概述
符号
设
T
=
{
t
:
t
∈
T
}
T={t: t in T}
T={t:t∈T} 为任意指标集,
{
(
Ω
t
,
F
t
)
:
t
∈
T
}
left{left(Omega_{t}, mathscr{F}_{t}right): t in Tright}
{(Ωt,Ft):t∈T} 为一族可测空间,
Ω
=
∏
t
∈
T
Ω
t
,
F
=
∏
t
∈
T
F
t
Omega=underset{t in T}{prod} Omega_{t}, quad mathscr{F}=underset{t in T}{prod}{mathcal{F}}_{t}
Ω=t∈T∏Ωt,F=t∈T∏Ft
又
T
1
⊂
T
T_{1} subset T
T1⊂T 为
T
T
T 的任一子集, 记
Ω
T
1
=
∏
t
∈
T
1
Ω
t
,
F
T
1
=
∏
t
∈
T
1
F
t
Omega_{T_{1}}=underset{t in T_{1}}{prod} Omega_{t}, quad mathscr{F}_{T_{1}}=underset{t in T_{1}}{prod} mathscr{F}_{t}
ΩT1=t∈T1∏Ωt,FT1=t∈T1∏Ft
也可记
Ω
=
Ω
T
,
F
=
F
T
.
Omega=Omega_{T}, mathscr{F}=mathscr{F}_{T} .
Ω=ΩT,F=FT.
对
A
∈
F
T
1
A in mathscr{F}_{T_{1}}
A∈FT1
B
1
=
{
(
ω
t
,
t
∈
T
)
∈
Ω
T
:
(
ω
α
,
α
∈
T
1
)
∈
A
}
B_{1}=left{left(omega_{t}, t in Tright) in Omega_{T}:left(omega_{alpha}, alpha in T_{1}right) in Aright}
B1={(ωt,t∈T)∈ΩT:(ωα,α∈T1)∈A}
称
B
1
B_{1}
B1 为
Ω
Omega
Ω 中以
A
A
A 为基底的柱集.
对
T
1
⊂
T
2
⊂
T
,
T_{1} subset T_{2} subset T,
T1⊂T2⊂T,
B
2
=
{
(
ω
t
,
t
∈
T
2
)
∈
Ω
T
2
:
(
ω
α
,
α
∈
T
1
)
∈
A
}
B_{2}=left{left(omega_{t}, t in T_{2}right) in Omega_{T_{2}}:left(omega_{alpha}, alpha in T_{1}right) in Aright}
B2={(ωt,t∈T2)∈ΩT2:(ωα,α∈T1)∈A}
称 B 2 B_{2} B2 为 Ω T 2 Omega_{T_{2}} ΩT2 中以 A A A 为基底的柱集。
以 F ‾ T 1 overline{mathscr{F}}_{T_{1}} FT1 和 F ‾ T 1 T 2 overline{mathscr{F}}_{T_{1}}^{T_{2}} FT1T2 分别表示 F T 1 mathscr{F}_{T_{1}} FT1中所有基底在 Ω Omega Ω和 Ω T 2 Omega_{T_{2}} ΩT2 中的柱集全体.
设
C
mathscr{C}
C 表示
Ω
Omega
Ω 中基底为有限维可测矩形的柱集全体,
A
=
⋃
T
1
⊂
T
F
‾
T
1
mathscr{A}=bigcup_{T_{1} subset T} overline{mathscr{F}}_{T_{1}}
A=T1⊂T⋃FT1
表示
Ω
Omega
Ω 中有限维可测基底的柱集全体, 则由1.3的结果可知,
C
⊂
A
,
mathscr{C} subset mathscr{A},
C⊂A, 且
A
mathscr{A}
A 为域,
σ
(
C
)
=
σ
(
A
)
=
F
sigma(mathscr{C})=sigma(mathscr{A})=mathscr{F}
σ(C)=σ(A)=F
若
T
1
⊂
T
2
⊂
T
,
T_{1} subset T_{2} subset T,
T1⊂T2⊂T, 规定
Ω
T
2
Omega_{T_{2}}
ΩT2 到
Ω
T
1
Omega_{T_{1}}
ΩT1 的映照
π
T
1
T
2
pi_{T_{1}}^{T_{2}}
πT1T2 如下:
π
T
1
T
2
{
ω
t
:
t
∈
T
2
}
=
{
ω
t
:
t
∈
T
1
}
pi_{T_{1}}^{T_{2}}left{omega_{t}: t in T_{2}right}=left{omega_{t}: t in T_{1}right}
πT1T2{ωt:t∈T2}={ωt:t∈T1}
则
π
T
1
T
2
pi_{T_{1}}^{T_{2}}
πT1T2 是
Ω
T
2
Omega_{T_{2}}
ΩT2 到
Ω
T
1
Omega_{T_{1}}
ΩT1 的投影.
