分布模型的均值与方差

随机变量分布模型的数字特征

离散型

分布律
$$P(X=x_i)=p_i$$
均值
$$E(X)=\sum _i{x}_i{p}_i$$
方差
$$D(X)=\sum _i(x_i-E(X){)}^2=E(X^2)-E^2(X)$$

两点分布$X\sim B(1,p)$

分布律$P(X=k)=p^k(1-p{)}^{1-k},k=0,1$

均值$E(X)=p$

$$E(X)=0\times (1-p)+1\times p=p$$

均值$D(X)=p(1-p)$

方均值
$$E(X^2)=0^2(1-p)+1^2\times p=p$$
方差
$$D(X)=E(X^2)-E^2(X)=p-p^2=p(1-p)$$

二项分布$X\sim B(n,p)$

分布律$P(X=k)=C_n^k{p}^k{q}^{n-k},q+p=1,k=0,1,2,...$

均值$E(X)=np$

证明:
1.n次独立重复实验,每次成功概率为p,视为每次成功了p个实验,那么n次就成功了np个实验
2.更严谨的证明
$$\begin{array}{rl}E(X)& =\sum _{k=0}^{n}k{p}_k\\ & =\sum _{k=0}^{n}k{C}_n^k{p}^k{q}^{n-k}\\ & =\sum _{k=0}^{n}k\frac{n!}{k!(n-k)!}{p}^k{q}^{n-k}\\ & =np\sum _{k=1}^n\frac{(n-1)!}{(k-1)!((n-1)-(k-1))!}{p}^{k-1}{q}^{(n-1)-(k-1)}\\ & =np\sum _{k=1}^n{C}_{n-1}^k{p}^{k-1}{q}^{(n-1)-(k-1)}\\ & =np(p+q{)}^{n-1}//二项展开式逆用\\ & =np{1}^{n-1}\\ & =np\end{array}$$

方差$D(X)=npq$

证明:
方均值:
$E(X^2)=E(X(X-1)+X)=E(X(X-1))+E(X)$
问题转化为求$E(X(X-1))$
$$\begin{array}{rl}E(X(X-1))& =\sum _{k=1}^{n}k(k-1)p_k\\ & =\sum _{k=1}^{n}k(k-1)C_n^k{p}^k{q}^{n-k}\\ & =\sum _{k=1}^{n}k(k-1)\frac{n!}{k!(n-k)!}{p}^k{q}^{n-k}\\ & =n(n-1)p^2\sum _{k=2}^n\frac{(n-2)!}{(k-2)!((n-2)-(k-2))!}{p}^{k-2}{q}^{(n-2)-(k-2)}\\ & =n(n-1)p^2\sum _{k=2}^n{C}_{n-2}^k{p}^{k-2}{q}^{(n-2)-(k-2)}\\ & =n(n-1)p^2(p+q{)}^{n-2}\\ & =n(n-1)p^2\end{array}$$
那么方均值为$E(X^2)=E(X(X-1))+E(X)=n(n-1)p^2+np=n^2-n{p}^2+np$
那么方差$D(X)=E(X^2)-E^2(X)=(n(n-1)-n^2)p^2+np=np(1-p)=npq$

泊松分布$X\sim P(\lambda )$

分布律$P(X=k)=\frac{{\lambda }^k}{k!}{e}^{-\lambda },\lambda >0,k=0,1,2,...$

均值$E(X)=\lambda$

$$\begin{array}{rl}E(X)& =e^{-\lambda }\sum _{k=0}k\frac{{\lambda }^k}{k!}\\ & =\lambda e^{-\lambda }\sum _{k=1}\frac{{\lambda }^{k-1}}{(k-1)!}\\ & =\lambda e^{-\lambda }\sum _{k=0}\frac{{\lambda }^k}{k!}\end{array}$$
$y=e^x$的麦克劳林展开式为$$e^x=1+x+\frac{1}{2!}{x}^2+\frac{1}{3!}{x}^3+...=\sum _{i=0}^n\frac{x^i}{i!}$$
故$$\sum _{k=0}\frac{{\lambda }^k}{k!}=e^{\lambda }$$
那么均值为$E(X)=\lambda e^{-\lambda }{\sum }_{k=0}\frac{{\lambda }^k}{k!}=\lambda$

