极限理论是概率论中备受关注的问题之一, 中心极限定理和大数律的研究, 可上溯至十八世纪初, 不过那时的结果主要针对的是单序列的部分和(求和的个数是确定的). Kolmogorov在20世纪三四十年代, 对行内独立的二重序列确定和的极限定理(包括中心极限定理和重对数律等)做了系统的研究, 并得出一系列奠基性的结果[1-2].
在此之后, 由于实际的需要, 中心极限定理的研究对象开始从单纯的行内独立的随机变量序列的确定和转向随机个随机变量之和.如胡迪鹤[3]通过单序列随机和的中心极限定理结果, 将Donsker不变原理推广到分枝过程的情形.
1970—1971年, Heyde[4]得到关于分枝过程的一系列极限定理结果, 其中包括中心极限定理和重对数律.其后一些工作讨论了随机环境中分枝过程的极限理论[5-6].
1 背景知识本文讨论一致渐进可忽略(u.a.n.)二重随机变量序列的随机和的P-收敛性, 包括变化环境中的分枝过程的P-收敛性.对二重随机变量序列的确定和而言, P-收敛性的讨论通常是弱收敛研究的铺垫[2].
令
定义1.1 令Z0=1且对于任意的n≥0,
${Z_{n + 1}} = \left\{ \begin{array}{l} \sum\limits_{j = 1}^{{Z_n}} {{X_{n,j}}\left( x \right),\;{Z_n} \ne 0;} \\ 0,\;\;\;\;\;\;\;\;\;\;\;\;\;\;{Z_n} = 0, \end{array} \right.$ |
其中,
定义1.2 设
$\forall \epsilon > 0,\mathop {\lim }\limits_{n \to \infty } \mathop {\max }\limits_{1 \le k \le {\xi _n}} P\left( {\left| {{V_{n,k}}} \right| \ge \epsilon} \right) = 0\;\;\;{\rm{a}}{\rm{.s}}{\rm{.}}$ |
注1.1 当ξn≡kn时(kn为正整数,
引理1.1 沿用前面的符号, 设{Zn, n≥0}为定义1.1中所定义的变化环境中的上临界的分枝过程, {Xn, j:j≥1}的期望为μn.假定Zn→∞a.s.当
${Y_{n,j}}: = \frac{{{X_{n,j}} - {\mu _n}}}{{f\left( {{Z_n}} \right)}},$ |
是关于{Zn, n≥0}的u.a.n.序列.
证明 首先根据上临界分枝过程的性质和pn(0)≡0的假设知: Zn→∞a.s., 当n→∞.
现在只需要对于任意的
$\mathop {\lim }\limits_{n \to \infty } \mathop {\max }\limits_{1 \le j \le {Z_n}} P\left( {\frac{{\left| {{X_{n,j}} - {\mu _n}} \right|}}{{f\left( {{Z_n}} \right)}} \ge \epsilon} \right) = 0\;{\rm{a}}{\rm{.s}}{\rm{.}}$ |
给定
$\begin{array}{l} P\left( {\frac{{\left| {{X_{n,j}} - {\mu _n}} \right|}}{{f\left( {{Z_n}} \right)}} \ge \epsilon} \right)\\ \le P\left( {\frac{{\left| {{X_{n,j}} - {\mu _n}} \right|}}{{f\left( {{Z_n}} \right)}} \ge \epsilon,{\mathit{\Omega} _0}} \right) + \eta \\ \le \frac{1}{{\epsilon f\left( C \right)}}E\left( {\left| {{X_{n,j}} - {\mu _n}} \right|} \right) + \eta \\ \le \frac{M}{{\epsilon f\left( C \right)}} + \eta , \end{array}$ |
令C→∞, 则有对a.s.ω
$\mathop {\lim }\limits_{n \to \infty } \mathop {\max }\limits_{1 \le j \le {Z_n}} P\left( {\frac{{\left| {{X_{n,j}}{\mu _n}} \right|}}{{f\left( {{Z_n}} \right)}} \ge \epsilon} \right) \le \eta ,$ |
而η是任意的, 故