对
A
∈
F
T
1
,
A in mathscr{F}_{T_{1}},
A∈FT1, 由于
(
π
T
1
T
2
)
−
1
A
=
A
×
Ω
T
2
T
1
left(pi_{T_{1}}^{T_{2}}right)^{-1} A=A times Omega_{T_{2} backslash T_{1}}
(πT1T2)−1A=A×ΩT2T1
因此
π
T
1
T
2
pi_{T_{1}}^{T_{2}}
πT1T2 是
(
Ω
T
2
,
F
T
2
)
left(Omega_{T_{2}}, mathscr{F}_{T_{2}}right)
(ΩT2,FT2) 到
(
Ω
T
1
,
F
T
1
)
left(Omega_{T_{1}}, mathscr{F}_{T_{1}}right)
(ΩT1,FT1) 的可测映照, 且
(
π
T
1
T
2
)
−
1
F
T
1
=
F
‾
T
1
T
2
left(pi_{T_{1}}^{T_{2}}right)^{-1} mathscr{F}_{T_{1}}=overline{mathscr{F}}_{T_{1}}^{T_{2}}
(πT1T2)−1FT1=FT1T2
(证明:把
F
‾
T
1
T
2
overline{mathscr{F}}_{T_{1}}^{T_{2}}
FT1T2的元素写出来就可以了)
若
P
T
2
P_{T_{2}}
PT2 为
(
Ω
T
2
,
F
T
2
)
left(Omega_{T_{2}}, mathscr{F}_{T_{2}}right)
(ΩT2,FT2) 上的概率, 则
P
T
2
(
π
T
1
T
2
)
−
1
P_{T_{2}}left(pi_{T_{1}}^{T_{2}}right)^{-1}
PT2(πT1T2)−1 为
(
Ω
T
1
,
F
T
1
)
left(Omega_{T_{1}}, mathscr{F}_{T_{1}}right)
(ΩT1,FT1) 上的概率.
(证明:
P
T
2
(
π
T
1
T
2
)
−
1
(
Ω
T
1
)
=
P
T
2
Ω
T
2
=
1
P_{T_{2}}left(pi_{T_{1}}^{T_{2}}right)^{-1}(Omega_{T_{1}})=P_{T_{2}}Omega_{T_{2}}=1
PT2(πT1T2)−1(ΩT1)=PT2ΩT2=1)
反之,对
B
∈
F
‾
T
1
T
2
B in overline{mathscr{F}}_{T_{1}}^{T_{2}}
B∈FT1T2 必存在完全确定的
A
=
π
T
1
T
2
(
B
)
∈
F
T
1
,
A=pi_{T_{1}}^{T_{2}}(B) in mathscr{F}_{T_{1}},
A=πT1T2(B)∈FT1, 使
B
=
(
π
T
1
T
2
)
−
1
A
.
B=left(pi_{T_{1}}^{T_{2}}right)^{-1} A .
B=(πT1T2)−1A. 所以,
对
(
Ω
T
1
,
F
T
1
)
left(Omega_{T_{1}}, mathscr{F}_{T_{1}}right)
(ΩT1,FT1) 上的概率
P
T
1
,
P_{T_{1}},
PT1, 由
Q
(
B
)
=
Q
(
(
π
T
1
T
2
)
−
1
(
A
)
)
=
P
(
A
)
Q(B)=Qleft(left(pi_{T_{1}}^{T_{2}}right)^{-1}(A)right)=P(A)
Q(B)=Q((πT1T2)−1(A))=P(A)
亦可完全确定地在
(
Ω
T
2
,
F
‾
T
1
T
2
)
left(Omega_{T_{2}}, overline{mathscr{F}}_{T_{1}}^{T_{2}}right)
(ΩT2,FT1T2) 上规定一个概率
Q
.
Q .
Q. 特别地, 由
P
T
1
P_{T_{1}}
PT1 可在
(
Ω
,
F
‾
T
1
)
left(Omega, overline{mathcal{F}}_{T_{1}}right)
(Ω,FT1)
上规定一个概率.
