Seite - 208 - in Differential Geometrical Theory of Statistics
Bild der Seite - 208 -
Text der Seite - 208 -
Entropy2016,18, 433
Definition 49. Adatum [U,ΦU×φU,PU,γUU∗] as in the last definition is called a local statistical chart
of [E,π,M,D].
Figure4 is representswhatare crucial steps toward theserachof characteristic invariants, viz
invariants encoding the points of themoduli space of statisticalmodels. At the present Figure 4
describes themoduli spaceof thecategoryFB(Γ,Ξ)
Before dealing with morphisms of the category GM(Ξ,Ω)we introduce a relevant global
geometrical invariant.
8.3.2. TheGlobalProbabilityDensityofaStatisticalModel
WeconsideraCOMPLETE(ormaximalstatistical)atlasofanobject [E,π,M,D] (of thecategory
GM(Ξ,Ω)),namely
AΦ=[Uj,Φj,φj,Pj,γij].
ThefamilyUj isanopencoveringofM. ThepairEj×Uj is thedomainof the localchart(Φj×φj).
Wehave
Ej=EUj.
IfUi∩Uj =∅ thenonehas
φj(x)= γ˜ji ·φij(x) ∀x∈Ui∩Uj.
InparticularA=(Uj,φj) isanaffineatlasof the locallyflatmanifold (M,D).Wehave
Φj(Ey∗)=φj(y∗)×Ξ ∀y∗ ∈Uj.
Thereforeweset
[Ey∗,Ωy∗]=Φ−1j [[φj(y∗)×Ξ],Ω].
TheatlasAΦ satisfiesrequirements(ρ1.1),(ρ1.2)and(ρ1.3). InEUj thelocalfunctionpj isdefinedby
pj=Pj◦Φj.
Wesuppose that
Ui∩Uj =∅.
By thevirtueofof [ρ1.3]onehas
pi(e)= pj(e)
forall e∈EUi∩Uj.
Therebythereexistsauniquefunction
E e→ p(e)∈R
whoserestrictiontoEj coincideswith pj. TherestrictiontoEx isdenotedby px. The triple
(Ex,Ωx,px)
isaprobabilityspace.
Definition50. The function
E e→ p(e)∈R
is called theprobabilitydensityof themodel [E,π,M,D].
208
Differential Geometrical Theory of Statistics
- Titel
- Differential Geometrical Theory of Statistics
- Autoren
- Frédéric Barbaresco
- Frank Nielsen
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2017
- Sprache
- englisch
- Lizenz
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03842-425-3
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 476
- Schlagwörter
- Entropy, Coding Theory, Maximum entropy, Information geometry, Computational Information Geometry, Hessian Geometry, Divergence Geometry, Information topology, Cohomology, Shape Space, Statistical physics, Thermodynamics
- Kategorien
- Naturwissenschaften Physik