注1.2 类似于上述证明可得, 在同样条件下
$\mathop {\lim }\limits_{n \to \infty } \mathop {\max }\limits_{1 \le j \le {Z_n}} P\left( {\frac{{\left| {{X_{n,j}}{\mu _n}} \right|}}{{f\left( {{Z_n}} \right)}} \ge \epsilon\left| {{Z_n}} \right.} \right) = 0\;\;{\rm{a}}{\rm{.s}}{\rm{.}}$ |
此时称
注1.3 令Zn, k(j)为定义1.1中的分枝过程第n代第j个个体的第k代子孙的数量, 记
命题2.1 设
1)
2)在任意的有限区间[a, b]上对a.s.ω有,
3)在任意的有限区间[a, b]上对a.s.ω有,
4)对任意的t,
5)对任意的t,
6)对任意的t,
证明 1)
$\begin{array}{c} {\mathop{\rm Re}\nolimits} \left( {1 - {f_{n,k}}\left( t \right)} \right)\\ = \int_{ - \infty }^\infty {\left( {1 - \cos xt} \right){\rm{d}}{F_{n,k}}\left( x \right)} \\ = \int_{\left| x \right| < \epsilon } {\left( {1 - \cos \;xt} \right){\rm{d}}{F_{n,k}}\left( x \right) + } \\ \int_{\left| x \right| \ge \epsilon} {\left( {1 - \cos xt} \right){\rm{d}}{F_{n,k}}\left( x \right)} \\ \le \frac{{{\epsilon^2}{T^2}}}{2} + \int_{\left| x \right| \ge \epsilon} {2{\rm{d}}{F_{n,k}}\left( x \right)} \\ = \frac{{{\epsilon^2}{T^2}}}{2} + 2P\left( {\left| {{X_{n,k}}} \right| \ge \epsilon} \right), \end{array}$ |
由1)知存在Ω0与t无关且P(Ω0)=1, 使得
$\mathop {\lim }\limits_{n \to \infty } \mathop {\max }\limits_{1 \le k \le {\xi _n}} P\left( {\left| {{V_{n,k}}} \right| \ge \epsilon} \right) = 0,$ |
故有
$\mathop {\lim \;\;\sup }\limits_{n \to \infty } \mathop {\max }\limits_{1 \le k \le {\xi _n}} {\mathop{\rm Re}\nolimits} \left( {1 - {f_{n,k}}\left( t \right)} \right) \le \frac{{{\epsilon^2}{T^2}}}{2},$ |
上述收敛性关于
2)
$1 - {f_{n,k}}\left( t \right) = {\mathop{\rm Re}\nolimits} \left( {1 - {f_{n,k}}\left( t \right)} \right) - \int_{ - \infty }^\infty {{\rm{i}}\sin xt{\rm{d}}{F_{n,k}}\left( x \right)} ,$ |
又因为
$\begin{array}{l} {\left| {\int_{ - \infty }^\infty {{\rm{i}}\sin xt{\rm{d}}{F_{n,k}}\left( x \right)} } \right|^2}\\ \le \int_{ - \infty }^\infty {{{\sin }^2}tx{\rm{d}}{F_{n,k}}\left( x \right)} \\ = \frac{1}{2}\int_{ - \infty }^\infty {\left( {1 - \cos 2tx} \right){\rm{d}}{F_{n,k}}\left( x \right)} \\ = \frac{1}{2}{\mathop{\rm Re}\nolimits} \left( {1 - {f_{n,k}}\left( {2t} \right)} \right), \end{array}$ |
故由2)可得3).
