数学物理学报  2016, Vol. 36 Issue (3): 448-455   PDF (290 KB)    
扩展功能
加入收藏夹
复制引文信息
加入引用管理器
Email Alert
RSS
本文作者相关文章
李炜
END序列加权和的完全收敛性
李炜     
仲恺农业工程学院计算科学学院 广州 510225
摘要: 该文把Sung[1]的一个关于同分布NA随机变量序列加权和最大值完全收敛性结果部分推广到了END随机变量序列情形. 由于已有文献所用的工具是部分和最大值指数型不等式或部分和最大值Rosenthal型矩不等式,而对于END而言相应的不等式是否成立至今未知,因此原有的证明方法已失效. 该文将应用END随机变量序列部分和的Rosenthal型矩不等式,结合三段截尾法,获得了理想的结果. 该文的证明方法不同于已有的结果.
关键词: END序列     加权和     完全收敛性    
Complete Convergence for Weighted Sums Under END Setup
Li Wei     
College of Computation Science, Zhongkai University of Agriculture and Engineering, Guangzhou 510225
Abstract: The complete convergence for the weighted sums of identically distributed NA random variables in [1] is improved and extended under the END setup. The main tool in [1] is the maximum Rosenthal's type moemnt inequality, but it is unknown whether the kind of moment inequality holds or not for END, so our method is different from [1].
Key words: END sequence     Weighted sum     Complete convergence    
1 引言和结果

在统计中,很多统计量,如最小二乘估计、水手刀法统计、密度核估计、递归密度核估计、非线性回归核估计等等,都表现为随机变量序列加权和形式,因此对随机变量序列加权和极限性质的研究是有必要的.

最近Sung[1]对同分布NA随机变量序列加权和获得了如下完全收敛性结果.

定理A 设$1<\alpha\leq 2$,$\gamma>0$,$\{X,X_{n},n\ge 1\}$为同分布的NA随机变量序列,$\{a_{nk},n\geq1,$ $ 1\leq k\leq n\}$为常数序列满足

\[\mathop {\sup }\limits_{n \ge 1} \sum\limits_{k = 1}^n | {a_{nk}}{|^\alpha } < \infty .\] (1.1)
如果$EX=0$且
$\left\{\begin{array}{ll} E|X|^\alpha<\infty,&\ \ \mbox{如果}\ \alpha>\gamma,\\ E|X|^\alpha\log(1+|X|)<\infty,&\ \ \mbox{如果}\ \alpha=\gamma,\\ E|X|^\gamma<\infty,&\ \ \mbox{如果}\ \alpha<\gamma, \end{array}\right.$ (1.2)
则对任意$\varepsilon>0$有
\[\sum\limits_{n = 1}^\infty {{n^{ - 1}}} P(\mathop {\max }\limits_{1 \le j \le n} |\sum\limits_{k = 1}^j {{a_{nk}}} {X_k}| > \varepsilon {\log ^{1/\gamma }}n) < \infty .\] (1.3)

完全收敛性的概念是由Hsu和Robbins[2]最先提出并加以研究的. 从那时开始就已吸引了众多学者的关注,至今已取得了丰富的成果.

(1.3)式最先由Wang等[3]对NOD随机变量序列在很强的指数阶矩条件$E\exp(h|X|^\gamma)<\infty$下获得的,其中$h>0$.Sung[1]在NA情形下获了比较完美的结果,也即是上面的定理A. Zhou等[4],Sung[5],Wu等[6]把定理A推广到了$\rho^*$ -混合随机变量序列情形. Huang等[7],Shen[8],Zhang等[9]在NOD情形下获得了部分和的完全收敛性,但比部分和最大值的完全收敛性,即(1.3)式,要弱. Sung[1]的证明方法本质上用到了NA序列部分和最大值的指数不等式. Zhou等[4],Sung[5],Wu等[6]则是用到了$\rho^*$ -混合序列部分和最大值的Rosenthal型矩不等式,且Wu等[6]在其文中指出只要某种序列具有部分和最大值的Rosenthal型矩不等式,那么定理A总是成立的. Huang等[7],Shen[8],Zhang等[9]等使用的工具是NOD序列部分和的Rosenthal型矩不等式,同样只要某种序列满足部分和的Rosenthal型矩不等式,相应的结果也是成立的.

