On an extension of Lesche stability Aziz El Kaabouchi † Institut Supérieur des Matériaux et Mécaniques Avancés du Mans 44 Avenue Bartholdi, 72000 Le Mans, France Abstract In this paper, we give a new method for proving the Lesche stability of several functionals (Incomplete entropy, Tsallis entropy, entropy). We prove also that the Incomplete entropy for 0 q 1 are Incomplete stable q expectation entropy, q expectation Quantum-Group value and Rényi for all 0 q . Finally, we prove that the new value is stable 1 for all 0 1. Notations and definitions We denote by E f R n . For N N * and 0 , we define the function d , N : R N R N R by nN* 1/ N d , N ( p, p ' ) p i p i' . i 1 We recall that for 1 , d , N is a distance on R N p, p' R N , d , N ( p, p' ) 0 p p' p, p' R N d , N ( p, p' ) d , N ( p' , p) p, p' , p' ' R N , d , N ( p, p' ' ) d , N ( p, p' ) d , N ( p' , p' ' ) for 1 , d , N is a quasi-distance on R N . p, p' R N , d , N ( p, p' ) 0 p p' p, p' R N d , N ( p, p' ) d , N ( p' , p) 1 p, p' , p' ' R , d , N ( p, p' ' ) N 2 1 d , N ( p, p' ) d , N ( p' , p' ' ) Let C be a function defined on A a subset of E f . For N N * and 0 , we denote by C N ,max the number defined by (when it exists) C N ,max sup C ( p) , p A R N . Definition 1 A Function C defined on A a subset of E f is said to be stable if C posses the following property 0 , 0 , N N* , p, p' A R N , d , N ( p, p' ) C ( p) C ( p' ) . C N ,max Remark. The Lesche stability correspond to stability for 1 . Proposition 2. Let 0 and C be a function defined on a subset of E f . If C is stable then C is stable. Proof. It’s easy to see that for all x ]0,1[ , x x and thus if d , N ( p, p' ) then d , N ( p, p' ) / . Proposition 3. If C1 ,C 2 are stable with C1 0 , C2 0 and , R , then C1 C 2 is stable. Proof. We can suppose that C1 0 , C2 0 and , R * . We remark that for all N N * and 0 , 2 C1 C 2 ( p) C1 C 2 ( p' ) C1 C 2 N ,max C1 ( p) C1 ( p' ) C 2 ( p) C 2 ( p' ) C1 C 2 N ,max C1 ( p) C1 ( p' ) C 2 ( p) C 2 ( p' ) C1 C 2 N ,max C1 C 2 N ,max C1 ( p) C1 ( p' ) C 2 ( p) C 2 ( p' ) C1 N ,max C 2 N ,max C1 ( p) C1 ( p' ) C 2 ( p) C 2 ( p' ) C1 N ,max C 2 N ,max Let 0 , there exists 1 0 , 2 0 , such that N N* , p, p' A R N , d , N ( p, p ' ) 1 C1 ( p) C1 ( p' ) C ( p) C 2 ( p' ) and d , N ( p, p ' ) 2 2 . C1 N ,max C 2 N ,max 2 2 We put min 1 , 2 0 , p, p' A R N , d , N ( p, p' ) d , N ( p, p' ) 1 and d , N ( p, p' ) 2 Consequently C1 C 2 ( p) C1 C 2 ( p' ) C1 C 2 N ,max 2 2 . Definition 4. Let q 0, \ 1 . The incomplete entropy S qI is defined on A nN* S qI ( p) 1 N i 1 pi 1 q n p [0,1] , p n i i 1 q 1 by for all p [0,1] N . Theorem 5. For all q 0, \ 1 , the incomplete entropy S qI is stable for all 1 . Proof. It’s easy to see that i) S I q N , max 1 S qI (1 / N )1 / q , , (1 / N )1 / q ii) For all q 0, \ 1 , the function x 1 N 1 q 1 1 x Indeed if q 0,1 , then lim 1 x 1 x 1 1 q 1 . is bounded on 2, 1 and if q 1 , then lim 1 x 1 x Let M 0 , such that for all N N * \ 1 , 1 1 N For all 0 , there exists M implies 1 q 1 q 1 1 q 1 1 q 0. M . 