0
$\begingroup$

I'm given the problem:

The function $f:\mathbb R^n\to\mathbb R$ is convex, $a\in\mathbb R^n$, $b\in\mathbb R$: $$\mathcal B = \lbrace x\in\mathbb R^n : f(x) = 0,\langle a,x\rangle\leq b,x\geq 0\rbrace$$

It's easy to check that if $x_1,x_2 \in B$, then $(1-λ)x_1 + λx_2$ satisfies the last two requirements which are $\langle \alpha,x\rangle \le b $, and $x \ge 0$.

But when I tried to verify the first requirement I get stuck.

This is what I get:

$$f((1-\lambda)x_1+\lambda x_2)\leq (1-\lambda)f(x_1)+\lambda f(x_2)$$ Since $f(x_1) = f(x_2) = 0$, we have that: $$f((1-\lambda x_1)+\lambda x_2)\leq (1-\lambda) 0+\lambda 0 = 0$$

but I don't know how can I prove it is equal to $0$.

Can anyone give a hint?

$\endgroup$
  • $\begingroup$ how is the $x \ge 0$ used in the proof as it does not make sense since $x \in \mathbb{R^n}$ $\endgroup$ – DeepSea Dec 11 '16 at 5:30
  • 1
    $\begingroup$ @DeepSea It probably means that each element of the vector $x$ is non-negative. $\endgroup$ – Theoretical Economist Dec 11 '16 at 5:39
  • $\begingroup$ Don't you think it is $f(x)>0$ instead of $f(x)=0$ ? In this way, the problem boils down to the fact that the intersection of convex sets (here 3 of them) is convex. $\endgroup$ – Jean Marie Dec 11 '16 at 5:56
  • $\begingroup$ @JeanMarie I don't think convexity of $f$ guarantees that $\{ x:f(x)>0 \}$ is convex. It does guarantee that $\{ x:f(x)\leq 0 \}$ is convex, though. $\endgroup$ – Theoretical Economist Dec 11 '16 at 7:25
  • $\begingroup$ @Theoretical Economist Oops, you are perfectly right. Waiting for Chen's reaction now... $\endgroup$ – Jean Marie Dec 11 '16 at 7:44
2
$\begingroup$

I don't think this is true, unless you impose additional conditions.

Suppose, for example, that $n=1$, $f(x)=(x-2)^2 - 4$, $a=1$ and $b=4$. Then,

$$ \mathcal{B}=\{0,4 \}$$

which is not convex.

$\endgroup$
  • $\begingroup$ Wait why is that not convex? $\endgroup$ – IAmNoOne Dec 11 '16 at 5:46
  • $\begingroup$ $0\lambda + 4(1-\lambda)$ is not in $\mathcal{B}$ for $\lambda \in (0,1)$, no? $\endgroup$ – Theoretical Economist Dec 11 '16 at 5:50
  • $\begingroup$ Oh okay on $B$ $\endgroup$ – IAmNoOne Dec 11 '16 at 5:52

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.