pith. sign in

arxiv: 2509.17827 · v4 · submitted 2025-09-22 · 📡 eess.SY · cs.SY

On Fast Attitude Filtering Using Matrix Fisher Distributions with Stability Guarantee

Shijie Wang , Haichao Gui , Rui Zhong This is my paper

Pith reviewed 2026-05-18 14:43 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords attitude estimationmatrix Fisher distributionright-invariant errornonlinear filteringSO(3)stability analysisBayesian filterattitude filtering
0
0 comments X

The pith

Closed-form attitude filters with matrix Fisher distributions and right-invariant errors achieve almost global asymptotic stability on SO(3).

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

This paper analyzes how matrix Fisher distributions evolve under Bayes' rule and isolates two essential properties that support reliable Bayesian attitude filtering on the special orthogonal group. It introduces two approximate filters that use linearized error systems with right-invariant errors to obtain closed-form iterations while keeping those properties intact. The right-invariant version is proven almost globally asymptotically stable for any trajectory on SO(3) and locally exponentially stable for single-axis rotations. Simulations confirm that the filters match the accuracy of earlier MFD-based methods in difficult conditions but run in a small fraction of the time.

Core claim

By preserving the two essential properties of MFD distribution evolution along Bayes' rule through linearized right-invariant error systems, the proposed filter iterates in closed form and represents uncertainty globally with MFDs, which enables the proof of almost global asymptotic stability for any trajectory on SO(3) and supports reliable convergence from large initial errors.

What carries the argument

Linearized error system with right-invariant errors applied to the matrix Fisher distribution (MFD) on SO(3), which enables closed-form iteration and supplies the global uncertainty representation used for the stability proofs.

If this is right

  • The filter converges almost globally asymptotically for arbitrary initial conditions and any reference trajectory on SO(3).
  • Local exponential stability holds for single-axis rotations, with convergence rate influenced by the retained distribution properties.
  • Accuracy remains comparable to full Bayesian MFD filters under large uncertainties while computation time drops to roughly 1/5 to 1/100 of prior MFD implementations.

Where Pith is reading between the lines

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

  • The stability result could support attitude control loops on spacecraft or UAVs that must recover from poor initial orientation estimates.
  • The right-invariant linearization technique may extend to filtering problems on other compact Lie groups where directional statistics are used.
  • Hardware experiments with real IMU data could test whether the predicted convergence rates hold outside simulation.

Load-bearing premise

The two essential properties of the MFD distribution evolution along Bayes' rule remain intact when the filter is approximated via linearized error systems with right-invariant errors.

What would settle it

A specific trajectory on SO(3) together with a large initial attitude error for which the filter state diverges instead of converging to the true attitude.

Figures

Figures reproduced from arXiv: 2509.17827 by Haichao Gui, Rui Zhong, Shijie Wang.

