Abstract
The dualistic structure of statistical manifolds in information geometry yields eight types of (possibly mixed type) geodesic triangles passing through three given points, the triangle vertices. The interior angles of geodesic triangles can sum up to \(\pi \) like in Euclidean/Mahalanobis flat geometry, or exhibit otherwise angle excesses or angle defects. In this work, we initiate the study of geodesic triangles in dually flat spaces, termed Bregman manifolds, where a generalized Pythagorean theorem holds. We consider non-self dual Bregman manifolds since Mahalanobis self-dual manifolds amount to Euclidean geometry. First, we show how to construct geodesic triangles with either one, two, or three interior right angles, whenever it is possible. Second, we report a construction of triples of points for which the dual Pythagorean theorems hold simultaneously at a point, yielding two dual pairs of dual-type geodesics with right angles at that point.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
More precisely, a geodesic \(\gamma _{pq}^\nabla (t)\) with respect to an affine connection \(\nabla \) satisfies \(\nabla _{\dot{\gamma }_{pq}} \dot{\gamma }_{pq}=0\). A \(\nabla \)-geodesic is an autoparallel curve at it is invariant by affine reparameterization of t (i.e., \(t'=at+b\)).
- 2.
We do consider at a triangle vertex only pairs of geodesics with interior angles linking the two other triangle vertices.
- 3.
Given an affine connection \(\nabla \), the \(\nabla \)-geodesic is an autoparallel curve [15].
- 4.
The notion of dual connections of information geometry is more general than the notion of conjugate connections of affine differential geometry [10] which stems from dual affine immersions.
- 5.
In differential geometry, the tangent plane at a point p is the space of all linear derivations that satisfies the Leibniz’s rule. A basis \(\{t_i\}_i\) of \(T_p\) is such that \(t_i(f)=\frac{\partial f}{\partial \theta ^i}\).
- 6.
A each quadrilateral vertex, we have 4 geodesics defining 6 interior angles between them.
References
Akaho, S., Hino, H., Murata, N.: On a convergence property of a geometrical algorithm for statistical manifolds (2019). arXiv:1909.12644
Amari, S.: Information geometry of positive measures and positive-definite matrices: decomposable dually flat structure. Entropy 16(4), 2131–2145 (2014)
Amari, S.-I.: Information Geometry and its Applications, vol. 194. Springer, Berlin (2016)
Banerjee, A., Merugu, S., Dhillon, I.S., Ghosh, J.: Clustering with Bregman divergences. J. Mach. Learn. Res. 6, 1705–1749 (2005)
Boissonnat, J.-D., Nielsen, F., Nock, R.: Bregman Voronoi diagrams. Discret. Comput. Geom. 44(2), 281–307 (2010)
Crouzeix, J.-P.: A relationship between the second derivatives of a convex function and of its conjugate. Math. Program. 13(1), 364–365 (1977)
Edelsbrunner, H., Wagner, H.: Topological data analysis with Bregman divergences. In: 33rd International Symposium on Computational Geometry (SoCG 2017). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik (2017)
Eguchi, S., et al.: Geometry of minimum contrast. Hiroshima Math. J. 22(3), 631–647 (1992)
Eisenhart, L.P.: Riemannian Geometry. Princeton University Press, Princeton (1997)
Kurose, T.: Dual connections and affine geometry. Math. Z. 203(1), 115–121 (1990)
Lauritzen, S.L.: Statistical manifolds. Differ. Geom. Stat. Inference 10, 163–216 (1987)
Mahalanobis, P.C.: On the generalized distance in statistics. Proc. Natl. Inst. Sci. (Calcutta) 2, 49–55 (1936)
Nielsen, F.: Legendre transformation and information geometry (2010)
Nielsen, F.: Cramér-Rao lower bound and information geometry. In: Connected at Infinity II, pp. 18–37. Springer, Berlin (2013)
Nielsen, F.: An elementary introduction to information geometry (2018). arXiv:1808.08271
Nielsen, F., Boltz, S.: The Burbea-Rao and Bhattacharyya centroids. IEEE Trans. Inf. Theory 57(8), 5455–5466 (2011)
Nielsen, F., Bhatia, R.: Matrix Information Geometry. Springer, Berlin (2013)
Nielsen, F., Garcia, V. Statistical exponential families: a digest with flash cards (2009). arXiv:0911.4863
Nielsen, F., Hadjeres, G.: Monte Carlo information geometry: the dually flat case (2018). arXiv:1803.07225
Nielsen, F., Nock, R.: Sided and symmetrized Bregman centroids. IEEE Trans. Inf. Theory 55(6), 2882–2904 (2009)
Nielsen, F., Nock, R. Skew Jensen-Bregman Voronoi diagrams. In: Transactions on Computational Science, vol. XIV, pp. 102–128. Springer, Berlin (2011)
Nielsen, F., Nock, R.: On the geometry of mixtures of prescribed distributions. In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2861–2865. IEEE (2018)
Nielsen, F., Muzellec, B., Nock, R.: Classification with mixtures of curved Mahalanobis metrics. In: 2016 IEEE International Conference on Image Processing (ICIP), pp. 241–245. IEEE (2016)
Nock, R., Magdalou, B., Briys, E., Nielsen, F.: Mining matrix data with Bregman matrix divergences for portfolio selection. In: Matrix Information Geometry, pp. 373–402. Springer, Berlin (2013)
Norden, A.P.: On pairs of conjugate parallel displacements in multidimensional spaces. Dokl. Akad. Nauk SSSR 49(9), 1345–1347 (1945)
Pietra, S.F., Pietra, V.D., Lafferty, J.: Inducing features of random fields. IEEE Trans. Pattern Anal. Mach. Intell. 19(4), 380–393 (1997)
Rosenfeld, B.A.: A History of Non-Euclidean Geometry: Evolution of the Concept of a Geometric Space, vol. 12. Springer Science & Business Media (2012)
Sen, R.N.: Parallel displacement and scalar product of vectors. In: Proceedings of the National Institute of Sciences of India, vol. 14, p. 45. National Institute of Sciences of India (1948)
Shima, H.: The Geometry of Hessian Structures. World Scientific (2007)
Sturmfels, B.: Solving Systems of Polynomial Equations, vol. 97. American Mathematical Society (2002)
Wolf, J.A.: Spaces of Constant Curvature, vol. 372. American Mathematical Society (1972)
Wong, T.-K.L.: Logarithmic divergences from optimal transport and Rényi geometry. Inf. Geom. 1(1), 39–78 (2018)
Acknowledgements
Figures were programmed using processing.org
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Ethics declarations
Views and opinions expressed are those of the authors and do not necessarily represent official positions of their respective companies.