1/ 0 , such that for all N N * , p, p' A R N , d , N ( p, p ' ) 3 S qI ( p ) S qI ( p ' ) S I q N , max p N i 1 1 N M N i 1 ' i pi 1 1 q N p i' p i i 1 1 N p i' p i M N i 1 1 1 q p i' p i M d , N ( p, p ' ) M Definition 6. Let q 0, \ 1 . S q ( p) 1 N i 1 The p i q Tsallis entropy Sq is defined A on n p [0,1] , * nN n p i 1 i 1 by 1 by for all p [0,1] N . q 1 Theorem 7. For all q 0, \ 1 , the Tsallis entropy S q is stable for all 1 . Corollary 8 a) For all 1,1 , the entropy is stable for all 1 . b) For all q 0, \ 1 , the Quantum-Group entropy is stable for all 1 . Proof. a) Let 1,1 . S ( p) N i 1 entropy The pi pi pi 2 S is defined A on n p [0,1] , p nN* n i 1 i for all p [0,1] N . 1 S 1 ( p) S1 ( p) and we conclude by proposition 3. 2 n n p [ 0 , 1 ] , p i 1 is defined by b) Let q 0, \ 1 . The Quantum-Group entropy is defined on A i 1 nN* It’s easy to see that S ( p) p N S qQG ( p) i 1 q i p i 1 / q 1 q q for all p [0,1] N . 1 q q q 1 It’s easy to see that S qQG ( p) S q ( p) S 1 ( p) S q ( p) S 1 ( p) and we conclude by 1 q 1 q 1 q 1 q 1 q q proposition 3. For the proof of the theorem 7, we have need of the following lemmas. Lemma 9. x 1 q is bounded on 2, by M 1 0 ; 1 x 1 q 1 b) For all q 1 , the function x is bounded on 2, by M 2 0 . 1 x 1 q a) For all q 0,1 , the function x Indeed if q 0,1 , then lim x 1 q x 1 x 1 q 1 and if q 1 , then lim 1 x 1 x 1 q 4 0. Lemma 10. If 1 , then p p d Let 0 , N N * , pi , pi' 0,1N . N a) ' i i ( p, p' ) ; ,N i 1 b) If 1 , then p p N 1 d1, N ( p, p' ) ; p p d1, N ( p, p' ) ; N ' i i i 1 N c) If 1 , then ' i i i 1 N d) If 1 , then p i p i' N 1 1 d , N ( p, p' ). i 1 Proof. a) We put I1 i 1,, N, pi' pi and I 2 i 1,, N, pi pi' . p' p' p' p' p' p' For all i I1 , we have: 1 i 1 i i 1 i i 1 i pi pi pi pi pi pi pi pi' p i p i' And by We conclude that for i 1, , N , pi p p p ' i i i pi' . we deduce that all , we deduce that i I2 , we have: . pi pi' p N b) By a) we have, symmetry N pi' i 1 i pi pi' N pi' , and thus a) is proved. p i 1 i pi' 1 , and by Holder’s Inequality, i 1 1 1 1 , we obtain 1 / 1 /(1 ) N p i pi' i 1 N 1 pi pi' i 1 1 / N 11 /(1 ) i 1 1 N pi pi' N 1 . i 1 c) The function x x is continuous on [0,1] , differentiable on ]0,1[ , by the Mean-Value Theorem, we have p i 1, , N , ci pi , pi' , such that pi pi' We deduce that i 1, , N , pi pi' p p N Consequently, i ' i i 1 N d) By c) we have p i p i' i 1 N i 1 N Consequently, N p i p i' i 1 p i p i' N 1 pi pi' . 1/ N p i p i' 1 , and by Holder’s inequality we obtain i 1 N /( 1) 1 i 1 1 pi' ci 1 , d 1, N ( p, p' ) . N p i p i' i 1 N p i p i' 1 p i p i' i 1 i d , N ( p, p ' ) . i 1 Proof of theorem 7. 5 1 1 N 1 1 d , N ( p, p ' ) . max It’s easy to see that S q S q 1 / N ), , (1 / N For q 0,1 , let 0 , there exists M1 d 1, N ( p, p ' ) implies S q ( p) S q ( p' ) S q max 1 /(q ) N pq p' i i i 1 1 N 1q . 1 q q 1 N 1 q 1 , p, p' A R N , 0 , such that for all N N * \ N p iq N N 1 q p i p i' i 1 q p i' i 1 1 N 1 q N 1 q d , N ( p, p ' ) q 1 N 1 q q 1 N 1 q N 1 q d , N ( p, p' ) q 1 N For q 1 , let 0 , there exists qM 2 d 1, N ( p, p ' ) implies S q max M 1 q 1/ 0 , such that for all N S q ( p) S q ( p' ) 1 q i 1 1 N 1 q N q p i p i' i 1 q p iq p i' N N* , p, p' A R N , 1 N 1 q q d , N ( p, p ' ) 1 N 1 q qM 2 Thus, the theorem 6 is proved. The stability of the Incomplete q expectation