命题 2.5.3
设对
T
T
T 的有限子集
T
1
=
{
t
1
,
⋯
,
t
n
}
⊂
T
,
T_{1}=left{t_{1}, cdots, t_{n}right} subset T,
T1={t1,⋯,tn}⊂T, 以
P
T
1
P_{T_{1}}
PT1 表示
(
Ω
T
1
,
F
T
1
)
left(Omega_{T_{1}}, mathscr{F}_{T_{1}}right)
(ΩT1,FT1)
上的概率测度, 则:
对测度族
{
P
T
1
,
left{P_{T_{1}},right.
{PT1, 有限
T
1
⊂
T
}
,
left.T_{1} subset Tright},
T1⊂T}, 在
(
Ω
,
F
)
(Omega, mathscr{F})
(Ω,F) 上存在非负有限可加集函数
P
,
P,
P, 对每个有限
T
1
⊂
T
T_{1} subset Tquad
T1⊂T满足
P
(
π
T
1
T
)
−
1
=
P
T
1
Pleft(pi_{T_{1}}^{T}right)^{-1}=P_{T_{1}}quad
P(πT1T)−1=PT1
的充要条件是
{
P
T
1
,
T
1
⊂
T
}
left{P_{T_{1}}, T_{1} subset Tright}
{PT1,T1⊂T} 满足下列相容性条件:对
T
T
T 任意有限子集
T
1
,
T
2
T_{1}, T_{2}
T1,T2,
T
1
⊂
T
2
,
T_{1} subset T_{2},
T1⊂T2, 有
P
T
2
(
π
T
1
T
2
)
−
1
=
P
T
1
P_{T_{2}}left(pi_{T_{1}}^{T_{2}}right)^{-1}=P_{T_{1}}
PT2(πT1T2)−1=PT1
证明:
⇒
Rightarrow
⇒ 由于
π
T
1
T
2
∘
π
T
2
T
,
pi_{T_{1}}^{T_{2}} circ pi_{T_{2}}^{T},
πT1T2∘πT2T, 故对
A
∈
F
T
1
,
A in mathscr{F}_{T_{1}},
A∈FT1,
(
π
T
1
T
)
−
1
A
=
(
π
T
2
T
)
−
1
(
π
T
1
T
2
)
−
1
A
left(pi_{T_{1}}^{T}right)^{-1} A=left(pi_{T_{2}}^{T}right)^{-1}left(pi_{T_{1}}^{T_{2}}right)^{-1} A
(πT1T)−1A=(πT2T)−1(πT1T2)−1A
因此, 若存在
P
P
P 满足相容性, 必有
P
T
1
(
A
)
=
P
(
(
π
T
1
T
)
−
1
A
)
=
P
(
(
π
T
2
T
)
−
1
(
π
T
1
T
2
)
−
1
A
)
=
P
T
2
(
(
π
T
1
T
2
)
−
1
A
)
P_{T_{1}}(A)=Pleft(left(pi_{T_{1}}^{T}right)^{-1} Aright)=Pleft(left(pi_{T_{2}}^{T}right)^{-1}left(pi_{T_{1}}^{T_{2}}right)^{-1} Aright)=P_{T_{2}}left(left(pi_{T_{1}}^{T_{2}}right)^{-1} Aright)
PT1(A)=P((πT1T)−1A)=P((πT2T)−1(πT1T2)−1A)=PT2((πT1T2)−1A)
所以
⇒
Rightarrow
⇒方向成立。
⇐
Leftarrowquad
⇐首先找到测度
P
P
P。考虑
B
∈
A
B in mathscr{A}
B∈A:
B
=
A
1
×
Ω
T
−
T
1
=
A
2
×
Ω
T
−
T
2
,
A
1
∈
F
T
1
,
A
2
∈
F
T
2
B=A_{1} times Omega_{T-T_{1}}=A_{2} times Omega_{T-T_{2}}, quad A_{1} in mathscr{F}_{T_{1}}, A_{2} in mathscr{F}_{T_{2}}
B=A1×ΩT−T1=A2×ΩT−T2,A1∈FT1,A2∈FT2
其中
T
1
,
T
2
T_{1}, T_{2}
T1,T2 都是
T
T
T 的有限子集,则可取
S
=
T
1
∪
T
2
S=T_{1} cup T_{2}
S=T1∪T2,将
B
B
B 记为
(
(
π
T
1
S
)
−
1
A
1
)
×
Ω
T
−
S
=
A
1
×
Ω
T
−
T
1
=
B
=
A
2
×
Ω
T
−
T
2
=
(
(
π
T
2
S
)
−
1
A
2
)
×
Ω
T
−