3)
5)
$\int_{ - \infty }^\infty {\frac{{{x^2}}}{{1 + {x^2}}}{\rm{d}}F\left( x \right) \le M\left( t \right)\int_0^t {\left( {1 - {\mathop{\rm Re}\nolimits} \left( {f\left( v \right)} \right)} \right){\rm{d}}v,\forall t > 0.} } $ |
其中,
于是
$\begin{array}{l} \mathop {\max }\limits_{1 \le k \le {\xi _n}} \int_{ - \infty }^\infty {\frac{{{x^2}}}{{1 + {x^2}}}{\rm{d}}{F_{n,k}}\left( x \right)} \\ \le \mathop {\max }\limits_{1 \le k \le {\xi _n}} M\left( t \right)\int_0^t {{\mathop{\rm Re}\nolimits} } \left( {1 - {f_{n,k}}\left( v \right)} \right){\rm{d}}v\\ \le M\left( t \right)\int_0^t {\mathop {\max }\limits_{1 \le k{\xi _n}} {\mathop{\rm Re}\nolimits} \left( {1 - {f_{n,k}}\left( v \right)} \right){\rm{d}}v,\forall t > 0,} \end{array}$ |
由Lebesgue控制收敛定理, 对上式两端取极限, 立得6).
6)
$\begin{array}{l} \mathop {\max }\limits_{1 \le k \le {\xi _n}} P\left( {\left| {{v_{n,k}}} \right| \ge \epsilon} \right)\\ = \mathop {\max }\limits_{1 \le k \le {\xi _n}} \int_{\left| x \right| \ge \epsilon} {\frac{{1 + {x^2}}}{{{x^2}}}\frac{{{x^2}}}{{1 + {x^2}}}{\rm{d}}{F_{n,k}}\left( x \right)} \\ \le \frac{{1 + {\epsilon^2}}}{{{\epsilon^2}}}\mathop {\max }\limits_{1 \le k \le {\xi _n}} \int_{\left| x \right| \ge \epsilon} {\frac{{{x^2}}}{{1 + {x^2}}}{\rm{d}}{F_{n,k}}\left( x \right),} \end{array}$ |
由6)立得1).
注2.1 经典的u.a.n.序列的相应结果见文献[7].
命题2.2 设
1)
2)
3)
4)
证明 1)不妨设对于所有的
$\mathop {\max }\limits_{1 \le k \le {\xi _n}\left( \omega \right)} P\left( {\left| {{V_{n,k}}} \right| \ge \epsilon} \right) < \frac{1}{2}$ |
而
$\mathop {\max }\limits_{1 \le k \le {\xi _n}\left( \omega \right)} \left| {{m_{n,k}}} \right| < \epsilon,n \ge N,\omega \in \mathit{\Omega} ,$ |
由
2)任给r>0, τ>0以及
$\begin{array}{l} \mathop {\max }\limits_{1 \le k \le {\xi _n}} \int_{\left| x \right| < \tau } {{{\left| x \right|}^r}{\rm{d}}{F_{n,k}}\left( x \right)} \\ \le \mathop {\max }\limits_{1 \le K \le {\xi _n}} \int_{\left| x \right| < \epsilon} {{{\left| x \right|}^r}{\rm{d}}{F_{n,k}}\left( x \right) + } \\ \mathop {\max }\limits_{1 \le k \le {\xi _n}} \int_{\epsilon \le \left| x \right| < \tau } {{{\left| x \right|}^r}{\rm{d}}{F_{n,k}}\left( x \right)} \\ \le {\epsilon^r} + {\tau ^r}\mathop {\max }\limits_{1 \le k \le {\xi _n}} P\left( {\left| {{V_{n,k}}} \right| \ge \epsilon} \right), \end{array}$ |
再根据
3)在2)中取r=1即可.
4)由3)可知
注2.2 该命题推广了经典u.a.n.序列的相应结论, 见文献[7].
