第三章 Poisson过程(1)
1.Poisson过程的定义
记 N ( s , t ] N(s,t] N(s,t]为 ( s , t ] (s,t] (s,t]时间段内事件发生的次数,也就是一个计数过程, N ( t ) = N ( 0 , t ] N(t)=N(0,t] N(t)=N(0,t],则满足以下要求的随机过程 N = ( N ( t ) , t ≥ 0 ) \boldsymbol N =(N(t),t\ge 0) N=(N(t),t≥0)是Poisson过程:
-
初始条件: N ( 0 ) = 0 , N ( t ) ≥ 0 N(0)=0,N(t)\ge 0 N(0)=0,N(t)≥0;
-
独立增量:对任意 0 < t 1 < ⋯ < t k 0<t_1<\cdots<t_k 0<t1<⋯<tk,有 N ( t 1 ) , N ( t 2 ) − N ( t 1 ) , ⋯ , N ( t k ) − N ( t k − 1 ) N(t_1),N(t_2)-N(t_1),\cdots,N(t_k)-N(t_{k-1}) N(t1),N(t2)−N(t1),⋯,N(tk)−N(tk−1)相互独立;
-
平稳增量:对任意 s < t s<t s<t, N ( t ) − N ( s ) N(t)-N(s) N(t)−N(s)与 N ( t − s ) N(t-s) N(t−s)具有相同的分布;
-
稀有性:存在一个正常数 λ > 0 \lambda >0 λ>0,使得对任何 t > 0 t>0 t>0有
P ( N ( t , t + Δ ( t ) ) = 1 ) = λ Δ ( t ) + o ( Δ ( t ) ) P ( N ( t , t + Δ ( t ) ) ≥ 1 ) = o ( Δ ( t ) ) P(N(t,t+\Delta(t))=1)=\lambda \Delta(t)+o(\Delta (t))\\ P(N(t,t+\Delta(t))\ge1)=o(\Delta(t)) P(N(t,t+Δ(t))=1)=λΔ(t)+o(Δ(t))P(N(t,t+Δ(t))≥1)=o(Δ(t))
这里 Δ ( t ) \Delta(t) Δ(t)是一个相对小的时间增量,意味着在微小时刻几乎不可能同时发生两个及以上的事件。
由以上1~4条件保证的泊松过程,满足以下的分布规律:
P
(
N
(
t
)
=
k
)
=
(
λ
t
)
k
k
!
e
−
λ
t
,
k
=
0
,
1
,
2
,
⋯
P(N(t)=k)=\frac{(\lambda t)^k}{k!}e^{-\lambda t},\quad k=0,1,2,\cdots
P(N(t)=k)=k!(λt)ke−λt,k=0,1,2,⋯
要证明这个结论,将
[
0
,
t
]
[0,t]
[0,t]区间划分成
n
n
n个小区间,使得每个区间长度为
t
/
n
t/n
t/n足够小,这样
P
(
N
(
t
)
=
k
)
P(N(t)=k)
P(N(t)=k)就相当于
n
n
n个小区间中有
k
k
k个小区间取1,剩下的取0(由稀缺性,每个小区间不超过2)。由独立增量性,每个小区间互相独立且同分布,即
P
(
N
(
t
/
n
)
=
1
)
=
λ
t
n
+
o
(
t
n
)
,
P
(
N
(
t
/
n
)
=
0
)
=
1
−
λ
t
n
+
o
(
t
n
)
,
P
(
N
(
t
)
=
k
)
=
lim
n
→
∞
C
n
k
(
λ
t
n
+
o
(
t
n
)
)
k
(
1
−
λ
t
n
+
o
t
n
)
n
−
k
=
lim
n
→
∞
n
!
