1
$\begingroup$

I am trying to find a solution for following

\begin{eqnarray} \text{minimize }~~ -\sum_{i=1}^K \frac{1}{a_i+b_i2^{(-2x_i)}} \\ \text{s.t.} ~~~ \sum_{i=1}^Kx_i=C \end{eqnarray}

where $a_i,b_i,x_i\geq 0$.

when I check the convexity it requires $b_i < a_i 2^{(2x_i)}$ or $x_i>log_2(\sqrt{b_i/a_i})$.

applying KKT conditions yield the solution like following:

\begin{eqnarray} x_i^*=\frac 12\cdot log_2\left({\frac{b_i}{\lambda-2a_i-\sqrt{\lambda^2-4\lambda a_i}}}\right) \end{eqnarray} where $\lambda$ is some Lagrange multiplier.

I have two questions:

1- if all $x_i^*$'s satisfy the constraint $x_i^*>log_2(\sqrt{b_i/a_i})$ can I claim that the solution is optimal?

2-what if some of $x_i^*$'s violate the constraint $x_i^*>log_2(\sqrt{b_i/a_i})$? Can I still claim that solution is optimal? if not so what is the optimal solution !

my simulations corroborate that the solution coming from KKT is optimal for all scenarios, but I am confused with the math.

$\endgroup$
2
  • $\begingroup$ Did you use Lagrange multipliers for the non-negativity contraints $x_i\ge 0$? Can we rule out that some $x_i=0$? $\endgroup$ Oct 3 '15 at 0:38
  • $\begingroup$ @SergioParreiras actually I didn't use Lagrange multipliers for $x_i>0$, and rule out $x_i<0$ , does it help? $\endgroup$
    – Alireza
    Oct 5 '15 at 15:20
0
$\begingroup$

Take a look at chapter 2.G. of "Implicit functions and solution mappings" from Dontchev and Rockafellar. There you can find rather general criteria which guarantee (at least) local optimality.

$\endgroup$
1
  • $\begingroup$ the book has useful stuff, but for my case I am seeking global optimality, the book have very general theorem mainly on local optimality $\endgroup$
    – Alireza
    Oct 5 '15 at 21:36

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.