4 随机向量，随机变量的独立性_随机变量向量-CSDN博客

本文链接：https://blog.csdn.net/baidu_18548009/article/details/117732199

文章目录

1 随机向量及分布函数
- 1.1 随机向量
- 1.2 (联合) 分布函数 (Jonint C.D.F)
2 常见的多元随机变量
- 2.1 离散型
- 2.2 连续型
3 边际分布及条件分布 (Conditional Distribution)
- 3.1 边际分布
- 3.2 条件分布
3 随机变量的独立性

1 随机向量及分布函数

1.1 随机向量

若随机变量 $\xi_1 ( \omega ), \xi_2 ( \omega ), ..., \xi_n( \omega )$ 定义在同一概率空间 $\Omega, \mathcal{F}, \mathcal{P} )$ 上, 则称

$\vec{\xi} (\omega) = \big( \xi_1 ( \omega ), \xi_2 ( \omega ), ..., \xi_n( \omega ) \big)$

构成一个 $n$ 维随机向 (变) 量.

1.2 (联合) 分布函数 (Jonint C.D.F)

$\vec{\xi} (\omega) \sim F (x_1, x_2, ..., x_n) = P \{ \xi_1 ( \omega ) \leq x_1, ..., \xi_n ( \omega ) \leq x_n \}.$

其中:
$\mathbb{R}^n \mapsto [0, 1]$ ;
$\{ \omega: \xi_i (\omega) \leq x_i, ~ i = 1, 2, ..., n \} \in \mathcal{F}$ ;
$\cap_{i = 1}^n \{ \xi_n^{-1} (x_i) \} \in Borel ~ \mathbb{R}^n$ .

性质:

$F(\vec{x})$ 右连续;
$\displaystyle \lim_{x_i \to - \infty} F(x_1, ..., x_n) = 0$ ;
( $\because ~ \forall ~ i \in \{ 1, 2, ..., n \}, ~ \cap_{j = 1}^n \{ \xi_j \leq x_j \} \subset \{ \xi_i \leq x_i \}$ , 且 $\displaystyle \lim_{x_i \to - \infty} F(x_i) = 0$ )
$\displaystyle \lim_{x_i \to + \infty} F(x_1, ..., x_n) = F_{n-1} (x_1, ..., x_{i-1}, x_{i+1}, ...x_n)$ .
即 $(\xi_1, ..., \xi_{i-1}, \xi_{i+1}, ..., \xi_n)$ 的 C.D.F.
特别地 $\xi_i \sim F_i (x_i) = \displaystyle \lim_{x_j \to + \infty} F(x_1, ..., x_i, ..., x_n), ~ j \neq i$ , ;称为它的边际(Marginal)分布.
单调性: 关于每个分量单调不减.
一元: $\forall ~ x_1 < x_2, ~ P(x_1 < \xi \leq x_2) = F(x_2) - F(x_1) > 0$ ;
二元: $\forall ~ x_1' < x_1, ~ x_2' < x_2$ ,
$\begin{aligned} &P(x_1' < \xi_1 \leq x_1, ~ x_2' < \xi_2 \leq x_2) = \\ &F(x_1, x_2) - F(x_1', x_2) - F(x_1, x_2') + F(x_1', x_2'); \end{aligned}$
$n$ 元: $\forall ~ x_1' < x_1, ..., ~ x_n' < x_n$ ,
$\begin{aligned} &P(x_1' < \xi_1 \leq x_1, ..., ~ x_n' < \xi_n \leq x_n) = \\ &F(x_1, ..., x_n) - \sum F(含有1个'的) + \sum F(含有2个'的) \\ &+ ... +(-1)^n \sum F(含有n个'的); \end{aligned}$

满足以上四条性质的函数必是某随机变量的分布函数.

2 常见的多元随机变量

2.1 离散型

多项分布: (Multinomial Distribution)
$n$ 次试验中 $A_i$ 发生 $n_i$ 次, $i = 1, . . ., r$ , 即一次试验结果有 $r$ 种情况.
$P(A_i) = p_i$ , $\sum_{i = 1}^r n_i$ , $\sum_{i = 1}^r p_i = 1$ ,
$\begin{aligned} P \{ \xi_1 = n_1, ..., ~\xi_r = n_r \} = \frac{n!}{n_1!...n_r!} p_1^{n_1}...p_r^{n_r} \end{aligned}$ .
超几何分布: (Hypergeometric Distribution)
袋中共有 $N$ 只球, 第 $i$ 种球有 $N_i$ 只, 共取出 $n$ 只球, 第 $i$ 种球取出了 $n_i$ 只.
$\sum_{i = 1}^r N_i$ , $\sum_{i = 1}^r n_i$ , $i = 1, . . ., r$ , 即共有 $i$ 种球.
$\begin{aligned} P \{ \xi_1 = n_1, ..., ~\xi_r = n_r \} = \frac{N \choose n}{ {N_1 \choose n_1} {N_2 \choose n_2}... {N_r \choose n_r} } \end{aligned}$ .

