数学物理学报  2016, Vol. 36 Issue (2): 353-361   PDF (319 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
王伟刚
杨广宇
高振龙
有限矩条件下变化环境中分枝过程的收敛定理
王伟刚1, 杨广宇2, 高振龙3    
1. 浙江工商大学统计与数学学院 杭州 310018;
2. 郑州大学数学学院 郑州 450001;
3. 曲阜师范大学数学科学学院 山东 曲阜 273165
摘要: 该文研究了变化环境中分枝过程的收敛定理.在环境分布不独立的情况下,给定环境分布的矩条件,证明了WnLt收敛到W,并且W>0,a.s.,以此为基础,给出了该过程Zn的中心极限定理,以及log Zn的重对数律.这些结果对研究其它的渐进性质以及偏差理论都有重要的意义.
关键词: 随机环境     分枝过程     中心极限定理     重对数律    
Convergence of Branching Process in Varying Environments when All Moments Being Finite
Wang Weigang1, Yang Guangyu2, Gao Zhenlong3     
1. School of Statistics and Mathematics, Zhejiang Gongshang University, Hangzhou 310018;
2. School of mathematics, Zhengzhou university, Zhengzhou 450001;
3. School of mathematics, Qufu normal university, Shandong Qufu 273165
Foundation Item: Supported by the Natural Science Foundation of Zhejiang Province (LY13A010003), the NSFC (11371321, 11201420), Humanities and Social Sciences Foundation of Ministry of Education (10YJC790091), Foundation of Zhejiang Educational Committee (Y201326953) and Zhejiang Gongshang University Institute of Humanities and Social Sciences (Statistics)
Abstract: In this paper, we studied convergence theorems of the branching processes in varying environments. When the environment is not independence, at the moment conditions of the environment, we prove that $W_n \mathop{\longrightarrow}\limits^{L^t}W$ and $W>0,$ a.s. firstly, and then we give the central limit theorem of the process, at last we give the law of the iterated logarithm of $\log Z_n$. Those results are very important to the other asymptotic properties and deviations of the process.
Key words: Branching process     Random environments     Central limit theorem     LIL    
1 引言

令${\Bbb N}=\{0,1,2,\cdots\}$,${\Bbb Z}_+=\{1,2,\cdots\}$,$(\Omega,{\cal F},P)$ 为一概率空间,$(\Theta,{\cal B})$为环境空间,$({\cal X},{\cal A})$ 为状态空间,其中${\cal X}={\Bbb N}$,${\cal A}$ 为${\cal X}$离散$\sigma$域. 对于$\forall \theta\in\Theta$,$\{p_n(\theta);n\in{\Bbb N}\}$ 为概率分布列,且满足

$$\sum_{j=0}^{+\infty}jp_j(\theta)<+\infty,0\le p_0(\theta)+p_1(\theta)<1. $$ 对于任意$\theta\in\Theta$,对应有概率母函数

$$f_\theta(s)=p_0+p_1s+p_2s^2+\cdots+p_ns^n+\cdots. $$

定义1.1 令$\vec{\xi}=\{\xi_n,n\in{\Bbb N}\}$是$(\Omega,{\cal F},P)$上取值于$(\Theta,{\cal B})$的随机变量序列,$\vec{Z}=\{Z_n,n\in{\Bbb N}\}$和$\{X_{n_i},n\in{\Bbb N},i\in{\Bbb Z}_+\}$是$(\Omega,{\cal F},P)$上取值于$({\cal X},{\cal A})$的随机变量序列,且满足

$$Z_{n+1}=\sum_{i=1}^{Z_n}X_{n_i},P(X_{n_i}=j|\vec{\xi})=p_j(\xi_n),j\in{\cal X}, $$ $$P(X_{n_i}=j_{n_i},1\le i\le m,0\le n<k|\vec{\xi})=\prod_{n=0}^k\prod_{i=1}^mP(X_{n_i}=j_{n_i}|\vec{\xi}),j_{n_i}\in {\cal X}. $$ 则称$\vec{Z}$是随机环境$\vec{\xi}$中的分枝过程(BPRE) (参见文献[1]).

