欢迎光临散文网 会员登陆 & 注册

拉格朗日函数,对偶函数,(凸)共轭函数

2023-06-28 11:37 作者:师翊  | 我要投稿

考虑一个最优化问题:

%5Cbegin%7Barray%7D%7Bl%7D%7B%5Coperatorname*%7Bmin%7D~f(x)%7D%5C%5C%20%7Bs.t.~l_%7Bi%7D(x)%5Cleqslant0%7D%26%7Bi%3D1%2C...%2Cm%7D%5C%5C%5Cqquad%7Bh_%7Bj%7D(x)%3D0%7D%26%7Bj%3D1%2C...%2Cn%7D%5C%5C%20%5Cend%7Barray%7D

该最优化问题的定义域:

D%3D%5Ctext%7Bdom%7Df%5Cleft(%20x%20%5Cright)%20%5Ccap%20%5Ctext%7Bdom%7Dl_i%5Cleft(%20x%20%5Cright)%20%5Ccap%20%5Ctext%7Bdom%7Dh_i%5Cleft(%20x%20%5Cright)%20

上述问题的拉格朗日函数为:

L%5Cleft(%20x%2C%5Clambda%20%2C%5Cmu%20%5Cright)%20%3Df_0%5Cleft(%20x%20%5Cright)%20%2B%5Csum_%7Bi%3D1%7D%5Em%7B%5Clambda%20_il_i%7D%5Cleft(%20x%20%5Cright)%20%2B%5Csum_%7Bj%3D1%7D%5En%7B%5Cmu%20_j%7Dh_j%5Cleft(%20x%20%5Cright)%20

该问题的对偶函数为:

g%5Cleft(%20%5Clambda%20%2C%5Cmu%20%5Cright)%20%3D%5Cunderset%7Bx%5Cin%20D%7D%7B%5Ctext%7Binf%7D%7DL%5Cleft(%20x%2C%5Clambda%20%2C%5Cmu%20%5Cright)%20%3D%5Cunderset%7Bx%5Cin%20D%7D%7B%5Ctext%7Binf%7D%7D%5Cleft%5C%7B%20f%5Cleft(%20x%20%5Cright)%20%2B%5Csum_%7Bi%3D1%7D%5Em%7B%5Clambda%20_il_i%5Cleft(%20x%20%5Cright)%20%2B%5Csum_%7Bj%3D1%7D%5En%7B%5Cmu%20_ih_i%5Cleft(%20x%20%5Cright)%7D%7D%20%5Cright%5C%7D

f(x)的(凸)共轭函数的定义为:

f%5E*%5Cleft(%20y%20%5Cright)%20%3D%5Cunderset%7Bx%5Cin%20%5Ctext%7Bdom%7Df%7D%7B%5Ctext%7Bsup%7D%7D%5Cleft(%20y%5ETx-f%5Cleft(%20x%20%5Cright)%20%5Cright)%20%3D-%5Cunderset%7Bx%5Cin%20%5Ctext%7Bdom%7Df%7D%7B%5Ctext%7Binf%7D%7D%5Cleft(%20f%5Cleft(%20x%20%5Cright)%20-y%5ETx%20%5Cright)%20

下面推导共轭函数和对偶函数的关系。首先将约束条件矩阵化,考虑如下最优化问题:

%5Cmin%5Ctext%7B%20%7Df%5Cleft(%20%5Cboldsymbol%7Bx%7D%20%5Cright)%20%0A%5C%5Cs.t.%5C%20%5Cboldsymbol%7BAx%7D%5Cle%20%5Cboldsymbol%7Bb%7D%0A%0A%5C%20%5C%20%5C%20%5C%20%5C%5C%5Cboldsymbol%7BCx%7D%3D%5Cboldsymbol%7Bd%7D