Figure 1
Figure 1. Figure 1: Matrix Fisher distributions with different parameters. The marginal distribution for each column of R is shown on the unit sphere as red, green and blue shades, as presented in [33]. The red, green and blue lines represent the first, second, and third columns of the mean attitude, respectively, and the black line represents the axis w0. (a) M(F1): κ1 = 100, θ¯1 = 0; (b) M(F2): κ2 = 4, θ¯2 = 0; (c) M(F3): κ… view at source ↗
Figure 3
Figure 3. Figure 3: The trends of θ¯+ on κm/κ− with different ∆θ¯. Blue: θ¯+ given by MFD-based Bayesian filters. Red: θ¯+ given by CGD-based Bayesian filters. Green: the mean angle with higher confidence in θ¯− and θ¯m axis w0 when the mean attitude of the prior and the likelihood distributions are consistent. Meanwhile, ¯θ + MFD and ¯θ + CGD are also different [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: The trend of κ+ as a function of ∆θ¯. Blue: κ+ given by MFD￾based Bayesian filters. Red: κ+ given by CGD-based Bayesian filters In this subsection, the parameters given by MFD-based Bayesian filters are denoted by κ + MFD and ¯θ + MFD, while the parameters given by CGD-based Bayesian filters are denoted by κ + CGD and ¯θ + CGD. Comparing κ + MFD and κ + CGD, it can be seen from Table I and [PITH_FULL_IMAG… view at source ↗
Figure 4
Figure 4. Figure 4: The posterior distribution with several direct attitude measure￾ments. A. Attitude estimation error following MFDs We construct the following probability model to characterize the uncertainty of random attitudes with invariant attitude error following MFDs on SO(3). Definition IV.1. The discrepancy of the random matrix R ∈ SO(3) with respect to R¯ ∈ SO(3) is characterized by the left￾or right-invariant err… view at source ↗
Figure 5
Figure 5. Figure 5: Average error for the case of large initial errors with non￾isotropic measurement errors. The shadow represents an envelope of 95% confidence. The attitude uncertainty is calculated as the square root of the first diagonal term of the attitude covariance matrix in the inertial frame as [36]. As shown in [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Average error for the case of direct attitude measurements with non-isotropic noise: Nm = diag[100, 0, 0]. The two proposed filters maintain almost the same estima￾tion accuracy as, and have significantly faster computation speed than the BF-MFD and BF-AMFD. All four MFD￾based filters achieve smaller steady-state errors compared to the IEKF. This indicates that although FNF-R and FNF-L utilize the lineariz… view at source ↗
read the original abstract

This paper addresses two interrelated problems of the nonlinear filtering mechanism and fast attitude filtering with the matrix Fisher distribution (MFD) on the special orthogonal group. By analyzing the distribution evolution along Bayes' rule, we reveal two essential properties that enhance the performance of Bayesian attitude filters with MFDs, particularly in challenging conditions. Benefiting from the new understanding of the filtering mechanism associated with MFDs, two closed-form filters with MFDs are then proposed. These filters avoid the burdensome computations in previous MFD-based filters by introducing linearized error systems with right-invariant errors but retaining the two advantageous properties. The proposed filter with right-invariant error is proven to be almost globally asymptotically stable for any trajectory on $SO(3)$ leveraging its closed-form iteration and global uncertainty representation with MFDs. Moreover, we further prove the local exponential stability of the filter for single-axis rotations to reveal the effect of the two properties on the convergence rate. These stability results support the performance of the proposed filter with large initial error from a theoretical viewpoint, which to our knowledge, is not achieved by existing directional statistics-based filters. Numerical simulations demonstrate that proposed filters are as accurate as recent MFD-based Bayesian filters in challenging circumstances but consume far less computation time (about 1/5 to 1/100 of previous MFD-based attitude filters).

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 / 2 minor

Summary. The manuscript analyzes the evolution of Matrix Fisher Distributions (MFDs) along Bayes' rule on SO(3) to identify two essential properties that enhance Bayesian attitude filtering. It proposes two closed-form filters that employ linearized error systems with right-invariant errors to retain these properties while avoiding expensive computations of prior MFD filters. The right-invariant-error filter is proven almost globally asymptotically stable for arbitrary trajectories on SO(3), with an additional local exponential stability result for single-axis rotations; simulations show accuracy comparable to recent MFD-based filters at 1/5 to 1/100 the computation time.

Significance. If the stability transfer from the exact MFD case to the linearized implementation holds, the work supplies a computationally efficient attitude filter with global uncertainty representation and almost-global stability guarantees, a combination not achieved by existing directional-statistics filters. The closed-form iteration and explicit stability analysis for large initial errors constitute clear strengths.

major comments (2)
  1. [§5] §5 (stability analysis): The almost-global asymptotic stability claim for the right-invariant-error filter rests on the linearized error system preserving the two essential MFD properties (mean/concentration updates and global uncertainty representation) revealed from exact Bayes evolution. Because linearization is performed locally around the current estimate, it is not evident that these properties remain unperturbed for large initial errors or arbitrary trajectories; this assumption is load-bearing for the global result and requires explicit justification or a supporting lemma.
  2. [§4] §4 (filter derivation): The closed-form iteration is obtained by substituting the linearized right-invariant error dynamics into the MFD update; however, the manuscript does not quantify the approximation error in the concentration matrix or the mean direction, which could affect whether the two advantageous properties are exactly retained rather than approximately.
minor comments (2)
  1. [Simulation section] Table 1 or simulation section: the reported computation-time ratios (1/5 to 1/100) would be more convincing if the exact hardware, matrix dimensions, and number of Monte-Carlo runs were stated explicitly.
  2. [§3] Notation: the symbols for the two essential properties should be introduced once in §3 and used consistently thereafter to avoid ambiguity when referring to their preservation in the linearized filter.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address the major comments point by point below, outlining how we will strengthen the presentation and analysis in the revised version.

