南晓杰
![通讯作者](http://html.rhhz.net/ZGKXYDXXB/images/REcor.gif)
![nanxiaojie12@mails.ucas.ac.cn](http://html.rhhz.net/ZGKXYDXXB/images/REemail.gif)
中国科学院大学数学科学学院, 北京 100049
摘要: 借助Berry-Esseen引理和Asmussen对条件Borel-Cantelli 引理的重要推广, 在变化环境中上临界分枝过程的每一代每一个个体的后代个体总数的 2 阶矩有一致上下界的情况下, 得到变化环境中分枝过程的重对数律, 从而改进了在相应的2+
δ阶矩有限条件下的证明.
关键词: 变化环境分枝过程重对数律
The law of the iterated logarithm(abbr.LIL) of the classical Galton-Watson process was firstly proved by Heyde
[1] under the condition that the 2+
δ moment of the process is finite. In the same year,Heyde and Leslie
[2] again obtained the LIL under the condition that the second moment is finite. Later,Asmussen
[3] gave another proof via a very delicate truncation procedure and Kronecker lemma. The proof of Huggins
[4] is based on the Skorohod embedding techniques and new properties of Brownian motion and stopping times.
Gao
[5] proved the LIL of the super-critical Galton-Watson processes in varying environment satisfied that there is a uniform upper bound for the 2+
δ moment of the number of the offspring of each individual of each generation. In addition,the author pointed out a mistake in the proof of Theorem 1 in Heyde and Leslie
[2]. Enlightened by the proof of the LIL of the classical Galton-Watson process in Asmussen
[3],we obtain the LIL of the super-critical Galton-Watson processes in varying environment under the condition that the second moment has a uniform upper bound and a uniform lower bound.
1 Main resultLet
Z0≡1 and for all
n≥1,define
${{Z}_{n+1}}=\left\{ \begin{array}{*{35}{l}} \sum\limits_{j=1}^{{{Z}_{n}}}{{{X}_{n,j}}}, & \text{if }{{Z}_{n}}\ne 0; \\ 0, & \text{if }{{Z}_{n}}=0, \\\end{array} \right.$ |
where {
Xn,j;
n≥0,
j≥1} are independent and for each
n≥0,{
Xn,j,j≥1} have the same distribution {
pn(k),k∈
N }.
pn(k) denotes the probability of
k offspring produced by an individual of the
n’th generation and N is the set of non-negative integers. Then {
Zn,n≥0} is said to be a Galton-Watson process in varying environment(GWVE).
Let the generating functions of
Zn and
Xn,j are respectively
fn(s) and
$\phi $n(s),and let
mn and
μn are respectively their expectations. Then
$\begin{align} & {{f}_{n}}\left( s \right)={{\phi }_{0}}\left( {{\phi }_{1}}\left( \cdots {{\phi }_{n-1}}\left( s \right)\cdots \right) \right), \\ & {{m}_{n}}=f{{'}_{n}}\left( 1 \right)=\prod\limits_{i=0}^{n-1}{\phi {{'}_{i}}\left( 1 \right)}=\prod\limits_{i=0}^{n-1}{{{\mu }_{i}}.} \\ \end{align}$ |
From now on,we always assume that ∏
k=nn-1μk=1,0<m
n<∞,
$\forall $n≥0. It is known that {
Wn:=Zn/mn,n≥0} is a nonnegative martingale and there exists a nonnegative random variable W so that lim
n→∞Wn=
W a.s.. Moreover,if sup
nE (
Wn2)<∞,then E
W=1 and
σ2:=Var(
W)=∑
n=0∞δn2/(
μn2mn)<∞. These results can be found in Fearn
[6].
