Finding minimizer of sum of quadratic matrix form minus log(det(X))

127 Views Asked by At

I am looking to solve the following problem. The first task is to comment on the convexity/concavity of the problem and then I need to compute minimizer of the problem.

$$\min f \left( \Psi \right) = ​​\left( \displaystyle\sum​​_{​​i=1}^N y_{​​i}^T\Psi y_{​​i}​​ \right) - \log | \Psi | \quad \text{subject to} \quad \Psi \geq 0$$


My attempt so far:

The first term is a sum of quadratic forms hence convex. $-\log|\Psi|$ is also a convex function in $\Psi$. hence, the problem seems convex to me.

In order to solve it, I have taken derivative w.r.t $\Psi$ and equated it to zero which gives me

$$\sum_{​​i=1}^N​​ ​​( y_{​​i}y_{​​i}​​^T) - (\Psi^{-1})^T = 0$$

So using this, I get

$$\Psi^* =\left(( \sum​​_{​​i=1}^N​​ \left( y_{​​i} y_{​​i}​​^T \right) \right)^{T})^{-1}$$

However, I'm not too confident about this approach. Any help in this regard would be highly appreciated.