S
left(left(pi_{T_{1}}^{S}right)^{-1} A_{1}right) times Omega_{T-S}=A_{1} times Omega_{T-T_{1}}=B=A_{2} times Omega_{T-T_{2}}=left(left(pi_{T_{2}}^{S}right)^{-1} A_{2}right) times Omega_{T-S}
((πT1S)−1A1)×ΩT−S=A1×ΩT−T1=B=A2×ΩT−T2=((πT2S)−1A2)×ΩT−S
因此
(
π
T
1
S
)
−
1
A
1
=
(
π
T
2
S
)
−
1
A
2
,
left(pi_{T_{1}}^{S}right)^{-1} A_{1}=left(pi_{T_{2}}^{S}right)^{-1} A_{2},
(πT1S)−1A1=(πT2S)−1A2, 由
(
2.5.15
)
(2.5 .15)
(2.5.15) 式,
P
T
1
(
A
1
)
=
P
S
(
(
π
T
1
S
)
−
1
A
1
)
=
P
S
(
(
π
T
2
S
)
−
1
A
2
)
=
P
T
2
(
A
2
)
P_{T_{1}}left(A_{1}right)=P_{S}left(left(pi_{T_{1}}^{S}right)^{-1} A_{1}right)=P_{S}left(left(pi_{T_{2}}^{S}right)^{-1} A_{2}right)=P_{T_{2}}left(A_{2}right)
PT1(A1)=PS((πT1S)−1A1)=PS((πT2S)−1A2)=PT2(A2)
所以对有限维可测基底柱集
B
B
B, 虽然其基底不是唯一的, 但取其任一基底, 规定
P
(
B
)
=
P
T
1
(
A
1
)
P(B)=P_{T_{1}}left(A_{1}right)
P(B)=PT1(A1)
是
F
T
1
mathscr{F}_{T_{1}}
FT1上的概率,因此也就是
F
‾
T
1
overline{mathscr{F}}_{T_{1}}
FT1 上的概率,再证明
A
mathscr{A}
A上的有限可加性,注意这里
P
P
P还只是一个有限可加的集函数,不是测度。
定理 2.5.4
{
(
Ω
t
,
F
t
,
P
t
)
,
t
∈
T
}
left{left(Omega_{t}, mathscr{F}_{t}, P_{t}right), t in Tright}
{(Ωt,Ft,Pt),t∈T} 为一族概率空间, 则在乘积可测空间
(
Ω
T
,
F
T
)
left(Omega_{T}, mathscr{F}_{T}right)
(ΩT,FT) 上存在唯一概率测度
P
,
P,
P, 满足对每个有限
T
1
⊂
T
T_{1} subset T
T1⊂T
P
(
π
T
1
T
)
−
1
=
∏
t
∈
T
1
P
t
Pleft(pi_{T_{1}}^{T}right)^{-1}=underset{t in T_{1}}{mathrm{prod}} P_{t}
P(πT1T)−1=t∈T1∏Pt
注意,
∏
t
∈
T
1
P
t
underset{t in T_{1}}{mathrm{prod}} P_{t}
t∈T1∏Pt 是有限乘积空间的测度,可以用于Fubini定理。
证明:所以若 P P P 存在,因为有限维可测矩形柱集全体 C mathscr{C} C 是一个 π pi π 类, σ ( C ) = F sigma(mathscr{C})=mathscr{F} σ(C)=F ,又由 P ( π T 1 T ) − 1 = ∏ t ∈ T 1 P t Pleft(pi_{T_{1}}^{T}right)^{-1}=underset{t in T_{1}}{mathrm{prod}} P_{t} P(πT1T)−1=t∈T1∏Pt 可以保证在有限维可测矩形柱集上的是唯一确定的,因此唯一性可以保证。
对有限 T 1 ⊂ T , T_{1} subset T, T1⊂T, 取 P T 1 = ∏ t ∈ T 1 P t , P_{T_{1}}=underset{t in T_{1}}{prod} P_{t}, PT1=t∈T1∏Pt, 则容易说明 { P T 1 , left{P_{T_{1}},right. {PT1, 有限 T 1 ⊂ T } left.T_{1} subset Tright} T1⊂T} 满足相容性条件,因此由命题 2.5.3可知,可规定 P P P使之在 A mathscr{A} A 上为有限可加的,且 P ( Ω T ) = 1 Pleft(Omega_{T}right)=1 P(ΩT)=1。因而还只需要证明 P P P 在 A mathscr{A} A 上是 σ sigma σ 可加的。
