康托洛维奇不等式

标量形式

$a_i,{\lambda }_i>0(i=1,2,\cdots ,n)$,${\sum }_{i=1}^n{a}_i=1$,${\lambda }_1\le {\lambda }_2\le \cdots \le {\lambda }_n$,则
$$(\sum _{i=1}^n{\lambda }_i{a}_i)(\sum _{i=1}^n{\lambda }_i^{-1}{a}_i)\le \frac{({\lambda }_1+{\lambda }_n{)}^2}{4{\lambda }_1{\lambda }_n}$$
证明：
$$f(x)=(\sum _{i=1}^n{\lambda }_i^{-1}{a}_i)x^2-\frac{{\lambda }_1+{\lambda }_n}{\sqrt{{\lambda }_1{\lambda }_n}}x+\sum _{i=1}^n{\lambda }_i{a}_i$$
$$\begin{array}{rl}f(\sqrt{{\lambda }_1{\lambda }_n})& =(\sum _{i=1}^n{\lambda }_i^{-1}{a}_i){\lambda }_1{\lambda }_n-({\lambda }_1+{\lambda }_n)+\sum _{i=1}^n{\lambda }_i{a}_i\\ & ={\lambda }_n{a}_1+{\lambda }_1{a}_n+(\sum _{i=2}^{n-1}{\lambda }_i^{-1}{a}_i){\lambda }_1{\lambda }_n-({\lambda }_1+{\lambda }_n)+{\lambda }_1{a}_1+{\lambda }_n{a}_n+\sum _{i=2}^{n-1}{\lambda }_i{a}_i\\ & =(a_1+a_n-1)({\lambda }_1+{\lambda }_n)+(\sum _{i=2}^{n-1}{\lambda }_i^{-1}{a}_i){\lambda }_1{\lambda }_n+\sum _{i=2}^{n-1}{\lambda }_i{a}_i\\ & =(a_1+a_n-\sum _{i=1}^n{a}_i)({\lambda }_1+{\lambda }_n)+(\sum _{i=2}^{n-1}{\lambda }_i^{-1}{a}_i){\lambda }_1{\lambda }_n+\sum _{i=2}^{n-1}{\lambda }_i{a}_i\\ & =-\sum _{i=2}^{n-1}{a}_i({\lambda }_1+{\lambda }_n)+(\sum _{i=2}^{n-1}{\lambda }_i^{-1}{a}_i){\lambda }_1{\lambda }_n+\sum _{i=2}^{n-1}{\lambda }_i{a}_i\\ & =\sum _{i=2}^{n-1}{a}_i({\lambda }_1+{\lambda }_n+{\lambda }_i^{-1}{\lambda }_1{\lambda }_n+{\lambda }_i)\\ & =\sum _{i=2}^{n-1}{a}_i\frac{{\lambda }_i^2+({\lambda }_1+{\lambda }_n){\lambda }_i+{\lambda }_1{\lambda }_n}{{\lambda }_i}\\ & =\sum _{i=2}^{n-1}{a}_i\frac{({\lambda }_i-{\lambda }_1)({\lambda }_i-{\lambda }_n)}{{\lambda }_i}\\ & \le 0\end{array}$$
$f(x)$是抛物线，开口向上，且至少有一个零点,所以$f(x)=0$的判别式$\Delta =b^2-4ac\ge 0$
从而
$$(\sum _{i=1}^n{\lambda }_i{a}_i)(\sum _{i=1}^n{\lambda }_i^{-1}{a}_i)\le \frac{({\lambda }_1+{\lambda }_n{)}^2}{4{\lambda }_1{\lambda }_n}$$