Lemma 1.1 (Decomposition Lemma 1) Let {
Zn,n≥0} be a GWVE,then
$\forall $n≥0,
r≥1 we have
${{Z}_{n+r}}-{{m}_{n,r}}{{Z}_{n}}=\left\{ \begin{array}{*{35}{l}} \sum\limits_{j=1}^{{{Z}_{n}}}{\left( Z_{n,r}^{\left( j \right)}-{{m}_{n,r}} \right)}, & \text{if }{{Z}_{n}}\ne 0; \\ 0, & \text{if }{{Z}_{n}}=0, \\\end{array} \right.$ |
where
Zn,r(j) represents the number of
r’th generation offspring of the j’th of the Z
n individuals of the n’th generation,and {
Zn,r(j),j≥1} are independent and identically distributed and independent of
Zn. Furthermore,
$\begin{align} & {{m}_{n,r}}:=\text{E}\left( Z_{n,r}^{\left( j \right)} \right)=\prod\limits_{j=n}^{n+r-1}{{{\mu }_{j}},} \\ & \sigma _{n,r}^{2}:=\text{Var}\left( Z_{n,r}^{\left( j \right)} \right)={{\left( {{m}_{n,r}} \right)}^{2}}\sum\limits_{j=n}^{n+r-1}{\frac{\delta _{j}^{2}}{\mu _{j}^{2}{{m}_{n,j-n}}}}. \\ \end{align}$ |
Proof See Ref.
[5].
Remark 1.1 Zn,1(j)=
Xn,j,
m0,r=
mr,
mn,1=μn,
mn,0=1,
σ0,r2=Var(
Zr)and
σn,12=
δn2.
Lemma 1.2(Decomposition Lemma 2) Let {
Zn,n≥0} be a GWVE,then
$\forall $n≥0
${{m}_{n}}W-{{Z}_{n}}=\left\{ \begin{array}{*{35}{l}} \sum\limits_{j=1}^{{{Z}_{n}}}{\left( W_{n}^{\left( j \right)}-1 \right)}, & \text{if }{{Z}_{n}}\ne 0; \\ 0, & \text{if }{{Z}_{n}}=0, \\\end{array} \right.$ |
where {
Wn(j),j≥1} are independent and identically distributed and independent of
Zn. If
$\sum\limits_{j=n}^{\infty }{\frac{\delta _{j}^{2}}{\mu _{j}^{2}\prod\nolimits_{k=n}^{j-1}{{{\mu }_{k}}}}}<\infty ,\forall n\ge 0,$ | (1) |
then
$\text{E}\left( W_{n}^{\left( j \right)} \right)=1\text{ and }\sigma _{j}^{2}:=\text{Var}\left( W_{n}^{\left( j \right)} \right)=\sum\limits_{j=n}^{\infty }{\frac{\delta _{j}^{2}}{\mu _{j}^{2}{{m}_{n,j-n}}}}.$ |
Proof See Ref.
[5].
Remark 1.2 W
0(1)=
W,
σ02=
σ2=Var(
W)=
$\sum\limits_{j=0}^{\infty }{\delta _{j}^{2}/\left( \mu _{j}^{2}{{m}_{j}} \right)}$.
Now assume that there exist four constants
α,β,τ,γ with
β>
α>1 and
τ>
γ>0 such that
$\forall $n≥0
$\alpha \le {{\mu }_{n}}\le \beta ,{{\gamma }^{2}}\le \delta _{n}^{2}\le {{\tau }^{2}},$ | (2) |
$\sum\limits_{n=0}^{\infty }{\int_{\left| y \right|>{{\zeta }^{n}}}{{{y}^{2}}\text{d}{{F}_{n}}\left( y \right)}}<\infty ,$ | (3) |
$\sum\limits_{n=0}^{\infty }{\int_{\left| y \right|>{{\zeta }^{n}}}{{{y}^{2}}\text{d}{{G}_{n}}\left( y \right)}}<\infty ,$ | (4) |
where 1<
ζ<
α1/4,and
Fn is the distribution of
Zn,r(j)-
mn,r in Decomposition Lemma 1,and
Gn is the distribution of
Wn(j)-1 in Decomposition Lemma 2.
