Aug 15, 2016 - Taylor and Inoue had proved the strong law of large numbers for independent ... space, which is both the extension of the result in [1] for ...
Advances in Pure Mathematics, 2016, 6, 583-592 Published Online August 2016 in SciRes. http://www.scirp.org/journal/apm http://dx.doi.org/10.4236/apm.2016.69047
Strong Laws of Large Numbers for Fuzzy Set-Valued Random Variables in Gα Space Lamei Shen, Li Guan College of Applied Sciences, Beijing University of Technology, Beijing, China Received 7 July 2016; accepted 12 August 2016; published 15 August 2016 Copyright © 2016 by authors and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY). http://creativecommons.org/licenses/by/4.0/
Abstract In this paper, we shall present the strong laws of large numbers for fuzzy set-valued random variables in the sense of dH∞ . The results are based on the result of single-valued random variables obtained by Taylor [1] and set-valued random variables obtained by Li Guan [2].
Keywords Laws of Large Numbers, Fuzzy Set-Valued Random Variable, Hausdorff Metric
1. Introduction With the development of set-valued stochastic theory, it has become a new branch of probability theory. And limits theory is one of the most important theories in probability and statistics. Many scholars have done a lot of research in this aspect. For example, Artstein and Vitale in [3] had proved the strong law of large numbers for independent and identically distributed random variables by embedding theory. Hiai in [4] had extended it to separable Banach space. Taylor and Inoue had proved the strong law of large numbers for independent random variable in the Banach space in [5]. Many other scholars also had done lots of works in the laws of large numbers for set-valued random variables. In [2], Li proved the strong laws of large numbers for set-valued random variables in Gα space in the sense of dH metric. As we know, the fuzzy set is an extension of the set. And the concept of fuzzy set-valued random variables is a natural generalization of that of set-valued random variables, so it is necessary to discuss convergence theorems of fuzzy set-valued random sequence. The limits of theories for fuzzy set-valued random sequences are also been discussed by many researchers. Colubi et al. [6], Feng [7] and Molchanov [8] proved the strong laws of large numbers for fuzzy set-valued random variables; Puri and Ralescu [9], Li and Ogura [10] proved converHow to cite this paper: Shen, L.M. and Guan, L. (2016) Strong Laws of Large Numbers for Fuzzy Set-Valued Random Variables in Gα Space. Advances in Pure Mathematics, 6, 583-592. http://dx.doi.org/10.4236/apm.2016.69047
L. M. Shen, L. Guan
gence theorems for fuzzy set-valued martingales. Li and Ogura [11] proved the SLLN of [12] in the sense of d H∞ by using the “sandwich” method. Guan and Li [13] proved the SLLN for weighted sums of fuzzy setvalued random variables in the sense of d H∞ which used the same method. In this paper, what we concerned are the convergence theorems of fuzzy set-valued sequence in Gα space in the sense of d H∞ . The purpose of this paper is to prove the strong laws of large numbers for fuzzy set-valued random variables in Gα space, which is both the extension of the result in [1] for single-valued random sequence and also the extension in [2] for set-valued random sequence. This paper is organized as follows. In Section 2, we shall briefly introduce some concepts and basic results of set-valued and fuzzy set-valued random variables. In Section 3, I shall prove the strong laws of large numbers for fuzzy set-valued random variables in Gα space, which is in the sense of Hausdorff metric d H∞ .
