3
$\begingroup$

I am dealing with convex optimization problems. There are some useful theories for optimization problems where real-valued vector parameter, e.g., $x \in \mathbb{R}^n$, is considered. I manage to apply the theories to my semidefinite programming problem. I mean, I verify it by simulations. However, to prove the convergence of the algorithm analytically, I need to verify whether the set of Hermitian positive semidefinite matrices, where $\bf{X}=\bf{X}^{\rm H}$ and $\bf{X} \succeq 0$, is compact given that ${\rm trace}\{\bf{X}\} \leq 1$. Can anyone help me with this or at least point out a direction I should go for? Thank you very much in advance.

$\endgroup$
6
$\begingroup$

Think of it as a subset of $\mathbb R^{n^2}$. You need to show that the set is closed and bounded. Closed is easy. For boundedness, show that each matrix entry is bounded by $1$. You know this is true for the diagonal entries (because they are all non-negative and add up to 1). Now use the fact that each $2\times 2$ submatrix $\left[\begin{matrix} a_{ii} & a_{ij} \\ a_{ij} & a_{jj} \end{matrix}\right]$ is positive semi-definite, and hence its determinant is non-negative.

Let me add one more thing. That the condition $X$ is positive semidefinite is closed under limits is perhaps most easily shown by noting that this is true if and only if all the diagonal minor matrices have non-negative determinant.

$\endgroup$
  • $\begingroup$ 1) by a subset of $R^n2$, you mean vec{X} \in R^n2 or X can be written as real-valued representation such that X \in R^(n x n). $\endgroup$ – user111614 Nov 29 '13 at 16:16
  • $\begingroup$ I am not sure what you mean by the latter statement. I definitely mean the former. $\endgroup$ – Stephen Montgomery-Smith Nov 29 '13 at 16:18

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.