For any given
r≥1,define
Yn,j:=
Zn,r(j)-
mn,r,
Y′n,j:=
Yn,jI(|
Yn,j|≤
$\sqrt{{{m}_{n}}}$),
Vn,j:=Wn(j)-1,
V′n,j:=
Vn,jI(|
Vn,j|≤
$\sqrt{{{m}_{n}}}$).
Theorem 1.1 Let {
Zn,n≥0} be a GWVE. Suppose that
pn(0)=0,
$\forall $n≥0. If the conditions (2),(3),and (4) are satisfied,and
$\text{Var}(Y{{\prime }_{n,j}})/\text{Var}({{Y}_{n,j}})\to 1,\text{as }n\to \infty ,$ | (5) |
$\text{Var}(V{{\prime }_{n,j}})/\text{Var}({{V}_{n,j}})\to 1,\text{as }n\to \infty ,$ | (6) |
then for all
r≥1,with probability one we have
$\underset{n\to \infty }{\mathop{\lim \sup }}\,(\underset{n\to \infty }{\mathop{\text{lim }\!\!~\!\!\text{ inf}}}\,)\frac{{{Z}_{n+r}}-{{m}_{n,r}}{{Z}_{n}}}{{{(2\sigma _{n,r}^{2}{{Z}_{n}}\text{log}n)}^{1/2}}}=1\left( -1 \right),$ | (7) |
$\underset{n\to \infty }{\mathop{\lim \sup }}\,(\underset{n\to \infty }{\mathop{\text{lim }\!\!~\!\!\text{ inf}}}\,)\frac{{{m}_{n}}W-{{Z}_{n}}}{{{(2\sigma _{n}^{2}{{Z}_{n}}\text{log}n)}^{1/2}}}=1\left( -1 \right).$ | (8) |
Remark 1.3 Since
pn(0)>0,
$\forall $n≥0,one has
W>0 a.s.,hence
Zn=O(
mn)a.s..
Remark 1.4 We can obtain
αn≤
mn≤
βn and Eq.(1) from the condition (2). According Remark 1.3,we know that log
Zn-
nlog
m→log
W a.s., which means loglog
Zn/log
n→1a.s., so log
n can be substituted by log log
Zn in Eq.(7) and Eq.(8).
$\begin{align} & \int_{\left| y \right|>{{\zeta }^{n}}}{{{\left| y \right|}^{2}}\text{d}{{F}_{n}}\left( y \right)}\le \frac{1}{{{\left( \log {{\zeta }^{n}} \right)}^{1+\delta }}}\int_{\left| y \right|>{{\zeta }^{n}}}{{{\left| y \right|}^{2}}{{\left( \log \left| y \right| \right)}^{\text{1+}\delta }}\text{d}{{F}_{n}}\left( y \right)}, \\ & \int_{\left| y \right|>{{\zeta }^{n}}}{{{\left| y \right|}^{2}}\text{d}{{G}_{n}}\left( y \right)}\le \frac{1}{{{\left( \log {{\zeta }^{n}} \right)}^{1+\delta }}}\int_{\left| y \right|>{{\zeta }^{n}}}{{{\left| y \right|}^{2}}{{\left( \log \left| y \right| \right)}^{\text{1+}\delta }}\text{d}{{G}_{n}}\left( y \right)}, \\ \end{align}$ |
where 0<
δ<1,we can get the conditions (3) and (4) under the following conditions (9) and (10) are satisfied:
$\underset{n\ge 0}{\mathop{\text{sup}}}\,\int_{y\in R}{{{\left| y \right|}^{2}}{{(log\left| y \right|)}^{1+\delta }}\text{d}{{F}_{n}}\left( y \right)<\infty },$ | (9) |
$\underset{n\ge 0}{\mathop{\text{sup}}}\,\int_{y\in R}{{{\left| y \right|}^{2}}{{(log\left| y \right|)}^{1+\delta }}\text{d}{{G}_{n}}\left( y \right)<\infty },$ | (10) |
However (9) and (10) are weaker than (1.14) in Ref.
[5].
