2
$\begingroup$

I have an optimization problem in the form of

[\begin{array}{l} \mathop {{\rm{Minimize}}}\limits_{\bf{X}} \,\,\,2\left| \delta \right|\sqrt {{\rm{Tr}}\left( {{\bf{A}}{{\bf{X}}^2}} \right)} {\rm{ - }}Tr\left( {{\bf{AX}}} \right)\\ {\rm{Subject to Tr}}\left( {\bf{X}} \right){\rm{ - }}{P_{th}} \le 0\\ & \,\,\,\,\,\,\,\,\,\,{\bf{X}} \succ {\bf{0}} \end{array}]

Where ${\bf{X}} = {\bf{W}}{{\bf{W}}^H}$ and ${\bf{A}} = {\bf{F}}{{\bf{F}}^H}$ which ${\bf{W}} \in {C^{N \times N}}$ and ${\bf{F}} \in {C^{M \times N}}$ , $\delta \in C$ and M, N are scalars. First of all I want to know whether my cost function is convex or concave, specifically the first term. Secondly, how can I convert my optimization problem in a convex form that I could use CVX package to solve it?

$\endgroup$

1 Answer 1

2
$\begingroup$

Your objective is convex, and can be written in conic form (to be precise, a mixed SDP and SOCP). By cyclicity of the trace, and since $X$ is Hermitian $$ Tr(AX^2)=Tr(XAX^H)=Tr(XF(XF)^H) $$ Now, the square root of this term is nothing but the Euclidean norm of $vec(XF)$, which can be written in second-order conic form $$ (z,t)\in L^{N^2},\quad z=vec(XF). $$ Replace the first term in the objective by $2|\delta|t$ and that is a valid input for CVX!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.