根据多元正态分布的性质:AX+d~N(Aμ+d,AΣA')
分别计算出Aμ+d和AΣA’得:
Aμ+d =
(
2
1
)
\binom{2}{1}
(12)
AΣA’ = ( 3 − 1 − 1 1 ) \left(\begin{matrix}3&-1\\-1&1\\\end{matrix}\right) (3−1−11)
(1) 记A=(1 1),B=(1 -1),则X1+X2=AX,X1-X2=BX;
根据多元正态分布的性质:协方差为零的分量间相互独立
,只需求得AX与BX的协方差为零即可证明相互独立;
Cov(AX,BX) = AΣB'
= σ²(1+ρ 1+ρ)(1 -1)’ = 0
得证
(2) 即:求AX和BX的分布
AX~N(Aμ,AΣA’),BX ~N(Bμ,BΣB’)
所以X1+X2~N( μ1+μ2,2σ²(1+ρ)),X1-X2 ~N( μ1-μ2,2σ²(1-ρ))
和2-2类似
(1)记A=(
I
p
{I_p}
Ip
I
p
{I_p}
Ip),B=(
I
p
{I_p}
Ip -
I
p
{I_p}
Ip),则X1+X2=AX,X1-X2=BX;
Cov(AX,BX)=AΣB’=(
I
p
{I_p}
Ip
I
p
{I_p}
Ip)
(
Σ
1
Σ
2
Σ
2
Σ
1
)
\left(\begin{matrix} Σ1 &Σ2\\Σ2&Σ1\\\end{matrix}\right)
(Σ1Σ2Σ2Σ1) (
I
p
{I_p}
Ip
−
I
p
{-I_p}
−Ip)’ = 0
得证
解法二:
记Y= ( I p I p I p − I p ) \left(\begin{matrix} {I_p} &{I_p}\\{I_p}&{-I_p}\\\end{matrix}\right) (IpIpIp−Ip) ( X 1 X 2 ) \binom{X1} {X2} (X2X1)= ( Y 1 Y 2 ) \binom{Y1} {Y2} (Y2Y1),则Y的协方差矩阵Var(Y)= ( I p I p I p − I p ) \left(\begin{matrix} {I_p} &{I_p}\\{I_p}&{-I_p}\\\end{matrix}\right) (IpIpIp−Ip) ( Σ 1 Σ 2 Σ 2 Σ 1 ) \left(\begin{matrix} Σ1 &Σ2\\Σ2&Σ1\\\end{matrix}\right) (Σ1Σ2Σ2Σ1) ( I p I p I p − I p ) \left(\begin{matrix} {I_p} &{I_p}\\{I_p}&{-I_p}\\\end{matrix}\right) (IpIpIp−Ip)’ =
(
2
(
Σ
1
+
Σ
2
)
0
0
2
(
Σ
1
−
Σ
2
)
)
\left(\begin{matrix} 2(Σ1+Σ2) &0\\0&2(Σ1-Σ2)\\\end{matrix}\right)
(2(Σ1+Σ2)002(Σ1−Σ2))
因为协方差矩阵为分块对角阵,所以Y1和Y2相互独立,即X1+X2和X1-X2相互独立。
(2)
X1+X2~N(μ1+μ2,2(Σ1+Σ2) ),X1-X2 ~(μ1+μ2,2(Σ1-Σ2))
(1) 容易看出A=(3 -2 1)使得AX=3X1-2X2+X3,即求AX的分布:AX~N(Aμ,AΣA’)
(2) X3 = (0 0 1)X,a’(X1 X2)’ = (a1 a2 0)(X1 X2 X3)’ = (a1 a2 0)X
则:X3 - a’(X1 X2)’ = (-a1 -a2 1)X
记:B = (0 0 1) , C = (-a1 -a2 1),则X3=BX,X3 - a’(X1 X2)‘=CX
根据多元正态分布的性质:协方差为零的分量间相互独立
,则令Cov(BX,CX)为0即可求出满足题意的二维向量
Cov(BX,CX)=BΣC’=…= -a1-2a2+2 = 0
所以当a1+2a2=2时两向量相互独立。
(2)(3)(4)是相互独立的,(1)(5)不是
①必要性:
∵ ( Y Z ) \binom{Y} {Z} (ZY)= ( A B ) \binom{A} {B} (BA)X+ ( d c ) \binom{d} {c} (cd)
∴Y和Z服从多元正态分布
而Cov(Y,Z)=Cov(AX+d,BX+c)=AΣB’=0,根据多元正态分布的性质,可知Y与Z独立;
②充分性:
∵ ( Y Z ) \binom{Y} {Z} (ZY)= ( A B ) \binom{A} {B} (BA)X+ ( d c ) \binom{d} {c} (cd)
∴Y和Z服从多元正态分布
而Y与Z独立,可推出AΣB’=0
实际上就是要说明Y与Z服从多元正态分布。
比较系数即可:
已知二元正态分布公式:
则:σ1²σ2²(1-ρ²)=1,σ1²=1,σ2²=2 … …
σ1=1,σ2=
2
\sqrt{2}
2,ρ=1/
2
\sqrt{2}
2,μ1=4,μ2=3;
所以均值向量μ=
(
4
3
)
\binom{4}{3}
(34),协方差阵Σ=
(
1
−
1
−
1
2
)
\left(\begin{matrix}1&-1\\-1&2\\\end{matrix}\right)
(1−1−12)
(1)即证E(Z)=μ:
E(Z)=E(
∑
i
=
1
n
c
i
X
i
\sum\limits_{i=1}^n{c_iX_i}
i=1∑nciXi)=
∑
i
=
1
n
c
i
E
(
X
i
\sum\limits_{i=1}^n{c_iE(X_i}
i=1∑nciE(Xi)=E(
X
i
{X_i}
Xi)=μ
(2)
由(1)知:
μ
z
{μ_z}
μz=μ
Σ z {Σ_z} Σz=Var( ∑ i = 1 n c i X i \sum\limits_{i=1}^n{c_iX_i} i=1∑nciXi)= ∑ i = 1 n c i 2 V a r ( X i ) \sum\limits_{i=1}^n{c_i²Var(X_i)} i=1∑nci2Var(Xi)= ∑ i = 1 n c i 2 Σ \sum\limits_{i=1}^n{c_i²Σ} i=1∑nci2Σ=c’cΣ
因此,Z~ N p {N_p} Np(μ,c’cΣ)
r代码:
# 样本均值
mean_value <- colMeans(data[,c("X1","X2","X3")])
print(mean_value)
# 样本离差阵
centered_data <- scale(data[,c("X1","X2","X3")], scale = FALSE)
deviation_matrix <- t(centered_data) %*% centered_data
print(deviation_matrix)
# 样本协方差阵
covariance_matrix <- cov(data[,c("X1","X2","X3")],use = "everything")
print(covariance_matrix)
# 样本相关阵
correlation_matrix <- cor(data[,c("X1","X2","X3")],use = "all.obs")
print(correlation_matrix)
用Python也可以
结果: