Open Access

Some strong limit theorems for arrays of rowwise negatively orthant-dependent random variables

Journal of Inequalities and Applications20112011:93

https://doi.org/10.1186/1029-242X-2011-93

Received: 18 April 2011

Accepted: 26 October 2011

Published: 26 October 2011

Abstract

In this article, the strong limit theorems for arrays of rowwise negatively orthant-dependent random variables are studied. Some sufficient conditions for strong law of large numbers for an array of rowwise negatively orthant-dependent random variables without assumptions of identical distribution and stochastic domination are presented. As an application, the Chung-type strong law of large numbers for arrays of rowwise negatively orthant-dependent random variables is obtained.

MR(2000) Subject Classification: 60F15

Keywords

negatively orthant-dependent sequence array of rowwise negatively orthant-dependent random variables strong law of large numbers

1 Introduction

Let {X n , n ≥ 1} be a sequence of random variables defined on a fixed probability space ( Ω , F , P ) with value in a real space . We say that the sequence {X n , n ≥ 1} satisfies the strong law of large numbers if there exist some increasing sequence {a n , n ≥ 1} and some sequence {c n , n ≥ 1} such that
1 a n i = 1 n ( X i - c i ) 0 a.s. as  n .

Many authors have extended the strong law of large numbers for sequences of random variables to the case of triangular array of random variables and arrays of rowwise random variables. For more details about the strong law of large numbers for triangular array of random variables and arrays of rowwise random variables, one can refer to Gut [1], and so forth. In the case of independence, Hu and Taylor [2] proved the following strong law of large numbers.

Theorem 1.1. Let {X ni : 1 ≤ in, n ≥ 1} be a triangular array of rowwise independent random variables. Let {a n , n ≥ 1} be a sequence of positive real numbers such that 0 < a n ∞. Let g(t) be a positive, even function such that g(|t|)/|t| p is an increasing function of |t| and g(|t|)/|t|p+1is a decreasing function of |t|, respectively, that is,
g ( | t | ) | t | p , g ( | t | ) | t | p + 1 as | t |
for some nonnegative integer p. If p ≥ 2 and
E X n i = 0 , n = 1 i = 1 n E g ( | X n i | ) g ( a n ) < , n = 1 i = 1 n E X n i a n 2 2 k < ,
where k is a positive integer, then
1 a n i = 1 n X n i 0 a.s.
In this article, we will consider the strong law of large numbers for arrays of rowwise negatively associated (NA) random variables. A finite collection of random variables X1, X2, . . . , X n is said to be negatively orthant dependent (NOD) if
P ( X 1 > x 1 , X 2 > x 2 , , X n > x n ) i = 1 n P ( X i > x i )
and
P ( X 1 x 1 , X 2 x 2 , , X n x n ) i = 1 n P ( X i x i )

for all x1, x2, . . . , x n . An infinite sequence {X n , n ≥ 1} is said to be NOD if every finite subcollection is NOD.

An array of random variables {X ni , i ≥ 1, n ≥ 1} is called rowwise NOD random variables if for every n ≥ 1, {X ni , i ≥ 1} is a sequence of NOD random variables.

The concept of NOD sequence was introduced by Joag-Dev and Proschan [3]. Obviously, independent random variables are NOD. Joag-Dev and Proschan [3] pointed out that NA (one can refer to Joag-Dev and Proschan [3]) random variables are NOD. They also presented an example in which X = (X1, X2, X3, X4) possesses NOD, but does not possess NA. So we can see that NOD is weaker than NA. A number of limit theorems for NOD random variables have been established by many authors. We refer to Volodin [4] for the Kolmogorov exponential inequality, Asadian et al. [5] for the Rosental's-type inequality, Amini et al. [6, 7], Klesov et al. [8], and Li et al. [9] for almost sure convergence, Amini and Bozorgnia [10, 11], Kuczmaszewska [12], Taylor et al. [13], Zarei and Jabbari [14] and Wu [15] for complete convergence, and so on.

The main purpose of this article is to study the strong limit theorems for arrays of rowwise NOD random variables. As an application, the Chung-type strong law of large numbers for arrays of rowwise NOD random variables is obtained. We will give some sufficient conditions for strong law of large numbers for an array of rowwise NOD random variables without assumptions of identical distribution and stochastic domination. The results presented in this article are obtained using the truncated method and the Rosental's-type inequality of NOD random variables.

