跳转至

预备知识

可测空间、概率空间与概率测度

\(\renewcommand{\geq}{\geqslant}\renewcommand{\leq}{\leqslant}\) 当一个试验的结果无法预先确定时,称该实验为随机试验。随机试验可能出现的所有结果集合构成样本空间\(\Omega\),每个可能的结果称为样本点\(\omega\)\(\Omega\)的子集构成的集合称为集类\(\mathcal F\)

可测空间

\(\mathcal F\)为由\(\Omega\)的某些子集构成的非空集类,若满足以下条件:

  1. \(A\in \mathcal F\),则\(A^C = \Omega - A\in \mathcal F\)
  2. \(A_n\in \mathcal F\),则\(\bigcup\limits_{n=1}^\infty A_n \in \mathcal F\)

\(\mathcal F\)\(\sigma\)\((\Omega, \mathcal F)\)可测空间

\(\sigma\)\(\mathcal F\)\(\cap, \cup, -\)运算封闭,任何元素经过可列次运算后仍属于\(\mathcal F\)

对于集类\(\mathcal A\),包含\(\mathcal A\)\(\sigma\)域的交称为\(\mathcal A\)生成的\(\sigma\)域,记作\(\sigma(\mathcal A)\)。如:\(\sigma(\{\varnothing, A, \Omega\}) = \{\varnothing, A, A^C, \Omega\}\)。特殊地,记\(\mathcal B = \sigma(\{(-\infty, \alpha], \forall \alpha \in \mathbb R\})\)为Borel域。Borel域解决了样本空间在\(\mathbb R\)上连续的问题。可以证明,\(\forall a < b, [a, b]\in \mathcal B, (a, b]\in \mathcal B, [a, b)\in \mathcal B, (a, b)\in \mathcal B\)。定义\(\mathcal B[a, b]\)为限制在\([a, b]\)上的Borel域。

概率测度与概率空间

\((\Omega, \mathcal F)\)为可测空间,\(P: \mathcal F\rightarrow [0, 1]\)为定义在\(\mathcal F\)上的集函数。且\(P\)满足

  1. 非负性:\(\forall A\in \mathcal F, P(A)\geq 0\)
  2. 规一性:\(P(\Omega) = 1\)
  3. 可列可加性:若\(\forall i\in \mathbb N, A_i\in \mathcal F\),且\(\forall i\not = j, A_i\cap A_j=\varnothing\),则

    \[ P\left(\bigcup_{n=0}^\infty A_n\right) = \sum_{n=0}^\infty P(A_n) \]

\(P\)为可测空间\((\Omega, \mathcal F)\)上的概率测度\((\Omega, \mathcal F, P)\)概率空间\(\mathcal F\)事件域\(A\in \mathcal F\)(随机)事件

概率测度\(P\)满足如下性质

  1. 有限可加性:可以令\(A_{n+1} = A_{n+2} = \cdots = \varnothing\),结合可列可加性推出
  2. \(P(A^C) = 1 - P(A)\)
    • \(P(\varnothing) = 0\)
  3. 集合的包含关系:\(A\subset B\Rightarrow P(A)\leq P(B)\)
  4. 容斥原理:

    \[ P\left(\bigcup_{i=1}^n A_i\right) = \sum_{i=1}^n P(A_i) - \sum_{1\leq i < j\leq n}P(A_i\cap A_j) + \cdots + (-1)^{n+1}P(A_1\cap \cdots \cap A_n) \]
    • \(P(A\cup B) = P(A) + P(B) - P(A\cap B)\)
    • \(P(A - B) = P(A\cup B) - P(B) = P(A) - P(A\cap B)\)
    • \(P\left(\bigcup_{i=1}^n A_i\right)\leq \sum_{i=1}^n P(A_i)\)

满足\(A_n\subset A_{n+1}\)的事件列\(\{A_n, n\geq 1\}\)称为单调增序列,满足\(A_n\supset A_{n+1}\)的事件列\(\{A_n, n\geq 1\}\)称为单调减序列。由此可以定义事件列的极限:

  1. \(\{A_n, n\geq 1\}\)为单调增序列,则\(\lim\limits_{n\rightarrow \infty}A_n = \bigcup_{i=1}^\infty A_i\)
  2. \(\{A_n, n\geq 1\}\)为单调减序列,则\(\lim\limits_{n\rightarrow \infty}A_n = \bigcap_{i=1}^\infty A_i\)

事件列的极限满足

\[ \lim_{n\rightarrow\infty}P(A_n) = P\left(\lim_{n\rightarrow\infty} A_n\right) \]

