5
$\begingroup$

This is so called a hyperbolic set

$$\{(x_1, x_2) \in \mathbb{R}^2_{+} \mid x_1 x_2 \geq 1\}$$

We proceed to prove that it is convex by showing that a convex combination of points (a line segment) will lie in the set

Suppose $x = (x_1, x_2)$, $y = (y_1, y_2)$ and $x \geq y$ in the elementwise sense

Then set: $z = \theta(x_1,x_2) + (1-\theta)(y_1, y_2)$

$z = (y_1,y_2) + \theta(x_1 - y_1, x_2 - y_2)$

Clearly, any $z_1,z_2 \in z$ lies in the hyperbolic set (i.e. $z_1z_2 \geq 1$)

How do we proceed to prove the case when $x \not \geq y$ in the elementwise sense?

$\endgroup$
  • 1
    $\begingroup$ You miss some requirement since now $(1,1)$ and $(-1,-1)$ are both in the set, but the origin is not. $\endgroup$ – Dirk Sep 19 '15 at 14:06
2
$\begingroup$

Let $(x_1,y_1)$ and $(x_2, y_2)$ in our set, and $\alpha \in [0,1]$. We need to show that $\alpha (x_1,y_1) + (1-\alpha) (x_2, y_2)$ is in our set, i.e. that $$\alpha^2 x_1y_1 + (1-\alpha)^2 x_2y_2 + \alpha(1-\alpha)(x_1y_2+x_2y_1) \geq 1$$

Hint: notice that $x_1y_1 \geq 1, x_2y_2 \geq 1$ and prove that $x_1y_2 + x_2y_1 \geq 2$ to conclude.

$\endgroup$
2
$\begingroup$

Say $f(x)=\frac{1}{x}$, so our set is $\{(x,y): y\geq f(x)\}$. We know that $f(x)$ is convex, that is for $0\leq\theta\leq 1$ $$f(\theta x+(1-\theta)y)\leq \theta f(x)+(1-\theta)f(y).$$

Now take $(x_1,y_1)$ and $(x_2,y_2)$ in the set. In other words $y_1\geq f(x_1)$ and $y_2\geq f(x_2)$. For $0\leq\theta\leq 1$ $$\theta y_1+(1-\theta)y_2\geq\theta f(x_1)+(1-\theta)f(x_2)\geq f(\theta x_1+(1-\theta)x_2)$$ so it is in the set.

$\endgroup$
  • $\begingroup$ Can you help me clarify some confusion? The domain of your example is $\mathbb{R}$ and the domain in the question is $\mathbb{R^2}$. Why are you using a different domain for your proof ? and how does it relate to the domain in the question? $\endgroup$ – Sother Sep 8 '17 at 16:19
1
$\begingroup$

Your hyperbolic set is defined by the following linear matrix inequality (LMI)

$$\begin{bmatrix} x_1 & 1\\ 1 & x_2\end{bmatrix} \succeq \mathrm O_2$$

which, using Sylvester's criterion, produces the following conjunction of (non-strict) inequalities

$$x_1 \geq 0 \land x_2 \geq 0 \land x_1 x_2 - 1 \geq 0$$

Hence, the hyperbolic set is a spectrahedron and, thus, convex.

$\endgroup$

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.