Remark 1.6 The condition (5) holds naturally for a classical super-critical Galton-Watson branching process {
Zn,n≥0} with E(
Z1log
Z1)<∞. Moreover,if there exists a random variable
Y∈
L2(
Ω,
F , P) so that |
Yn,1|≤
Y,then Eq.(5) can be deduced. Since ∑
nP(|
Yn,1|>
$\sqrt{{{m}_{n}}}$)<∞,we almost surely have
$Y{{\prime }_{n,1}}-{{Y}_{n,1}}\to 0~\text{and}~{{(Y{{\prime }_{n,1}})}^{2}}-{{({{Y}_{n,1}})}^{2}}\to 0.$ |
By the dominated convergence theorem,we have
$\text{Var}(Y{{\prime }_{n,1}})-\text{Var}({{Y}_{n,1}})=\text{Var}(Y{{\prime }_{n,j}})-\text{Var}({{Y}_{n,j}})\to 0~\text{a}\text{.s}.$ |
hence the condition (5) holds. For Eq.(6) we have similar results.
2 Basic lemmas
Lemma 2.1 Let {
Fn,n≥0} be an increasing sequence of
σ-algebras and {
Tn,n≥0} a (not necessarily adapted) random variable sequence such that
$\sum\limits_{n=0}^{\infty }{{{\Delta }_{n}}:=}\sum\limits_{n=0}^{\infty }{\underset{y\in R}{\mathop{\sup }}\,\left| \text{P}\left( {{T}_{n}}\le y|{{F}_{n}} \right)-\Phi \left( y \right) \right|}<\infty ,$ |
where
Φ(y) is the distribution function of
N(0,1). Then
$\underset{n\to \infty }{\mathop{\lim \sup }}\,\frac{{{T}_{n}}}{{{\left( 2\log n \right)}^{1/2}}}\le 1\text{a}\text{.s}.,$ |
with the inequality replaced by equality if
Tn is measurable with respect to
Fn+k for some
1≤k<∞.
Proof See Ref.
[3].
Lemma 2.2 (Berry-Esseen Lemma)
Let{
Xn,n≥1} be an independent and identically distributed random variable sequence such that E
Xn=0,
E Xn2=
σ2>0 and E |
Xn|
3<∞. Denote
Sn:=∑
k=1nXk. Then
$\underset{x\in \text{R}}{\mathop{\text{sup}}}\,\left| \text{P}\left( \frac{{{S}_{n}}}{\sigma \sqrt{n}}<x \right)-\Phi \left( x \right) \right|\le A\frac{\text{E}{{\left| {{X}_{1}} \right|}^{3}}}{{{\sigma }^{3}}\sqrt{n}},$ |
where
Φ(x) is the standard normal distribution and
A is a positive constant that is called the Berry-Esseen constant.
Proof See Ref.
[7],P124.
Lemma 2.3 (Kronecker Lemma)Let {
bn} be an increasing sequence of positive real numbers with
bn→∞,and let {
xn} be a sequence of real numbers with ∑
n=1∞xn=
x(finite). Then
$\frac{1}{{{b}_{n}}}\sum\limits_{j=1}^{n}{{{b}_{j}}{{x}_{j}}}\to 0,\text{as }n\to \infty .$ |
Proof See Ref.
[7],P63.