可以证明,单点集\(\{a\}\)为事件列\(\{[a, a+1/n], n > 0\}\)的极限。

Borel-Cantelli引理

\(\{A_n, n\geq 1\}\)为事件序列,满足\(\sum_{i=1}^\infty P(A_i) < \infty\),则

\[ P\left(\lim_{i\rightarrow\infty}\sup A_i\right) \triangleq P\left(\bigcap_{n=1}^\infty \bigcup_{i=n}^\infty A_i\right) = 0 \]

定义条件概率

\[ P(A|B) = \frac{P(A\cap B)}{P(B)} \]

若事件\(A, B\)满足\(P(A\cap B) = P(A)P(B)\),则称事件\(A, B\)相互独立,且\(P(A|B) = P(A)\)。同理可以推广至\(n\)个事件相互独立的情况。设\(A_1, \cdots, A_n\in \mathcal F\),若对于其中任意\(k\)个事件,都有

\[ P(A_{i_1}\cap\cdots\cap A_{i_k}) = P(A_{i_1})\cdots P(A_{i_k}) \]

\(A_1, \cdots, A_n\)相互独立。

\(\{A_n, n\geq 1\}\)为相互独立的事件序列,且\(\sum_{n=1}^\infty P(A_n) = \infty\),则

\[ P\left(\lim_{i\rightarrow\infty}\sup A_i\right) \triangleq P\left(\bigcap_{n=1}^\infty \bigcup_{i=n}^\infty A_i\right) = 1 \]

随机变量与分布函数

设样本空间为\(\Omega\)

随机变量

\((\Omega, \mathcal F, P)\)为概率空间,\(X(\omega)\)为定义在\(\Omega\)上的单值实函数,即\(X: \Omega\rightarrow \mathbb R\),若\(\forall a\in \mathbb R\),有\(\{\omega: X(\omega) \leq a\}\in \mathcal F\),称\(X(\omega)\)为随机变量,简记为\(X\)。定义\(F(x) = P(X\leq x) = P(X\in (-\infty, X])\)\(X\)的分布函数。

  • 离散型随机变量:随机变量\(X\)的可能取值的全体是可列集或者有限集。
  • 连续型随机变量:若\(\forall B\in \mathcal F\),存在一个函数\(f(x)\)满足

    \[ P(X\in B) = \int_B f(x)\mathrm dx \]

    \(X\)为连续型随机变量,\(f(x)\)\(X\)的概率密度函数。

概率密度函数与概率分布函数

根据概率分布函数的定义,有

\[ P(x < X\leq x + h) = F(x + h) - F(x) = \int_{x}^{x + h} f(x)\mathrm dx = f(x)h + o(h) \]

\(h\rightarrow 0\)并取极限,得到

\[ \frac{\mathrm dF(x)}{\mathrm dx} = \lim_{h\rightarrow 0} \frac{F(x + h) - F(x)}{h} = f(x) \]

对于二维随机变量\(X, Y\),定义联合分布函数为\(F(x, y) = P(X\leq x, Y\leq y)\),边缘分布为\(F_X(x) = P(X\leq x), F_Y = P(Y\leq y)\)。同理可以定义概率密度函数\(f(x, y)\)

\[ F(x, y) = \int_{-\infty}^x\int_{-\infty}^y f(v, u)\mathrm du\mathrm dv \]

\(F(x, y) = F_X(x)F_Y(y)\),称\(X\)\(Y\)相互独立。

数字特征

\(X\)为随机变量,分布函数为\(F(x)\),若\(\int_{-\infty}^\infty |x|\mathrm dF(x)\)存在,则定义\(X\)的期望\(E(X)\)

\[ E(X) = \int_{-\infty}^\infty x\mathrm dF(x) \]

数学期望\(E(X)\)满足

  • 可加性

    \[ E\left(\sum_{i=1}^n c_iX_i\right) = \sum_{i=1}^n c_iE(X_i) \]
  • 函数

    \[ E(g(x)) = \int_{-\infty}^\infty g(x)\mathrm dF(x) \]

对于离散型随机变量,有

\[ E(X) = \sum_{i=1}^\infty x_nP(X = x_n) \]

对于连续型随机变量,有

\[ E(X) = \int_{-\infty}^\infty xf(x)\mathrm dx \]

若随机变量\(X, Y\)相互独立,则有

\[ \begin{aligned} E(XY) &= \int_{-\infty}^\infty \int_{-\infty}^\infty xyf(x, y)\mathrm dx\mathrm dy \\ &= \int_{-\infty}^\infty \int_{-\infty}^\infty xyf_X(x)f_Y(y)\mathrm dx\mathrm dy \\ &= \int_{-\infty}^\infty xf_X(x)\mathrm dx\int_{-\infty}^\infty yf_Y(y)\mathrm dy \\ &= E(X)E(Y) \end{aligned} \]

