Page - 273 - in Differential Geometrical Theory of Statistics
Image of the Page - 273 -
Text of the Page - 273 -
Entropy2016,18, 407
whose inverse is the so called theKaniadakis’κ-logarithm logk : (0,∞)→R,which isgivenby
logκ(u)= { uκ−u−κ
2κ , ifκ =0,
ln(u), ifκ=0.
It is clear thatexpκ(·) satisfies (a1)and (a2). Letu0: T→ (0,∞)beanymeasurable function
forwhich∫
Texpκ(u0)dμ<∞.Wewill showthatu0 satisfies expression (1). Foranyu∈Randα≥1,wecanwrite
expκ(αu)=α 1/|κ|(|κ|u+ √
1/α2+ |κ|2u2)1/|κ|
≤α1/|κ|(|κ|u+ √
1+ |κ|2u2)1/|κ|
=α1/|κ|expκ(u),
whereweused thatexpκ(·)= exp−κ(·). Then,weconclude that ∫
Texpκ(αu0)dμ<∞ for allα≥0. Fixany
measurable functionc: T→R such that∫T ϕ(c)dμ=1. For
eachλ>0,wehave∫
T expκ(c+λu0)dμ≤ 1
2 ∫
T expκ(2c)dμ+ 1
2 ∫
T expκ(2λu0)dμ
≤21/|κ|−1 ∫
T expκ(c)dμ+2 1/|κ|−1 ∫
T expκ(λu0)dμ
<∞,
which shows that expκ(·) satisfies (a3). Therefore, the Kaniadakis’ κ-exponential expκ(·) is an example
of ϕ-function.
Therestrictionthat ∫
T ϕ(c)dμ=1canbeweakened,asasserted in thenext result.
Lemma1. Let c˜: T→Rbeanymeasurable functionsuchthat∫T ϕ(c˜)dμ<∞. Then,∫T ϕ(c˜+λu0)dμ<∞
for allλ>0.
Proof. Notice that if ∫
T ϕ(c˜)dμ≥1, then ∫
T ϕ(c˜−αu0)dμ=1forsomeα>1. Fromthedefinitionof
u0, it follows that ∫
T ϕ(c˜+λu0)dμ= ∫
T ϕ(c+(α+λ)u0)dμ<∞,where c= c˜−αu0. Nowassume
that ∫
T ϕ(c˜)dμ < 1. Consider any measurable set A ⊆ T with measure 0 < μ(A) < μ(T).
Letu: T→ [0,∞)beameasurable function supportedon A satisfying ϕ(c˜+u)1A = [ϕ(c˜)+α]1A,
whereα=(1−∫T ϕ(c˜)dμ)/μ(A). Defining c=(c˜+u)1A+ c˜1T\A,wesee that∫T ϕ(c)dμ=1. Bythe
definitionofu0,wecanwrite∫
T ϕ(c˜+λu0)dμ≤ ∫
T ϕ(c+λu0)dμ<∞, foranyλ>0,
which is thedesiredresult.
AsaconsequenceofLemma1,condition(a3)canbereplacedbythe followingone:
(a3’) Thereexistsameasurable functionu0: T→ (0,∞)
suchthat∫
T ϕ(c+λu0)dμ<∞, forallλ>0, (2)
foreachmeasurable function c: T→R forwhich∫T ϕ(c)dμ<∞.
Without the equivalence between conditions (a3) and (a3’), we could not generalize
Rényidivergence in themannerwepropose. In fact,ϕ-functionscouldbedefineddirectly in terms
of (a3’), without mentioning (a3). We chose to begin with (a3) because this condition appeared
initially in [12].
273
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