矩阵形式

矩阵$G\in S_{++}^n$,其特征值${\lambda }_1\le {\lambda }_2\le \cdots \le {\lambda }_n,$则，$\forall x\in R^n$,有
$$\frac{(x^{T}Gx)(x^T{G}^{-1}x)}{(x^{T}x{)}^2}\le \frac{({\lambda }_1+{\lambda }_n{)}^2}{4{\lambda }_1{\lambda }_n}$$
证明：
$G\in S_{++}^n$,实对称矩阵必可对角化
存在正交矩阵$P$,使得$G=P^T\Lambda P$,其中$\Lambda P=diag({\lambda }_1,{\lambda }_2,\cdots ,{\lambda }_n)$
设$y=Px,a_i=y_i^2$
$$\begin{array}{rl}\frac{(x^{T}Gx)(x^T{G}^{-1}x)}{(x^{T}x{)}^2}& =\frac{(x^T{P}^T\Lambda Px)(x^T{P}^T{\Lambda }^{-1}Px)}{(x^T{P}^{T}Px{)}^2}\\ & =\frac{(y^T\Lambda y)(y^T{\Lambda }^{-1}y)}{(y^{T}y{)}^2}\\ & =\frac{(y^T\Lambda y)(y^T{\Lambda }^{-1}y)}{(y^{T}y{)}^2}\\ & =\frac{({\sum }_{i=1}^n{\lambda }_i{y}_i^2)({\sum }_{i=1}^n{\lambda }_i^{-1}{y}_i^2)}{({\sum }_{i=1}^n{y}_i^2{)}^2}\\ & =\frac{({\sum }_{i=1}^n{\lambda }_i{a}_i)({\sum }_{i=1}^n{\lambda }_i^{-1}{a}_i)}{({\sum }_{i=1}^n{a}_i{)}^2}\\ & =(\sum _{i=1}^n{\lambda }_i\frac{a_i}{{\sum }_{j=1}^n{a}_j})(\sum _{i=1}^n{\lambda }_i^{-1}\frac{a_i}{{\sum }_{i=j}^n{a}_j})(\text{其中}\sum _{i=1}^n\frac{a_i}{{\sum }_{j=1}^n{a}_j}=1)\\ & \le \frac{({\lambda }_1+{\lambda }_n{)}^2}{4{\lambda }_1{\lambda }_n}(\text{标量形式})\end{array}$$

积分形式

$f(x)$在$\left[a,b\right]$上是连续正值函数，$M=\underset{a\le x\le b}{\max }f(x)$,$m=\underset{a\le x\le b}{\min }f(x)$,则
$${\int }_a^{b}f(x)\mathrm{d}x{\int }_a^b\frac{1}{f(x)}\mathrm{d}x\le \frac{(M+m{)}^2}{4Mm}(b-a{)}^2$$
证明：
$$\begin{array}{rl}g(t)& ={\int }_a^{b}f(x)\mathrm{d}x{t}^2-\frac{M+m}{\sqrt{mM}}(b-a)t+{\int }_a^b\frac{1}{f(x)}\mathrm{d}x\\ & ={\int }_a^{b}f(x)\mathrm{d}x{t}^2-{\int }_a^b\frac{M+m}{\sqrt{mM}}t\mathrm{d}x+{\int }_a^b\frac{1}{f(x)}\mathrm{d}x\\ & ={\int }_a^b(f(x)t^2-\frac{M+m}{\sqrt{mM}}t+\frac{1}{f(x)})\mathrm{d}x\\ & ={\int }_a^b\frac{\sqrt{mM}{f}^2(x)t^2-(M+m)f(x)t+\sqrt{mM}}{\sqrt{mM}f(x)}\mathrm{d}x\\ & ={\int }_a^b\frac{t^2(f(x)-\frac{1}{Mt})(f(x)-\frac{1}{mt})}{f(x)}\mathrm{d}x\end{array}$$
$$\frac{1}{Mt}=m\Rightarrow t=\frac{1}{Mm}\Rightarrow g(\frac{1}{Mm})={\int }_a^b\frac{(f(x)-m)(f(x)-M)}{(MN{)}^{2}f(x)}\mathrm{d}x\le 0$$
$g(t)$是抛物线，开口向上，且至少有一个零点,所以$g(t)=0$的判别式$\Delta =b^2-4ac\ge 0$
从而
$${\int }_a^{b}f(x)\mathrm{d}x{\int }_a^b\frac{1}{f(x)}\mathrm{d}x\le \frac{(M+m{)}^2}{4Mm}(b-a{)}^2$$