K.K.T. conditions, Lagrangian gradient not defined for zero.

469 Views Asked by At

When I write the K.K.T. conditions for the problem I have, I get the following expression for the gradient of the Lagrangian:

$$\frac{\partial \mathcal{L}}{\partial x} = - \frac{\sqrt{x} + \sqrt{y}}{\sqrt{x}\left(1+(\sqrt{x}+\sqrt{y})^2\right)} + A = 0$$

$$\frac{\partial \mathcal{L}}{\partial y} = - \frac{\sqrt{x} + \sqrt{y}}{\sqrt{y}\left(1+(\sqrt{x}+\sqrt{y})^2\right)} + B = 0$$

Where $A$ and $B$ are the functions of the corresponding Lagrangian multipliers only.

Since Lagrangian partial derivatives are not defined for $x=0$ and $y=0$, can I say that therefore $x \neq 0$ and $y \neq 0$ for the optimal solution. In other words, I need to have strict inequalities $x > 0$ and $y > 0$ instead of $x \geq 0$ and $y \geq 0$ at the optimum point.

Thank you very much in advance, Lazar

I provide the complete set of K.K.T. conditions in the sequel:

$\frac{\partial \mathcal{L}}{\partial x_k}, \frac{\partial \mathcal{L}}{\partial y_k} = 0$

$\sum_{k=1}^{n} x_k \leq C_n$ $\text{for}\; n=1\ldots N$

$\sum_{k=1}^{n} y_k \leq D_n$ $\text{for}\; n=1\ldots N$

$x_k,y_k \geq 0$

$\lambda_n, \nu_n, \mu_k, \xi_k \geq 0$

$\lambda_n \left(\sum_{k=1}^{n} x_k - C_n\right) = 0$ $\text{for}\; n=1\ldots N$

$\nu_n \left(\sum_{k=1}^{n} y_k - D_n\right) = 0$ $\text{for}\; n=1\ldots N$

$-\mu_k x_k = 0$ $\text{for}\; k=1\ldots N$

$-\xi_k y_k = 0$ $\text{for}\; k=1\ldots N$

I want to see whether the KKT conditions (derivatives given at the beginning) imply $x \neq 0$ and $y \neq 0$, resulting in $\mu = 0$ and $\xi = 0$ from two equations at the end.

Thank you.