2. Preliminaries on Set-Valued Random Variables Throughout this paper, we assume that ( Ω, , µ ) is a complete probability space, ( X, ⋅ ) is a real separable Banach space, K ( X ) is the family of all nonempty closed subsets of X , and K b ( X ) ( K k ( X ) ) is the family
of all non-empty bounded closed(compact) subsets of X , and K kc ( X ) is the family of all non-empty compact convex subsets of X . Let A and B be two nonempty subsets of X and let λ ∈ , the set of all real numbers. We define addition and scalar multiplication by
A + B = {a + b : a ∈ A, b ∈ B}
{λ a : a ∈ A}
= λA The Hausdorff metric on K ( X ) is defined by
{
d H ( A, B )= max sup inf a − b , sup inf a − b a∈A b∈B
b∈B a∈A
}
for A, B ∈ K ( X ) . For an A in K ( X ) , let A K = d H ({0} , A ) . The metric space ( K b ( X ) , d H ) is complete, and K bc ( X ) is a closed subset of
(Kb (X ) , dH )
(cf. [14],
Theorems 1.1.2 and 1.1.3). For more general hyperspaces, more topological properties of hyperspaces, readers may refer to the books [15] and [14]. For each A ∈ K ( X ) , define the support function by
(
)
= s x* , A sup x* , a , x* ∈ X* , *
where X is the dual space of X . Let S* denote the unit sphere of X* , C S* as v C = sup x∗ ∈S∗ .
( )
a∈A
the all continuous functions of S* , and the norm is defined
The following is the equivalent definition of Hausdorff metric. For each A, B ∈ K bc ( X ) ,
{(
) (
)
}
d H ( A, B ) = sup s x* , A − s x* , A : x* ∈ S * .
A set-valued mapping F : Ω → K ( X ) is called a set-valued random variable (or a random set, or a multifunction) if, for each open subset O of X , F −1 ( O= ) {ω ∈ Ω : F (ω ) ∩ O ≠ ∅} ∈ . For each set-valued random variable F, the expectation of F, denoted by E [ F ] , is defined by
= E [F ] where
∫Ω fdµ
{
{∫
Ω
}
fdµ : f ∈ S F ,
is the usual Bochner integral in L1 [ Ω, X ] , the family of integrable X -valued random variables,
}
and S F = f ∈ L1 [ Ω; X ] : f (ω ) ∈ F (ω ) , a.e. ( µ ) .
584
L. M. Shen, L. Guan
Let Fk ( X ) denote the family of all functions v : X → [ 0,1] which satisfy the following conditions: 1) The level set v1= { x ∈ X : v ( x )= 1} ≠ ∅ . 2) Each v is upper semicontinuous, i.e. for each α ∈ ( 0,1] , the α level set vα = { x ∈ X : v ( x ) ≥ α } is a closed subset of X . cl { x ∈ X : v ( x ) > 0} is compact. 3) The support set v0+ = A function v in Fk ( X ) is called convex if it satisfies
v ( λ x + (1 − λ ) y ) ≥ min {v ( x ) , v ( y )} , for any x, y ∈ X, λ ∈ ( 0,1] .
Let Fkc ( X ) be the subset of all convex fuzzy sets in Fk ( X ) . It is known that v is convex in the above sense if and only if, for any α ∈ ( 0,1] , the level set vα is a convex subset of X (cf. Theorem 3.2.1 of [16]). For any v ∈ Fk ( X ) , the closed convex hull cov ∈ Fkc ( X ) of v is defined by the relation
( cov )
α
= covα for all α ∈ ( 0,1] .
For any two fuzzy sets ν ,ν , define 1
2
(ν
1
+ν 2
) ( x=)
{
}
sup α ∈ ( 0,1] : x ∈ν α1 + ν α2 ,
for any x ∈ X. Similarly for a fuzzy set ν and a real number λ , define
( λν )( x ) =sup {α ∈ ( 0,1] : x ∈ λν α } , for any x ∈ X. The following two metrics in Fk ( X ) which are extensions of the Hausdorff metric dH are often used (cf. [17] and [18], or [14]): for v1 , v 2 ∈ Fk ( X ) ,
(
)
(
)
(
)
d H∞ v1 , v 2 = sup d H vα1 , vα2 , α ∈( 0,1] 1
(
)
d H1 v1 , v 2 = ∫0 d H vα1 , vα2 dα . ∞ = v F : d= Denote supα >0 vα H ( v, I 0 )
(
)
K
, where I 0 is the fuzzy set taking value one at 0 and zero for all
x ≠ 0 . The space Fk ( X ) , d H∞ is a complete metric space (cf. [18], or [14]: Theorem 5.1.6) but not separable (cf. [17], or [14]: Remark 5.1.7). It is well known that vα = β 0 , there exists a finite 0 = α 0 < α1 < < α M = 1 , such that
(
)
d H ν α j ,ν α j −1 + ≤ ε , for all 1, , M .
