Seite - 348 - in Differential Geometrical Theory of Statistics
Bild der Seite - 348 -
Text der Seite - 348 -
Entropy2016,18, 396
Onealreadyencounters theproblemofdensityestimation in thehyperbolicspace forelectrical
impedance [15],networks [16]andradarsignals [17]. In [18], ageneralizationof theGaussian lawon
thehyperbolicspacewasproposed.Apart from[19],whereauthorsproposeageneralizationof the
Gaussian law,probabilitydensityestimationontheSiegel spacehasnotyetbeenaddressed.
Thecontributionsof thepaperare the following. Wereviewthemainnonparametricdensity
estimationtechniquesontheSiegeldisk.Weprovidesomerathersimpleexplicit expressionsof the
kernelsdefinedbyPelletier in [1]. These expressionsmake thekerneldensity estimation themost
adaptedmethod.Wepresentvisual resultsofestimateddensities in thesimplecasewhere theSiegel
diskreduces to thePoincarédisk.
Thepaperbeginswithan introduction to theSiegel space inSection2. Section3reviewsthemain
non-parametricdensityestimation techniquesontheSiegel space. Section3.3contains theoriginal
resultsof thepaper. Section4presentsanapplicationtoradardataestimation.
2.TheSiegelSpace
This sectionpresents facts about theSiegel space. The interested reader canfindmoredetails
in [20,21]. ThenecessarybackgroundonLiegroupsandsymmetric spacecanbefoundin[22].
2.1. TheSiegelUpperHalfSpace
TheSiegelupperhalf space isageneralizationof thePoincaréupperhalf space (see [23]) fora
descriptionof thehyperbolic space. LetSym(n)bethespaceof real symmetricmatricesofsizen×n
andSym+(n) thesetof real symmetricpositivedefinitematricesofsizen×n. TheSiegelupperhalf
space isdefinedby
Hn={Z=X+ iY|X∈Sym(n),Y∈Sym+(n)} .
Hn isequippedwith the followingmetric:
ds=2tr(Y−1dZY−1dZ).
Thesetof real symplecticmatricesSp(n,R) isdefinedby
g∈Sp(n,R)⇔ gtJg= J,
where
J= (
0 In
−In 0 )
,
and In is then×n identitymatrix. Sp(n,R) is a subgroupofSL2n(R), the setof 2n×2n invertible
matrices of determinant 1. Let g = (
A B
C D )
∈ Sp(n,R). Themetric ds is invariant under the
followingactionofSp(n,R),
g.Z=(AZ+B)(CZ+D)−1.
Thisaction is transitive, i.e.,
∀Z∈Hn,∃g∈Sp(n,R),g.iI=Z.
The stabilizerK of iI is the set of elements g of Sp(n,R)whose action leaves iI fixed. K is a
subgroupofSp(n,R)calledthe isotropygroup.Wecanverify that
K= {(
A B
−B A )
,A+ iB∈SU(n) }
.
348
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