A. Notations
A. Notations
- F :
-
Strictly convex and \(C^3\) real-valued function
- \(F^*\) :
-
Dual Legendre-Fenchel convex conjugate
- \(\theta (p)=(\theta ^1(p),\ldots ,\theta ^D(p))\) :
-
primal coordinates of point p
- \(\eta (p)=(\eta _1(p),\ldots ,\eta _D(p))\) :
-
dual coordinates of point p
- \(\theta _{ab}=\theta _a-\theta _b\) :
-
notational shortcut
- \(\eta _{ab}=\eta _a-\eta _b\) :
-
notational shortcut
- \(D_F(p:q)\) :
-
Divergence between points
- \(B_F(\theta (p):\theta (q))\) :
-
Bregman divergence
- \(A_F(\theta (p):\eta (q))\) :
-
Fenchel-Young divergence
- \((M,g,\nabla ,\nabla ^*)\) :
-
Dually flat space (Bregman manifold)
- \(T_p\) :
-
Tangent plane at p
- \(g_p(u,v)\) :
-
inner product between two vectors u and v of \(T_p\)
- [\(g_{ij}\):
-
=[\(g(e_i,e_j)\)]\(_{ij}=\nabla ^{2F}(\theta )\)] Riemannian metric
- [\({g^*}^{ij}\):
-
=[\(g^*({e^*}^i,{e^*}^j)\)]\(_{ij}=\nabla ^2F^*(\eta )\)] dual Riemannian metric
- \(\prod _{p,q}(v)\) :
-
primal parallel transport of \(v\in T_p\) to \(T_q\)
- \(\prod _{p,q}^*(v)\) :
-
dual parallel transport of \(v\in T_p\) to \(T_q\)
- \(\gamma _{ab}(t)\) :
-
Primal geodesic: \(\theta (\gamma _{ab}(t))=(1-t)\theta (a)+t\theta (b)\)
- \(\gamma _{ab}(t)^*\) :
-
Dual geodesic: \(\eta (\gamma _{ab}^*(t))=(1-t)\eta (a)+t\eta (b)\)
- \((v)_B\) :
-
vector components in basis B, arranged in a D-tuple
- [v:
-
\(_B\)] vector components in basis B, arranged in a D-dimensional column vector
- \(B_p=\{e_i=\partial _i=\frac{\partial }{\partial \theta ^i}\}\) :
-
natural basis at \(T_p\)
- \(B_p^*=\{{e^*}^i=\partial ^i=\frac{\partial }{\partial \eta _i}\}_i\) :
-
reciprocal basis at \(T_p\) so that \(g(e_i,{e^*}^j)=\delta _{i}^j\)
- \(v_{ab}=\frac{d}{\mathrm {d}t}\gamma _{ab}(0)=\dot{\gamma }_{ab}(0)\) :
-
tangent vector of \(\gamma _{ab}(t)\) at a with contravariant components \(\theta (b)-\theta (a)\)
- \(v_{ab}^*=\frac{d}{\mathrm {d}t}\gamma _{ab}^*(0)=\dot{\gamma }_{ab}^*(0)\) :
-
tangent vector of \(\gamma _{ab}^*(t)\) at a with covariant components \(\eta (b)-\eta (a)\)
- [\(v^i\):
-
\(_B\)] contravariant components of vector v, \(v^i=g(v,{e^*}^i)\)
- [\(v_i\):
-
\(_B\)] covariant components of vector v (meaning \([v]_{B^*}\)), \(v_i=g(v,e_i)\)
- \(g_p(u,v)\) :
-
inner product at \(T_p\) of two vectors: \(g_p(u,v)=u_i v^i=u^i v_i\).
Rights and permissions
Copyright information
© 2021 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Nielsen, F. (2021). On Geodesic Triangles with Right Angles in a Dually Flat Space. In: Nielsen, F. (eds) Progress in Information Geometry. Signals and Communication Technology. Springer, Cham. https://doi.org/10.1007/978-3-030-65459-7_7
Download citation
DOI: https://doi.org/10.1007/978-3-030-65459-7_7
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-65458-0
Online ISBN: 978-3-030-65459-7
eBook Packages: Physics and AstronomyPhysics and Astronomy (R0)