方差$D(X)=\lambda$

观察到分布律分母上为$k!$,直接求方均值不容易求,但是利用二项分布求错位均值的思想可以维持分母的阶乘形式
$$\begin{array}{rl}E(X(X-1))& =e^{-\lambda }\sum _{k=1}k(k-1)\frac{{\lambda }^k}{k!}\\ & ={\lambda }^2{e}^{-\lambda }\sum _{k=2}\frac{{\lambda }^{k-2}}{(k-2)!}\\ & ={\lambda }^2{e}^{-\lambda }\sum _{k=0}\frac{e^k}{k!}\\ & ={\lambda }^2{e}^{-\lambda }{e}^{\lambda }\\ & ={\lambda }^2\end{array}$$
则方均值为$E(X^2)=E(X(X-1))+E(X)={\lambda }^2+\lambda$
方差为$D(X)=E(X^2)-E^2(X)=\lambda$

几何分布$X\sim GE(p)$

分布律$P(X=k)=q^{k-1}p(k=1,2,...,p+q=1)$

均值$E(X)=\frac{1}{p}$

$$E(X)=p\sum _{k=1}k{q}^{k-1}$$
令生成函数$f_n(x)$为:
$$f_n(x)=\sum _{n}n{x}^{n-1}$$其中$0<x<1$
则$$\int f_n(x)dx=\sum _n{x}^n=\frac{1(1-x^n)}{1-x}=\frac{1}{1-x}(当0<x<1,\underset{n\to \infty }{\lim }{x}^n=0)$$
那么$$f_n(x)=(\int f_n(x)dx{)}^{\prime }=(\frac{1}{1-x}{)}^{\prime }=\frac{1}{(1-x{)}^2}$$
令$x=q$得到:
$$f_n(q)=\sum _{n}n{q}^{n-1}=\frac{1}{(1-q{)}^2}=\frac{1}{p^2}$$
故$$E(X)=p\sum _{k=1}k{q}^{k-1}=p\times \frac{1}{p^2}=\frac{1}{p}$$

方差$D(X)=\frac{1-p}{p^2}$

$$E((X+1)X)=p\sum _{k=0}(k+1)k{q}^{k-1}$$

令生成函数$$f_n(x)=\sum _n(n+1)n{x}^{n-1}$$

则$$\int f_n(x)dx=\int (\sum _n(n+1)n{x}^{n-1})dx=\sum _n\int (n+1)n{x}^{n-1}dx=\sum _n(n+1)x^n$$

$$\begin{array}{rl}\int (\int f_n(x)dx)dx& =\sum _n{x}^{n+1}\\ & =\frac{x(1-x^n)}{1-x}\\ & =\frac{x}{1-x}\end{array}$$

$$\int f_n(x)dx=(\frac{x}{1-x}{)}^{\prime }=\frac{1}{(1-x{)}^2}$$

$$f_n(x)=(\frac{1}{(1-x{)}^2}{)}^{\prime }=-\frac{2}{(x-1{)}^3}=\frac{2}{(1-x{)}^3}$$