那么一个具有挑战性的问题是,如果只知道某序列满足部分和的Rosenthal型矩不等式,是否能得到(1.3)式这样的结果?

本文的目的是在更广泛的END随机变量序列情形下,利用部分和的Rosenthal型矩不等式,部分获得最大值的完全收敛性结果. 先来介绍END的概念.

定义1.1 称随机变量序列$\{X_n,n\geq1\}$是END(extended negatively dependent)的,如果存在$M>0$(称$M$为控制常数),使对任意$n\geq2$及任意实数$ x_1,x_2,\cdots,x_n$有

$$P(X_k>x_i,k=1,2,\cdots,n)\leq M\prod^n_{k=1}P(X_k>x_k) $$及 $$P(X_k\leq x_i,k=1,2,\cdots,n)\leq M\prod^n_{k=1}P(X_k\leq x_k). $$END这一概念是由Liu[10]在2009年引入的,当$M=1$时即为NOD序列的定义. 因此RND是包含NOD序列和NA序列在内的非常广泛的随机序列,在金融、保险、可靠性分析、多元统计分析和时间序列分析中有广泛的应用,因此已越来越引起关注. 如Chen等[11]获得了随机变量END序列的Kolomogorov强大数定律,Chen等[12],Liu[13],Cheng等[14],Wang等[15]等,在END结构下讨论了风险模型的破产概率问题,Shen[16]获得了END随机变量序列部分和的Rosenthal型矩不等式,Wang等[17]和Qiu等[18]讨论了在其它条件下的完全收敛性,等等.

下面来介绍本文的主要结果,必要的引理及定理的证明放到第2节.

定理1.1 $1<\alpha<2$,$\alpha<\gamma$,$\{X,X_{n},n\ge 1\}$为同分布的END随机变量序列满足$EX=0$及$E|X|^\gamma<\infty$,又设常数序列$\{a_{nk},n\geq1,1\le k\le n\}$满足(1.1)式. 则对任意$\varepsilon>0$,(1.3)式成立.

推论1.1 设$1<\alpha<2$,$\alpha<\gamma$,$\{X,X_{n},n\ge 1\}$为同分布的END随机变量序列满足$EX=0$及$E|X|^\gamma<\infty$.

(1) 若常序列$\{b_n,n\geq1\}$满足$\sum\limits^n_{k=1}|b_k|^\alpha=O(n)$,则对任意$\varepsilon>0$,

$\sum^\infty_{n=1}n^{-1}P\bigg\{\max_{1\leq m\leq n}\bigg|\sum^m_{k=1}b_kX_k\bigg|>\varepsilon n^{1/\alpha}(\log n)^{1/\gamma}\bigg\}<\infty.$ (1.4)
进而有
$\frac{\sum\limits^n_{k=1}b_kX_k}{n^{1/\alpha}(\log n)^{1/\gamma}}\rightarrow 0,\ \ {\rm a.s..}$ (1.5)

(2) 若常序列$\{b_n,n\geq1\}$满足$\sum\limits^\infty_{n=1}|b_n|^\alpha<\infty$,则对任意$\varepsilon>0$,

$$\sum^\infty_{n=1}n^{-1}P\bigg\{\max_{1\leq m\leq n}\bigg|\sum^m_{k=1}b_kX_k\bigg|>\varepsilon (\log n)^{1/\gamma}\bigg\}<\infty. $$进而有 $$\frac{\sum\limits^n_{k=1}b_kX_k}{(\log n)^{1/\gamma}}\rightarrow 0,\ \ {\rm a.s..} $$

本文约定,$C$总代表正常数,在不同的地方可以代表不同的值.

2 引理及主要结果的证明

定理的证明需要下面的引理.

