Page - 110 - in Differential Geometrical Theory of Statistics
Image of the Page - 110 -
Text of the Page - 110 -
Entropy2016,18, 386
Let transformationkμ(θ)definedonΘ(u) interiorofDμ= { θ∈E∗,Lμ<∞ }
:
kμ(θ)= logLμ(θ) (C2)
naturalexponential familiesaregivenby:
F(μ)= {
P(θ,μ)(dx)= e〈θ,x〉−kμ(θ)μ(dx),θ∈Θ(μ) }
(C3)
with injective function(domainofmeans):
k′μ(θ)=
E xP(θ,μ)μ(dx) (C4)
the inverse function:
ψμ :MF→Θ(μ)withMF= Im (
k′μ(Θ(μ)) )
(C5)
andtheCovarianceoperator:
VF(m)= k′′μ (
ψμ(m) )
= (
ψ′μ(m) )−1
, m∈MF (C6)
MeasuregeneretadbyafamilyF is thengivenby:
F(μ)=F(μ′)⇔∃(a,b)∈E∗×R,suchthatμ′(dx)= e〈a,x〉+bμ(dx) (C7)
Let F an exponential family of E generated by μ and ϕ : x → gϕx+vϕ with gϕ ∈ GL(E)
automorphismsofEandvϕ ∈ E, then the familyϕ(F)= {ϕ(P(θ,μ)) ,θ∈Θ(μ)} is anexponential
famillyofEgeneratedbyϕ(μ)
DefinitionC1.Anexponential familyF is invariantbyagroupG(affinegroupofE), if
∀ϕ∈G,ϕ(F)=F : ∀μ,F(ϕ(μ))=F(μ) (C8)
(the contrarycouldbe false)
ThenMurielCasalishas established the following theorem:
TheoremC1(Casalis).Let F=F(μ)anexponential familyofEandGaffinegroupofE, thenF is invariant
byG if andonly:
∃a :G→E∗,∃b :G→R, suchthat:
∀(ϕ,ϕ′)∈G2, ⎧⎨⎩ a(ϕϕ′)=tg−1ϕ a(ϕ′)+a(ϕ)
b(ϕϕ′)= b(ϕ)+b(ϕ′)− 〈
a(ϕ′) ,g−1ϕ vϕ 〉
∀ϕ∈G,ϕ(μ)(dx)= e〈a(ϕ),x〉+b(ϕ)μ(dx) (C9)
WhenG isa linear subgroup, b is a character ofGanda couldbeobtainedby thehelpof cohomologyof
Liegroups.
IfwedefineactionofGonE∗ by:
g ·x=tg−1x,g∈G,x∈E∗ (C10)
It canbeverified that:
a(g1g2)= g1 ·a(g2)+a(g1) (C11)
110
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