The optimal estimation problems for linear dynamic systems, and in particular for systems with aftereffect, reduce to different variational problems. The type and complexity of these variational problems depend on the process model, the model of uncertainties, and the estimation performance criterion. A solution of a variational problem determines an optimal estimator. In addition, frequently the optimal algorithm for one noise model must operate under another, more complex assumption about noise. Hence, simplified algorithms must be used. It is important to evaluate the level of nonoptimality for the simplified algorithms. Since the original variational problems can be very difficult, the estimate of nonoptimality must be obtained without solving the original variational problem. In this book, guaranteed levels of nonoptimality for simplified estimation and control algorithms are constructed. To obtain these levels the duality theory for convex extremal problems is used. Audience: This book will be of interest to applied mathematicians, researchers and engineers who deal with estimation and control systems. The material, which requires a good knowledge of calculus, is also suitable for a two-semester graduate or postgraduate course.

When solving the control and design problems in aerospace and naval engi neering, energetics, economics, biology, etc., we need to know the state of investigated dynamic processes. The presence of inherent uncertainties in the description of these processes and of noises in measurement devices leads to the necessity to construct the estimators for corresponding dynamic systems. The estimators recover the required information about system state from mea surement data. An attempt to solve the estimation problems in an optimal way results in the formulation of different variational problems. The type and complexity of these variational problems depend on the process model, the model of uncertainties, and the estimation performance criterion. A solution of variational problem determines an optimal estimator. Howerever, there exist at least two reasons why we use nonoptimal esti mators. The first reason is that the numerical algorithms for solving the corresponding variational problems can be very difficult for numerical imple mentation. For example, the dimension of these algorithms can be very high."