0
$\begingroup$

$$C = conv\{xx^T | x\succeq0\}$$

As far as I know, $x\succeq0$ means x is a symmetric, positive semidefinite matrix, therefore the $xx^T$ here is also symmetric and positive semidefinite.

As the "Positive Semedefinite Cone" is defined as $\mathbb{S}^{n}_{+} = \{\mathbf{X}\in\mathbb{S}^{n}: \mathbf{X}\succeq\mathbf{0}\}$, then $\{xx^T | x\succeq0\}$ should've already been a convex cone here.

How to understand this $C$, does it over-define a convex set?

Thank you!

$\endgroup$
  • $\begingroup$ I'm not sure what the question is... what property of $C$ do you want to understand? $\endgroup$ – user7530 Jan 16 '18 at 8:43
  • $\begingroup$ sorry for the vagueness, I just added some explanation $\endgroup$ – XJY95 Jan 16 '18 at 8:53
  • $\begingroup$ You've written $xx^{T}$ with a lower case $x$, which would normally refer to a vector $x$ rather than a matrix. $x \succeq 0$ could be an odd way of saying that the elements of the vector $x$ are nonnegative. I wonder if you're misreading your source. It would help if you could link to it. $\endgroup$ – Brian Borchers Jan 17 '18 at 5:22
  • $\begingroup$ sorry I missed this one, x here is actually a vector $\endgroup$ – XJY95 Jan 22 '18 at 3:08
1
$\begingroup$

I don't know what "over-define" means. Certainly each non-extreme element of any convex set $C$ has more than one representation as a convex combination of elements of $C$, so it is over-defined in one sense. I also don't completely understand your definition of $C$.

The set $D=\operatorname{conv}\{XX^T: X\in \mathbb S_+^n\}$ is equal to the set $\mathbb S_+^n$ as follows. Among the elements of $\mathbb S_+^n$ are the rank $1$ matrices of form $X=vv^T$ for $v\in\mathbb R^n$. But the convex hull of such matrices is the set of psd matrices, so $D \subseteq \mathbb S_+^n$. Since elements of $\mathbb S_+^n$ posessess psd square roots, each element $Y\in\mathbb S_+^n$ can be written in the form $Y=XX=XX^T$ for some $X\in\mathbb S_+^n$. Hence $\mathbb S_+^n\subseteq D$. Hence $D=\mathbb S_+^n$.

If your $C$ is my $D$ then it true that your definition of $C$ is another equivalent characterization of the psd matrices.

$\endgroup$
  • $\begingroup$ Thank you! It's kind of wired that I was told the x here is just a vector... $\endgroup$ – XJY95 Jan 17 '18 at 8:39

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.