read point-by-point responses

  1. Referee: [§5] §5 (stability analysis): The almost-global asymptotic stability claim for the right-invariant-error filter rests on the linearized error system preserving the two essential MFD properties (mean/concentration updates and global uncertainty representation) revealed from exact Bayes evolution. Because linearization is performed locally around the current estimate, it is not evident that these properties remain unperturbed for large initial errors or arbitrary trajectories; this assumption is load-bearing for the global result and requires explicit justification or a supporting lemma.

    Authors: We agree that the stability argument would benefit from explicit justification of how the linearized error system preserves the essential MFD properties for large initial errors. In the revised manuscript we will insert a supporting lemma in Section 5 that bounds the linearization-induced perturbation on the mean and concentration updates. The lemma will show that, because the right-invariant error is used and the MFD continues to represent global uncertainty on SO(3), the almost-global asymptotic stability established for the exact Bayes evolution carries over to the closed-form filter; the perturbation vanishes asymptotically as the estimate converges, allowing the same Lyapunov argument to apply from almost all initial conditions. revision: yes

  2. Referee: [§4] §4 (filter derivation): The closed-form iteration is obtained by substituting the linearized right-invariant error dynamics into the MFD update; however, the manuscript does not quantify the approximation error in the concentration matrix or the mean direction, which could affect whether the two advantageous properties are exactly retained rather than approximately.

    Authors: The referee correctly identifies that the approximation error introduced by linearization is not quantified. We will add a short error analysis subsection in Section 4 that derives first-order bounds on the resulting perturbations to the concentration matrix and mean direction. These bounds will be expressed in terms of the linearization residual and the current concentration level; we will then discuss how the right-invariant formulation keeps the two advantageous properties intact to first order, thereby supporting the subsequent stability claims while making the approximate nature of the implementation explicit. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation builds on independent Bayes-rule analysis and explicit stability proof

full rationale

The paper first derives two essential properties directly from exact MFD evolution under Bayes' rule, then constructs closed-form filters via linearized right-invariant errors that are stated to retain those properties. The almost-global asymptotic stability result for arbitrary SO(3) trajectories is proven using the closed-form iteration and global MFD uncertainty representation. No step reduces by construction to a fitted parameter, self-defined quantity, or load-bearing self-citation; the linearization is an explicit design choice for speed whose preservation of the properties is asserted as part of the filter definition rather than smuggled in. The chain is therefore self-contained against external benchmarks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Limited visibility from abstract only; the work relies on standard properties of matrix Fisher distributions and linearization approximations without introducing new free parameters or entities in the summary.

axioms (2)
  • domain assumption Matrix Fisher distribution can represent global uncertainty on SO(3) and evolves according to Bayes' rule with two essential properties that improve filter performance.
    Invoked in the analysis of distribution evolution to justify the new filters.
  • ad hoc to paper Linearized error systems with right-invariant errors preserve the two advantageous properties of the full MFD Bayesian update.
    Central modeling choice that enables closed-form fast filters.

pith-pipeline@v0.9.0 · 5769 in / 1123 out tokens · 52641 ms · 2026-05-18T14:43:26.030870+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.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 1 internal anchor

  1. [1]

    Nonlinear compleme ntary filters on the special orthogonal group,

    R. Mahony, T. Hamel, and J.-M. Pflimlin, “Nonlinear compleme ntary filters on the special orthogonal group,” IEEE Transactions on automatic control, vol. 53, no. 5, pp. 1203–1218, 2008

    work page 2008

  2. [2]