定义随机变量\(X\)的方差为\(\sigma_X^2 = D(X) = E(X - E(X)) = E(X^2) - E^2(X)\),定义随机变量\(X, Y\)的协方差为\(\mathrm{cov}(X, Y) = E(XY) - E(X)E(Y)\),相关系数为\(\rho = \mathrm{cov}(X, Y) / (\sigma_X\sigma_Y)\)\(k\)阶矩为\(E(X^k)\)

定义随机变量\(X\)的矩母函数为\(\psi(t) = E\left(e^{tX}\right)\),特征函数为\(\phi(t) = E\left(e^{\mathbf itX}\right)\)

矩母函数满足如下性质

  1. \(E\left(X^k\right) = \psi^{(k)}(0)\)
  2. 设随机变量\(X, Y\)的分布函数为\(F_X(t), F_Y(t)\),矩母函数为\(\psi_X(t), \psi_Y(t)\),则\(\psi_X(t) = \psi_Y(t) \Leftrightarrow F_X(t) = F_Y(t)\)

特征函数满足如下性质

  1. 设随机变量\(X, Y\)的分布函数为\(F_X(t), F_Y(t)\),特征函数为\(\phi_X(t), \phi_Y(t)\),则\(\phi_X(t) = \phi_Y(t) \Leftrightarrow F_X(t) = F_Y(t)\)
  2. \(i^kE(X^k) = \phi^{(k)}(0)\)
  3. \(X, Y\)相互独立,有\(\phi_{X+Y}(t) = \phi_{X}(t) + \phi_{Y}(t)\)

常见随机变量的分布

  1. 二项分布:\(X\sim B(n, p)\)

    \[ P(X = k) = \binom{n}{k} p^k(1 - p)^{n - k} \]
  2. 泊松分布:\(X\sim P(\lambda)\)

    \[ P(X = k) = \frac{\lambda^k}{k!}e^{-\lambda} \]
  3. 几何分布:\(X\sim G(p)\)

    \[ P(X = k) = (1 - p)^{k - 1}p \]
  4. 均匀分布:\(X\sim U(a, b)\)

    \[ f(x) = \left\{\begin{aligned} & \frac{1}{b-a} & a < x < b \\ & 0 & \text{otherwise} \end{aligned}\right. \]
  5. 正态分布:\(X\sim N(\mu, \sigma^2)\)

    \[ f(x) = \frac{1}{\sqrt{2\pi}\sigma} e^{-\frac{(x-\mu)^2}{2\sigma^2}} \]
  6. 指数分布:\(X\sim E(\lambda)\)

    \[ f(x) = \left\{\begin{aligned} & \lambda e^{-\lambda x} & x\geq 0 \\ & 0 & x < 0 \end{aligned}\right. \]

条件数学期望

对于连续随机变量,条件数学期望\(E(X|Y = y)\)的定义如下:

\[ E(X|Y = y) = \int_{-\infty}^\infty x\frac{f(x, y)}{f_Y(y)}\mathrm dx \]

定义\(E(X|Y)\)\(X\)关于\(Y\)的条件数学期望

\[ E(X|Y) = \sum_j \boldsymbol 1_{\{Y = y_j\}}(\omega)E(X|Y=y_j) \]

对于离散随机变量,条件数学期望\(E(X|Y = y)\)的定义如下:

\[ E(X|Y = y) = \sum_i x_iP(X=x_i | Y=y) \]

显然,\(E(X|Y = y)\)\(y\)的函数,定义随机变量\(E(X|Y)\)\(X\)关于\(Y\)的条件期望,若\(E(X|Y)\)满足

  1. \(E(X|Y)\)\(Y\)的函数,当\(Y = y\)时,\(E(X|Y) = E(X|Y = y)\)
  2. \(\forall D\in \mathcal B\),有

    \[ E[E(X|Y)|Y\in D] = E(X|Y\in D) \]

条件数学期望的性质

条件数学期望满足如下性质:

  1. \(E(E(X|Y)) = EX\)
  2. \(X, Y\)相互独立,则\(E(X|Y) = EX\)
  3. \(E[g(X)h(Y)|Y] = h(Y)E(g(X)|Y)\quad \text{a.s.}\)

随机过程

随机过程是一族无穷多个,相互有关的随机变量。设\(X(t, \omega)\)为随机变量,其中\(t\in T\subset \mathbb R\)为参数。称\(X_T = \{X(t, \omega), t\in T\}\)为随机过程。

  1. \(X(t, \omega)\)可以简记为\(X(t)\)
  2. \(T\)为可列集时,称\(X_T\)为随机序列;
  3. \(X_T\)的取值范围称为状态空间\(S\)
  4. \(X(\cdot, \omega), \omega\in \Omega\)是关于\(t\)的函数,称为轨道。

评论