Seite - 413 - in Differential Geometrical Theory of Statistics
Bild der Seite - 413 -
Text der Seite - 413 -
Entropy2016,18, 425
onthesymplectic spaceT∗FM. Letting πˆdenote theprojectiononthebundleT∗FM→FM, (8)gives
H(z)= 1
2 〈gFM(z)|gFM(z)〉= 12‖z◦hπˆ(z)◦ πˆ(z)‖ 2
(Rd)∗= 1
2 d
∑
i=1 ξ(Hi(u))2.
Normalgeodesics insub-Riemannianmanifoldssatisfy theHamilton–Jacobiequations [23]with
Hamiltonianflow
z˙t=XH=Ω#dH(z) (13)
whereΩhere is thecanonical symplectic formonT∗FM (e.g., [25]).Wedenote (13) theMPPequations,
andweletprojectionsxt=πT∗FM(zt)ofminimizingcurvessatisfying(13)bedenotednormalMPPs.
Thesystem(13)has2(d+d2)degreesof freedom, incontrast to theusual2ddegreesof freedomforthe
classicalgeodesicequation.Of these,d2 describes thecurrent frameat time t,while theremainingd2
allowsthecurve to“twist”whilestillbeinghorizontal.Wewill see thiseffectvisualized inSection6.
Ina local canonical trivializationz=(u,ξ), (13)gives theHamilton–Jacobiequations
u˙= ∂ξH(u,ξ)= gFM(u,ξ)= hu (
u(ξ(H1(u)), . . . ,ξ(Hd(u)))T )
ξ˙=−∂uH(u,ξ)=−∂u12‖ξ◦hu◦u‖ 2
(Rd)∗=−∂u 1
2 d
∑
i=1 ξ(Hi(u))2. (14)
Using(3),wehavefor thesecondequation
ξ˙=− d
∑
i=1 ξ(Hi(u))ξ(∂uhu(uei))
=− d
∑
i=1 ξ(Hi(u))ξ (
ψ(u,Ru(uei,π∗(∂u)))+∂hu(uei)ψ ( u,ψ−1(C(∂u)) )
+∂hu(uei)hu(π∗(∂u)) )
=−ξ(ψ(u,Ru(π∗(u˙),π∗(∂u)))+∂u˙ψ(u,ψ−1(C(∂u)))+∂u˙hu(π∗(∂u))). (15)
Here∂u˙denotesu-derivative in thedirection u˙, equivalently∂u˙hu(v)= ∂t(hu)(v).While thefirst
equationof (14) involvesonly thehorizontalpartofξ, thesecondequationcouples theverticalpartof
ξ throughtheevaluationofξonthe termψ(u,Ru(π∗(u˙),π∗(∂u)). If theconnection iscurvature-free,
whichinnon-flatcasesimpliesthat itcarriestorsion, thisvertical termvanishes.Conversely,whenM is
Riemannian,C thegRLevi–Civitaconnection,andu0 isgRorthonormal,gFM(hu(v),hu(w))= gR(v,w)
forallv,w∈Tπ(ut)M. In thiscase,anormalMPPπ(ut)willbeaRiemanniangRgeodesic.
4.2. Evolution inCoordinates
Incoordinatesu=(xi,uiα,ξi,ξiα) forT ∗FM,wecanequivalentlywrite
x˙i= gijξj+g ijβξjβ =W ijξj−WihΓjβh ξjβ
X˙iα= g iαjξj+g iαjβξjβ =−ΓiαhWhjξj+ΓiαkWkhΓ jβ
h ξjβ
ξ˙i=−12 (
∂yig hk
y ξhξk+∂yig hkδ
y ξhξkδ+∂yig hγk
y ξhγξk+∂yig hγkδ
y ξhγξkδ )
ξ˙iα =− 1
2 (
∂yiαg hk
y ξhξk+∂yiαg hkδ
y ξhξkδ+∂yiαg hγk
y ξhγξk+∂yiαg hγkδ
y ξhγξkδ )
withΓhγk,i for∂yiΓ hγ
k , andwhere
∂ylg ij=0 , ∂ylg ijβ =−WihΓjβh,l , ∂ylgiαj=−Γiαh,lWhj , ∂ylgiαjβ =Γiαk,lWkhΓ jβ
h +Γ iα
kW khΓ jβ
h,l ,
413
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