标签:引入 最优化问题 matrix alpha sum inf under 问题: set
原始问题:
假设$f(x),c_{i}(x),h_{j}(x)$是定义在$R^{n}$上的连续可微函数,考虑约束最优化问题:
$\underset{x\in R^{n}}{min}f(x)$
$s.t. \ c_{i}\leq 0,i=1,2,3...k$
$h_{j}= 0,j=1,2,3...l$
引入拉格朗日函数:
$L(x,\alpha ,\beta)=f(x)+\sum_{i=1}^{k} \alpha_{i}c_{i}(x)+\sum_{j=1}^{l}\beta_{j}h_{j}(x)$
这里,$\alpha_{i},\beta_{j}$是拉格朗日乘子,$\alpha \geq 0$。则原始问题:
$\theta_{p}(x)=\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max}L(x,\alpha,\beta)$
如果$c_{i},h_{j}$不满足条件则:
$\left\{\begin{matrix}c_{i}(x)>0 \Rightarrow \alpha_{i}c_{i} \rightarrow + \infty \\ h_{j}(x)\neq 0 \Rightarrow \beta_{j}h_{j} \rightarrow + \infty \end{matrix}\right.$
$\theta_{p}(x)=+\infty $
所以:
$\theta_{p}(x)= \left\{\begin{matrix}f(x),c_{i}\leq 0,h_{j}= 0 \\ +\infty ,other \end{matrix}\right.$
$\underset{x}{min}\theta_{p}(x)=\underset{x}{min}\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max}L(x,\alpha,\beta)$
原始问题最优解是:
$p^{*}=\underset{x}{min}\theta_{p}(x)$
对偶问题:
定义:
$\theta_{D}(\alpha,\beta)=\underset{x}{min}L(x,\alpha,\beta)$
再考虑极大化$\theta_{D}(\alpha,\beta)=\underset{x}{min}L(x,\alpha,\beta)$:
$\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max} \theta_{D}(\alpha,\beta)=\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max} \underset{x}{min}L(x,\alpha,\beta)$
这是广义拉格朗日函数的极大极小问题
所以最优解为:
$d^{*}=\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max}\theta_{D}(\alpha,\beta)$
原始问题和对偶问题的关系:
若原始问题和对偶问题都有最优解,则:
$d^{*}=\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max} \underset{x}{min} \ L(x,\alpha,\beta) \leq \underset{x}{min} \underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max} L(x,\alpha,\beta)=p^{*}$
证明:
$\theta_{D}(\alpha,\beta)=\underset{x}{min}L(x,\alpha,\beta)\leq L(x,\alpha,\beta)\leq \underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max}L(x,\alpha,\beta)=\theta _{p}(x)$
即:
$\theta_{D}(\alpha,\beta)\leq \theta _{p}(x)$
由于原始问题和对偶问题均有最优解,所以:
$\underset{\alpha_{i},\beta_{j},\alpha_{i}\geq 0}{max}\theta_{D}(\alpha,\beta)\leq \underset{x}{min}\ \theta _{p}(x)$
$x^{*},\alpha^{*},\beta^{*}$是最优解,如果它们符合KKT条件则$p^{*}=d^{*}$:
$\triangledown_{x}L(x^{*},\alpha^{*},\beta^{*})=0$
$\triangledown_{\alpha}L(x^{*},\alpha^{*},\beta^{*})=0$
$\triangledown_{\beta}L(x^{*},\alpha^{*},\beta^{*})=0$
$a_{i}^{*}c_{i}(x^{*})=0,i=1,2..k$
$c_{i}(x^{*})\leq 0,i=1,2..k$
$a_{i}^{*}\geq 0,i=1,2..k$
$h_{j}(x^{*})=0,j=1,2..l$
标签:引入 最优化问题 matrix alpha sum inf under 问题: set
原文地址:https://www.cnblogs.com/xcxy-boke/p/11456392.html