引理2.1 (参见文献[10]) 设$X_1,X_2,\cdots,X_n$是END序列,$f_1,f_2,\cdots,f_n$全部是单调增(或单调减)函数. 则$f_1(X_1),f_2(X_2),\cdots,f_n(X_n)$是END的.

注2.1 引理2.1中$f_1(X_1),f_2(X_2),\cdots,f_n(X_n)$的控制常数可以与$X_1,X_2,\cdots,X_n$的控制常数相同.这一点在定理的证明中至关重要.

下面的引理就是END序列的Rosenthal不等式,可参见文献[16].

引理2.2 对任意$s\geq 2$,存在正常数$C_s$使得对任意END序列$\{X_n,n\geq1\}$均有

$$E\bigg|\sum_{k=1}^{n}(X_k-EX_k)\bigg |^s \le C_s\bigg\{\sum_{k=1}^{n}E|X_k|^s+ \bigg(\sum_{k=1}^{n }E|X_k|^2\bigg)^{s/2}\bigg\},\ \ \forall\ n\geq1. $$ 由引理2.2及类似文献 [19,定理2.3.1]的讨论有下面结论.

引理2.3 对任意$s\geq 2$,存在正常数$C_s$使得对任意END序列$\{X_n,n\geq1\}$均有

$$E\max_{1\le m\le n}\bigg|\sum_{k=1}^m(X_k-EX_k)\bigg|^s \le C_s(\log n)^s\bigg\{\sum_{k=1}^{n}E|X_k|^s+ \bigg(\sum_{k=1}^{n }E|X_k|^2\bigg)^{s/2}\bigg\},\ \ \forall\ n\geq1. $$

注2.2 引理2.2中常数$C_s$不但与$s$有关,还与控制常数$M$有关. 引理2.3中常数$C_s$同样如此.

定理1.1的证明 不妨设$a_{nk}>0$,$\sum\limits_{i=1}^n |a_{ni}|^\alpha\le 1.$对$1\le k\le n$及$n\ge 1$,令

\begin{eqnarray*}X_{nk}(1)&=&a_{ni} X_kI\left(|a_{ni}X_k|\le (\log n)^{1/\gamma-\beta}\right)+(\log n)^{1/\gamma-\beta}I\left(a_{nk}X_k> (\log n)^{1/\gamma-\beta}\right)\\&&-(\log n)^{1/\gamma-\beta}I\left(a_{nk}X_k<- (\log n)^{1/\gamma-\beta}\right),\end{eqnarray*} $${X_{nk}}(2) = \left( {{a_{nk}}{X_k} - {{(\log n)}^{1/\gamma - \beta }}} \right)I\left( {{{(\log n)}^{1/\gamma - \beta }} < {a_{nk}}{X_k} \le \varepsilon {{(\log n)}^{1/\gamma }}/(4N)} \right), $$ $${X_{nk}}(3) = \left( {{a_{nk}}{X_k} + {{(\log n)}^{1/\gamma - \beta }}} \right)I\left( { - \varepsilon {{(\log n)}^{1/\gamma }}/(4N) \le {a_{nk}}{X_k} < - {{(\log n)}^{1.\gamma - \beta }}} \right), $$\begin{eqnarray*}X_{n4}(4)&=&\left(a_{nk}X_k-(\log n)^{1/\gamma-\beta}\right)I\left(a_{nk}X_k>\varepsilon(\log n)^{1/\gamma}/(4N) \right)\\&& +\left(a_{nk}X_k+(\log n)^{1/\gamma-\beta}\right)I\left(a_{nk}X_k<-\varepsilon(\log n)^{1/\gamma}/(4N) \right),\end{eqnarray*}其中$0<\beta<1/\gamma$,$N$是一正整数(其值将在后面确定). 则 $$a_{nk}X_k=X_{nk}(1)+X_{nk}(2)+X_{nk}(3)+X_{nk}(4) $$且由引理2.1,$\{X_{nk}(1),1\le k\le n\}$也是END的. 令$b_n=(\log n)^{1/\gamma}$,注意到\begin{eqnarray*}&&\sum_{n=1}^\infty \frac{1}{n} P\bigg( \max_{1\le m \le n}\bigg|\sum_{k=1}^m a_{nk}X_k\bigg|>b_n \varepsilon \bigg)\\&\le& \sum_{n=1}^\infty \frac{1}{n} P\bigg( \max_{1\le m \le n}\bigg|\sum_{k=1}^m X_{nk}(1)\bigg|>b_n \varepsilon/4 \bigg) +\sum_{n=1}^\infty \frac{1}{n} P\bigg( \max_{1\le m \le n} \bigg|\sum_{k=1}^m X_{nk}(2) \bigg|>b_n \varepsilon/4 \bigg)\\&& + \sum_{n=1}^\infty \frac{1}{n} P\bigg( \max_{1\le m \le n}\bigg|\sum_{k=1}^m X_{nk}(3) \bigg|>b_n \varepsilon/4 \bigg)+\sum_{n=1}^\infty \frac{1}{n} P\bigg( \max_{1\le m \le n} \bigg|\sum_{k=1}^m X_{nk}(4) \bigg|>b_n \varepsilon/4 \bigg)\\&:=&I_1+I_2+I_3+I_4.\end{eqnarray*}因此要证(1.3)式成立,只须证$I_l<\infty$,$l=1,2,3,4$.

