashi and T. Okuda. On Invariant Conjugate Symmetric Statistical Structures on the Space of Zero-Mean Multivariate Normal Distributions. InGeometric Science of Information. Part I, vol- ume 16033 ofLecture Notes in Comput. Sci., pages 65–72. Springer, Cham, 2026.https://doi.org/10. 1007/978-3-032-03918-7_7

    2026

  23. [23]

    Kobayashi and K

    S. Kobayashi and K. Nomizu.Foundations of Differential Geometry. Vol I. Interscience Publishers (a division of John Wiley & Sons, Inc.), New York-London, 1963

    1963

  24. [24]

    Kobayashi and K

    S. Kobayashi and K. Nomizu.Foundations of Differential Geometry. Vol. II, volume No. 15 ofInterscience Tracts in Pure and Applied Mathematics. Interscience Publishers John Wiley & Sons, Inc., New York-London- Sydney, 1969

    1969

  25. [25]

    Kobayashi and Y

    S.-P. Kobayashi and Y. Ohno. A characterization of the alpha-connections on the statistical manifold of multivariate normal distributions.Osaka J. Math., 62(2):329–349, 2025.https://doi.org/10.18910/101132

    work page doi:10.18910/101132 2025

  26. [26]

    Kodama, A

    H. Kodama, A. Takahara, and H. Tamaru. The space of left-invariant metrics on a Lie group up to isometry and scaling.Manuscripta Math., 135(1-2):229–243, 2011.https://doi.org/10.1007/s00229-010-0419-4

    work page doi:10.1007/s00229-010-0419-4 2011

  27. [27]

    Y. Kondo. A classification of left-invariant pseudo-Riemannian metrics on some nilpotent Lie groups.Hi- roshima Math. J., 52(3):333–356, 2022.https://doi.org/10.32917/h2021054

    work page doi:10.32917/h2021054 2022

  28. [28]

    Kondo and H

    Y. Kondo and H. Tamaru. A classification of left-invariant Lorentzian metrics on some nilpotent Lie groups. Tohoku Math. J. (2), 75(1):89–117, 2023.https://doi.org/10.2748/tmj.20211122

    work page doi:10.2748/tmj.20211122 2023

  29. [29]

    A. Kubo, K. Onda, Y. Taketomi, and H. Tamaru. On the moduli spaces of left-invariant pseudo- Riemannian metrics on Lie groups.Hiroshima Math. J., 46(3):357–374, 2016.https://doi.org/10.32917/ hmj/1487991627

    work page arXiv 2016

  30. [30]

    T. Kurose. Dual connections and affine geometry.Math. Z., 203(1):115–121, 1990.https://doi.org/10. 1007/BF02570725

    1990

  31. [31]

    T. Kurose. On the divergences of 1-conformally flat statistical manifolds.Tohoku Math. J. (2), 46(3):427–433, 1994.https://doi.org/10.2748/tmj/1178225722

    work page doi:10.2748/tmj/1178225722 1994

  32. [32]

    S. L. Lauritzen. Statistical Manifolds. InDifferential geometry in statistical inference, volume 10, pages 163–216. Institute of Mathematical Statistics, Hayward, 1987.https://doi.org/10.1214/lnms/1215467061

    work page doi:10.1214/lnms/1215467061 1987

  33. [33]

    J. M. Lee.Introduction to Smooth Manifolds, volume 218 ofGraduate Texts in Mathematics. Springer, New York, second edition, 2013.https://doi.org/10.1007/978-1-4419-9982-5

    work page doi:10.1007/978-1-4419-9982-5 2013

  34. [34]

    Matsuzoe

    H. Matsuzoe. Statistical manifolds and affine differential geometry. InProbabilistic approach to geometry, volume 57 ofAdv. Stud. Pure Math., pages 303–321. Math. Soc. Japan, Tokyo, 2010.https://doi.org/10. 2969/aspm/05710303 40 HIDEYUKI ISHI, HIKOZO KOBAYASHI, TAKAYUKI OKUDA

    work page arXiv 2010

  35. [35]

    A. F. S. Mitchell. The information matrix, skewness tensor andα-connections for the general multivariate elliptic distribution.Ann. Inst. Statist. Math., 41(2):289–304, 1989.https://doi.org/10.1007/BF00049397

    work page doi:10.1007/bf00049397 1989

  36. [36]

    B. Opozda. Bochner’s technique for statistical structures.Ann. Global Anal. Geom., 48(4):357–395, 2015. https://doi.org/10.1007/s10455-015-9475-z

    work page doi:10.1007/s10455-015-9475-z 2015

  37. [37]

    B. Opozda. A sectional curvature for statistical structures.Linear Algebra Appl., 497:134–161, 2016.https: //doi.org/10.1016/j.laa.2016.02.021

    work page doi:10.1016/j.laa.2016.02.021 2016

  38. [38]

    B. Opozda. Curvature bounded conjugate symmetric statistical structures with complete metric.Ann. Global Anal. Geom., 55(4):687–702, 2019.https://doi.org/10.1007/s10455-019-09647-y

    work page doi:10.1007/s10455-019-09647-y 2019

  39. [39]

