Find the solution using the KKT conditions.

  • MHB
  • Thread starter mathmari
  • Start date
  • Tags
    Conditions
In summary, the conversation discusses finding the solution to a nonlinear problem using the KKT conditions. The solution is found to be $(0,9)$ after considering two cases, one where $μ_1=0$ and the other where $μ_1\ne0$. This solution is unique and minimizes the objective function.
  • #1
mathmari
Gold Member
MHB
5,049
7
Hi! :eek:

Given then following problem:
$$ \min [-x_1 - (x_2 -9)^2]$$
subject to $$4x_1^2+x_2^2 \leq 400$$
I have to find the solution using the KKT conditions.

I have found that the solution is $(0,9)$. Is my result correct?
 
Mathematics news on Phys.org
  • #2
mathmari said:
Hi! :eek:

Given then following problem:
$$ \min [-x_1 - (x_2 -9)^2]$$
subject to $$4x_1^2+x_2^2 \leq 400$$
I have to find the solution using the KKT conditions.

I have found that the solution is $(0,9)$. Is my result correct?

Hey! :eek:

I don't understand?? :confused:

How about for instance $(0,-9)$?
Doesn't it have a lower minimum while still satisfying the condition?
 
  • #3
I like Serena said:
Hey! :eek:

I don't understand?? :confused:

How about for instance $(0,-9)$?
Doesn't it have a lower minimum while still satisfying the condition?

Having the nonlinear problem:
$ \max f(x)$
$ g_i (x) \leq 0$
$x \geq 0$

the KKT conditions are:
$$μ_i \geq 0$$
$$\frac{\partial f}{\partial x_j}(x^*)-\sum_{i=1}^{p}μ_i \frac{\partial g_i}{\partial x_j}(x^*) \leq 0$$
$$x_j^*(\frac{\partial f}{\partial x_j}(x^*)-\sum_{i=1}^{p}μ_i \frac{\partial g_i}{\partial x_j}(x^*))=0$$
$$g_i(x^*) \leq 0$$
$$μ_i g_i (x^*) =0$$
$$x_j^* \geq 0$$

So the solution cannot have negative values.
 
  • #4
mathmari said:
Having the nonlinear problem:
$ \max f(x)$
$ g_i (x) \leq 0$
$x \geq 0$

the KKT conditions are:
$$μ_i \geq 0$$
$$\frac{\partial f}{\partial x_j}(x^*)-\sum_{i=1}^{p}μ_i \frac{\partial g_i}{\partial x_j}(x^*) \leq 0$$
$$x_j^*(\frac{\partial f}{\partial x_j}(x^*)-\sum_{i=1}^{p}μ_i \frac{\partial g_i}{\partial x_j}(x^*))=0$$
$$g_i(x^*) \leq 0$$
$$μ_i g_i (x^*) =0$$
$$x_j^* \geq 0$$

So the solution cannot have negative values.

Okay... how about (10, 0) then?
Better or worse?
 
  • #5
I like Serena said:
Okay... how about (10, 0) then?
Better or worse?

(Thinking)
Since the objective function at the point (10,0) is -91, and with (0,9) it's 0, I assume that yours is better..

I did the following:
The KKT conditions for this proble are:
KKT1: $μ_1 \geq 0$
KKT2: $-1-μ_1 8x_1 \leq 0$
KKT3: $-2(x_2-9)-μ_1 2 x_2 \leq 0$
KKT4: $ x_1 (-1-μ_1 8x_1 )=0$
KKT5: $ x_2 (-2(x_2-9)-μ_1 2 x_2 )=0$
KKT6: $ 4x_1^2+x_2^2-400 \leq 0$
KKT7: $ μ_1 (4x_1^2+x_2^2-400 )=0$
KKT8: $ x_1 \geq 0$
KKT9: $ x_2 \geq 0$

Case1: $μ_1=0$
KKT4: $-x_1=0 \Rightarrow x_1=0$
KKT5: $-2x_2(x_2-9)=0 \Rightarrow x_2=0 \text{ or } x_2=9$
For $x_2=0$, at KKT3:$18 \leq 0$ that is not true, so $x_2 \neq 0$
Since, for $μ_1=0$, $x_1=0$ and $x_2=9$ all conditions are satisfied, $(0,9)$ is a solution. Then I showed that the function $f$ is concave, so this solution is unique.

Have I done something wrong? :confused:
 
  • #6
mathmari said:
(Thinking)
Since the objective function at the point (10,0) is -91, and with (0,9) it's 0, I assume that yours is better..

I did the following:
The KKT conditions for this proble are:
KKT1: $μ_1 \geq 0$
KKT2: $-1-μ_1 8x_1 \leq 0$
KKT3: $-2(x_2-9)-μ_1 2 x_2 \leq 0$
KKT4: $ x_1 (-1-μ_1 8x_1 )=0$
KKT5: $ x_2 (-2(x_2-9)-μ_1 2 x_2 )=0$
KKT6: $ 4x_1^2+x_2^2-400 \leq 0$
KKT7: $ μ_1 (4x_1^2+x_2^2-400 )=0$
KKT8: $ x_1 \geq 0$
KKT9: $ x_2 \geq 0$

Case1: $μ_1=0$
KKT4: $-x_1=0 \Rightarrow x_1=0$
KKT5: $-2x_2(x_2-9)=0 \Rightarrow x_2=0 \text{ or } x_2=9$
For $x_2=0$, at KKT3:$18 \leq 0$ that is not true, so $x_2 \neq 0$
Since, for $μ_1=0$, $x_1=0$ and $x_2=9$ all conditions are satisfied, $(0,9)$ is a solution. Then I showed that the function $f$ is concave, so this solution is unique.

Have I done something wrong? :confused:
You have found the unique point $(x_1,x_2) = (0,9)$ that maximises the objective function. But you want to minimise it. So it looks as though you need to consider Case2: $μ_1\ne0$.
 
  • #7
Opalg said:
You have found the unique point $(x_1,x_2) = (0,9)$ that maximises the objective function. But you want to minimise it. So it looks as though you need to consider Case2: $μ_1\ne0$.

Ok...Thank you..! :eek:
 

FAQ: Find the solution using the KKT conditions.

What are the KKT conditions?

The Karush-Kuhn-Tucker (KKT) conditions are necessary conditions for a point to be a solution to a constrained optimization problem. They are a generalization of the Lagrange multiplier conditions and take into account both equality and inequality constraints.

How do the KKT conditions help find a solution?

The KKT conditions provide a set of equations that must be satisfied at the optimal solution of a constrained optimization problem. By solving these equations, we can determine the optimal values of the decision variables and the Lagrange multipliers, which represent the sensitivity of the constraints.

Are the KKT conditions applicable to all optimization problems?

No, the KKT conditions are only applicable to convex optimization problems, which are problems where the objective function and constraints are convex. For non-convex problems, the KKT conditions may not hold, and other methods must be used to find a solution.

How do the KKT conditions handle inequality constraints?

The KKT conditions use the concept of complementary slackness to handle inequality constraints. This means that if a constraint is active (holds with equality) at the optimal solution, its corresponding Lagrange multiplier is non-zero. If a constraint is inactive (holds with inequality) at the optimal solution, its corresponding Lagrange multiplier is zero.

Can the KKT conditions be used for both linear and nonlinear optimization problems?

Yes, the KKT conditions can be used for both linear and nonlinear optimization problems as long as they are convex. For linear problems, the KKT conditions reduce to the standard Lagrange multiplier conditions. For nonlinear problems, the KKT conditions provide additional information about the optimal solution.

Similar threads

Replies
3
Views
1K
Replies
9
Views
5K
Replies
5
Views
2K
Replies
3
Views
675
Replies
4
Views
873
Replies
6
Views
2K
Back
Top