Markov‘s inequality和Chebyshev‘s inequality

定义

with $m>0$:

Markov’s inequality
$$X⩾0,\mathrm{Pr}(X>m)⩽\frac{\mathbb{E}(X)}{m}$$

Chebyshev’s inequality
$$X\in \mathbb{R},\mathrm{Pr}(|X-\mathbb{E}(X)|>m)⩽\frac{\mathrm{V}\mathrm{a}\mathrm{r}(X)}{m^2}$$

one sided Chebyshev’s inequality
$$\begin{gathered} X\in \mathbb{R},\mathrm{Pr}((X-\mathbb{E}(X))>m)⩽\frac{\mathrm{V}\mathrm{a}\mathrm{r}(X)}{\mathrm{V}\mathrm{a}\mathrm{r}(X)+m^2} \\ X\in \mathbb{R},\mathrm{Pr}((X-\mathbb{E}(X))<-m)⩽\frac{\mathrm{V}\mathrm{a}\mathrm{r}(X)}{\mathrm{V}\mathrm{a}\mathrm{r}(X)+m^2} \end{gathered}$$

分析

理论上, Chebyshev’s inequality 和 one sided Chebyshev’s inequality 都是 Markov’s inequality 的推论; 在一定条件下, Chebyshev’s inequality 和 one sided Chebyshev’s inequality 能够互相转化.

直觉上(不严谨的) “Markov’s inequality 强于 Chebyshev’s inequality 强于 one sided Chebyshev’s inequality”, 但是实际计算结果往往是“越强的不等式的界越松”.
A能推导出B(A比B强)但是A的界却比B的界松, 这看起来违背直觉, 但实际上是合理的, 因为A能推导出B的过程不是简单的“Pr(𝒳⩾𝓂)⩽A(𝓂)⩽B(𝓂)”, 而是通过对𝒳进行变换, 挤掉分布中的水分得到更紧的界, “Pr(𝒳⩾𝓂)⩽A(𝓂), Pr(φ(𝒳)⩾φ(𝓂))⩽A(φ(𝓂))=B(𝓂), 通过恰当选择φ实现A(φ(𝓂))=B(𝓂)⩽A(𝓂)”.

附录(mermaid流程图绘制)

%%{init: {'theme': 'base', 'themeVariables': { 'fontFamily': 'MS Serif', 'fontSize' : '12px'}}}%%

flowchart LR
    subgraph corollary
        direction TB
        C[Chebyshev's inequality]
        OC[one sided Chebyshev's inequality]
        C -.->|"Var[&Xscr;] < &mscr;&sup2;<br/>Pr(&Xscr; - &Eopf;&Xscr; < -&xscr;) = Pr(&Xscr; - &Eopf;&Xscr; > &xscr;)"| OC
        OC -->|"Var[&Xscr;] > &mscr;&sup2;"| C
    end
    M[Markov's inequality]
    M -->|"implies"| corollary