|
设$D \subset R^n$是开集,$f: D \to R$,$\phi: D \to R^m$,$n=m+r$,并改用行向量记$x=(x_1,\cdots,x_n=(x_1,\cdots,x_r,x_{r+1},\cdots,x_{r+m})=(y,z)$,$y \in R^r$,$z \in R^m$。现在讨论在条件
$$\phi(x)=\phi(y,z)=0$$
限制下,求函数$f(x)=f(y,z)$的极值。对于这个条件极值问题,它的Largrange函数为
$$L(x,\lambda)=L(y,z,\lambda)=f(y,z)+\lambda^T\phi(y,z),$$
其中$\lambda=(\lambda_1,\cdots,\lambda_m)^T$为Largrange乘数向量。于是Largrange乘数法的向量形式是:
定理 对以上所设的函数$f$,$\phi$若满足条件:
(i)$f$,$\phi$在$D$内有连续导数;
(ii)$\phi(x_0)=\phi(y_0,z_0)=0$;
(iii)${\rm rank} \phi'(x_0)={\rm rank}[\phi_y'(y_0,z_0),\phi_z'(y_0,z_0)]=m$;
(iv)$x_0=(y_0,z_0)$是$f$在条件
$$\phi(x)=\phi(y,z)=0$$
下的条件极值点,
则存在$\Lambda_0 \in R^m$,使得$(x_0,\Lambda_0)$是
$$L(x,\lambda)=L(y,z,\lambda)=f(y,z)+\lambda^T\phi(y,z),$$
所设函数$L$的稳定点即满足
$$L'(x_0,\Lambda_0)=[L_x(x_0,\Lambda_0)+L_\lambda(x_0,\Lambda_0)]=0。$$
但因$L_\lambda(x_0,\Lambda_0)=[\phi(x_0)]^T=0$(条件(ii)),故上式等同于
$$L_x(x_0,\Lambda_0)=f'(x_0)+\Lambda_0^T \phi'(x_0)=0。$$ |
|