定理2.1 令
(a)
(b)
则存在随机变量序列
$\sum\limits_{j=1}^{{{\xi }_{n}}}{{{V}_{n,k}}-{{V}_{n}}\xrightarrow{P}0,}\ 当n\to \infty .$ |
另一方面, 若存在形如h(ξn)的随机变量(h是Borel可测函数), 对任意的
$\mathop {\lim }\limits_{n \to \infty } \left( {\left| {\sum\limits_{j = 1}^{{\xi _n}} {{V_{n,j}} - h\left( {{\xi _n}} \right)} } \right| \ge \epsilon\left| {{\xi _n}} \right.} \right) = 0\;\;{\rm{a}}{\rm{.s}}{\rm{.,}}$ |
则(a)与(b)成立.
证明 首先证明第1部分.令
$\begin{align} & {{U}_{n}}=\sum\limits_{j=1}^{{{\xi }_{n}}}{{{V}_{n,j}},V_{n,k}^{\left( 1 \right)}={{V}_{n,k}}-{{m}_{n,k}},} \\ & F_{n,k}^{\left( 1 \right)}\left( x \right)=P\left. \left( V_{n,k}^{\left( 1 \right)}\le x \right) \right),V_{n,k}^{\left( 2 \right)}=V_{n,k}^{\left( 1 \right)}{{I}_{\left\{ \left| V_{n,k}^{\left( 1 \right)} \right|\le 1 \right\}}}, \\ & {{V}_{n}}=\sum\limits_{j=1}^{{{\xi }_{n}}}{\left( {{m}_{n,j}}+EV_{n,j}^{\left( 2 \right)} \right),}U_{n}^{\left( 1 \right)}=\sum\limits_{j=1}^{{{\xi }_{n}}}{V_{n,j}^{\left( 1 \right)},} \\ & U_{n}^{\left( 2 \right)}=\sum\limits_{j=1}^{{{\xi }_{n}}}{V_{n,j}^{\left( 2 \right)},{{B}_{n}}=\left\{ U_{n}^{\left( 1 \right)}=U_{n}^{\left( 2 \right)} \right\}.} \\ \end{align}$ |
由于
$\begin{array}{l} P\left( {B_n^c \cap {\mathit{\Omega} _0}} \right)\\ = P\left( {\sum\limits_{j = 1}^{{\xi _n}} {V_{n,j}^{\left( 1 \right)} \ne \sum\limits_{j = 1}^{{\xi _n}} {V_{n,k}^{\left( 2 \right)},{\xi _n} = k,{\mathit{\Omega} _0}} } } \right)\\ = \sum\limits_{k = 1}^\infty {P\left( {{\xi _n} = k,{\mathit{\Omega} _0}} \right)P\left( {\sum\limits_{j = 1}^k {V_{n,j}^{\left( 1 \right)} \ne \sum\limits_{j = 1}^k {V_{n,j}^{\left( 2 \right)}} } } \right)} \\ \le \sum\limits_{k = 1}^\infty {P\left( {{\xi _n} = k,{\mathit{\Omega} _0}} \right)\sum\limits_{j = 1}^k {P\left( {V_{n,j}^{\left( 1 \right)} \ne V_{n,j}^{\left( 2 \right)}} \right)} } \\ = \sum\limits_{k = 1}^\infty {P\left( {{\xi _n} = k,{\mathit{\Omega} _0}} \right)\sum\limits_{j = 1}^k {P\left( {\left| {V_{n,j}^{\left( 1 \right)}} \right| > 1} \right)} } \\ = \sum\limits_{k = 1}^\infty {P\left( {{\xi _n} = k,{\mathit{\Omega} _0}} \right)\sum\limits_{j = 1}^k {\int_{\left| x \right| > 1} {{\rm{d}}{F_{n,j}}\left( {x + {m_{n,j}}} \right)} } }\\ = E\left( {\sum\limits_{j = 1}^{{\xi _n}} {\int_{\left| x \right| > 1} {{\rm{d}}{F_{n,j}}\left( {x + {m_{n,j}}} \right),{\mathit{\Omega} _0}} } } \right), \end{array}$ | (1) |
再利用控制收敛定理可得,
