节点文献

相依随机变量的强收敛性

【作者】 刘庆

【导师】 苏中根;

【作者基本信息】 浙江大学 , 概率论与数理统计, 2007, 硕士

【摘要】 本文是我在硕士阶段完成的,主要研究了相依随机变量的强大数定律和完全收敛性。全文共分三章:第一章两两PQD序列的强大数律和完全收敛性两两PQD序列的概念是Lehmann提出的,它是一类比较广泛的随机变量序列,包含常见的相伴序列。对这方面的研究已经有了一些结果,但不如相伴序列多。Matula得到了两两PQD序列几乎处处中心极限定理,严继高等给出了两两PQD序列的Jamison型加权部分和的强稳定性,陆凤彬给出了与PA列类似的完全收敛性。在第一章中,我们主要考虑了两两PQD序列的Marcinkiewicz型强大数律和另一形式完全收敛性,得到如下结果:定理0.1设{Xn;n≥1}是均值为0的两两PQD列,且sum from i=1 to∞v1/2(2i)<∞。令1≤p<2;φ:R→R+是非负连续的偶函数,(?)φ(x)=∞,对某个1<s≤2使得(?)↑且φ(x)/xs↓,x→∞。如果sum from n=1 to∞log2 n[(?)Eφ(|Xi|/φ(n1/p)]1/2<∞则1/n1/psum from i=1 to n Xi→0,a.s. n→∞。定理0.2设{Xn;n≥1}是两两PQD列,对(?)ε>0,有sum from j=1 to∞VarXj/j2+sum from j≠k=1 to∞Cov(Xj,Xk)/j·k<∞和sum from j≠k=1 to∞P(|sum from i=1 to j(Xi-EXi)|≥jε,|sum from i=1 to k(Xi-EXi)|≥kε)<∞。则sum from n=1 to∞(logn)-2P(|sum from i=1 to n(Xi-EXi)|≥ε)<∞。第二章一类PA序列非单调函数的强大数律和完全收敛性1984年Newman给出了如下定义:定义0.1设f和f1是定义在Rn上两实值函数,则f《f1当且仅当f+f1,f-f1均为逐点单调不减。特别地,如果f《f1,则f1为逐点非降的。设{Xn;n≥1}为PA序列,令(ⅰ)Yn=fn(X1,X2…,Xn)(ⅱ)(?)n=(?)n(X2,X2…,Xn) (0.0.1)(ⅲ)fn《(?)n(ⅳ)EYn2<∞,E(?)n2<∞,n∈N当条件(ⅰ)-(ⅳ)成立时,我们记为Yn《(?)n。其中fn,(?)n为实函数,且仅依赖于有限个Xn。关于{Yn;n≥1}的极限性质已经有了一些结果。Matula(2001)证明了{Yn;n≥1}的强大数律和中心极限定理,Dewan和Rao(2006)年得到了Hajek-Renyi型不等式及推广的三级数定理。在第二章我们得到如下结果:定理0.3{Xn;n≥1}为PA列,Yn,(?)n如(0.0.1)式定义,且Yn《(?)n。gn(x)是偶函数列,且在区间x>0取正值不减,而且对每一个n满足下列条件之一:(ⅰ)区间x>0中,x/gn(x)不减;(ⅱ)区间x>0中,x/gn(x),gn(x)/x2都不增且EYn=E(?)n=0。此外{αn;n≥1}是常数列,满足0<αn↑∞,且sum from n=1 to∞Egn(Yn)/gn(an)<∞和sum from n=1 to∞(Egn((?)n)/gn(an1/2<∞。则当n→∞时1/αn sum from k=1 to n Yk→0 a.s。