令$x=q$得到:
$$f_n(q)=\sum _k(k+1)k{q}^{k-1}=\frac{2}{(1-q{)}^3}=\frac{2}{p^3}$$

$$E((X+1)X)=p\sum _{k=0}(k+1)k{q}^{k-1}=\frac{2p}{p^3}=\frac{2}{p^2}$$

$$E(X^2)=E((X+1)X)-E(X)=\frac{2}{p^2}-\frac{1}{p}$$

$$D(X)=E(X^2)-E^2(X)=(\frac{2}{p^2}-\frac{1}{p})-\frac{1}{p^2}=\frac{1-p}{p^2}$$

超几何分布$X\sim H(n,M,N)$

分布律$P(X=k)=\frac{C_M^k{C}_{N-M}^{n-k}}{C_N^n},k=0,1,...,min(M,n)$

意义:N个球中有M个黑球,N-M个红球,从中随意挑出n个不放回,挑出k个黑球的概率

均值$E(X)=\frac{nM}{N}$

首先证明一个引理:
$$C_n^k=\frac{n}{k}{C}_{n-1}^{k-1}$$
证明:
$$左侧=C_n^k=\frac{n!}{k!(n-k)!}=\frac{n}{k}\frac{(n-1)!}{(k-1)!((n-1)-(k-1))!}=\frac{n}{k}{C}_{n-1}^{k-1}=右侧$$
该引理在证明超几何分布均值中的应用
令$m=min(M,n)$即每次模球时保证有黑球
$$\begin{array}{rl}E(X)& =\sum _{k=0}^{m}k{p}_k\\ & =\sum _{k=0}^{m}k\frac{C_M^k{C}_{N-M}^{n-k}}{C_N^n}\\ & =\sum _{k=1}^{m}k\frac{\frac{M}{k}{C}_{M-1}^{k-1}{C}_{(N-1)-(M-1)}^{(n-1)-(k-1)}}{\frac{N}{n}{C}_{N-1}^{n-1}}\\ & =\frac{nM}{N}\sum _{k=1}^m\frac{C_{M-1}^{k-1}{C}_{(N-1)-(M-1)}^{(n-1)-(k-1)}}{C_{N-1}^{n-1}}\\ & =\frac{nM}{N}\sum _{k=0}^{m-1}{p}_k(从M-1个黑球,N-M个红球中挑出n个不放回,黑球数为1,2,...,M-1的全概率显然为1)\\ & =\frac{nM}{N}\end{array}$$

连续型

设概率密度函数为$f(x)$
$$E(X)={\int }_{-\infty }^{+\infty }xf(x)dx$$

$$D(X)={\int }_{-\infty }^{+\infty }(x-E(X){)}^{2}f(x)dx$$

均匀分布$X\sim U(a,b)$

概率密度:$f(x)=\{\begin{array}{l}\frac{1}{b-a},a<x\le b\\ 0,x\le a或x>b\end{array}$

均值$E(X)=\frac{a+b}{2}$

$$E(X)={\int }_a^b\frac{x}{b-a}dx=\frac{a+b}{2}$$

方差$D(X)=\frac{(b-a{)}^2}{12}$

$$D(X)={\int }_a^b(x-\frac{a+b}{2}{)}^2\frac{1}{b-a}dx=\frac{(b-a{)}^2}{12}$$

正态分布$X\sim N(\mu ,{\sigma }^2)$

概率密度$f(x)=\frac{1}{\sqrt{2\pi }\sigma }{e}^{-\frac{(x-\mu {)}^2}{2{\sigma }^2}},-\infty <x<+\infty$

均值$E(X)=\mu$

$$\begin{array}{rl}E(X)& =\frac{1}{\sqrt{2\pi }\sigma }{\int }_{-\infty }^{+\infty }x{e}^{-\frac{(x-\mu {)}^2}{2{\sigma }^2}}dx\\ & =\frac{1}{\sqrt{2\pi }\sigma }(-{\sigma }^2{\int }_{-\infty }^{+\infty }d(e^{-\frac{(x-\mu {)}^2}{2{\sigma }^2}})+\mu {\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2{\sigma }^2}}dt)\\ & =\frac{\mu }{\sqrt{2\pi }\sigma }{\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2{\sigma }^2}dt}\end{array}$$
$$\begin{array}{rl}& {\int }_{-\infty }^{+\infty }{e}^{a{x}^2}dx\\ & =\sqrt{({\int }_{-\infty }^{+\infty }{e}^{a{x}^2}dx{)}^2}\\ & =\sqrt{{\int }_{-\infty }^{+\infty }{e}^{a{x}^2}dx{\int }_{-\infty }^{+\infty }{e}^{a{y}^2}dy}\\ & =\sqrt{\iint e^{a(x^2+y^2)}dxdy}\\ & =\sqrt{{\int }_0^{2\pi }d\theta {\int }_0^{\infty }{e}^{a{\rho }^2}\rho d\rho }\\ & =\sqrt{\pi {\int }_0^{\infty }{e}^{a{\rho }^2}d{\rho }^2}\\ & =\sqrt{\frac{\pi }{a}{\int }_0^{\infty }d{e}^{a{\rho }^2}}\end{array}$$
当$a<0$得到$\sqrt{-\frac{\pi }{a}}$
令$a=-\frac{1}{2{\sigma }^2}$得到:
$${\int }_{-\infty }^{+\infty }{e}^{-\frac{1}{2{\sigma }^2}{x}^2}dx=\sqrt{-\pi \times (-2{\sigma }^2)}=\sqrt{2\pi }\sigma$$
$$\begin{array}{rl}E(X)& =\frac{\mu }{\sqrt{2\pi }\sigma }{\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2{\sigma }^2}dt}=\mu \end{array}$$