Then for α j −1 < α < α j ,
1 n 1 n 1 n d H ∑ν αk ,ν α ≤ d H ∑ν αk j ,ν α j −1+ + d H ∑ν αk j −1+ ,ν α j = n k 1= n k 1 = n k 1 n n 1 1 ≤ d H ∑ν αk j ,ν α j + d H ∑ν αk j −1+ ,ν α j −1+ + 2d H ν α j ,ν α j −1+ n n = k 1= k1
(
)
Consequently,
1 n 1 n 1 n sup d H ∑ν αk ,ν α ≤ max d H ∑ν αk j ,ν α j + max d H ∑ν αk j −1+ ,ν α j −1+ + 2ε . α∈( 0,1] = n k 1= 1≤ j ≤ M n k 1 = 1≤ j ≤ M n k 1 Since the first two terms on the right hand converge to 0 in probability one, we have
1 n lim sup sup d H ∑ν αk ,ν α ≤ 2ε , n→∞ α ∈( 0,1] n k =1 but ε is arbitrary and the result follows. Theorem 3.6. Let X be a Banach space which satisfies the condition of Gα , let X n : n ≥ 1 dent fuzzy set-valued random variables in Fk ( X ) , such that E X n = I 0 for any n. If
{
∑ E φ0 ( ∞
X
j =1
j F
}
be indepen-
) < +∞,
where φ0 ( t ) = t1+α for 0 ≤ t ≤ 1 and φ0 ( t ) = t for t ≥ 1, then
∞
∑X j
converges with probability 1 in the sense
j =1
of d H∞ . Proof. Define j = U j X= I j , W j X jI
{X
F
}
≤1
{
}
j Note that X= W j + U j for each j, and both W j : j ≥ 1
fuzzy set-valued random variables. When
X
j F
{X
and
> 1 , we have W
j F
}
>1
{U
j F
. j
}
: j ≥1
= X
j F
are independent sequence of
, and φ0
( X )= j
F
X
j F
. Then,
for any m, n m m m E ∑W j ≤ ∑ E W j = ∑ E φ0 X j F . F = =j n jn j n= F
(
∞
And from
∑ E φ0 j =1
X
j
< ∞ , we know that E F
m
∑W j j =1
587
)
: m ≥ 1 is a Cauchy sequence. So, we have F
L. M. Shen, L. Guan
E
m
∑W j j =1
converges as m → ∞. F
Since convergence in the mean implied convergence in probability, Ito and Nisios result in [9] for independent random elements (cf. Section 4.5) provides that m
∑W j j =1
converges in probability 1 . F
So, for any n, m ≥ 1, m > n, by triangle inequality we have m n n m n j j j j j ∞ d H∞= ∑W , ∑W dH ∑W , ∑W + ∑ W = j 1 =j 1 = j 1 =j 1 =j n +1 n n m ≤ d H∞ ∑ W j , ∑ W j + d H∞ I 0 ∑ W j j= 1 j= 1 j= n +1 m = d H∞ I 0 , ∑ W j j= n +1
=
m
∑Wj
j = n +1
→ 0, a.e. as n, m → ∞. F
n It means ∑ W j : n ≥ 1 is a Cauchy sequence in the sense of d H∞ . By the completeness of j =1 n
we have
∑W j j =1
(F (X) , d ) , k
∞ H
converges almost everywhere in the sense of d H∞ . n
Next we shall prove that
∑U j j =1
converges in the sense of d H∞ . Firstly, we assume that
