Manufacturing process real-time optimization has been one of the main digital twins’ operations. It is of utmost importance to the processes, since it enables the feedback of a digital twin towards the real world. However, it is quite difficult to be implemented, since it requires modelling of the process, adaptivity of both the model and the process, real-time communication and link to other functionalities. Under the framework of formalizing such activities, the current work attempts to categorize the types of manufacturing process real-time optimization and show their limitations. For the sake of simplicity, generic process models are adopted and then the requirements for the process control are given, driving the aforementioned definitions. Specific numerical examples are used to illustrate the definitions, while the latter presented herein span all categories of real-time optimization as well as all manufacturing performance indicators. Finally, both mathematically and physics-wise, the limitations are discussed.