先证$I_1<\infty$. 由$EX_i=EX=0$及$E|X|^\alpha\leq (E|X|^\gamma)^{\alpha/\gamma}<\infty$,有

\begin{eqnarray*}&&b_n^{-1} \max_{1\le m\le n} \bigg|\sum_{k=1}^m EX_{nk}(1)\bigg|\\&\le& b_n^{-1} \max_{1\le m\le n} \bigg|\sum_{k=1}^m a_{nk} EX_k I\left(|a_{nk}X_k|\le (\log n)^{1/\gamma-\beta}\right)\bigg|\\&& +(\log n)^{-\beta}\sum_{k=1}^n P\left(|a_{nk}X_k|> (\log n)^{1/\gamma-\beta}\right)\\&=&b_n^{-1} \max_{1\le m\le n} \bigg|\sum_{k=1}^m a_{nk} EX_k I\left(|a_{nk}X_k|> (\log n)^{1/\gamma-\beta}\right)\bigg|\\&& +(\log n)^{-\beta}\sum_{k=1}^n P\left(|a_{nk}X_k|> (\log n)^{1/\gamma-\beta}\right)\\&\le& 2 b_n^{-1} \sum_{k=1}^n |a_{nk}| E|X_k|I\left(|a_{nk}X_k|> (\log n)^{1/\gamma-\beta}\right)\\&\le& 2 b_n^{-1} (\log n)^{(1/\gamma-\beta)(1-\alpha)}\sum_{k=1}^n |a_{nk}|^\alpha E|X_k|^\alpha I\left(|a_{nk}X_k|> (\log n)^{1/\gamma-\beta}\right)\\&\le& 2 E|X|^\alpha (\log n)^{\beta(\alpha-1)-\alpha/\gamma}\to 0.\end{eqnarray*}因此要证$I_1<\infty$,只须证 $$J=\sum_{n=1}^\infty \frac{1}{n} P\bigg( \max_{1\le m \le n} \bigg|\sum_{k=1}^m (X_{nk}(1)-EX_{nk}(1)) \bigg|>b_n \varepsilon/8\bigg)<\infty. $$对$n\ge 1$,$t>0$ (其值将在后面确定),令 $$A_n=\left\{1\le k\le n:~ |a_{nk}|\le (\log n)^{-t}\right\},~~ B_n=\left\{1\le k\le n:~ |a_{nk}|> (\log n)^{-t}\right\}. $$注意到 $$1\ge \sum_{k=1}^n |a_{nk}|^\alpha\ge \sum_{k\in B_n} |a_{nk}|^\alpha\ge (\log n)^{-t\alpha} \sharp B_n, $$于是有
$\sharp B_n\le (\log n)^{t\alpha}.$ (2.1)