本文恒设 $\vec{Z}$是随机环境 $\vec{\xi}$中的分枝过程,并且$Z_0=1$.令$\vec{\Theta}=\Theta^{{\Bbb N}}$,$T$为$\vec{\Theta}$的转移算子,即$T^k(\vec{\theta})=(\theta_k,\theta_{k+1},\cdots)$.对任意$\vec{\theta}\in\vec{\Theta}$,简记$P_{\vec{\theta}}(\cdot )=P(\cdot|\vec{\xi}=\vec{\theta})$,$P,P_{\vec{\theta}}$相对应的期望分别为$E,E_{\vec{\theta}}$.令$Q=P\circ\vec{\xi}^{-1}$为$(\vec{\Theta},\vec{{\cal B}})$上的概率测度,满足

$$Q(\vec{B})=P(\vec{\xi}\in\vec{{\cal B}}),\forall \vec{B}\in\vec{{\cal B}}. $$

给出如下矩条件(A)(B)(C)(D),及条件(E):

(A) 任意$n\in{\Bbb N}$,存在$d_n>1$,使得对于$Q,$ a.s. $\vec{\theta},\forall k\in {\Bbb N},E_{T^k\vec{\theta}}Z_1^n<d_n.$

(B) 存在$d_2$,使得对于$Q,$ a.s. $\vec{\theta},\forall k\in {\Bbb N},E_{T^k\vec{\theta}}Z_1^2\le d_2.$

(C) 存在$t>2$,$d_t>1$,使得对于$Q,$ a.s. $\vec{\theta},\forall k\in {\Bbb N},E_{T^k\vec{\theta}}Z_1^t\le d_t.$

(D) 存在$u>1$,使得$E_{T^k\vec{\theta}}Z_1\ge u,$ $Q$,a.s. $\vec{\theta}$.

(E) 对于任意$\theta\in\Theta$,恒有$p_{0}(\theta)=0$.

由于$Z_1$为非负整数值随机变量,故对任意$0<s\le t$有$Z_1^s\le Z_1^t,$ a.s.,所以若存在$t>0$,使得$ E_{T^k\vec{\theta}}Z_1^d<d_t<\infty,$$Q,$ a.s. $\vec{\theta}$,则$\forall 0< s\le t$有,$ E_{T^k\vec{\theta}}Z_1^s\le E_{T^k\vec{\theta}}Z_1^t<d_t<\infty,$$Q,$ a.s. $\vec{\theta}.$ 故条件(A)等价于任意$t>0$,存在$d_t>1$,使得$ E_{T^k\vec{\theta}}Z_1^t<d_t<+\infty,$$Q,$ a.s. $\vec{\theta}.$

2 主要结论

文献[3]中,当环境独立同分布时,在某给定条件下(例如$EZ_1^a<\infty,\forall a\ge 1$),得到结论:存在$C(t)\in(0,\infty)$使得

$$\lim_{n\rightarrow\infty}\frac{EZ_n^t}{(Em_0^t)^n}=C(t), $$ 其中$m_0=E(Z_1|\xi_0)$. 这给求 $\log Z_n$的Laplace变换$\Lambda(t)=\lim\limits_{n\rightarrow\infty}\frac{1}{n}\log EZ_n^t$,并进一步利用Cärtner-Ellis定理求$\log Z_n$的大偏差带来很大便利. 在环境不平稳的条件下,随机环境中分枝过程的大偏差理论就要复杂很多. 文章[8] 讨论了其偏差性质. 另外,很多文献,例如文献[6],在环境平稳遍历的条件下得到了

$$P(W>0)=P(\lim_{n\rightarrow\infty}Z_n=\infty). $$ 因此,在$p_0(\xi)=0$,a.s 的条件下,$W_n=\frac{Z_n}{m_n}\rightarrow W>0$,a.s. 本文在给定条件下,利用引理3.5 同样证明了$W>0$,a.s.