(
n
−
k
)
!
n
k
(
λ
t
)
k
k
!
e
−
(
n
−
k
)
λ
t
n
=
(
λ
t
)
k
k
!
e
−
λ
t
\begin{aligned} P(N(t/n)=1)=&\frac{\lambda t}{n}+o(\frac tn),\\ P(N(t/n)=0)=&1-\frac{\lambda t}{n}+o(\frac tn),\\ \\ P(N(t)=k)=&\lim_{n\to \infty}C_n^k (\frac{\lambda t}{n}+o(\frac tn))^k(1-\frac{\lambda t}{n}+o\frac{t}{n})^{n-k}\\ =&\lim_{n\to \infty }\frac{n!}{(n-k)!n^k}\frac{(\lambda t)^k}{k!}e^{-\frac{(n-k)\lambda t}{n}}\\ =&\frac{(\lambda t)^k}{k!}e^{-\lambda t} \end{aligned}
P(N(t/n)=1)=P(N(t/n)=0)=P(N(t)=k)===nλt+o(nt),1−nλt+o(nt),n→∞limCnk(nλt+o(nt))k(1−nλt+ont)n−kn→∞lim(n−k)!nkn!k!(λt)ke−n(n−k)λtk!(λt)ke−λt
可以看到,参数为
λ
\lambda
λ的泊松过程
N
\boldsymbol N
N,有
N
(
t
)
∼
P
(
λ
t
)
N(t)\sim P(\lambda t)
N(t)∼P(λt)。
2.泊松过程的基本性质
由于泊松过程
N
=
(
N
(
t
)
,
t
∈
T
)
\boldsymbol N=(N(t),t\in T)
N=(N(t),t∈T)满足
N
(
t
)
∼
P
(
λ
t
)
N(t)\sim P(\lambda t)
N(t)∼P(λt),因此计算泊松过程的均值、方差和特征函数如下:
μ
N
(
t
)
=
E
(
N
(
t
)
)
=
λ
t
,
σ
N
2
(
t
)
=
D
(
N
(
t
)
)
=
λ
t
,
ϕ
N
(
t
)
(
x
)
=
E
e
i
x
N
(
t
)
=
∑
k
=
0
∞
e
i
x
k
(
λ
t
)
k
k
!
e
−
λ
t
=
e
−
λ
t
∑
k
=
0
∞
(
e
i
x
λ
t
)
k
k
!
=
e
λ
t
(
e
i
x
−
1
)
\mu_N(t)=E(N(t))=\lambda t,\\ \sigma^2_N(t)=D(N(t))=\lambda t,\\ \begin{aligned} &\phi_{N(t)}(x)=Ee^{ixN(t)}\\=&\sum_{k=0}^\infty e^{ixk}\frac{(\lambda t)^k}{k!}e^{-\lambda t}\\ =&e^{-\lambda t}\sum_{k=0}^\infty \frac{(e^{ix}\lambda t)^k}{k!}\\ =&e^{\lambda t(e^{ix}-1)} \end{aligned}
μN(t)=E(N(t))=λt,σN2(t)=D(N(t))=λt,===ϕN(t)(x)=EeixN(t)k=0∑∞eixkk!(λt)ke−λte−λtk=0∑∞k!(eixλt)keλt(eix−1)
因此
λ
=
E
[
N
(
t
)
]
/
t
\lambda =E[N(t)]/t
λ=E[N(t)]/t,如果将此计数过程看作一个服务系统,则参数
λ
\lambda
λ表明服务的平均繁忙程度。
下面假设
s
<
t
s<t
s<t,计算泊松过程的自相关函数和自协方差函数如下:
r
N
(
s
,
t
)
=
E
[
N
(
s
)
N
(
t
)
]
=
E
[
N
(
s
)
N
(
t
−
s
+
s
)
]
=
E
[
N
(
s
)
N
(
t
−
s
)
]
+
E
[
N
(
s
)
]
2
=
E
[
N
(
s
)
]
E
[
N
(
t
−
s
)
]
+
D
[
N
(
s
)
]
+
[
E
N
(
s
)
]
2
=
λ
s
λ
(
t
−
s
)
+
λ
s
+
(
λ
s
)
2
=
λ
2
s
t
+
λ
s
C
o
v
N
(
s
,
t
)
=
r
N