我们将分别在$\gamma<2$和$\gamma\ge 2$情形下来证明$J<\infty$. 先设$\alpha<\gamma<2$,由引理2.2及(2.1)式有\begin{eqnarray*}J&\le& 64 \varepsilon^{-2} \sum_{n=1}^\infty n^{-1} b_n^{-2} E\max_{1\le m \le n} \bigg|\sum_{k=1}^m (X_{nk}(1)-EX_{nk}(1))\bigg|^2\\&\le &C\sum_{n=1}^\infty n^{-1} b_n^{-2} E\max_{1\le m \le n} \bigg\{\bigg|\sum_{k\le m,k\in A_n} (X_{nk}(1)-EX_{nk}(1))\bigg|^2 \\&& +\bigg|\sum_{k\le m,k\in B_n} (X_{nk}(1)-EX_{nk}(1))\bigg|^2 \bigg\}\\&\le& C \sum_{n=1}^\infty n^{-1} b_n^{-2} (\log n)^2 \sum_{k\in A_n} EX^2_{nk}(1)+C\sum_{n=1}^\infty n^{-1} b_n^{-2} (\log \log n)^2 \sum_{k\in B_n} EX^2_{nk}(1)\\&:=&J_1+J_2.\end{eqnarray*}因$\alpha<\gamma<2$,可取$t$足够大使得$t(\gamma-\alpha)+\beta(2-\gamma)>2$,于是有\begin{eqnarray*}J_1&=&C\sum_{n=1}^\infty n^{-1} b_n^{-2} (\log n)^2 \sum_{k\in A_n}\Big\{a_{nk}^2EX_k^2I\left(|a_{nk}X_k|\le (\log n)^{1/\gamma-\beta}\right)\\&& +b_n^2(\log n)^{-2\beta}P\left(|a_{nk}X_k|>(\log n)^{1/\gamma-\beta}\right)\Big\}\\&\le& C \sum_{n=1}^\infty n^{-1} b_n^{-\gamma} (\log n)^{2-\beta(2-\gamma)}\sum_{k\in A_n} |a_{nk}|^\gamma E|X_k|^\gamma\\&\le& C \sum_{n=1}^\infty n^{-1} b_n^{-\gamma} (\log n)^{2-\beta(2-\gamma)-t(\gamma-\alpha)}\sum_{k\in A_n} |a_{nk}|^\alpha E|X_k|^\gamma\\&\le &C E|X|^\gamma \sum_{n=1}^\infty n^{-1}(\log n)^{1-\beta(2-\gamma)-t(\gamma-\alpha)}<\infty.\end{eqnarray*}类似地有\begin{eqnarray*}J_2&\le& C \sum_{n=1}^\infty n^{-1} b_n^{-\gamma} (\log \log n)^2 (\log n)^{-\beta(2-\gamma)}\sum_{k\in B_n} |a_{nk}|^\gamma E|X_k|^\gamma\\&\le &C E|X|^\gamma \sum_{n=1}^\infty n^{-1} b_n^{-\gamma} (\log \log n)^2 (\log n)^{-\beta(2-\gamma)}\\&=&C E|X|^\gamma \sum_{n=1}^\infty n^{-1} (\log \log n)^2 (\log n)^{-1-\beta(2-\gamma)}<\infty.\end{eqnarray*}

