My main question is that when I use Lagrange Multipliers/KKT conditions to perform optimization with threshold constraints, I seem to get contradictory FOC.

Here is a characteristic example: take an optimization problem like the following:

$$ \begin{align*} &\max_{x,y} 2x+y-(x-y)^2\\ &s.t.\\ &x,y \geq 0\\ &x+y \leq I \end{align*} $$ For some value $I>0$. If I translate this directly to a lagrangian I believe the KKT conditions should give that I need to optimize: $$ \mathcal{L}(x,y,\mu_1, \mu_2, \mu_3)=2x+y-(x-y)^2 + \mu_1(-x) + \mu_2(-y)+\mu_3(x+y-I) $$ However partial derivatives w.r.t $\mu_1, \mu_2,\mu_3$ gives: $$\begin{align*} &-x=0\\ &-y=0\\ &x+y-I=0\\ \end{align*}$$ which gives a contradiction for $I>0$.

Could someone help me understand why I seem to get this contradiction?


1 Answer 1


Instead of differentiating w.r.t. the multipliers you should properly write the complementary slackness condition (defined in your link). In your case this is

$$-\mu_1 x=0$$

$$-\mu_2 y=0$$

$$\mu_3 (x+y-I)=0$$

This says that either the constraint binds with equality or the multiplier is zero. If you just differentiate with respect to $\mu_1$ for instance you would incorrectly conclude that always $x=0$, while this can be strictly positive.

  • $\begingroup$ Ahh got it. So at an optimum we check for FOC for the term we wish to optimize but complementary slackness for inequality terms? $\endgroup$
    – naveace
    Feb 21, 2022 at 0:19
  • $\begingroup$ @naveace Yep correct. $\endgroup$
    – fes
    Feb 21, 2022 at 9:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.