2.2 连续型

连续型随机变量密度函数 $p(x_1, ..., x_n)$ 的定义: 存在非负函数 $p(x_1, ..., x_n)$ 使得

$F(x_1, ..., x_n) = \int_{- \infty}^{x_1} ... \int_{- \infty}^{x_n} p(y_1, ..., y_n) ~ dy_1...dy_n.$

均匀分布: (Uniform Distribution)
$\vec{x} \sim N(\vec{\mu}, \Sigma)$
有限可测区域 $\in \mathbb{R}^n$ , 其测度为 $S$ .
$p(x_1, x_2, ..., x_n) = \begin{cases} \frac{1}{S}, ~~~ (x_1, x_2, ..., x_n) \in G; \\ 0, ~~~~ (x_1, x_2, ..., x_n) \notin G. \end{cases}$
多元正太分布: (Multivariate Normal Distribution)
$\begin{aligned} p(x_1, x_2, ..., x_n) = p(\vec{x}) = \frac{1}{(2 \pi)^{\frac{n}{2}} |\Sigma|^{\frac{1}{2} } } exp \{ - \frac{1}{2} (\vec{x} - \vec{\mu}) \Sigma^{-1} (\vec{x} - \vec{\mu}) \} \end{aligned}$
其中 $\vec{\mu} = (\mu_1, \mu_2, ..., \mu_n)$ , $\Sigma = (\sigma_{ij})$ 为 $n$ 阶正定对称矩阵.

$n = 1$ 时, $\mu \in \mathbb{R}$ , $\Sigma = \sigma^2 > 0$ ;
$n = 2$ 时, $\sim N(\mu_1, \mu_2, \sigma_1^2, \sigma_2^2, \rho)$ ,
$\vec{\mu} = (\mu_1, \mu_2)$ , $\Sigma = \begin{pmatrix} \sigma_1^2 & \rho \sigma_1 \sigma_2 \\ \rho \sigma_1 \sigma_2 & \sigma_2^2 \end{pmatrix}$ ,
$\begin{aligned} p(x, y) = \frac{1}{2 \pi \sigma_1 \sigma_2 \sqrt{1 - \rho^2}} exp \{ -\frac{1}{2(1 - \rho)^2} \big[ \frac{(x - \mu_1)^2}{\sigma_1^2} - 2 \rho \frac{(x - \mu_1)(y - \mu_2)}{\sigma_1 \sigma_2} + \frac{(y - \mu_2)^2}{\sigma_2^2} \big] \} \end{aligned}$ ,
其典型分解为
$\begin{aligned} p(x, y) = \frac{1}{\sqrt{2 \pi} \sigma_1} exp \{- \frac{(x - \mu_1)^2}{2 \sigma_1^2}\} \cdot \frac{1}{\sqrt{2 \pi} \sigma_2 \sqrt{1 - \rho^2}} exp \{ - \frac{ \big[ y - \big( \mu_2 + \rho \frac{\sigma_1}{\sigma_2} (x - \mu_1) \big) \big]^2 }{2 \sigma_2^2 (1 - \rho^2)} \} \end{aligned}$ .

3 边际分布及条件分布 (Conditional Distribution)

3.1 边际分布

设多元随机变量 $\vec{x} \sim F (\vec{x})$ , 则 $x_j$ 的边际分布为 $F_j (x_j) = F (+ \infty, + \infty, ..., x_j, ...)$ .

