Why the optimal value is different

In my optimal solution, i use sequential convex approximation method, and i should find a feasible initial point, the problem is that when i use different initial point, i get different optimal value.

Do you have a proof that cannot happen?

If the problem is non-convex, SCA will produce a locally optimal solution … if you’re lucky. That locally optimal solution can depend on the initial point, and may or may not be globally optimal.

if the problem is convex, there’s not much point in using SCA, unless it is applied to a convex problem which is not DCP-representable (can’t be input into CVX).