(
s
,
t
)
−
μ
N
(
s
)
μ
N
(
t
)
=
λ
2
s
t
+
λ
s
−
λ
2
s
t
=
λ
s
\begin{aligned} &r_N(s,t)=E[N(s)N(t)]\\ =&E[N(s)N(t-s+s)]\\ =&E[N(s)N(t-s)]+E[N(s)]^2\\ =&E[N(s)]E[N(t-s)]+D[N(s)]+[EN(s)]^2\\ =&\lambda s\lambda (t-s)+\lambda s+(\lambda s)^2\\ =&\lambda^2st+\lambda s\\ \\ &Cov_N(s,t)=r_N(s,t)-\mu_N(s)\mu_N(t)\\ =&\lambda^2st+\lambda s-\lambda^2 st\\ =&\lambda s \end{aligned}
=======rN(s,t)=E[N(s)N(t)]E[N(s)N(t−s+s)]E[N(s)N(t−s)]+E[N(s)]2E[N(s)]E[N(t−s)]+D[N(s)]+[EN(s)]2λsλ(t−s)+λs+(λs)2λ2st+λsCovN(s,t)=rN(s,t)−μN(s)μN(t)λ2st+λs−λ2stλs
如果没给定
s
,
t
s,t
s,t的大小关系,则
C
o
v
N
(
s
,
t
)
=
λ
(
m
∧
n
)
Cov_N(s,t)=\lambda(m\land n)
CovN(s,t)=λ(m∧n),这里
∧
\land
∧表示两者取小,
∨
\lor
∨表示两者取大。
接下来计算Poisson过程的联合分布,由于其具有独立增量性,所以
P
(
N
(
t
1
)
=
k
1
,
⋯
,
N
(
t
m
)
=
k
m
)
=
P
(
N
(
t
1
)
=
k
1
)
⋯
P
(
N
(
t
m
)
−
N
(
t
m
−
1
)
=
k
m
−
k
m
−
1
)
=
(
λ
t
1
)
k
1
k
1
!
[
λ
(
t
2
−
t
1
)
]
k
2
−
k
1
(
k
2
−
k
1
)
!
⋯
[
λ
(
t
m
−
t
m
−
1
)
]
k
m
−
k
m
−
1
(
k
m
−
k
m
−
1
)
!
e
−
λ
t
m
\begin{aligned} &P(N(t_1)=k_1,\cdots,N(t_m)=k_m)\\ =&P(N(t_1)=k_1)\cdots P(N(t_m)-N(t_{m-1})=k_m-k_{m-1})\\ =&\frac{(\lambda t_1)^{k_1}}{k_1!}\frac{[\lambda(t_2-t_1)]^{k_2-k_1}}{(k_2-k_1)!}\cdots\frac{[\lambda(t_m-t_{m-1})]^{k_m-k_{m-1}}}{(k_m-k_{m-1})!}e^{-\lambda t_m}\\ \end{aligned}
==P(N(t1)=k1,⋯,N(tm)=km)P(N(t1)=k1)⋯P(N(tm)−N(tm−1)=km−km−1)k1!(λt1)k1(k2−k1)![λ(t2−t1)]k2−k1⋯(km−km−1)![λ(tm−tm−1)]km−km−1e−λtm
对于条件分布,有两种情况,现假设
s
<
t
,
k
≤
m
s<t,k\le m
s<t,k≤m,则
P
(
N
(
t
)
=
m
∣
N
(
s
)
=
k
)
=
P
(
N
(
t
)
=
m
,
N
(
s
)
=
k
)
P
(
N
(
s
)
=
k
)
=
(
λ
s
)
k
k
!
e
−
λ
s
[
λ
(
t
−
s
)
]
m
−
k
(
m
−
k
)
!
e
−
λ
(
t
−
s
)
(
λ
s
)
k
k
!
e
−
λ
s
=
[
λ
(
t
−
s
)
]
m
−
k
(
m
−
k
)
!
e
−
λ
(
t
−
s
)
P
(
N
(
s
)
=
k
∣
N
(
t
)
=
m
)
=
P
(
N
(
t
)
=
m
,
N
(
s
)
=
k
)
P
(
N
(
t
)
=
m
)
=
(
λ
s
)
k
k
!
e
−
λ
s
[
λ
(
t
−
s
)
]
m
−
k
(
m
−
k
)
!
e
−
λ
(
t
−
s
)
(
λ
t
)
m
m
!
e
−
λ
t
=
m
!
k
!
