节点文献

一些随机变量序列部分和的大偏差定理

Large Deviations for the Partial Sums of Several Random Variable Sequences

【作者】 王学军

【导师】 胡舒合;

【作者基本信息】 安徽大学 , 概率论与数理统计, 2007, 硕士

【摘要】 近两个世纪以来,有关随机变量序列部分和的各种收敛性问题,如大数定律和中心极限定理等,一直是概率极限理论研究的主要问题,而关于随机变量序列部分和的大偏差却研究得很少。设{Xn,n≥1}是定义在概率空间(Ω,F,μ)上的随机变量序列,Sn=sum from i=1 to n Xi,Xn∈Lp,n≥1,1≤p<∞。如果{Xn,n≥1}是独立同分布(i.i.d.)的随机变量序列,则由弱大数定律知(?)μ(|Sn|>nx)=0,x>0。更一般地,如果随机变量序列{Xn,n≥1}是强平稳的,则遍历性定理蕴含了上述结果仍然正确、有关μ(|Sn|>nx)的收敛速度的问题,已经引起了很多学者的关注,在这中间,Nagaev(Theory Probab.Appl.10(1965),214-235)得到了估计:μ(|Sn|>n)=o(n1-p),1≤p<∞,Xi∈Lp,Lesigne和Volny(Stochastic Process.Appl.96(2001),143-159)又证明了上述估计是最优的;如果{Xn,n≥1}是鞅差序列,Lesigne和Volny(Stochastic Process.Appl.96(2001),143-159)证明了:如果supi E(e|Xi|)<∞,则存在常数c>0,使得μ(|Sn|>n)≤e-cn1/3,这个估计对于强平稳和遍历的鞅差序列来说是最优的;如果鞅差序列{Xn,n≥1}满足:Xi∈Lp,2≤p<∞,Lesigne和Volny(Stochastic Process.Appl.96(2001),143-159)又得到了估计:μ(|Sn|>n)≤cn-p/2,并且还证明了这个估计对于强平稳和遍历的鞅差序列来说也是最优的;Yulin Li(Statistics and Probability Letters,62(2003),317-321)又将此结果推广到p∈(1,2]的情形,利用Burkholder不等式、Cr不等式和鞅的极大值不等式得到了估计:μ(|Sn|>n)≤cn1-p,在一定情况下,这个估计是最优的。本文主要利用ρ混合序列、φ混合序列、(?)混合序列、(?)混合序列、NA序列、M-Z序列和线性过程序列的一些矩不等式,研究了它们的部分和序列Sn的大偏差定理,并且得到了与独立序列和鞅差序列类似的大偏差定理。

【Abstract】 In recent two centuries, kinds of convergence properties for the partial sums of randomvariable sequences, such as strong law of large numbers and central limit theorem, havebeen the key subjects for probability limit theory research. But large deviations for thepartial sums of random variable sequences have seldom been studied.Let {Xn, n≥1} be a random variable sequences defined on a fixed probability space(Ω,F,μ), and let Sn =sum from i=1 to n Xi, Xn∈Lp, n≥1, 1≤p<∞. If {Xn, n≥1} isindependent and identically distributed(i.i.d.), the weak law of large numbers asserts that (?)μ(|Sn|>nx)=0,x>0.More generally, if the sequence {Xn, n≥1} is stationary(in the strong sense), thenthe ergodic theorem asserts that the result is still true. In recent years, some authors havepaid much attentions to the problem of growth rate ofμ(|Sn|>nx), for example, Na-gaev(Theory Probab.Appl.10(1965), 214-235) got the estimationμ(|Sn|>n)= o(n1-p) forXi∈Lp, 1≤p<∞, and Lesigne and Volny (Stochastic Process. Appl. 96(2001), 143-159)gave a simple proof that the estimate of Nagaev can’t be improved; If {Xn, n≥1} is a mar-tingale difference sequence, Lesigne and Volny (Stochastic Process. Appl. 96(2001), 143-159) proved that if supi E(eXi)<∞, then there exists a constant c>0 such thatμ(|Sn|>n)≤e-cn1/3, this bound is optimal for the class of martingale difference sequenceswhich are also strictly stationary and ergodic; If the sequence {Xn, n≥1}is bounded inLp, 2≤p<∞, then Lesigne and Volny (Stochastic Process. Appl. 96(2001), 143-159)got the estimationμ(|Sn|>n)≤cn-p/2 which is again optional for strictly station-ary and ergodic sequences of martingale difference; Yulin Li(Statistics and ProbabilityLetters, 62(2003), 317-321) generalized the result to the case for 1<p≤2, by usingBurkholder’s inequality, Cr-inequality and martingale maximal inequality, he obtainedμ(|Sn|>n)≤cn1-p, these are optimal in a certain sense.In the paper, we study the large deviations for the partial sums ofρ-mixing se- quence,φ-mixing sequence,(?)-mixing sequence,(?)-mixing sequence, NA sequence, M-Z-type sequence and Linear process sequence using some moment inequalities, and obtainthe similar results optimal upper bounds forμ(|Sn|>n) as those for independent andidentically distributed sequence and martingale difference sequence.

  • 【网络出版投稿人】 安徽大学
  • 【网络出版年期】2007年 06期
  • 【分类号】O211
  • 【下载频次】102
节点文献中: 

本文链接的文献网络图示:

本文的引文网络