The main results of this article are depending on the following lemmas:

Lemma 1.1 (cf. Bozorgnia et al.[16]). Let random variables X1, X2, . . . , X n be NOD, f1, f2, . . . , f n be all nondecreasing (or all nonincreasing) functions, then random variables f1(X1), f2(X2), . . . , f n (X n ) are NOD.

Lemma 1.2 (cf. Asadian et al.[5]). Let p ≥ 2 and {X n , n ≥ 1} be a sequence of NOD random variables with EX n = 0 and E|X n | p <for every n ≥ 1. Then, there exists a positive constant C = C(p) depending only on p such that for every n ≥ 1
E i = 1 n X i p C i = 1 n E | X i | p + i = 1 n E X i 2 p 2 .

Throughout the article, let I(A) be the indicator function of the set A. C denotes a positive constant which may be different in various places.

2 Main results

In this section, we will give some sufficient conditions for strong law of large numbers for an array of rowwise NOD random variables without assumptions of identical distribution and stochastic domination. Our main results are as follows.

Theorem 2.1. Let {X ni : i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. Let {g n (t), n ≥ 1} be a sequence of positive, even functions such that g n (|t|) is an increasing function of |t| and g n (|t|)/|t| is a decreasing function of |t| for every n ≥ 1, respectively, that is
g n ( | t | ) , g n ( | t | ) | t | as | t | .
If
n = 1 i = 1 n E g n ( | X n i | ) g n ( a n ) < ,
(2.1)
then for any ε > 0,
n = 1 P 1 a n i = 1 n X n i > ε < .
(2.2)
Proof. For fixed n ≥ 1, define
X i ( n ) = - a n I ( X n i < - a n ) + X n i I ( | X n i | a n ) + a n I ( X n i > a n ) , i 1 , T j ( n ) = 1 a n i = 1 j X i ( n ) - E X i ( n ) , j = 1 , 2 , , n .
By Lemma 1.1, we can see that for fixed n ≥ 1, { X i ( n ) , i 1 } is still a sequence of NOD random variables. It is easy to check that for any ε > 0,
1 a n i = 1 n X n i > ε max 1 i n | X n i | > a n 1 a n i = 1 n X i ( n ) > ε ,
which implies that
P 1 a n i = 1 n X n i > ε P max 1 i n | X n i | > a n + P 1 a n i = 1 n X i ( n ) > ε i = 1 n P | X n i | > a n + P T n ( n ) > ε - 1 a n i = 1 n E X i ( n ) .
(2.3)
First, we will show that
1 a n i = 1 n E X i ( n ) 0 as n .
(2.4)
Actually, by conditions g n (|t|) ↑, g n (|t|)/|t| ↓ as |t| ↑ and (2.1), we have that
1 a n i = 1 n E X i ( n ) i = 1 n P | X n i | > a n + 1 a n i = 1 n E | X n i | I ( | X n i | a n ) i = 1 n E g n ( | X n i | ) g n ( a n ) + i = 1 n E g n ( | X n i | ) I ( | X n i | a n ) g n ( a n ) 2 i = 1 n E g n ( | X n i | ) g n ( a n ) 0 as n ,
which implies (2.4). It follows from (2.3) and (2.4) that for n large enough,
P 1 a n i = 1 n X n i > ε i = 1 n P | X n i | > a n + P T n ( n ) > ε 2 .
Hence, to prove (2.2), we only need to show that
n = 1 i = 1 n P | X n i | > a n <
(2.5)
and
n = 1 P T n ( n ) > ε 2 < .
(2.6)
The conditions g n (|t|) ↑ as |t| ↑ and (2.1) yield that
n = 1 i = 1 n P | X n i | > a n n = 1 i = 1 n E g n ( | X n i | ) g n ( a n ) < ,

which implies (2.5).

By Markov's inequality, Lemma 1.2 (for p = 2), g n (|t|) ↑, g n (|t|)/|t| ↓ as |t| ↑ and (2.1), we can get that
n = 1 P T n ( n ) > ε 2 C n = 1 E T n ( n ) 2 C n = 1 1 a n 2 i = 1 n E X i ( n ) 2 C n = 1 i = 1 n P | X n i | > a n + C n = 1 i = 1 n E | X n i | 2 I ( | X n i | a n ) a n 2 C n = 1 i = 1 n | E g n ( | X n i | ) g n ( a n ) + C n = 1 i = 1 n E | X n i | 2 I ( | X n i | a n ) a n 2 C n = 1 i = 1 n E g n ( | X n i | ) g n ( a n ) + C n = 1 i = 1 n E g n ( | X n i | ) I ( | X n i | a n ) g n ( a n ) C n = 1 i = 1 n E g n ( | X n i | ) g n ( a n ) < ,