定理0.4{Xn;n≥1}为PA列,Yn,(?)n如(0.0.1)式定义,且Yn《(?)n。若对(?)ε>0,sum from j=1 to∞Var(?)j/j2+sum from j≠k=1 to∞Cov((?)j,(?))k)/j·k<∞和sum from j≠k=1 to∞P(|sum from i=1 to j(Yi-EYi)|≥jε,|sum from i=1 to k(Yi-EYi)|≥kε)<∞。则sum from n=1 to∞P(|sum from i=1 to n(Yi-EYi)|≥ε)<∞。第三章相依线性过程的完全收敛性假设{Xi;-∞<i<∞}是随机变量序列,{αi;-∞<i<∞}是绝对可加的实数序列,定义线性过程{Yk;k≥1}:Yk=sum from i=-∞to∞αi+kXi。(0.0.2)我们称(0.0.2)式定义的线性过程为滑动平均过程。滑动平均过程是时间序列的重要研究对象,许多作者在适当的假设下得到了相应的极限性质。例如Burton和Dehling(1990)在条件Eexp(tX1)<∞下得到了大偏差原理;Ibragimov(1962)建立了的中心极限定理;Yang(1996)得到重对数律;Li et al.(1992)得到完全收敛性。本章第二节{Xi;—∞<i<∞}为两两NQD序列时的上述形式的完全收敛性。得到如下结果:定理0.5设1<p<2,1/2<α≤1,αp≥1。若(ⅰ)h(x)>0(x>0)为→∞时的缓变函数;(ⅱ){αi;—∞<i<∞}是绝对可加的实数序列;(ⅲ)EX1=0,EX12<∞。则E|X1|ph(|X11/α|)<∞蕴涵sum from n=1 to∞nαp-2h(n)P(|sum from k=1 to n Yk|≥nαε)<∞,(?)ε>0。设{Yt;t∈Z+}是概率空间(Ω,(?),P)上如下定义的线性过程:Yt=sum from j=0 to∞αjXt-j。(0.0.3)其中{αj;j≥0}是实数列,sum from j=0 to∞|αj|<∞;{Xt;t∈Z+}是随机过程且EXt=0,0<EXt2<∞。我们称(0.0.3)式定义的线性过程为一般线性过程。显然,滑动平均过程是一般线性过程的特殊情况。此类线性过程在时间序列分析中起着特殊的作用,而且由于它在经济,工程,物理科学等方面的广泛应用而备受关注。大量的结果都是通过对{Xt;t∈Z+}施加各种各样的条件得到的。Fakhre Zaberi和Lee建立了独立同分布时的CLT;1997年他们又证明了强混合条件下的FCLT;Tae-Sun Kim和Baek建立了平稳LPQD过程时的CLT;而Tae-Sun Kim,Mi-Hwa Ko和Dong Ho Park得到了LPQD和PA过程时的强大数定律。本章第三节我们考虑了正相依样本下的线性过程的完全收敛性,得到如下结果:定理0.6设1<p<2,1/2<α≤1,αp≥1。若(ⅰ)h(x)>0(x>0)为x→∞时的缓变函数;(ⅱ){αj;j≥0}是实数列且sum from j=0 to∞|αj|<∞;(ⅲ){Xn;n≥1}是被X0所界的PA列,EXn=0且sum from i=1 to∞v1/2(2i)<∞。则E|X0|ph(|X01/α|)<∞蕴涵sum from n=1 to∞nαp-2h(n)P(|Sn|≥εnα)<∞,(?)ε>0。同时,对于两两PQD样本的一般线性过程,我们也得到类似定理0.6的结果。

