The theory of economic models of decentralization is developed from a point of view quite different from the virtual planning phases of economic systems as displayed by market mechanism type procedures. The Central unit transmits appropriate information to each divisional unit which then acts (commits resources) according to optimization with respect to variables under its control. Our theory presents a natural hierarchy of coherently decentralized systems which is based on the increasing amounts of information which are to be transmitted to divisional units for proper decentralized decision making. In general, this information involves more than prices alone and the notion of preemptive goals is introduced for this purpose by means of examining dual convex programming problems. Employing our precise specifications, uniqueness of divisional optima (in contrast to certain trivial cases of decentralization) is not required. Thus, additional flexibility is available to divisions without violating company goals. Further we show that our procedure is a robust one in the sense that approximate fulfillment of the preemptive goals, e.g., non-optimal but close to optimal divisional solutions, results in small deviations from optimal profit.