- 随机过程本质就是一组与时间相关的随机变量,描述系统在 一系列时刻所处的状态。因此,我们首先补充关于多维随机变量(随机向量)的内容
- 注:本文中加粗大写字母代表矩阵(如 A , B , L \pmb{A,B,L} A,B,L) 或随机向量(如 X , Y \pmb{X,Y} X,Y);加粗小写字母代表数的向量(如 a , b \pmb{a,b} a,b);普通大写字母代表一个随机变量(如 X , Y X,Y X,Y)
1. 多维 r.v. 的期望和方差
- 记多维 r.v.s. 为
X
=
[
X
1
X
2
⋮
X
n
]
,
Y
=
[
Y
1
Y
2
⋮
Y
n
]
\pmb{X} = \begin{bmatrix}X_1\\X_2\\\vdots \\X_n \end{bmatrix}, \pmb{Y} = \begin{bmatrix}Y_1\\Y_2\\\vdots \\Y_n \end{bmatrix}
X=
X1X2⋮Xn
,Y=
Y1Y2⋮Yn
期望
: E X = [ E X 1 E X 2 ⋮ E X n ] E\pmb{X} = \begin{bmatrix}EX_1\\EX_2\\\vdots \\EX_n \end{bmatrix} EX= EX1EX2⋮EXn 方差(矩阵)
:
Var ( X ) = ( Cov ( X i , X j ) ) n × n = ( E [ ( X i − E X i ) ( X j − E X j ) ] ) n × n = E [ ( X − E X ) ( X − E X ) ⊤ ] \begin{aligned} \text{Var}(\pmb{X}) &= \big(\text{Cov}(X_i,X_j)\big)_{n\times n} \\ &=\big(E\big[(X_i-EX_i)(X_j-EX_j)\big]\big)_{n\times n} \\ &= E\big[(\pmb{X}-E\pmb{X})(\pmb{X}-E\pmb{X})^\top \big]\end{aligned} Var(X)=(Cov(Xi,Xj))n×n=(E[(Xi−EXi)(Xj−EXj)])n×n=E[(X−EX)(X−EX)⊤]协方差(矩阵)
:
Cov ( X , Y ) = ( Cov ( X i , Y j ) ) n × n = ( E [ ( X i − E X i ) ( Y j − E Y j ) ] ) n × n = E [ ( X − E X ) ( Y − E Y ) ⊤ ] \begin{aligned} \text{Cov}(\pmb{X},\pmb{Y}) &= (\text{Cov}(X_i,Y_j))_{n\times n} \\ &=\big(E\big[(X_i-EX_i)(Y_j-EY_j)\big]\big)_{n\times n} \\ &= E\big[(\pmb{X}-E\pmb{X})(\pmb{Y}-E\pmb{Y})^\top\big] \end{aligned} Cov(X,Y)=(Cov(Xi,Yj))n×n=(E[(Xi−EXi)(Yj−EYj)])n×n=E[(X−EX)(Y−EY)⊤]
- 方差矩阵性质:
Var
(
A
X
)
=
A
Var
(
X
)
A
⊤
\text{Var}(\pmb{AX}) = \pmb{A} \text{Var}(\pmb{X})\pmb{A}^\top
Var(AX)=AVar(X)A⊤,证明如下
Var ( A X ) = E [ ( A X − E A X ) ( A X − E A X ) ⊤ ] = E [ ( A X − A E X ) ( A X − A E X ) ⊤ ] = A E [ ( X − E X ) ( A ( X − E X ) ⊤ ) ] = A E [ ( X − E X ) ( X − E X ) ⊤ A ⊤ ] = A E [ ( X − E X ) ( X − E X ) ⊤ ] A ⊤ = A Var ( X ) A ⊤ \begin{aligned} \text{Var}(\pmb{AX}) &= E\big[(\pmb{AX}-E\pmb{AX})(\pmb{AX}-E\pmb{AX})^\top\big]\\ &= E\big[(\pmb{AX}-\pmb{A}E\pmb{X})(\pmb{AX}-\pmb{A}E\pmb{X})^\top\big]\\ &= \pmb{A}E\big[(\pmb{X}-E\pmb{X})(\pmb{A}(\pmb{X}-E\pmb{X})^\top)\big]\\ &= \pmb{A}E\big[(\pmb{X}-E\pmb{X})(\pmb{X}-E\pmb{X})^\top\pmb{A}^\top\big]\\ &= \pmb{A}E\big[(\pmb{X}-E\pmb{X})(\pmb{X}-E\pmb{X})^\top\big]\pmb{A}^\top\\ &=\pmb{A} \text{Var}(\pmb{X})\pmb{A}^\top \end{aligned} Var(AX)=E[(AX−EAX)(AX−EAX)⊤]=E[(AX−AEX)(AX−AEX)⊤]=AE[(X−EX)(A(X−EX)⊤)]=AE[(X−EX)(X−EX)⊤A⊤]=AE[(X−EX)(X−EX)⊤]A⊤=AVar(X)A⊤ 说明:随机变量的线性组合仍是随机变量,对于一组随机变量,或者 “随机向量” 而言,可以对它们进行线性组合得到一个新的 “随机向量”,并用矩阵形式表示为 A r × n X n × 1 = X r × 1 ′ \pmb{A}_{r\times n}\pmb{X}_{n\times 1} = \pmb{X}'_{r\times 1} Ar×nXn×1=Xr×1′,即上式中 A X \pmb{AX} AX
2. 多维 r.v. 的分布函数
- 对于多维 r.v.s.(随机向量)
X
=
[
X
1
X
2
⋮
X
n
]
\pmb{X} = \begin{bmatrix}X_1\\X_2\\\vdots \\X_n \end{bmatrix}
X=
X1X2⋮Xn
,其分布函数定义为
F ( X ) = F ( X 1 , X 2 , . . . , X n ) = P ( X 1 ≤ x 1 , X 2 ≤ x 2 , . . . , X n ≤ x n ) = P ( X ≤ x ) F(\pmb{X}) = F(X_1,X_2,...,X_n) = P(X_1\leq x_1,X_2 \leq x_2,...,X_n\leq x_n) = P(\pmb{X} \leq \pmb{x}) F(X)=F(X1,X2,...,Xn)=P(X1≤x1,X2≤x2,...,Xn≤xn)=P(X≤x) - 两道例题
- 令 X t = ξ c o s t X_t = \xi cost Xt=ξcost,其中 P ( ξ = 1 ) = P ( ξ = 2 ) = P ( ξ = 3 ) = 1 3 , t ∈ R P(\xi=1)=P(\xi=2)=P(\xi=3)=\frac{1}{3},t\in\mathbb{R} P(ξ=1)=P(ξ=2)=P(ξ=3)=31,t∈R,这样就可以通过选取不同的实数 t t t 得到任意维随机向量 { X t } \{X_t\} {Xt}。求一维分布 F ( x ; π 4 ) , F ( x ; π 2 ) F(x;\frac{\pi}{4}),F(x;\frac{\pi}{2}) F(x;4π),F(x;2π) 和二维分布 F ( x 1 , x 2 ; 0 , π 3 ) F(x_1,x_2;0,\frac{\pi}{3}) F(x1,x2;0,3π)
- 令
X
t
=
ξ
+
η
t
X_t = \xi +\eta t
Xt=ξ+ηt,其中
ξ
,
η
\xi,\eta
ξ,η 为互相独立的标准正态分布,
t
∈
R
t\in\mathbb{R}
t∈R,这样就可以通过选取不同的实数
t
t
t 得到任意维随机向量
{
X
t
}
\{X_t\}
{Xt},求一维分布和二维分布
注:右边第1题答案中第3问,分布列应该是 X t = π 3 = 1 2 , 1 , 3 2 X_{t=\frac{\pi}{3}} = \frac{1}{2},1,\frac{3}{2} Xt=3π=21,1,23,订正一下
3. 多维 r.v. 的函数(统计量)的概率密度
- 以二维情况为例,设二维 r.v.
