Page - 317 - in Differential Geometrical Theory of Statistics
Image of the Page - 317 -
Text of the Page - 317 -
Entropy2017,19, 7
Theconnectionââ is called thedual connectionofâwithrespect tog. The triplet (S,ââ,g) isalso
astatisticalmanifold,which iscalled thedual statisticalmanifoldof (S,â,g). Thecubic formisgivenby
thedifferenceof twoafïŹneconnectionsââ andâ:
C(X,Y,Z)= g(ââXYââXY,Z).
WedeïŹne generalized conformal structures for statisticalmanifolds followed toKurose [18].
Twostatisticalmanifolds (S,â,g)and (S,âÂŻ, gÂŻ)are said tobe1-conformally equivalent if thereexists
a functionλ :SâR++ suchthat
g¯(X,Y) = λg(X,Y), (6)
âÂŻXY = âXYâg(X,Y)gradg(lnλ), (7)
wheregradg(lnλ) is thegradientvectorïŹeldof lnλwith respect to g, that is, g(X, lnλ) =X(lnλ).
We say that (S,â,g) is 1-conformally ïŹat if (S,â,g) is locally 1-conformally equivalent to a ïŹat
statisticalmanifold.
Twostatisticalmanifolds (S,â,g)and (S,âÂŻ, gÂŻ)aresaid tobe (â1)-conformally equivalent if there
existsa functionλ :SâR++ suchthat
g¯(X,Y) = λg(X,Y),
âÂŻXY = âXY+d(lnλ)(Y)X+d(lnλ)(X)Y, (8)
where d(lnλ)(X) = X(lnλ). If two statistical manifolds (S,â,g) and (S,âÂŻ, gÂŻ) are
1-conformally equivalent, then their dual statistical manifolds (S,ââ,g) and (S,âÂŻâ, gÂŻ) are
(â1)-conformallyequivalent.
Proposition2. If two statisticalmanifolds (S,â,g) and (S,âÂŻ, gÂŻ) are1-conformally equivalent, then their
cubic formshave the followingrelation:
1
λ C¯(X,Y,Z)=C(X,Y,Z)+g(Y,Z)d(lnλ)(X)+g(Z,X)d(lnλ)(Y)+g(X,Y)d(lnλ)(Z).
Proof. FromEquations (7)and(8),weobtain
âÂŻXY=âXY+d(lnλ)(Y)X+d(lnλ)(X)Y+g(X,Y)gradg(lnλ).
By takinganinnerproductwithrespect tog,weobtain theresult.
5. StatisticalManifoldStructuresonq-ExponentialFamilies
In thissection,weconsiderstatisticalmanifoldstructuresonaq-exponential family. It isknown
thataq-exponential familynaturallyhasat least threekindsofstatisticalmanifoldstructures (cf. [6,8]).
Wereformulate these structures fromtheviewpointof the sequenceof escortdistributions. In this
paper,weomit thedetailsabout informationgeometry. See [21,22] for furtherdetails.
Firstly, we review basic facts about q-exponential family. Let Sq be a q-exponential family.
The normalization Ï(Ξ) on Sq is convex, but may not be strictly convex. In fact, we obtain the
followingproposition.
317
Differential Geometrical Theory of Statistics
- Title
- Differential Geometrical Theory of Statistics
- Authors
- Frédéric Barbaresco
- Frank Nielsen
- Editor
- MDPI
- Location
- Basel
- Date
- 2017
- Language
- English
- License
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Size
- 17.0 x 24.4 cm
- Pages
- 476
- Keywords
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Categories
- Naturwissenschaften Physik