$\begin{array}{c} P\left( {\left| {{U_n} - {V_n}} \right| \ge \epsilon,{\mathit{\Omega} _0}} \right)\\ = P\left[ {\left\{ {\left| {{U_n} - {V_n}} \right| \ge \epsilon} \right\} \cap {B_n} \cap {\mathit{\Omega} _0}} \right] + \\ P\left[ {\left\{ {\left| {{U_n} - {V_n}} \right| \ge \epsilon} \right\} \cap B_n^c \cap {\mathit{\Omega} _0}} \right], \end{array}$ | (2) |
前面已证等式右边第2项收敛到0.对第1项利用Chebyshev不等式, 可以得到
$\begin{array}{l} P\left[ {\left\{ {\left| {{U_n} - {V_n}} \right| \ge \epsilon} \right\} \cap {B_n} \cap {\mathit{\Omega} _0}} \right]\\ = P\left[ {\left\{ {\left| {\sum\limits_{j = 1}^{{\xi _n}} {\left( {V_{n,k}^{\left( 2 \right)} - EV_{n,k}^{\left( 2 \right)}} \right)} } \right| \ge \epsilon} \right\} \cap {B_n} \cap {\mathit{\Omega} _0}} \right]\\ \le \sum\limits_{k = 1}^\infty {P\left( {{\xi _n} = k,{\mathit{\Omega} _0}} \right)P\left( {\left| {\sum\limits_{j = 1}^k {\left( {V_{n,k}^{\left( 2 \right)} - EV_{n,j}^{\left( 2 \right)}} \right)} } \right| \ge \epsilon} \right)} \\ \le \sum\limits_{k = 1}^\infty {P\left( {{\xi _n} = k,{\mathit{\Omega} _0}} \right)\frac{1}{{{\epsilon^2}}}\sum\limits_{j = 1}^k {{\rm{Var}}\left( {V_{n,k}^{\left( 2 \right)}} \right)} } \\ = \frac{1}{{{\epsilon^2}}}E\left( {\sum\limits_{k = 1}^{{\xi _n}} {{\rm{Var}}\left( {V_{n,j}^{\left( 2 \right)}} \right),{\mathit{\Omega} _0}} } \right)\\ = \frac{1}{{{\epsilon^2}}}E\left( {\sum\limits_{j = 1}^{{\xi _n}} {\int_{\left| x \right| \le 1} {{x^2}{\rm{d}}{F_{n,j}}\left( {x + {m_{n,j}}} \right),{\mathit{\Omega} _0}} } } \right), \end{array}$ |
因为在Ω0上
$\mathop {\lim }\limits_{n \to \infty } P\left( {\left| {{U_n} - {V_n}} \right| \ge \epsilon} \right) = 0.$ |
因此
$\sum\limits_{j=1}^{{{\xi }_{n}}}{{{V}_{n,j}}-{{V}_{n}}}\xrightarrow{P}0,n\to \infty .$ |
另一方面, 对于每个
$P\left( \left| \sum\limits_{j=1}^{{{\xi }_{n}}}{{{V}_{n,j}}-h\left( {{\xi }_{n}} \right)} \right|\ge \epsilon \left| {{\xi }_{n}} \right. \right)\to 0\ \ \text{a}\text{.s}\text{.}$ |
可得
$P\left( \sum\limits_{j=1}^{{{\xi }_{n}}}{{{V}_{n,j}}-h\left( {{\xi }_{n}} \right)\le x\left| {{\xi }_{n}} \right.} \right)\text{a}\text{.s}\text{.}\left\{ \begin{align} & 1,x>0, \\ & 0, < 0. \\ \end{align} \right.$ |
故任取T>0, 下式对u∈I=[-T, T]一致成立
$\mathop {\lim }\limits_{n \to \infty ]} E{\mkern 1mu} \left( {\exp \left( {{\rm{i}}u\left( {\sum\limits_{j = 1}^{{\xi _n}} {{V_{n,j}} - h\left( {{\xi _n}} \right)} } \right)} \right)\left| {{\xi _n}} \right.} \right)\left( \omega \right) = 1\;\;{\rm{a}}{\rm{.s}}{\rm{.}}$ |
而