本文的主要结果是在环境的各阶矩有限的条件下,给出过程的中心极限定理[4, 6]和大偏差定理[2, 3, 5].简记$u_n=E_{T^k\vec{\theta}}(Z_1)$,令$m_n=\prod\limits_{k=0}^{n-1}u_k$,$W_n=\frac{Z_n}{m_n}$,易知$W_n$为非负上鞅,所以存在非负随机变量$W$,使得$W_n\mathop{\longrightarrow}\limits^{\rm a.s.}W$,若$W>0,$ a.s.,由于$\frac{1}{n}\log Z_n=\frac{1}{n}\log W_n+\frac{1}{n}\log m_n$,因此$\frac{1}{n}\log Z_n$和$\frac{1}{n}\log m_n$有相似的偏差性质.因而$W>0$是否几乎处处成立以及$W$的矩是很多学者关心的问题(参见文献[3]).本文主要是在给定子孙分布一阶矩和二阶矩有界的条件下给出了$W_n$的$L^2$收敛性和$W>0,$ a.s.,并给出了过程的其它一些渐近性质. 具体有如下主要结论. 在不致引起混淆的情况下简记$P$,a.s.为a.s..

定理2.1 在条件(B)、(D)、(E)下,存在$f_2>0$,使得$EW_n^2\le f_2$,从而存在随机变量$W$,使得$W_n \mathop{\longrightarrow}\limits^{L^2} W$,并且$W>0$,a.s..

定理2.2 在条件(C)、(D)、(E)下,

$$\lim_{n\rightarrow\infty}P\bigg(\frac{Z_n-m_nW}{\sigma(T^n\vec{\xi})\sqrt{Z_n}}\le x\bigg)=\Phi(x),\forall x\in {\Bbb R}, $$ 其中$\Phi(x)$为$N(0,1)$的分布函数,$\sigma^2(\vec{\xi})=Var(W|\vec{\xi})$.

定理2.3 在条件(B)、(D)、(E)下,并且$\{\xi_i,i\ge0\}$为独立随机变量,存在$a>0$,使得 $Var \log u_1(\xi_k)\ge a$,则

$$\limsup_{n\rightarrow\infty}(\liminf)\frac{\log Z_n-a_n}{\sqrt{2s_n^2\log\log s_n^2}}=1(-1),{\rm a.s.},$$ 其中$a_n=\sum\limits_{k=0}^{n-1}E\log m_k$,$s_n^2=\sum\limits_{k=0}^{n-1}Var(\log m_k)$.

条件(E)是为了防止$0$做分母和对$0$取对数,在此条件下过程$Z_n$是单调非减的.

例2.4 令$p_1(\theta_1)=\frac{1}{2},p_3(\theta_1)=\frac{1}{2}$; $p_1(\theta_2)=\frac{1}{4},p_5(\theta_2)=\frac{3}{4}$;$p_1(\theta_3)=\frac{1}{8},p_9(\theta_3)=\frac{7}{8}$. 环境过程$\xi_0,\xi_1,\cdots$独立,并且$P(\xi_{2n}=\theta_1)=P(\xi_{2n}=\theta_2)=\frac{1}{2}$,$P(\xi_{2n+1}=\theta_1)=P(\xi_{2n+1}=\theta_3)=\frac{1}{2}$. 容易验证条件(A)--(E)成立,从而定理2.1--2.3成立. 并且

$$E\log u_{2n}=\frac{3}{2}\log2,E\log u_{2n+1}=2\log2. $$ $$Var\log u_{2n}=\frac{1}{4}\log^22,Var\log u_{2n+1}=\log^22. $$ $$E\log m_n=\sum_{k=0}^{n-1}E\log u_k\thicksim\left(\frac{7}{4}\log2\right)n,a_n=\sum_{k=0}^{n-1}E\log m_k\thicksim\left(\frac{7}{8}\log2\right)n^2. $$ $$Var\log m_n=\sum_{k=0}^{n-1}Var \log u_k\thicksim\left(\frac{5}{8}\log^22\right)n,s_n^2=\sum_{k=0}^{n-1}Var\log m_k\thicksim\left(\frac{5}{16}\log^22\right)n^2. $$ $$\sqrt{2s_n^2\log\log s_n^2}\thicksim\log2\sqrt{\frac{5}{8}\log\log n}\cdot n. $$ 因此由定理2.3

$$\limsup_{n\rightarrow\infty}(\liminf)\frac{\log Z_n-\frac{7}{8}\log2\cdot n^2}{\log2\sqrt{\frac{5}{8}\log\log n}\cdot n}=1(-1),{\rm a.s.}. $$
3 定理证明

引理3.1 在条件(A)和(D)下,任意$t>0$,存在$f(t)$使得$EW_n^t\le f(t)$,即$EW_n^t$对$n\ge0$有界.