3 Proof of Theorem 1.1
Proof Denote
F0:=
σ(
Z0) and
Fn:=
σ{
Xk,j;0≤
k≤
n-1,
j≥1}. Then
$\forall $n≥1,
Fn is the
σ-algebra generated by the individuals of previous
n-1 generations. First prove Eq.(7). We only need to show
$\underset{n\to \infty }{\mathop{\lim ~\sup }}\,\frac{\overline{{{Z}_{n+r}}}-{{m}_{n,r}}\overline{{{Z}_{n}}}}{{{(2{{\sigma }^{2}}_{n,r}{{Z}_{n}}logn)}^{1/2}}}=1~a.s..$ | (11) |
In fact,if Eq.(11) is true,let
Zn=-
Zn,n≥0,then we have
$\underset{n\to \infty }{\mathop{\lim ~\sup }}\,\frac{{{Z}_{n+r}}-{{m}_{n,r}}{{Z}_{n}}}{{{(2{{\sigma }^{2}}_{n,r}{{Z}_{n}}logn)}^{1/2}}}=1~a.s.,$ | (12) |
which in fact is
$\underset{n\to \infty }{\mathop{\lim \inf }}\,\frac{{{Z}_{n+r}}-{{m}_{n,r}}{{Z}_{n}}}{{{\left( 2\sigma _{n,r}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}=-1\text{ a}\text{.s}..$ |
Define
$\begin{align} & \widetilde{{{Y}_{n,j}}}:=Y{{'}_{n,j}}-\text{E}Y{{'}_{n,j}}, \\ & \widetilde{{{S}_{n}}}:=\sum\limits_{j=1}^{{{Z}_{n}}}{\widetilde{{{Y}_{n,j}}}}, \\ & {{\widetilde{{{\omega }_{n}}}}^{2}}:\text{=Var}\left( \widetilde{{{S}_{n}}}|{{F}_{n}} \right)={{Z}_{n}}\text{Var}\left( \widetilde{{{Y}_{n,j}}} \right), \\ & {{T}_{n}}:=\widetilde{{{S}_{n}}}/\widetilde{{{\omega }_{n}}}. \\ \end{align}$ |
By a standard moment inequality,
$\begin{align} & \text{E}\left( {{\left| \text{ }\widetilde{{{Y}_{n,j}}^{\prime }} \right|}^{3}} \right)\le \\ & \text{E}\left( {{\left| {{Y}^{\prime }}_{n,j} \right|}^{3}} \right)+3\text{E}\left( \left| {{Y}^{\prime }}_{n,j} \right| \right){{\left( \text{E}\left| {{Y}^{\prime }}_{n,j} \right| \right)}^{2}}+ \\ & 3\text{E}\left( \left| {{Y}^{\prime }}_{n,j} \right| \right)\text{E}\left( {{Y}^{'}}_{n,j}^{2} \right)+\text{E}\left( {{\left| {{Y}^{\prime }}_{n,j} \right|}^{3}} \right) \\ & \le 8\text{E}\left( {{\left| {{Y}^{\prime }}_{n,j} \right|}^{3}} \right)=8\int_{\left| y \right|\le \sqrt{{{m}_{n}}}}{{{\left| y \right|}^{\text{3}}}\text{d}{{F}_{n}}\left( y \right)}. \\ \end{align}$ |
Letting A be the Berry-Essen constant,by the Berry-Esseen Lemma we have
$\begin{align} & {{\Delta }_{n}}:=\underset{y\in \mathbb{R}}{\mathop{\text{sup}}}\,\left| \mathbb{P}({{T}_{n}}\le y \right|{{F}_{n}})-\Phi \left( y \right)| \\ & \le 8A\frac{{{Z}_{n}}}{\widetilde{{{\omega }_{n}}^{3}}}\int_{\left| y \right|\le {{m}_{n}}}{{{\left| y \right|}^{3}}}d{{F}_{n}}\left( y \right). \\ \end{align}$ | (13) |
By the condition (2) we can deduce that there exist a uniform upper and a uniform lower bound only dependent on r for
