Well, the change in f is the value of f at the upperpoint here, which is x0+delta x, and minus its value at the lower point P, which is f of x0, divided by delta x.
The lower-level problem is to maximize the loading margin under a load change direction and the upper-level problem is to find the closest critical point subject to the constraints of load change.