下面来考虑情形$\gamma\ge 2$. 取$q>\max\{2,\gamma\}$,由引理2.3及(2.1)式有

\begin{eqnarray*}J&\le &C \sum_{n=1}^\infty n^{-1} b_n^{-q} (\log n)^q \bigg\{ \sum_{k\in A_n} E\left|X_{nk}(1)\right|^q+ \bigg(\sum_{k\in A_n} EX_{nk}^2(1)\bigg)^{q/2}\bigg\}\\&& +C \sum_{n=1}^\infty n^{-1} b_n^{-q} (\log \log n)^q \bigg\{ \sum_{k\in B_n} E\left|X_{nk}(1)\right|^q+ \bigg(\sum_{k\in B_n} EX_{nk}^2(1)\bigg)^{q/2}\bigg\}\\&:=&J_3+J_4+J_5+J_6.\end{eqnarray*}类似于$J_1<\infty$的证明,取$t$足够大,使$t(\gamma-\alpha)+\beta(q-\gamma)>q$,有\begin{eqnarray*}J_3&\le& C \sum_{n=1}^\infty n^{-1} b_n^{-\gamma} (\log n)^{q-\beta(q-\gamma)}\sum_{k\in A_n} |a_{nk}|^\gamma E|X_k|^\gamma\\&\le &C \sum_{n=1}^\infty n^{-1} b_n^{-\gamma} (\log n)^{q-\beta(q-\gamma)-t(\gamma-\alpha)}\sum_{k\in A_n} |a_{nk}|^\alpha E|X_k|^\gamma\\&\le &C E|X|^\gamma \sum_{n=1}^\infty n^{-1}(\log n)^{-1+q-\beta(q-\gamma)-t(\gamma-\alpha)}<\infty.\end{eqnarray*}同样,取$t$足够大,使$t(2-\alpha)q/2+q/\gamma-q>1$,有\begin{eqnarray*}J_4&\le& C \sum_{n=1}^\infty n^{-1} b_n^{-q} (\log n)^q \bigg( \sum_{k\in A_n} a_{nk}^2 EX_k^2\bigg)^{q/2}\\&\le &C \left(EX^2\right)^{q/2} \sum_{n=1}^\infty n^{-1} b_n^{-q} (\log n)^q \bigg( (\log n)^{-t(2-\alpha)} \sum_{k\in A_n} |a_{nk}|^\alpha\bigg)^{q/2}\\&\le &C \left(E|X|^\gamma\right)^{q/\gamma} \sum_{n=1}^\infty n^{-1} (\log n)^{q-q/\gamma-t(2-\alpha)q/2}.\end{eqnarray*}同样有 $$J_5\le C E|X|^\gamma \sum_{n=1}^\infty n^{-1} (\log \log n)^q (\log n )^{-1-\beta(q-\gamma)}<\infty, $$ $$J_6\le C \left(E|X|^\gamma\right)^{q/\gamma} \sum_{n=1}^\infty n^{-1}(\log \log n)^q (\log n)^{-q/\gamma}<\infty. $$

现在来证明$I_2<\infty$. 因$0\le X_{nk}(2)\le b_n\varepsilon/(4N),$ $\sum\limits_{k=1}^n X_{nk}(2)>b_n \varepsilon/4$,这就意味着$\{X_{nk}(2),1\leq k\leq n\}$中至少有$N$个$X_{nk}(2)$不为零. 于是

\begin{eqnarray*}&&P\bigg( \max_{1\le m \le n} \bigg|\sum_{k=1}^m X_{nk}(2) \bigg|>b_n \varepsilon/4\bigg)\\&=&P\bigg( \sum_{k=1}^n X_{nk}(2)>b_n \varepsilon/4 \bigg)\\&\le&\sum_{1\le k_1<\cdots < k_N\le n} P\left( a_{n,k_1}X_{k_1}>b_n (\log n)^{-\beta},\cdots,a_{n,k_N}X_{k_N}>b_n (\log n)^{-\beta}\right)\\&\le &M\sum_{1\le k_1<\cdots < k_N\le n} P\left( a_{n,k_1}X_{k_1}>b_n (\log n)^{-\beta}\right)\cdots P\left(a_{n,k_N}X_{k_N}>b_n (\log n)^{-\beta}\right)\\&\le &M\bigg( \sum_{k=1}^n P\left( a_{nk}X_k >b_n (\log n)^{-\beta}\right)\bigg)^N\\&\le &M\bigg( b_n^{-\gamma} E|X|^\gamma (\log n)^{\beta\gamma} \sum_{k=1}^n |a_{nk}|^\gamma \bigg)^N\\&\le& M\left(E|X|^\gamma\right)^N (\log n)^{(-1+\beta\gamma)N}.\end{eqnarray*} 取$N$足够大使得$(1-\beta\gamma)N>1$,就有$I_2<\infty$.

