Page - 24 - in Differential Geometrical Theory of Statistics
Image of the Page - 24 -
Text of the Page - 24 -
Entropy2016,18, 370
Theentropy s(ρ)ofaprobabilitydensityρhasveryremarkablevariationalpropertiesdiscussed
in the followingdefinitionsandproposition.
Definition17. Letρbe thedensityof a smoothstatistical state onasymplecticmanifold (M,ω).
1. For each function f defined on M, taking its values inR or in some finite-dimensional vector space,
such that the integral on the righthandsideof the equality
Eρ(f)= ∫
M fρdλω
converges, thevalueEρ(f)of that integral is called themeanvalueof f with respect toρ.
2. Let f be a smooth function on M, taking its values inR or in some finite-dimensional vector space,
satisfying theproperties statedabove.Asmooth infinitesimalvariationofρwithfixedmeanvalueof f is a
smoothmap,definedontheproduct ]−ε,ε[×M,withvalues inR+,where ε>0,
(τ,z) → ρ(τ,z) , τ∈]−ε,ε[, z∈M ,
such that
• forτ=0andanyz∈M,ρ(0,z)= ρ(z),
• for eachτ∈]−ε,ε[ , z → ρτ(z)= ρ(τ,z) is a smoothprobabilitydensityonMsuch that
Eρτ(f)= ∫
M ρτ fdλω=Eρ(f) .
3. The entropy function s is said to be stationary at the probability density ρ with respect to smooth
infinitesimal variations of ρ with fixed mean value of f, if for any smooth infinitesimal variation
(τ,z) → ρ(τ,z)ofρwithfixedmeanvalueof f
ds(ρτ)
dτ ∣∣∣
τ=0 =0.
Proposition 8. Let H : M→R be a smoothHamiltonian on a symplecticmanifold (M,ω) and ρ be the
densityofasmoothstatistical stateonMsuchthat the integraldefiningthemeanvalueEρ(H)ofHwithrespect
toρ converges. The entropy functions is stationaryatρwithrespect to smooth infinitesimalvariationsofρwith
fixedmeanvalueofH, if andonly if there exists a real b∈R such that, for all z∈M,
ρ(z)= 1
P(b) exp (−bH(z)) , with P(b)= ∫
M exp(−bH)dλω .
Proof. Letτ → ρτbeasmoothinfinitesimalvariationofρwithfixedmeanvalueofH. Since ∫
M ρτdλω
and ∫
M ρτHdλω donotdependonτ, it satisfies, forallτ∈]−ε,ε[
,∫
M ∂ρ(τ,z)
∂τ dλω(z)=0, ∫
M ∂ρ(τ,z)
∂τ H(z)dλω(z)=0.
Moreoveraneasycalculation leads to
ds(ρτ)
dτ ∣∣∣
τ=0 =− ∫
M ∂ρ(τ,z)
∂τ ∣∣∣
τ=0 (1+ log (
ρ(z) )
dλω(z) .
Awellknownresult incalculusofvariationsshowsthat theentropyfunction s is stationaryatρ
withrespect tosmooth infinitesimalvariationsofρwithfixedmeanvalueofH, if andonly if there
exist tworealconstants aandb, calledLagrangemultipliers, suchthat, forallz∈M,
1+ log(ρ)+a+bH=0,
24
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