第五章
5.1 大数定律
5.1.1 切比雪夫不等式
定理:若
E
(
X
)
E(X)
E(X),
D
(
X
)
D(X)
D(X)存在,
∀
ϵ
>
0
\forall \epsilon>0
∀ϵ>0,
P
(
∣
X
−
E
(
X
)
∣
≥
ϵ
)
≤
D
(
X
)
ϵ
2
P(|X-E(X)|\geq\epsilon)\leq\displaystyle\frac{D(X)}{\epsilon^2}
P(∣X−E(X)∣≥ϵ)≤ϵ2D(X)
(变形:)
P
(
∣
X
−
E
(
X
)
∣
≥
ϵ
)
≥
1
−
D
(
X
)
ϵ
2
P(|X-E(X)|\geq\epsilon)\geq1-\displaystyle\frac{D(X)}{\epsilon^2}
P(∣X−E(X)∣≥ϵ)≥1−ϵ2D(X)
例
【例1】白细胞平均:7300,标准差:700,求白细胞数在5200~9400的概率
解:
E
(
X
)
=
7300
,
D
(
X
)
=
490000
E(X)=7300,D(X)=490000
E(X)=7300,D(X)=490000
9400
−
7300
=
7300
−
5200
=
2100
9400-7300=7300-5200=2100
9400−7300=7300−5200=2100
故
ϵ
=
2100
\epsilon=2100
ϵ=2100
P
(
∣
X
−
7300
∣
≤
2100
)
≥
1
−
490000
210
0
2
P(|X-7300|\leq2100)\geq1-\displaystyle\frac{490000}{2100^2}
P(∣X−7300∣≤2100)≥1−21002490000
【例2】 E ( X ) = μ , D ( X ) = σ 2 > 0 E(X)=\mu,D(X)=\sigma^2>0 E(X)=μ,D(X)=σ2>0,求 P ( ∣ X − μ ∣ ≥ 3 σ ) P(|X-\mu|\geq3\sigma) P(∣X−μ∣≥3σ)
解:
P
(
∣
X
−
μ
∣
≥
3
σ
)
≤
D
(
X
)
(
3
σ
)
2
=
σ
2
9
σ
2
=
1
9
P(|X-\mu|\geq3\sigma)\leq\displaystyle\frac{D(X)}{(3\sigma)^2}=\displaystyle\frac{\sigma^2}{9\sigma^2}=\displaystyle\frac{1}{9}
P(∣X−μ∣≥3σ)≤(3σ)2D(X)=9σ2σ2=91
5.1.2 切比雪夫大数定律
收敛:
a
n
→
a
a_n\to a
an→a:
∀
ϵ
>
0
,
∃
N
>
0
,
n
>
N
\forall\epsilon>0,\exist N>0,n>N
∀ϵ>0,∃N>0,n>N时,
∣
a
n
−
a
∣
<
ϵ
|a_n-a|<\epsilon
∣an−a∣<ϵ
依概率收敛:
x
n
→
P
a
\displaystyle x_n\overset{P}{\to}a
xn→Pa:
∀
ϵ
>
0
,
∃
N
>
0
,
n
>
N
\forall\epsilon>0,\exist N>0,n>N
∀ϵ>0,∃N>0,n>N时,
lim
n
→
∞
P
(
∣
x
n
−
a
∣
<
ϵ
)
=
1
\lim\limits_{n\to\infin}P(|x_n-a|<\epsilon)=1
n→∞limP(∣xn−a∣<ϵ)=1
伯努利大数定律:
n重伯努利实验,A发生了
m
n
m_n
mn次,
lim
n
→
∞
P
(
∣
m
n
n
−
p
∣
<
ϵ
)
=
1
\lim\limits_{n\to\infin}P(|\displaystyle\frac{m_n}{n}-p|<\epsilon)=1
n→∞limP(∣nmn−p∣<ϵ)=1
(变形:)
lim
n
→
∞
P
(
∣
m
n
n
−
p
∣
≥
ϵ
)
=
0
\lim\limits_{n\to\infin}P(|\displaystyle\frac{m_n}{n}-p|\geq\epsilon)=0
n→∞limP(∣nmn−p∣≥ϵ)=0
即:频率收敛于概率
设
x
1
,
x
2
,
⋯
,
x
n
x_1,x_2,\cdots,x_n
x1,x2,⋯,xn独立同分布,
x
i
=
{
1
发
生
0
不
发
生
x_i=\begin{cases}1 & 发生\\0 & 不发生 \end{cases}