(
m
−
k
)
!
s
k
(
t
−
s
)
m
−
k
t
m
=
C
m
k
(
s
t
)
k
(
1
−
s
t
)
m
−
k
\begin{aligned} &P(N(t)=m|N(s)=k)\\ =&\frac{P(N(t)=m,N(s)=k)}{P(N(s)=k)}\\ =&\frac{\frac{(\lambda s)^k}{k!}e^{-\lambda s}\frac{[\lambda(t-s)]^{m-k}}{(m-k)!}e^{-\lambda (t-s)}}{\frac{(\lambda s)^k}{k!}e^{-\lambda s}}\\ =&\frac{[\lambda (t-s)]^{m-k}}{(m-k)!}e^{-\lambda(t-s)} \\ \\ &P(N(s)=k|N(t)=m)\\ =&\frac{P(N(t)=m,N(s)=k)}{P(N(t)=m)}\\ =&\frac{\frac{(\lambda s)^k}{k!}e^{-\lambda s}\frac{[\lambda(t-s)]^{m-k}}{(m-k)!}e^{-\lambda (t-s)}}{\frac{(\lambda t)^m}{m!}e^{-\lambda t}}\\ =&\frac{m!}{k!(m-k)!}\frac{s^k(t-s)^{m-k}}{t^m}\\ =&C_m^k(\frac st)^k (1-\frac st)^{m-k} \end{aligned}
=======P(N(t)=m∣N(s)=k)P(N(s)=k)P(N(t)=m,N(s)=k)k!(λs)ke−λsk!(λs)ke−λs(m−k)![λ(t−s)]m−ke−λ(t−s)(m−k)![λ(t−s)]m−ke−λ(t−s)P(N(s)=k∣N(t)=m)P(N(t)=m)P(N(t)=m,N(s)=k)m!(λt)me−λtk!(λs)ke−λs(m−k)![λ(t−s)]m−ke−λ(t−s)k!(m−k)!m!tmsk(t−s)m−kCmk(ts)k(1−ts)m−k
可以看到第二个条件分布是二项分布的形式,这可以解释为
t
t
t时间内到达的
m
m
m个顾客独立选择服务,有
k
k
k个选择在
(
0
,
s
]
(0,s]
(0,s]时间段内服务,选择的概率自然是两段时间的长度比
s
/
t
s/t
s/t。
3.到达时刻与时间间隔
记 S i S_i Si为第 i i i个事件的到达时刻,且记 S 0 = 0 S_0=0 S0=0。 T i = S i − S i − 1 T_i=S_i-S_{i-1} Ti=Si−Si−1为第 i − 1 i-1 i−1个事件和第 i i i个事件间隔。 S i , T i , N ( t ) S_i,T_i,N(t) Si,Ti,N(t)之间存在一系列关联。
S
n
>
t
S_n>t
Sn>t,意味着第
n
n
n个事件在时刻
t
t
t以后发生,也就意味着
t
t
t时刻及以前发生的事件个数不大于
n
−
1
n-1
n−1,也就是
S
n
>
t
⇔
N
(
t
)
≤
n
−
1
P
(
S
n
>
t
)
=
P
(
N
(
t
)
≤
n
−
1
)
=
∑
k
=
0
n
−
1
(
λ
t
)
k
k
!
e
−
λ
t
S_n>t\Leftrightarrow N(t)\le n-1\\ P(S_n>t)=P(N(t)\le n-1)=\sum_{k=0}^{n-1}\frac{(\lambda t)^k}{k!}e^{-\lambda t}
Sn>t⇔N(t)≤n−1P(Sn>t)=P(N(t)≤n−1)=k=0∑n−1k!(λt)ke−λt
这样就得到
S
n
S_n
Sn的分布函数和密度函数
F
S
n
(
t
)
=
P
(
S
n
≤
t
)
=
1
−
∑
k
=
0
n
−
1
(
λ
t
)
k
k
!
e
−
λ
t
=
e
−
λ
t
∑
k
=
n
∞
(
λ
t
)
k
k
!
p
S
n
(
t
)
=
F
S
n
′
(
t
)
=
−
e
−
λ
t
∑
k
=
n
∞
λ
k
+
1
t
k
k
!
+
e
−
λ
t
∑
k
=
n
∞
λ
k
t
k
−
1
(
k
−
1
)
!
