Page - 87 - in Differential Geometrical Theory of Statistics
Image of the Page - 87 -
Text of the Page - 87 -
Entropy2016,18, 386
TheFisher informationdefinesametricgiventoNn aRiemannianmanifoldstructure. The inner
productof twotangentvectors (η1,Ω1)∈Tn, (η2,Ω2)∈Tn at thepoint (μ,Σ)∈Nn isgivenby:
g(μ,Σ)) ((η1,Ω1) ,(η1,Ω1))=η T
1Σ −1η2+ 1
2 Tr (
Σ−1Ω1Σ−1Ω2 )
(180)
NielsChristianBangJespersonhasprovedthat the transformationmodelonRnwithparameter
set Rn×Sym+n are exactly those of the form pμ,Σ = fμ,Σλ where λ is the Lebesque measure,
where fμ,Σ(x)= h (
(x−μ)TΣ−1(x−μ) )
/det(Σ)1/2 and h : [0,+∞[→R+ isacontinuousfunction
with +∞
0 h(s)s n
2−1ds<+∞. Distributionswithdensitiesof this formarecalledellipticdistributions.
To improve understanding of tools, wewill considerGA(n) as a sub-group of affine group,
thatcouldbedefinedbyamatrixLiegroupGaf f , thatacts formultivariateGaussianlaws,as illustrated
inFigure11:
[
Y
1 ]
= [
R1/2 m
0 1 ][
X
1 ]
= [
R1/2X+m
1 ]
, ⎧⎪⎪⎨⎪⎪⎩ (m,R)∈Rn×Sym+(n)
M= [
R1/2 m
0 1 ]
∈Gaf f
X≈ℵ(0, I)→Y≈ℵ(m,R) (181)
Wecanverify thatM is aLiegroupwith classicalproperties, thatproductofMpreserves the
structure, theassociativity, thenon-commutativity,andtheexistenceofneutralelement:
M1 ·M2= [
R1/21 m1
0 1 ][
R1/22 m2
0 1 ]
= [
R1/21 R 1/2
2 R 1/2
1 m2+m1
0 1 ]
M2 ·M1= [
R1/22 m2
0 1 ][
R1/21 m1
0 1 ]
= [
R1/22 R 1/2
1 R 1/2
2 m1+m2
0 1 ] ⎫⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎭
⇒ ⎧⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎩ M1 ·M2∈Gaf f
M2 ·M1∈Gaf f
M1 ·M2 =M2 ·M1
M1 ·(M2 ·M3)= (M1 ·M2) ·M3
M1 · I=M1 (182)
Wecanalsoobserve that the inversepreserves thestructure:
M= [
R1/2 m
0 1 ]
⇒M−1R =M−1L =M−1= [
R−1/2 −R−1/2m
0 1 ]
∈Gaf f (183)
To thisLiegroupwecanassociateaLiealgebrawhoseunderlyingvector space is the tangent
spaceof theLiegroupat the identityelementandwhichcompletelycaptures the local structureof
thegroup. ThisLiegroupactssmoothlyonthemanifold,andactsonthevectorfields.Anytangent
vectorat the identityofaLiegroupcanbeextendedtoa left (respectivelyright) invariantvectorfield
by left (respectivelyright) translating the tangentvector tootherpointsof themanifold. This identifies
the tangentspaceat the identityg=TI(G)withthespaceof left invariantvectorfields,andtherefore
makes the tangentspaceat the identity intoaLiealgebra, calledtheLiealgebraofG.
LG : {
Gaf f →Gaf f
M →LMN=M ·N andRG : {
Gaf f →Gaf f
M →RMN=N ·M (184)
87
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