X
=
(
X
1
,
X
2
)
\pmb{X}=(X_1,X_2)
X=(X1,X2) 的概率密度函数为
f X ( x 1 , x 2 ) = { f X G ( x 1 , x 2 ) , ( x 1 , x 2 ) ∈ G ⊂ R 0 , ( x 1 , x 2 ) ∉ G f_{\mathbf{X}}(x_1,x_2) = \left\{ \begin{aligned} &f_{\mathbf{X}}^G(x_1,x_2) &&,(x_1,x_2)\in G \subset \mathbb{R} \\ &0 &&,(x_1,x_2)\notin G\\ \end{aligned} \right. fX(x1,x2)={fXG(x1,x2)0,(x1,x2)∈G⊂R,(x1,x2)∈/G 当有 { Y 1 = g 1 ( X 1 , X 2 ) Y 2 = g 2 ( X 1 , X 2 ) \left\{\begin{aligned}&Y_1 = g_1(X_1,X_2) \\&Y_2 = g_2(X_1,X_2) \end{aligned}\right. {Y1=g1(X1,X2)Y2=g2(X1,X2) 时,求 Y = ( Y 1 , Y 2 ) \pmb{Y} = (Y_1,Y_2) Y=(Y1,Y2) 的概率密度函数 f Y ( y 1 , y 2 ) f_{\mathbf{Y}}(y_1,y_2) fY(y1,y2) - 遵循以下过程求解
- 依题意,对于
X
1
,
X
2
X_1,X_2
X1,X2 的任意取值
x
1
,
x
2
x_1,x_2
x1,x2,变换
T
T
T 为
T : { y 1 = g 1 ( x 1 , x 2 ) y 2 = g 2 ( x 1 , x 2 ) T:\left\{ \begin{aligned} &y_1 = g_1(x_1,x_2) \\ &y_2 = g_2(x_1,x_2) \end{aligned} \right. T:{y1=g1(x1,x2)y2=g2(x1,x2) 利用变换 T T T,把取值范围变换过来,即利用所有概率非0的 ( x 1 , x 2 ) (x_1,x_2) (x1,x2) 找出所有概率非零的 ( y 1 , y 2 ) (y_1,y_2) (y1,y2)
G ∗ = T ( G ) G^* = T(G) G∗=T(G) - 下面求
T
T
T 的反函数(即利用
y
\pmb{y}
y 表示
x
\pmb{x}
x),先假设有唯一的反函数
{ x 1 = x 1 ( y 1 , y 2 ) x 2 = x 2 ( y 1 , y 2 ) \left\{ \begin{aligned} &x_1 = x_1(y_1,y_2) \\ &x_2 = x_2(y_1,y_2) \end{aligned} \right. {x1=x1(y1,y2)x2=x2(y1,y2) - 计算 Jacobi 行列式 (Jacobi 矩阵的行列式)
∣ J ∣ = ∣ ∂ ( x 1 , x 2 ) ∂ ( y 1 , y 2 ) ∣ = ∣ ∂ x 1 ∂ y 1 ∂ x 2 ∂ y 1 ∂ x 1 ∂ y 2 ∂ x 2 ∂ y 2 ∣ |\pmb{J}| = \begin{vmatrix} \frac{\partial(x_1,x_2)}{\partial(y_1,y_2)}\end{vmatrix} = \begin{vmatrix} \frac{\partial x_1}{\partial y_1} & \frac{\partial x_2}{\partial y_1} \\ \frac{\partial x_1}{\partial y_2} & \frac{\partial x_2}{\partial y_2} \end{vmatrix} ∣J∣= ∂(y1,y2)∂(x1,x2) = ∂y1∂x1∂y2∂x1∂y1∂x2∂y2∂x2 得到 Y = { Y 1 , Y 2 } \pmb{Y} = \{Y_1,Y_2\} Y={Y1,Y2} 的 p.d.f. 为
f Y ( y 1 , y 2 ) = { f X G ( x 1 ( y 1 , y 2 ) , x 2 ( y 1 , y 2 ) ) ∣ J ∣ , ( y 1 , y 2 ) ∈ G ∗ 0 , ( y 1 , y 2 ) ∉ G ∗ f_{\mathbf{Y}}(y_1,y_2) = \left\{ \begin{aligned} &f_{\mathbf{X}}^G\big(x_1(y_1,y_2),x_2(y_1,y_2)\big) |\pmb{J}| &&,(y_1,y_2)\in G^* \\ &0 &&,(y_1,y_2)\notin G^*\\ \end{aligned} \right. fY(y1,y2)={fXG(x1(y1,y2),x2(y1,y2))∣J∣0,(y1,y2)∈G∗,(y1,y2)∈/G∗ - 若第 2 步中的反函数不唯一,则对每个反函数
x
(
i
)
\pmb{x}^{(i)}
x(i) 计算对应的 Jacobi 行列式
∣
J
(
i
)
∣
,
i
=
1
,
2
,
.