=
λ
n
t
n
−
1
(
n
−
1
)
!
e
−
λ
t
=
λ
n
Γ
(
n
)
t
n
−
1
e
−
λ
t
,
t
>
0
\begin{aligned} &F_{S_n}(t)=P(S_n\le t)\\ =&1-\sum_{k=0}^{n-1}\frac{(\lambda t)^k}{k!}e^{-\lambda t}\\ =&e^{-\lambda t}\sum_{k=n}^\infty \frac{(\lambda t)^k}{k!}\\ \\ &p_{S_n}(t)=F'_{S_n}(t)\\ =&-e^{-\lambda t}\sum_{k=n}^\infty \frac{\lambda ^{k+1} t^k }{k!}+e^{-\lambda t}\sum_{k=n}^\infty \frac{\lambda ^k t^{k-1}}{(k-1)!}\\ =&\frac{\lambda ^n t^{n-1}}{(n-1)!}e^{-\lambda t}\\ =&\frac{\lambda^n}{\Gamma(n)}t^{n-1}e^{-\lambda t},\quad t>0 \end{aligned}
=====FSn(t)=P(Sn≤t)1−k=0∑n−1k!(λt)ke−λte−λtk=n∑∞k!(λt)kpSn(t)=FSn′(t)−e−λtk=n∑∞k!λk+1tk+e−λtk=n∑∞(k−1)!λktk−1(n−1)!λntn−1e−λtΓ(n)λntn−1e−λt,t>0
这里得到
S
n
∼
Γ
(
n
,
λ
)
S_n\sim \Gamma(n,\lambda)
Sn∼Γ(n,λ)。
同时,由于
T
i
=
S
i
−
S
i
−
1
T_i=S_i-S_{i-1}
Ti=Si−Si−1,可以计算第
i
−
1
i-1
i−1个事件到第
i
i
i个事件的时间间隔,得到
P
(
T
n
>
t
∣
S
n
−
1
=
s
)
=
P
(
S
n
>
t
+
s
∣
S
n
−
1
=
s
)
=
P
(
N
(
t
+
s
)
<
n
∣
N
(
s
)
=
n
−
1
,
N
(
s
−
0
)
<
n
−
1
)
=
P
(
N
(
s
,
t
+
s
)
<
1
)
=
P
(
N
(
t
)
=
0
)
=
e
−
λ
t
\begin{aligned} &P(T_n>t|S_{n-1}=s)\\ =&P(S_n>t+s|S_{n-1}=s)\\ =&P(N(t+s)<n|N(s)=n-1,N(s-0)<n-1)\\ =&P(N(s,t+s)<1)\\ =&P(N(t)=0)=e^{-\lambda t} \end{aligned}
====P(Tn>t∣Sn−1=s)P(Sn>t+s∣Sn−1=s)P(N(t+s)<n∣N(s)=n−1,N(s−0)<n−1)P(N(s,t+s)<1)P(N(t)=0)=e−λt
这里可以看到
P
(
T
n
>
t
∣
S
n
=
s
)
P(T_n>t|S_n=s)
P(Tn>t∣Sn=s)与
s
s
s无关,因此有
T
n
T_n
Tn与
S
n
−
1
S_{n-1}
Sn−1独立,并且
F
T
n
(
t
)
=
1
−
P
(
T
n
>
t
)
=
1
−
e
−
λ
t
;
p
T
n
(
t
)
=
λ
e
−
λ
t
.
F_{T_n}(t)=1-P(T_n>t)=1-e^{-\lambda t};\\ p_{T_n}(t)=\lambda e^{-\lambda t}.
FTn(t)=1−P(Tn>t)=1−e−λt;pTn(t)=λe−λt.
这里得到
T
n
∼
E
(
λ
)
=
Γ
(
1
,
λ
)
T_n\sim E(\lambda )=\Gamma(1,\lambda)
Tn∼E(λ)=Γ(1,λ)。因此有这样的结论:
-
事件发生的间隔 T 1 , ⋯ , T n T_1,\cdots,T_n T1,⋯,Tn为相互独立的随机变量,它们独立同分布于 E ( λ ) E(\lambda) E(λ);
-
S n = T 1 + ⋯ + T n S_n=T_1+\cdots+T_n Sn=T1+⋯+Tn为第 n n n个事件的发生时间,它服从 Γ ( n , λ ) \Gamma(n,\lambda ) Γ(n,λ)。
-
如果 X 1 , X 2 , ⋯ X_1,X_2,\cdots X1,X2,⋯是一列独立同分布的 E ( λ ) E(\lambda) E(λ)随机变量,定义 S 0 = 0 S_0=0 S0=0, S n = ∑ i = 1 n X i S_n=\sum_{i=1}^n X_i Sn=∑i=1nXi, N ( 0 ) = 0 N(0)=0 N(0)=0,则随机过程 N = ( N ( t ) , t ∈ T ) \boldsymbol N=(N(t),t\in T) N=(N(t),t∈T)是泊松过程,这里
N ( t ) = max { n : S n ≤ t } . N(t)=\max\{n:S_n\le t\}. N(t)=max{n:Sn≤t}.