对于$t\ge 1$,只须证明对任意$m\in{\Bbb N}$,有$EW_n^m\le f(m)$. 用归纳法证明.

对于$m=0$,显然$EW_n^0$有界. 假设对$k\le m-1$有$EW_n^k\le f(k)$.当$k=m$时,

$$Z_{n+1}^m=\bigg(\sum_{k=1}^{Z_n}X_{n,k}\bigg)^m\\=\sum_{k=1}^{Z_n}X_{n,k}^m+\sum_{1\le i<j\le Z_n,k_i+k_j=m}X_{n,i}^{k_i}X_{n,j}^{k_j}+\sum_{1\le i<j<l\le Z_n,k_i+k_j+k_l=m}X_{n,i}^{k_i}X_{n,j}^{k_j}X_{n,l}^{k_l}\\+\cdots + \sum_{1<i_1<i_2<\cdots<i_m}X_{n,i_1}X_{n,i_2}\cdots X_{n,i_m},$$ 所以对$Q$ a.s. $\vec{\theta}$

$$E_{\vec{\theta}}Z_{n+1}^m\le E_{\vec{\theta}}Z_nd_m+E_{\vec{\theta}}Z_n^2m^2d_m^2+\cdots+E_{\vec{\theta}}Z_n^{m-1}m^{m-1}d_m^{m-1}+E_{\vec{\theta}}Z_n^mu_n^m,$$ 又$u_1(\theta_k)\ge u>1$,可得$m_n\ge u^n$,a.s.,所以

$$E_{\vec{\theta}}W_{n+1}^m=E_{\vec{\theta}}\frac{Z_{n+1}^m}{m_{n+1}^m}\\\le\frac{1}{m_{n+1}^{m-1}}d_m+E_{\vec{\theta}}W_n^2\frac{1}{m_{n+1}^{m-2}}m^2d_m^2+\cdots+E_{\vec{\theta}}W_n^{m-1}\frac{1}{m_{n+1}}m^{m-1}d_m^{m-1}+E_{\vec{\theta}}W_n^m\\\le\frac{1}{m_{n+1}^{m-1}}d_m+\frac{1}{m_{n+1}^{m-2}}f(2)m^2d_m^2+\cdots+\frac{1}{m_{n+1}}f(m-1)m^{m-1}d_m^{m-1}+E_{\vec{\theta}}W_n^m\\\le\bigg(\sum_{k=1}^{n+1}\frac{1}{m_k^{m-1}}\bigg)d_m+\bigg(\sum_{k=1}^{n+1}\frac{1}{m_k^{m-2}}\bigg)f(2)m^2d_m^2\\+\cdots+\bigg(\sum_{k=1}^{n+1}\frac{1}{m_k}\bigg)f(m-1)m^{m-1}d_m^{m-1}\\\le\bigg(\sum_{k=1}^{+\infty}\frac{1}{u^{k(m-1)}}\bigg)d_m+\bigg(\sum_{k=1}^{+\infty}\frac{1}{u^{k(m-2)}}\bigg)f(2)m^2d_m^2\\+\cdots+\bigg(\sum_{k=1}^{+\infty}\frac{1}{u^k}\bigg)f(m-1)m^{m-1}d_m^{m-1}\\\triangleq f(m).$$ 因此有,$EW_n^m\le f(m)$.

对于$0<t<1$,由$x^t$的凹性,$EW_n^t\le (EW_n)^t=1$,引理得证.

引理3.2 在条件(B)和(D)下,对a.s. $\omega$,存在$N$,任意$n>N$有

$$Z_n<Z_{n+1}. $$

参见文献[8,引理3.3].