为此, 在下面将证明, 对
A
mathscr{A}
A 中每个递减到
∅
varnothing
∅ 的序列
{
A
n
}
left{A_{n}right}
{An} 都有
lim
n
→
∞
P
(
A
n
)
=
0
lim _{n rightarrow infty} Pleft(A_{n}right)=0
n→∞limP(An)=0
由于 A n A_{n} An 都是有限维柱集, 因此不妨设有 T T T 的可列子集 { t n , n ≥ 1 } left{t_{n}, n geq 1right} {tn,n≥1} 使每个 A n A_{n} An 都是基底在 F t 1 ⋯ t n mathscr{F}_{t_{1} cdots t_{n}} Ft1⋯tn 的有限维柱集。(可数个可数集合的并是可数的,如果 A n A_{n} An 的基底不够则用 Ω t n Omega_{t_{n}} Ωtn 代替)。
若
T
n
T_{n}
Tn 为
T
T
T 的有限子集, 记
Ω
T
n
=
∏
t
∈
T
n
Ω
t
,
P
T
n
=
∏
t
∈
T
n
P
t
Omega_{T_{n}}=underset{t in T_{n}}{prod} Omega_{t}, quad P_{T_{n}}=underset{t in T_{n}}{prod} P_{t}
ΩTn=t∈Tn∏Ωt,PTn=t∈Tn∏Pt
又令
P
T
−
T
n
P_{T-T_{n}}
PT−Tn 为
Ω
T
−
T
n
Omega_{T-T_{n}}
ΩT−Tn 上对有限维可测柱集规定的可加集函数, 它满足
P
T
−
T
n
(
π
u
1
⋯
u
n
T
−
T
n
)
−
1
=
∏
i
=
1
n
P
u
i
,
u
1
⋯
,
u
n
∈
T
−
T
n
P_{T-T_{n}}left(pi_{u_{1} cdots u_{n}}^{T-T_{n}}right)^{-1}=prod_{i=1}^{n} P_{u_{i}}, quad u_{1} cdots, u_{n} in T-T_{n}
PT−Tn(πu1⋯unT−Tn)−1=i=1∏nPui,u1⋯,un∈T−Tn
由 Fubini 定理,对有限维可测柱集 A = B × Ω T − T k , B ∈ F T k A=B times Omega_{T-T_{k}}, B in mathscr{F}_{T_{k}} A=B×ΩT−Tk,B∈FTk 若 T n = { t 1 , ⋯ , t n } ⊂ T k , A ( ω t 1 , ⋯ , ω t n ) T_{n}=left{t_{1}, cdots, t_{n}right} subset T_{k}, Aleft(omega_{t_{1}}, cdots, omega_{t_{n}}right) Tn={t1,⋯,tn}⊂Tk,A(ωt1,⋯,ωtn) 表示 A A A 的 ( ω t 1 , ⋯ , ω t n ) left(omega_{t_{1}}, cdots, omega_{t_{n}}right) (ωt1,⋯,ωtn) 截口。 A ( ω t 1 , ⋯ , ω t n ) Aleft(omega_{t_{1}}, cdots, omega_{t_{n}}right) A(ωt1,⋯,ωtn)表示 A A A 集合里固定 ( ω t 1 , ⋯ , ω t n ) left(omega_{t_{1}}, cdots, omega_{t_{n}}right) (ωt1,⋯,ωtn) 的那些元素。
P
T
k
=
P
T
k
−
T
n
×
P
T
n
P_{T_{k}}=P_{T_{k}-T_{n}} times P_{T_{n}}
PTk=PTk−Tn×PTn 都是有限维乘积空间上的测度,所以
P
T
k
(
B
)
=
∬
I
B
d
P
T
k
−
T
n
d
P
T
n
=
∫
P
T
k
−
T
n
(
B
(
ω
t
1
,
⋯
,
ω
t
n
)
)
d
P
T
n
P_{T_{k}}(B)=iint I_{B} mathrm{d} P_{T_{k}-T_{n}} mathrm{d} P_{T_{n}} \ =int P_{T_{k}-T_{n}}left(Bleft(omega_{t_{1}}, cdots, omega_{t_{n}}right)right) mathrm{d} P_{T_{n}}
PTk(B)=∬IBdPTk−TndPTn=∫PTk−Tn(B(ωt1,⋯,ωtn))dPTn