which implies (2.6). This completes the proof of the theorem. □

Corollary 2.1. Under the conditions of Theorem 2.1,
1 a n i = 1 n X n i 0 a . s .
Theorem 2.2. Let {X ni : i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. Let {g n (t), n ≥ 1} be a sequence of nonnegative, even functions such that g n (|t|) is an increasing function of |t| for every n ≥ 1. Assume that there exists a constant δ > 0 such that g n (t) ≥ δt for 0 < t ≤ 1. If
n = 1 i = 1 n E g n X n i a n < ,
(2.7)

then for any ε > 0, (2.2) holds true.

Proof. We use the same notations as that in Theorem 2.1. The proof is similar to that of Theorem 2.1.

First, we will show that (2.4) holds true. In fact, by the conditions g n (t) ≥ δt for 0 < t ≤ 1 and (2.7), we have that
1 a n i = 1 n E X i ( n ) i = 1 n P | X n i | > a n + i = 1 n E | X n i | a n I ( | X n i | a n ) 1 δ i = 1 n E g n X n i a n + 1 δ i = 1 n E g n X n i a n I ( | X n i | a n ) 2 δ i = 1 n E g n X n i a n 0 as n ,

which implies (2.4).

According to the proof of Theorem 2.1, we only need to prove that (2.5) and (2.6) hold true.

When |X ni | > a n > 0, we have g n X n i a n g n ( 1 ) δ , which yields that
P ( | X n i | > a n ) = E I ( | X n i | > a n ) 1 δ E g n X n i a n .
Hence,
n = 1 i = 1 n P | X n i | > a n 1 δ n = 1 i = 1 n E g n X n i a n < ,

which implies (2.5).

By Markov's inequality, Lemma 1.2 (for p = 2), g n (t) ≥ δt for 0 < t ≤ 1 and (2.7), we can get that
n = 1 P T n ( n ) > ε 2 C n = 1 i = 1 n P | X n i | > a n + C n = 1 i = 1 n E | X n i | 2 I ( | X n i | a n ) a n 2 C + C n = 1 i = 1 n E | X n i | I ( | X n i | a n ) a n C + C n = 1 i = 1 n E g n X n i a n I ( | X n i | a n ) C + C n = 1 i = 1 n E g n X n i a n < ,

which implies (2.6). This completes the proof of the theorem. □

Corollary 2.2. Let {X ni , i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. If there exists a constant β (0, 1] such that
n = 1 i = 1 n E | X n i | β | a n | β + | X n i | β < ,

then (2.2) holds true.

Proof. In Theorem 2.2, we take
g n ( t ) | t | β 1 + | t | β , 0 < β 1 , n 1 .
It is easy to check that {g n (t), n ≥ 1} is a sequence of nonnegative, even functions such that g n (|t|) is an increasing function of |t| for every n ≥ 1. And
g n ( t ) 1 2 t β 1 2 t , 0 < t 1 , 0 < β 1 .

Therefore, by Theorem 2.2, we can easily get (2.2). □

Corollary 2.3. Under the conditions of Theorem 2.2 or Corollary 2.2,
1 a n i = 1 n X n i 0 a . s .

Theorem 2.3. Let {X ni : i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. EX ni = 0, i ≥ 1, n ≥ 1. Let {g n (x), n ≥ 1} be a sequence of nonnegative, even functions. Assume that there exist β (1, 2] and δ > 0 such that g n (x) ≥ δx β for 0 < x ≤ 1 and there exists a δ > 0 such that g n (x) ≥ δx for x > 1. If (2.7) satisfies, then for any ε > 0, (2.2) holds true.

Proof. We use the same notations as that in Theorem 2.1. The proof is similar to that of Theorem 2.1.

First, we will show that (2.4) holds true. Actually, by the conditions EX ni = 0, g n (x) ≥ δx for x > 1 and (2.7), we have that
1 a n i = 1 n E X i ( n ) i = 1 n P | X n i | > a n + 1 a n i = 1 n E X n i I ( | X n i | > a n ) 2 i = 1 n E | X n i | a n I ( | X n i | > a n ) 2 δ i = 1 n E g n X n i a n I ( | X n i | > a n ) 2 δ i = 1 n E g n X n i a n 0 as n ,

which implies (2.4). Hence, to prove (2.2), we only need to show that (2.5) and (2.6) hold true.