σn,r2. So there exist positive and finite constants
C1 and
C2 which are only dependent on r such that
${{C}_{1}}\le \underset{n\to \infty }{\mathop{\lim ~ \text{inf}}}\,\frac{{{Z}_{n}}}{\widetilde{{{\omega }_{n}}^{2}}}\le \underset{n\to \infty }{\mathop{\lim ~\sup }}\,\frac{{{Z}_{n}}}{\widetilde{{{\omega }_{n}}^{2}}}\le {{C}_{2}}.$ | (14) |
From Remark 1.3 we know that
Zn=O(m
n)a.s.,hence
${{\widetilde{{{\omega }_{n}}}}^{2}}$=O(
mn)a.s..Combining with Eq.(13) and Eq.(14),we have
$\begin{align} & \sum\limits_{n=0}^{\infty }{\frac{1}{\sqrt{{{m}_{n}}}}}{{\int }_{\left| y \right|\le \sqrt{{{m}_{n}}}}}{{\left| y \right|}^{3}}d{{F}_{n}}\left( y \right) \\ & =\sum\limits_{n=0}^{\infty }{\frac{1}{\sqrt{{{m}_{n}}}}}\left( \sum\limits_{j=1}^{{{\xi }^{n}}}{{{\int }_{j-1 <\left| y \right|\le j}}}{{\left| y \right|}^{3}}d{{F}_{n}}{{\left( y \right)}^{n}} \right)+ \\ & \sum\limits_{n=0}^{\infty }{\frac{1}{\sqrt{{{m}_{n}}}}}\left( \sum\limits_{j={{\zeta }^{n}}+1}^{\sqrt{{{m}_{n}}}}{{{\int }_{j-1<\left| y \right|\le j}}}{{\left| y \right|}^{3}}d{{F}_{n}}\left( y \right) \right) \\ & \le \sum\limits_{n=0}^{\infty }{\sum\limits_{j=1}^{{{\xi }^{n}}}{{{\int }_{j-1<\left| y \right|\le j}}}}{{\left| y \right|}^{2}}d{{F}_{n}}\left( y \right)+ \\ & \sum\limits_{n=0}^{\infty }{\sum\limits_{j={{\zeta }^{n}}+1}^{\sqrt{{{m}_{n}}}}{{{\int }_{j-1<\left| y \right|\le j}}}}{{\left| y \right|}^{2}}d{{F}_{n}}\left( y \right)) \\ & \le \sum\limits_{n=0}^{\infty }{{{C}_{3}}}{{({{\zeta }^{2}}/\sqrt{\alpha })}^{n}}+\sum\limits_{n=0}^{\infty }{{{C}_{3}}}{{\int }_{\left| y \right|>{{\zeta }^{n}}}}{{\left| y \right|}^{2}}d{{F}_{n}}\left( y \right), \\ \end{align}$ | (15) |
where
C3>0 is a constant. By using the condition (3),Eq.(13) and Eq.(15) we have ∑△
n<∞ a.s.. Again applying Lemma 2.1 one eventually has
$\underset{n\to \infty }{\mathop{\lim \sup }}\,\frac{{{T}_{n}}}{{{\left( 2\log n \right)}^{1/2}}}\le 1,\text{a}\text{.s}..$ |
In addition,since
Tn is measurable with respect to
Fn+r,the above inequality should be replaced by equality.
${{S}_{n}}=\sum\limits_{j=1}^{{{Z}_{n}}}{{{Y}_{n,j}}}=\sum\limits_{j=1}^{{{Z}_{n}}}{\left\{ \widetilde{{{Y}_{n,j}}}+{{Y}_{n,j}}-Y{{'}_{n,j}}+\text{E}Y{{'}_{n,j}} \right\}}.$ |
Thus it suffices to verify
$\underset{n\to \infty }{\mathop{\lim \sup }}\,\frac{{{S}_{n}}}{{{\left( 2\sigma _{n,r}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}=\underset{n\to \infty }{\mathop{\lim \sup }}\,\frac{{{T}_{n}}}{{{\left( 2\log n \right)}^{1/2}}}.$ |
Noting that