引理3.3 对于$0<p<1$,存在$l\in{\Bbb N}$,使得$\sum\limits_{n=1}^{+\infty}C_n^{[\frac{n}{l}]}p^{[n-\frac{n}{l}]}<+\infty.$

因为$\lim\limits_{n\rightarrow\infty}nep^n=0$,可取$l\in{\Bbb N}$,使得$lep^l<1$.要证$\sum\limits_{n=1}^{+\infty}C_n^{[\frac{n}{l}]}p^{[n-\frac{n}{l}]}<+\infty,$只须证明

$$\sum_{n=1}^{+\infty}C_{ln}^np^{l(n-1)}<+\infty,\cdots,\sum_{n=1}^{+\infty}C_{ln+l-1}^np^{ln-1}<+\infty. $$ 因为

$$\lim_{n\rightarrow\infty}\frac{C_{l(n+1)}^{n+1}}{C_{ln}^n}p^l=\lim_{n\rightarrow\infty}\frac{[l(n+1)]!}{(n+1)![(l-1)(n+1)]!}\cdot\frac{n![(l-1)n]!}{(ln)!}p^l\\=\lim_{n\rightarrow\infty}\frac{l(n+1)[l(n+1)-1]\cdots(ln+1)}{(n+1)[(l-1)(n+1)]\cdots[(l-1)n+1]}p^l\\=l(1+\frac{1}{l-1})^{l-1}p^l\\\le lep^l<1,$$ 由正项级数收敛判别法,$\sum\limits_{n=1}^{+\infty}C_{ln}^np^{ln}<+\infty$收敛,其它项类似可证. 引理3.4 在条件(B)和(D)下,存在$a>0$,对a.s. $\omega$,存在$N$,任意$n>N$有 $$Z_{n+1}\ge Z_n(1+a). $$

由文献[8,引理3.2],在条件(B)和(D)下,存在$0<p<1$,使得$p_0(\theta_k)<p<1$,$Q,$ a.s. $\vec{\theta}$.由引理3.3存在$l\in{\Bbb N}$,使得$\sum\limits_{n=1}^{+\infty}C_n^{[\frac{n}{l}]}p^{[n-\frac{n}{l}]}<+\infty.$令$a=\frac{1}{l}$. 由引理3.2对于足够大的$n$,有$Z_n<Z_{n+1},$ a.s.,不妨设任意$n\ge0$,有$Z_n<Z_{n+1}$,a.s.,对于足够大的$n$,

$$P(\{Z_{n+1}<Z_n(1+a)\})=E[P(Z_{n+1}<Z_n(1+a)|Z_n)]\\\le EC_{Z_n}^{[Z_n-Z_na]}p^{[Z_n(1-a)]}\\\le C_n^{[n-na]}p^{[n(1-a)]}\\\le C_n^{[\frac{n}{l}]+1}p^{[n-\frac{n}{l}]}.$$ 所以

$$P\bigg(\bigcup_{n=1}^{+\infty}\{Z_{n+1}<Z_n(1+a)\}\bigg)\le\sum_{n=1}^{+\infty}P(\{Z_{n+1}<Z_n(1+a)\})\\\le C_n^{[\frac{n}{l}]+1}p^{[n-\frac{n}{l}]}<+\infty.$$ 由Borel-Cantelli引理,此即

$$P(\{Z_{n+1}<Z_n(1+a),{\rm i.o.}\})=0. $$ 本引理得证.

引理3.5 在条件(B)和(D)下,对a.s. $\omega$,存在$N$,任意$n>N$有

$$Z_{n+1}\ge Z_n(u_n-\frac{1}{n^2}). $$

由引理条件,存在$b>0$,使得$Var(Z_1|\vec{\xi})<b$,a.s.,令

$$\tau=\inf\{N;\forall n\ge N,Z_{n+1}\ge Z_n(1+a)\}, $$ 由引理3.4,$P(\tau<+\infty)=1$.

