0
$\begingroup$

Say I intend to maximize a real-valued, strictly concave, twice-differentiable function:

\begin{equation} f:\mathbb{R^n} \rightarrow \mathbb{R} \end{equation}

If the problem is unconstrained, that is: \begin{equation} \underset{x_1,x_2,\dots,x_n}{max} f(x_1,x_2,\dots,x_n) \end{equation} I would take the first derivative, and set equal to zero. This would yield the unique global optimum due to the strict concavity of the problem.

Say I impose a non-negativity constraint, such that my problem becomes:

\begin{aligned} & \underset{x_1,x_2,\dots,x_n}{max} & f(x_1,x_2,\dots,x_n) \\ & & x_1,x_2,\dots,x_n \geq 0 \end{aligned}

Will it be sufficient to take the solution to the unconstrained problem and set any negative $x_i$'s (if any) equal to 0? Or does that depend entirely on what my function look like, regardless of the real-value, strictly concave, and twice-differentiable attributes of it?

In other words, can I bypass using Kuhn-Tucker conditions due to the nature of the function?

$\endgroup$
3
$\begingroup$

Will it be sufficient to take the solution to the unconstrained problem and set any negative $x_i$'s (if any) equal to $0$?

NO.

the function

$$f(x,y) = -(x+y+1)^2 - (x-2y+2)^2$$

is a counterexample. The global maximum of the function is clearly at $x=\frac43, y=-\frac13$, where the otherwise negative function hits $0$. Proving that the function is positive definite should also be fairly simple.

However, the point $(\frac43, 0)$ is not the maximum of the function on $\{x\geq 0,y\geq 0\}$, since $f(\frac43, 0) = -16.\dot 5$ while $f(0, 0.6) = -3.2$ which is a greater value.


Naturally, if all the $x_i$ of the global maximum are positive, then yes, your global maximum is also a maximum on $\mathbb R_+^n$. However, if one of the values is negative, you cannot simply project your point the way you suggest.

$\endgroup$
  • $\begingroup$ @DrT I was missing a square in the definition of the function, I think the function is now strictly concave. $\endgroup$ – 5xum Aug 1 '18 at 9:51
  • $\begingroup$ I accidentally deleted my previous comment. Appreciate the answer @5xum! $\endgroup$ – DrT Aug 1 '18 at 9:52

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.