Many physical processes are expensive or impracticable to observe directly, and certainly they can only be observed in operation under conditions that actually exist. Almost any industrial process, almost any natural biological, physical, or social environment, and almost any large computing system is quite complex. Such as, building a large chemical plant based on a newly discovered sequence of reactions without knowing that it will work on a large scale much as it does in the laboratory would not be wise. In such a case it is worthwhile to model such processes and to study assumption that the surrogate for the the model extensively, with the behavior of the model is a valid behavior of the process itself.