$\begin{align} & \frac{{{S}_{n}}}{{{\left( 2\sigma _{n,r}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}=\frac{{{T}_{n}}}{{{\left( 2\log n \right)}^{1/2}}}{{\left( \frac{\text{Var}(Y{{\prime }_{n,j}})}{\sigma _{n,r}^{2}} \right)}^{1/2}}+ \\ & \frac{\sum\nolimits_{j=1}^{{{Z}_{n}}}{\left\{ {{Y}_{n,j}}-Y{{\prime }_{n,j}} \right\}}}{{{\left( 2\sigma _{n,r}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}+\frac{\sum\nolimits_{j=1}^{{{Z}_{n}}}{\text{E}Y{{\prime }_{n,j}}}}{{{\left( 2\sigma _{n,r}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}, \\ \end{align}$ |
it suffices to verify that
$Var(Y{{\prime }_{n,j}})/{{\sigma }^{2}}_{n,r}\to 1,$ | (16) |
$\frac{\sum _{j=1}^{{{Z}_{n}}}\{{{Y}_{n,j}}-Y{{\prime }_{n,j}}\}}{{{({{Z}_{n}}logn)}^{1/2}}}\to 0,$ | (17) |
$\frac{\sum _{j=1}^{{{Z}_{n}}}\mathbb{E}Y{{\prime }_{n,j}}}{{{({{Z}_{n}}logn)}^{1/2}}}\to 0.$ | (18) |
By the condition (5) we know that Eq.(16) holds. For Eq.(17) and Eq.(18),by Kronecer Lemma it only needs to prove that
$\sum\limits_{n=0}^{\infty }{\frac{1}{\sqrt{{{m}_{n}}logn}}}\sum\limits_{j=1}^{{{Z}_{n}}}{\left| {{Y}_{n,j}}-Y{{\prime }_{n,j}} \right|}<\infty ,$ | (19) |
$\sum\limits_{n=0}^{\infty }{\frac{1}{\sqrt{{{m}_{n}}logn}}}\sum\limits_{j=1}^{{{Z}_{n}}}{\left| \mathbb{E}Y{{\prime }_{n,j}} \right|}<\infty .$ | (20) |
Since
$\begin{align} & \left| \text{E}Y{{'}_{n,j}} \right|=\left| \text{E}\left( Y{{'}_{n,j}}-{{Y}_{n,j}} \right) \right|\le \text{E}\left| Y{{'}_{n,j}}-{{Y}_{n,j}} \right| \\ & =\text{E}\left| {{Y}_{n,j}} \right|I\left( \left| {{Y}_{n,j}} \right| \right)>\sqrt{{{m}_{n}}}=\int_{\left| y \right|\sqrt{{{m}_{n}}}}{\left| y \right|}\text{d}{{F}_{n}}\left( y \right), \\ \end{align}$ |
and noting that
Zn=O(
mn)a.s., it suffices for Eq.(19) and Eq.(20) that
$\sum\limits_{n=0}^{\infty }{\frac{{{m}_{n}}}{\sqrt{{{m}_{n}}logn}}}{{\int }_{\left| y \right|>\sqrt{{{m}_{n}}}}}\left| y \right|d{{F}_{n}}\left( y \right)<\infty ,$ | (21) |
(for the first,taking the mean). And Eq.(21) certainly holds since even
$\begin{align} & \sum\limits_{n=0}^{\infty }{\sqrt{{{m}_{n}}}}\int_{\left| y \right|\sqrt{{{m}_{n}}}}{\left| y \right|}\text{d}{{F}_{n}}\left( y \right)\le \sum\limits_{n=0}^{\infty }{\int_{\left| y \right|\sqrt{{{m}_{n}}}}{{{y}^{2}}}\text{d}{{F}_{n}}\left( y \right)} \\ & \le \sum\limits_{n=0}^{\infty }{\int_{\left| y \right|>{{\xi }^{2n}}}{{{\left| y \right|}^{2}}}\text{d}{{F}_{n}}\left( y \right)}<\infty . \\ \end{align}$ |
Therefore,Eq.(17) and Eq.(18) hold.