xi={10发生不发生,
则
E
(
X
i
)
=
p
,
D
(
X
i
)
=
p
(
1
−
p
)
,
m
n
=
∑
i
=
1
n
x
i
E(X_i)=p,D(X_i)=p(1-p),m_n=\sum\limits_{i=1}^{n}x_i
E(Xi)=p,D(Xi)=p(1−p),mn=i=1∑nxi,
m n n = 1 n ∑ i = 1 n x i , p = E ( 1 n ∑ x i ) = 1 n ∑ i = 1 n E ( X i ) \displaystyle\frac{m_n}{n}=\frac{1}{n}\sum\limits_{i=1}^{n}x_i,p=E(\displaystyle\frac{1}{n}\sum x_i)=\frac{1}{n}\sum\limits_{i=1}^{n}E(X_i) nmn=n1i=1∑nxi,p=E(n1∑xi)=n1i=1∑nE(Xi),
lim n → ∞ P ( ∣ 1 n ∑ i = 1 n x i − 1 n ∑ i = 1 n E ( X i ) ∣ < ϵ ) = 1 \displaystyle\lim\limits_{n\to\infin}P(|\frac{1}{n}\sum\limits_{i=1}^{n}x_i-\frac{1}{n}\sum\limits_{i=1}^{n}E(X_i)|<\epsilon)=1 n→∞limP(∣n1i=1∑nxi−n1i=1∑nE(Xi)∣<ϵ)=1
切比雪夫大数定律: x 1 , x 2 , ⋯ , x n x_1,x_2,\cdots,x_n x1,x2,⋯,xn是两两不相关的变量, E ( X i ) E(X_i) E(Xi)和 D ( X i ) D(X_i) D(Xi)都存在,方差有界, D ( X i ) ≤ M D(X_i)\leq M D(Xi)≤M, ∀ ϵ > 0 , lim n → ∞ P ( ∣ 1 n ∑ i = 1 n x i − 1 n ∑ i = 1 n E ( X i ) ∣ < ϵ ) = 1 \forall\epsilon>0,\displaystyle\lim\limits_{n\to\infin}P(|\frac{1}{n}\sum\limits_{i=1}^{n}x_i-\frac{1}{n}\sum\limits_{i=1}^{n}E(X_i)|<\epsilon)=1 ∀ϵ>0,n→∞limP(∣n1i=1∑nxi−n1i=1∑nE(Xi)∣<ϵ)=1
即:变量的均值收敛于期望的均值
推论: x 1 , x 2 , ⋯ , x n x_1,x_2,\cdots,x_n x1,x2,⋯,xn独立同分布, E ( X i ) = μ , D ( X i ) = σ 2 , ∀ ϵ > 0 , lim n → ∞ P ( ∣ 1 n ∑ i = 1 n x i − μ ∣ < ϵ ) = 1 E(X_i)=\mu,D(X_i)=\sigma^2,\forall\epsilon>0,\displaystyle\lim\limits_{n\to\infin}P(|\frac{1}{n}\sum\limits_{i=1}^{n}x_i-\mu|<\epsilon)=1 E(Xi)=μ,D(Xi)=σ2,∀ϵ>0,n→∞limP(∣n1i=1∑nxi−μ∣<ϵ)=1
辛钦大数定律: x 1 , x 2 , ⋯ , x n x_1,x_2,\cdots,x_n x1,x2,⋯,xn独立同分布, E ( X i ) = μ E(X_i)=\mu E(Xi)=μ,方差无要求, ∀ ϵ > 0 , lim n → ∞ P ( ∣ 1 n ∑ i = 1 n x i − μ ∣ < ϵ ) = 1 \forall\epsilon>0,\displaystyle\lim\limits_{n\to\infin}P(|\frac{1}{n}\sum\limits_{i=1}^{n}x_i-\mu|<\epsilon)=1 ∀ϵ>0,n→∞limP(∣n1i=1∑nxi−μ∣<ϵ)=1
即:平均数收敛于期望
5.2 中心极限定理
现象由大量相互独立的因素影响
大量独立同分布的变量和的极限分布是正态分布
