矩量母函数介绍

1 矩量母函数

 矩量母函数又称矩母函数（Moment Generating Function）又称动差生成函数，是一种构造函数，其定义为：

随机变量$X$是连续型随机变量时，其矩量母函数为：$$M_X(t)=\mathrm{E}(e^{tX})={\int }_{-\infty }^{+\infty }{e}^{tx}f(x)dx$$随机变量$X$是离散型随机变量时，其矩量母函数为：$$M_X(t)=\mathrm{E}(e^{tX})=\sum _{x_i}{e}^{t{x}_i}p(x_i)$$

 由泰勒级数可知$$e^x=1+x+\frac{x^2}{2!}+\frac{x^3}{3!}+\frac{x^4}{4!}+\cdots +\frac{x^n}{n!}+\cdots$$得到：$$\begin{array}{rl}{M}_X(t)& ={\int }_{-\infty }^{+\infty }(1+xt+\frac{x^2{t}^2}{2!}+\frac{x^3{t}^3}{3!}+\frac{x^4{t}^4}{4!}+\cdots +\frac{x^n{t}^n}{n!}+\cdots )f(x)dx\\ & ={\int }_{-\infty }^{+\infty }f(x)dx+t{\int }_{-\infty }^{+\infty }xf(x)dx+\frac{t^2}{2!}{\int }_{-\infty }^{+\infty }{x}^{2}f(x)dx+\cdots \frac{t^n}{n!}{\int }_{-\infty }^{+\infty }{x}^{n}f(x)dx+\cdots \\ & =t^0{M}_0+t^1{M}_1+\frac{t^2}{2!}{M}_2+\cdots +\frac{t^n}{n!}{M}_n+\cdots \end{array}$$其中，$M_n$即为$X$的$n$阶中心距。
 矩量母函数对$t$求$n$阶导可得$$\begin{array}{rl}{M}_X^{(n)}(t)& =(t^0{M}_0{)}^{(n)}+(t^1{M}_1{)}^{(n)}+{\left(\frac{t^2}{2!}{M}_2\right)}^{(n)}+\cdots +{\left(\frac{t^n}{n!}{M}_n\right)}^{(n)}+{\left(\frac{t^{(n+1)}}{(n+1)!}{M}_{n+1}\right)}^{(n)}+\cdots \\ & =0+0+0+\cdots +M_n+\frac{(n+1)\cdot n\cdot (n-1)\cdots 2\cdot t}{(n+1)!}{M}_{n+1}+\cdots \end{array}$$当$t=0$时，则有
$$\begin{array}{rl}{M}_X^{(n)}(0)& =0+0+0+\cdots +M_n+\frac{(n+1)\cdot n\cdot (n-1)\cdots 2\cdot 0}{(n+1)!}{M}_{n+1}+\cdots 0+\cdots \\ & =M_n\end{array}$$由此可知随机变量$X$的均值和方差分别为：$$\mathrm{E}(X)=M_X^{(1)}(0)={\int }_{-\infty }^{+\infty }xf(x)dx=M_1$$$$\mathrm{V}\mathrm{a}\mathrm{r}(X)=\mathrm{E}(X^2)-[\mathrm{E}(X){]}^2={\int }_{-\infty }^{+\infty }{x}^{2}p(x)dx-{\left({\int }_{-\infty }^{+\infty }xp(x)dx\right)}^2=M_2-(M_1{)}^2$$

2 参数为$n$和$p$的二项分布

 离散随机变量$X$服从参数为$n$和$p$的二项分布，则其矩母函数为$$\begin{array}{rl}\varphi (t)& =\mathrm{E}[e^{tX}]=\sum _{k=0}^n{e}^{tk}\left(\begin{array}{c}n\\ k\end{array}\right)p^k(1-p{)}^{n-k}\\ & =\sum _{k=0}^n\left(\begin{array}{c}n\\ k\end{array}\right)(p{e}^t{)}^k(1-p{)}^{n-k}\\ & =(p{e}^t+1-p{)}^n\end{array}$$因此$${\varphi }^{\prime }(t)=n(p{e}^t+1-p{)}^{n-1}p{e}^t$$所以则有$$\mathrm{E}[X]={\varphi }^{\prime }(0)=np$$求二阶导则有$${\varphi }^{\prime \prime }(t)=n(n-1)(p{e}^t+1-p{)}^{n-2}(p{e}^t{)}^2+n(p{e}^t+1-p{)}^{n-1}p{e}^t$$所以$$\mathrm{E}[X^2]={\varphi }^{\prime \prime }(0)=n(n-1)p^2+np$$因此，$X$的方差为$$\mathrm{V}\mathrm{a}\mathrm{r}(X)=\mathrm{E}[X^2]-(\mathrm{E}[X]{)}^2=n(n-1)p^2+np-n^2{p}^2=np(1-p)$$