方差$D(X)={\sigma }^2$

$$\begin{array}{rl}D(X)& =E((X-E(X){)}^2)\\ & ={\int }_{-\infty }^{+\infty }(x-\mu {)}^{2}f(x)dx\\ & ={\int }_{-\infty }^{+\infty }(x-\mu {)}^2\frac{1}{\sqrt{2\pi }\sigma }{e}^{-\frac{(x-\mu {)}^2}{2{\sigma }^2}}dx\end{array}$$
令$t=\frac{x-\mu }{\sigma }$
$$\begin{array}{rl}& {\int }_{-\infty }^{+\infty }(x-\mu {)}^2\frac{1}{\sqrt{2\pi }\sigma }{e}^{-\frac{(x-\mu {)}^2}{2{\sigma }^2}}dx\\ & =\frac{1}{\sqrt{2\pi }\sigma }{\int }_{-\infty }^{+\infty }{\sigma }^2{t}^2{e}^{-\frac{t^2}{2}}\sigma dt\\ & =\frac{{\sigma }^2}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }{t}^2{e}^{-\frac{t^2}{2}}dt\\ & =\frac{{\sigma }^2}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }\frac{t}{2}{e}^{-\frac{t^2}{2}}d{t}^2\\ & =-\frac{{\sigma }^2}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }td(e^{-\frac{t^2}{2}})\\ & =-\frac{{\sigma }^2}{\sqrt{2\pi }}(t{d}^{-\frac{t^2}{2}}{|}_{-\infty }^{+\infty }-{\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2}}dt)\\ & =\frac{{\sigma }^2}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2}}dt\end{array}$$
又
$$\begin{array}{rl}& {\int }_{-\infty }^{+\infty }{e}^{a{x}^2}dx=\sqrt{\frac{\pi }{a}{\int }_0^{\infty }d{e}^{a{\rho }^2}}=\end{array}=\sqrt{-\frac{\pi }{a}},a<0$$
令$a=-\frac{1}{2}$得到
$${\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2}}dt=\sqrt{-\pi \times (-2)}=\sqrt{2\pi }$$
故方差为
$$D(X)=\frac{{\sigma }^2}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }{e}^{-\frac{t^2}{2}}dt={\sigma }^2$$

指数分布$X\sim E(\lambda )$

概率密度$f(x)=\{\begin{array}{l}\lambda e^{-\lambda x},x\ge 0\\ 0,x<0\end{array}$

均值$E(X)=\frac{1}{\lambda }$

$$E(X)=\lambda {\int }_0^{+\infty }x{e}^{-\lambda x}dx$$
令$t=-\lambda x$
$$\begin{array}{rl}E(X)& =-\frac{1}{\lambda }{\int }_{-\infty }^{0}t{e}^{t}dt\\ & =-\frac{1}{\lambda }(t-1)e^t{|}_{-\infty }^0\\ & =\frac{1}{\lambda }\end{array}$$

方差$D(X)=\frac{1}{{\lambda }^2}$

方均值
$$E(X^2)=\lambda {\int }_0^{+\infty }{x}^2{e}^{-\lambda x}dx=\frac{1}{{\lambda }^2}{\int }_{-\infty }^0{t}^2{e}^{t}dt=\frac{1}{{\lambda }^2}(t^2-2t+2)e^t{|}_{-\infty }^0=\frac{2}{{\lambda }^2}$$
方差
$$D(X)=E(X^2)-E^2(X)=\frac{1}{{\lambda }^2}$$