Base Change for Fisher Metrics: Case of the q−Gaussian Inverse Distribution

Authors: Gabriel I. Loaiza O., Carlos A. Cadavid M., Juan C. Arango P.

Abstract:

It is known that the Riemannian manifold determined by the family of inverse Gaussian distributions endowed with the Fisher metric has negative constant curvature κ = −1/2 , as does the family of usual Gaussian distributions. In the present paper, firstly we arrive at this result by following a different path, much simpler than the previous ones. We first put the family in exponential form, thus endowing the family with a new set of parameters, or coordinates, θ1, θ2; then we determine the matrix of the Fisher metric in terms of these parameters; and finally we compute this matrix in the original parameters. Secondly, we define the Inverse q−Gaussian distribution family (q < 3), as the family obtained by replacing the usual exponential function by the Tsallis q−exponential function in the expression for the Inverse Gaussian distribution, and observe that it supports two possible geometries, the Fisher and the q−Fisher geometry. And finally, we apply our strategy to obtain results about the Fisher and q−Fisher geometry of the Inverse q−Gaussian distribution family, similar to the ones obtained in the case of the Inverse Gaussian distribution family.

Keywords: Base of Changes, Information Geometry, Inverse Gaussian distribution, Inverse q-Gaussian distribution, Statistical Manifolds.

References:

[1] Amari, Shun-ichi and Ohara, Atsumi. Geometry of q-Exponential Family of Probability Distributions. Entropy 13, 2011, pp. 1170-1185. DOI: 10.3390/e13061170.
[2] Amari, Shun-ichi and Nagaoka, Hiroshi. Methods of Information Geometry. Translation of Mathematics Monographs, volume 191. American Mathematics Society, Oxford University Press, 2000.
[3] Cadavid, Carlos; Loaiza, Gabriel, and Arango, Juan Carlos. Geodesic in the manifold generated by the q-Gaussian distribution. Differential Geometry and its Applications. It is currently under review, submitted since April in 2021.
[4] Chhikara, Raj and Folks, Leroy. The inverse Gaussian distribution: theory, methodology and applications. Marcel Dekker, Inc., New York, 1989.
[5] Dodson, C.T.J. Information Geometry. School Mathematics, Manchester University. DOI:10.1109/TMI.2010.2086464.
[6] Ghoshdastidar, Debarghya. On some Statistical Properties of Multivariate q-Gaussian Distribution and its application to Smoothed Functional Algorithms. International Symposium on Information Theory. IEEE, 2012.
[7] Loaiza; Gabriel and Quiceno; H´ector. A q-exponential statistical Banach manifold. Journal of Mathematical Analysis and Applications, 398, 2013. DOI: https://doi.org/10.1016/j.jmaa.2012.08.046.
[8] Matsuzoe, Hiroshi and Ohara, Atsumi. Geometry for a q-exponential families. WSPC - Proceedings, April 16, 2011. DOI: https://doi.org/10.1142/9789814355476 0004.
[9] Naudts, Jan. Generalised Thermostatistics. Springer, London, 2011. DOI: https://doi.org/10.1007/978-0-85729-355-8.
[10] Naudts, Jan. The q-exponential family in statistical physics. Central European Journal of Physics September 2009, Volume 7, Issue 3, pp 405-413. DOI: 10.1088/1742-6596/201/1/012003.
[11] Tanaya, Daiki, Tanaka, Masaru and Matsuzoe, Hiroshi. Notes on geometry of q-normal distributions. WSPC - Proceedings, May 07, 2011. DOI: https://doi.org/10.1142/9789814355476 0009.
[12] Tsallis, Constantino. Introduction to Nonextensive Statistical Mechanics. Approaching a complex world. Springer, 2009. DOI 10.1007/978-0-387-85359-8.
[13] Wani, J. K. and Kare, D. G. Note on a Characterization of the Inverse Gaussian Distribution. The Annals of Mathematical Statistics, 1970, Vol. 41, No. 3.
[14] Zhang, Zhenning; Sun, Huafei and Zhong, Fengwei. Information geometry power inverse Gaussian distribution. Applied Sciences, Vol.9, 2007, pp. 194-203.