{U } j
fuzzy set-valued random variables. Then by the equivalent definition of Hausdorff metric, we have 1+α m m = E sup ∑ U βj E ∑U j j n= β ∈( 0,1] j n = F
1+α
K
m = E sup d H1+α ∑ U βj , {0} j =n β ∈( 0,1] = E sup sup β ∈( 0,1] x∗∈S ∗
m s x∗ , ∑U βj j =n
1+α
For any fixed n, m, there exists a sequence xk∗ ∈ S ∗ , such that m m lim s xk∗ , ∑ U βj = sup s x∗ , ∑ U βj . k →∞ = j n= j n x∗∈S ∗
That means there exist a sequence xk∗ ∈ S ∗ , such that 1+α 1+α m ∗ m j j = E sup lim s xk , ∑ U β . E ∑U β ∈( 0,1] k →∞ j n= = j n F
Then by Cr inequality, dominated convergence theorem and Lemma 3.2, we have
588
are all convex
E
L. M. Shen, L. Guan
1+α
1+α
m = E sup lim s xk∗ , ∑ U βj ∑U k →∞ j n= j n F β ∈( 0,1] m
j
1+α
m ≤ E sup lim ∑ s xk∗ , U βj β ∈( 0,1] k →∞ j = n
(
)
1+α
m ≤ E lim sup ∑ s xk∗ , U βj k →∞ β ∈( 0,1] j = n
(
)
1+α
m ≤ lim E sup ∑ s xk∗ , U βj − E s xk∗ , U βj + E s xk∗ , U βj k →∞ β ∈ 0,1 ( ] j =n
(
)
(
)
(
)
1+α
m = lim E lim ∑ s xk∗ , U βji − E s xk∗ , U βji + E s xk∗ , U βji k →∞ i →∞ j =n
(
)
(
)
(
)
1+α
m = lim lim E ∑ s xk∗ , U βji − E s xk∗ , U βji + E s xk∗ , U βji k →∞ i →∞ j =n
(
)
(
)
(
)
1+α
m m ≤ lim lim E ∑ s xk∗ , U βji − E s xk∗ , U βji + ∑ E s xk∗ , U βji k →∞ i →∞ = j n=j n
(
)
(
)
(
)
1+α 1+α m m ∗ j ∗ j ∗ j ≤ lim lim 2 E ∑ s xk , U βi − E s xk , U βi + ∑ E s xk , U β k →∞ i →∞ = j n j n=
(
1+α
)
(
)
(
)
1+α m 1+α m ∗ j ∗ j ∗ j E s xk , U βi ≤ lim lim 2 A∑ E s xk , U βi − E s xk , U βi + ∑ k →∞ i →∞ = j n j n=
(
1+α
)
(
)
(
)
1+α m m 1+α 1+α m 1+α j 1+α j j ∗ ∗ ∗ ≤ lim lim 2 2 A∑ E s xk , U βi + 2 A E s x , U + E s x , W ∑ ∑ k βi k βi k →∞ i →∞ = j n j n= = j n
(
1+α
)
(
)
(
)
1+α m m 1+α 1+α m 1+α ∗ j 1+α ∗ j ∗ j 2 , ≤ lim lim 2 2 A∑ E s xk , U βi + A E s x U + E s x , W ∑ ∑ k βi k βi k →∞ i →∞ = = j n j n= j n
(
1+α
)
(
)
1+α m 1+α m 2 +α j j ∗ ∗ lim lim 2 2 A∑ E s xk , U βi E s x , W + ∑ k βi k →∞ i →∞ = j n= j n
(
1+α
)
(
)
1+α m 1+α m 2 +α j j ∗ ∗ E sup s x , W ≤ lim 2 2 A∑ E sup s x , U βi + ∑ βi i →∞ j n= = x*∈S * j n x*∈S *
(
1+α
(
)
1+α
m m 1+α + E W j = lim 2 22 +α A∑ E U βji ∑ K i →∞ j n βi = j n= m m 1+α + E W j = 2 2 2 +α A ∑ E U j ∑ F j n j n= = 1+α
m m ≤ 2 22 +α A∑ E φ0 X j + ∑ E φ0 F = j n= j n 1+α
(
)
for each n and m.