The conditions g n (x) ≥ δx for x > 1 and (2.1) yield that
n = 1 i = 1 n P | X n i | > a n = n = 1 i = 1 n E I | X n i | > a n n = 1 i = 1 n E | X n i | a n I | X n i | > a n 1 δ n = 1 i = 1 n E g n X n i a n I | X n i | > a n 1 δ n = 1 i = 1 n E g n X n i a n < ,

which implies (2.5).

By Markov's inequality, Lemma 1.2 (for p = 2), g n (x) ≥ δx β for 1 < β ≤ 2, 0 < x ≤ 1 and (2.7), we can get that
n = 1 P T n ( n ) > ε 2 C n = 1 i = 1 n P | X n i | > a n + C n = 1 i = 1 n E | X n i | 2 I ( | X n i | a n ) a n 2 C + C n = 1 i = 1 n E | X n i | β I ( | X n i | a n ) a n β C + C n = 1 i = 1 n E g n X n i a n I ( | X n i | a n ) C + C n = 1 i = 1 n E g n X n i a n < ,

which implies (2.6). This completes the proof of the theorem. □

Corollary 2.4. Let {X ni , i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. EX ni = 0, i ≥ 1, n ≥ 1. If there exists a constant β (1, 2] such that
n = 1 i = 1 n E | X n i | β a n | X n i | β - 1 + a n β < ,

then (2.2) holds true.

Proof. In Theorem 2.3, we take
g n ( x ) x β 1 + x β - 1 , 1 < β 2 , n 1 .
It is easy to check that {g n (x), n ≥ 1} is a sequence of nonnegative, even functions satisfying
g n ( x ) 1 2 x β , 0 < x 1 , 1 < β 2 a n d g n ( x ) 1 2 x , x > 1 .

Therefore, by Theorem 2.3, we can easily get (2.2). □

Furthermore, by Corollaries 2.2 and 2.4, we can get the following important Chungtype strong law of large numbers for arrays of rowwise NOD random variables.

Corollary 2.5. Let {X ni , i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. If there exists some β (0, 2] such that
n = 1 i = 1 n E X n i β a n β < ,
(2.8)

and EX ni = 0, i ≥ 1, n ≥ 1 if β (1, 2], then (2.2) holds true and 1 a n i = 1 n X n i 0 a . s . .

For β ≥ 2, we have the following result.

Theorem 2.4. Let {X ni : i ≥ 1, n ≥ 1} be an array of rowwise NOD random variables and {a n , n ≥ 1} be a sequence of positive real numbers. Let {g n (x), n ≥ 1} be a sequence of nonnegative, even functions. Assume that there exists some β ≥ 2 such that g n (x) ≥ δx β for x > 0. If
n = 1 i = 1 n E g n X n i a n 1 β < ,
(2.8a)

then for any ε > 0, (2.2) holds true.

Proof. We use the same notations as that in Theorem 2.1. The proof is similar to that of Theorem 2.1. It is easily seen that (2.8) implies that
n = 1 i = 1 n E g n X n i M a n <
(2.9)
and
n = 1 i = 1 n E g n X n i M a n 2 β < .
(2.10)
First, we will show that (2.4) holds true. In fact, by Hölder's inequality, g n (x) ≥ δx β for x > 0, (2.8) and (2.9), we have that
1 a n i = 1 n E X i ( n ) i = 1 n P X n i > a n + i = 1 n E X n i a n I ( X n i a n ) i = 1 n E X n i β a n β I ( X n i > a n ) + i = 1 n E X n i β a n β I ( X n i a n ) 1 β C i = 1 n E g n X n i a n + C i = 1 n E g n X n i a n I ( X n i a n ) 1 β C i = 1 n E g n X n i a n + C i = 1 n E g n X n i a n 1 β 0 as n ,

which implies (2.4). To prove (2.2), we only need to show that (2.5) and (2.6) hold true.

By the condition g n (x) ≥ δx β for x > 0 again and (2.9), we have
n = 1 i = 1 n P X n i > a n = n = 1 i = 1 n E I ( X n i > a n ) n = 1 i = 1 n E X n i β a n β I ( X n i > a n ) 1 δ n = 1 i = 1 n E g n X n i a n < ,

which implies (2.5).

