Page - 313 - in Differential Geometrical Theory of Statistics
Image of the Page - 313 -
Text of the Page - 313 -
Entropy2017,19, 7
In this paper,we introduce a sequence of escort distributions, thenwe consider a sequential
structure of escort expectations. It is known that a deformed exponential family naturally has at
least threekindofdifferentstatisticalmanifoldstructures [6,11]. Thenweshowthatsuchstatistical
manifoldstructurescanbeobtainedfromasequential structureofescortexpectations. Inparticular,
weshowthataFishermetriconaq-exponential familycanbeobtainedfromthedeformedexpectations
withrespect to thesecondescortdistribution,andacubic form(oranAmari–Chentsovtensorfield,
equivalently) isobtainedfromthedeformedexpectationswithrespect to the thirdescortdistribution.
This paper is written based on the proceeding paper [7]. However, this paper focuses on
deformedexpectationsofaq-exponential family,whereas thepreviouspaper focusedondeformed
independences.Weremarkthatseveralauthorshavebeenstudyingdeformedexpectationsrecently.
See [12,13], forexample.
2.DeformedExponentialFamilies
In thispaper,weassumethatallobjectsaresmooth forsimplicity. Letusreviewpreliminary facts
aboutdeformedexponential functionsanddeformedexponential families. Formoredetails, see [2,6],
forexample.Historically,Tsallis [14] introducedthenotionofq-exponential functionandNaudts [5]
introducedthenotionofq-exponential family togetherwitha furthergeneralization. Suchahistorical
note isprovidedin[2].
LetR++ be the set of all positive real numbers,R++ := {x ∈ R|x> 0}. Let χ be a strictly
increasing function from R++ to R++. We define a χ-logarithm function or a deformed logarithm
functionby
lnχ s := ∫ s
1 1
χ(t) dt.
The inverseof lnχ s is calledaχ-exponential functionoradeformed exponential function,which is
definedby
expχ t :=1+ ∫ t
0 u(s)ds,
where the functionu(s) isgivenbyu(lnχ s)=χ(s).
Fromnowon,we suppose that χ is a power function, that is, χ(t) = tq. Then thedeformed
logarithmandthedeformedexponentialaredefinedby
lnq s := s1−q−1
1−q , (s>0),
expq t :=(1+(1−q)t) 1
1−q , (1+(1−q)t>0).
We say that lnq s is a q-logarithm function and expq t is a q-exponential function. In this case,
the functionu(s) isgivenby
u(s)=(1+(1−q)s) q
1−q ={expq s}q.
Bytakingalimitq→1, thesefunctionscoincidewiththestandardlogarithmlnsandthestandard
exponentialexpt, respectively.
AstatisticalmodelSq is calledaq-exponential family if
Sq := {
p(x,θ) ∣∣∣∣∣p(x;θ)= expq [
n
∑
i=1 θiFi(x)−ψ(θ) ]
, θ∈Θ⊂Rn }
, (1)
whereF1(x), . . . ,Fn(x)are functionsonasamplespaceΩ, θ= t(θ1, . . . ,θn) isaparameter, andψ(θ)
is thenormalizationwithrespect to theparameter θ. Undersuitableconditions,Sq is regardedasa
manifoldwithalocalcoordinatesystem{θ1, . . . ,θn}. Inthiscase,wecall{θi}anaturalcoordinatesystem.
313
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