马尔科夫不等式和坎泰利不等式的证明

马尔科夫不等式(Markov’s inequality)
对于随机变量$X$,有
$$P\left(\left|X\right|⩾\epsilon \right)⩽\frac{E{\left|X\right|}^k}{{\epsilon }^k},\epsilon >0,k<\infty$$
证明：
$$\begin{gathered} P\left(\left|X\right|⩾\epsilon \right)={\int }_{|x|⩾\epsilon }f\left(x\right)dx⩽{\int }_{|x|⩾\epsilon }\frac{{\left|x\right|}^k}{{\epsilon }^k}f\left(x\right)dx \\ ⩽\frac{1}{{\epsilon }^k}{\int }_{-\infty }^{+\infty }{\left|x\right|}^{k}f\left(x\right)dx=\frac{E{\left|X\right|}^k}{{\epsilon }^k} \end{gathered}$$
坎泰利不等式(Cantelli’s inequality)
对于随机变量$X$($X>\mu$)，存在有限均值和方差，则有
$$P\left(X-\mu ⩾\epsilon \right)⩽\frac{{\sigma }^2}{{\sigma }^2+{\epsilon }^2}$$

证明：
利用马尔科夫不等式令$Y$等于$X-\mu$，则有$$P\left(Y⩾\epsilon \right)⩽\frac{E\left(Y^2\right)}{{\epsilon }^2}$$取$\beta >0$,有$$P\left(Y+\beta ⩾\epsilon +\beta \right)⩽\frac{E{\left(Y+\beta \right)}^2}{{\left(\epsilon +\beta \right)}^2}$$$$E{\left(Y+\beta \right)}^2=E\left(Y^2+2Y\beta +{\beta }^2\right)={\sigma }^2+{\beta }^2\left(EY=0\right)$$,所以$$P\left(Y+\beta ⩾\epsilon +\beta \right)⩽\frac{{\sigma }^2+{\beta }^2}{{\left(\epsilon +\beta \right)}^2}$$令$$g\left(\beta \right)=\frac{{\sigma }^2+{\beta }^2}{{\left(\epsilon +\beta \right)}^2}$$
则$$g\prime \left(\beta \right)=\frac{2\left(\epsilon \beta -{\sigma }^2\right)}{{\left(\epsilon +\beta \right)}^3}$$
当$\beta =\frac{{\sigma }^2}{\epsilon }$时$g\left(\beta \right)$有最小值，
$$g_{min}\left(\beta \right)=g\left(\frac{{\sigma }^2}{\epsilon }\right)=\frac{{\sigma }^2+\frac{{\sigma }^4}{{\epsilon }^2}}{{\left(\epsilon +\frac{{\sigma }^2}{\epsilon }\right)}^2}=\frac{{\sigma }^2}{{\epsilon }^2+{\sigma }^2}$$
因此，$$P\left(Y⩾\epsilon \right)⩽\frac{{\sigma }^2}{{\sigma }^2+{\epsilon }^2}$$