    Pedro Castellanos Moscoso

    L. Pedro Castellanos Moscoso. Left-invariant symplectic structures on diagonal almost abelian Lie groups. Hiroshima Math. J., 52(3):357–378, 2022.https://doi.org/10.32917/h2021055

    work page doi:10.32917/h2021055 2022

  40. [40]

    Pedro Castellanos Moscoso and H

    L. Pedro Castellanos Moscoso and H. Tamaru. A classification of left-invariant symplectic structures on some Lie groups.Beitr. Algebra Geom., 64(2):471–491, 2023.https://doi.org/10.1007/s13366-022-00643-1

    work page doi:10.1007/s13366-022-00643-1 2023

  41. [41]

    S. M. Salamon. Complex structures on nilpotent Lie algebras.J. Pure Appl. Algebra, 157(2-3):311–333, 2001. https://doi.org/10.1016/S0022-4049(00)00033-5

    work page doi:10.1016/s0022-4049(00)00033-5 2001

  42. [42]

    Shima.The Geometry of Hessian Structures

    H. Shima.The Geometry of Hessian Structures. World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 2007.https://doi.org/10.1142/9789812707536

    work page doi:10.1142/9789812707536 2007

  43. [43]

    Shima and J

    H. Shima and J. H. Hao. Geometry associated with normal distributions.Osaka J. Math., 37(2):509–517, 2000

    2000

  44. [44]

    L. T. Skovgaard. A Riemannian Geometry of the Multivariate Normal Model.Scand. J. Statist., 11(4):211– 223, 1984

    1984

  45. [45]

    ˇSukilovi´ c and S

    T. ˇSukilovi´ c and S. Vukmirovi´ c. Geometry of cotangent bundle of Heisenberg group.Differential Geom. Appl., 88:Paper No. 101997, 11, 2023.https://doi.org/10.1016/j.difgeo.2023.101997

    work page doi:10.1016/j.difgeo.2023.101997 2023

  46. [46]

    ˇSukilovi´ c, S

    T. ˇSukilovi´ c, S. Vukmirovi´ c, and N. Bokan. On the moduli space of left-invariant metrics on the cotangent bundle of the Heisenberg group.Rev. Un. Mat. Argentina, 68(2):485–518, 2025.https://doi.org/10.33044/ revuma.3426

    2025

  47. [47]

    K. Takano. Geodesics on statistical models of the multivariate normal distribution.Tensor (N.S.), 67(2):162– 169, 2006

    2006

  48. [48]

    Taketomi

    Y. Taketomi. Examples of hyperpolar actions of the automorphism groups of Lie algebras.Topology Appl., 196:904–910, 2015.https://doi.org/10.1016/j.topol.2015.05.057

    work page doi:10.1016/j.topol.2015.05.057 2015

  49. [49]

    Taketomi

    Y. Taketomi. On a Riemannian submanifold whose slice representation has no nonzero fixed points.Hiroshima Math. J., 48(1):1–20, 2018.https://doi.org/10.32917/hmj/1520478020

    work page doi:10.32917/hmj/1520478020 2018

  50. [50]

    Taketomi and H

    Y. Taketomi and H. Tamaru. On the nonexistence of left-invariant Ricci solitons—a conjecture and examples. Transform. Groups, 23(1):257–270, 2018.https://doi.org/10.1007/s00031-017-9439-4

    work page doi:10.1007/s00031-017-9439-4 2018

  51. [51]

    S. V. Thuong. Metrics on 4-dimensional unimodular Lie groups.Ann. Global Anal. Geom., 51(2):109–128, 2017.https://doi.org/10.1007/s10455-016-9527-z

    work page doi:10.1007/s10455-016-9527-z 2017

  52. [52]

    J. F. Torres. Moduli space of bi-invariant metrics.Bol. Soc. Mat. Mex. (3), 30(2):Paper No. 49, 17, 2024. https://doi.org/10.1007/s40590-024-00622-7

    work page doi:10.1007/s40590-024-00622-7 2024

  53. [53]

    F. W. Warner.Foundations of Differentiable Manifolds and Lie Groups, volume 94 ofGraduate Texts in Mathematics. Springer-Verlag, New York-Berlin, 1983.https://doi.org/10.1007/978-1-4757-1799-0

    work page doi:10.1007/978-1-4757-1799-0 1983

  54. [54]

    Warner.Harmonic Analysis on Semi-Simple Lie Groups I, volume 188 ofDie Grundlehren der mathematischen Wissenschaften

    G. Warner.Harmonic Analysis on Semi-Simple Lie Groups I, volume 188 ofDie Grundlehren der mathematischen Wissenschaften. Springer-Verlag, New York-Heidelberg, 1972.https://doi.org/10.1007/ 978-3-642-50275-0 (H. Ishi)Department of Mathematics, Graduate School of Science, Osaka Metropolitan Univer- sity, 3-3-138, Sugimoto, Sumiyoshi-ku, Osaka, 558-8585, Jap...

    1972