    Higher order nonlinear compl ementary filtering on lie groups,

    D. E. Zlotnik and J. R. Forbes, “Higher order nonlinear compl ementary filtering on lie groups,” IEEE Transactions on Automatic Control , vol. 64, no. 5, pp. 1772–1783, 2018

    work page 2018

  3. [3]

    Globally asymptotically s table attitude observer on so (3),

    T.-H. Wu, E. Kaufman, and T. Lee, “Globally asymptotically s table attitude observer on so (3),” in 2015 54th IEEE Conference on Decision and Control (CDC) , pp. 2164–2168, IEEE, 2015

    work page 2015

  4. [4]

    Hybrid global ex ponential stabilization on so (3),

    S. Berkane, A. Abdessameud, and A. Tayebi, “Hybrid global ex ponential stabilization on so (3),” Automatica, vol. 81, pp. 279–285, 2017

    work page 2017

  5. [5]

    Hybrid attitude and gyro-bias observer design on so(3),

    S. Berkane, A. Abdessameud, and A. Tayebi, “Hybrid attitude and gyro-bias observer design on so(3),” IEEE Transactions on Automatic Control, vol. 62, no. 11, pp. 6044–6050, 2017

    work page 2017

  6. [6]

    Hybrid nonlinear observers for inert ial navi- gation using landmark measurements,

    M. Wang and A. Tayebi, “Hybrid nonlinear observers for inert ial navi- gation using landmark measurements,” IEEE Transactions on Automatic Control, vol. 65, no. 12, pp. 5173–5188, 2020

    work page 2020

  7. [7]

    Nonlinear attitude estimation using inter- mittent and multirate vector measurements,

    M. Wang and A. Tayebi, “Nonlinear attitude estimation using inter- mittent and multirate vector measurements,” IEEE Transactions on Automatic Control, vol. 69, no. 8, pp. 5231–5245, 2023

    work page 2023

  8. [8]

    Kalman Filtering f or Spacecraft Attitude Estimation,

    E. Lefferts, F. Markley, and M. Shuster, “Kalman Filtering f or Spacecraft Attitude Estimation,” Journal of Guidance, Control, and Dynamics , vol. 5, pp. 417–429, Sept. 1982

    work page 1982

  9. [9]

    Attitude error representations for kalman fi ltering,

    F. L. Markley, “Attitude error representations for kalman fi ltering,” Journal of guidance, control, and dynamics , vol. 26, no. 2, pp. 311– 317, 2003

    work page 2003

  10. [10]

    Intrinsic Filtering on Lie Group s With Applications to Attitude Estimation,

    A. Barrau and S. Bonnabel, “Intrinsic Filtering on Lie Group s With Applications to Attitude Estimation,” IEEE Transactions on Automatic Control, vol. 60, pp. 436–449, Feb. 2015

    work page 2015

  11. [11]

    The Invariant Extended Kalman Fi lter as a Stable Observer,

    A. Barrau and S. Bonnabel, “The Invariant Extended Kalman Fi lter as a Stable Observer,” IEEE Transactions on Automatic Control , vol. 62, pp. 1797–1812, Apr. 2017

    work page 2017

  12. [12]

    Stochastic observers on Lie grou ps: A tutorial,

    A. Barrau and S. Bonnabel, “Stochastic observers on Lie grou ps: A tutorial,” in 2018 IEEE Conference on Decision and Control (CDC) , pp. 1264–1269, IEEE, 2018

    work page 2018

  13. [13]

    The invariant extended kalman fil ter as a stable observer,

    A. Barrau and S. Bonnabel, “The invariant extended kalman fil ter as a stable observer,” IEEE Transactions on Automatic Control , vol. 62, no. 4, pp. 1797–1812, 2016

    work page 2016

  14. [14]

    Extended Kalman Filtering With N onlinear Equality Constraints: A Geometric Approach,

    A. Barrau and S. Bonnabel, “Extended Kalman Filtering With N onlinear Equality Constraints: A Geometric Approach,” IEEE Transactions on Automatic Control, vol. 65, pp. 2325–2338, June 2020

    work page 2020

  15. [15]

