Abstract.
The following important question is frequently overlooked: Which variables should we select to control? It is shown that the idea of selecting the variables that achieve self-optimizing control provides a link between steady-state optimization, feedback control, time scale separation and uncertainty. In summary, the basic idea is to turn the optimization problem into a setpoint problem, and we show that a good candidate variable for a controlled output should have the following properties:

1. Its optimal value is insensitive to disturbances.
2. On the other hand, the variable itself should be sensitive to changes in the manipulated variables (inputs).
3. It should be easy to control accurately.