定理: X 1 , X 2 , ⋯ , x n X_1,X_2,\cdots,x_n X1,X2,⋯,xn独立同分布, E ( X i ) = μ , D ( X i ) = σ 2 , 0 < σ 2 < + ∞ E(X_i)=\mu,D(X_i)=\sigma^2,0<\sigma^2<+\infin E(Xi)=μ,D(Xi)=σ2,0<σ2<+∞,则 lim n → ∞ P ( ∑ i = 1 n x i − n μ n σ ≤ x ) = Φ 0 ( x ) \displaystyle\lim\limits_{n\to\infin}P(\frac{\sum\limits_{i=1}^{n}x_i-n\mu}{\sqrt{n}\sigma}\leq x)=\Phi_{0}(x) n→∞limP(nσi=1∑nxi−nμ≤x)=Φ0(x)
设 Y = ∑ i = 1 n x i Y=\sum\limits_{i=1}^{n}x_i Y=i=1∑nxi,则 E ( Y ) = n μ , D ( Y ) = n σ 2 E(Y)=n\mu,D(Y)=n\sigma^2 E(Y)=nμ,D(Y)=nσ2
∑ i = 1 n x i − n μ n σ \displaystyle\frac{\sum\limits_{i=1}^{n}x_i-n\mu}{\sqrt{n}\sigma} nσi=1∑nxi−nμ可按照 N ( 0 , 1 ) N(0,1) N(0,1)近似, ∑ i = 1 n x i \sum\limits_{i=1}^{n}x_i i=1∑nxi按照 N ( n μ , n σ 2 ) N(n\mu,n\sigma^2) N(nμ,nσ2)计算
例
【例1】顾客100人,在
[
0
,
60
]
[0,60]
[0,60]均匀分布,独立,日销售额超过3500元的概率
解:
设
X
i
X_i
Xi为第
i
i
i人的消费,
E
(
X
i
)
=
30
,
D
(
X
i
)
=
6
0
2
÷
12
=
300
E(X_i)=30,D(X_i)=60^2\div12=300
E(Xi)=30,D(Xi)=602÷12=300
根据中心极限定理,
∑
x
i
−
3000
100
3
∼
近
似
N
(
0
,
1
)
\displaystyle\frac{\sum x_i-3000}{100\sqrt{3}}\overset{近似}{\sim}N(0,1)
1003∑xi−3000∼近似N(0,1)
P ( ∑ i = 1 100 x i > 3500 ) = 1 − P ( ∑ x i ≤ 3500 ) = 1 − P ( ∑ x i − 3000 100 3 ≤ 3500 − 3000 100 3 ) = 1 − Φ 0 ( 2.887 ) P(\sum\limits_{i=1}^{100}x_i>3500)=1-P(\sum x_i\leq3500)=1-P(\frac{\sum x_i-3000}{100\sqrt{3}}\leq\frac{3500-3000}{100\sqrt{3}})=1-\Phi_0(2.887) P(i=1∑100xi>3500)=1−P(∑xi≤3500)=1−P(1003∑xi−3000≤10033500−3000)=1−Φ0(2.887)
【例2】某人射击环数的分布律为:
X
10
9
8
P
0.5
0.3
0.2
\begin{array}{c|c} X & 10 & 9 & 8\\ \hline P & 0.5 & 0.3 & 0.2 \end{array}
XP100.590.380.2
求100次射击总环数为915~945的概率
解:
设
X
i
X_i
Xi为第
i
i
i次射击环数
E
(
X
)
=
9.3
,
D
(
X
)
=
0.61
E(X)=9.3,D(X)=0.61
E(X)=9.3,D(X)=0.61
∑
x
i
−
930
61
∼
近
似
N
(
0
,
1
)
\displaystyle\frac{\sum x_i-930}{\sqrt{61}}\overset{近似}{\sim}N(0,1)
61∑xi−930∼近似N(0,1)