589
K
1+α
1+α
F
(X ) j
F
1+α
)
(
)
L. M. Shen, L. Guan
Then, we know E
m
∑U j =1
1+α j F
is a Cauchy sequence. Hence, ∞
Thus by the similar way as above to prove
∑W j
{E ∑
m
Uj
j =1
F
} is a Cauchy sequence.
converges with probability 1 in the sense of d H∞ . We also
j =1
can prove that ∞
∑U j
converges
j =1
with probability 1 in the sense of d H∞ . In fact, for each n ≤ m , n n m n j m j ∞ j j j d H∞= ∑U , ∑U dH ∑U , ∑U + ∑ U = j 1 =j 1 = j 1 =j 1 =j n +1
≤
m
∑U j
j= n +1
F
→ 0, a.e.
as n, m → ∞.
So, we can prove that ∞
∑X j
converges
j =1
{ }
n
with probability 1 in the sense of d H∞ . If U j
∑ coU j
are not convex, we can prove
converges with proba-
j =1
bility 1 in the sense of d H∞ as above, and by the Lemma 3.5, we can prove that
n
∑U j
converges with proba-
j =1
bility 1 in the sense of d H∞ . Then the result was proved. From Theorem 3.6, we can easily obtain the following corollary. Corollary 3.7. Let X be a separable Banach space which is Gα for some 0 < α ≤ 1 . Let X n : n ≥ 1 be a sequence of independent fuzzy set-valued random variables in Fk ( X ) , such that E X n = I 0 for each n. If
{
1, 2, , are continuous and such that φn : + → + , n =
α n ⊂ + the convergence of ∞
∑
E φn
φn ( t ) t1+α and are non-decreasing, then for each t φn ( t )
( X ) n
φn (α n )
n =1
F
implies that ∞
Xn
∑α n =1
n
∞ H
converges with probability one in the sense of d . Proof. Let Xj Xj and W j Uj = I j I =
αj
If
Xn
F
{X
F
≤α j
> α n , by the non-decreasing property of
}
αj
φn ( t ) , we have t
(
n φn (α n ) φn X ≤ αn Xn
F
590
}
F
).
{X
j F
>α j
}
.
L. M. Shen, L. Guan
That is
Xn
αn Xn
If
F
F
≤
(
φn X n
F
φn (α n )
).
(4.1)
t1+α , we have φn ( t )
≤ α n , by the non-decreasing property of Xn
(
1+α F
φn X n
F
)
≤
α n1+α . φn (α n )
That is
Xn
α
1+α F
1+α n
≤
(
φn X n
F
φn (α n )
).
(4.2) ∞
∞
Then as the similar proof of Theorem 3.6, we can prove both
∑U j j =1
and
∑W j
converges with probability
j =1
one in the sense of d H∞ , and the result was obtained.
Acknowledgements We thank the Editor and the referee for their comments. Research of Li Guan is funded by the NSFC (11301015, 11571024, 11401016).
References [1]
Taylor, R.L. (1978) Lecture Notes in Mathematics. Springer-Verlag, 672.
[2]
Li, G. (2015) A Strong Law of Large Numbers for Set-Valued Random Variables in Gα Space. Journal of Applied Mathematics and Physics, 3, 797-801. http://dx.doi.org/10.4236/jamp.2015.37097
[3]
Artstein, Z. and Vitale, R.A. (1975) A Strong Law of Large Numbers for Random Compact Sets. Annals of Probability, 3, 879-882. http://dx.doi.org/10.1214/aop/1176996275
[4]
Hiai, F. (1984) Strong Laws of Large Numbers for Multivalued Random Variables, Multifunctions and Integrands. In: Salinetti, G., Ed., Lecture Notes in Mathematics, Vol. 1091, Springer, Berlin, 160-172.
[5]
Taylor, R.L. and Inoue, H. (1985) A Strong Law of Large Numbers for Random Sets in Banach Spaces. Bulletin of the Institute of Mathematics Academia Sinica, 13, 403-409.