$\begin{align} & E\left( \exp \left( \text{i}u\left( \sum\limits_{j=1}^{{{\xi }_{n}}}{Vn,j-h\left( {{\xi }_{n}} \right)} \right) \right)\left| {{\xi }_{n}} \right. \right) \\ & \ \ \ =\exp \left\{ -\text{i}uh\left( {{\xi }_{n}} \right) \right\}\prod\limits_{j=1}^{{{\xi }_{n}}}{{{f}_{n,j}}\left( u \right)}, \\ \end{align}$ |
其中,fn, j(u)为Vn, j的特征函数, 于是对于a.s.ω,
$\underset{n\to \infty }{\mathop{\lim }}\,\sum\limits_{j=1}^{{{\xi }_{n}}}{\log }\left( 1-\left( 1-{{\left| {{f}_{n,j}}\left( u \right) \right|}^{2}} \right) \right)=0\ \ \text{a}\text{.s}\text{.}$ | (3) |
根据命题2.1, 对于a.s.ω,
$\begin{align} & \text{-}\sum\limits_{j=1}^{{{\xi }_{n}}}{\log \left( 1-\left( 1-{{\left| {{f}_{n,j}}\left( u \right) \right|}^{2}} \right) \right)\ge \sum\limits_{j=1}^{{{\xi }_{n}}}{\left( 1-{{\left| {{f}_{n,j}}\left( u \right) \right|}^{2}} \right)}} \\ & \ \ \ge 0,\forall u\in I,n\ge {{n}_{0}}={{n}_{0}}\left( \omega \right), \\ \end{align}$ |
利用式(3)有:对于a.s.ω,
$\underset{n\to \infty }{\mathop{\lim }}\,\int_{-1}^{1}{\sum\limits_{j=1}^{{{\xi }_{n}}}{\left( 1-{{\left| {{f}_{n,j}}\left( u \right) \right|}^{2}} \right)\text{d}u=0\ \ \ \text{a}\text{.s}\text{.}}}$ | (4) |
令
$\int_{-1}^{1}{\left( 1-{{\left| {{f}_{n,j}}\left( u \right) \right|}^{2}} \right)}\text{d}u=2\int_{-\infty }^{\infty }{\left( 1-\frac{\sin \ x}{x} \right)\text{d}\ {{G}_{n,j}}\left( x \right).}$ |
接下来分析一些简单的估计, 首先存在正常数K1, 使得|x|>1时,
$1-\frac{\sin x}{x}=1-\frac{1}{x}\left( x-\frac{{{x}^{3}}}{3!}+\frac{{{x}^{5}}}{5!}-+\cdots \right)\ge {{K}_{2}}{{x}^{2}}.$ |
令K3=min{K1, K2}>0, 令
$\begin{align} & \int\limits_{-1}^{1}{\sum\limits_{j=1}^{{{\xi }_{n}}}{\left( 1-{{\left| {{f}_{n,j}}\left( u \right) \right|}^{2}} \right)\text{d}u}} \\ & \ge {{K}_{3}}\left\{ \sum\limits_{j=1}^{{{\xi }_{n}}}{\int_{\left| x \right|\le 1}{{{x}^{2}}\text{d}{{G}_{n,j}}\left( x \right)+\sum\limits_{j=1}^{{{\xi }_{n}}}{\int_{\left| x \right|>1}{\text{d}{{G}_{n,j}}\left( x \right)}}}} \right\} \\ & ={{K}_{3}}\sum\limits_{j=1}^{{{\xi }_{n}}}{E\psi \left( {{{\bar{V}}}_{n,j}}-{{V}_{n,j}} \right)\ge \frac{{{K}_{3}}}{2}\sum\limits_{j=1}^{{{\xi }_{n}}}{E\psi \left( {{{\bar{V}}}_{n,j}}-{{m}_{n,j}} \right)}} \\ & =\frac{{{K}_{3}}}{2}\sum\limits_{j=1}^{{{\xi }_{n}}}{\left\{ \int_{\left| x \right|>1}{\text{d}{{F}_{n,j}}\left( x+{{m}_{n,j}} \right)}+ \right.} \\ & \left. \int_{\left| x \right|\le 1}{{{x}^{2}}\text{d}{{F}_{n,j}}\left( x+{{m}_{n,j}} \right)} \right\}\ge 0. \\ \end{align}$ |
由于括号内两式均非负, 对两边取极限得到(a)和(b).