$I_3<\infty$的证明与$I_2<\infty$类似.

最后来证$I_4<\infty$. 由Chebyshev不等式,(1.1)式及标准的计算有

\begin{eqnarray*}I_4&\le& \sum_{n=1}^\infty n^{-1} \sum_{k=1}^n |a_{nk}|^\alpha E|X_k|^\alpha I(|a_{nk}X_k|>b_n\varepsilon/(4N))\\&\le &\sum^\infty_{n=1}n^{-1}E|X|^\alpha I(|X|>b_n\varepsilon/(4N))\\&\le &CE|X|^\gamma<\infty.\end{eqnarray*}这就完成了定理的证明.
参考文献
[1] Sung S H. Strong convergence for weighted sums of random variables. Stat Papers, 2011, 52:447-454
[2] Hsu P, Robbins H. Complete convergence and the law of large numbers. Proc Natl Acad Sci USA, 1947, 33:25-31
[3] Wang X, Hu S, Yang W. Complete convergence for arrays of rowwise negatively orthant dependent random variables. Rev R Acad Cien Serie A Mat, 2012, 106:234-245
[4] Zhou X C, Tan C C, Lin J G. On the strong laws for weighted sums of ρ*-mixing random variables. J Ineq Appl, 2011, DOI:10.1155/2011/157816
[5] Sung S H. On the strong convergence for weighted sums of ρ*-mixing random variables. Stat Papers, 2013, 54:773-781
[6] Wu Y F, Sung S H, Volodin A. A note on the rates of convergence for weighted sums of ρ*-mixing random variables. Lith Math J, 2014, 54:220-228
[7] Huang H, Wang D. A note on the strong limit theorem for weighted sums of sequence of negatively dependent random variables. J Ineq Appl, 2012, DOI:10.1186/1029-242X-2012-233
[8] Shen A. On the strong convergence rate for weighted sums of arrays of rowwise negatively orthant dependent random variables. Rev R Acad Cien Serie A Mat, 2013, 107:257-271
[9] Zhang Q, Wang D. A note on the rate of strong convergence for weighted sums of arrays of rowwise negatively orthant dependent random variables. Discrete Dyn Nat Soc, 2014, Article ID:308702
[10] Liu L. Precise large deviations for dependent random variables with heavy tails. Statist Probab Lett, 2009, 79(9):1290-1298
[11] Chen Y, Chen A Y, Ng K W. The strong law of large numbers for extended negatively dependent random variables. J Appl Probab, 2010, 47:908-922
[12] Chen Y, Yuen K C, Ng K W. Precise large deviations of random sums in presence of negative dependence and consistent variation. Methodol Comput Appl Probab, 2011, 13:821-833
[13] Liu L. Necessary and sufficient conditions for moderate deviations of dependent random variables with heavy tails. Sci China Math, 2010, 53(5):1421-1434
[14] Cheng F Y, Li N. Asymptotics for the tail probability of random sums with a heavy tailed random number and extended negatively dependent summands. Chin Ann Math Ser B, 2014, 35:69-78
[15] Wang K, Wang Y, Gao Q. Uniform asymptotics for the finite-time ruin probability of a dependent risk model with a constant interest rate. Methodol Comput Appl Probab, 2013, 15:109-124
[16] Shen A T. Probability inequalities for END sequence and their applications. J Ineq Appl, 2011, DOI:10.1186/1029-242X-2011-98
[17] Wang X J, Hu T C, Volodin A, Hu S H. Complete convergence for weighted sums and arrays of rowwise extended negatively dependent random variables. Communications in Statistics Theory and Methods, 2013, 42:2391-2401
[18] Qiu D, Chen P, Giuliano R, Volodin A. On the complete convergence for arrays of rowwise extended negatively dependent random variables. J Korean Math Soc, 2013, 50(2):379-392
[19] Stout W F. Almost Sure Convergence. New York:Academic, 1974