由于
(
π
T
k
−
T
n
T
)
−
1
B
(
ω
t
1
,
⋯
,
ω
t
n
)
=
A
(
ω
t
1
,
⋯
,
ω
t
n
)
left(pi_{T_{k}-T_{n}}^{T}right)^{-1}Bleft(omega_{t_{1}}, cdots, omega_{t_{n}}right)=Aleft(omega_{t_{1}}, cdots, omega_{t_{n}}right)
(πTk−TnT)−1B(ωt1,⋯,ωtn)=A(ωt1,⋯,ωtn)
所以
P
(
A
)
=
P
(
(
π
T
k
T
)
−
1
B
)
=
P
T
k
(
B
)
=
∫
P
T
−
T
n
(
A
(
ω
t
1
,
⋯
,
ω
t
n
)
)
d
P
T
n
begin{aligned} P(A) &=Pleft(left(pi_{T_{k}}^{T}right)^{-1} Bright)=P_{T_{k}}(B) & \ &=int P_{T-T_{n}}left(Aleft(omega_{t_{1}}, cdots, omega_{t_{n}}right)right) mathrm{d} P_{T_{n}} end{aligned}
P(A)=P((πTkT)−1B)=PTk(B)=∫PT−Tn(A(ωt1,⋯,ωtn))dPTn
下面用反证法来证明
lim
n
→
∞
P
(
A
n
)
=
0
lim _{n rightarrow infty} Pleft(A_{n}right)=0
n→∞limP(An)=0
若对某递减序列 { A n } left{A_{n}right} {An} 不成立,则有某个 ε > 0 varepsilon>0 ε>0 对每个 n n n 有 P ( A n ) ≥ ε Pleft(A_{n}right) geq varepsilon P(An)≥ε。
由于 A n A_{n} An 都是有限维柱集, 因此不妨设有 T T T 的可列子集 { t n , n ≥ 1 } left{t_{n}, n geq 1right} {tn,n≥1} 使每个 A n A_{n} An 都是基底在 F t 1 ⋯ t n mathscr{F}_{t_{1} cdots t_{n}} Ft1⋯tn 的有限维柱集。(可数个可数集合的并是可数的,如果 A n A_{n} An 的基底不够则用 Ω t n Omega_{t_{n}} Ωtn 代替)。
对于柱集列 { A n } left{A_{n}right} {An} 做简单调整后, 我们总可以选取参数列 { t 1 ⋯ t n } ⊂ T left{t_{1} cdots t_{n}right} subset T {t1⋯tn}⊂T,设柱集 A n A_{n} An 的基底是 { t 1 ⋯ t n } left{t_{1} cdots t_{n}right} {t1⋯tn},由于 { A n } left{A_{n}right} {An} 是递减的,那么柱集 A n + 1 A_{n+1} An+1 的基底包含 { t 1 ⋯ t n } left{t_{1} cdots t_{n}right} {t1⋯tn}。由此我们可以得到一个递增的基底序列: { t 1 ⋯ t n } ⊂ { t 1 ⋯ t n , ⋯ t s } ⊂ ⋯ left{t_{1} cdots t_{n}right}subset left{t_{1} cdots t_{n}, cdots t_sright} subset cdots {t1⋯tn}⊂{t1⋯tn,⋯ts}⊂⋯。规定 { A n } left{A_{n}right} {An} 的基底都在 T T T 上,用这些有限的基底分别作用上去:
令
B
n
=
{
(
ω
t
1
…
ω
t
n
1
)
:
P
T
−
{
t
1
⋯
t
n
1
}
(
A
n
(
ω
t
1
…
ω
t
n
1
)
)
≥
ε
/
2
}
B_{n}=left{left(omega_{t_{1}}dotsomega_{t_{n_1 }}right): P_{T-left{t_{1} cdots t_{n_1}right}}left(A_{n}left(omega_{t_{1}}dotsomega_{t_{n_1}}right)right) geq varepsilon / 2right}
Bn={(ωt1…ωtn1):PT−{t1⋯tn1}(An(ωt1…ωtn1))≥ε/2}
代入上面Fubini定理得到的公式:
ε
≤
P
(
A
n
)
=
∫
P
T
−
{
t
1
⋯
t
n
1
}
(
A
(
ω
t
1
,
⋯
,