要证明 N ( t ) N(t) N(t)是泊松过程,首先要证明 N ( t ) ∼ P ( λ t ) N(t)\sim P(\lambda t) N(t)∼P(λt),为此,有
P ( N ( t ) = k ) = P ( S k ≤ t < S k + 1 ) = ∫ 0 t P ( t < S k + 1 ∣ S k = s ) p S k ( s ) d s = ∫ 0 t P ( X k + 1 > t − s ∣ S k = s ) p S k ( s ) d s = ∫ 0 t e − λ ( t − s ) λ k Γ ( k ) s k − 1 e − λ s d s = ( λ t ) k k ! e − λ t \begin{aligned} &P(N(t)=k)\\ =& P(S_k\le t < S_{k+1})\\ =& \int_0^tP(t<S_{k+1}|S_k=s)p_{S_k}(s)ds\\ =& \int_0^t P(X_{k+1}>t-s|S_k=s)p_{S_k}(s)ds\\ =& \int_0^t e^{-\lambda(t-s)}\frac{\lambda ^k}{\Gamma(k)}s^{k-1}e^{-\lambda s}ds\\ =&\frac{(\lambda t)^k}{k!}e^{-\lambda t} \end{aligned} =====P(N(t)=k)P(Sk≤t<Sk+1)∫0tP(t<Sk+1∣Sk=s)pSk(s)ds∫0tP(Xk+1>t−s∣Sk=s)pSk(s)ds∫0te−λ(t−s)Γ(k)λksk−1e−λsdsk!(λt)ke−λt
这里用到全概率公式: P ( A ) = ∫ − ∞ ∞ P ( A ∣ X = x ) p ( x ) d x P(A)=\int_{-\infty}^\infty P(A|X=x)p(x)dx P(A)=∫−∞∞P(A∣X=x)p(x)dx。还要证明其独立平稳增量性,即证明
P ( N ( t ) − N ( s ) < m , N ( s ) = k ) = P ( N ( t − s ) < m ) ⋅ P ( N ( s ) = k ) P(N(t)-N(s)<m, N(s)=k)=P(N(t-s)<m)\cdot P(N(s)=k) P(N(t)−N(s)<m,N(s)=k)=P(N(t−s)<m)⋅P(N(s)=k)
这里
P ( N ( t ) − N ( s ) < m , N ( s ) = k ) = P ( N ( t ) < k + m , N ( s ) = k ) = P ( S k + m > t , S k ≤ s < S k + 1 ) = P ( S k + m > t , S k ≤ s ) − P ( S k + m > t , S k + 1 ≤ s ) P ( S k + m > t , S k ≤ s ) = P ( S k + m − S k > t − S k , S k ≤ s ) = ∫ 0 s P ( S m > t − u ) p S k ( u ) d u \begin{aligned} &P(N(t)-N(s)<m, N(s)=k)\\ =&P(N(t)<k+m,N(s)=k)\\ =&P(S_{k+m}>t,S_k\le s<S_{k+1})\\ =& P(S_{k+m}>t,S_k \le s)-P(S_{k+m}>t,S_{k+1}\le s)\\ \\ &P(S_{k+m}>t,S_k\le s)\\ =&P(S_{k+m}-S_k >t-S_k,S_k\le s)\\ =& \int_0^s P(S_m>t-u) p_{S_k}(u)du\\ \end{aligned} =====P(N(t)−N(s)<m,N(s)=k)P(N(t)<k+m,N(s)=k)P(Sk+m>t,Sk≤s<Sk+1)P(Sk+m>t,Sk≤s)−P(Sk+m>t,Sk+1≤s)P(Sk+m>t,Sk≤s)P(Sk+m−Sk>t−Sk,Sk≤s)∫0sP(Sm>t−u)pSk(u)du
注意到前面我们证明了 N ( t ) ∼ P ( λ t ) < n ⇔ S n ∼ Γ ( n , λ ) > t N(t)\sim P(\lambda t)<n\Leftrightarrow S_n\sim \Gamma(n,\lambda )>t N(t)∼P(λt)<n⇔Sn∼Γ(n,λ)>t,因此可以对上述式子中的概率进行替换,即 P ( S m > t − u ) = P ( X < m ) P(S_m>t-u)=P(X<m) P(Sm>t−u)=P(X<m),这里 X ∼ P ( λ ( t − u ) ) X\sim P(\lambda (t-u)) X∼P(λ(t−u)),所以