$$P\bigg(\bigcup_{n=N+k}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}|\tau=N\bigg)\\ \le\sum_{n=N+k}^{+\infty}P(\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}|\tau=N)\\\le\sum_{n=N+k}^{+\infty}P(\{|Z_{n+1}-Z_nu_n|>\frac{Z_n}{n^2}\}|\tau=N)\\\le\sum_{n=N+k}^{+\infty}E\bigg(\frac{Z_n\delta_{\xi_n}^2n^4}{Z_n^2}|\tau=N\bigg)\\\le \sum_{n=k}^{+\infty}\frac{b(N+n)^4}{(1+a)^n}.$$ 从而

$$P\bigg(\bigcap_{k=1}^{+\infty}\bigcup_{n=N+k}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}|\tau=N\bigg)\\=\lim_{k\rightarrow\infty}P\bigg(\bigcup_{n=N+k}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}|\tau=N\bigg)\\=\lim_{k\rightarrow\infty}\sum_{n=k}^{+\infty}\frac{b(N+n)^4}{(1+a)^n}=0.$$ 所以可得

$$P\bigg(\bigcap_{m=1}^{+\infty}\bigcup_{n=m}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}\bigg)\\=P\bigg(\bigcap_{m=1}^{+\infty}\bigcup_{n=m}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\},\tau<\infty\bigg)\\=\sum_{k=1}^{+\infty}P(\tau=k)P\bigg(\bigcap_{m=1}^{+\infty}\bigcup_{n=m}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}|\tau=k\bigg)\\=\sum_{k=1}^{+\infty}P(\tau=k)P\bigg(\bigcap_{m=N}^{+\infty}\bigcup_{n=m}^{+\infty}\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2})\}|\tau=k\bigg)=0.$$ 亦即

$$P(\{Z_{n+1}<Z_n(u_n-\frac{1}{n^2}),{\rm i.o.}\})=0, $$ 本引理得证.

证明定理2.1$W_n$关于$\sigma$代数${\cal F}_n=\sigma\{Z_0,Z_1,\cdots,Z_{n-1},\xi_0,\cdots\xi_{n-1}\}$为非负鞅,从而存在$W$,使得$W_n\rightarrow W$,a.s.,由引理3.1的证明,$\sup\limits_{n\ge0}EW_n^2\le f(2)$,所以由鞅的收敛定理,$W_n \mathop{\longrightarrow}\limits^{L^2}W$. 令

$$\tau=\inf\{N;\forall n\ge N,Z_{n+1}>Z_n(1+a)\}, $$ 由引理3.5,$P(\tau<+\infty)=1$. 所以

$$ \lim_{n\rightarrow\infty}W_n=\lim_{n\rightarrow\infty}\frac{Z_n}{m_n}\ge \lim_{n\rightarrow\infty}\frac{\prod\limits_{k=\tau}^n(u_k-\frac{1}{k^2})\cdot Z_{\tau}} {\prod\limits_{k=\tau}^nu_k\cdot\prod\limits_{k=0}^{\tau-1}u_k}\\ = \lim_{n\rightarrow\infty}\prod_{k=\tau}^n(1-\frac{1}{k^2u_k})\cdot\frac{Z_{\tau}}{\prod\limits_{k=0}^\tau u_k}\\ \ge\lim_{n\rightarrow\infty}\prod_{k=\tau}^n(1-\frac{1}{k^2u})\cdot\frac{Z_{\tau}}{\prod\limits_{k=0}^\tau u_k}\\ ={\rm e}^{\sum\limits_{k=\tau}^{\infty}\log(1-\frac{1}{k^2u})}\cdot\frac{Z_{\tau}}{\prod\limits_{k=0}^\tau u_k}>0. $$ 故有,$W>0$,a.s.,本定理得证.

由定理2.1的证明以及引理3.1,我们有如下推论.

推论3.6 在条件(A)、(D)、(E)下,存在$f_t>0$,使得$EW_n^t\le f_t$,从而存在随机变量$W$,使得$W_n \mathop{\longrightarrow}\limits^{L^t} W$,并且$W>0,$ a.s..