离散型: $n = 2$ 时, $p_{ij} = P(x = x_i, y = y_j)$ , $i, j = 1, 2, . . .$ .
$x$ 的边际分布为
$\begin{aligned} P(x = x_i) &= P(x = x_i, \cup_{j = 1}^{\infty} \{ y = y_j \}) \\ &= \sum_{j = 1}^{\infty} P(x = x_i, y = y_j) \\ &= \sum_{j = 1}^{\infty} p_{ij} = p_{i.}.\end{aligned}$
连续型:
$\begin{aligned} F_t(x_t) &= F(+ \infty, + \infty, ..., x_t, ...) \\ &= \int_{- \infty}^{x_t} \bigg[ \int_{- \infty}^{+ \infty} ... \int_{- \infty}^{+ \infty} f(\vec{y}) ~ dy_1 dy_2 ... dy_{t-1} dy_{t+1} ... \bigg]~ dy_t \end{aligned}$
混合型:
eg. $\sim N(0, 1)$ , $\begin{cases} -1, ~~~ x < -1; \\ x, ~~~ -1 \leq x < 1; \\ 1, ~~~ x \geq 1. \end{cases}$

3.2 条件分布

以二元条件分布为例 $y_0) = P(X \leq x ~ | ~ Y = y_0)$ .

离散型: $\begin{aligned} P(x = x_{i} ~ | ~ y = y_{j}) = \frac{P(x = x_i, y = y_i)}{P(y = y_i)} = \frac{p_{ij}}{p_{.j}} \end{aligned}$ ;
连续型:
$\begin{aligned} &P(x \leq X < x + \triangle x ~ | ~ y \leq Y < y + \triangle y) \\ = & \frac{P(x \leq X < x + \triangle x, ~ y \leq Y < y + \triangle y)}{P( y \leq Y < y + \triangle y)} \\ = & \frac{\int_x^{x + \triangle x} \int_{y}^{y + \triangle y} f(s, t) ~ dt ds}{\int_{- \infty}^{+ \infty} \int_{y}^{y + \triangle y} f(s, t) ~ dt ds}; \end{aligned}$
$\begin{aligned} P(x ~ | ~ y) &= \lim_{{ \begin{gathered} \triangle x \to 0 \\ \triangle y \to 0 \end{gathered}}} \frac{1}{\triangle x} P(x \leq X < x + \triangle x ~ | ~ y \leq Y < y + \triangle y) \\ &=\lim_{{ \begin{gathered} \triangle x \to 0 \\ \triangle y \to 0 \end{gathered}}} \frac{\frac{1}{\triangle x \triangle y} \int_x^{x + \triangle x} \int_{y}^{y + \triangle y} f(s, t) ~ dt ds}{\int_{- \infty}^{+ \infty} \big[ \frac{1}{\triangle y} \int_{y}^{y + \triangle y} f(s, t) ~ dt \big] ds} \\ &= \frac{f(x, y)}{\int_{- \infty}^{+ \infty} f(s, y) ~ ds} = \frac{f(x, y)}{f_y (y)}. \end{aligned}$
若 $f_{y}(y) = 0$ , 则 $p (x ∣ y) = 0$ .

例: 二元正态分布 $\sim N(\mu_1, \mu_2, \sigma_1^2, \sigma_2^2, \rho)$
边际分布: $\sim N(\mu_1, \sigma_1^2)$ ; $\sim N(\mu_2, \sigma_2^2)$ ;
条件分布: $x_0 \sim N \big( \mu_2 + \rho \frac{\sigma_2}{\sigma_1} (x_0 - \mu_1), \sigma_2^2 (1 - \rho^2) \big)$ .
当 $\rho = 0$ (即 $x$ , $y$ 独立) 时, 即为 $Y$ 的边际分布.

3 随机变量的独立性

对于 $\mathbb{R}$ 上的任意 Borel 集 $A$ , $B$ …
$\in A, Y \in B, ...) = P(X \in A) P(Y \in B) ...,$
则称 $A$ , $B$ … 独立.
特别地, $\infty , x]$ , $\infty , y]$ … 时,
$\leq x, Y \leq y, ...) = P(X \leq x) P(Y \leq y) ....$
以两个变量为例, 即 $F(x, y) = F_x (x) F_y (y)$ 对 $\forall ~ x, y \in \mathbb{R}$ 成立.
即定义域中任意点满足: 联合分布 = $\prod$ 边际分布.

离散型
$x_1, x_2, ... x_n)^T$ 对 $A_1, ..., A_n$ ,
$P(x_1 \in A_1, ... , x_n \in A_n) = P(x_1 \in A_1)... P(x_n \in A_n)$ .
连续型
$\begin{aligned} f(x_1, ..., x_n) = \frac{\partial^n F(x_1, ..., x_n)}{\partial x_1 ... \partial x_n} = f_{x_1} (x_1)... f_{x_n} (x_n) \end{aligned}.$
对于几乎处处 $\vec{x} \in \mathbb{R}^n$ 成立.
eg. $\begin{cases} f(s_0) + t, ~~~ s = s_0; \\ f(s), ~~~~~~~~~~~ s \neq s_0. \end{cases}$