P ( S k + m > t , S k ≤ s ) = P ( S k + m − S k > t − S k , S k ≤ s ) = ∫ 0 s P ( S m > t − u ) p S k ( u ) d u = ∫ 0 s P { P ( λ ( t − u ) < m ) } p s k ( u ) d u = ∫ 0 s ∑ l = 0 m − 1 λ l ( t − u ) l l ! e − λ ( t − u ) λ k ( k − 1 ) ! u k − 1 e − λ u d u = e − λ t ∑ l = 0 m − 1 λ k + l ( k − 1 ) ! l ! ∫ 0 s u k − 1 ( t − u ) l d u \begin{aligned} &P(S_{k+m}>t,S_k\le s)\\ =&P(S_{k+m}-S_k >t-S_k,S_k\le s)\\ =& \int_0^s P(S_m>t-u) p_{S_k}(u)du\\ =&\int_0^sP\{P(\lambda(t-u)<m)\}p_{s_k}(u)du\\ =&\int_0^s \sum_{l=0}^{m-1}\frac{\lambda^l(t-u)^l}{l!}e^{-\lambda (t-u)}\frac{\lambda ^k}{(k-1)!}u^{k-1}e^{-\lambda u}du\\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{(k-1)!l!}\int_0^s u^{k-1}(t-u)^l du \end{aligned} =====P(Sk+m>t,Sk≤s)P(Sk+m−Sk>t−Sk,Sk≤s)∫0sP(Sm>t−u)pSk(u)du∫0sP{P(λ(t−u)<m)}psk(u)du∫0sl=0∑m−1l!λl(t−u)le−λ(t−u)(k−1)!λkuk−1e−λudue−λtl=0∑m−1(k−1)!l!λk+l∫0suk−1(t−u)ldu
类似地有
P ( S k + m > t , S k + 1 ≤ s ) = ∫ 0 s P ( S m − 1 > t − u ) p S k + 1 ( u ) d u = ∫ 0 s P { P ( λ ( t − u ) ) < m − 1 } p S k + 1 ( u ) d u = ∫ 0 s ∑ l = 0 m − 2 λ l ( t − u ) l l ! e − λ ( t − u ) λ k + 1 k ! u k e − λ u d u = e − λ t ∑ l = 0 m − 2 λ k + 1 + l k ! ⋅ l ! ∫ 0 s u k ( t − u ) l d u \begin{aligned} &P(S_{k+m}>t,S_{k+1}\le s)\\ =& \int_0^s P(S_{m-1}>t-u)p_{S_{k+1}}(u)du\\ =& \int_0^s P\{P(\lambda (t-u))<m-1\}p_{S_{k+1}}(u)du\\ =&\int_0^s \sum_{l=0}^{m-2} \frac{\lambda^l (t-u)^l}{l!}e^{-\lambda(t-u)}\frac{\lambda ^{k+1}}{k!}u^ke^{-\lambda u}du\\ =&e^{-\lambda t}\sum_{l=0}^{m-2}\frac{\lambda ^{k+1+l}}{k!\cdot l!}\int_0^s u^k (t-u)^ldu \end{aligned} ====P(Sk+m>t,Sk+1≤s)∫0sP(Sm−1>t−u)pSk+1(u)du∫0sP{P(λ(t−u))<m−1}pSk+1(u)du∫0sl=0∑m−2l!λl(t−u)le−λ(t−u)k!λk+1uke−λudue−λtl=0∑m−2k!⋅l!λk+1+l∫0suk(t−u)ldu
而
P ( S k + m > t , S k ≤ s ) = e − λ t ∑ l = 0 m − 1 λ k + l ( k − 1 ) ! l ! ∫ 0 s u k − 1 ( t − u ) l d u = e − λ t ∑ l = 0 m − 1 λ k + l k ! l ! ∫ 0 s k u k − 1 ( t − u ) l d u = e − λ t ∑ l = 0 m − 1 λ k + l k ! l ! ( s k ( t − s ) l + l ∫ 0 s u k ( t − u ) l − 1 d u ) = e − λ t ∑ l = 0 m − 1 λ k + l k ! l ! s k ( t − s ) l + e − λ t ∑ l = 1 m − 1 λ k + l k ! ( l − 1 ) ! ∫ 0 s u k ( t − u ) l − 1 d u = e − λ t ∑ l = 0 m − 1 λ k + l k ! l ! s k ( t − s ) l + e − λ t ∑ l = 0 m − 2 λ k + l + 1 k ! ⋅ l ! ∫ 0 s u k ( t − u ) l d u = e − λ t ∑ l = 0 m − 1 λ k + l k ! l ! s k ( t − s ) l + P ( S k + m > t , S k + 1 ≤ s ) \begin{aligned} &P(S_{k+m}>t,S_k\le s)\\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{(k-1)!l!}\int_0^s u^{k-1}(t-u)^l du \\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{k!l!}\int_0^s ku^{k-1 }(t-u)^ldu\\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{k!l!}\left(s^{k}(t-s)^l+l\int_0^s u^k (t-u)^{l-1}du\right)\\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{k!l!}s^{k}(t-s)^l \\&+e^{-\lambda t}\sum_{l=1}^{m-1}\frac{\lambda ^{k+l}}{k!(l-1)!}\int_0^su^k(t-u)^{l-1}du\\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{k!l!}s^{k}(t-s)^l+e^{-\lambda t}\sum_{l=0}^{m-2}\frac{\lambda ^{k+l+1}}{k!\cdot l!}\int_0^s u^k(t-u)^ldu\\ =&e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda^{k+l}}{k!l!}s^{k}(t-s)^l+P(S_{k+m}>t,S_{k+1}\le s) \end{aligned} ======P(Sk+m>t,Sk≤s)e−λtl=0∑m−1(k−1)!l!λk+l∫0suk−1(t−u)ldue−λtl=0∑m−1k!l!λk+l∫0skuk−1(t−u)ldue−λtl=0∑m−1k!l!λk+l(sk(t−s)l+l∫0suk(t−u)l−1du)e−λtl=0∑m−1k!l!λk+lsk(t−s)l+e−λtl=1∑m−1k!(l−1)!λk+l∫0suk(t−u)l−1due−λtl=0∑m−1k!l!λk+lsk(t−s)l+e−λtl=0∑m−2k!⋅l!λk+l+1∫0suk(t−u)ldue−λtl=0∑m−1k!l!λk+lsk(t−s)l+P(Sk+m>t,Sk+1≤s)
因此有
P ( N ( t ) − N ( s ) < m , N ( s ) = k ) = P ( S k + m > t , S k ≤ s ) − P ( S k + m > t , S k + 1 ≤ s ) = e − λ t ∑ l = 0 m − 1 λ k + l k ! l ! s k ( t − s ) l = ( λ k s k k ! e − λ s ) ( ∑ l = 0 m − 1 λ l ( t − s ) l l ! e − λ ( t − s ) ) = P ( N ( s ) = k ) ⋅ P ( N ( t − s ) < m ) \begin{aligned} &P(N(t)-N(s)<m, N(s)=k)\\ =& P(S_{k+m}>t,S_k \le s)-P(S_{k+m}>t,S_{k+1}\le s)\\ =& e^{-\lambda t}\sum_{l=0}^{m-1}\frac{\lambda ^{k+l}}{k!l!}s^{k}(t-s)^l\\ =&\left(\frac{\lambda ^ks^k}{k!}e^{-\lambda s}\right) \left(\sum_{l=0}^{m-1} \frac{\lambda ^{l}(t-s)^{l}}{l!}e^{-\lambda (t-s)}\right)\\ =&P(N(s)=k)\cdot P(N(t-s)<m) \end{aligned} ====P(N(t)−N(s)<m,N(s)=k)P(Sk+m>t,Sk≤s)−P(Sk+m>t,Sk+1≤s)e−λtl=0∑m−1k!l!λk+lsk(t−s)l(k!λkske−λs)(l=0∑m−1l!λl(t−s)le−λ(t−s))P(N(s)=k)⋅P(N(t−s)<m)