推论3.7 在条件(A)、(D)、(E)下,若环境过程$\vec{\xi}$独立同分布,则 $$\lim_{n\rightarrow\infty}\frac{1}{n}\log P(\frac{\log Z_n}{n}\ge x)=-\Lambda^{*}(x),(\forall x\ge E\log \mu_0), $$ 其中$\Lambda(t)=\log E\mu_0^t,\Lambda^{*}=\sup\{tx-\Lambda(t),t\ge0\}$.

$$\tilde{Q}_0(d\theta)=\frac{\mu(\theta)^tQ_0(d\theta)}{E\mu_0^t},\tilde{Q}=\tilde{Q}_0^{\otimes{\Bbb N}},\tilde{{\Bbb P}}={\Bbb P}_{\vec{\xi}}\otimes \vec{Q}. $$ 易知,在测度$\tilde{Q}$下$\vec{\xi}$仍然满足条件(A)、(D)、(E),故由推论3.6,在测度$\tilde{{\Bbb P}}$下$W_n\mathop{\longrightarrow}\limits^{L^t}W$,并且$\tilde{E}W^t\le f(t)$,$W>0,$ a.s.,所以$0<\tilde{E}W^t<\infty$. 又$\frac{EZ_n^t}{(E\mu_0^t)^n}=\tilde{E}W_n^t$,可得

$$\lim\limits_{n\rightarrow\infty}\frac{1}{n}\log EZ_n^t=\lim_{n\rightarrow\infty}\frac{1}{n}\tilde{E}W_n^t+\lim_{n\rightarrow\infty}\frac{1}{n}\log (E\mu_0^t)^n= \log E\mu_0^t,(\forall t\ge 0). $$ 证毕.

由文献[4,定理6.1],可证本推论.

引理3.8 对任意$n>0$,$\{\xi_{n,j},j\ge1\}$为一列期望为$0$方差为$1$的独立同分布随机变量,$N_n$为一列正整数值随机变量,令$L_n=\sum\limits_{j=1}^{N_n}\xi_{n,j}$,$G_n(x)=P(\frac{L_n}{\sqrt{N_n}}\le x)$,如果存在$\delta,$ $M>0$,使得$E|\xi_{n,j}|^{2+\delta}\le M$,则

$$\sup_{x}|G_n(x)-\Phi(x)|\le CE(N_n^{-\frac{\delta}{2}}), $$ 其中$\Phi(x)$为标准正态分布的分布函数.

证明参加文献[5,引理3.1].

证明定理2.2 $Z_n-m_nW$可以写成

$$Z_n-m_nW=\sum_{j=1}^{Z_n}(1-W_n^j)~~~ {\rm a.s.}, $$ 其中$W_n^j$关于$j\ge0$独立同分布,并且

$$P(W_n^j\in\cdot|T^n\vec{\xi})=P(W\in\cdot|\vec{\xi}). $$ 所以在条件$\vec{\xi}$下,$\frac{1-W_n^j}{\sigma(T^n\vec{\xi})}$,关于$j\ge0$为期望为$0$,方差为$1$的独立同分布随机变量序列.又由引理3.1及定理条件,

$$E_{\vec{\xi}}\bigg|\frac{1-W_n^{j}}{\sigma(T^n\vec{\xi})}\bigg|^t=E_{T^n\vec{\xi}}\bigg|\frac{1-W}{\sigma(\vec{\xi})}\bigg|^t<+\infty,\exists t>2. $$ 所以由引理3.6可得

$$\bigg|P_{\vec{\xi}}\bigg(\frac{Z_n-m_nW}{\sigma(T^n\vec{\xi})\sqrt{Z_n}}\le x\bigg)-\Phi(x)\bigg|=\bigg|P_{\vec{\xi}}\bigg(\frac{\sum\limits_{j=1}^{Z_n}(1-W_n^j)}{\sigma(T^n\vec{\xi})\sqrt{Z_n}}\le x\bigg)-\Phi(x)\bigg|\le CE_{\vec{\xi}}(Z_n^{-\frac{t-2}{2}}), $$ 故有

$$\bigg|P\bigg(\frac{Z_n-m_nW}{\sigma(T^n\vec{\xi})\sqrt{Z_n}}\le x\bigg)-\Phi(x)\bigg|\le E\bigg|P_{\vec{\xi}}\bigg(\frac{Z_n-m_nW}{\sigma(T^n\vec{\xi})\sqrt{Z_n}}\le x\bigg)-\Phi(x)\bigg| \le CE(Z_n^{-\frac{t-2}{2}}). $$ 因此

$$\lim_{n\rightarrow\infty}P\bigg(\frac{Z_n-m_nW}{\sigma(T^n\vec{\xi})\sqrt{Z_n}}\le x\bigg)=\Phi(x),\forall x\in {\Bbb R}. $$ 定理得证.