下面讨论另一种情形.仍然沿用前面的符号.设{Zn:n≥0}为定义1.1所定义的变化环境中的上临界的分枝过程, 记
定理2.2 假设Zn→∞a.s., 当n→∞, 若下面两个条件成立:
(a′)
(b′)
则存在随机变量序列
$\mathop {\lim }\limits_{n \to \infty } P\left( {\left| {\sum\limits_{j = 1}^{{Z_n}} {{Y_{n,j}} - {H_n}} } \right| \ge \epsilon\left| {{Z_n}} \right.} \right) = 0\;\;\;{\rm{a}}.{\rm{s}}.,$ |
进而
$\sum\limits_{j=1}^{{{Z}_{n}}}{{{Y}_{n,j}}-{{H}_{n}}\xrightarrow{P}0,}为n\to \infty .$ |
证明 首先令
$\begin{align} & Y_{n,k}^{\left( 1 \right)}={{Y}_{n,k}}-{{v}_{n,k}},F_{n,k}^{\left( 1 \right)}\left( x \right)=P\left( Y_{n,k}^{(1)}\le x\left| {{Z}_{n}} \right. \right), \\ & Y_{n,k}^{\left( 2 \right)}=Y_{n,k}^{\left( 1 \right)}{{I}_{\left\{ \left| Y_{n,k}^{\left( 1 \right)} \right|\le 1 \right\}}},{{H}_{n}}=\sum\limits_{j=1}^{{{Z}_{n}}}{\left( {{v}_{n,j}}+EY_{n,j}^{\left( 2 \right)} \right),} \\ & {{Y}_{n}}=\sum\limits_{j=1}^{{{Z}_{n}}}{{{Y}_{n,j}},Y_{n}^{\left( 1 \right)}=\sum\limits_{j=1}^{{{Z}_{n}}}{Y_{n,j}^{\left( 1 \right)}},} \\ & Y_{n}^{\left( 2 \right)}=\sum\limits_{j=1}^{{{Z}_{n}}}{Y_{n,j}^{\left( 2 \right)},{{C}_{n}}=\left\{ Y_{n}^{\left( 1 \right)}=Y_{n}^{\left( 2 \right)} \right\}.} \\ \end{align}$ |
注意到
$\begin{align} & P\left( C_{n}^{c}\left| {{Z}_{n}} \right. \right)=P\left( Y_{n}^{\left( 1 \right)}\ne Y_{n}^{\left( 2 \right)}\left| {{Z}_{n}} \right. \right), \\ & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ =P\left( \sum\limits_{j=1}^{{{Z}_{n}}}{Y_{n,j}^{\left( 1 \right)}\ne \sum\limits_{j=1}^{{{Z}_{n}}}{Y_{n,j}^{\left( 2 \right)}\left| {{Z}_{n}} \right.}} \right), \\ & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \le \sum\limits_{k=1}^{{{Z}_{n}}}{P\left( Y_{n,j}^{\left( 1 \right)}\ne Y_{n,j}^{\left( 2 \right)}\left| {{Z}_{n}} \right. \right),} \\ & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ =\sum\limits_{j=1}^{{{Z}_{n}}}{\int_{\left| x \right|>1}{\text{d}{{F}_{n,j}}\left( x+{{v}_{n,j}} \right),}} \\ \end{align}$ |
根据(a′)知,