其对偶函数为:

g%5Cleft(%20%5Cboldsymbol%7B%5Clambda%20%2C%5Cmu%20%7D%20%5Cright)%20%3D%5Cunderset%7Bx%5Cin%20D%7D%7B%5Ctext%7Binf%7D%7D%5C%7Bf%5Cleft(%20%5Cboldsymbol%7Bx%7D%20%5Cright)%20%2B%5Cboldsymbol%7B%5Clambda%20%7D%5ET%5Cleft(%20%5Cboldsymbol%7BAx%7D-%5Cboldsymbol%7Bb%7D%20%5Cright)%20%2B%5Cboldsymbol%7B%5Cmu%20%7D%5ET%5Cleft(%20%5Cboldsymbol%7BCx%7D-%5Cboldsymbol%7Bd%7D%20%5Cright)%20%5C%7D%0A%5C%5C%3D%5Cunderset%7Bx%5Cin%20D%7D%7B%5Ctext%7Binf%7D%7D%5Cleft%5C%7B%20f%5Cleft(%20%5Cboldsymbol%7Bx%7D%20%5Cright)%20%2B%5Cleft(%20%5Cboldsymbol%7BA%7D%5ET%5Cboldsymbol%7B%5Clambda%20%7D%2B%5Cboldsymbol%7BC%7D%5ET%5Cboldsymbol%7B%5Cmu%20%7D%20%5Cright)%20%5ET%5Cboldsymbol%7Bx%7D%20%5Cright%5C%7D%20-%5Cboldsymbol%7B%5Clambda%20%7D%5ET%5Cboldsymbol%7Bb%7D-%5Cboldsymbol%7B%5Cmu%20%7D%5ET%5Cboldsymbol%7Bd%7D%0A%5C%5C%3D%5Cunderset%7Bx%5Cin%20D%7D%7B%5Ctext%7Binf%7D%7D%5Cleft%5C%7B%20f%5Cleft(%20%5Cboldsymbol%7Bx%7D%20%5Cright)%20-%5Cleft(%20-%5Cboldsymbol%7BA%7D%5ET%5Cboldsymbol%7B%5Clambda%20%7D-%5Cboldsymbol%7BC%7D%5ET%5Cboldsymbol%7B%5Cmu%20%7D%20%5Cright)%20%5ET%5Cboldsymbol%7Bx%7D%20%5Cright%5C%7D%20-%5Cboldsymbol%7B%5Clambda%20%7D%5ET%5Cboldsymbol%7Bb%7D-%5Cboldsymbol%7B%5Cmu%20%7D%5ET%5Cboldsymbol%7Bd%7D%0A%5C%5C%5Ctext%7B%E8%A7%82%E5%AF%9F%E5%AF%B9%E5%81%B6%E5%87%BD%E6%95%B0%E5%AE%9A%E4%B9%89%EF%BC%8C%E5%8D%B3%E4%B8%8A%E5%BC%8F%E4%B8%AD%7D%5Cleft(%20-%5Cboldsymbol%7BA%7D%5ET%5Cboldsymbol%7B%5Clambda%20%7D-%5Cboldsymbol%7BC%7D%5ET%5Cboldsymbol%7B%5Cmu%20%7D%20%5Cright)%20%5Ctext%7B%E6%98%AF%E5%AF%B9%E5%81%B6%E5%87%BD%E6%95%B0%E4%B8%AD%E7%9A%84%7Dy%0A%5C%5C%3D-f%5E*%5Cleft(%20-%5Cboldsymbol%7BA%7D%5ET%5Cboldsymbol%7B%5Clambda%20%7D-%5Cboldsymbol%7BC%7D%5ET%5Cboldsymbol%7B%5Cmu%20%7D%20%5Cright)%20-%5Cboldsymbol%7B%5Clambda%20%7D%5ET%5Cboldsymbol%7Bb%7D-%5Cboldsymbol%7B%5Cmu%20%7D%5ET%5Cboldsymbol%7Bd%7D%0A

因此对偶函数可以通过共轭函数表示。而共轭函数又可以通过原函数表示。那么共轭函数将对偶函数和原函数连接起来。





拉格朗日函数,对偶函数,(凸)共轭函数的评论 (共 条)

分享到微博请遵守国家法律