Web-Books
in the Austria-Forum
Austria-Forum
Web-Books
Naturwissenschaften
Physik
Differential Geometrical Theory of Statistics
Page - 396 -
  • User
  • Version
    • full version
    • text only version
  • Language
    • Deutsch - German
    • English

Page - 396 - in Differential Geometrical Theory of Statistics

Image of the Page - 396 -

Image of the Page - 396 - in Differential Geometrical Theory of Statistics

Text of the Page - 396 -

Entropy2016,9, 337 As expected, onlymoves normal to the trajectorywill change at first order the value of the criterion: thedisplacementof thecurveγjwill thusbeperformedat t in thenormalbundle toγj and isgiven,upto the (∑Ni=1 li) −1 term,by: ∫ Ω ( γj(t)−x ‖γj(t)−x‖ ) N K′ (‖γj(t)−x‖) log(d˜(x))dx‖γ′j(t)‖ − (∫ Ω K (‖γj(t)−x‖) log(d˜(x))dx) ( γ′′j (t) ‖γ′j(t)‖ ) N + (∫ Ω d˜(x) log(d˜(x))dx )( γ′′j (t) ‖γ′j(t)‖ ) N . (19) Thefirst termintheexpressionwill favormoves towardsareasofhighdensity,while thesecond andthirdonesaremovingalongnormalvectorandwill straightenthe trajectory. This lastpointcanbe enlightenedbyconsideringthecaseofasingleplanarcurvewithfixedendpoints. Proposition3. Let a,bbefixedpoints inR2 andKbeakernel as in (7). Thesegment [a,b] is a criticalpoint for the entropyassociatedwith the curve systeminR2 consistingof single smoothpathswith endpoints a,b. Proof. Let the segment [a,b] be parametrized as γ: t ∈ [0,1] → a+ tvwith v the vector (b− a). Startingwith theexpression(19), it is clear that thesecondandthirdtermsoccurring in the formula will vanishas thesecondderivativeofγ is zero. Letube theunitnormalvector toγ. Anypoint x inR2 canbewrittenasx= a+θv+ξu, θ,ξ∈R, so thatγ(t)−x=(t−θ)v−ξuand‖γ(t)−x‖=√ (t−θ)2‖b−a‖2+ξ2. Thechangeofvariablesx→ (θ,ξ)has Jacobian‖v‖= ‖b−a‖. Forafixed t∈ [0,1], itbecomes: ∫ R2 ( γ(t)−x ‖γ(t)−x‖ ) N K′(‖γ(t)−x‖) log(d˜(x))dx‖γ′(t)‖= ‖b−a‖2 ∫ R ∫ R −ξ√ (t−θ)2‖b−a‖2+ξ2K ′ (√ (t−θ)2‖b−a‖2+ξ2 ) log(d˜(θ,ξ))dξdθ. (20) Thedensity d˜ for theγcurve isexpressed inξ,θ coordinatesas: ∫ [0,1] K (√ (t−θ)2‖b−a‖2+ξ2 ) dt andisanevenfunction inξ. Thesameis true forK′(‖γ(t)−x‖). Finally, themapping: ξ → −ξ√ (t−θ)2‖b−a‖2+ξ2 is odd for afixed θ, so that thewhole integrand is oddas a functionof ξ. By theFubini theorem, integratingfirst inξwill thereforeyieldavanishing integral,provingtheassertion. The result still holds inRq, the only different aspect being that x is now expanded as x = a+θv+∑ q−1 i=1 ξiui withui, i = 1,. . . ,q−1 anorthonormal basis of the orthogonal complement of Rv inRq. Rewritingγ(t)−x=(t−θ)v−∑q−1i=1 ξiui and‖γ(t)−x‖= √ (t−θ)2‖b−a‖2+∑q−1i=1 ξ2i , thesameparityargumentcanbeappliedonanyof thecomponentsξi, i=1,. . . ,q−1, showingthat the integral isvanishing. Theeffectofcurvestraighteningispresentwhenminimizingtheentropyofawholecurvesystem, but is counterbalancedbythegatheringeffect.Dependingonthechoiceof thekernelbandwidth,one or theothereffect isdominant: straightening ispreeminent for lowvalues,beingtheonlyremaining 396
back to the  book Differential Geometrical Theory of Statistics"
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
Web-Books
Library
Privacy
Imprint
Austria-Forum
Austria-Forum
Web-Books
Differential Geometrical Theory of Statistics