.
.
,
n
|\pmb{J}^{(i)}|,i=1,2,...,n
∣J(i)∣,i=1,2,...,n,最后得到
Y
=
{
Y
1
,
Y
2
}
\pmb{Y} = \{Y_1,Y_2\}
Y={Y1,Y2} 的 p.d.f. 为
f Y ( y 1 , y 2 ) = { ∑ i = 1 n f X G ( x 1 ( i ) ( y 1 , y 2 ) , x 2 ( i ) ( y 1 , y 2 ) ) ∣ J ( i ) ∣ , ( y 1 , y 2 ) ∈ G ∗ 0 , ( y 1 , y 2 ) ∉ G ∗ f_{\mathbf{Y}}(y_1,y_2) = \left\{ \begin{aligned} &\sum_{i=1}^nf_{\mathbf{X}}^G\big(x_1^{(i)}(y_1,y_2),x_2^{(i)}(y_1,y_2)\big) |\pmb{J}^{(i)}| &&,(y_1,y_2)\in G^* \\ &0 &&,(y_1,y_2)\notin G^*\\ \end{aligned} \right. fY(y1,y2)=⎩ ⎨ ⎧i=1∑nfXG(x1(i)(y1,y2),x2(i)(y1,y2))∣J(i)∣0,(y1,y2)∈G∗,(y1,y2)∈/G∗
- 依题意,对于
X
1
,
X
2
X_1,X_2
X1,X2 的任意取值
x
1
,
x
2
x_1,x_2
x1,x2,变换
T
T
T 为
- 例题:
-
设 r.v.s { X , Y } \{X,Y\} {X,Y} 有 X , Y ∼ ϵ ( 1 ) X,Y\sim \epsilon(1) X,Y∼ϵ(1) 且相互独立,对于如下线性变换,求 g ( u , v ) g(u,v) g(u,v) 的 p.d.f
{ U = X + Y V = X / Y \left\{ \begin{aligned} &U = X+Y \\ &V = X/Y \end{aligned} \right. {U=X+YV=X/Y 注意 X , Y X,Y X,Y 独立 ⇔ F X , Y ( x , y ) = F X ( x ) F Y ( y ) \Leftrightarrow F_{X,Y}(x,y) = F_X(x)F_Y(y) ⇔FX,Y(x,y)=FX(x)FY(y),答案如下
-
设 r.v.s { X , Y } \{X,Y\} {X,Y} 有 X , Y ∼ N ( 0 , σ 2 ) X,Y\sim N(0,\sigma^2) X,Y∼N(0,σ2) 且相互独立,对于如下线性变换,求 g ( u , v ) g(u,v) g(u,v) 的 p.d.f
{ U = X 2 + Y 2 V = X / Y \left\{ \begin{aligned} &U = \sqrt{X^2+Y^2} \\ &V = X/Y \end{aligned} \right. {U=X2+Y2V=X/Y 注意 X , Y X,Y X,Y 独立 ⇔ F X , Y ( x , y ) = F X ( x ) F Y ( y ) \Leftrightarrow F_{X,Y}(x,y) = F_X(x)F_Y(y) ⇔FX,Y(x,y)=FX(x)FY(y),答案如下
-