The purpose of the proposed research is to develop new concepts, methods, and algorithms for the decentralized control of complex nonlinear systems. These systems are characterized by high dimensions, uncertainty, and information structure constraints. The main thrust of the research is directed toward the development of new decomposition methods and algorithms for a "piece-by-piece" design of control and estimation structures for complex plants. These algorithms can be implemented by using the modern parallel multiprocessing architectures. Of special interest are the graph- theoretical methods for tearing of dynamic systems because of their inherent simplicity and numerical reliability. The partitioning framework includes the overlapping decompositions of nonlinear systems involving the unorthodox notion of subsystems that share common parts. This type of decomposition can succeed where the standard disjoint compositions fail.