x + inv_pos(y) is convex, so can be minimized in CVX, but not maximized.
As to Dinkelbach formulations?
I presume there are other unstated constraints; otherwise, the optimal solution is trivial by inspection (x at upper bound, y at lower bound).