In order to prove Eq.(8),recall
Vn,j:=
Wn(j)-1. Repeating the proof of Eq.(11) and noting that there doesn’t exist 1≤
r<∞ so that
Tn is measurable with respect to
Fn+r,we have
$\underset{n\to \infty }{\mathop{\lim ~\sup }}\,$ | (22) |
Note that
$\begin{align} & \frac{{{m}_{n}}\left( W-{{W}_{n}} \right)}{{{\left( 2\sigma _{n}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}=\frac{{{m}_{n+k}}\left( W-{{W}_{n+k}} \right)}{2\sigma _{n+k}^{2}{{Z}_{n+k}}\log \left( n+k \right)}\cdot \\ & {{\left( \frac{\sigma _{n+k}^{2}{{Z}_{n+k}}\log \left( n+k \right)}{\sigma _{n}^{2}{{Z}_{n}}\log n} \right)}^{1/2}}\frac{1}{{{m}_{n+k}}}+ \\ & \frac{{{m}_{n+k}}\left( {{W}_{n+k}}-{{W}_{n}} \right)}{{{\left( 2\sigma _{n,k}^{2}{{Z}_{n}}\log n \right)}^{1/2}}}{{\left( \frac{\sigma _{n,k}^{2}}{\sigma _{n}^{2}} \right)}^{1/2}}\frac{1}{{{m}_{n+k}}}. \\ \end{align}$ |
So the lim sup part is at least
$\begin{align} & -\underset{n\to \infty }{\mathop{\lim }}\,{{\left( \frac{\sigma _{n+k}^{2}{{Z}_{n+k}}\log \left( n+k \right)}{\sigma _{n}^{2}{{Z}_{n}}\log n} \right)}^{1/2}}\frac{1}{{{m}_{n+k}}}+ \\ & \underset{n\to \infty }{\mathop{\lim }}\,{{\left( \frac{\sigma _{n+k}^{2}}{\sigma _{n}^{2}} \right)}^{1/2}}\frac{1}{{{m}_{n,k}}}. \\ \end{align}$ |
Again since the first item of the above expression is smaller than
Cα-k/2 and the second item
$\frac{\sigma _{n+k}^{2}}{\sigma _{n}^{2}}\frac{1}{{{\left( {{m}_{n,k}} \right)}^{2}}}=\left( \sum\limits_{j=n}^{n+k-1}{\frac{\delta _{j}^{2}}{\mu _{j}^{2}{{m}_{n,j-n}}}} \right)/\left( \sum\limits_{j=n}^{\infty }{\frac{\delta _{j}^{2}}{\mu _{j}^{2}{{m}_{n,j-n}}}} \right)=\frac{1}{1+{{b}_{n,k}}},$ |
where
bn,k<
Cα-k and
C>0 is a constant,the lim sup part of Eq.(8) is at least 1 as
k→∞. The lim inf part of Eq.(8) can be proved in the same way as for Eq.(12).□
I am very grateful to my supervisor Prof. HU for intensive discussion with me and also benefit much from the communications with Gao Zhenlong,Liang Longyue,Zhang Zhiyang,and Zhao Rongjie.
References[1] | Heyde C C. Some almost sure convergence theorems for branching processes[J].Z Wahrscheinlichkeitstheorie verw Geb, 1971, 20:189–192.DOI:10.1007/BF00534900 |
[2] | Heyde C C, Leslie J R. Improved classical limit analogues for Galton-Watson processes with or without immigration[J].Bull Austral Math Soc, 1971, 5:145–155.DOI:10.1017/S0004972700047018 |
[3] | Asmussen S. Almost sure behavior of linear functionals of supercritical branching processes[J].Transactions of The American Mathematical Society, 1977, 231(1):233–248.DOI:10.1090/S0002-9947-1977-0440719-1 |
[4] | Huggins R M. Laws of the iterated logarithm for time changed brownian motion with an application to branching processes[J].Ann Probability, 1985, 13(4):1148–1156.DOI:10.1214/aop/1176992801 |
[5] | Gao Z L. Limit theorems for Galton-Watson processes in random environments[D]. Beijing: Graduate University of Chinese Academy of Sciences, 2011(in Chinese). |
[6] | Fearn D H. Galton-Watson processes with generation dependence[J].Proc Sixth Berkeley Symp Math Statist Probab, 1971, 2:159–172. |
[7] | Durrett R. Probability:theory and examples[M].3rd ed.Beijing: World Book Inc, 2011. |