引理3.9 $\{X_i,i\ge1\}$为独立随机变量序列,其中每个随机变量的期望为$0$,$S_n=\sum\limits_{k=1}^nX_k$,$s_n^2=\sum\limits_{k=1}^nEX_i^2$,若存在$\delta>0$,$K<\infty$,使得

$$\liminf_{n\rightarrow\infty}\frac{s_n^2}{n}>0,E|X_i|^{2+\delta}\le K<\infty, $$ 则有重对数律成立,即

$$\limsup_{n\rightarrow\infty}(\liminf)\frac{S_n}{\sqrt{2s_n^2\log\log s_n^2}}=1(-1). $$

参见文献[9,推论5.2.3].

证明定理2.3 由定理条件知$\log u_1(\xi_k)$,$k\ge0$为独立随机变量,可得

$$0<\log u\le\log u_1(\xi_k)\le\log d_1,{\rm a.s.}. $$ 令$s_n^2=\sum\limits_{k=0}^{n-1}Var\log u_1(\xi_k)$,有

$$\liminf_{n\rightarrow\infty}\frac{s_n^2}{n}\ge \liminf_{n\rightarrow\infty}\frac{na}{n}=a. $$ 对于$\delta>0$,有

$$E|\log u_1(\xi_k)-E\log u_1(\xi_k)|^{2+\delta}\le E|\log d_1-\log u|^{2+\delta}\\= |\log d_1-\log u|^{2+\delta}<+\infty.$$ 由引理3.7,$\log u_1(\xi_k)$满足重对数律,即

$$\limsup_{n\rightarrow\infty}(\liminf)\frac{\sum\limits_{k=0}^{n-1}(\log u_1(\xi_k)-E\log u_1(\xi_k))}{\sqrt{2s_n^2\log\log s_n^2}}=1(-1). $$ 由定理2.1,$W>0$,a.s.,所以

$$\limsup_{n\rightarrow\infty}\frac{\log Z_n-a_n}{\sqrt{2s_n^2\log\log s_n^2}}=\limsup_{n\rightarrow\infty}\frac{\log W_n+\log m_n-a_n}{\sqrt{2s_n^2\log\log s_n^2}}\\=\limsup_{n\rightarrow\infty}\frac{\log m_n-a_n}{\sqrt{2s_n^2\log\log s_n^2}}=1,$$ 同理

$$\liminf_{n\rightarrow\infty}\frac{\log Z_n-a_n}{\sqrt{2s_n^2\log\log s_n^2}}=-1. $$ 定理得证.

参考文献
[1] Tanny David. Limit theorems for branching in a random environment. Ann Prob, 1977, 5(1):100-116
[2] Dembo A, Zeitouni O. Large Deviation Techniques and Applications. New York:Springer, 1998
[3] Huang Chunmao, Liu Quansheng. Moments, moderate and large deviations for a branching process in random environment. Stochastic Proc Appl, 2012, 122:522-545
[4] Liu Quansheng, Rouault A. Limit theorems for Mandelbort's multiplicative cascades. Ann Appl Probab, 2000, 10:218-239
[5] Gao Zhenlong, Hu Xiaoyu. Limit theorems for a Galton-Watson process in the i.i.d random environment. Acta Mathematica Scientia, 2012, 32B(3):1193-1205
[6] Wang Hesong, Gao Zhiqiang, Liu Qusheng. Central limit theorems for a supercritical branching processes in a random environment. Stat Prob Lett, 2011, 81:539-547
[7] Wang Weigang. Bounds of deviation for branching chains in random environments. Acta Math Sinica, 2011, 27(5):897-904
[8] 王伟刚,明瑞星,甘建军. 变化环境下Galton-Watson过程的极限理论. 数学学报, 2014, 57(10):923-930 Wang Weigang, Ming Ruixing, Gan Jianjun. The limits theorems of Galton-Watson process in varying environments. Acta Mathematica Sinica (Chinese Series), 2014, 57(10):923-930
[9] Stout William F. Almost Sure Convergence. New York:Academic Press, 1974