[6]
Colubi, A., López-Díaz, M., Domnguez-Menchero, J.S. and Gil, M.A. (1999) A Generalized Strong Law of Large Numbers. Probability Theory and Related Fields, 114, 401-417. http://dx.doi.org/10.1007/s004400050229
[7]
Feng, Y. (2004) Strong Law of Large Numbers for Stationary Sequences of Random Upper Semicontinuous Functions. Stochastic Analysis and Applications, 22, 1067-1083. http://dx.doi.org/10.1081/SAP-120037631
[8]
Molchanov, I. (1999) On Strong Laws of Large Numbers for Random Upper Semicontinuous Functions. Journal of Mathematical Analysis and Applications, 235, 249-355. http://dx.doi.org/10.1006/jmaa.1999.6403
[9]
Puri, M.L. and Ralescu, D.A. (1991) Convergence Theorem for Fuzzy Martingales. Journal of Mathematical Analysis and Applications, 160, 107-121. http://dx.doi.org/10.1016/0022-247X(91)90293-9
[10] Li, S. and Ogura, Y. (2003) A Convergence Theorem of Fuzzy Valued Martingale in the Extended Hausdorff Metric H∞. Fuzzy Sets and Systems, 135, 391-399. http://dx.doi.org/10.1016/S0165-0114(02)00145-8 [11] Li, S. and Ogura, Y. (2003) Strong Laws of Numbers for Independent Fuzzy Set-Valued Random Variables. Fuzzy Sets and Systems, 157, 2569-2578. http://dx.doi.org/10.1016/j.fss.2003.06.011 [12] Inoue, H. (1991) A Strong Law of Large Numbers for Fuzzy Random Sets. Fuzzy Sets and Systems, 41, 285-291. http://dx.doi.org/10.1016/0165-0114(91)90132-A [13] Guan, L. and Li, S. (2004) Laws of Large Numbers for Weighted Sums of Fuzzy Set-Valued Random Variables. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 12, 811-825.
591
L. M. Shen, L. Guan
http://dx.doi.org/10.1142/S0218488504003223
[14] Li, S., Ogura, Y. and Kreinovich, V. (2002) Limit Theorems and Applications of Set-Valued and Fuzzy Set-Valued Random Variables. Kluwer Academic Publishers, Dordrecht. http://dx.doi.org/10.1007/978-94-015-9932-0 [15] Beer, G. (1993) Topologies on Closed and Closed Convex Sets. Mathematics and Its Applications. Kluwer Academic Publishers, Dordrecht, Holland. http://dx.doi.org/10.1007/978-94-015-8149-3 [16] Chen, Y. (1984) Fuzzy Systems and Mathematics. Huazhong Institute Press of Science and Technology, Wuhan. (In Chinese) [17] Klement, E.P., Puri, L.M. and Ralescu, D.A. (1986) Limit Theorems for Fuzzy Random Variables. Proceedings of the Royal Society of London A, 407, 171-182. http://dx.doi.org/10.1098/rspa.1986.0091 [18] Puri, M.L. and Ralescu, D.A. (1986) Fuzzy Random Variables. Journal of Mathematical Analysis and Applications, 114, 406-422. http://dx.doi.org/10.1016/0022-247X(86)90093-4 [19] Li, S. and Ogura, Y. (1996) Fuzzy Random Variables, Conditional Expectations and Fuzzy Martingales. Journal of Fuzzy Mathematics, 4, 905-927.
Submit or recommend next manuscript to SCIRP and we will provide best service for you: Accepting pre-submission inquiries through Email, Facebook, LinkedIn, Twitter, etc. A wide selection of journals (inclusive of 9 subjects, more than 200 journals) Providing 24-hour high-quality service User-friendly online submission system Fair and swift peer-review system Efficient typesetting and proofreading procedure Display of the result of downloads and visits, as well as the number of cited articles Maximum dissemination of your research work
Submit your manuscript at: http://papersubmission.scirp.org/
592