3 均值为$\lambda$的泊松分布

 离散随机变量$X$服从均值为$\lambda$的泊松分布，则其矩母函数为 ϕ ( t ) = E [ e t X ] = ∑ k = 0 ∞ e t n e − λ λ n n ! = e − λ ∑ n = 0 ∞ ( λ e t ) n n ! = e − λ e λ e t = exp ⁡ { λ ( e t − 1 ) } \phi(t)=\mathrm{E}[e^{tX}]=\sum\limits_{k=0}^\infty\frac{e^{tn}e^{-\lambda}\lambda^n}{n!}=e^{-\lambda}\sum\limits_{n=0}^{\infty}\frac{(\lambda e^t)^n}{n!}=e^{-\lambda}e^{\lambda e^t}=\exp\{\lambda(e^t-1)\} ϕ(t)=E[etX]=k=0∑∞​n!etne−λλn​=e−λn=0∑∞​n!(λet)n​=e−λeλet=exp{λ(et−1)}求微分可得 ϕ ′ ( t ) = λ e t exp ⁡ { λ ( e t − 1 ) } ϕ ′ ′ ( t ) = ( λ e t ) 2 exp ⁡ { λ ( e t − 1 ) } + λ e t { λ ( e t − 1 ) } \begin{aligned}\phi^{\prime}(t)&=\lambda e^t\exp\{\lambda(e^t-1)\}\\\phi^{\prime\prime}(t)&=(\lambda e^t)^2 \exp\{\lambda(e^t-1)\}+\lambda e^t\{\lambda(e^t-1)\}\end{aligned} ϕ′(t)ϕ′′(t)​=λetexp{λ(et−1)}=(λet)2exp{λ(et−1)}+λet{λ(et−1)}​所以则有$$\begin{array}{rl}\mathrm{E}[X]& ={\varphi }^{\prime }(0)=\lambda \\ \mathrm{E}[X]& ={\varphi }^{\prime \prime }(0)={\lambda }^2+\lambda \\ \mathrm{V}\mathrm{a}\mathrm{r}(X)& =\mathrm{E}[X^2]-(\mathrm{E}[X]{)}^2=\lambda \end{array}$$因此，泊松分布的均值和方差都是$\lambda$。

4 参数为$\lambda$的指数分布

 离散随机变量$X$服从参数为$\lambda$的指数分布，则其矩母函数为$$\varphi (t)=\mathrm{E}[e^{tX}]={\int }_0^{\infty }{e}^{tx}\lambda e^{-\lambda x}dx=\lambda {\int }_0^{\infty }{e}^{-(\lambda -t)x}dx=\frac{\lambda }{\lambda -t},t<\lambda$$从上面的推导可以发现，对于指数分布，$\varphi (t)$只对小于$\lambda$的$t$值定义。对$\varphi (t)$微分可以得到$${\varphi }^{\prime }(t)=\frac{\lambda }{(\lambda -t{)}^2},{\varphi }^{\prime \prime }(t)=\frac{2\lambda }{(\lambda -t{)}^3}$$因此$$\mathrm{E}[X]={\varphi }^{\prime }(0)=\frac{1}{\lambda },\mathrm{E}[X^2]={\varphi }^{\prime \prime }(0)=\frac{2}{{\lambda }^2}$$于是$X$的方差为$$\mathrm{V}\mathrm{a}\mathrm{r}(X)=\mathrm{E}[X^2]-(\mathrm{E}[X]{)}^2=\frac{1}{{\lambda }^2}$$

5 参数为$\mu$和${\delta }^2$的正态分布

 标准正态随机变量$X$的矩母函数如下所示$$\begin{array}{rl}\mathrm{E}[e^{tX}]& =\frac{1}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }{e}^{tx}{e}^{-x^2/2}dx=\frac{1}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }{e}^{-(x^2-2tx)}dx\\ & =e^{t^2/2}\frac{1}{\sqrt{2\pi }}{\int }_{-\infty }^{+\infty }{e}^{-(x-t{)}^2/2}dx=e^{t^2/2}\end{array}$$如果$X$是标准正态分布，那么$Z=\sigma X+\mu$就是参数为$\mu$和${\sigma }^2$的正态分布，则有$$\varphi (t)=\mathrm{E}[e^{tZ}]=\mathrm{E}[e^{t(\sigma X+\mu )}]=e^{tu}\mathrm{E}[e^{t\sigma X}]=\exp \left\{\frac{{\sigma }^2{t}^2}{2}+\mu t\right\}$$经过微分可以得到 ϕ ′ ( t ) = ( μ + t σ 2 ) exp ⁡ { σ 2 t 2 2 + μ t } ϕ ′ ′ ( t ) = ( μ + t σ 2 ) 2 exp ⁡ { σ 2 t 2 2 + μ t } + σ 2 exp ⁡ { σ 2 t 2 2 + μ t } \begin{aligned}\phi^{\prime}(t)&=(\mu+t\sigma^2)\exp\{\frac{\sigma^2 t^2}{2}+\mu t\}\\\phi^{\prime\prime}(t)&=(\mu+ t\sigma^2)^2\exp\{\frac{\sigma^2 t^2}{2}+\mu t\}+\sigma^2 \exp \left\{\frac{\sigma^2 t^2}{2}+\mu t\right\}\end{aligned} ϕ′(t)ϕ′′(t)​=(μ+tσ2)exp{2σ2t2​+μt}=(μ+tσ2)2exp{2σ2t2​+μt}+σ2exp{2σ2t2​+μt}​所以则有$$\mathrm{E}[X]={\varphi }^{\prime }(0)=\mu ,\mathrm{E}[X^2]={\varphi }^{\prime \prime }(0)={\mu }^2+{\sigma }^2$$方差为$$\mathrm{V}\mathrm{a}\mathrm{r}(\mathrm{X})=\mathrm{E}[X^2]-(\mathrm{E}[X]{)}^2={\sigma }^2$$