ω
t
n
1
)
)
d
P
{
t
1
⋯
t
n
1
}
=
∫
B
n
P
T
−
{
t
1
⋯
t
n
1
}
(
A
(
ω
t
1
,
⋯
,
ω
t
n
1
)
)
d
P
{
t
1
⋯
t
n
1
}
+
∫
B
n
c
P
T
−
{
t
1
⋯
t
n
1
}
(
A
(
ω
t
1
,
⋯
,
ω
t
n
1
)
)
d
P
{
t
1
⋯
t
n
1
}
≤
P
{
t
1
⋯
t
n
1
}
(
B
n
)
+
ε
/
2
begin{aligned} varepsilon leq P(A_n) &=int P_{T-left{t_{1} cdots t_{n_1}right}}left(Aleft(omega_{t_{1}}, cdots, omega_{t_{n_1}}right)right) mathrm{d} P_{left{t_{1} cdots t_{n_1}right}}\ &=int_{B_{n}}P_{T-left{t_{1} cdots t_{n_1}right}}left(Aleft(omega_{t_{1}}, cdots, omega_{t_{n_1}}right)right) mathrm{d} P_{left{t_{1} cdots t_{n_1}right}}+int_{B_{n}^{c}} P_{T-left{t_{1} cdots t_{n_1}right}}left(Aleft(omega_{t_{1}}, cdots, omega_{t_{n_1}}right)right) mathrm{d} P_{left{t_{1} cdots t_{n_1}right}} \&leq P_{left{t_{1} cdots t_{n_1}right}}left(B_{n}right)+varepsilon / 2 end{aligned}
ε≤P(An)=∫PT−{t1⋯tn1}(A(ωt1,⋯,ωtn1))dP{t1⋯tn1}=∫BnPT−{t1⋯tn1}(A(ωt1,⋯,ωtn1))dP{t1⋯tn1}+∫BncPT−{t1⋯tn1}(A(ωt1,⋯,ωtn1))dP{t1⋯tn1}≤P{t1⋯tn1}(Bn)+ε/2
(把概率放大成1)
故
P
{
t
1
⋯
t
n
1
}
(
B
n
)
≥
ε
/
2
P_{left{t_{1} cdots t_{n_1}right}}left(B_{n}right) geq varepsilon / 2
P{t1⋯tn1}(Bn)≥ε/2。 又由于
{
B
n
}
left{B_{n}right}
{Bn} 为
F
{
t
1
⋯
t
n
1
}
mathscr{F}_{left{t_{1} cdots t_{n_1}right}}
F{t1⋯tn1} 中递减序列(
A
n
A_{n}
An 递减),因而由
P
{
t
1
⋯
t
n
1
}
P_{left{t_{1} cdots t_{n_1}right}}
P{t1⋯tn1} 为
(
Ω
{
t
1
⋯
t
n
1
}
,
F
{
t
1
⋯
t
n
1
}
)
left(Omega_{left{t_{1} cdots t_{n_1}right}}, mathscr{F}_{left{t_{1} cdots t_{n_1}right}}right)
(Ω{t1⋯tn1},F{t1⋯tn1}) 上的概率可知,必存在
ω
ˉ
{
t
1
⋯
t
n
1
}
∈
⋂
n
B
n
bar{omega}_{{left{t_{1} cdots t_{n_1}right}}} in bigcap_{n} B_{n}
ωˉ{t1⋯tn1}∈⋂nBn 这时
P
T
−
{
t
1
⋯
t
n
1
}
(
A
n
(
ω
ˉ
t
1
…
ω
ˉ
t
n
1
)
)
≥
ε
/
2
,
n
≥
1
P_{T-{left{t_{1} cdots t_{n_1}right}}}left(A_{n}left(bar{omega}_{t_{1}}dots bar{omega}_{t_{n_1}}right)right) geq varepsilon / 2, quad n geq 1
PT−{t1⋯tn1}(An(ωˉt1…ωˉtn1))≥ε/2,n≥1
把刚才用于
Ω
T
,
A
n
,
ε
Omega_{T}, A_{n}, varepsilon
ΩT,An,ε 的论证, 现在用于
Ω
T
−
{
t
1
⋯
t
n
1
}
,
A
n
(
ω
ˉ
t
1
…
ω
ˉ
t
n
1
)
,
ε
/
2
,
