One of the main challenges in the optimization-based synthesis of technological schemes is the definition of the superstructure of alternatives that contains the best solution. Most of the works reported have been based on the hand representa­tion of the superstructure for each particular problem without following general rules. To generalize this procedure, the definition of the representation type for the superstructure is needed. Among the main types of representation proposed are the state-task network and the state-equipment network. In the first type, two classes of nodes (states and tasks) are used for the representation; the assignment of equipment is dealt implicitly through the model. In the second type of repre­sentation, the states and the equipments are employed as the nodes; the tasks in this case are treated implicitly through the model (Grossmann et al., 2000).

Other crucial aspect during optimization of superstructures consists in how to generate in a systematic way the superstructure so that all the alternatives of interest are included. One of the trends in this field is the automatic generation of superstructures, which has been developed for the case of lineal process net­works employing an algorithm based on graphs and ensuring a search space large enough to include the optimal solution (Friedler et al., 1993). Fraga (1998) and Fraga et al. (2000) have developed the Jacaranda system that is able to solve a process synthesis problem through the automatic generation of the superstructure at the same time that it executes the search procedure. The Jacaranda system is based on the use of an implicit enumeration procedure that generates and ana­lyzes a directed graph representing the synthesis problem.

The level of detail of the optimization model also plays an important role for the optimization-based approach. In general, the models can be classified into three main classes: aggregated models, short-cut models, and rigorous models (Grossmann et al., 2000). The first class of models employs high-level representa­tions in which the synthesis problem is simplified by one aspect or objective that tends to dominate the problem as mentioned above. Models for predicting the minimum utilities and the minimum amount of units in a heat exchange and mass exchange networks belong in this category. Short-cut models are employed in superstructures with a high level of detail and involve the optimization of capital and operating costs, but in which the performance of individual units is predicted with relatively simple nonlinear models in order to reduce the computational effort. Among the examples of this type, distillation sequences and technological schemes can be highlighted. Finally, rigorous models are also applied in detailed superstructures for predicting the behavior of the units as in the case of the syn­thesis of distillation trains for separation of ideal and nonideal mixtures.