I'm a bit confused about Lagrange multipliers. I know it works wonders if I only have equality constraints. Whenever I have inequality constraints, or both, I use Kuhn-Tucker conditions and it does the job.

But my question is, can I solve a inequality constraint problem using only Lagrange multiplier?

For example:

min $f(x)=x^4$ with $x \leq -1$.

$L(x,\lambda)=x^4+ \lambda(x+1)$, gives:

$4x^3+ \lambda =0$ and $x+1=0 \Leftrightarrow x=-1 , \lambda =4$

Is this correct? I just assumed equality.. What happens if the constraint is not active?


Introducing the slack variable $\epsilon$ we can transform an inequality into an equality. So with this idea we can write the equivalent Lagrangian

$$ L(x,\lambda,\epsilon) = x^4+\lambda(x+1+\epsilon^2) $$

The stationary points are solved by

$$ L_x = 4x^3+\lambda = 0\\ L_{\lambda} = x+1+\epsilon^2 = 0\\ L_{\epsilon} = 2\lambda\epsilon = 0 $$

Now, depending of the values for $\epsilon,\lambda$ we can qualify those stationary points.


After solving we have the feasible solution $x = -1,\ \,\lambda = 4,\ \ \epsilon = 0$ which indicates that the restriction actuates.

  • 1
    $\begingroup$ I've seen this somewhere also. Is this the same as assuming that if constraint is active then $\lambda \geq 0$ and if its inactive we have $\lambda=0$? What I just said comes from what you did right? $\endgroup$ – Amateur Mathematician Dec 5 '18 at 23:03
  • $\begingroup$ forgot to tag you $\endgroup$ – Amateur Mathematician Dec 5 '18 at 23:40
  • $\begingroup$ @AmateurMathematician $\epsilon = 0$ indicates that the restriction is active. $\lambda > 0$ indicates that the objective function gradient $4 x^3$ points in the same direction as the restriction, at the solution point. $\endgroup$ – Cesareo Dec 5 '18 at 23:51
  • $\begingroup$ The stationary points are just the KKT conditions. $\endgroup$ – LinAlg Dec 6 '18 at 1:22
  • $\begingroup$ @Cesareo can tou recommend me a book on this 'slack variable'? Also why do the slack variable squared? $\endgroup$ – Amateur Mathematician Dec 6 '18 at 8:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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