Omega_{T-left{t_{1}cdots t_{n_1}right}}, A_{n}left(bar{omega}_{t_{1}}dots bar{omega}_{t_{n_1}}right), varepsilon / 2,
ΩT−{t1⋯tn1},An(ωˉt1…ωˉtn1),ε/2, 可得
ω
ˉ
t
n
1
+
1
⋯
t
n
2
bar{omega}_{t_{n_1+1}cdots t_{n_2}}
ωˉtn1+1⋯tn2 满足:
P
T
−
{
t
1
⋯
t
n
1
⋯
t
n
2
}
(
A
n
(
ω
ˉ
t
1
⋯
t
n
1
,
ω
ˉ
t
n
1
+
1
⋯
t
n
2
)
)
≥
ε
/
4
,
n
≥
1
P_{T-left{t_{1}cdots t_{n_1}cdots t_{n_2}right}}left(A_{n}left(bar{omega}_{t_{1}cdots t_{n_1}}, bar{omega}_{t_{n_1+1}cdots t_{n_2}}right)right) geq varepsilon / 4, quad n geq 1
PT−{t1⋯tn1⋯tn2}(An(ωˉt1⋯tn1,ωˉtn1+1⋯tn2))≥ε/4,n≥1
继续这一过程可得
P
T
−
T
j
(
A
n
(
ω
^
1
,
⋯
,
ω
^
j
)
)
≥
ε
/
2
j
,
n
≥
1
,
j
≥
1
P_{T-T_j}left(A_{n}left(hat{omega}_{1}, cdots, hat{omega}_{j}right)right) geq varepsilon / 2^{j}, quad n geq 1, quad j geq 1
PT−Tj(An(ω^1,⋯,ω^j))≥ε/2j,n≥1,j≥1
这里:
T
j
=
{
t
1
⋯
t
n
1
⋯
t
n
j
}
,
ω
^
j
=
ω
ˉ
t
n
j
−
1
+
1
⋯
t
n
j
T_j={t_1cdots t_{n_1}cdots t_{n_j}}, hat{omega}_{j}=bar{omega}_{t_{n_{j-1}+1}cdots t_{n_j}}
Tj={t1⋯tn1⋯tnj},ω^j=ωˉtnj−1+1⋯tnj
构造 ω ˉ = ( ω t , t ∈ T ) bar{omega}=left(omega_{t}, t in Tright) ωˉ=(ωt,t∈T),如果 t ∉ T j , j ≥ 1 t notin T_{j}, j geq 1 t∈/Tj,j≥1,分量在 Ω t Omega_{t} Ωt 任取,如果 t ∈ T j , ∃ j ≥ 1 t in T_{j}, exist j geq 1 t∈Tj,∃j≥1,那么分量取 ω ^ j hat{omega}_j ω^j。
接着将证明
ω
ˉ
∈
⋂
n
=
1
∞
A
n
bar{omega}in bigcap_{n=1}^{infty} A_{n}
ωˉ∈n=1⋂∞An
由递减性,
A
1
A_{1}
A1 是基底在
F
T
1
mathscr{F}_{T_{1}}
FT1 的柱集,
P
T
−
T
1
(
A
1
(
ω
^
1
)
)
≥
ε
/
2
P_{T-T_{1}}left(A_{1}left(hat{omega}_{1}right)right) geq varepsilon / 2
PT−T1(A1(ω^1))≥ε/2。故
A
1
(
ω
^
1
)
A_{1}left(hat{omega}_{1}right)
A1(ω^1)非空,
ω
ˉ
∈
A
1
bar{omega} in A_{1}
ωˉ∈A1。 同理
P
T
−
T
j
(
A
j
(
w
^
1
,
⋯
,
ω
^
j
)
)
≥
ε
/
2
j
P_{T-T_j}left(A_{j}left(hat{w}_{1}, cdots, hat{omega}_{j}right)right) geq varepsilon / 2^{j}
PT−Tj(Aj(w^1,⋯,ω^j))≥ε/2j
故
A
j
(
ω
^
1
,
⋯
,
ω
^
j
)
A_{j}left(hat{omega}_{1}, cdots, hat{omega}_{j}right)
Aj(ω^1,⋯,ω^j) 非空(
A
j
A_j
Aj 是柱集,它的一个截口非空 )。所以有
ω
ˉ
∈
A
j
bar{omega} in A_{j}
ωˉ∈Aj 。这和
A
n
A_n
An 递减到空集矛盾。
最后
以上就是欣慰白猫为你收集整理的无限维乘积空间(彼此独立)的测度(完)的全部内容,希望文章能够帮你解决无限维乘积空间(彼此独立)的测度(完)所遇到的程序开发问题。
如果觉得靠谱客网站的内容还不错,欢迎将靠谱客网站推荐给程序员好友。
发表评论 取消回复