    The geometry of navigation probl ems,

    A. Barrau and S. Bonnabel, “The geometry of navigation probl ems,” IEEE Transactions on Automatic Control , vol. 68, no. 2, pp. 689–704, 2022

    work page 2022

  16. [16]

    Quaternion Invariant Extended Kalman Filtering for Spacecraft Attitude Estimation,

    H. Gui and A. H. J. de Ruiter, “Quaternion Invariant Extended Kalman Filtering for Spacecraft Attitude Estimation,” Journal of Guidance, Control, and Dynamics , vol. 41, pp. 863–878, Apr. 2018

    work page 2018

  17. [17]

    An EKF-SLAM algorithm with consistency properties

    A. Barrau and S. Bonnabel, “An EKF-SLAM algorithm with consi stency properties,” arXiv preprint arXiv:1510.06263 , 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  18. [18]

    InGVIO: A Consistent Invarian t Filter for Fast and High-Accuracy GNSS-Visual-Inertial Odometry ,

    C. Liu, C. Jiang, and H. Wang, “InGVIO: A Consistent Invarian t Filter for Fast and High-Accuracy GNSS-Visual-Inertial Odometry ,” IEEE Robotics and Automation Letters , vol. 8, pp. 1850–1857, Mar. 2023

    work page 2023

  19. [19]

    A Right Invariant Extended Kalman Filter for Object Based SLA M,

    Y . Song, Z. Zhang, J. Wu, Y . Wang, L. Zhao, and S. Huang, “A Right Invariant Extended Kalman Filter for Object Based SLA M,” IEEE Robotics and Automation Letters , vol. 7, pp. 1316–1323, Apr. 2022

    work page 2022

  20. [20]

    G. S. Chirikjian, Stochastic models, information theory, and Lie groups, volume 2: Analytic methods and modern applications , vol. 2. Springer Science & Business Media, 2011

    work page 2011

  21. [21]

    Invariant Kalman Filtering,

    A. Barrau and S. Bonnabel, “Invariant Kalman Filtering,” Annual Review of Control, Robotics, and Autonomous Systems, vol. 1, pp. 237–257, May 2018

    work page 2018

  22. [22]

    Error propagation on the Eucli dean group with applications to manipulator kinematics,

    Y . Wang and G. S. Chirikjian, “Error propagation on the Eucli dean group with applications to manipulator kinematics,” IEEE Transactions on Robotics , vol. 22, no. 4, pp. 591–602, 2006

    work page 2006

  23. [23]

    Nonparametric second-order t heory of error propagation on motion groups,

    Y . Wang and G. S. Chirikjian, “Nonparametric second-order t heory of error propagation on motion groups,” The International journal of robotics research, vol. 27, no. 11-12, pp. 1258–1273, 2008

    work page 2008

  24. [24]

    Bayesian fusi on on lie groups,

    K. C. Wolfe, M. Mashner, and G. S. Chirikjian, “Bayesian fusi on on lie groups,” Journal of Algebraic Statistics , vol. 2, no. 1, 2011

    work page 2011

  25. [25]

    Conti nuous- discrete extended kalman filter on matrix lie groups using co ncentrated gaussian distributions,

    G. Bourmaud, R. M´ egret, M. Arnaudon, and A. Giremus, “Conti nuous- discrete extended kalman filter on matrix lie groups using co ncentrated gaussian distributions,” Journal of Mathematical Imaging and Vision , vol. 51, pp. 209 – 228, 2014

    work page 2014

  26. [26]

    Matrix Fisher–Gaussian distribution on SO(3) × Rn and Bayesian attitude estimation,

    W. Wang and T. Lee, “Matrix Fisher–Gaussian distribution on SO(3) × Rn and Bayesian attitude estimation,” IEEE Transactions on Automatic Control, vol. 67, no. 5, pp. 2175–2191, 2021

    work page 2021

  27. [27]

    K. V . Mardia and P . E. Jupp, eds., Directional Statistics. Wiley Series in Probability and Statistics, Hoboken, NJ, USA: John Wiley & Sons, Inc., Jan. 1999

    work page 1999

  28. [28]