Definition 11. Let q 0, \ 1 et C a element of E f The q expectation Incomplete p p [0,1] N , N p i i 1 q of C is defined by C q ( p) N i 1 p i q C i 1 where N N * . Theorem 12 a) For q 1 , the Incomplete q expectation of C is stable for all 0 1 . b) For q 0,1 , the Incomplete q expectation of C is stable for all 0 q . c) For q 0,1 , the Incomplete q expectation of C is 1 unstable. Proof We have N C q N , max max C q ( p ) , p p [0,1] N , p i q 1 i 1 N N max p i q C i , where p satisfies p i q 1 i 1 i 1 q max C i 1 max C i i 1,, N i 1,, N 6 for all a) For q 1 , let 0 , there exists q d , N ( p, p ' ) implies 1/ N C q ( p) C q ( p' ) C q N ,max p iq p i' p p N q ' q i q i Ci i 1 N N* , 0 , such that for all C q N ,max p, p' A R N , Ci i 1 max C i 1q i 1,, N N q p i p i' i 1 i 1 qd 1, N ( p, p' ) qd , N ( p, p' ) q p p N ' q i q i b) For q 0,1 , let 0 , there exists 1 / 0 , such that for all N N * , d , N ( p, p ' ) implies p p N C q ( p) C q ( p' ) C q N ,max C q N ,max max C i 1q q p i p i' i 1 d , N ( p, p ' ) ' q i q i i 1 d , N ( p, p' ) q 2 1q 1 c) Let q 0,1 , and 0 , for all 0 , there is N 2W 2 E 2 1 p pi i 1,, N , p i W * 2 N , let p, p' defined by 1/ q if i is odd and p i 0 if i is even p ' p i' Ci i 1 i 1,, N N p p N ' q i q i Ci i 1 p p N ' q i q i p, p' A R N , 1 , p i' W i 1,, N 1/ q pi We have N q 1 1 / q W 1 1 , W W i 1 W p q i i 1 N d1, N ( p, p' ) i 1 q 1 1 / q W 1 1 W W i 1 N W q p i' i 1 1 pi pi' N W 1/ q 2W 11 / q C Ci i 1,, N , Ci 1 if i is odd and C i 0 if i is even W i 1 C q N ,max W1 1 W 0 C q ( p) C q ( p' ) W W And 1 1 / q C q N ,max W 1 1 q 7 W 11 / q 1 . Consequently, the Incomplete q expectation is 1 unstable. The stability of the Rényi entropy Definition 13 1/ q Let q 0, \ 1 . The Rényi Rq entropy is defined nN Rq ( p) ln N i 1 pi q 1 q n p [0,1] , p A on * n i 1 i 1 by for all p [0,1] N . Theorem 14 For q 0,1 , the Rényi entropy R q is stable for all 0 q . Proof. max It’s easy to see that Rq Rq 1 / N ), , (1 / N ln N 1q ln N . 1 q For q 0,1 , let 0 , there exists 1 / 0 , such that for all N N * \ 1,2, p, p' A R N , d , N ( p, p ' ) implies N q N ln p i ln p i' R q ( p) R q ( p' ) i 1 i 1 Rq max ln N q N q pi , By the Mean-Value Theorem, there exists c q , N i 1 N q N ln p i ln p i' i 1 i 1 N Because i 1 p N p iq 1 and ' q i 1 , then i 1 N q N ln p i ln p i' i 1 i 1 ln N By a) of lemma 10, we obtain 1 c q, N q Rq ( p) Rq ( p' ) Rq max q p N ' q i i 1 such that N p N p iq i 1 ' q i i 1 1 c q, N 1 , consequently N p N p iq i 1 i 1 ln N d q , N ( p, p ' ) ' q i 1 c q, N p p N q i ' q i i 1 q d , N ( p, p' ) . References [1] S. Abe, Instability of q expectation value; [2] S. Abe, Stability of Tsallis entrpy and instabilities of Rényi and normalized Tsallis entropies: A basis for q exponential distributions [3] S. Abe, G. Kaniadakis, A. M. Scarfone; Stability of generalized entropies [4] G. Kaniadakis, A. M. Scrafone, Lesche Stability of Entropy. arXiv:cond-mat/0310728v2 [5] B Lesche, Instabilities of Rényi Entropies. Journal of Statistical Physics, Vol. 27, No. 2, 1982 [6] M F. Curado, F. D. Nobre, On the stability of analytic entropic form. Physica A 335 (2004) 94-106 8
© Copyright 2026 Paperzz