$\begin{align} & P\left( \left\{ \left| \sum\limits_{k=1}^{{{Z}_{n}}}{{{Y}_{n,j}}-{{H}_{n}}} \right|\ge \epsilon \right\},{{C}_{n}}\left| {{Z}_{n}} \right. \right) \\ & =P\left( \left| \sum\limits_{j=1}^{{{Z}_{n}}}{\left( Y_{n,j}^{\left( 2 \right)}-EY_{n,j}^{\left( 2 \right)} \right)\left| \ge \epsilon,{{C}_{n}} \right.} \right|{{Z}_{n}} \right) \\ & \le P\left( \left| \sum\limits_{j=1}^{{{Z}_{n}}}{\left( Y_{n,j}^{\left( 2 \right)}-EY_{n,j}^{\left( 2 \right)} \right)} \right|\ge \epsilon\left| {{Z}_{n}} \right. \right) \\ & \le \sum\limits_{k=1}^{{{Z}_{n}}}{P\left( \left| Y_{n,j}^{\left( 2 \right)}-EY_{n,j}^{\left( 2 \right)} \right|\ge \epsilon\left| {{Z}_{n}} \right. \right)} \\ & =\sum\limits_{k=1}^{\infty }{\frac{1}{{{\epsilon}^{2}}}\text{Var}\left( Y_{n,j}^{\left( 2 \right)}\left| {{Z}_{n}} \right. \right)} \\ & \le \sum\limits_{k=1}^{\infty }{\frac{1}{{{\epsilon}^{2}}}E\left( {{\left| Y_{n,j}^{\left( 2 \right)} \right|}^{2}}\left| {{Z}_{n}} \right. \right)} \\ & =\frac{1}{{{\epsilon}^{2}}}\sum\limits_{k=1}^{\infty }{\int_{\left| x \right|\le 1}{{{x}^{2}}\text{d}{{F}_{n,j}}\left( x+{{v}_{n,j}} \right),}} \\ \end{align}$ |
由(b′),
$\mathop {\lim }\limits_{n \to \infty } {\mkern 1mu} P\left( {\left| {\sum\limits_{j = 1}^{{Z_n}} {{Y_{n,j}} - {H_n}} } \right| \ge \epsilon\left| {{Z_n}} \right.} \right) = 0\;\;\;{\rm{a}}{\rm{.s}}{\rm{.}}$ |
利用控制收敛定理得
$\sum\limits_{j=1}^{{{Z}_{n}}}{{{Y}_{n,j}}-{{H}_{n}}\xrightarrow{P}0,当n\to \infty .}$ |
[1] | Gendenko B V, Kolmogorov A N. Limit distributions for sums of independent random variables[M]. Chung K L, trans. Cambridge:Addison-Wesley, 1954. |
[2] | Tucker H G. A graduate course in probability[M]. New York and London: Academic Press, 1967 . |
[3] | 胡迪鹤. 不变原理及其在分枝过程中的应用[J]. 北京大学学报:自然科学版 , 1964 (1) :1–27. |
[4] | Heyde C C. Some central limit analogues for supercritical Galton-Watson processes[J]. Appl Prob , 1971, 8 (1) :52–59. |
[5] | 高振龙.随机环境中Galton-Watson分枝过程的极限定理[D].北京:中国科学院研究生院, 2011. |
[6] | Gao Z L, Hu X Y. Limit theorems for a Galton-Watson process in the i.i.d. random environment[J]. Acta Mathematica Scientia, 2012, 32(3):1193-1205. |
[7] | 胡迪鹤. 分析概率论[M]. 北京: 科学出版社, 1997 . |
[8] | Athreya K B, Ney P E. Branching processes[M]. Berlin, Heidelberg, New York: Springer-Verlag, 1972 . |