By Markov's inequality, Lemma 1.2 (for p = 2), g n (x) ≥ δx β for x > 0 and (2.10), we can get that
n = 1 P T n ( n ) > ε 2 C n = 1 i = 1 n P X n i > a n + C n = 1 i = 1 n E X n i 2 I ( X n i a n ) a n 2 C + C n = 1 i = 1 n E X n i β a n β I ( X n i a n ) 2 β C + C n = 1 i = 1 n E g n X n i a n I ( X n i a n ) 2 β C + C n = 1 i = 1 n E g n X n i a n 2 β < ,

which implies (2.6). This completes the proof of the theorem. □

Declarations

Acknowledgements

The author was most grateful to the Editor Andrei Volodin and anonymous referees for careful reading of the manuscript and valuable suggestions which helped in improving an earlier version of this article.

Supported by the National Natural Science Foundation of China (11171001, 71071002) and the Academic Innovation Team of Anhui University (KJTD001B).

Authors’ Affiliations

(1)
School of Mathematical Science, Anhui University

References

  1. Gut A: Complete convergence for arrays. Periodica Math Hungarica 1992,25(1):51–75. 10.1007/BF02454383MathSciNetView ArticleGoogle Scholar
  2. Hu TC, Taylor RL: On the strong law for arrays and for the bootstrap mean and variance. Int J Math Math Sci 1997,20(2):375–382. 10.1155/S0161171297000483MathSciNetView ArticleGoogle Scholar
  3. Joag-Dev K, Proschan F: Negative association of random variables with applications. Ann Stat 1983,11(1):286–295. 10.1214/aos/1176346079MathSciNetView ArticleGoogle Scholar
  4. Volodin A: On the Kolmogorov exponential inequality for negatively dependent random variables. Pakistan J Stat 2002, 18: 249–254.MathSciNetGoogle Scholar
  5. Asadian N, Fakoor V, Bozorgnia A: Rosental's type inequalities for negatively orthant dependent random variables. J Iranian Stat Soc 2006,5(1–2):66–75.Google Scholar
  6. Amini M, Azarnoosh HA, Bozorgnia A: The strong law of large numbers for negatively dependent generalized Gaussian random variables. Stochast Anal Appl 2004, 22: 893–901.MathSciNetView ArticleGoogle Scholar
  7. Amini M, Zarei H, Bozorgnia A: Some strong limit theorems of weighted sums for negatively dependent generalized Gaussian random variables. Stat Probab Lett 2007, 77: 1106–1110. 10.1016/j.spl.2007.01.015MathSciNetView ArticleGoogle Scholar
  8. Klesov O, Rosalsky A, Volodin A: On the almost sure growth rate of sums of lower negatively dependent nonnegative random variables. Stat Probab Lett 2005, 71: 193–202. 10.1016/j.spl.2004.10.027MathSciNetView ArticleGoogle Scholar
  9. Li D, Rosalsky A, Volodin A: On the strong law of large numbers for sequences of pairwise negative quadrant dependent random variables. Bull Inst Math Acad Sin (New Ser) 2006,1(2):281–305.MathSciNetGoogle Scholar
  10. Amini M, Bozorgnia A: Negatively dependent bounded random variable probability inequalities and the strong law of large numbers. J Appl Math Stochast Anal 2000, 13: 261–267. 10.1155/S104895330000023XMathSciNetView ArticleGoogle Scholar
  11. Amini M, Bozorgnia A: Complete convergence for negatively dependent random variables. J Appl Math Stochast Anal 2003, 16: 121–126. 10.1155/S104895330300008XView ArticleGoogle Scholar
  12. Kuczmaszewska A: On some conditions for complete convergence for arrays of rowwise negatively dependent random variables. Stochast Anal Appl 2006, 24: 1083–1095. 10.1080/07362990600958754MathSciNetView ArticleGoogle Scholar
  13. Taylor RL, Patterson RF, Bozorgnia A: A strong law of large numbers for arrays of rowwise negatively dependent random variables. Stochast Anal Appl 2002, 20: 643–656. 10.1081/SAP-120004118MathSciNetView ArticleGoogle Scholar
  14. Zarei H, Jabbari H: Complete convergence of weighted sums under negative dependence. Stat Pap 2009.Google Scholar
  15. Wu QY: Complete convergence for negatively dependent sequences of random variables. J Inequal Appl 2010, 2010: 10. Article ID 507293Google Scholar
  16. Bozorgnia A, Patterson RF, Taylor RL: Limit theorems for dependent random variables. World Congress Nonlinear Analysts'92 1996, 1639–1650.Google Scholar

Copyright

© Shen; licensee Springer. 2011

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.