    Ley and T

    C. Ley and T. V erdebout, Modern Directional Statistics . Chapman and Hall/CRC, 1 ed., Aug. 2017

    work page 2017

  29. [29]

    Ley and T

    C. Ley and T. V erdebout, eds., Applied Directional Statistics . Boca Raton, FL: CRC Press, 2019

    work page 2019

  30. [30]

    Orientation statistics,

    T. D. Downs, “Orientation statistics,” Biometrika, vol. 59, no. 3, pp. 665– 676, 1972

    work page 1972

  31. [31]

    The von Mises-Fisher Matrix Di stribu- tion in Orientation Statistics,

    C. G. Khatri and K. V . Mardia, “The von Mises-Fisher Matrix Di stribu- tion in Orientation Statistics,” Journal of the Royal Statistical Society: Series B (Methodological) , vol. 39, pp. 95–106, Sept. 1977

    work page 1977

  32. [32]

    An antipodally symmetric distribution on the s phere,

    C. Bingham, “An antipodally symmetric distribution on the s phere,” The Annals of Statistics , pp. 1201–1225, 1974

    work page 1974

  33. [33]

    Bayesian attitude estimation with the matrix Fishe r distribution on SO (3),

    T. Lee, “Bayesian attitude estimation with the matrix Fishe r distribution on SO (3),” IEEE Transactions on Automatic Control , vol. 63, no. 10, pp. 3377–3392, 2018

    work page 2018

  34. [34]

    Unscented vo n Mises– Fisher Filtering,

    G. Kurz, I. Gilitschenski, and U. D. Hanebeck, “Unscented vo n Mises– Fisher Filtering,” IEEE Signal Processing Letters , vol. 23, pp. 463–467, Apr. 2016

    work page 2016

  35. [35]

    Bingham-Gaussian Distribution on S3 × Rn for Unscented Attitude Estimation,

    W. Wang and T. Lee, “Bingham-Gaussian Distribution on S3 × Rn for Unscented Attitude Estimation,” in 2021 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Syst ems (MFI) , pp. 1–7, IEEE, 2021

    work page 2021

  36. [36]

    Matrix Fisher-Gaussian distribution on SO(3) × Rn for attitude estimation with a gyro bias,

    W. Wang and T. Lee, “Matrix Fisher-Gaussian distribution on SO(3) × Rn for attitude estimation with a gyro bias,” arXiv preprint arXiv:2003.02180, 2020. AUTHOR et al.: TITLE 17

    work page arXiv 2003

  37. [37]

    Bayesian Attitude Estimation with Approximate Mat rix Fisher Distributions on SO(3),

    T. Lee, “Bayesian Attitude Estimation with Approximate Mat rix Fisher Distributions on SO(3),” in 2018 IEEE Conference on Decision and Control (CDC), (Miami Beach, FL), pp. 5319–5325, IEEE, Dec. 2018

    work page 2018

  38. [38]

    Attitude determination using vector observ ations and the singular value decomposition,

    F. L. Markley, “Attitude determination using vector observ ations and the singular value decomposition,” Journal of the Astronautical Sciences , vol. 36, no. 3, pp. 245–258, 1988

    work page 1988

  39. [39]

    G. S. Chirikjian, Engineering Applications of Noncommutative Har- monic Analysis: With Emphasis on Rotation and Motion Groups . CRC press, 2000

    work page 2000

  40. [40]

    Dispersion on a sphere,

    R. A. Fisher, “Dispersion on a sphere,” Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences , vol. 217, no. 1130, pp. 295–305, 1953

    work page 1953

  41. [41]

    Matrix Fisher distribution

    T. Lee, “Matrix Fisher distribution.” Available: https://github.com/fdcl-gwu/Matrix-Fisher-Distribut ion,

    work page

  42. [42]

    Gaussian statistics for pala eomagnetic vectors,

    J. J. Love and C. G. Constable, “Gaussian statistics for pala eomagnetic vectors,” Geophysical Journal International, vol. 152, pp. 515–565, Mar. 2003

    work page 2003