P ( 915 ≤ ∑ x i ≤ 945 ) = P ( 915 − 930 61 ≤ ∑ x i − 930 61 ≤ 945 − 930 61 ) = P ( − 1.92 ≤ ∑ x i − 930 61 ≤ 1.92 ) = 2 Φ 0 ( 1.92 ) − 1 P(915\leq\sum x_i\leq945)=P(\displaystyle\frac{915-930}{\sqrt{61}}\leq\frac{\sum x_i-930}{\sqrt{61}}\leq\frac{945-930}{\sqrt{61}})=P(-1.92\leq\displaystyle\frac{\sum x_i-930}{\sqrt{61}}\leq1.92)=2\Phi_0(1.92)-1 P(915≤∑xi≤945)=P(61915−930≤61∑xi−930≤61945−930)=P(−1.92≤61∑xi−930≤1.92)=2Φ0(1.92)−1
棣莫弗-拉普拉斯定理: Y n Y_n Yn是参数为n,p的二项分布,则 lim n → ∞ P ( Y n − n p n p ( 1 − p ) ≤ x ) = Φ 0 ( x ) \displaystyle\lim\limits_{n\to\infin}P(\frac{Y_n-np}{\sqrt{np(1-p)}}\leq x)=\Phi_0(x) n→∞limP(np(1−p)Yn−np≤x)=Φ0(x)
例
【例3】保险中每人死亡概率0.005,有10000人参加,求死亡人数超过70人的概率
解:设 X X X表示死亡人数
- 二项分布法:
P ( X ≤ 70 ) ∑ k = 0 70 C 10000 k 0.00 5 k 0.99 5 10000 − k P(X\leq70)\displaystyle\sum\limits_{k=0}^{70}C_{10000}^{k}0.005^{k}0.995^{10000-k} P(X≤70)k=0∑70C10000k0.005k0.99510000−k - 正态分布法:
P ( X ≤ 70 ) = P ( X − n p n p ( 1 − p ) ≤ 70 − n p n p ( 1 − p ) ) = Φ 0 ( 2.84 ) P(X\leq70)=P(\frac{X-np}{\sqrt{np(1-p)}}\leq\frac{70-np}{\sqrt{np(1-p)}})=\Phi_0(2.84) P(X≤70)=P(np(1−p)X−np≤np(1−p)70−np)=Φ0(2.84)
(变形:) P ( X = k ) = P ( k − 1 2 < X < k + 1 2 ) = P ( k − 1 2 − n p n p ( 1 − p ) < X − n p n p ( 1 − p ) < k + 1 2 − n p n p ( 1 − p ) ) = Φ 0 ( k + 1 2 − n p n p ( 1 − p ) ) − Φ 0 ( k − 1 2 − n p n p ( 1 − p ) ) P(X=k)=P(k-\frac{1}{2}<X<k+\frac{1}{2})=P(\frac{k-\frac{1}{2}-np}{\sqrt{np(1-p)}}<\frac{X-np}{\sqrt{np(1-p)}}<\frac{k+\frac{1}{2}-np}{\sqrt{np(1-p)}})=\Phi_{0}(\frac{k+\frac{1}{2}-np}{\sqrt{np(1-p)}})-\Phi_0(\frac{k-\frac{1}{2}-np}{\sqrt{np(1-p)}}) P(X=k)=P(k−21<X<k+21)=P(np(1−p)k−21−np<np(1−p)X−np<np(1−p)k+21−np)=Φ0(np(1−p)k+21−np)−Φ0(np(1−p)k−21−np)
例
【例4】击中飞机的概率0.01,500发子弹,求恰好3发能击中的概率
解:
n
=
500
,
p
=
0.01
n=500,p=0.01
n=500,p=0.01
P
(
X
=
5
)
=
P
(
4.5
<
X
<
5.5
)
=
P
(
4.5
−
5
500
×
0.01
×
0.99
<
X
<
5.5
−
5
500
×
0.01
×
0.99
)
=
0.178
P(X=5)=P(4.5<X<5.5)=P(\frac{4.5-5}{\sqrt{500\times0.01\times0.99}}<X<\frac{5.5-5}{\sqrt{500\times0.01\times0.99}})=0.178
P(X=5)=P(4.5<X<5.5)=P(500×0.01×0.994.5−5<X<500×0.01×0.995.5−5)=0.178
二项分布近似:
n
n
n大,
n
p
np
np适中——泊松分布
n
n
n大,
n
p
np
np大——正态分布