【Abstract】 This thesis is finished during my master of science, mainly discusses SLLN andcomplete convergence for dependent random variables. It consist of three chapters:In chapterⅠ, we mainly discuss the strong law of large numbers and completeconvergence for pairwise PQD sequences.The concept of pairwise PQD sequences was introduced by Lehmann. It is a kindof generic sequences, including associated sequences. Matula established a almost ev-erywhere central limit theorem; Yan Jigao proved a strong convergence for Jamison-wise weighted sums of pairwise PQD sequences; Lu Fengbin obtained a completeconvergence which is similar to PA sequences. In this chapter, the following resultsare obtained:Theorem0.1 Suppose{Xn; n≥1}is a pairwise PQD sequence with mean 0, andsum from i=1 to∞v1/2(2i)<∞. Let 1≤p<2;φ: R→R+ is nonnegative, even and continuousfunction, (?)φ(x)=∞, for some 1<s≤2,φ(x)/x↑andφ(x)/x8↓, x→∞. Assume sum from n=1 to∞log2 n[(?) Eφ(|Xi|)/φ(n1/p]1/2<∞Then 1/n1/p sum from i=1 to n Xi→0, a.s. n→∞.Theorem0.2 Suppose{Xn; n≥1} is a pairwise PQD sequence, for(?)ε>0, sum from j=1 to∞VarXj/j2+sum from j≠k=1 to∞Cov(Xj, Kk)/j·k<∞and sum from j≠k=1 to∞P(|sum from i=1 to j(Xi-EXi)|≥jε, |sum from i=1 to k(Xi-EXi)|≥kε)<∞.Then sum from n=1 to∞(logn)-2P(|sum from i=1 to n(Xi-EXi)|≥ε)<∞.In chapterⅡ, we mainly discussed the SLLN for a sequence of nonmonotonic func-tions of associated random variables.The following definition was given by Newman in 1984:Definition0.1 Let f and f1 be two real-valued functions defined on Rn, then f<<f1 if and only if f+f1, f-f1 are both nondecreasing componentwise. In particular, if f<<f1, then f1 will be nondecreasing componentwise.Let{Xn; n≥1} be a PA sequences. Let(ⅰ) Yn=fn(X1, X2,…, Xn)(ⅱ) (?)n=(?)n(X1, X2,…, Xn) (0.0.4)(ⅲ) fn<<(?)n(ⅳ) EYn2<∞, E(?)n2<∞, n∈NIf the conditions (ⅰ)-(ⅳ) hold, we write Yn<<(?)n. There are some limiting resultson{Yn; n≥1}. Matula(2001) proved SLLN and CLT for {Yn; n≥1}, Dewan andRao(2006)obtained Hajek-Renyi-tpye inequlity. We prove the following result:Theorem0.3 {Xn; n≥1} is a PA sequence, Yn, (?)n is defined in (0.0.1), andYn<<(?)n. gn(X) is even functions, and is positive and nondecreasing when x>0, forevery n satisfies the alternative assumption that:(ⅰ) x/gn(x) is nondecreasing in (0,∞);(ⅱ) x/gn(x), gn(x)/x2 are nonincreasing in (0,∞). Meanwhile EYn=E(?)n=0.In addition, {an; n≥1} is a sequence of real numbers, with 0<an↑∞, sum from n=1 to∞Egn(Yn)/gn(an)<∞and sum from n=1 to∞(Egn((?)n))/gn(an1/2<∞. Then when n→∞, 1/an sum from k=1 to n Yk→0 a.s.In addition, we prove a complete convergence for {Yn; n≥1} similar to Theo-rem0.2.In chapterⅢ, a complete convergence for linear processes under dependent as-sumption is discussed.Assume that {Xi; -∞<i<∞} is a doubly infinite sequence, let {ai; -∞<i<∞} be an absolutely summable sequence of real numbers, and {Yk; k≥1}: Yk=sum from i=-∞to∞ai+kXi. (0.0.5)Linear processes defined as (0.0.5) are called moving average processes. Manylimiting results were obtained for moving average processes {Yk; k≥1}. Burton andDehling(1990) obtained large deviation principle assuming Eexp(tX1)<∞; Ibragimov(1962) established CLT; Li et al.(1992) obtained a complete convergence. In sectionⅡ, weprove a complete convergence when{Xi; -∞<i<∞} is a pairwise NQD sequence:Theorem0.5 Supposel<p<2, 1/2<α<1,αp≥1. Let(ⅰ) h(x)>0(x>0) be a slowly varying function, when x→∞;(ⅱ) {ai; -∞<i<∞} be an absolutely summable sequence of real numbers;(ⅲ) EX1=0, EX12<∞. ThenE|X1|ph(|X11/α|)<∞imply sum from n=1 to∞nαp-2h(n)P(|sum from k=1 to n Yk|≥nαε)<∞, (?)ε>0.Let{Yt; t∈Z+} be a linear process defined on a probability space (Ω, (?), P): Yt=sum from j=0 to∞ajXt-j. (0.0.6)where {aj; j≥0} is a sequence of real numbers, sum from j=0 to∞|aj|<∞; {Xt; t∈Z+} is a processand EXt=0, 0<EXt2<∞.Linear processes defined as (0.0.6) are called general linear processes. Obviously, moving average processes are special cases of general linear processes. The linear pro-cesses are of special importance in time series analysis. Fakhre Zaberi and Lee obtainedCLT under i.i.d assumption; in 1997, they obtained FCLT under the strong mixingcondition; Tae-Sun Kim, Mi-Hwa Ko and Dong Ho Park proved SLLN under theLPQD and PA condition on {Xt; t∈Z+}. In sectionⅢ, we discuss the linear pro-cess under positive dependence condition on {Xt; t∈Z+}, and obtain the followingresult:Therorem0.6 Supposel<p<2, 1/2<α≤1,αp≥1. Let(ⅰ) h(x)>0(x>0) be a slowly varying function, when x→∞;(ⅱ) {aj; j≥0} be a sequence of real numbers and sum from j=0 to∞|aj|<∞;(ⅲ) {Xn; n≥1} is a PA sequence bounded by X0, EXn =0 and sum from i=1 to∞v1/2(2i)<∞.Then E|X0|Ph(|X01/α|)<∞imply sum from n=1 to∞nαp-2h(n)P(|Sn|≥εnα)<∞, (?)ε>0.Meanwhile, we obtain a result under pairwise PQD condition on {Xt; t∈Z+}similar to the above.

  • 【网络出版投稿人】 浙江大学
  • 【网络出版年期】2007年 06期
  • 【分类号】O211
  • 【下载频次】81
节点文献中: 

本文链接的文献网络图示:

本文的引文网络