I have attached the original problem and my model. I used a lot of 2*2 semidefinte matrix inequality to express the sum of square. It takes cvx 9 seconds to model a small problem, and it takes MOSEK 0.2 seconds to solve the problem.
I want to express the sum of square in rotated quadratic cones. However, someone posted a problem that it takes cvx a lot of time to build a lot of rotated quadratic cones. The problem was solved by using geo_mean.
How can I formulate this problem more efficiently? Is geo_mean better than rotated quadratic cone for my problem?