Multiple or scalar Lagrange multipler, which is a simpler method?

81 Views Asked by At

Without loss of generality, consider minimizing

$F(x,y,z)$

subject to constraints

$G(x,y,z)=0$

$H(x,y,z)=0$

Multiple constraints can be reduced to a single one:

$G^2+H^2=0$

Then, Lagrange method generates equations:

$ {\partial F \over \partial x} + 2 \lambda (G {\partial G \over \partial x}+H {\partial H \over \partial x}) = 0$

$ {\partial F \over \partial y} + 2 \lambda (G {\partial G \over \partial y}+H {\partial H \over \partial y}) = 0$

$ {\partial F \over \partial z} + 2 \lambda (G {\partial G \over \partial z}+H {\partial H \over \partial z}) = 0$

When comparing this to equations produced with canonical method with multiple Lagrange multipliers, it seems that each of the component is proportional to this scalar $\lambda$. That is

$\lambda_G = 2 G \lambda$

$\lambda_H = 2 H \lambda$

I wonder if this observation is mathematical folklore.

1

There are 1 best solutions below

0
On

The problem is, that your transformation of the constraints will render constraint qualifications unsatisfiable. In particular LICQ is never satisfied, since the derivative of $G^2 + H^2$ is zero if $G = H = 0$.

Moreover, if $G = H = 0$ is satisfied, you obtain $\lambda_G = \lambda_H = 0$ with your approach.